1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
5 Copyright (C) 1993-1999, 2000, 2001 Free Software Foundation, Inc.
6 This file is part of the GNU C Library.
8 The GNU C Library is free software; you can redistribute it and/or
9 modify it under the terms of the GNU Lesser General Public
10 License as published by the Free Software Foundation; either
11 version 2.1 of the License, or (at your option) any later version.
13 The GNU C Library is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public
19 License along with the GNU C Library; if not, write to the Free
20 Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
23 /* AIX requires this to be the first thing in the file. */
24 #if defined _AIX && !defined REGEX_MALLOC
35 #ifdef USE_INCLUDED_REGEX
37 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
38 # define PARAMS(args) args
40 # define PARAMS(args) ()
42 #endif /* Not PARAMS. */
44 #ifndef INSIDE_RECURSION
46 # if defined STDC_HEADERS && !defined emacs
49 /* We need this for `regex.h', and perhaps for the Emacs include files. */
50 # include <sys/types.h>
54 #define HAVE_WCTYPE_H 0
57 # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
59 /* For platform which support the ISO C amendement 1 functionality we
60 support user defined character classes. */
61 # if defined _LIBC || (defined WIDE_CHAR_SUPPORT && WIDE_CHAR_SUPPORT)
62 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
68 /* We have to keep the namespace clean. */
69 # define regfree(preg) __regfree (preg)
70 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
71 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
72 # define regerror(errcode, preg, errbuf, errbuf_size) \
73 __regerror(errcode, preg, errbuf, errbuf_size)
74 # define re_set_registers(bu, re, nu, st, en) \
75 __re_set_registers (bu, re, nu, st, en)
76 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
77 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
78 # define re_match(bufp, string, size, pos, regs) \
79 __re_match (bufp, string, size, pos, regs)
80 # define re_search(bufp, string, size, startpos, range, regs) \
81 __re_search (bufp, string, size, startpos, range, regs)
82 # define re_compile_pattern(pattern, length, bufp) \
83 __re_compile_pattern (pattern, length, bufp)
84 # define re_set_syntax(syntax) __re_set_syntax (syntax)
85 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
86 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
87 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
89 # define btowc __btowc
91 /* We are also using some library internals. */
92 # include <locale/localeinfo.h>
93 # include <locale/elem-hash.h>
94 # include <langinfo.h>
95 # include <locale/coll-lookup.h>
98 /* This is for other GNU distributions with internationalized messages. */
99 # if (defined HAVE_LIBINTL_H && HAVE_LIBINTL_H && defined ENABLE_NLS && ENABLE_NLS) || defined _LIBC
100 # include <libintl.h>
103 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
106 # define gettext(msgid) (msgid)
109 # ifndef gettext_noop
110 /* This define is so xgettext can find the internationalizable
112 # define gettext_noop(String) String
115 /* The `emacs' switch turns on certain matching commands
116 that make sense only in Emacs. */
123 # else /* not emacs */
125 /* If we are not linking with Emacs proper,
126 we can't use the relocating allocator
127 even if config.h says that we can. */
130 # if defined STDC_HEADERS || defined _LIBC
137 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
138 If nothing else has been done, use the method below. */
139 # ifdef INHIBIT_STRING_HEADER
140 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
141 # if !defined bzero && !defined bcopy
142 # undef INHIBIT_STRING_HEADER
147 /* This is the normal way of making sure we have a bcopy and a bzero.
148 This is used in most programs--a few other programs avoid this
149 by defining INHIBIT_STRING_HEADER. */
150 # ifndef INHIBIT_STRING_HEADER
151 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
155 # define bzero(s, n) (memset (s, '\0', n), (s))
157 # define bzero(s, n) __bzero (s, n)
161 # include <strings.h>
163 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
166 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
171 /* Define the syntax stuff for \<, \>, etc. */
173 /* This must be nonzero for the wordchar and notwordchar pattern
174 commands in re_match_2. */
179 # ifdef SWITCH_ENUM_BUG
180 # define SWITCH_ENUM_CAST(x) ((int)(x))
182 # define SWITCH_ENUM_CAST(x) (x)
185 # endif /* not emacs */
187 # if defined _LIBC || (defined HAVE_LIMITS_H && HAVE_LIMITS_H)
192 # define MB_LEN_MAX 1
195 /* Get the interface, including the syntax bits. */
198 /* isalpha etc. are used for the character classes. */
201 /* Jim Meyering writes:
203 "... Some ctype macros are valid only for character codes that
204 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
205 using /bin/cc or gcc but without giving an ansi option). So, all
206 ctype uses should be through macros like ISPRINT... If
207 STDC_HEADERS is defined, then autoconf has verified that the ctype
208 macros don't need to be guarded with references to isascii. ...
209 Defining isascii to 1 should let any compiler worth its salt
210 eliminate the && through constant folding."
211 Solaris defines some of these symbols so we must undefine them first. */
214 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
215 # define ISASCII(c) 1
217 # define ISASCII(c) isascii(c)
221 # define ISBLANK(c) (ISASCII (c) && isblank (c))
223 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
226 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
228 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
232 # define ISPRINT(c) (ISASCII (c) && isprint (c))
233 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
234 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
235 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
236 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
237 # define ISLOWER(c) (ISASCII (c) && islower (c))
238 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
239 # define ISSPACE(c) (ISASCII (c) && isspace (c))
240 # define ISUPPER(c) (ISASCII (c) && isupper (c))
241 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
244 # define TOLOWER(c) _tolower(c)
246 # define TOLOWER(c) tolower(c)
250 # define NULL (void *)0
253 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
254 since ours (we hope) works properly with all combinations of
255 machines, compilers, `char' and `unsigned char' argument types.
256 (Per Bothner suggested the basic approach.) */
257 # undef SIGN_EXTEND_CHAR
258 # if defined __STDC__ && __STDC__
259 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
260 # else /* not __STDC__ */
261 /* As in Harbison and Steele. */
262 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
266 /* How many characters in the character set. */
267 # define CHAR_SET_SIZE 256
271 extern char *re_syntax_table
;
273 # else /* not SYNTAX_TABLE */
275 static char re_syntax_table
[CHAR_SET_SIZE
];
277 static void init_syntax_once
PARAMS ((void));
287 bzero (re_syntax_table
, sizeof re_syntax_table
);
289 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
291 re_syntax_table
[c
] = Sword
;
293 re_syntax_table
['_'] = Sword
;
298 # endif /* not SYNTAX_TABLE */
300 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
304 /* Integer type for pointers. */
305 # if !defined _LIBC && !defined HAVE_UINTPTR_T
306 typedef unsigned long int uintptr_t;
309 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
310 use `alloca' instead of `malloc'. This is because using malloc in
311 re_search* or re_match* could cause memory leaks when C-g is used in
312 Emacs; also, malloc is slower and causes storage fragmentation. On
313 the other hand, malloc is more portable, and easier to debug.
315 Because we sometimes use alloca, some routines have to be macros,
316 not functions -- `alloca'-allocated space disappears at the end of the
317 function it is called in. */
321 # define REGEX_ALLOCATE malloc
322 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
323 # define REGEX_FREE free
325 # else /* not REGEX_MALLOC */
327 /* Emacs already defines alloca, sometimes. */
330 /* Make alloca work the best possible way. */
332 # define alloca __builtin_alloca
333 # else /* not __GNUC__ */
334 # if defined HAVE_ALLOCA_H && HAVE_ALLOCA_H
336 # endif /* HAVE_ALLOCA_H */
337 # endif /* not __GNUC__ */
339 # endif /* not alloca */
341 # define REGEX_ALLOCATE alloca
343 /* Assumes a `char *destination' variable. */
344 # define REGEX_REALLOCATE(source, osize, nsize) \
345 (destination = (char *) alloca (nsize), \
346 memcpy (destination, source, osize))
348 /* No need to do anything to free, after alloca. */
349 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
351 # endif /* not REGEX_MALLOC */
353 /* Define how to allocate the failure stack. */
355 # if defined REL_ALLOC && defined REGEX_MALLOC
357 # define REGEX_ALLOCATE_STACK(size) \
358 r_alloc (&failure_stack_ptr, (size))
359 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
360 r_re_alloc (&failure_stack_ptr, (nsize))
361 # define REGEX_FREE_STACK(ptr) \
362 r_alloc_free (&failure_stack_ptr)
364 # else /* not using relocating allocator */
368 # define REGEX_ALLOCATE_STACK malloc
369 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
370 # define REGEX_FREE_STACK free
372 # else /* not REGEX_MALLOC */
374 # define REGEX_ALLOCATE_STACK alloca
376 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
377 REGEX_REALLOCATE (source, osize, nsize)
378 /* No need to explicitly free anything. */
379 # define REGEX_FREE_STACK(arg)
381 # endif /* not REGEX_MALLOC */
382 # endif /* not using relocating allocator */
385 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
386 `string1' or just past its end. This works if PTR is NULL, which is
388 # define FIRST_STRING_P(ptr) \
389 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
391 /* (Re)Allocate N items of type T using malloc, or fail. */
392 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
393 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
394 # define RETALLOC_IF(addr, n, t) \
395 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
396 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
398 # define BYTEWIDTH 8 /* In bits. */
400 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
404 # define MAX(a, b) ((a) > (b) ? (a) : (b))
405 # define MIN(a, b) ((a) < (b) ? (a) : (b))
407 typedef char boolean
;
411 static reg_errcode_t byte_regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
413 struct re_pattern_buffer
*bufp
));
415 static int byte_re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
416 const char *string1
, int size1
,
417 const char *string2
, int size2
,
419 struct re_registers
*regs
,
421 static int byte_re_search_2
PARAMS ((struct re_pattern_buffer
*bufp
,
422 const char *string1
, int size1
,
423 const char *string2
, int size2
,
424 int startpos
, int range
,
425 struct re_registers
*regs
, int stop
));
426 static int byte_re_compile_fastmap
PARAMS ((struct re_pattern_buffer
*bufp
));
429 static reg_errcode_t wcs_regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
431 struct re_pattern_buffer
*bufp
));
434 static int wcs_re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
435 const char *cstring1
, int csize1
,
436 const char *cstring2
, int csize2
,
438 struct re_registers
*regs
,
440 wchar_t *string1
, int size1
,
441 wchar_t *string2
, int size2
,
442 int *mbs_offset1
, int *mbs_offset2
));
443 static int wcs_re_search_2
PARAMS ((struct re_pattern_buffer
*bufp
,
444 const char *string1
, int size1
,
445 const char *string2
, int size2
,
446 int startpos
, int range
,
447 struct re_registers
*regs
, int stop
));
448 static int wcs_re_compile_fastmap
PARAMS ((struct re_pattern_buffer
*bufp
));
451 /* These are the command codes that appear in compiled regular
452 expressions. Some opcodes are followed by argument bytes. A
453 command code can specify any interpretation whatsoever for its
454 arguments. Zero bytes may appear in the compiled regular expression. */
460 /* Succeed right away--no more backtracking. */
463 /* Followed by one byte giving n, then by n literal bytes. */
467 /* Same as exactn, but contains binary data. */
471 /* Matches any (more or less) character. */
474 /* Matches any one char belonging to specified set. First
475 following byte is number of bitmap bytes. Then come bytes
476 for a bitmap saying which chars are in. Bits in each byte
477 are ordered low-bit-first. A character is in the set if its
478 bit is 1. A character too large to have a bit in the map is
479 automatically not in the set. */
480 /* ifdef MBS_SUPPORT, following element is length of character
481 classes, length of collating symbols, length of equivalence
482 classes, length of character ranges, and length of characters.
483 Next, character class element, collating symbols elements,
484 equivalence class elements, range elements, and character
486 See regex_compile function. */
489 /* Same parameters as charset, but match any character that is
490 not one of those specified. */
493 /* Start remembering the text that is matched, for storing in a
494 register. Followed by one byte with the register number, in
495 the range 0 to one less than the pattern buffer's re_nsub
496 field. Then followed by one byte with the number of groups
497 inner to this one. (This last has to be part of the
498 start_memory only because we need it in the on_failure_jump
502 /* Stop remembering the text that is matched and store it in a
503 memory register. Followed by one byte with the register
504 number, in the range 0 to one less than `re_nsub' in the
505 pattern buffer, and one byte with the number of inner groups,
506 just like `start_memory'. (We need the number of inner
507 groups here because we don't have any easy way of finding the
508 corresponding start_memory when we're at a stop_memory.) */
511 /* Match a duplicate of something remembered. Followed by one
512 byte containing the register number. */
515 /* Fail unless at beginning of line. */
518 /* Fail unless at end of line. */
521 /* Succeeds if at beginning of buffer (if emacs) or at beginning
522 of string to be matched (if not). */
525 /* Analogously, for end of buffer/string. */
528 /* Followed by two byte relative address to which to jump. */
531 /* Same as jump, but marks the end of an alternative. */
534 /* Followed by two-byte relative address of place to resume at
535 in case of failure. */
536 /* ifdef MBS_SUPPORT, the size of address is 1. */
539 /* Like on_failure_jump, but pushes a placeholder instead of the
540 current string position when executed. */
541 on_failure_keep_string_jump
,
543 /* Throw away latest failure point and then jump to following
544 two-byte relative address. */
545 /* ifdef MBS_SUPPORT, the size of address is 1. */
548 /* Change to pop_failure_jump if know won't have to backtrack to
549 match; otherwise change to jump. This is used to jump
550 back to the beginning of a repeat. If what follows this jump
551 clearly won't match what the repeat does, such that we can be
552 sure that there is no use backtracking out of repetitions
553 already matched, then we change it to a pop_failure_jump.
554 Followed by two-byte address. */
555 /* ifdef MBS_SUPPORT, the size of address is 1. */
558 /* Jump to following two-byte address, and push a dummy failure
559 point. This failure point will be thrown away if an attempt
560 is made to use it for a failure. A `+' construct makes this
561 before the first repeat. Also used as an intermediary kind
562 of jump when compiling an alternative. */
563 /* ifdef MBS_SUPPORT, the size of address is 1. */
566 /* Push a dummy failure point and continue. Used at the end of
570 /* Followed by two-byte relative address and two-byte number n.
571 After matching N times, jump to the address upon failure. */
572 /* ifdef MBS_SUPPORT, the size of address is 1. */
575 /* Followed by two-byte relative address, and two-byte number n.
576 Jump to the address N times, then fail. */
577 /* ifdef MBS_SUPPORT, the size of address is 1. */
580 /* Set the following two-byte relative address to the
581 subsequent two-byte number. The address *includes* the two
583 /* ifdef MBS_SUPPORT, the size of address is 1. */
586 wordchar
, /* Matches any word-constituent character. */
587 notwordchar
, /* Matches any char that is not a word-constituent. */
589 wordbeg
, /* Succeeds if at word beginning. */
590 wordend
, /* Succeeds if at word end. */
592 wordbound
, /* Succeeds if at a word boundary. */
593 notwordbound
/* Succeeds if not at a word boundary. */
596 ,before_dot
, /* Succeeds if before point. */
597 at_dot
, /* Succeeds if at point. */
598 after_dot
, /* Succeeds if after point. */
600 /* Matches any character whose syntax is specified. Followed by
601 a byte which contains a syntax code, e.g., Sword. */
604 /* Matches any character whose syntax is not that specified. */
608 #endif /* not INSIDE_RECURSION */
613 # define UCHAR_T unsigned char
614 # define COMPILED_BUFFER_VAR bufp->buffer
615 # define OFFSET_ADDRESS_SIZE 2
616 # define PREFIX(name) byte_##name
617 # define ARG_PREFIX(name) name
618 # define PUT_CHAR(c) putchar (c)
621 # define CHAR_T wchar_t
622 # define UCHAR_T wchar_t
623 # define COMPILED_BUFFER_VAR wc_buffer
624 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
625 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
626 # define PREFIX(name) wcs_##name
627 # define ARG_PREFIX(name) c##name
628 /* Should we use wide stream?? */
629 # define PUT_CHAR(c) printf ("%C", c);
635 # define INSIDE_RECURSION
637 # undef INSIDE_RECURSION
640 # define INSIDE_RECURSION
642 # undef INSIDE_RECURSION
646 #ifdef INSIDE_RECURSION
647 /* Common operations on the compiled pattern. */
649 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
650 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
653 # define STORE_NUMBER(destination, number) \
655 *(destination) = (UCHAR_T)(number); \
658 # define STORE_NUMBER(destination, number) \
660 (destination)[0] = (number) & 0377; \
661 (destination)[1] = (number) >> 8; \
665 /* Same as STORE_NUMBER, except increment DESTINATION to
666 the byte after where the number is stored. Therefore, DESTINATION
667 must be an lvalue. */
668 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
670 # define STORE_NUMBER_AND_INCR(destination, number) \
672 STORE_NUMBER (destination, number); \
673 (destination) += OFFSET_ADDRESS_SIZE; \
676 /* Put into DESTINATION a number stored in two contiguous bytes starting
678 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
681 # define EXTRACT_NUMBER(destination, source) \
683 (destination) = *(source); \
686 # define EXTRACT_NUMBER(destination, source) \
688 (destination) = *(source) & 0377; \
689 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
694 static void PREFIX(extract_number
) _RE_ARGS ((int *dest
, UCHAR_T
*source
));
696 PREFIX(extract_number
) (dest
, source
)
703 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
704 *dest
= *source
& 0377;
709 # ifndef EXTRACT_MACROS /* To debug the macros. */
710 # undef EXTRACT_NUMBER
711 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
712 # endif /* not EXTRACT_MACROS */
716 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
717 SOURCE must be an lvalue. */
719 # define EXTRACT_NUMBER_AND_INCR(destination, source) \
721 EXTRACT_NUMBER (destination, source); \
722 (source) += OFFSET_ADDRESS_SIZE; \
726 static void PREFIX(extract_number_and_incr
) _RE_ARGS ((int *destination
,
729 PREFIX(extract_number_and_incr
) (destination
, source
)
733 PREFIX(extract_number
) (destination
, *source
);
734 *source
+= OFFSET_ADDRESS_SIZE
;
737 # ifndef EXTRACT_MACROS
738 # undef EXTRACT_NUMBER_AND_INCR
739 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
740 PREFIX(extract_number_and_incr) (&dest, &src)
741 # endif /* not EXTRACT_MACROS */
747 /* If DEBUG is defined, Regex prints many voluminous messages about what
748 it is doing (if the variable `debug' is nonzero). If linked with the
749 main program in `iregex.c', you can enter patterns and strings
750 interactively. And if linked with the main program in `main.c' and
751 the other test files, you can run the already-written tests. */
755 # ifndef DEFINED_ONCE
757 /* We use standard I/O for debugging. */
760 /* It is useful to test things that ``must'' be true when debugging. */
765 # define DEBUG_STATEMENT(e) e
766 # define DEBUG_PRINT1(x) if (debug) printf (x)
767 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
768 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
769 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
770 # endif /* not DEFINED_ONCE */
772 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
773 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
774 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
775 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
778 /* Print the fastmap in human-readable form. */
780 # ifndef DEFINED_ONCE
782 print_fastmap (fastmap
)
785 unsigned was_a_range
= 0;
788 while (i
< (1 << BYTEWIDTH
))
794 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
808 # endif /* not DEFINED_ONCE */
811 /* Print a compiled pattern string in human-readable form, starting at
812 the START pointer into it and ending just before the pointer END. */
815 PREFIX(print_partial_compiled_pattern
) (start
, end
)
830 /* Loop over pattern commands. */
834 printf ("%td:\t", p
- start
);
836 printf ("%ld:\t", (long int) (p
- start
));
839 switch ((re_opcode_t
) *p
++)
847 printf ("/exactn/%d", mcnt
);
859 printf ("/exactn_bin/%d", mcnt
);
862 printf("/%lx", (long int) *p
++);
866 # endif /* MBS_SUPPORT */
870 printf ("/start_memory/%d/%ld", mcnt
, (long int) *p
++);
875 printf ("/stop_memory/%d/%ld", mcnt
, (long int) *p
++);
879 printf ("/duplicate/%ld", (long int) *p
++);
892 printf ("/charset [%s",
893 (re_opcode_t
) *(workp
- 1) == charset_not
? "^" : "");
895 length
= *workp
++; /* the length of char_classes */
896 for (i
=0 ; i
<length
; i
++)
897 printf("[:%lx:]", (long int) *p
++);
898 length
= *workp
++; /* the length of collating_symbol */
899 for (i
=0 ; i
<length
;)
903 PUT_CHAR((i
++,*p
++));
907 length
= *workp
++; /* the length of equivalence_class */
908 for (i
=0 ; i
<length
;)
912 PUT_CHAR((i
++,*p
++));
916 length
= *workp
++; /* the length of char_range */
917 for (i
=0 ; i
<length
; i
++)
919 wchar_t range_start
= *p
++;
920 wchar_t range_end
= *p
++;
921 printf("%C-%C", range_start
, range_end
);
923 length
= *workp
++; /* the length of char */
924 for (i
=0 ; i
<length
; i
++)
928 register int c
, last
= -100;
929 register int in_range
= 0;
931 printf ("/charset [%s",
932 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
934 assert (p
+ *p
< pend
);
936 for (c
= 0; c
< 256; c
++)
938 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
940 /* Are we starting a range? */
941 if (last
+ 1 == c
&& ! in_range
)
946 /* Have we broken a range? */
947 else if (last
+ 1 != c
&& in_range
)
977 case on_failure_jump
:
978 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
980 printf ("/on_failure_jump to %td", p
+ mcnt
- start
);
982 printf ("/on_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
986 case on_failure_keep_string_jump
:
987 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
989 printf ("/on_failure_keep_string_jump to %td", p
+ mcnt
- start
);
991 printf ("/on_failure_keep_string_jump to %ld",
992 (long int) (p
+ mcnt
- start
));
996 case dummy_failure_jump
:
997 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
999 printf ("/dummy_failure_jump to %td", p
+ mcnt
- start
);
1001 printf ("/dummy_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
1005 case push_dummy_failure
:
1006 printf ("/push_dummy_failure");
1009 case maybe_pop_jump
:
1010 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1012 printf ("/maybe_pop_jump to %td", p
+ mcnt
- start
);
1014 printf ("/maybe_pop_jump to %ld", (long int) (p
+ mcnt
- start
));
1018 case pop_failure_jump
:
1019 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1021 printf ("/pop_failure_jump to %td", p
+ mcnt
- start
);
1023 printf ("/pop_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
1028 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1030 printf ("/jump_past_alt to %td", p
+ mcnt
- start
);
1032 printf ("/jump_past_alt to %ld", (long int) (p
+ mcnt
- start
));
1037 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1039 printf ("/jump to %td", p
+ mcnt
- start
);
1041 printf ("/jump to %ld", (long int) (p
+ mcnt
- start
));
1046 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1048 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1050 printf ("/succeed_n to %td, %d times", p1
- start
, mcnt2
);
1052 printf ("/succeed_n to %ld, %d times",
1053 (long int) (p1
- start
), mcnt2
);
1058 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1060 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1061 printf ("/jump_n to %d, %d times", p1
- start
, mcnt2
);
1065 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1067 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1069 printf ("/set_number_at location %td to %d", p1
- start
, mcnt2
);
1071 printf ("/set_number_at location %ld to %d",
1072 (long int) (p1
- start
), mcnt2
);
1077 printf ("/wordbound");
1081 printf ("/notwordbound");
1085 printf ("/wordbeg");
1089 printf ("/wordend");
1094 printf ("/before_dot");
1102 printf ("/after_dot");
1106 printf ("/syntaxspec");
1108 printf ("/%d", mcnt
);
1112 printf ("/notsyntaxspec");
1114 printf ("/%d", mcnt
);
1119 printf ("/wordchar");
1123 printf ("/notwordchar");
1135 printf ("?%ld", (long int) *(p
-1));
1142 printf ("%td:\tend of pattern.\n", p
- start
);
1144 printf ("%ld:\tend of pattern.\n", (long int) (p
- start
));
1150 PREFIX(print_compiled_pattern
) (bufp
)
1151 struct re_pattern_buffer
*bufp
;
1153 UCHAR_T
*buffer
= (UCHAR_T
*) bufp
->buffer
;
1155 PREFIX(print_partial_compiled_pattern
) (buffer
, buffer
1156 + bufp
->used
/ sizeof(UCHAR_T
));
1157 printf ("%ld bytes used/%ld bytes allocated.\n",
1158 bufp
->used
, bufp
->allocated
);
1160 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1162 printf ("fastmap: ");
1163 print_fastmap (bufp
->fastmap
);
1167 printf ("re_nsub: %Zd\t", bufp
->re_nsub
);
1169 printf ("re_nsub: %ld\t", (long int) bufp
->re_nsub
);
1171 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1172 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1173 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
1174 printf ("no_sub: %d\t", bufp
->no_sub
);
1175 printf ("not_bol: %d\t", bufp
->not_bol
);
1176 printf ("not_eol: %d\t", bufp
->not_eol
);
1177 printf ("syntax: %lx\n", bufp
->syntax
);
1178 /* Perhaps we should print the translate table? */
1183 PREFIX(print_double_string
) (where
, string1
, size1
, string2
, size2
)
1184 const CHAR_T
*where
;
1185 const CHAR_T
*string1
;
1186 const CHAR_T
*string2
;
1198 if (FIRST_STRING_P (where
))
1200 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1201 PUT_CHAR (string1
[this_char
]);
1207 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1209 PUT_CHAR (string2
[this_char
]);
1212 fputs ("...", stdout
);
1219 # ifndef DEFINED_ONCE
1228 # else /* not DEBUG */
1230 # ifndef DEFINED_ONCE
1234 # define DEBUG_STATEMENT(e)
1235 # define DEBUG_PRINT1(x)
1236 # define DEBUG_PRINT2(x1, x2)
1237 # define DEBUG_PRINT3(x1, x2, x3)
1238 # define DEBUG_PRINT4(x1, x2, x3, x4)
1239 # endif /* not DEFINED_ONCE */
1240 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1241 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1243 # endif /* not DEBUG */
1248 /* This convert a multibyte string to a wide character string.
1249 And write their correspondances to offset_buffer(see below)
1250 and write whether each wchar_t is binary data to is_binary.
1251 This assume invalid multibyte sequences as binary data.
1252 We assume offset_buffer and is_binary is already allocated
1255 static size_t convert_mbs_to_wcs (CHAR_T
*dest
, const unsigned char* src
,
1256 size_t len
, int *offset_buffer
,
1259 convert_mbs_to_wcs (dest
, src
, len
, offset_buffer
, is_binary
)
1261 const unsigned char* src
;
1262 size_t len
; /* the length of multibyte string. */
1264 /* It hold correspondances between src(char string) and
1265 dest(wchar_t string) for optimization.
1267 dest = {'X', 'Y', 'Z'}
1268 (each "xxx", "y" and "zz" represent one multibyte character
1269 corresponding to 'X', 'Y' and 'Z'.)
1270 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1276 wchar_t *pdest
= dest
;
1277 const unsigned char *psrc
= src
;
1278 size_t wc_count
= 0;
1282 size_t mb_remain
= len
;
1283 size_t mb_count
= 0;
1285 /* Initialize the conversion state. */
1286 memset (&mbs
, 0, sizeof (mbstate_t));
1288 offset_buffer
[0] = 0;
1289 for( ; mb_remain
> 0 ; ++wc_count
, ++pdest
, mb_remain
-= consumed
,
1293 consumed
= __mbrtowc (pdest
, psrc
, mb_remain
, &mbs
);
1295 consumed
= mbrtowc (pdest
, psrc
, mb_remain
, &mbs
);
1299 /* failed to convert. maybe src contains binary data.
1300 So we consume 1 byte manualy. */
1304 is_binary
[wc_count
] = TRUE
;
1307 is_binary
[wc_count
] = FALSE
;
1308 /* In sjis encoding, we use yen sign as escape character in
1309 place of reverse solidus. So we convert 0x5c(yen sign in
1310 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1311 solidus in UCS2). */
1312 if (consumed
== 1 && (int) *psrc
== 0x5c && (int) *pdest
== 0xa5)
1313 *pdest
= (wchar_t) *psrc
;
1315 offset_buffer
[wc_count
+ 1] = mb_count
+= consumed
;
1318 /* Fill remain of the buffer with sentinel. */
1319 for (i
= wc_count
+ 1 ; i
<= len
; i
++)
1320 offset_buffer
[i
] = mb_count
+ 1;
1327 #else /* not INSIDE_RECURSION */
1329 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1330 also be assigned to arbitrarily: each pattern buffer stores its own
1331 syntax, so it can be changed between regex compilations. */
1332 /* This has no initializer because initialized variables in Emacs
1333 become read-only after dumping. */
1334 reg_syntax_t re_syntax_options
;
1337 /* Specify the precise syntax of regexps for compilation. This provides
1338 for compatibility for various utilities which historically have
1339 different, incompatible syntaxes.
1341 The argument SYNTAX is a bit mask comprised of the various bits
1342 defined in regex.h. We return the old syntax. */
1345 re_set_syntax (syntax
)
1346 reg_syntax_t syntax
;
1348 reg_syntax_t ret
= re_syntax_options
;
1350 re_syntax_options
= syntax
;
1352 if (syntax
& RE_DEBUG
)
1354 else if (debug
) /* was on but now is not */
1360 weak_alias (__re_set_syntax
, re_set_syntax
)
1363 /* This table gives an error message for each of the error codes listed
1364 in regex.h. Obviously the order here has to be same as there.
1365 POSIX doesn't require that we do anything for REG_NOERROR,
1366 but why not be nice? */
1368 static const char re_error_msgid
[] =
1370 # define REG_NOERROR_IDX 0
1371 gettext_noop ("Success") /* REG_NOERROR */
1373 # define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1374 gettext_noop ("No match") /* REG_NOMATCH */
1376 # define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1377 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1379 # define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1380 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1382 # define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1383 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1385 # define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1386 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1388 # define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1389 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1391 # define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1392 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1394 # define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1395 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1397 # define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1398 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1400 # define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1401 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1403 # define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1404 gettext_noop ("Invalid range end") /* REG_ERANGE */
1406 # define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1407 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1409 # define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1410 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1412 # define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1413 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1415 # define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1416 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1418 # define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1419 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1422 static const size_t re_error_msgid_idx
[] =
1443 #endif /* INSIDE_RECURSION */
1445 #ifndef DEFINED_ONCE
1446 /* Avoiding alloca during matching, to placate r_alloc. */
1448 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1449 searching and matching functions should not call alloca. On some
1450 systems, alloca is implemented in terms of malloc, and if we're
1451 using the relocating allocator routines, then malloc could cause a
1452 relocation, which might (if the strings being searched are in the
1453 ralloc heap) shift the data out from underneath the regexp
1456 Here's another reason to avoid allocation: Emacs
1457 processes input from X in a signal handler; processing X input may
1458 call malloc; if input arrives while a matching routine is calling
1459 malloc, then we're scrod. But Emacs can't just block input while
1460 calling matching routines; then we don't notice interrupts when
1461 they come in. So, Emacs blocks input around all regexp calls
1462 except the matching calls, which it leaves unprotected, in the
1463 faith that they will not malloc. */
1465 /* Normally, this is fine. */
1466 # define MATCH_MAY_ALLOCATE
1468 /* When using GNU C, we are not REALLY using the C alloca, no matter
1469 what config.h may say. So don't take precautions for it. */
1474 /* The match routines may not allocate if (1) they would do it with malloc
1475 and (2) it's not safe for them to use malloc.
1476 Note that if REL_ALLOC is defined, matching would not use malloc for the
1477 failure stack, but we would still use it for the register vectors;
1478 so REL_ALLOC should not affect this. */
1479 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1480 # undef MATCH_MAY_ALLOCATE
1482 #endif /* not DEFINED_ONCE */
1484 #ifdef INSIDE_RECURSION
1485 /* Failure stack declarations and macros; both re_compile_fastmap and
1486 re_match_2 use a failure stack. These have to be macros because of
1487 REGEX_ALLOCATE_STACK. */
1490 /* Number of failure points for which to initially allocate space
1491 when matching. If this number is exceeded, we allocate more
1492 space, so it is not a hard limit. */
1493 # ifndef INIT_FAILURE_ALLOC
1494 # define INIT_FAILURE_ALLOC 5
1497 /* Roughly the maximum number of failure points on the stack. Would be
1498 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1499 This is a variable only so users of regex can assign to it; we never
1500 change it ourselves. */
1502 # ifdef INT_IS_16BIT
1504 # ifndef DEFINED_ONCE
1505 # if defined MATCH_MAY_ALLOCATE
1506 /* 4400 was enough to cause a crash on Alpha OSF/1,
1507 whose default stack limit is 2mb. */
1508 long int re_max_failures
= 4000;
1510 long int re_max_failures
= 2000;
1514 union PREFIX(fail_stack_elt
)
1520 typedef union PREFIX(fail_stack_elt
) PREFIX(fail_stack_elt_t
);
1524 PREFIX(fail_stack_elt_t
) *stack
;
1525 unsigned long int size
;
1526 unsigned long int avail
; /* Offset of next open position. */
1527 } PREFIX(fail_stack_type
);
1529 # else /* not INT_IS_16BIT */
1531 # ifndef DEFINED_ONCE
1532 # if defined MATCH_MAY_ALLOCATE
1533 /* 4400 was enough to cause a crash on Alpha OSF/1,
1534 whose default stack limit is 2mb. */
1535 int re_max_failures
= 4000;
1537 int re_max_failures
= 2000;
1541 union PREFIX(fail_stack_elt
)
1547 typedef union PREFIX(fail_stack_elt
) PREFIX(fail_stack_elt_t
);
1551 PREFIX(fail_stack_elt_t
) *stack
;
1553 unsigned avail
; /* Offset of next open position. */
1554 } PREFIX(fail_stack_type
);
1556 # endif /* INT_IS_16BIT */
1558 # ifndef DEFINED_ONCE
1559 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1560 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1561 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1565 /* Define macros to initialize and free the failure stack.
1566 Do `return -2' if the alloc fails. */
1568 # ifdef MATCH_MAY_ALLOCATE
1569 # define INIT_FAIL_STACK() \
1571 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1572 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1574 if (fail_stack.stack == NULL) \
1577 fail_stack.size = INIT_FAILURE_ALLOC; \
1578 fail_stack.avail = 0; \
1581 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1583 # define INIT_FAIL_STACK() \
1585 fail_stack.avail = 0; \
1588 # define RESET_FAIL_STACK()
1592 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1594 Return 1 if succeeds, and 0 if either ran out of memory
1595 allocating space for it or it was already too large.
1597 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1599 # define DOUBLE_FAIL_STACK(fail_stack) \
1600 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1602 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1603 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1604 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1605 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1607 (fail_stack).stack == NULL \
1609 : ((fail_stack).size <<= 1, \
1613 /* Push pointer POINTER on FAIL_STACK.
1614 Return 1 if was able to do so and 0 if ran out of memory allocating
1616 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1617 ((FAIL_STACK_FULL () \
1618 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1620 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1623 /* Push a pointer value onto the failure stack.
1624 Assumes the variable `fail_stack'. Probably should only
1625 be called from within `PUSH_FAILURE_POINT'. */
1626 # define PUSH_FAILURE_POINTER(item) \
1627 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1629 /* This pushes an integer-valued item onto the failure stack.
1630 Assumes the variable `fail_stack'. Probably should only
1631 be called from within `PUSH_FAILURE_POINT'. */
1632 # define PUSH_FAILURE_INT(item) \
1633 fail_stack.stack[fail_stack.avail++].integer = (item)
1635 /* Push a fail_stack_elt_t value onto the failure stack.
1636 Assumes the variable `fail_stack'. Probably should only
1637 be called from within `PUSH_FAILURE_POINT'. */
1638 # define PUSH_FAILURE_ELT(item) \
1639 fail_stack.stack[fail_stack.avail++] = (item)
1641 /* These three POP... operations complement the three PUSH... operations.
1642 All assume that `fail_stack' is nonempty. */
1643 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1644 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1645 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1647 /* Used to omit pushing failure point id's when we're not debugging. */
1649 # define DEBUG_PUSH PUSH_FAILURE_INT
1650 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1652 # define DEBUG_PUSH(item)
1653 # define DEBUG_POP(item_addr)
1657 /* Push the information about the state we will need
1658 if we ever fail back to it.
1660 Requires variables fail_stack, regstart, regend, reg_info, and
1661 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1664 Does `return FAILURE_CODE' if runs out of memory. */
1666 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1668 char *destination; \
1669 /* Must be int, so when we don't save any registers, the arithmetic \
1670 of 0 + -1 isn't done as unsigned. */ \
1671 /* Can't be int, since there is not a shred of a guarantee that int \
1672 is wide enough to hold a value of something to which pointer can \
1674 active_reg_t this_reg; \
1676 DEBUG_STATEMENT (failure_id++); \
1677 DEBUG_STATEMENT (nfailure_points_pushed++); \
1678 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1679 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1680 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1682 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1683 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1685 /* Ensure we have enough space allocated for what we will push. */ \
1686 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1688 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1689 return failure_code; \
1691 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1692 (fail_stack).size); \
1693 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1696 /* Push the info, starting with the registers. */ \
1697 DEBUG_PRINT1 ("\n"); \
1700 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1703 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1704 DEBUG_STATEMENT (num_regs_pushed++); \
1706 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1707 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1709 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1710 PUSH_FAILURE_POINTER (regend[this_reg]); \
1712 DEBUG_PRINT2 (" info: %p\n ", \
1713 reg_info[this_reg].word.pointer); \
1714 DEBUG_PRINT2 (" match_null=%d", \
1715 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1716 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1717 DEBUG_PRINT2 (" matched_something=%d", \
1718 MATCHED_SOMETHING (reg_info[this_reg])); \
1719 DEBUG_PRINT2 (" ever_matched=%d", \
1720 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1721 DEBUG_PRINT1 ("\n"); \
1722 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1725 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1726 PUSH_FAILURE_INT (lowest_active_reg); \
1728 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1729 PUSH_FAILURE_INT (highest_active_reg); \
1731 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1732 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1733 PUSH_FAILURE_POINTER (pattern_place); \
1735 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1736 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1738 DEBUG_PRINT1 ("'\n"); \
1739 PUSH_FAILURE_POINTER (string_place); \
1741 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1742 DEBUG_PUSH (failure_id); \
1745 # ifndef DEFINED_ONCE
1746 /* This is the number of items that are pushed and popped on the stack
1747 for each register. */
1748 # define NUM_REG_ITEMS 3
1750 /* Individual items aside from the registers. */
1752 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1754 # define NUM_NONREG_ITEMS 4
1757 /* We push at most this many items on the stack. */
1758 /* We used to use (num_regs - 1), which is the number of registers
1759 this regexp will save; but that was changed to 5
1760 to avoid stack overflow for a regexp with lots of parens. */
1761 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1763 /* We actually push this many items. */
1764 # define NUM_FAILURE_ITEMS \
1766 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1770 /* How many items can still be added to the stack without overflowing it. */
1771 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1772 # endif /* not DEFINED_ONCE */
1775 /* Pops what PUSH_FAIL_STACK pushes.
1777 We restore into the parameters, all of which should be lvalues:
1778 STR -- the saved data position.
1779 PAT -- the saved pattern position.
1780 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1781 REGSTART, REGEND -- arrays of string positions.
1782 REG_INFO -- array of information about each subexpression.
1784 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1785 `pend', `string1', `size1', `string2', and `size2'. */
1786 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1788 DEBUG_STATEMENT (unsigned failure_id;) \
1789 active_reg_t this_reg; \
1790 const UCHAR_T *string_temp; \
1792 assert (!FAIL_STACK_EMPTY ()); \
1794 /* Remove failure points and point to how many regs pushed. */ \
1795 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1796 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1797 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1799 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1801 DEBUG_POP (&failure_id); \
1802 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1804 /* If the saved string location is NULL, it came from an \
1805 on_failure_keep_string_jump opcode, and we want to throw away the \
1806 saved NULL, thus retaining our current position in the string. */ \
1807 string_temp = POP_FAILURE_POINTER (); \
1808 if (string_temp != NULL) \
1809 str = (const CHAR_T *) string_temp; \
1811 DEBUG_PRINT2 (" Popping string %p: `", str); \
1812 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1813 DEBUG_PRINT1 ("'\n"); \
1815 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1816 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1817 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1819 /* Restore register info. */ \
1820 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1821 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1823 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1824 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1827 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1829 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1831 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1832 DEBUG_PRINT2 (" info: %p\n", \
1833 reg_info[this_reg].word.pointer); \
1835 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1836 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1838 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1839 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1843 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1845 reg_info[this_reg].word.integer = 0; \
1846 regend[this_reg] = 0; \
1847 regstart[this_reg] = 0; \
1849 highest_active_reg = high_reg; \
1852 set_regs_matched_done = 0; \
1853 DEBUG_STATEMENT (nfailure_points_popped++); \
1854 } /* POP_FAILURE_POINT */
1856 /* Structure for per-register (a.k.a. per-group) information.
1857 Other register information, such as the
1858 starting and ending positions (which are addresses), and the list of
1859 inner groups (which is a bits list) are maintained in separate
1862 We are making a (strictly speaking) nonportable assumption here: that
1863 the compiler will pack our bit fields into something that fits into
1864 the type of `word', i.e., is something that fits into one item on the
1868 /* Declarations and macros for re_match_2. */
1872 PREFIX(fail_stack_elt_t
) word
;
1875 /* This field is one if this group can match the empty string,
1876 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1877 # define MATCH_NULL_UNSET_VALUE 3
1878 unsigned match_null_string_p
: 2;
1879 unsigned is_active
: 1;
1880 unsigned matched_something
: 1;
1881 unsigned ever_matched_something
: 1;
1883 } PREFIX(register_info_type
);
1885 # ifndef DEFINED_ONCE
1886 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1887 # define IS_ACTIVE(R) ((R).bits.is_active)
1888 # define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1889 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1892 /* Call this when have matched a real character; it sets `matched' flags
1893 for the subexpressions which we are currently inside. Also records
1894 that those subexprs have matched. */
1895 # define SET_REGS_MATCHED() \
1898 if (!set_regs_matched_done) \
1901 set_regs_matched_done = 1; \
1902 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1904 MATCHED_SOMETHING (reg_info[r]) \
1905 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1911 # endif /* not DEFINED_ONCE */
1913 /* Registers are set to a sentinel when they haven't yet matched. */
1914 static CHAR_T
PREFIX(reg_unset_dummy
);
1915 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1916 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1918 /* Subroutine declarations and macros for regex_compile. */
1919 static void PREFIX(store_op1
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
, int arg
));
1920 static void PREFIX(store_op2
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
,
1921 int arg1
, int arg2
));
1922 static void PREFIX(insert_op1
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
,
1923 int arg
, UCHAR_T
*end
));
1924 static void PREFIX(insert_op2
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
,
1925 int arg1
, int arg2
, UCHAR_T
*end
));
1926 static boolean
PREFIX(at_begline_loc_p
) _RE_ARGS ((const CHAR_T
*pattern
,
1928 reg_syntax_t syntax
));
1929 static boolean
PREFIX(at_endline_loc_p
) _RE_ARGS ((const CHAR_T
*p
,
1931 reg_syntax_t syntax
));
1933 static reg_errcode_t wcs_compile_range
_RE_ARGS ((CHAR_T range_start
,
1934 const CHAR_T
**p_ptr
,
1937 reg_syntax_t syntax
,
1940 static void insert_space
_RE_ARGS ((int num
, CHAR_T
*loc
, CHAR_T
*end
));
1942 static reg_errcode_t byte_compile_range
_RE_ARGS ((unsigned int range_start
,
1946 reg_syntax_t syntax
,
1950 /* Fetch the next character in the uncompiled pattern---translating it
1951 if necessary. Also cast from a signed character in the constant
1952 string passed to us by the user to an unsigned char that we can use
1953 as an array index (in, e.g., `translate'). */
1954 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1955 because it is impossible to allocate 4GB array for some encodings
1956 which have 4 byte character_set like UCS4. */
1959 # define PATFETCH(c) \
1960 do {if (p == pend) return REG_EEND; \
1961 c = (UCHAR_T) *p++; \
1962 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1965 # define PATFETCH(c) \
1966 do {if (p == pend) return REG_EEND; \
1967 c = (unsigned char) *p++; \
1968 if (translate) c = (unsigned char) translate[c]; \
1973 /* Fetch the next character in the uncompiled pattern, with no
1975 # define PATFETCH_RAW(c) \
1976 do {if (p == pend) return REG_EEND; \
1977 c = (UCHAR_T) *p++; \
1980 /* Go backwards one character in the pattern. */
1981 # define PATUNFETCH p--
1984 /* If `translate' is non-null, return translate[D], else just D. We
1985 cast the subscript to translate because some data is declared as
1986 `char *', to avoid warnings when a string constant is passed. But
1987 when we use a character as a subscript we must make it unsigned. */
1988 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1989 because it is impossible to allocate 4GB array for some encodings
1990 which have 4 byte character_set like UCS4. */
1994 # define TRANSLATE(d) \
1995 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1996 ? (char) translate[(unsigned char) (d)] : (d))
1998 # define TRANSLATE(d) \
1999 (translate ? (char) translate[(unsigned char) (d)] : (d))
2004 /* Macros for outputting the compiled pattern into `buffer'. */
2006 /* If the buffer isn't allocated when it comes in, use this. */
2007 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
2009 /* Make sure we have at least N more bytes of space in buffer. */
2011 # define GET_BUFFER_SPACE(n) \
2012 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
2013 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
2016 # define GET_BUFFER_SPACE(n) \
2017 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
2021 /* Make sure we have one more byte of buffer space and then add C to it. */
2022 # define BUF_PUSH(c) \
2024 GET_BUFFER_SPACE (1); \
2025 *b++ = (UCHAR_T) (c); \
2029 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
2030 # define BUF_PUSH_2(c1, c2) \
2032 GET_BUFFER_SPACE (2); \
2033 *b++ = (UCHAR_T) (c1); \
2034 *b++ = (UCHAR_T) (c2); \
2038 /* As with BUF_PUSH_2, except for three bytes. */
2039 # define BUF_PUSH_3(c1, c2, c3) \
2041 GET_BUFFER_SPACE (3); \
2042 *b++ = (UCHAR_T) (c1); \
2043 *b++ = (UCHAR_T) (c2); \
2044 *b++ = (UCHAR_T) (c3); \
2047 /* Store a jump with opcode OP at LOC to location TO. We store a
2048 relative address offset by the three bytes the jump itself occupies. */
2049 # define STORE_JUMP(op, loc, to) \
2050 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
2052 /* Likewise, for a two-argument jump. */
2053 # define STORE_JUMP2(op, loc, to, arg) \
2054 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
2056 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
2057 # define INSERT_JUMP(op, loc, to) \
2058 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
2060 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
2061 # define INSERT_JUMP2(op, loc, to, arg) \
2062 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
2065 /* This is not an arbitrary limit: the arguments which represent offsets
2066 into the pattern are two bytes long. So if 2^16 bytes turns out to
2067 be too small, many things would have to change. */
2068 /* Any other compiler which, like MSC, has allocation limit below 2^16
2069 bytes will have to use approach similar to what was done below for
2070 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2071 reallocating to 0 bytes. Such thing is not going to work too well.
2072 You have been warned!! */
2073 # ifndef DEFINED_ONCE
2074 # if defined _MSC_VER && !defined WIN32
2075 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2076 The REALLOC define eliminates a flurry of conversion warnings,
2077 but is not required. */
2078 # define MAX_BUF_SIZE 65500L
2079 # define REALLOC(p,s) realloc ((p), (size_t) (s))
2081 # define MAX_BUF_SIZE (1L << 16)
2082 # define REALLOC(p,s) realloc ((p), (s))
2085 /* Extend the buffer by twice its current size via realloc and
2086 reset the pointers that pointed into the old block to point to the
2087 correct places in the new one. If extending the buffer results in it
2088 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2089 # if defined __BOUNDED_POINTERS__ && __BOUNDED_POINTERS__
2090 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2091 # define MOVE_BUFFER_POINTER(P) \
2092 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2093 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2096 SET_HIGH_BOUND (b); \
2097 SET_HIGH_BOUND (begalt); \
2098 if (fixup_alt_jump) \
2099 SET_HIGH_BOUND (fixup_alt_jump); \
2101 SET_HIGH_BOUND (laststart); \
2102 if (pending_exact) \
2103 SET_HIGH_BOUND (pending_exact); \
2106 # define MOVE_BUFFER_POINTER(P) (P) += incr
2107 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
2109 # endif /* not DEFINED_ONCE */
2112 # define EXTEND_BUFFER() \
2114 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2116 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2118 bufp->allocated <<= 1; \
2119 if (bufp->allocated > MAX_BUF_SIZE) \
2120 bufp->allocated = MAX_BUF_SIZE; \
2121 /* How many characters the new buffer can have? */ \
2122 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2123 if (wchar_count == 0) wchar_count = 1; \
2124 /* Truncate the buffer to CHAR_T align. */ \
2125 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2126 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2127 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2128 if (COMPILED_BUFFER_VAR == NULL) \
2129 return REG_ESPACE; \
2130 /* If the buffer moved, move all the pointers into it. */ \
2131 if (old_buffer != COMPILED_BUFFER_VAR) \
2133 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2134 MOVE_BUFFER_POINTER (b); \
2135 MOVE_BUFFER_POINTER (begalt); \
2136 if (fixup_alt_jump) \
2137 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2139 MOVE_BUFFER_POINTER (laststart); \
2140 if (pending_exact) \
2141 MOVE_BUFFER_POINTER (pending_exact); \
2143 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2146 # define EXTEND_BUFFER() \
2148 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2149 if (bufp->allocated == MAX_BUF_SIZE) \
2151 bufp->allocated <<= 1; \
2152 if (bufp->allocated > MAX_BUF_SIZE) \
2153 bufp->allocated = MAX_BUF_SIZE; \
2154 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2156 if (COMPILED_BUFFER_VAR == NULL) \
2157 return REG_ESPACE; \
2158 /* If the buffer moved, move all the pointers into it. */ \
2159 if (old_buffer != COMPILED_BUFFER_VAR) \
2161 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2162 MOVE_BUFFER_POINTER (b); \
2163 MOVE_BUFFER_POINTER (begalt); \
2164 if (fixup_alt_jump) \
2165 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2167 MOVE_BUFFER_POINTER (laststart); \
2168 if (pending_exact) \
2169 MOVE_BUFFER_POINTER (pending_exact); \
2171 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2175 # ifndef DEFINED_ONCE
2176 /* Since we have one byte reserved for the register number argument to
2177 {start,stop}_memory, the maximum number of groups we can report
2178 things about is what fits in that byte. */
2179 # define MAX_REGNUM 255
2181 /* But patterns can have more than `MAX_REGNUM' registers. We just
2182 ignore the excess. */
2183 typedef unsigned regnum_t
;
2186 /* Macros for the compile stack. */
2188 /* Since offsets can go either forwards or backwards, this type needs to
2189 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2190 /* int may be not enough when sizeof(int) == 2. */
2191 typedef long pattern_offset_t
;
2195 pattern_offset_t begalt_offset
;
2196 pattern_offset_t fixup_alt_jump
;
2197 pattern_offset_t inner_group_offset
;
2198 pattern_offset_t laststart_offset
;
2200 } compile_stack_elt_t
;
2205 compile_stack_elt_t
*stack
;
2207 unsigned avail
; /* Offset of next open position. */
2208 } compile_stack_type
;
2211 # define INIT_COMPILE_STACK_SIZE 32
2213 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2214 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2216 /* The next available element. */
2217 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2219 # endif /* not DEFINED_ONCE */
2221 /* Set the bit for character C in a list. */
2222 # ifndef DEFINED_ONCE
2223 # define SET_LIST_BIT(c) \
2224 (b[((unsigned char) (c)) / BYTEWIDTH] \
2225 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2226 # endif /* DEFINED_ONCE */
2228 /* Get the next unsigned number in the uncompiled pattern. */
2229 # define GET_UNSIGNED_NUMBER(num) \
2234 if (c < '0' || c > '9') \
2236 if (num <= RE_DUP_MAX) \
2240 num = num * 10 + c - '0'; \
2245 # ifndef DEFINED_ONCE
2246 # if defined _LIBC || (defined WIDE_CHAR_SUPPORT && WIDE_CHAR_SUPPORT)
2247 /* The GNU C library provides support for user-defined character classes
2248 and the functions from ISO C amendement 1. */
2249 # ifdef CHARCLASS_NAME_MAX
2250 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2252 /* This shouldn't happen but some implementation might still have this
2253 problem. Use a reasonable default value. */
2254 # define CHAR_CLASS_MAX_LENGTH 256
2258 # define IS_CHAR_CLASS(string) __wctype (string)
2260 # define IS_CHAR_CLASS(string) wctype (string)
2263 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2265 # define IS_CHAR_CLASS(string) \
2266 (STREQ (string, "alpha") || STREQ (string, "upper") \
2267 || STREQ (string, "lower") || STREQ (string, "digit") \
2268 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2269 || STREQ (string, "space") || STREQ (string, "print") \
2270 || STREQ (string, "punct") || STREQ (string, "graph") \
2271 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2273 # endif /* DEFINED_ONCE */
2275 # ifndef MATCH_MAY_ALLOCATE
2277 /* If we cannot allocate large objects within re_match_2_internal,
2278 we make the fail stack and register vectors global.
2279 The fail stack, we grow to the maximum size when a regexp
2281 The register vectors, we adjust in size each time we
2282 compile a regexp, according to the number of registers it needs. */
2284 static PREFIX(fail_stack_type
) fail_stack
;
2286 /* Size with which the following vectors are currently allocated.
2287 That is so we can make them bigger as needed,
2288 but never make them smaller. */
2289 # ifdef DEFINED_ONCE
2290 static int regs_allocated_size
;
2292 static const char ** regstart
, ** regend
;
2293 static const char ** old_regstart
, ** old_regend
;
2294 static const char **best_regstart
, **best_regend
;
2295 static const char **reg_dummy
;
2296 # endif /* DEFINED_ONCE */
2298 static PREFIX(register_info_type
) *PREFIX(reg_info
);
2299 static PREFIX(register_info_type
) *PREFIX(reg_info_dummy
);
2301 /* Make the register vectors big enough for NUM_REGS registers,
2302 but don't make them smaller. */
2305 PREFIX(regex_grow_registers
) (num_regs
)
2308 if (num_regs
> regs_allocated_size
)
2310 RETALLOC_IF (regstart
, num_regs
, const char *);
2311 RETALLOC_IF (regend
, num_regs
, const char *);
2312 RETALLOC_IF (old_regstart
, num_regs
, const char *);
2313 RETALLOC_IF (old_regend
, num_regs
, const char *);
2314 RETALLOC_IF (best_regstart
, num_regs
, const char *);
2315 RETALLOC_IF (best_regend
, num_regs
, const char *);
2316 RETALLOC_IF (PREFIX(reg_info
), num_regs
, PREFIX(register_info_type
));
2317 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
2318 RETALLOC_IF (PREFIX(reg_info_dummy
), num_regs
, PREFIX(register_info_type
));
2320 regs_allocated_size
= num_regs
;
2324 # endif /* not MATCH_MAY_ALLOCATE */
2326 # ifndef DEFINED_ONCE
2327 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2330 # endif /* not DEFINED_ONCE */
2332 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2333 Returns one of error codes defined in `regex.h', or zero for success.
2335 Assumes the `allocated' (and perhaps `buffer') and `translate'
2336 fields are set in BUFP on entry.
2338 If it succeeds, results are put in BUFP (if it returns an error, the
2339 contents of BUFP are undefined):
2340 `buffer' is the compiled pattern;
2341 `syntax' is set to SYNTAX;
2342 `used' is set to the length of the compiled pattern;
2343 `fastmap_accurate' is zero;
2344 `re_nsub' is the number of subexpressions in PATTERN;
2345 `not_bol' and `not_eol' are zero;
2347 The `fastmap' and `newline_anchor' fields are neither
2348 examined nor set. */
2350 /* Return, freeing storage we allocated. */
2352 # define FREE_STACK_RETURN(value) \
2353 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2355 # define FREE_STACK_RETURN(value) \
2356 return (free (compile_stack.stack), value)
2359 static reg_errcode_t
2360 PREFIX(regex_compile
) (ARG_PREFIX(pattern
), ARG_PREFIX(size
), syntax
, bufp
)
2361 const char *ARG_PREFIX(pattern
);
2362 size_t ARG_PREFIX(size
);
2363 reg_syntax_t syntax
;
2364 struct re_pattern_buffer
*bufp
;
2366 /* We fetch characters from PATTERN here. Even though PATTERN is
2367 `char *' (i.e., signed), we declare these variables as unsigned, so
2368 they can be reliably used as array indices. */
2369 register UCHAR_T c
, c1
;
2372 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2373 CHAR_T
*pattern
, *COMPILED_BUFFER_VAR
;
2375 /* offset buffer for optimization. See convert_mbs_to_wc. */
2376 int *mbs_offset
= NULL
;
2377 /* It hold whether each wchar_t is binary data or not. */
2378 char *is_binary
= NULL
;
2379 /* A flag whether exactn is handling binary data or not. */
2380 char is_exactn_bin
= FALSE
;
2383 /* A random temporary spot in PATTERN. */
2386 /* Points to the end of the buffer, where we should append. */
2387 register UCHAR_T
*b
;
2389 /* Keeps track of unclosed groups. */
2390 compile_stack_type compile_stack
;
2392 /* Points to the current (ending) position in the pattern. */
2397 const CHAR_T
*p
= pattern
;
2398 const CHAR_T
*pend
= pattern
+ size
;
2401 /* How to translate the characters in the pattern. */
2402 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2404 /* Address of the count-byte of the most recently inserted `exactn'
2405 command. This makes it possible to tell if a new exact-match
2406 character can be added to that command or if the character requires
2407 a new `exactn' command. */
2408 UCHAR_T
*pending_exact
= 0;
2410 /* Address of start of the most recently finished expression.
2411 This tells, e.g., postfix * where to find the start of its
2412 operand. Reset at the beginning of groups and alternatives. */
2413 UCHAR_T
*laststart
= 0;
2415 /* Address of beginning of regexp, or inside of last group. */
2418 /* Address of the place where a forward jump should go to the end of
2419 the containing expression. Each alternative of an `or' -- except the
2420 last -- ends with a forward jump of this sort. */
2421 UCHAR_T
*fixup_alt_jump
= 0;
2423 /* Counts open-groups as they are encountered. Remembered for the
2424 matching close-group on the compile stack, so the same register
2425 number is put in the stop_memory as the start_memory. */
2426 regnum_t regnum
= 0;
2429 /* Initialize the wchar_t PATTERN and offset_buffer. */
2430 p
= pend
= pattern
= TALLOC(csize
+ 1, CHAR_T
);
2431 mbs_offset
= TALLOC(csize
+ 1, int);
2432 is_binary
= TALLOC(csize
+ 1, char);
2433 if (pattern
== NULL
|| mbs_offset
== NULL
|| is_binary
== NULL
)
2440 pattern
[csize
] = L
'\0'; /* sentinel */
2441 size
= convert_mbs_to_wcs(pattern
, cpattern
, csize
, mbs_offset
, is_binary
);
2453 DEBUG_PRINT1 ("\nCompiling pattern: ");
2456 unsigned debug_count
;
2458 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2459 PUT_CHAR (pattern
[debug_count
]);
2464 /* Initialize the compile stack. */
2465 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2466 if (compile_stack
.stack
== NULL
)
2476 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2477 compile_stack
.avail
= 0;
2479 /* Initialize the pattern buffer. */
2480 bufp
->syntax
= syntax
;
2481 bufp
->fastmap_accurate
= 0;
2482 bufp
->not_bol
= bufp
->not_eol
= 0;
2484 /* Set `used' to zero, so that if we return an error, the pattern
2485 printer (for debugging) will think there's no pattern. We reset it
2489 /* Always count groups, whether or not bufp->no_sub is set. */
2492 #if !defined emacs && !defined SYNTAX_TABLE
2493 /* Initialize the syntax table. */
2494 init_syntax_once ();
2497 if (bufp
->allocated
== 0)
2500 { /* If zero allocated, but buffer is non-null, try to realloc
2501 enough space. This loses if buffer's address is bogus, but
2502 that is the user's responsibility. */
2504 /* Free bufp->buffer and allocate an array for wchar_t pattern
2507 COMPILED_BUFFER_VAR
= TALLOC (INIT_BUF_SIZE
/sizeof(UCHAR_T
),
2510 RETALLOC (COMPILED_BUFFER_VAR
, INIT_BUF_SIZE
, UCHAR_T
);
2514 { /* Caller did not allocate a buffer. Do it for them. */
2515 COMPILED_BUFFER_VAR
= TALLOC (INIT_BUF_SIZE
/ sizeof(UCHAR_T
),
2519 if (!COMPILED_BUFFER_VAR
) FREE_STACK_RETURN (REG_ESPACE
);
2521 bufp
->buffer
= (char*)COMPILED_BUFFER_VAR
;
2523 bufp
->allocated
= INIT_BUF_SIZE
;
2527 COMPILED_BUFFER_VAR
= (UCHAR_T
*) bufp
->buffer
;
2530 begalt
= b
= COMPILED_BUFFER_VAR
;
2532 /* Loop through the uncompiled pattern until we're at the end. */
2541 if ( /* If at start of pattern, it's an operator. */
2543 /* If context independent, it's an operator. */
2544 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2545 /* Otherwise, depends on what's come before. */
2546 || PREFIX(at_begline_loc_p
) (pattern
, p
, syntax
))
2556 if ( /* If at end of pattern, it's an operator. */
2558 /* If context independent, it's an operator. */
2559 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2560 /* Otherwise, depends on what's next. */
2561 || PREFIX(at_endline_loc_p
) (p
, pend
, syntax
))
2571 if ((syntax
& RE_BK_PLUS_QM
)
2572 || (syntax
& RE_LIMITED_OPS
))
2576 /* If there is no previous pattern... */
2579 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2580 FREE_STACK_RETURN (REG_BADRPT
);
2581 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2586 /* Are we optimizing this jump? */
2587 boolean keep_string_p
= false;
2589 /* 1 means zero (many) matches is allowed. */
2590 char zero_times_ok
= 0, many_times_ok
= 0;
2592 /* If there is a sequence of repetition chars, collapse it
2593 down to just one (the right one). We can't combine
2594 interval operators with these because of, e.g., `a{2}*',
2595 which should only match an even number of `a's. */
2599 zero_times_ok
|= c
!= '+';
2600 many_times_ok
|= c
!= '?';
2608 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
2611 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
2613 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2616 if (!(c1
== '+' || c1
== '?'))
2631 /* If we get here, we found another repeat character. */
2634 /* Star, etc. applied to an empty pattern is equivalent
2635 to an empty pattern. */
2639 /* Now we know whether or not zero matches is allowed
2640 and also whether or not two or more matches is allowed. */
2642 { /* More than one repetition is allowed, so put in at the
2643 end a backward relative jump from `b' to before the next
2644 jump we're going to put in below (which jumps from
2645 laststart to after this jump).
2647 But if we are at the `*' in the exact sequence `.*\n',
2648 insert an unconditional jump backwards to the .,
2649 instead of the beginning of the loop. This way we only
2650 push a failure point once, instead of every time
2651 through the loop. */
2652 assert (p
- 1 > pattern
);
2654 /* Allocate the space for the jump. */
2655 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2657 /* We know we are not at the first character of the pattern,
2658 because laststart was nonzero. And we've already
2659 incremented `p', by the way, to be the character after
2660 the `*'. Do we have to do something analogous here
2661 for null bytes, because of RE_DOT_NOT_NULL? */
2662 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2664 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2665 && !(syntax
& RE_DOT_NEWLINE
))
2666 { /* We have .*\n. */
2667 STORE_JUMP (jump
, b
, laststart
);
2668 keep_string_p
= true;
2671 /* Anything else. */
2672 STORE_JUMP (maybe_pop_jump
, b
, laststart
-
2673 (1 + OFFSET_ADDRESS_SIZE
));
2675 /* We've added more stuff to the buffer. */
2676 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2679 /* On failure, jump from laststart to b + 3, which will be the
2680 end of the buffer after this jump is inserted. */
2681 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2683 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2684 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2686 laststart
, b
+ 1 + OFFSET_ADDRESS_SIZE
);
2688 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2692 /* At least one repetition is required, so insert a
2693 `dummy_failure_jump' before the initial
2694 `on_failure_jump' instruction of the loop. This
2695 effects a skip over that instruction the first time
2696 we hit that loop. */
2697 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2698 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+
2699 2 + 2 * OFFSET_ADDRESS_SIZE
);
2700 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2714 boolean had_char_class
= false;
2716 CHAR_T range_start
= 0xffffffff;
2718 unsigned int range_start
= 0xffffffff;
2720 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2723 /* We assume a charset(_not) structure as a wchar_t array.
2724 charset[0] = (re_opcode_t) charset(_not)
2725 charset[1] = l (= length of char_classes)
2726 charset[2] = m (= length of collating_symbols)
2727 charset[3] = n (= length of equivalence_classes)
2728 charset[4] = o (= length of char_ranges)
2729 charset[5] = p (= length of chars)
2731 charset[6] = char_class (wctype_t)
2732 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2734 charset[l+5] = char_class (wctype_t)
2736 charset[l+6] = collating_symbol (wchar_t)
2738 charset[l+m+5] = collating_symbol (wchar_t)
2739 ifdef _LIBC we use the index if
2740 _NL_COLLATE_SYMB_EXTRAMB instead of
2743 charset[l+m+6] = equivalence_classes (wchar_t)
2745 charset[l+m+n+5] = equivalence_classes (wchar_t)
2746 ifdef _LIBC we use the index in
2747 _NL_COLLATE_WEIGHT instead of
2750 charset[l+m+n+6] = range_start
2751 charset[l+m+n+7] = range_end
2753 charset[l+m+n+2o+4] = range_start
2754 charset[l+m+n+2o+5] = range_end
2755 ifdef _LIBC we use the value looked up
2756 in _NL_COLLATE_COLLSEQ instead of
2759 charset[l+m+n+2o+6] = char
2761 charset[l+m+n+2o+p+5] = char
2765 /* We need at least 6 spaces: the opcode, the length of
2766 char_classes, the length of collating_symbols, the length of
2767 equivalence_classes, the length of char_ranges, the length of
2769 GET_BUFFER_SPACE (6);
2771 /* Save b as laststart. And We use laststart as the pointer
2772 to the first element of the charset here.
2773 In other words, laststart[i] indicates charset[i]. */
2776 /* We test `*p == '^' twice, instead of using an if
2777 statement, so we only need one BUF_PUSH. */
2778 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2782 /* Push the length of char_classes, the length of
2783 collating_symbols, the length of equivalence_classes, the
2784 length of char_ranges and the length of chars. */
2785 BUF_PUSH_3 (0, 0, 0);
2788 /* Remember the first position in the bracket expression. */
2791 /* charset_not matches newline according to a syntax bit. */
2792 if ((re_opcode_t
) b
[-6] == charset_not
2793 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2796 laststart
[5]++; /* Update the length of characters */
2799 /* Read in characters and ranges, setting map bits. */
2802 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2806 /* \ might escape characters inside [...] and [^...]. */
2807 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2809 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2813 laststart
[5]++; /* Update the length of chars */
2818 /* Could be the end of the bracket expression. If it's
2819 not (i.e., when the bracket expression is `[]' so
2820 far), the ']' character bit gets set way below. */
2821 if (c
== ']' && p
!= p1
+ 1)
2824 /* Look ahead to see if it's a range when the last thing
2825 was a character class. */
2826 if (had_char_class
&& c
== '-' && *p
!= ']')
2827 FREE_STACK_RETURN (REG_ERANGE
);
2829 /* Look ahead to see if it's a range when the last thing
2830 was a character: if this is a hyphen not at the
2831 beginning or the end of a list, then it's the range
2834 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2835 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2839 /* Allocate the space for range_start and range_end. */
2840 GET_BUFFER_SPACE (2);
2841 /* Update the pointer to indicate end of buffer. */
2843 ret
= wcs_compile_range (range_start
, &p
, pend
, translate
,
2844 syntax
, b
, laststart
);
2845 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2846 range_start
= 0xffffffff;
2848 else if (p
[0] == '-' && p
[1] != ']')
2849 { /* This handles ranges made up of characters only. */
2852 /* Move past the `-'. */
2854 /* Allocate the space for range_start and range_end. */
2855 GET_BUFFER_SPACE (2);
2856 /* Update the pointer to indicate end of buffer. */
2858 ret
= wcs_compile_range (c
, &p
, pend
, translate
, syntax
, b
,
2860 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2861 range_start
= 0xffffffff;
2864 /* See if we're at the beginning of a possible character
2866 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2867 { /* Leave room for the null. */
2868 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2873 /* If pattern is `[[:'. */
2874 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2879 if ((c
== ':' && *p
== ']') || p
== pend
)
2881 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2884 /* This is in any case an invalid class name. */
2889 /* If isn't a word bracketed by `[:' and `:]':
2890 undo the ending character, the letters, and leave
2891 the leading `:' and `[' (but store them as character). */
2892 if (c
== ':' && *p
== ']')
2897 /* Query the character class as wctype_t. */
2898 wt
= IS_CHAR_CLASS (str
);
2900 FREE_STACK_RETURN (REG_ECTYPE
);
2902 /* Throw away the ] at the end of the character
2906 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2908 /* Allocate the space for character class. */
2909 GET_BUFFER_SPACE(CHAR_CLASS_SIZE
);
2910 /* Update the pointer to indicate end of buffer. */
2911 b
+= CHAR_CLASS_SIZE
;
2912 /* Move data which follow character classes
2913 not to violate the data. */
2914 insert_space(CHAR_CLASS_SIZE
,
2915 laststart
+ 6 + laststart
[1],
2917 alignedp
= ((uintptr_t)(laststart
+ 6 + laststart
[1])
2918 + __alignof__(wctype_t) - 1)
2919 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2920 /* Store the character class. */
2921 *((wctype_t*)alignedp
) = wt
;
2922 /* Update length of char_classes */
2923 laststart
[1] += CHAR_CLASS_SIZE
;
2925 had_char_class
= true;
2934 laststart
[5] += 2; /* Update the length of characters */
2936 had_char_class
= false;
2939 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && (*p
== '='
2942 CHAR_T str
[128]; /* Should be large enough. */
2943 CHAR_T delim
= *p
; /* '=' or '.' */
2946 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
2951 /* If pattern is `[[=' or '[[.'. */
2952 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2957 if ((c
== delim
&& *p
== ']') || p
== pend
)
2959 if (c1
< sizeof (str
) - 1)
2962 /* This is in any case an invalid class name. */
2967 if (c
== delim
&& *p
== ']' && str
[0] != '\0')
2969 unsigned int i
, offset
;
2970 /* If we have no collation data we use the default
2971 collation in which each character is in a class
2972 by itself. It also means that ASCII is the
2973 character set and therefore we cannot have character
2974 with more than one byte in the multibyte
2977 /* If not defined _LIBC, we push the name and
2978 `\0' for the sake of matching performance. */
2979 int datasize
= c1
+ 1;
2987 FREE_STACK_RETURN (REG_ECOLLATE
);
2992 const int32_t *table
;
2993 const int32_t *weights
;
2994 const int32_t *extra
;
2995 const int32_t *indirect
;
2998 /* This #include defines a local function! */
2999 # include <locale/weightwc.h>
3003 /* We push the index for equivalence class. */
3006 table
= (const int32_t *)
3007 _NL_CURRENT (LC_COLLATE
,
3008 _NL_COLLATE_TABLEWC
);
3009 weights
= (const int32_t *)
3010 _NL_CURRENT (LC_COLLATE
,
3011 _NL_COLLATE_WEIGHTWC
);
3012 extra
= (const int32_t *)
3013 _NL_CURRENT (LC_COLLATE
,
3014 _NL_COLLATE_EXTRAWC
);
3015 indirect
= (const int32_t *)
3016 _NL_CURRENT (LC_COLLATE
,
3017 _NL_COLLATE_INDIRECTWC
);
3019 idx
= findidx ((const wint_t**)&cp
);
3020 if (idx
== 0 || cp
< (wint_t*) str
+ c1
)
3021 /* This is no valid character. */
3022 FREE_STACK_RETURN (REG_ECOLLATE
);
3024 str
[0] = (wchar_t)idx
;
3026 else /* delim == '.' */
3028 /* We push collation sequence value
3029 for collating symbol. */
3031 const int32_t *symb_table
;
3032 const unsigned char *extra
;
3039 /* We have to convert the name to a single-byte
3040 string. This is possible since the names
3041 consist of ASCII characters and the internal
3042 representation is UCS4. */
3043 for (i
= 0; i
< c1
; ++i
)
3044 char_str
[i
] = str
[i
];
3047 _NL_CURRENT_WORD (LC_COLLATE
,
3048 _NL_COLLATE_SYMB_HASH_SIZEMB
);
3049 symb_table
= (const int32_t *)
3050 _NL_CURRENT (LC_COLLATE
,
3051 _NL_COLLATE_SYMB_TABLEMB
);
3052 extra
= (const unsigned char *)
3053 _NL_CURRENT (LC_COLLATE
,
3054 _NL_COLLATE_SYMB_EXTRAMB
);
3056 /* Locate the character in the hashing table. */
3057 hash
= elem_hash (char_str
, c1
);
3060 elem
= hash
% table_size
;
3061 second
= hash
% (table_size
- 2);
3062 while (symb_table
[2 * elem
] != 0)
3064 /* First compare the hashing value. */
3065 if (symb_table
[2 * elem
] == hash
3066 && c1
== extra
[symb_table
[2 * elem
+ 1]]
3067 && memcmp (char_str
,
3068 &extra
[symb_table
[2 * elem
+ 1]
3071 /* Yep, this is the entry. */
3072 idx
= symb_table
[2 * elem
+ 1];
3073 idx
+= 1 + extra
[idx
];
3081 if (symb_table
[2 * elem
] != 0)
3083 /* Compute the index of the byte sequence
3085 idx
+= 1 + extra
[idx
];
3086 /* Adjust for the alignment. */
3087 idx
= (idx
+ 3) & ~3;
3089 str
[0] = (wchar_t) idx
+ 4;
3091 else if (symb_table
[2 * elem
] == 0 && c1
== 1)
3093 /* No valid character. Match it as a
3094 single byte character. */
3095 had_char_class
= false;
3097 /* Update the length of characters */
3099 range_start
= str
[0];
3101 /* Throw away the ] at the end of the
3102 collating symbol. */
3104 /* exit from the switch block. */
3108 FREE_STACK_RETURN (REG_ECOLLATE
);
3113 /* Throw away the ] at the end of the equivalence
3114 class (or collating symbol). */
3117 /* Allocate the space for the equivalence class
3118 (or collating symbol) (and '\0' if needed). */
3119 GET_BUFFER_SPACE(datasize
);
3120 /* Update the pointer to indicate end of buffer. */
3124 { /* equivalence class */
3125 /* Calculate the offset of char_ranges,
3126 which is next to equivalence_classes. */
3127 offset
= laststart
[1] + laststart
[2]
3130 insert_space(datasize
, laststart
+ offset
, b
- 1);
3132 /* Write the equivalence_class and \0. */
3133 for (i
= 0 ; i
< datasize
; i
++)
3134 laststart
[offset
+ i
] = str
[i
];
3136 /* Update the length of equivalence_classes. */
3137 laststart
[3] += datasize
;
3138 had_char_class
= true;
3140 else /* delim == '.' */
3141 { /* collating symbol */
3142 /* Calculate the offset of the equivalence_classes,
3143 which is next to collating_symbols. */
3144 offset
= laststart
[1] + laststart
[2] + 6;
3145 /* Insert space and write the collationg_symbol
3147 insert_space(datasize
, laststart
+ offset
, b
-1);
3148 for (i
= 0 ; i
< datasize
; i
++)
3149 laststart
[offset
+ i
] = str
[i
];
3151 /* In re_match_2_internal if range_start < -1, we
3152 assume -range_start is the offset of the
3153 collating symbol which is specified as
3154 the character of the range start. So we assign
3155 -(laststart[1] + laststart[2] + 6) to
3157 range_start
= -(laststart
[1] + laststart
[2] + 6);
3158 /* Update the length of collating_symbol. */
3159 laststart
[2] += datasize
;
3160 had_char_class
= false;
3170 laststart
[5] += 2; /* Update the length of characters */
3171 range_start
= delim
;
3172 had_char_class
= false;
3177 had_char_class
= false;
3179 laststart
[5]++; /* Update the length of characters */
3185 /* Ensure that we have enough space to push a charset: the
3186 opcode, the length count, and the bitset; 34 bytes in all. */
3187 GET_BUFFER_SPACE (34);
3191 /* We test `*p == '^' twice, instead of using an if
3192 statement, so we only need one BUF_PUSH. */
3193 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
3197 /* Remember the first position in the bracket expression. */
3200 /* Push the number of bytes in the bitmap. */
3201 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
3203 /* Clear the whole map. */
3204 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
3206 /* charset_not matches newline according to a syntax bit. */
3207 if ((re_opcode_t
) b
[-2] == charset_not
3208 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
3209 SET_LIST_BIT ('\n');
3211 /* Read in characters and ranges, setting map bits. */
3214 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3218 /* \ might escape characters inside [...] and [^...]. */
3219 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
3221 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3229 /* Could be the end of the bracket expression. If it's
3230 not (i.e., when the bracket expression is `[]' so
3231 far), the ']' character bit gets set way below. */
3232 if (c
== ']' && p
!= p1
+ 1)
3235 /* Look ahead to see if it's a range when the last thing
3236 was a character class. */
3237 if (had_char_class
&& c
== '-' && *p
!= ']')
3238 FREE_STACK_RETURN (REG_ERANGE
);
3240 /* Look ahead to see if it's a range when the last thing
3241 was a character: if this is a hyphen not at the
3242 beginning or the end of a list, then it's the range
3245 && !(p
- 2 >= pattern
&& p
[-2] == '[')
3246 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
3250 = byte_compile_range (range_start
, &p
, pend
, translate
,
3252 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
3253 range_start
= 0xffffffff;
3256 else if (p
[0] == '-' && p
[1] != ']')
3257 { /* This handles ranges made up of characters only. */
3260 /* Move past the `-'. */
3263 ret
= byte_compile_range (c
, &p
, pend
, translate
, syntax
, b
);
3264 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
3265 range_start
= 0xffffffff;
3268 /* See if we're at the beginning of a possible character
3271 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
3272 { /* Leave room for the null. */
3273 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
3278 /* If pattern is `[[:'. */
3279 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3284 if ((c
== ':' && *p
== ']') || p
== pend
)
3286 if (c1
< CHAR_CLASS_MAX_LENGTH
)
3289 /* This is in any case an invalid class name. */
3294 /* If isn't a word bracketed by `[:' and `:]':
3295 undo the ending character, the letters, and leave
3296 the leading `:' and `[' (but set bits for them). */
3297 if (c
== ':' && *p
== ']')
3299 # if defined _LIBC || (defined WIDE_CHAR_SUPPORT && WIDE_CHAR_SUPPORT)
3300 boolean is_lower
= STREQ (str
, "lower");
3301 boolean is_upper
= STREQ (str
, "upper");
3305 wt
= IS_CHAR_CLASS (str
);
3307 FREE_STACK_RETURN (REG_ECTYPE
);
3309 /* Throw away the ] at the end of the character
3313 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3315 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
3318 if (__iswctype (__btowc (ch
), wt
))
3321 if (iswctype (btowc (ch
), wt
))
3325 if (translate
&& (is_upper
|| is_lower
)
3326 && (ISUPPER (ch
) || ISLOWER (ch
)))
3330 had_char_class
= true;
3333 boolean is_alnum
= STREQ (str
, "alnum");
3334 boolean is_alpha
= STREQ (str
, "alpha");
3335 boolean is_blank
= STREQ (str
, "blank");
3336 boolean is_cntrl
= STREQ (str
, "cntrl");
3337 boolean is_digit
= STREQ (str
, "digit");
3338 boolean is_graph
= STREQ (str
, "graph");
3339 boolean is_lower
= STREQ (str
, "lower");
3340 boolean is_print
= STREQ (str
, "print");
3341 boolean is_punct
= STREQ (str
, "punct");
3342 boolean is_space
= STREQ (str
, "space");
3343 boolean is_upper
= STREQ (str
, "upper");
3344 boolean is_xdigit
= STREQ (str
, "xdigit");
3346 if (!IS_CHAR_CLASS (str
))
3347 FREE_STACK_RETURN (REG_ECTYPE
);
3349 /* Throw away the ] at the end of the character
3353 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3355 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
3357 /* This was split into 3 if's to
3358 avoid an arbitrary limit in some compiler. */
3359 if ( (is_alnum
&& ISALNUM (ch
))
3360 || (is_alpha
&& ISALPHA (ch
))
3361 || (is_blank
&& ISBLANK (ch
))
3362 || (is_cntrl
&& ISCNTRL (ch
)))
3364 if ( (is_digit
&& ISDIGIT (ch
))
3365 || (is_graph
&& ISGRAPH (ch
))
3366 || (is_lower
&& ISLOWER (ch
))
3367 || (is_print
&& ISPRINT (ch
)))
3369 if ( (is_punct
&& ISPUNCT (ch
))
3370 || (is_space
&& ISSPACE (ch
))
3371 || (is_upper
&& ISUPPER (ch
))
3372 || (is_xdigit
&& ISXDIGIT (ch
)))
3374 if ( translate
&& (is_upper
|| is_lower
)
3375 && (ISUPPER (ch
) || ISLOWER (ch
)))
3378 had_char_class
= true;
3379 # endif /* libc || wctype.h */
3389 had_char_class
= false;
3392 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '=')
3394 unsigned char str
[MB_LEN_MAX
+ 1];
3397 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3403 /* If pattern is `[[='. */
3404 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3409 if ((c
== '=' && *p
== ']') || p
== pend
)
3411 if (c1
< MB_LEN_MAX
)
3414 /* This is in any case an invalid class name. */
3419 if (c
== '=' && *p
== ']' && str
[0] != '\0')
3421 /* If we have no collation data we use the default
3422 collation in which each character is in a class
3423 by itself. It also means that ASCII is the
3424 character set and therefore we cannot have character
3425 with more than one byte in the multibyte
3432 FREE_STACK_RETURN (REG_ECOLLATE
);
3434 /* Throw away the ] at the end of the equivalence
3438 /* Set the bit for the character. */
3439 SET_LIST_BIT (str
[0]);
3444 /* Try to match the byte sequence in `str' against
3445 those known to the collate implementation.
3446 First find out whether the bytes in `str' are
3447 actually from exactly one character. */
3448 const int32_t *table
;
3449 const unsigned char *weights
;
3450 const unsigned char *extra
;
3451 const int32_t *indirect
;
3453 const unsigned char *cp
= str
;
3456 /* This #include defines a local function! */
3457 # include <locale/weight.h>
3459 table
= (const int32_t *)
3460 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEMB
);
3461 weights
= (const unsigned char *)
3462 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTMB
);
3463 extra
= (const unsigned char *)
3464 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAMB
);
3465 indirect
= (const int32_t *)
3466 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTMB
);
3468 idx
= findidx (&cp
);
3469 if (idx
== 0 || cp
< str
+ c1
)
3470 /* This is no valid character. */
3471 FREE_STACK_RETURN (REG_ECOLLATE
);
3473 /* Throw away the ] at the end of the equivalence
3477 /* Now we have to go throught the whole table
3478 and find all characters which have the same
3481 XXX Note that this is not entirely correct.
3482 we would have to match multibyte sequences
3483 but this is not possible with the current
3485 for (ch
= 1; ch
< 256; ++ch
)
3486 /* XXX This test would have to be changed if we
3487 would allow matching multibyte sequences. */
3490 int32_t idx2
= table
[ch
];
3491 size_t len
= weights
[idx2
];
3493 /* Test whether the lenghts match. */
3494 if (weights
[idx
] == len
)
3496 /* They do. New compare the bytes of
3501 && (weights
[idx
+ 1 + cnt
]
3502 == weights
[idx2
+ 1 + cnt
]))
3506 /* They match. Mark the character as
3513 had_char_class
= true;
3523 had_char_class
= false;
3526 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '.')
3528 unsigned char str
[128]; /* Should be large enough. */
3531 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3537 /* If pattern is `[[.'. */
3538 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3543 if ((c
== '.' && *p
== ']') || p
== pend
)
3545 if (c1
< sizeof (str
))
3548 /* This is in any case an invalid class name. */
3553 if (c
== '.' && *p
== ']' && str
[0] != '\0')
3555 /* If we have no collation data we use the default
3556 collation in which each character is the name
3557 for its own class which contains only the one
3558 character. It also means that ASCII is the
3559 character set and therefore we cannot have character
3560 with more than one byte in the multibyte
3567 FREE_STACK_RETURN (REG_ECOLLATE
);
3569 /* Throw away the ] at the end of the equivalence
3573 /* Set the bit for the character. */
3574 SET_LIST_BIT (str
[0]);
3575 range_start
= ((const unsigned char *) str
)[0];
3580 /* Try to match the byte sequence in `str' against
3581 those known to the collate implementation.
3582 First find out whether the bytes in `str' are
3583 actually from exactly one character. */
3585 const int32_t *symb_table
;
3586 const unsigned char *extra
;
3593 _NL_CURRENT_WORD (LC_COLLATE
,
3594 _NL_COLLATE_SYMB_HASH_SIZEMB
);
3595 symb_table
= (const int32_t *)
3596 _NL_CURRENT (LC_COLLATE
,
3597 _NL_COLLATE_SYMB_TABLEMB
);
3598 extra
= (const unsigned char *)
3599 _NL_CURRENT (LC_COLLATE
,
3600 _NL_COLLATE_SYMB_EXTRAMB
);
3602 /* Locate the character in the hashing table. */
3603 hash
= elem_hash (str
, c1
);
3606 elem
= hash
% table_size
;
3607 second
= hash
% (table_size
- 2);
3608 while (symb_table
[2 * elem
] != 0)
3610 /* First compare the hashing value. */
3611 if (symb_table
[2 * elem
] == hash
3612 && c1
== extra
[symb_table
[2 * elem
+ 1]]
3614 &extra
[symb_table
[2 * elem
+ 1]
3618 /* Yep, this is the entry. */
3619 idx
= symb_table
[2 * elem
+ 1];
3620 idx
+= 1 + extra
[idx
];
3628 if (symb_table
[2 * elem
] == 0)
3629 /* This is no valid character. */
3630 FREE_STACK_RETURN (REG_ECOLLATE
);
3632 /* Throw away the ] at the end of the equivalence
3636 /* Now add the multibyte character(s) we found
3639 XXX Note that this is not entirely correct.
3640 we would have to match multibyte sequences
3641 but this is not possible with the current
3642 implementation. Also, we have to match
3643 collating symbols, which expand to more than
3644 one file, as a whole and not allow the
3645 individual bytes. */
3648 range_start
= extra
[idx
];
3651 SET_LIST_BIT (extra
[idx
]);
3656 had_char_class
= false;
3666 had_char_class
= false;
3671 had_char_class
= false;
3677 /* Discard any (non)matching list bytes that are all 0 at the
3678 end of the map. Decrease the map-length byte too. */
3679 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3688 if (syntax
& RE_NO_BK_PARENS
)
3695 if (syntax
& RE_NO_BK_PARENS
)
3702 if (syntax
& RE_NEWLINE_ALT
)
3709 if (syntax
& RE_NO_BK_VBAR
)
3716 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3717 goto handle_interval
;
3723 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3725 /* Do not translate the character after the \, so that we can
3726 distinguish, e.g., \B from \b, even if we normally would
3727 translate, e.g., B to b. */
3733 if (syntax
& RE_NO_BK_PARENS
)
3734 goto normal_backslash
;
3740 if (COMPILE_STACK_FULL
)
3742 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3743 compile_stack_elt_t
);
3744 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3746 compile_stack
.size
<<= 1;
3749 /* These are the values to restore when we hit end of this
3750 group. They are all relative offsets, so that if the
3751 whole pattern moves because of realloc, they will still
3753 COMPILE_STACK_TOP
.begalt_offset
= begalt
- COMPILED_BUFFER_VAR
;
3754 COMPILE_STACK_TOP
.fixup_alt_jump
3755 = fixup_alt_jump
? fixup_alt_jump
- COMPILED_BUFFER_VAR
+ 1 : 0;
3756 COMPILE_STACK_TOP
.laststart_offset
= b
- COMPILED_BUFFER_VAR
;
3757 COMPILE_STACK_TOP
.regnum
= regnum
;
3759 /* We will eventually replace the 0 with the number of
3760 groups inner to this one. But do not push a
3761 start_memory for groups beyond the last one we can
3762 represent in the compiled pattern. */
3763 if (regnum
<= MAX_REGNUM
)
3765 COMPILE_STACK_TOP
.inner_group_offset
= b
3766 - COMPILED_BUFFER_VAR
+ 2;
3767 BUF_PUSH_3 (start_memory
, regnum
, 0);
3770 compile_stack
.avail
++;
3775 /* If we've reached MAX_REGNUM groups, then this open
3776 won't actually generate any code, so we'll have to
3777 clear pending_exact explicitly. */
3783 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3785 if (COMPILE_STACK_EMPTY
)
3787 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3788 goto normal_backslash
;
3790 FREE_STACK_RETURN (REG_ERPAREN
);
3795 { /* Push a dummy failure point at the end of the
3796 alternative for a possible future
3797 `pop_failure_jump' to pop. See comments at
3798 `push_dummy_failure' in `re_match_2'. */
3799 BUF_PUSH (push_dummy_failure
);
3801 /* We allocated space for this jump when we assigned
3802 to `fixup_alt_jump', in the `handle_alt' case below. */
3803 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
3806 /* See similar code for backslashed left paren above. */
3807 if (COMPILE_STACK_EMPTY
)
3809 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3812 FREE_STACK_RETURN (REG_ERPAREN
);
3815 /* Since we just checked for an empty stack above, this
3816 ``can't happen''. */
3817 assert (compile_stack
.avail
!= 0);
3819 /* We don't just want to restore into `regnum', because
3820 later groups should continue to be numbered higher,
3821 as in `(ab)c(de)' -- the second group is #2. */
3822 regnum_t this_group_regnum
;
3824 compile_stack
.avail
--;
3825 begalt
= COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.begalt_offset
;
3827 = COMPILE_STACK_TOP
.fixup_alt_jump
3828 ? COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3830 laststart
= COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.laststart_offset
;
3831 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
3832 /* If we've reached MAX_REGNUM groups, then this open
3833 won't actually generate any code, so we'll have to
3834 clear pending_exact explicitly. */
3837 /* We're at the end of the group, so now we know how many
3838 groups were inside this one. */
3839 if (this_group_regnum
<= MAX_REGNUM
)
3841 UCHAR_T
*inner_group_loc
3842 = COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.inner_group_offset
;
3844 *inner_group_loc
= regnum
- this_group_regnum
;
3845 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
3846 regnum
- this_group_regnum
);
3852 case '|': /* `\|'. */
3853 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3854 goto normal_backslash
;
3856 if (syntax
& RE_LIMITED_OPS
)
3859 /* Insert before the previous alternative a jump which
3860 jumps to this alternative if the former fails. */
3861 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3862 INSERT_JUMP (on_failure_jump
, begalt
,
3863 b
+ 2 + 2 * OFFSET_ADDRESS_SIZE
);
3865 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3867 /* The alternative before this one has a jump after it
3868 which gets executed if it gets matched. Adjust that
3869 jump so it will jump to this alternative's analogous
3870 jump (put in below, which in turn will jump to the next
3871 (if any) alternative's such jump, etc.). The last such
3872 jump jumps to the correct final destination. A picture:
3878 If we are at `b', then fixup_alt_jump right now points to a
3879 three-byte space after `a'. We'll put in the jump, set
3880 fixup_alt_jump to right after `b', and leave behind three
3881 bytes which we'll fill in when we get to after `c'. */
3884 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
3886 /* Mark and leave space for a jump after this alternative,
3887 to be filled in later either by next alternative or
3888 when know we're at the end of a series of alternatives. */
3890 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3891 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3899 /* If \{ is a literal. */
3900 if (!(syntax
& RE_INTERVALS
)
3901 /* If we're at `\{' and it's not the open-interval
3903 || (syntax
& RE_NO_BK_BRACES
))
3904 goto normal_backslash
;
3908 /* If got here, then the syntax allows intervals. */
3910 /* At least (most) this many matches must be made. */
3911 int lower_bound
= -1, upper_bound
= -1;
3913 /* Place in the uncompiled pattern (i.e., just after
3914 the '{') to go back to if the interval is invalid. */
3915 const CHAR_T
*beg_interval
= p
;
3918 goto invalid_interval
;
3920 GET_UNSIGNED_NUMBER (lower_bound
);
3924 GET_UNSIGNED_NUMBER (upper_bound
);
3925 if (upper_bound
< 0)
3926 upper_bound
= RE_DUP_MAX
;
3929 /* Interval such as `{1}' => match exactly once. */
3930 upper_bound
= lower_bound
;
3932 if (! (0 <= lower_bound
&& lower_bound
<= upper_bound
))
3933 goto invalid_interval
;
3935 if (!(syntax
& RE_NO_BK_BRACES
))
3937 if (c
!= '\\' || p
== pend
)
3938 goto invalid_interval
;
3943 goto invalid_interval
;
3945 /* If it's invalid to have no preceding re. */
3948 if (syntax
& RE_CONTEXT_INVALID_OPS
3949 && !(syntax
& RE_INVALID_INTERVAL_ORD
))
3950 FREE_STACK_RETURN (REG_BADRPT
);
3951 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3954 goto unfetch_interval
;
3957 /* We just parsed a valid interval. */
3959 if (RE_DUP_MAX
< upper_bound
)
3960 FREE_STACK_RETURN (REG_BADBR
);
3962 /* If the upper bound is zero, don't want to succeed at
3963 all; jump from `laststart' to `b + 3', which will be
3964 the end of the buffer after we insert the jump. */
3965 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3966 instead of 'b + 3'. */
3967 if (upper_bound
== 0)
3969 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3970 INSERT_JUMP (jump
, laststart
, b
+ 1
3971 + OFFSET_ADDRESS_SIZE
);
3972 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3975 /* Otherwise, we have a nontrivial interval. When
3976 we're all done, the pattern will look like:
3977 set_number_at <jump count> <upper bound>
3978 set_number_at <succeed_n count> <lower bound>
3979 succeed_n <after jump addr> <succeed_n count>
3981 jump_n <succeed_n addr> <jump count>
3982 (The upper bound and `jump_n' are omitted if
3983 `upper_bound' is 1, though.) */
3985 { /* If the upper bound is > 1, we need to insert
3986 more at the end of the loop. */
3987 unsigned nbytes
= 2 + 4 * OFFSET_ADDRESS_SIZE
+
3988 (upper_bound
> 1) * (2 + 4 * OFFSET_ADDRESS_SIZE
);
3990 GET_BUFFER_SPACE (nbytes
);
3992 /* Initialize lower bound of the `succeed_n', even
3993 though it will be set during matching by its
3994 attendant `set_number_at' (inserted next),
3995 because `re_compile_fastmap' needs to know.
3996 Jump to the `jump_n' we might insert below. */
3997 INSERT_JUMP2 (succeed_n
, laststart
,
3998 b
+ 1 + 2 * OFFSET_ADDRESS_SIZE
3999 + (upper_bound
> 1) * (1 + 2 * OFFSET_ADDRESS_SIZE
)
4001 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
4003 /* Code to initialize the lower bound. Insert
4004 before the `succeed_n'. The `5' is the last two
4005 bytes of this `set_number_at', plus 3 bytes of
4006 the following `succeed_n'. */
4007 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
4008 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
4009 of the following `succeed_n'. */
4010 PREFIX(insert_op2
) (set_number_at
, laststart
, 1
4011 + 2 * OFFSET_ADDRESS_SIZE
, lower_bound
, b
);
4012 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
4014 if (upper_bound
> 1)
4015 { /* More than one repetition is allowed, so
4016 append a backward jump to the `succeed_n'
4017 that starts this interval.
4019 When we've reached this during matching,
4020 we'll have matched the interval once, so
4021 jump back only `upper_bound - 1' times. */
4022 STORE_JUMP2 (jump_n
, b
, laststart
4023 + 2 * OFFSET_ADDRESS_SIZE
+ 1,
4025 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
4027 /* The location we want to set is the second
4028 parameter of the `jump_n'; that is `b-2' as
4029 an absolute address. `laststart' will be
4030 the `set_number_at' we're about to insert;
4031 `laststart+3' the number to set, the source
4032 for the relative address. But we are
4033 inserting into the middle of the pattern --
4034 so everything is getting moved up by 5.
4035 Conclusion: (b - 2) - (laststart + 3) + 5,
4036 i.e., b - laststart.
4038 We insert this at the beginning of the loop
4039 so that if we fail during matching, we'll
4040 reinitialize the bounds. */
4041 PREFIX(insert_op2
) (set_number_at
, laststart
,
4043 upper_bound
- 1, b
);
4044 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
4051 if (!(syntax
& RE_INVALID_INTERVAL_ORD
))
4052 FREE_STACK_RETURN (p
== pend
? REG_EBRACE
: REG_BADBR
);
4054 /* Match the characters as literals. */
4057 if (syntax
& RE_NO_BK_BRACES
)
4060 goto normal_backslash
;
4064 /* There is no way to specify the before_dot and after_dot
4065 operators. rms says this is ok. --karl */
4073 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
4079 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
4085 if (syntax
& RE_NO_GNU_OPS
)
4088 BUF_PUSH (wordchar
);
4093 if (syntax
& RE_NO_GNU_OPS
)
4096 BUF_PUSH (notwordchar
);
4101 if (syntax
& RE_NO_GNU_OPS
)
4107 if (syntax
& RE_NO_GNU_OPS
)
4113 if (syntax
& RE_NO_GNU_OPS
)
4115 BUF_PUSH (wordbound
);
4119 if (syntax
& RE_NO_GNU_OPS
)
4121 BUF_PUSH (notwordbound
);
4125 if (syntax
& RE_NO_GNU_OPS
)
4131 if (syntax
& RE_NO_GNU_OPS
)
4136 case '1': case '2': case '3': case '4': case '5':
4137 case '6': case '7': case '8': case '9':
4138 if (syntax
& RE_NO_BK_REFS
)
4144 FREE_STACK_RETURN (REG_ESUBREG
);
4146 /* Can't back reference to a subexpression if inside of it. */
4147 if (group_in_compile_stack (compile_stack
, (regnum_t
) c1
))
4151 BUF_PUSH_2 (duplicate
, c1
);
4157 if (syntax
& RE_BK_PLUS_QM
)
4160 goto normal_backslash
;
4164 /* You might think it would be useful for \ to mean
4165 not to translate; but if we don't translate it
4166 it will never match anything. */
4174 /* Expects the character in `c'. */
4176 /* If no exactn currently being built. */
4179 /* If last exactn handle binary(or character) and
4180 new exactn handle character(or binary). */
4181 || is_exactn_bin
!= is_binary
[p
- 1 - pattern
]
4184 /* If last exactn not at current position. */
4185 || pending_exact
+ *pending_exact
+ 1 != b
4187 /* We have only one byte following the exactn for the count. */
4188 || *pending_exact
== (1 << BYTEWIDTH
) - 1
4190 /* If followed by a repetition operator. */
4191 || *p
== '*' || *p
== '^'
4192 || ((syntax
& RE_BK_PLUS_QM
)
4193 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
4194 : (*p
== '+' || *p
== '?'))
4195 || ((syntax
& RE_INTERVALS
)
4196 && ((syntax
& RE_NO_BK_BRACES
)
4198 : (p
[0] == '\\' && p
[1] == '{'))))
4200 /* Start building a new exactn. */
4205 /* Is this exactn binary data or character? */
4206 is_exactn_bin
= is_binary
[p
- 1 - pattern
];
4208 BUF_PUSH_2 (exactn_bin
, 0);
4210 BUF_PUSH_2 (exactn
, 0);
4212 BUF_PUSH_2 (exactn
, 0);
4214 pending_exact
= b
- 1;
4221 } /* while p != pend */
4224 /* Through the pattern now. */
4227 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
4229 if (!COMPILE_STACK_EMPTY
)
4230 FREE_STACK_RETURN (REG_EPAREN
);
4232 /* If we don't want backtracking, force success
4233 the first time we reach the end of the compiled pattern. */
4234 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
4242 free (compile_stack
.stack
);
4244 /* We have succeeded; set the length of the buffer. */
4246 bufp
->used
= (uintptr_t) b
- (uintptr_t) COMPILED_BUFFER_VAR
;
4248 bufp
->used
= b
- bufp
->buffer
;
4254 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4255 PREFIX(print_compiled_pattern
) (bufp
);
4259 #ifndef MATCH_MAY_ALLOCATE
4260 /* Initialize the failure stack to the largest possible stack. This
4261 isn't necessary unless we're trying to avoid calling alloca in
4262 the search and match routines. */
4264 int num_regs
= bufp
->re_nsub
+ 1;
4266 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4267 is strictly greater than re_max_failures, the largest possible stack
4268 is 2 * re_max_failures failure points. */
4269 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
4271 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
4274 if (! fail_stack
.stack
)
4276 = (PREFIX(fail_stack_elt_t
) *) xmalloc (fail_stack
.size
4277 * sizeof (PREFIX(fail_stack_elt_t
)));
4280 = (PREFIX(fail_stack_elt_t
) *) xrealloc (fail_stack
.stack
,
4282 * sizeof (PREFIX(fail_stack_elt_t
))));
4283 # else /* not emacs */
4284 if (! fail_stack
.stack
)
4286 = (PREFIX(fail_stack_elt_t
) *) malloc (fail_stack
.size
4287 * sizeof (PREFIX(fail_stack_elt_t
)));
4290 = (PREFIX(fail_stack_elt_t
) *) realloc (fail_stack
.stack
,
4292 * sizeof (PREFIX(fail_stack_elt_t
))));
4293 # endif /* not emacs */
4296 PREFIX(regex_grow_registers
) (num_regs
);
4298 #endif /* not MATCH_MAY_ALLOCATE */
4301 } /* regex_compile */
4303 /* Subroutines for `regex_compile'. */
4305 /* Store OP at LOC followed by two-byte integer parameter ARG. */
4306 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4309 PREFIX(store_op1
) (op
, loc
, arg
)
4314 *loc
= (UCHAR_T
) op
;
4315 STORE_NUMBER (loc
+ 1, arg
);
4319 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4320 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4323 PREFIX(store_op2
) (op
, loc
, arg1
, arg2
)
4328 *loc
= (UCHAR_T
) op
;
4329 STORE_NUMBER (loc
+ 1, arg1
);
4330 STORE_NUMBER (loc
+ 1 + OFFSET_ADDRESS_SIZE
, arg2
);
4334 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
4335 for OP followed by two-byte integer parameter ARG. */
4336 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4339 PREFIX(insert_op1
) (op
, loc
, arg
, end
)
4345 register UCHAR_T
*pfrom
= end
;
4346 register UCHAR_T
*pto
= end
+ 1 + OFFSET_ADDRESS_SIZE
;
4348 while (pfrom
!= loc
)
4351 PREFIX(store_op1
) (op
, loc
, arg
);
4355 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4356 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4359 PREFIX(insert_op2
) (op
, loc
, arg1
, arg2
, end
)
4365 register UCHAR_T
*pfrom
= end
;
4366 register UCHAR_T
*pto
= end
+ 1 + 2 * OFFSET_ADDRESS_SIZE
;
4368 while (pfrom
!= loc
)
4371 PREFIX(store_op2
) (op
, loc
, arg1
, arg2
);
4375 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
4376 after an alternative or a begin-subexpression. We assume there is at
4377 least one character before the ^. */
4380 PREFIX(at_begline_loc_p
) (pattern
, p
, syntax
)
4381 const CHAR_T
*pattern
, *p
;
4382 reg_syntax_t syntax
;
4384 const CHAR_T
*prev
= p
- 2;
4385 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
4388 /* After a subexpression? */
4389 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
4390 /* After an alternative? */
4391 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
4395 /* The dual of at_begline_loc_p. This one is for $. We assume there is
4396 at least one character after the $, i.e., `P < PEND'. */
4399 PREFIX(at_endline_loc_p
) (p
, pend
, syntax
)
4400 const CHAR_T
*p
, *pend
;
4401 reg_syntax_t syntax
;
4403 const CHAR_T
*next
= p
;
4404 boolean next_backslash
= *next
== '\\';
4405 const CHAR_T
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
4408 /* Before a subexpression? */
4409 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
4410 : next_backslash
&& next_next
&& *next_next
== ')')
4411 /* Before an alternative? */
4412 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
4413 : next_backslash
&& next_next
&& *next_next
== '|');
4416 #else /* not INSIDE_RECURSION */
4418 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4419 false if it's not. */
4422 group_in_compile_stack (compile_stack
, regnum
)
4423 compile_stack_type compile_stack
;
4428 for (this_element
= compile_stack
.avail
- 1;
4431 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
4436 #endif /* not INSIDE_RECURSION */
4438 #ifdef INSIDE_RECURSION
4441 /* This insert space, which size is "num", into the pattern at "loc".
4442 "end" must point the end of the allocated buffer. */
4444 insert_space (num
, loc
, end
)
4449 register CHAR_T
*pto
= end
;
4450 register CHAR_T
*pfrom
= end
- num
;
4452 while (pfrom
>= loc
)
4458 static reg_errcode_t
4459 wcs_compile_range (range_start_char
, p_ptr
, pend
, translate
, syntax
, b
,
4461 CHAR_T range_start_char
;
4462 const CHAR_T
**p_ptr
, *pend
;
4463 CHAR_T
*char_set
, *b
;
4464 RE_TRANSLATE_TYPE translate
;
4465 reg_syntax_t syntax
;
4467 const CHAR_T
*p
= *p_ptr
;
4468 CHAR_T range_start
, range_end
;
4472 uint32_t start_val
, end_val
;
4478 nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
4481 const char *collseq
= (const char *) _NL_CURRENT(LC_COLLATE
,
4482 _NL_COLLATE_COLLSEQWC
);
4483 const unsigned char *extra
= (const unsigned char *)
4484 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
4486 if (range_start_char
< -1)
4488 /* range_start is a collating symbol. */
4490 /* Retreive the index and get collation sequence value. */
4491 wextra
= (int32_t*)(extra
+ char_set
[-range_start_char
]);
4492 start_val
= wextra
[1 + *wextra
];
4495 start_val
= collseq_table_lookup(collseq
, TRANSLATE(range_start_char
));
4497 end_val
= collseq_table_lookup (collseq
, TRANSLATE (p
[0]));
4499 /* Report an error if the range is empty and the syntax prohibits
4501 ret
= ((syntax
& RE_NO_EMPTY_RANGES
)
4502 && (start_val
> end_val
))? REG_ERANGE
: REG_NOERROR
;
4504 /* Insert space to the end of the char_ranges. */
4505 insert_space(2, b
- char_set
[5] - 2, b
- 1);
4506 *(b
- char_set
[5] - 2) = (wchar_t)start_val
;
4507 *(b
- char_set
[5] - 1) = (wchar_t)end_val
;
4508 char_set
[4]++; /* ranges_index */
4513 range_start
= (range_start_char
>= 0)? TRANSLATE (range_start_char
):
4515 range_end
= TRANSLATE (p
[0]);
4516 /* Report an error if the range is empty and the syntax prohibits
4518 ret
= ((syntax
& RE_NO_EMPTY_RANGES
)
4519 && (range_start
> range_end
))? REG_ERANGE
: REG_NOERROR
;
4521 /* Insert space to the end of the char_ranges. */
4522 insert_space(2, b
- char_set
[5] - 2, b
- 1);
4523 *(b
- char_set
[5] - 2) = range_start
;
4524 *(b
- char_set
[5] - 1) = range_end
;
4525 char_set
[4]++; /* ranges_index */
4527 /* Have to increment the pointer into the pattern string, so the
4528 caller isn't still at the ending character. */
4534 /* Read the ending character of a range (in a bracket expression) from the
4535 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4536 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4537 Then we set the translation of all bits between the starting and
4538 ending characters (inclusive) in the compiled pattern B.
4540 Return an error code.
4542 We use these short variable names so we can use the same macros as
4543 `regex_compile' itself. */
4545 static reg_errcode_t
4546 byte_compile_range (range_start_char
, p_ptr
, pend
, translate
, syntax
, b
)
4547 unsigned int range_start_char
;
4548 const char **p_ptr
, *pend
;
4549 RE_TRANSLATE_TYPE translate
;
4550 reg_syntax_t syntax
;
4554 const char *p
= *p_ptr
;
4556 # if defined _LIBC && _LIBC
4557 const unsigned char *collseq
;
4558 unsigned int start_colseq
;
4559 unsigned int end_colseq
;
4567 /* Have to increment the pointer into the pattern string, so the
4568 caller isn't still at the ending character. */
4571 /* Report an error if the range is empty and the syntax prohibits this. */
4572 ret
= syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
4574 # if defined _LIBC && _LIBC
4575 collseq
= (const unsigned char *) _NL_CURRENT (LC_COLLATE
,
4576 _NL_COLLATE_COLLSEQMB
);
4578 start_colseq
= collseq
[(unsigned char) TRANSLATE (range_start_char
)];
4579 end_colseq
= collseq
[(unsigned char) TRANSLATE (p
[0])];
4580 for (this_char
= 0; this_char
<= (unsigned char) -1; ++this_char
)
4582 unsigned int this_colseq
= collseq
[(unsigned char) TRANSLATE (this_char
)];
4584 if (start_colseq
<= this_colseq
&& this_colseq
<= end_colseq
)
4586 SET_LIST_BIT (TRANSLATE (this_char
));
4591 /* Here we see why `this_char' has to be larger than an `unsigned
4592 char' -- we would otherwise go into an infinite loop, since all
4593 characters <= 0xff. */
4594 range_start_char
= TRANSLATE (range_start_char
);
4595 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4596 and some compilers cast it to int implicitly, so following for_loop
4597 may fall to (almost) infinite loop.
4598 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4599 To avoid this, we cast p[0] to unsigned int and truncate it. */
4600 end_char
= ((unsigned)TRANSLATE(p
[0]) & ((1 << BYTEWIDTH
) - 1));
4602 for (this_char
= range_start_char
; this_char
<= end_char
; ++this_char
)
4604 SET_LIST_BIT (TRANSLATE (this_char
));
4613 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4614 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4615 characters can start a string that matches the pattern. This fastmap
4616 is used by re_search to skip quickly over impossible starting points.
4618 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4619 area as BUFP->fastmap.
4621 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4624 Returns 0 if we succeed, -2 if an internal error. */
4627 /* local function for re_compile_fastmap.
4628 truncate wchar_t character to char. */
4629 static unsigned char truncate_wchar (CHAR_T c
);
4631 static unsigned char
4635 unsigned char buf
[MB_CUR_MAX
];
4638 memset (&state
, '\0', sizeof (state
));
4640 retval
= __wcrtomb (buf
, c
, &state
);
4642 retval
= wcrtomb (buf
, c
, &state
);
4644 return retval
> 0 ? buf
[0] : (unsigned char) c
;
4649 PREFIX(re_compile_fastmap
) (bufp
)
4650 struct re_pattern_buffer
*bufp
;
4653 #ifdef MATCH_MAY_ALLOCATE
4654 PREFIX(fail_stack_type
) fail_stack
;
4656 #ifndef REGEX_MALLOC
4660 register char *fastmap
= bufp
->fastmap
;
4663 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4664 pattern to (char*) in regex_compile. */
4665 UCHAR_T
*pattern
= (UCHAR_T
*)bufp
->buffer
;
4666 register UCHAR_T
*pend
= (UCHAR_T
*) (bufp
->buffer
+ bufp
->used
);
4668 UCHAR_T
*pattern
= bufp
->buffer
;
4669 register UCHAR_T
*pend
= pattern
+ bufp
->used
;
4671 UCHAR_T
*p
= pattern
;
4674 /* This holds the pointer to the failure stack, when
4675 it is allocated relocatably. */
4676 fail_stack_elt_t
*failure_stack_ptr
;
4679 /* Assume that each path through the pattern can be null until
4680 proven otherwise. We set this false at the bottom of switch
4681 statement, to which we get only if a particular path doesn't
4682 match the empty string. */
4683 boolean path_can_be_null
= true;
4685 /* We aren't doing a `succeed_n' to begin with. */
4686 boolean succeed_n_p
= false;
4688 assert (fastmap
!= NULL
&& p
!= NULL
);
4691 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4692 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4693 bufp
->can_be_null
= 0;
4697 if (p
== pend
|| *p
== succeed
)
4699 /* We have reached the (effective) end of pattern. */
4700 if (!FAIL_STACK_EMPTY ())
4702 bufp
->can_be_null
|= path_can_be_null
;
4704 /* Reset for next path. */
4705 path_can_be_null
= true;
4707 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
4715 /* We should never be about to go beyond the end of the pattern. */
4718 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4721 /* I guess the idea here is to simply not bother with a fastmap
4722 if a backreference is used, since it's too hard to figure out
4723 the fastmap for the corresponding group. Setting
4724 `can_be_null' stops `re_search_2' from using the fastmap, so
4725 that is all we do. */
4727 bufp
->can_be_null
= 1;
4731 /* Following are the cases which match a character. These end
4736 fastmap
[truncate_wchar(p
[1])] = 1;
4750 /* It is hard to distinguish fastmap from (multi byte) characters
4751 which depends on current locale. */
4756 bufp
->can_be_null
= 1;
4760 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
4761 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
4767 /* Chars beyond end of map must be allowed. */
4768 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
4771 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
4772 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
4778 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4779 if (SYNTAX (j
) == Sword
)
4785 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4786 if (SYNTAX (j
) != Sword
)
4793 int fastmap_newline
= fastmap
['\n'];
4795 /* `.' matches anything ... */
4796 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4799 /* ... except perhaps newline. */
4800 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
4801 fastmap
['\n'] = fastmap_newline
;
4803 /* Return if we have already set `can_be_null'; if we have,
4804 then the fastmap is irrelevant. Something's wrong here. */
4805 else if (bufp
->can_be_null
)
4808 /* Otherwise, have to check alternative paths. */
4815 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4816 if (SYNTAX (j
) == (enum syntaxcode
) k
)
4823 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4824 if (SYNTAX (j
) != (enum syntaxcode
) k
)
4829 /* All cases after this match the empty string. These end with
4849 case push_dummy_failure
:
4854 case pop_failure_jump
:
4855 case maybe_pop_jump
:
4858 case dummy_failure_jump
:
4859 EXTRACT_NUMBER_AND_INCR (j
, p
);
4864 /* Jump backward implies we just went through the body of a
4865 loop and matched nothing. Opcode jumped to should be
4866 `on_failure_jump' or `succeed_n'. Just treat it like an
4867 ordinary jump. For a * loop, it has pushed its failure
4868 point already; if so, discard that as redundant. */
4869 if ((re_opcode_t
) *p
!= on_failure_jump
4870 && (re_opcode_t
) *p
!= succeed_n
)
4874 EXTRACT_NUMBER_AND_INCR (j
, p
);
4877 /* If what's on the stack is where we are now, pop it. */
4878 if (!FAIL_STACK_EMPTY ()
4879 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
4885 case on_failure_jump
:
4886 case on_failure_keep_string_jump
:
4887 handle_on_failure_jump
:
4888 EXTRACT_NUMBER_AND_INCR (j
, p
);
4890 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4891 end of the pattern. We don't want to push such a point,
4892 since when we restore it above, entering the switch will
4893 increment `p' past the end of the pattern. We don't need
4894 to push such a point since we obviously won't find any more
4895 fastmap entries beyond `pend'. Such a pattern can match
4896 the null string, though. */
4899 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
4901 RESET_FAIL_STACK ();
4906 bufp
->can_be_null
= 1;
4910 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
4911 succeed_n_p
= false;
4918 /* Get to the number of times to succeed. */
4919 p
+= OFFSET_ADDRESS_SIZE
;
4921 /* Increment p past the n for when k != 0. */
4922 EXTRACT_NUMBER_AND_INCR (k
, p
);
4925 p
-= 2 * OFFSET_ADDRESS_SIZE
;
4926 succeed_n_p
= true; /* Spaghetti code alert. */
4927 goto handle_on_failure_jump
;
4933 p
+= 2 * OFFSET_ADDRESS_SIZE
;
4944 abort (); /* We have listed all the cases. */
4947 /* Getting here means we have found the possible starting
4948 characters for one path of the pattern -- and that the empty
4949 string does not match. We need not follow this path further.
4950 Instead, look at the next alternative (remembered on the
4951 stack), or quit if no more. The test at the top of the loop
4952 does these things. */
4953 path_can_be_null
= false;
4957 /* Set `can_be_null' for the last path (also the first path, if the
4958 pattern is empty). */
4959 bufp
->can_be_null
|= path_can_be_null
;
4962 RESET_FAIL_STACK ();
4966 #else /* not INSIDE_RECURSION */
4969 re_compile_fastmap (bufp
)
4970 struct re_pattern_buffer
*bufp
;
4973 if (MB_CUR_MAX
!= 1)
4974 return wcs_re_compile_fastmap(bufp
);
4977 return byte_re_compile_fastmap(bufp
);
4978 } /* re_compile_fastmap */
4980 weak_alias (__re_compile_fastmap
, re_compile_fastmap
)
4984 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4985 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4986 this memory for recording register information. STARTS and ENDS
4987 must be allocated using the malloc library routine, and must each
4988 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4990 If NUM_REGS == 0, then subsequent matches should allocate their own
4993 Unless this function is called, the first search or match using
4994 PATTERN_BUFFER will allocate its own register data, without
4995 freeing the old data. */
4998 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4999 struct re_pattern_buffer
*bufp
;
5000 struct re_registers
*regs
;
5002 regoff_t
*starts
, *ends
;
5006 bufp
->regs_allocated
= REGS_REALLOCATE
;
5007 regs
->num_regs
= num_regs
;
5008 regs
->start
= starts
;
5013 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5015 regs
->start
= regs
->end
= (regoff_t
*) 0;
5019 weak_alias (__re_set_registers
, re_set_registers
)
5022 /* Searching routines. */
5024 /* Like re_search_2, below, but only one string is specified, and
5025 doesn't let you say where to stop matching. */
5028 re_search (bufp
, string
, size
, startpos
, range
, regs
)
5029 struct re_pattern_buffer
*bufp
;
5031 int size
, startpos
, range
;
5032 struct re_registers
*regs
;
5034 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
5038 weak_alias (__re_search
, re_search
)
5042 /* Using the compiled pattern in BUFP->buffer, first tries to match the
5043 virtual concatenation of STRING1 and STRING2, starting first at index
5044 STARTPOS, then at STARTPOS + 1, and so on.
5046 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
5048 RANGE is how far to scan while trying to match. RANGE = 0 means try
5049 only at STARTPOS; in general, the last start tried is STARTPOS +
5052 In REGS, return the indices of the virtual concatenation of STRING1
5053 and STRING2 that matched the entire BUFP->buffer and its contained
5056 Do not consider matching one past the index STOP in the virtual
5057 concatenation of STRING1 and STRING2.
5059 We return either the position in the strings at which the match was
5060 found, -1 if no match, or -2 if error (such as failure
5064 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
5065 struct re_pattern_buffer
*bufp
;
5066 const char *string1
, *string2
;
5070 struct re_registers
*regs
;
5074 if (MB_CUR_MAX
!= 1)
5075 return wcs_re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
,
5079 return byte_re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
,
5083 weak_alias (__re_search_2
, re_search_2
)
5086 #endif /* not INSIDE_RECURSION */
5088 #ifdef INSIDE_RECURSION
5090 #ifdef MATCH_MAY_ALLOCATE
5091 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5093 # define FREE_VAR(var) if (var) free (var); var = NULL
5097 # define MAX_ALLOCA_SIZE 2000
5099 # define FREE_WCS_BUFFERS() \
5101 if (size1 > MAX_ALLOCA_SIZE) \
5103 free (wcs_string1); \
5104 free (mbs_offset1); \
5108 FREE_VAR (wcs_string1); \
5109 FREE_VAR (mbs_offset1); \
5111 if (size2 > MAX_ALLOCA_SIZE) \
5113 free (wcs_string2); \
5114 free (mbs_offset2); \
5118 FREE_VAR (wcs_string2); \
5119 FREE_VAR (mbs_offset2); \
5127 PREFIX(re_search_2
) (bufp
, string1
, size1
, string2
, size2
, startpos
, range
,
5129 struct re_pattern_buffer
*bufp
;
5130 const char *string1
, *string2
;
5134 struct re_registers
*regs
;
5138 register char *fastmap
= bufp
->fastmap
;
5139 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5140 int total_size
= size1
+ size2
;
5141 int endpos
= startpos
+ range
;
5143 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5144 wchar_t *wcs_string1
= NULL
, *wcs_string2
= NULL
;
5145 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5146 int wcs_size1
= 0, wcs_size2
= 0;
5147 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5148 int *mbs_offset1
= NULL
, *mbs_offset2
= NULL
;
5149 /* They hold whether each wchar_t is binary data or not. */
5150 char *is_binary
= NULL
;
5153 /* Check for out-of-range STARTPOS. */
5154 if (startpos
< 0 || startpos
> total_size
)
5157 /* Fix up RANGE if it might eventually take us outside
5158 the virtual concatenation of STRING1 and STRING2.
5159 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5161 range
= 0 - startpos
;
5162 else if (endpos
> total_size
)
5163 range
= total_size
- startpos
;
5165 /* If the search isn't to be a backwards one, don't waste time in a
5166 search for a pattern that must be anchored. */
5167 if (bufp
->used
> 0 && range
> 0
5168 && ((re_opcode_t
) bufp
->buffer
[0] == begbuf
5169 /* `begline' is like `begbuf' if it cannot match at newlines. */
5170 || ((re_opcode_t
) bufp
->buffer
[0] == begline
5171 && !bufp
->newline_anchor
)))
5180 /* In a forward search for something that starts with \=.
5181 don't keep searching past point. */
5182 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
5184 range
= PT
- startpos
;
5190 /* Update the fastmap now if not correct already. */
5191 if (fastmap
&& !bufp
->fastmap_accurate
)
5192 if (re_compile_fastmap (bufp
) == -2)
5196 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5197 fill them with converted string. */
5200 if (size1
> MAX_ALLOCA_SIZE
)
5202 wcs_string1
= TALLOC (size1
+ 1, CHAR_T
);
5203 mbs_offset1
= TALLOC (size1
+ 1, int);
5204 is_binary
= TALLOC (size1
+ 1, char);
5208 wcs_string1
= REGEX_TALLOC (size1
+ 1, CHAR_T
);
5209 mbs_offset1
= REGEX_TALLOC (size1
+ 1, int);
5210 is_binary
= REGEX_TALLOC (size1
+ 1, char);
5212 if (!wcs_string1
|| !mbs_offset1
|| !is_binary
)
5214 if (size1
> MAX_ALLOCA_SIZE
)
5222 FREE_VAR (wcs_string1
);
5223 FREE_VAR (mbs_offset1
);
5224 FREE_VAR (is_binary
);
5228 wcs_size1
= convert_mbs_to_wcs(wcs_string1
, string1
, size1
,
5229 mbs_offset1
, is_binary
);
5230 wcs_string1
[wcs_size1
] = L
'\0'; /* for a sentinel */
5231 if (size1
> MAX_ALLOCA_SIZE
)
5234 FREE_VAR (is_binary
);
5238 if (size2
> MAX_ALLOCA_SIZE
)
5240 wcs_string2
= TALLOC (size2
+ 1, CHAR_T
);
5241 mbs_offset2
= TALLOC (size2
+ 1, int);
5242 is_binary
= TALLOC (size2
+ 1, char);
5246 wcs_string2
= REGEX_TALLOC (size2
+ 1, CHAR_T
);
5247 mbs_offset2
= REGEX_TALLOC (size2
+ 1, int);
5248 is_binary
= REGEX_TALLOC (size2
+ 1, char);
5250 if (!wcs_string2
|| !mbs_offset2
|| !is_binary
)
5252 FREE_WCS_BUFFERS ();
5253 if (size2
> MAX_ALLOCA_SIZE
)
5256 FREE_VAR (is_binary
);
5259 wcs_size2
= convert_mbs_to_wcs(wcs_string2
, string2
, size2
,
5260 mbs_offset2
, is_binary
);
5261 wcs_string2
[wcs_size2
] = L
'\0'; /* for a sentinel */
5262 if (size2
> MAX_ALLOCA_SIZE
)
5265 FREE_VAR (is_binary
);
5270 /* Loop through the string, looking for a place to start matching. */
5273 /* If a fastmap is supplied, skip quickly over characters that
5274 cannot be the start of a match. If the pattern can match the
5275 null string, however, we don't need to skip characters; we want
5276 the first null string. */
5277 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
5279 if (range
> 0) /* Searching forwards. */
5281 register const char *d
;
5282 register int lim
= 0;
5285 if (startpos
< size1
&& startpos
+ range
>= size1
)
5286 lim
= range
- (size1
- startpos
);
5288 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
5290 /* Written out as an if-else to avoid testing `translate'
5294 && !fastmap
[(unsigned char)
5295 translate
[(unsigned char) *d
++]])
5298 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
5301 startpos
+= irange
- range
;
5303 else /* Searching backwards. */
5305 register CHAR_T c
= (size1
== 0 || startpos
>= size1
5306 ? string2
[startpos
- size1
]
5307 : string1
[startpos
]);
5309 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
5314 /* If can't match the null string, and that's all we have left, fail. */
5315 if (range
>= 0 && startpos
== total_size
&& fastmap
5316 && !bufp
->can_be_null
)
5319 FREE_WCS_BUFFERS ();
5325 val
= wcs_re_match_2_internal (bufp
, string1
, size1
, string2
,
5326 size2
, startpos
, regs
, stop
,
5327 wcs_string1
, wcs_size1
,
5328 wcs_string2
, wcs_size2
,
5329 mbs_offset1
, mbs_offset2
);
5331 val
= byte_re_match_2_internal (bufp
, string1
, size1
, string2
,
5332 size2
, startpos
, regs
, stop
);
5335 #ifndef REGEX_MALLOC
5344 FREE_WCS_BUFFERS ();
5352 FREE_WCS_BUFFERS ();
5372 FREE_WCS_BUFFERS ();
5378 /* This converts PTR, a pointer into one of the search wchar_t strings
5379 `string1' and `string2' into an multibyte string offset from the
5380 beginning of that string. We use mbs_offset to optimize.
5381 See convert_mbs_to_wcs. */
5382 # define POINTER_TO_OFFSET(ptr) \
5383 (FIRST_STRING_P (ptr) \
5384 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5385 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5388 /* This converts PTR, a pointer into one of the search strings `string1'
5389 and `string2' into an offset from the beginning of that string. */
5390 # define POINTER_TO_OFFSET(ptr) \
5391 (FIRST_STRING_P (ptr) \
5392 ? ((regoff_t) ((ptr) - string1)) \
5393 : ((regoff_t) ((ptr) - string2 + size1)))
5396 /* Macros for dealing with the split strings in re_match_2. */
5398 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5400 /* Call before fetching a character with *d. This switches over to
5401 string2 if necessary. */
5402 #define PREFETCH() \
5405 /* End of string2 => fail. */ \
5406 if (dend == end_match_2) \
5408 /* End of string1 => advance to string2. */ \
5410 dend = end_match_2; \
5413 /* Test if at very beginning or at very end of the virtual concatenation
5414 of `string1' and `string2'. If only one string, it's `string2'. */
5415 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5416 #define AT_STRINGS_END(d) ((d) == end2)
5419 /* Test if D points to a character which is word-constituent. We have
5420 two special cases to check for: if past the end of string1, look at
5421 the first character in string2; and if before the beginning of
5422 string2, look at the last character in string1. */
5424 /* Use internationalized API instead of SYNTAX. */
5425 # define WORDCHAR_P(d) \
5426 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5427 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5428 || ((d) == end1 ? *string2 \
5429 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5431 # define WORDCHAR_P(d) \
5432 (SYNTAX ((d) == end1 ? *string2 \
5433 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5437 /* Disabled due to a compiler bug -- see comment at case wordbound */
5439 /* Test if the character before D and the one at D differ with respect
5440 to being word-constituent. */
5441 #define AT_WORD_BOUNDARY(d) \
5442 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5443 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5446 /* Free everything we malloc. */
5447 #ifdef MATCH_MAY_ALLOCATE
5449 # define FREE_VARIABLES() \
5451 REGEX_FREE_STACK (fail_stack.stack); \
5452 FREE_VAR (regstart); \
5453 FREE_VAR (regend); \
5454 FREE_VAR (old_regstart); \
5455 FREE_VAR (old_regend); \
5456 FREE_VAR (best_regstart); \
5457 FREE_VAR (best_regend); \
5458 FREE_VAR (reg_info); \
5459 FREE_VAR (reg_dummy); \
5460 FREE_VAR (reg_info_dummy); \
5461 if (!cant_free_wcs_buf) \
5463 FREE_VAR (string1); \
5464 FREE_VAR (string2); \
5465 FREE_VAR (mbs_offset1); \
5466 FREE_VAR (mbs_offset2); \
5470 # define FREE_VARIABLES() \
5472 REGEX_FREE_STACK (fail_stack.stack); \
5473 FREE_VAR (regstart); \
5474 FREE_VAR (regend); \
5475 FREE_VAR (old_regstart); \
5476 FREE_VAR (old_regend); \
5477 FREE_VAR (best_regstart); \
5478 FREE_VAR (best_regend); \
5479 FREE_VAR (reg_info); \
5480 FREE_VAR (reg_dummy); \
5481 FREE_VAR (reg_info_dummy); \
5486 # define FREE_VARIABLES() \
5488 if (!cant_free_wcs_buf) \
5490 FREE_VAR (string1); \
5491 FREE_VAR (string2); \
5492 FREE_VAR (mbs_offset1); \
5493 FREE_VAR (mbs_offset2); \
5497 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5499 #endif /* not MATCH_MAY_ALLOCATE */
5501 /* These values must meet several constraints. They must not be valid
5502 register values; since we have a limit of 255 registers (because
5503 we use only one byte in the pattern for the register number), we can
5504 use numbers larger than 255. They must differ by 1, because of
5505 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5506 be larger than the value for the highest register, so we do not try
5507 to actually save any registers when none are active. */
5508 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5509 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5511 #else /* not INSIDE_RECURSION */
5512 /* Matching routines. */
5514 #ifndef emacs /* Emacs never uses this. */
5515 /* re_match is like re_match_2 except it takes only a single string. */
5518 re_match (bufp
, string
, size
, pos
, regs
)
5519 struct re_pattern_buffer
*bufp
;
5522 struct re_registers
*regs
;
5526 if (MB_CUR_MAX
!= 1)
5527 result
= wcs_re_match_2_internal (bufp
, NULL
, 0, string
, size
,
5529 NULL
, 0, NULL
, 0, NULL
, NULL
);
5532 result
= byte_re_match_2_internal (bufp
, NULL
, 0, string
, size
,
5534 # ifndef REGEX_MALLOC
5542 weak_alias (__re_match
, re_match
)
5544 #endif /* not emacs */
5546 #endif /* not INSIDE_RECURSION */
5548 #ifdef INSIDE_RECURSION
5549 static boolean
PREFIX(group_match_null_string_p
) _RE_ARGS ((UCHAR_T
**p
,
5551 PREFIX(register_info_type
) *reg_info
));
5552 static boolean
PREFIX(alt_match_null_string_p
) _RE_ARGS ((UCHAR_T
*p
,
5554 PREFIX(register_info_type
) *reg_info
));
5555 static boolean
PREFIX(common_op_match_null_string_p
) _RE_ARGS ((UCHAR_T
**p
,
5557 PREFIX(register_info_type
) *reg_info
));
5558 static int PREFIX(bcmp_translate
) _RE_ARGS ((const CHAR_T
*s1
, const CHAR_T
*s2
,
5559 int len
, char *translate
));
5560 #else /* not INSIDE_RECURSION */
5562 /* re_match_2 matches the compiled pattern in BUFP against the
5563 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5564 and SIZE2, respectively). We start matching at POS, and stop
5567 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5568 store offsets for the substring each group matched in REGS. See the
5569 documentation for exactly how many groups we fill.
5571 We return -1 if no match, -2 if an internal error (such as the
5572 failure stack overflowing). Otherwise, we return the length of the
5573 matched substring. */
5576 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
5577 struct re_pattern_buffer
*bufp
;
5578 const char *string1
, *string2
;
5581 struct re_registers
*regs
;
5586 if (MB_CUR_MAX
!= 1)
5587 result
= wcs_re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
5589 NULL
, 0, NULL
, 0, NULL
, NULL
);
5592 result
= byte_re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
5595 #ifndef REGEX_MALLOC
5603 weak_alias (__re_match_2
, re_match_2
)
5606 #endif /* not INSIDE_RECURSION */
5608 #ifdef INSIDE_RECURSION
5611 static int count_mbs_length
PARAMS ((int *, int));
5613 /* This check the substring (from 0, to length) of the multibyte string,
5614 to which offset_buffer correspond. And count how many wchar_t_characters
5615 the substring occupy. We use offset_buffer to optimization.
5616 See convert_mbs_to_wcs. */
5619 count_mbs_length(offset_buffer
, length
)
5625 /* Check whether the size is valid. */
5629 if (offset_buffer
== NULL
)
5632 /* If there are no multibyte character, offset_buffer[i] == i.
5633 Optmize for this case. */
5634 if (offset_buffer
[length
] == length
)
5637 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5643 int middle
= (lower
+ upper
) / 2;
5644 if (middle
== lower
|| middle
== upper
)
5646 if (offset_buffer
[middle
] > length
)
5648 else if (offset_buffer
[middle
] < length
)
5658 /* This is a separate function so that we can force an alloca cleanup
5662 wcs_re_match_2_internal (bufp
, cstring1
, csize1
, cstring2
, csize2
, pos
,
5663 regs
, stop
, string1
, size1
, string2
, size2
,
5664 mbs_offset1
, mbs_offset2
)
5665 struct re_pattern_buffer
*bufp
;
5666 const char *cstring1
, *cstring2
;
5669 struct re_registers
*regs
;
5671 /* string1 == string2 == NULL means string1/2, size1/2 and
5672 mbs_offset1/2 need seting up in this function. */
5673 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5674 wchar_t *string1
, *string2
;
5675 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5677 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5678 int *mbs_offset1
, *mbs_offset2
;
5681 byte_re_match_2_internal (bufp
, string1
, size1
,string2
, size2
, pos
,
5683 struct re_pattern_buffer
*bufp
;
5684 const char *string1
, *string2
;
5687 struct re_registers
*regs
;
5691 /* General temporaries. */
5695 /* They hold whether each wchar_t is binary data or not. */
5696 char *is_binary
= NULL
;
5697 /* If true, we can't free string1/2, mbs_offset1/2. */
5698 int cant_free_wcs_buf
= 1;
5701 /* Just past the end of the corresponding string. */
5702 const CHAR_T
*end1
, *end2
;
5704 /* Pointers into string1 and string2, just past the last characters in
5705 each to consider matching. */
5706 const CHAR_T
*end_match_1
, *end_match_2
;
5708 /* Where we are in the data, and the end of the current string. */
5709 const CHAR_T
*d
, *dend
;
5711 /* Where we are in the pattern, and the end of the pattern. */
5713 UCHAR_T
*pattern
, *p
;
5714 register UCHAR_T
*pend
;
5716 UCHAR_T
*p
= bufp
->buffer
;
5717 register UCHAR_T
*pend
= p
+ bufp
->used
;
5720 /* Mark the opcode just after a start_memory, so we can test for an
5721 empty subpattern when we get to the stop_memory. */
5722 UCHAR_T
*just_past_start_mem
= 0;
5724 /* We use this to map every character in the string. */
5725 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5727 /* Failure point stack. Each place that can handle a failure further
5728 down the line pushes a failure point on this stack. It consists of
5729 restart, regend, and reg_info for all registers corresponding to
5730 the subexpressions we're currently inside, plus the number of such
5731 registers, and, finally, two char *'s. The first char * is where
5732 to resume scanning the pattern; the second one is where to resume
5733 scanning the strings. If the latter is zero, the failure point is
5734 a ``dummy''; if a failure happens and the failure point is a dummy,
5735 it gets discarded and the next next one is tried. */
5736 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5737 PREFIX(fail_stack_type
) fail_stack
;
5740 static unsigned failure_id
;
5741 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5745 /* This holds the pointer to the failure stack, when
5746 it is allocated relocatably. */
5747 fail_stack_elt_t
*failure_stack_ptr
;
5750 /* We fill all the registers internally, independent of what we
5751 return, for use in backreferences. The number here includes
5752 an element for register zero. */
5753 size_t num_regs
= bufp
->re_nsub
+ 1;
5755 /* The currently active registers. */
5756 active_reg_t lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
5757 active_reg_t highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
5759 /* Information on the contents of registers. These are pointers into
5760 the input strings; they record just what was matched (on this
5761 attempt) by a subexpression part of the pattern, that is, the
5762 regnum-th regstart pointer points to where in the pattern we began
5763 matching and the regnum-th regend points to right after where we
5764 stopped matching the regnum-th subexpression. (The zeroth register
5765 keeps track of what the whole pattern matches.) */
5766 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5767 const CHAR_T
**regstart
, **regend
;
5770 /* If a group that's operated upon by a repetition operator fails to
5771 match anything, then the register for its start will need to be
5772 restored because it will have been set to wherever in the string we
5773 are when we last see its open-group operator. Similarly for a
5775 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5776 const CHAR_T
**old_regstart
, **old_regend
;
5779 /* The is_active field of reg_info helps us keep track of which (possibly
5780 nested) subexpressions we are currently in. The matched_something
5781 field of reg_info[reg_num] helps us tell whether or not we have
5782 matched any of the pattern so far this time through the reg_num-th
5783 subexpression. These two fields get reset each time through any
5784 loop their register is in. */
5785 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5786 PREFIX(register_info_type
) *reg_info
;
5789 /* The following record the register info as found in the above
5790 variables when we find a match better than any we've seen before.
5791 This happens as we backtrack through the failure points, which in
5792 turn happens only if we have not yet matched the entire string. */
5793 unsigned best_regs_set
= false;
5794 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5795 const CHAR_T
**best_regstart
, **best_regend
;
5798 /* Logically, this is `best_regend[0]'. But we don't want to have to
5799 allocate space for that if we're not allocating space for anything
5800 else (see below). Also, we never need info about register 0 for
5801 any of the other register vectors, and it seems rather a kludge to
5802 treat `best_regend' differently than the rest. So we keep track of
5803 the end of the best match so far in a separate variable. We
5804 initialize this to NULL so that when we backtrack the first time
5805 and need to test it, it's not garbage. */
5806 const CHAR_T
*match_end
= NULL
;
5808 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5809 int set_regs_matched_done
= 0;
5811 /* Used when we pop values we don't care about. */
5812 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5813 const CHAR_T
**reg_dummy
;
5814 PREFIX(register_info_type
) *reg_info_dummy
;
5818 /* Counts the total number of registers pushed. */
5819 unsigned num_regs_pushed
= 0;
5822 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5826 #ifdef MATCH_MAY_ALLOCATE
5827 /* Do not bother to initialize all the register variables if there are
5828 no groups in the pattern, as it takes a fair amount of time. If
5829 there are groups, we include space for register 0 (the whole
5830 pattern), even though we never use it, since it simplifies the
5831 array indexing. We should fix this. */
5834 regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5835 regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5836 old_regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5837 old_regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5838 best_regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5839 best_regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5840 reg_info
= REGEX_TALLOC (num_regs
, PREFIX(register_info_type
));
5841 reg_dummy
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5842 reg_info_dummy
= REGEX_TALLOC (num_regs
, PREFIX(register_info_type
));
5844 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
5845 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
5853 /* We must initialize all our variables to NULL, so that
5854 `FREE_VARIABLES' doesn't try to free them. */
5855 regstart
= regend
= old_regstart
= old_regend
= best_regstart
5856 = best_regend
= reg_dummy
= NULL
;
5857 reg_info
= reg_info_dummy
= (PREFIX(register_info_type
) *) NULL
;
5859 #endif /* MATCH_MAY_ALLOCATE */
5861 /* The starting position is bogus. */
5863 if (pos
< 0 || pos
> csize1
+ csize2
)
5865 if (pos
< 0 || pos
> size1
+ size2
)
5873 /* Allocate wchar_t array for string1 and string2 and
5874 fill them with converted string. */
5875 if (string1
== NULL
&& string2
== NULL
)
5877 /* We need seting up buffers here. */
5879 /* We must free wcs buffers in this function. */
5880 cant_free_wcs_buf
= 0;
5884 string1
= REGEX_TALLOC (csize1
+ 1, CHAR_T
);
5885 mbs_offset1
= REGEX_TALLOC (csize1
+ 1, int);
5886 is_binary
= REGEX_TALLOC (csize1
+ 1, char);
5887 if (!string1
|| !mbs_offset1
|| !is_binary
)
5890 FREE_VAR (mbs_offset1
);
5891 FREE_VAR (is_binary
);
5897 string2
= REGEX_TALLOC (csize2
+ 1, CHAR_T
);
5898 mbs_offset2
= REGEX_TALLOC (csize2
+ 1, int);
5899 is_binary
= REGEX_TALLOC (csize2
+ 1, char);
5900 if (!string2
|| !mbs_offset2
|| !is_binary
)
5903 FREE_VAR (mbs_offset1
);
5905 FREE_VAR (mbs_offset2
);
5906 FREE_VAR (is_binary
);
5909 size2
= convert_mbs_to_wcs(string2
, cstring2
, csize2
,
5910 mbs_offset2
, is_binary
);
5911 string2
[size2
] = L
'\0'; /* for a sentinel */
5912 FREE_VAR (is_binary
);
5916 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5917 pattern to (char*) in regex_compile. */
5918 p
= pattern
= (CHAR_T
*)bufp
->buffer
;
5919 pend
= (CHAR_T
*)(bufp
->buffer
+ bufp
->used
);
5923 /* Initialize subexpression text positions to -1 to mark ones that no
5924 start_memory/stop_memory has been seen for. Also initialize the
5925 register information struct. */
5926 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
5928 regstart
[mcnt
] = regend
[mcnt
]
5929 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
5931 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
5932 IS_ACTIVE (reg_info
[mcnt
]) = 0;
5933 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
5934 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
5937 /* We move `string1' into `string2' if the latter's empty -- but not if
5938 `string1' is null. */
5939 if (size2
== 0 && string1
!= NULL
)
5946 mbs_offset2
= mbs_offset1
;
5952 end1
= string1
+ size1
;
5953 end2
= string2
+ size2
;
5955 /* Compute where to stop matching, within the two strings. */
5959 mcnt
= count_mbs_length(mbs_offset1
, stop
);
5960 end_match_1
= string1
+ mcnt
;
5961 end_match_2
= string2
;
5965 if (stop
> csize1
+ csize2
)
5966 stop
= csize1
+ csize2
;
5968 mcnt
= count_mbs_length(mbs_offset2
, stop
-csize1
);
5969 end_match_2
= string2
+ mcnt
;
5972 { /* count_mbs_length return error. */
5979 end_match_1
= string1
+ stop
;
5980 end_match_2
= string2
;
5985 end_match_2
= string2
+ stop
- size1
;
5989 /* `p' scans through the pattern as `d' scans through the data.
5990 `dend' is the end of the input string that `d' points within. `d'
5991 is advanced into the following input string whenever necessary, but
5992 this happens before fetching; therefore, at the beginning of the
5993 loop, `d' can be pointing at the end of a string, but it cannot
5996 if (size1
> 0 && pos
<= csize1
)
5998 mcnt
= count_mbs_length(mbs_offset1
, pos
);
6004 mcnt
= count_mbs_length(mbs_offset2
, pos
-csize1
);
6010 { /* count_mbs_length return error. */
6015 if (size1
> 0 && pos
<= size1
)
6022 d
= string2
+ pos
- size1
;
6027 DEBUG_PRINT1 ("The compiled pattern is:\n");
6028 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
6029 DEBUG_PRINT1 ("The string to match is: `");
6030 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
6031 DEBUG_PRINT1 ("'\n");
6033 /* This loops over pattern commands. It exits by returning from the
6034 function if the match is complete, or it drops through if the match
6035 fails at this starting point in the input data. */
6039 DEBUG_PRINT2 ("\n%p: ", p
);
6041 DEBUG_PRINT2 ("\n0x%x: ", p
);
6045 { /* End of pattern means we might have succeeded. */
6046 DEBUG_PRINT1 ("end of pattern ... ");
6048 /* If we haven't matched the entire string, and we want the
6049 longest match, try backtracking. */
6050 if (d
!= end_match_2
)
6052 /* 1 if this match ends in the same string (string1 or string2)
6053 as the best previous match. */
6054 boolean same_str_p
= (FIRST_STRING_P (match_end
)
6055 == MATCHING_IN_FIRST_STRING
);
6056 /* 1 if this match is the best seen so far. */
6057 boolean best_match_p
;
6059 /* AIX compiler got confused when this was combined
6060 with the previous declaration. */
6062 best_match_p
= d
> match_end
;
6064 best_match_p
= !MATCHING_IN_FIRST_STRING
;
6066 DEBUG_PRINT1 ("backtracking.\n");
6068 if (!FAIL_STACK_EMPTY ())
6069 { /* More failure points to try. */
6071 /* If exceeds best match so far, save it. */
6072 if (!best_regs_set
|| best_match_p
)
6074 best_regs_set
= true;
6077 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6079 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
6081 best_regstart
[mcnt
] = regstart
[mcnt
];
6082 best_regend
[mcnt
] = regend
[mcnt
];
6088 /* If no failure points, don't restore garbage. And if
6089 last match is real best match, don't restore second
6091 else if (best_regs_set
&& !best_match_p
)
6094 /* Restore best match. It may happen that `dend ==
6095 end_match_1' while the restored d is in string2.
6096 For example, the pattern `x.*y.*z' against the
6097 strings `x-' and `y-z-', if the two strings are
6098 not consecutive in memory. */
6099 DEBUG_PRINT1 ("Restoring best registers.\n");
6102 dend
= ((d
>= string1
&& d
<= end1
)
6103 ? end_match_1
: end_match_2
);
6105 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
6107 regstart
[mcnt
] = best_regstart
[mcnt
];
6108 regend
[mcnt
] = best_regend
[mcnt
];
6111 } /* d != end_match_2 */
6114 DEBUG_PRINT1 ("Accepting match.\n");
6115 /* If caller wants register contents data back, do it. */
6116 if (regs
&& !bufp
->no_sub
)
6118 /* Have the register data arrays been allocated? */
6119 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
6120 { /* No. So allocate them with malloc. We need one
6121 extra element beyond `num_regs' for the `-1' marker
6123 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
6124 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
6125 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
6126 if (regs
->start
== NULL
|| regs
->end
== NULL
)
6131 bufp
->regs_allocated
= REGS_REALLOCATE
;
6133 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
6134 { /* Yes. If we need more elements than were already
6135 allocated, reallocate them. If we need fewer, just
6137 if (regs
->num_regs
< num_regs
+ 1)
6139 regs
->num_regs
= num_regs
+ 1;
6140 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
6141 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
6142 if (regs
->start
== NULL
|| regs
->end
== NULL
)
6151 /* These braces fend off a "empty body in an else-statement"
6152 warning under GCC when assert expands to nothing. */
6153 assert (bufp
->regs_allocated
== REGS_FIXED
);
6156 /* Convert the pointer data in `regstart' and `regend' to
6157 indices. Register zero has to be set differently,
6158 since we haven't kept track of any info for it. */
6159 if (regs
->num_regs
> 0)
6161 regs
->start
[0] = pos
;
6163 if (MATCHING_IN_FIRST_STRING
)
6164 regs
->end
[0] = mbs_offset1
!= NULL
?
6165 mbs_offset1
[d
-string1
] : 0;
6167 regs
->end
[0] = csize1
+ (mbs_offset2
!= NULL
?
6168 mbs_offset2
[d
-string2
] : 0);
6170 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
6171 ? ((regoff_t
) (d
- string1
))
6172 : ((regoff_t
) (d
- string2
+ size1
)));
6176 /* Go through the first `min (num_regs, regs->num_regs)'
6177 registers, since that is all we initialized. */
6178 for (mcnt
= 1; (unsigned) mcnt
< MIN (num_regs
, regs
->num_regs
);
6181 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
6182 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
6186 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
6188 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
6192 /* If the regs structure we return has more elements than
6193 were in the pattern, set the extra elements to -1. If
6194 we (re)allocated the registers, this is the case,
6195 because we always allocate enough to have at least one
6197 for (mcnt
= num_regs
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
6198 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
6199 } /* regs && !bufp->no_sub */
6201 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6202 nfailure_points_pushed
, nfailure_points_popped
,
6203 nfailure_points_pushed
- nfailure_points_popped
);
6204 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
6207 if (MATCHING_IN_FIRST_STRING
)
6208 mcnt
= mbs_offset1
!= NULL
? mbs_offset1
[d
-string1
] : 0;
6210 mcnt
= (mbs_offset2
!= NULL
? mbs_offset2
[d
-string2
] : 0) +
6214 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
6219 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
6225 /* Otherwise match next pattern command. */
6226 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
6228 /* Ignore these. Used to ignore the n of succeed_n's which
6229 currently have n == 0. */
6231 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6235 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6238 /* Match the next n pattern characters exactly. The following
6239 byte in the pattern defines n, and the n bytes after that
6240 are the characters to match. */
6246 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
6248 /* This is written out as an if-else so we don't waste time
6249 testing `translate' inside the loop. */
6258 if ((UCHAR_T
) translate
[(unsigned char) *d
++]
6264 if (*d
++ != (CHAR_T
) *p
++)
6268 if ((UCHAR_T
) translate
[(unsigned char) *d
++]
6280 if (*d
++ != (CHAR_T
) *p
++) goto fail
;
6284 SET_REGS_MATCHED ();
6288 /* Match any character except possibly a newline or a null. */
6290 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6294 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
6295 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
6298 SET_REGS_MATCHED ();
6299 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d
);
6309 unsigned int i
, char_class_length
, coll_symbol_length
,
6310 equiv_class_length
, ranges_length
, chars_length
, length
;
6311 CHAR_T
*workp
, *workp2
, *charset_top
;
6312 #define WORK_BUFFER_SIZE 128
6313 CHAR_T str_buf
[WORK_BUFFER_SIZE
];
6318 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
6320 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6322 c
= TRANSLATE (*d
); /* The character to match. */
6325 nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
6327 charset_top
= p
- 1;
6328 char_class_length
= *p
++;
6329 coll_symbol_length
= *p
++;
6330 equiv_class_length
= *p
++;
6331 ranges_length
= *p
++;
6332 chars_length
= *p
++;
6333 /* p points charset[6], so the address of the next instruction
6334 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6335 where l=length of char_classes, m=length of collating_symbol,
6336 n=equivalence_class, o=length of char_range,
6337 p'=length of character. */
6339 /* Update p to indicate the next instruction. */
6340 p
+= char_class_length
+ coll_symbol_length
+ equiv_class_length
+
6341 2*ranges_length
+ chars_length
;
6343 /* match with char_class? */
6344 for (i
= 0; i
< char_class_length
; i
+= CHAR_CLASS_SIZE
)
6347 uintptr_t alignedp
= ((uintptr_t)workp
6348 + __alignof__(wctype_t) - 1)
6349 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6350 wctype
= *((wctype_t*)alignedp
);
6351 workp
+= CHAR_CLASS_SIZE
;
6353 if (__iswctype((wint_t)c
, wctype
))
6354 goto char_set_matched
;
6356 if (iswctype((wint_t)c
, wctype
))
6357 goto char_set_matched
;
6361 /* match with collating_symbol? */
6365 const unsigned char *extra
= (const unsigned char *)
6366 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
6368 for (workp2
= workp
+ coll_symbol_length
; workp
< workp2
;
6372 wextra
= (int32_t*)(extra
+ *workp
++);
6373 for (i
= 0; i
< *wextra
; ++i
)
6374 if (TRANSLATE(d
[i
]) != wextra
[1 + i
])
6379 /* Update d, however d will be incremented at
6380 char_set_matched:, we decrement d here. */
6382 goto char_set_matched
;
6386 else /* (nrules == 0) */
6388 /* If we can't look up collation data, we use wcscoll
6391 for (workp2
= workp
+ coll_symbol_length
; workp
< workp2
;)
6393 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6395 length
= __wcslen (workp
);
6397 length
= wcslen (workp
);
6400 /* If wcscoll(the collating symbol, whole string) > 0,
6401 any substring of the string never match with the
6402 collating symbol. */
6404 if (__wcscoll (workp
, d
) > 0)
6406 if (wcscoll (workp
, d
) > 0)
6409 workp
+= length
+ 1;
6413 /* First, we compare the collating symbol with
6414 the first character of the string.
6415 If it don't match, we add the next character to
6416 the compare buffer in turn. */
6417 for (i
= 0 ; i
< WORK_BUFFER_SIZE
-1 ; i
++, d
++)
6422 if (dend
== end_match_2
)
6428 /* add next character to the compare buffer. */
6429 str_buf
[i
] = TRANSLATE(*d
);
6430 str_buf
[i
+1] = '\0';
6433 match
= __wcscoll (workp
, str_buf
);
6435 match
= wcscoll (workp
, str_buf
);
6438 goto char_set_matched
;
6441 /* (str_buf > workp) indicate (str_buf + X > workp),
6442 because for all X (str_buf + X > str_buf).
6443 So we don't need continue this loop. */
6446 /* Otherwise(str_buf < workp),
6447 (str_buf+next_character) may equals (workp).
6448 So we continue this loop. */
6453 workp
+= length
+ 1;
6456 /* match with equivalence_class? */
6460 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6461 /* Try to match the equivalence class against
6462 those known to the collate implementation. */
6463 const int32_t *table
;
6464 const int32_t *weights
;
6465 const int32_t *extra
;
6466 const int32_t *indirect
;
6471 /* This #include defines a local function! */
6472 # include <locale/weightwc.h>
6474 table
= (const int32_t *)
6475 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEWC
);
6476 weights
= (const wint_t *)
6477 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTWC
);
6478 extra
= (const wint_t *)
6479 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAWC
);
6480 indirect
= (const int32_t *)
6481 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTWC
);
6483 /* Write 1 collating element to str_buf, and
6487 for (i
= 0 ; idx2
== 0 && i
< WORK_BUFFER_SIZE
- 1; i
++)
6489 cp
= (wint_t*)str_buf
;
6492 if (dend
== end_match_2
)
6497 str_buf
[i
] = TRANSLATE(*(d
+i
));
6498 str_buf
[i
+1] = '\0'; /* sentinel */
6499 idx2
= findidx ((const wint_t**)&cp
);
6502 /* Update d, however d will be incremented at
6503 char_set_matched:, we decrement d here. */
6504 d
= backup_d
+ ((wchar_t*)cp
- (wchar_t*)str_buf
- 1);
6507 if (dend
== end_match_2
)
6516 len
= weights
[idx2
];
6518 for (workp2
= workp
+ equiv_class_length
; workp
< workp2
;
6521 idx
= (int32_t)*workp
;
6522 /* We already checked idx != 0 in regex_compile. */
6524 if (idx2
!= 0 && len
== weights
[idx
])
6527 while (cnt
< len
&& (weights
[idx
+ 1 + cnt
]
6528 == weights
[idx2
+ 1 + cnt
]))
6532 goto char_set_matched
;
6539 else /* (nrules == 0) */
6541 /* If we can't look up collation data, we use wcscoll
6544 for (workp2
= workp
+ equiv_class_length
; workp
< workp2
;)
6546 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6548 length
= __wcslen (workp
);
6550 length
= wcslen (workp
);
6553 /* If wcscoll(the collating symbol, whole string) > 0,
6554 any substring of the string never match with the
6555 collating symbol. */
6557 if (__wcscoll (workp
, d
) > 0)
6559 if (wcscoll (workp
, d
) > 0)
6562 workp
+= length
+ 1;
6566 /* First, we compare the equivalence class with
6567 the first character of the string.
6568 If it don't match, we add the next character to
6569 the compare buffer in turn. */
6570 for (i
= 0 ; i
< WORK_BUFFER_SIZE
- 1 ; i
++, d
++)
6575 if (dend
== end_match_2
)
6581 /* add next character to the compare buffer. */
6582 str_buf
[i
] = TRANSLATE(*d
);
6583 str_buf
[i
+1] = '\0';
6586 match
= __wcscoll (workp
, str_buf
);
6588 match
= wcscoll (workp
, str_buf
);
6592 goto char_set_matched
;
6595 /* (str_buf > workp) indicate (str_buf + X > workp),
6596 because for all X (str_buf + X > str_buf).
6597 So we don't need continue this loop. */
6600 /* Otherwise(str_buf < workp),
6601 (str_buf+next_character) may equals (workp).
6602 So we continue this loop. */
6607 workp
+= length
+ 1;
6611 /* match with char_range? */
6615 uint32_t collseqval
;
6616 const char *collseq
= (const char *)
6617 _NL_CURRENT(LC_COLLATE
, _NL_COLLATE_COLLSEQWC
);
6619 collseqval
= collseq_table_lookup (collseq
, c
);
6621 for (; workp
< p
- chars_length
;)
6623 uint32_t start_val
, end_val
;
6625 /* We already compute the collation sequence value
6626 of the characters (or collating symbols). */
6627 start_val
= (uint32_t) *workp
++; /* range_start */
6628 end_val
= (uint32_t) *workp
++; /* range_end */
6630 if (start_val
<= collseqval
&& collseqval
<= end_val
)
6631 goto char_set_matched
;
6637 /* We set range_start_char at str_buf[0], range_end_char
6638 at str_buf[4], and compared char at str_buf[2]. */
6643 for (; workp
< p
- chars_length
;)
6645 wchar_t *range_start_char
, *range_end_char
;
6647 /* match if (range_start_char <= c <= range_end_char). */
6649 /* If range_start(or end) < 0, we assume -range_start(end)
6650 is the offset of the collating symbol which is specified
6651 as the character of the range start(end). */
6655 range_start_char
= charset_top
- (*workp
++);
6658 str_buf
[0] = *workp
++;
6659 range_start_char
= str_buf
;
6664 range_end_char
= charset_top
- (*workp
++);
6667 str_buf
[4] = *workp
++;
6668 range_end_char
= str_buf
+ 4;
6672 if (__wcscoll (range_start_char
, str_buf
+2) <= 0
6673 && __wcscoll (str_buf
+2, range_end_char
) <= 0)
6675 if (wcscoll (range_start_char
, str_buf
+2) <= 0
6676 && wcscoll (str_buf
+2, range_end_char
) <= 0)
6678 goto char_set_matched
;
6682 /* match with char? */
6683 for (; workp
< p
; workp
++)
6685 goto char_set_matched
;
6692 /* Cast to `unsigned' instead of `unsigned char' in case the
6693 bit list is a full 32 bytes long. */
6694 if (c
< (unsigned) (*p
* BYTEWIDTH
)
6695 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
6700 if (!not) goto fail
;
6701 #undef WORK_BUFFER_SIZE
6703 SET_REGS_MATCHED ();
6709 /* The beginning of a group is represented by start_memory.
6710 The arguments are the register number in the next byte, and the
6711 number of groups inner to this one in the next. The text
6712 matched within the group is recorded (in the internal
6713 registers data structure) under the register number. */
6715 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6716 (long int) *p
, (long int) p
[1]);
6718 /* Find out if this group can match the empty string. */
6719 p1
= p
; /* To send to group_match_null_string_p. */
6721 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
6722 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6723 = PREFIX(group_match_null_string_p
) (&p1
, pend
, reg_info
);
6725 /* Save the position in the string where we were the last time
6726 we were at this open-group operator in case the group is
6727 operated upon by a repetition operator, e.g., with `(a*)*b'
6728 against `ab'; then we want to ignore where we are now in
6729 the string in case this attempt to match fails. */
6730 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6731 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
6733 DEBUG_PRINT2 (" old_regstart: %d\n",
6734 POINTER_TO_OFFSET (old_regstart
[*p
]));
6737 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
6739 IS_ACTIVE (reg_info
[*p
]) = 1;
6740 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
6742 /* Clear this whenever we change the register activity status. */
6743 set_regs_matched_done
= 0;
6745 /* This is the new highest active register. */
6746 highest_active_reg
= *p
;
6748 /* If nothing was active before, this is the new lowest active
6750 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
6751 lowest_active_reg
= *p
;
6753 /* Move past the register number and inner group count. */
6755 just_past_start_mem
= p
;
6760 /* The stop_memory opcode represents the end of a group. Its
6761 arguments are the same as start_memory's: the register
6762 number, and the number of inner groups. */
6764 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6765 (long int) *p
, (long int) p
[1]);
6767 /* We need to save the string position the last time we were at
6768 this close-group operator in case the group is operated
6769 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6770 against `aba'; then we want to ignore where we are now in
6771 the string in case this attempt to match fails. */
6772 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6773 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
6775 DEBUG_PRINT2 (" old_regend: %d\n",
6776 POINTER_TO_OFFSET (old_regend
[*p
]));
6779 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
6781 /* This register isn't active anymore. */
6782 IS_ACTIVE (reg_info
[*p
]) = 0;
6784 /* Clear this whenever we change the register activity status. */
6785 set_regs_matched_done
= 0;
6787 /* If this was the only register active, nothing is active
6789 if (lowest_active_reg
== highest_active_reg
)
6791 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
6792 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
6795 { /* We must scan for the new highest active register, since
6796 it isn't necessarily one less than now: consider
6797 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6798 new highest active register is 1. */
6800 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
6803 /* If we end up at register zero, that means that we saved
6804 the registers as the result of an `on_failure_jump', not
6805 a `start_memory', and we jumped to past the innermost
6806 `stop_memory'. For example, in ((.)*) we save
6807 registers 1 and 2 as a result of the *, but when we pop
6808 back to the second ), we are at the stop_memory 1.
6809 Thus, nothing is active. */
6812 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
6813 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
6816 highest_active_reg
= r
;
6819 /* If just failed to match something this time around with a
6820 group that's operated on by a repetition operator, try to
6821 force exit from the ``loop'', and restore the register
6822 information for this group that we had before trying this
6824 if ((!MATCHED_SOMETHING (reg_info
[*p
])
6825 || just_past_start_mem
== p
- 1)
6828 boolean is_a_jump_n
= false;
6832 switch ((re_opcode_t
) *p1
++)
6836 case pop_failure_jump
:
6837 case maybe_pop_jump
:
6839 case dummy_failure_jump
:
6840 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
6842 p1
+= OFFSET_ADDRESS_SIZE
;
6850 /* If the next operation is a jump backwards in the pattern
6851 to an on_failure_jump right before the start_memory
6852 corresponding to this stop_memory, exit from the loop
6853 by forcing a failure after pushing on the stack the
6854 on_failure_jump's jump in the pattern, and d. */
6855 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
6856 && (re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == start_memory
6857 && p1
[2+OFFSET_ADDRESS_SIZE
] == *p
)
6859 /* If this group ever matched anything, then restore
6860 what its registers were before trying this last
6861 failed match, e.g., with `(a*)*b' against `ab' for
6862 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6863 against `aba' for regend[3].
6865 Also restore the registers for inner groups for,
6866 e.g., `((a*)(b*))*' against `aba' (register 3 would
6867 otherwise get trashed). */
6869 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
6873 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
6875 /* Restore this and inner groups' (if any) registers. */
6876 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
6879 regstart
[r
] = old_regstart
[r
];
6881 /* xx why this test? */
6882 if (old_regend
[r
] >= regstart
[r
])
6883 regend
[r
] = old_regend
[r
];
6887 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
6888 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
6894 /* Move past the register number and the inner group count. */
6899 /* \<digit> has been turned into a `duplicate' command which is
6900 followed by the numeric value of <digit> as the register number. */
6903 register const CHAR_T
*d2
, *dend2
;
6904 int regno
= *p
++; /* Get which register to match against. */
6905 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
6907 /* Can't back reference a group which we've never matched. */
6908 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
6911 /* Where in input to try to start matching. */
6912 d2
= regstart
[regno
];
6914 /* Where to stop matching; if both the place to start and
6915 the place to stop matching are in the same string, then
6916 set to the place to stop, otherwise, for now have to use
6917 the end of the first string. */
6919 dend2
= ((FIRST_STRING_P (regstart
[regno
])
6920 == FIRST_STRING_P (regend
[regno
]))
6921 ? regend
[regno
] : end_match_1
);
6924 /* If necessary, advance to next segment in register
6928 if (dend2
== end_match_2
) break;
6929 if (dend2
== regend
[regno
]) break;
6931 /* End of string1 => advance to string2. */
6933 dend2
= regend
[regno
];
6935 /* At end of register contents => success */
6936 if (d2
== dend2
) break;
6938 /* If necessary, advance to next segment in data. */
6941 /* How many characters left in this segment to match. */
6944 /* Want how many consecutive characters we can match in
6945 one shot, so, if necessary, adjust the count. */
6946 if (mcnt
> dend2
- d2
)
6949 /* Compare that many; failure if mismatch, else move
6952 ? PREFIX(bcmp_translate
) (d
, d2
, mcnt
, translate
)
6953 : memcmp (d
, d2
, mcnt
*sizeof(UCHAR_T
)))
6955 d
+= mcnt
, d2
+= mcnt
;
6957 /* Do this because we've match some characters. */
6958 SET_REGS_MATCHED ();
6964 /* begline matches the empty string at the beginning of the string
6965 (unless `not_bol' is set in `bufp'), and, if
6966 `newline_anchor' is set, after newlines. */
6968 DEBUG_PRINT1 ("EXECUTING begline.\n");
6970 if (AT_STRINGS_BEG (d
))
6972 if (!bufp
->not_bol
) break;
6974 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
6978 /* In all other cases, we fail. */
6982 /* endline is the dual of begline. */
6984 DEBUG_PRINT1 ("EXECUTING endline.\n");
6986 if (AT_STRINGS_END (d
))
6988 if (!bufp
->not_eol
) break;
6991 /* We have to ``prefetch'' the next character. */
6992 else if ((d
== end1
? *string2
: *d
) == '\n'
6993 && bufp
->newline_anchor
)
7000 /* Match at the very beginning of the data. */
7002 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
7003 if (AT_STRINGS_BEG (d
))
7008 /* Match at the very end of the data. */
7010 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
7011 if (AT_STRINGS_END (d
))
7016 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
7017 pushes NULL as the value for the string on the stack. Then
7018 `pop_failure_point' will keep the current value for the
7019 string, instead of restoring it. To see why, consider
7020 matching `foo\nbar' against `.*\n'. The .* matches the foo;
7021 then the . fails against the \n. But the next thing we want
7022 to do is match the \n against the \n; if we restored the
7023 string value, we would be back at the foo.
7025 Because this is used only in specific cases, we don't need to
7026 check all the things that `on_failure_jump' does, to make
7027 sure the right things get saved on the stack. Hence we don't
7028 share its code. The only reason to push anything on the
7029 stack at all is that otherwise we would have to change
7030 `anychar's code to do something besides goto fail in this
7031 case; that seems worse than this. */
7032 case on_failure_keep_string_jump
:
7033 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
7035 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7037 DEBUG_PRINT3 (" %d (to %p):\n", mcnt
, p
+ mcnt
);
7039 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
7042 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
7046 /* Uses of on_failure_jump:
7048 Each alternative starts with an on_failure_jump that points
7049 to the beginning of the next alternative. Each alternative
7050 except the last ends with a jump that in effect jumps past
7051 the rest of the alternatives. (They really jump to the
7052 ending jump of the following alternative, because tensioning
7053 these jumps is a hassle.)
7055 Repeats start with an on_failure_jump that points past both
7056 the repetition text and either the following jump or
7057 pop_failure_jump back to this on_failure_jump. */
7058 case on_failure_jump
:
7060 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
7062 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7064 DEBUG_PRINT3 (" %d (to %p)", mcnt
, p
+ mcnt
);
7066 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
7069 /* If this on_failure_jump comes right before a group (i.e.,
7070 the original * applied to a group), save the information
7071 for that group and all inner ones, so that if we fail back
7072 to this point, the group's information will be correct.
7073 For example, in \(a*\)*\1, we need the preceding group,
7074 and in \(zz\(a*\)b*\)\2, we need the inner group. */
7076 /* We can't use `p' to check ahead because we push
7077 a failure point to `p + mcnt' after we do this. */
7080 /* We need to skip no_op's before we look for the
7081 start_memory in case this on_failure_jump is happening as
7082 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7084 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
7087 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
7089 /* We have a new highest active register now. This will
7090 get reset at the start_memory we are about to get to,
7091 but we will have saved all the registers relevant to
7092 this repetition op, as described above. */
7093 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
7094 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
7095 lowest_active_reg
= *(p1
+ 1);
7098 DEBUG_PRINT1 (":\n");
7099 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
7103 /* A smart repeat ends with `maybe_pop_jump'.
7104 We change it to either `pop_failure_jump' or `jump'. */
7105 case maybe_pop_jump
:
7106 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7107 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
7109 register UCHAR_T
*p2
= p
;
7111 /* Compare the beginning of the repeat with what in the
7112 pattern follows its end. If we can establish that there
7113 is nothing that they would both match, i.e., that we
7114 would have to backtrack because of (as in, e.g., `a*a')
7115 then we can change to pop_failure_jump, because we'll
7116 never have to backtrack.
7118 This is not true in the case of alternatives: in
7119 `(a|ab)*' we do need to backtrack to the `ab' alternative
7120 (e.g., if the string was `ab'). But instead of trying to
7121 detect that here, the alternative has put on a dummy
7122 failure point which is what we will end up popping. */
7124 /* Skip over open/close-group commands.
7125 If what follows this loop is a ...+ construct,
7126 look at what begins its body, since we will have to
7127 match at least one of that. */
7131 && ((re_opcode_t
) *p2
== stop_memory
7132 || (re_opcode_t
) *p2
== start_memory
))
7134 else if (p2
+ 2 + 2 * OFFSET_ADDRESS_SIZE
< pend
7135 && (re_opcode_t
) *p2
== dummy_failure_jump
)
7136 p2
+= 2 + 2 * OFFSET_ADDRESS_SIZE
;
7142 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7143 to the `maybe_finalize_jump' of this case. Examine what
7146 /* If we're at the end of the pattern, we can change. */
7149 /* Consider what happens when matching ":\(.*\)"
7150 against ":/". I don't really understand this code
7152 p
[-(1+OFFSET_ADDRESS_SIZE
)] = (UCHAR_T
)
7155 (" End of pattern: change to `pop_failure_jump'.\n");
7158 else if ((re_opcode_t
) *p2
== exactn
7160 || (re_opcode_t
) *p2
== exactn_bin
7162 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
7165 = *p2
== (UCHAR_T
) endline
? '\n' : p2
[2];
7167 if (((re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == exactn
7169 || (re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == exactn_bin
7171 ) && p1
[3+OFFSET_ADDRESS_SIZE
] != c
)
7173 p
[-(1+OFFSET_ADDRESS_SIZE
)] = (UCHAR_T
)
7176 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7178 (wint_t) p1
[3+OFFSET_ADDRESS_SIZE
]);
7180 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7182 (char) p1
[3+OFFSET_ADDRESS_SIZE
]);
7187 else if ((re_opcode_t
) p1
[3] == charset
7188 || (re_opcode_t
) p1
[3] == charset_not
)
7190 int not = (re_opcode_t
) p1
[3] == charset_not
;
7192 if (c
< (unsigned) (p1
[4] * BYTEWIDTH
)
7193 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
7196 /* `not' is equal to 1 if c would match, which means
7197 that we can't change to pop_failure_jump. */
7200 p
[-3] = (unsigned char) pop_failure_jump
;
7201 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7204 #endif /* not WCHAR */
7207 else if ((re_opcode_t
) *p2
== charset
)
7209 /* We win if the first character of the loop is not part
7211 if ((re_opcode_t
) p1
[3] == exactn
7212 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
7213 && (p2
[2 + p1
[5] / BYTEWIDTH
]
7214 & (1 << (p1
[5] % BYTEWIDTH
)))))
7216 p
[-3] = (unsigned char) pop_failure_jump
;
7217 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7220 else if ((re_opcode_t
) p1
[3] == charset_not
)
7223 /* We win if the charset_not inside the loop
7224 lists every character listed in the charset after. */
7225 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
7226 if (! (p2
[2 + idx
] == 0
7227 || (idx
< (int) p1
[4]
7228 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
7233 p
[-3] = (unsigned char) pop_failure_jump
;
7234 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7237 else if ((re_opcode_t
) p1
[3] == charset
)
7240 /* We win if the charset inside the loop
7241 has no overlap with the one after the loop. */
7243 idx
< (int) p2
[1] && idx
< (int) p1
[4];
7245 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
7248 if (idx
== p2
[1] || idx
== p1
[4])
7250 p
[-3] = (unsigned char) pop_failure_jump
;
7251 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7255 #endif /* not WCHAR */
7257 p
-= OFFSET_ADDRESS_SIZE
; /* Point at relative address again. */
7258 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
7260 p
[-1] = (UCHAR_T
) jump
;
7261 DEBUG_PRINT1 (" Match => jump.\n");
7262 goto unconditional_jump
;
7264 /* Note fall through. */
7267 /* The end of a simple repeat has a pop_failure_jump back to
7268 its matching on_failure_jump, where the latter will push a
7269 failure point. The pop_failure_jump takes off failure
7270 points put on by this pop_failure_jump's matching
7271 on_failure_jump; we got through the pattern to here from the
7272 matching on_failure_jump, so didn't fail. */
7273 case pop_failure_jump
:
7275 /* We need to pass separate storage for the lowest and
7276 highest registers, even though we don't care about the
7277 actual values. Otherwise, we will restore only one
7278 register from the stack, since lowest will == highest in
7279 `pop_failure_point'. */
7280 active_reg_t dummy_low_reg
, dummy_high_reg
;
7281 UCHAR_T
*pdummy
= NULL
;
7282 const CHAR_T
*sdummy
= NULL
;
7284 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7285 POP_FAILURE_POINT (sdummy
, pdummy
,
7286 dummy_low_reg
, dummy_high_reg
,
7287 reg_dummy
, reg_dummy
, reg_info_dummy
);
7289 /* Note fall through. */
7293 DEBUG_PRINT2 ("\n%p: ", p
);
7295 DEBUG_PRINT2 ("\n0x%x: ", p
);
7297 /* Note fall through. */
7299 /* Unconditionally jump (without popping any failure points). */
7301 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
7302 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
7303 p
+= mcnt
; /* Do the jump. */
7305 DEBUG_PRINT2 ("(to %p).\n", p
);
7307 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
7312 /* We need this opcode so we can detect where alternatives end
7313 in `group_match_null_string_p' et al. */
7315 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7316 goto unconditional_jump
;
7319 /* Normally, the on_failure_jump pushes a failure point, which
7320 then gets popped at pop_failure_jump. We will end up at
7321 pop_failure_jump, also, and with a pattern of, say, `a+', we
7322 are skipping over the on_failure_jump, so we have to push
7323 something meaningless for pop_failure_jump to pop. */
7324 case dummy_failure_jump
:
7325 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7326 /* It doesn't matter what we push for the string here. What
7327 the code at `fail' tests is the value for the pattern. */
7328 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
7329 goto unconditional_jump
;
7332 /* At the end of an alternative, we need to push a dummy failure
7333 point in case we are followed by a `pop_failure_jump', because
7334 we don't want the failure point for the alternative to be
7335 popped. For example, matching `(a|ab)*' against `aab'
7336 requires that we match the `ab' alternative. */
7337 case push_dummy_failure
:
7338 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7339 /* See comments just above at `dummy_failure_jump' about the
7341 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
7344 /* Have to succeed matching what follows at least n times.
7345 After that, handle like `on_failure_jump'. */
7347 EXTRACT_NUMBER (mcnt
, p
+ OFFSET_ADDRESS_SIZE
);
7348 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
7351 /* Originally, this is how many times we HAVE to succeed. */
7355 p
+= OFFSET_ADDRESS_SIZE
;
7356 STORE_NUMBER_AND_INCR (p
, mcnt
);
7358 DEBUG_PRINT3 (" Setting %p to %d.\n", p
- OFFSET_ADDRESS_SIZE
7361 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
- OFFSET_ADDRESS_SIZE
7368 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7369 p
+ OFFSET_ADDRESS_SIZE
);
7371 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7372 p
+ OFFSET_ADDRESS_SIZE
);
7376 p
[1] = (UCHAR_T
) no_op
;
7378 p
[2] = (UCHAR_T
) no_op
;
7379 p
[3] = (UCHAR_T
) no_op
;
7386 EXTRACT_NUMBER (mcnt
, p
+ OFFSET_ADDRESS_SIZE
);
7387 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
7389 /* Originally, this is how many times we CAN jump. */
7393 STORE_NUMBER (p
+ OFFSET_ADDRESS_SIZE
, mcnt
);
7396 DEBUG_PRINT3 (" Setting %p to %d.\n", p
+ OFFSET_ADDRESS_SIZE
,
7399 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
+ OFFSET_ADDRESS_SIZE
,
7402 goto unconditional_jump
;
7404 /* If don't have to jump any more, skip over the rest of command. */
7406 p
+= 2 * OFFSET_ADDRESS_SIZE
;
7411 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7413 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7415 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7417 DEBUG_PRINT3 (" Setting %p to %d.\n", p1
, mcnt
);
7419 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
7421 STORE_NUMBER (p1
, mcnt
);
7426 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7427 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7428 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7429 macro and introducing temporary variables works around the bug. */
7432 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7433 if (AT_WORD_BOUNDARY (d
))
7438 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7439 if (AT_WORD_BOUNDARY (d
))
7445 boolean prevchar
, thischar
;
7447 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7448 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
7451 prevchar
= WORDCHAR_P (d
- 1);
7452 thischar
= WORDCHAR_P (d
);
7453 if (prevchar
!= thischar
)
7460 boolean prevchar
, thischar
;
7462 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7463 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
7466 prevchar
= WORDCHAR_P (d
- 1);
7467 thischar
= WORDCHAR_P (d
);
7468 if (prevchar
!= thischar
)
7475 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7476 if (!AT_STRINGS_END (d
) && WORDCHAR_P (d
)
7477 && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
7482 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7483 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
7484 && (AT_STRINGS_END (d
) || !WORDCHAR_P (d
)))
7490 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7491 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
7496 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7497 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
7502 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7503 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
7508 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
7513 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7517 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7519 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
7521 SET_REGS_MATCHED ();
7525 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
7527 goto matchnotsyntax
;
7530 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7534 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7536 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
7538 SET_REGS_MATCHED ();
7541 #else /* not emacs */
7543 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7545 if (!WORDCHAR_P (d
))
7547 SET_REGS_MATCHED ();
7552 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7556 SET_REGS_MATCHED ();
7559 #endif /* not emacs */
7564 continue; /* Successfully executed one pattern command; keep going. */
7567 /* We goto here if a matching operation fails. */
7569 if (!FAIL_STACK_EMPTY ())
7570 { /* A restart point is known. Restore to that state. */
7571 DEBUG_PRINT1 ("\nFAIL:\n");
7572 POP_FAILURE_POINT (d
, p
,
7573 lowest_active_reg
, highest_active_reg
,
7574 regstart
, regend
, reg_info
);
7576 /* If this failure point is a dummy, try the next one. */
7580 /* If we failed to the end of the pattern, don't examine *p. */
7584 boolean is_a_jump_n
= false;
7586 /* If failed to a backwards jump that's part of a repetition
7587 loop, need to pop this failure point and use the next one. */
7588 switch ((re_opcode_t
) *p
)
7592 case maybe_pop_jump
:
7593 case pop_failure_jump
:
7596 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7599 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
7601 && (re_opcode_t
) *p1
== on_failure_jump
))
7609 if (d
>= string1
&& d
<= end1
)
7613 break; /* Matching at this starting point really fails. */
7617 goto restore_best_regs
;
7621 return -1; /* Failure to match. */
7624 /* Subroutine definitions for re_match_2. */
7627 /* We are passed P pointing to a register number after a start_memory.
7629 Return true if the pattern up to the corresponding stop_memory can
7630 match the empty string, and false otherwise.
7632 If we find the matching stop_memory, sets P to point to one past its number.
7633 Otherwise, sets P to an undefined byte less than or equal to END.
7635 We don't handle duplicates properly (yet). */
7638 PREFIX(group_match_null_string_p
) (p
, end
, reg_info
)
7640 PREFIX(register_info_type
) *reg_info
;
7643 /* Point to after the args to the start_memory. */
7644 UCHAR_T
*p1
= *p
+ 2;
7648 /* Skip over opcodes that can match nothing, and return true or
7649 false, as appropriate, when we get to one that can't, or to the
7650 matching stop_memory. */
7652 switch ((re_opcode_t
) *p1
)
7654 /* Could be either a loop or a series of alternatives. */
7655 case on_failure_jump
:
7657 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7659 /* If the next operation is not a jump backwards in the
7664 /* Go through the on_failure_jumps of the alternatives,
7665 seeing if any of the alternatives cannot match nothing.
7666 The last alternative starts with only a jump,
7667 whereas the rest start with on_failure_jump and end
7668 with a jump, e.g., here is the pattern for `a|b|c':
7670 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7671 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7674 So, we have to first go through the first (n-1)
7675 alternatives and then deal with the last one separately. */
7678 /* Deal with the first (n-1) alternatives, which start
7679 with an on_failure_jump (see above) that jumps to right
7680 past a jump_past_alt. */
7682 while ((re_opcode_t
) p1
[mcnt
-(1+OFFSET_ADDRESS_SIZE
)] ==
7685 /* `mcnt' holds how many bytes long the alternative
7686 is, including the ending `jump_past_alt' and
7689 if (!PREFIX(alt_match_null_string_p
) (p1
, p1
+ mcnt
-
7690 (1 + OFFSET_ADDRESS_SIZE
),
7694 /* Move to right after this alternative, including the
7698 /* Break if it's the beginning of an n-th alternative
7699 that doesn't begin with an on_failure_jump. */
7700 if ((re_opcode_t
) *p1
!= on_failure_jump
)
7703 /* Still have to check that it's not an n-th
7704 alternative that starts with an on_failure_jump. */
7706 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7707 if ((re_opcode_t
) p1
[mcnt
-(1+OFFSET_ADDRESS_SIZE
)] !=
7710 /* Get to the beginning of the n-th alternative. */
7711 p1
-= 1 + OFFSET_ADDRESS_SIZE
;
7716 /* Deal with the last alternative: go back and get number
7717 of the `jump_past_alt' just before it. `mcnt' contains
7718 the length of the alternative. */
7719 EXTRACT_NUMBER (mcnt
, p1
- OFFSET_ADDRESS_SIZE
);
7721 if (!PREFIX(alt_match_null_string_p
) (p1
, p1
+ mcnt
, reg_info
))
7724 p1
+= mcnt
; /* Get past the n-th alternative. */
7730 assert (p1
[1] == **p
);
7736 if (!PREFIX(common_op_match_null_string_p
) (&p1
, end
, reg_info
))
7739 } /* while p1 < end */
7742 } /* group_match_null_string_p */
7745 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7746 It expects P to be the first byte of a single alternative and END one
7747 byte past the last. The alternative can contain groups. */
7750 PREFIX(alt_match_null_string_p
) (p
, end
, reg_info
)
7752 PREFIX(register_info_type
) *reg_info
;
7759 /* Skip over opcodes that can match nothing, and break when we get
7760 to one that can't. */
7762 switch ((re_opcode_t
) *p1
)
7765 case on_failure_jump
:
7767 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7772 if (!PREFIX(common_op_match_null_string_p
) (&p1
, end
, reg_info
))
7775 } /* while p1 < end */
7778 } /* alt_match_null_string_p */
7781 /* Deals with the ops common to group_match_null_string_p and
7782 alt_match_null_string_p.
7784 Sets P to one after the op and its arguments, if any. */
7787 PREFIX(common_op_match_null_string_p
) (p
, end
, reg_info
)
7789 PREFIX(register_info_type
) *reg_info
;
7796 switch ((re_opcode_t
) *p1
++)
7816 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
7817 ret
= PREFIX(group_match_null_string_p
) (&p1
, end
, reg_info
);
7819 /* Have to set this here in case we're checking a group which
7820 contains a group and a back reference to it. */
7822 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
7823 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
7829 /* If this is an optimized succeed_n for zero times, make the jump. */
7831 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7839 /* Get to the number of times to succeed. */
7840 p1
+= OFFSET_ADDRESS_SIZE
;
7841 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7845 p1
-= 2 * OFFSET_ADDRESS_SIZE
;
7846 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7854 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
7859 p1
+= 2 * OFFSET_ADDRESS_SIZE
;
7862 /* All other opcodes mean we cannot match the empty string. */
7868 } /* common_op_match_null_string_p */
7871 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7872 bytes; nonzero otherwise. */
7875 PREFIX(bcmp_translate
) (s1
, s2
, len
, translate
)
7876 const CHAR_T
*s1
, *s2
;
7878 RE_TRANSLATE_TYPE translate
;
7880 register const UCHAR_T
*p1
= (const UCHAR_T
*) s1
;
7881 register const UCHAR_T
*p2
= (const UCHAR_T
*) s2
;
7885 if (((*p1
<=0xff)?translate
[*p1
++]:*p1
++)
7886 != ((*p2
<=0xff)?translate
[*p2
++]:*p2
++))
7889 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
7897 #else /* not INSIDE_RECURSION */
7899 /* Entry points for GNU code. */
7901 /* re_compile_pattern is the GNU regular expression compiler: it
7902 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7903 Returns 0 if the pattern was valid, otherwise an error string.
7905 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7906 are set in BUFP on entry.
7908 We call regex_compile to do the actual compilation. */
7911 re_compile_pattern (pattern
, length
, bufp
)
7912 const char *pattern
;
7914 struct re_pattern_buffer
*bufp
;
7918 /* GNU code is written to assume at least RE_NREGS registers will be set
7919 (and at least one extra will be -1). */
7920 bufp
->regs_allocated
= REGS_UNALLOCATED
;
7922 /* And GNU code determines whether or not to get register information
7923 by passing null for the REGS argument to re_match, etc., not by
7927 /* Match anchors at newline. */
7928 bufp
->newline_anchor
= 1;
7931 if (MB_CUR_MAX
!= 1)
7932 ret
= wcs_regex_compile (pattern
, length
, re_syntax_options
, bufp
);
7935 ret
= byte_regex_compile (pattern
, length
, re_syntax_options
, bufp
);
7939 return gettext (re_error_msgid
+ re_error_msgid_idx
[(int) ret
]);
7942 weak_alias (__re_compile_pattern
, re_compile_pattern
)
7945 /* Entry points compatible with 4.2 BSD regex library. We don't define
7946 them unless specifically requested. */
7948 #if defined _REGEX_RE_COMP || defined _LIBC
7950 /* BSD has one and only one pattern buffer. */
7951 static struct re_pattern_buffer re_comp_buf
;
7955 /* Make these definitions weak in libc, so POSIX programs can redefine
7956 these names if they don't use our functions, and still use
7957 regcomp/regexec below without link errors. */
7967 if (!re_comp_buf
.buffer
)
7968 return gettext ("No previous regular expression");
7972 if (!re_comp_buf
.buffer
)
7974 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
7975 if (re_comp_buf
.buffer
== NULL
)
7976 return (char *) gettext (re_error_msgid
7977 + re_error_msgid_idx
[(int) REG_ESPACE
]);
7978 re_comp_buf
.allocated
= 200;
7980 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
7981 if (re_comp_buf
.fastmap
== NULL
)
7982 return (char *) gettext (re_error_msgid
7983 + re_error_msgid_idx
[(int) REG_ESPACE
]);
7986 /* Since `re_exec' always passes NULL for the `regs' argument, we
7987 don't need to initialize the pattern buffer fields which affect it. */
7989 /* Match anchors at newlines. */
7990 re_comp_buf
.newline_anchor
= 1;
7993 if (MB_CUR_MAX
!= 1)
7994 ret
= wcs_regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
7997 ret
= byte_regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
8002 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
8003 return (char *) gettext (re_error_msgid
+ re_error_msgid_idx
[(int) ret
]);
8014 const int len
= strlen (s
);
8016 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
8019 #endif /* _REGEX_RE_COMP */
8021 /* POSIX.2 functions. Don't define these for Emacs. */
8025 /* regcomp takes a regular expression as a string and compiles it.
8027 PREG is a regex_t *. We do not expect any fields to be initialized,
8028 since POSIX says we shouldn't. Thus, we set
8030 `buffer' to the compiled pattern;
8031 `used' to the length of the compiled pattern;
8032 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
8033 REG_EXTENDED bit in CFLAGS is set; otherwise, to
8034 RE_SYNTAX_POSIX_BASIC;
8035 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
8036 `fastmap' to an allocated space for the fastmap;
8037 `fastmap_accurate' to zero;
8038 `re_nsub' to the number of subexpressions in PATTERN.
8040 PATTERN is the address of the pattern string.
8042 CFLAGS is a series of bits which affect compilation.
8044 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
8045 use POSIX basic syntax.
8047 If REG_NEWLINE is set, then . and [^...] don't match newline.
8048 Also, regexec will try a match beginning after every newline.
8050 If REG_ICASE is set, then we considers upper- and lowercase
8051 versions of letters to be equivalent when matching.
8053 If REG_NOSUB is set, then when PREG is passed to regexec, that
8054 routine will report only success or failure, and nothing about the
8057 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
8058 the return codes and their meanings.) */
8061 regcomp (preg
, pattern
, cflags
)
8063 const char *pattern
;
8068 = (cflags
& REG_EXTENDED
) ?
8069 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
8071 /* regex_compile will allocate the space for the compiled pattern. */
8073 preg
->allocated
= 0;
8076 /* Try to allocate space for the fastmap. */
8077 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
8079 if (cflags
& REG_ICASE
)
8084 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
8085 * sizeof (*(RE_TRANSLATE_TYPE
)0));
8086 if (preg
->translate
== NULL
)
8087 return (int) REG_ESPACE
;
8089 /* Map uppercase characters to corresponding lowercase ones. */
8090 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
8091 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
8094 preg
->translate
= NULL
;
8096 /* If REG_NEWLINE is set, newlines are treated differently. */
8097 if (cflags
& REG_NEWLINE
)
8098 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
8099 syntax
&= ~RE_DOT_NEWLINE
;
8100 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
8101 /* It also changes the matching behavior. */
8102 preg
->newline_anchor
= 1;
8105 preg
->newline_anchor
= 0;
8107 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
8109 /* POSIX says a null character in the pattern terminates it, so we
8110 can use strlen here in compiling the pattern. */
8112 if (MB_CUR_MAX
!= 1)
8113 ret
= wcs_regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
8116 ret
= byte_regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
8118 /* POSIX doesn't distinguish between an unmatched open-group and an
8119 unmatched close-group: both are REG_EPAREN. */
8120 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
8122 if (ret
== REG_NOERROR
&& preg
->fastmap
)
8124 /* Compute the fastmap now, since regexec cannot modify the pattern
8126 if (re_compile_fastmap (preg
) == -2)
8128 /* Some error occurred while computing the fastmap, just forget
8130 free (preg
->fastmap
);
8131 preg
->fastmap
= NULL
;
8138 weak_alias (__regcomp
, regcomp
)
8142 /* regexec searches for a given pattern, specified by PREG, in the
8145 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8146 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8147 least NMATCH elements, and we set them to the offsets of the
8148 corresponding matched substrings.
8150 EFLAGS specifies `execution flags' which affect matching: if
8151 REG_NOTBOL is set, then ^ does not match at the beginning of the
8152 string; if REG_NOTEOL is set, then $ does not match at the end.
8154 We return 0 if we find a match and REG_NOMATCH if not. */
8157 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
8158 const regex_t
*preg
;
8161 regmatch_t pmatch
[];
8165 struct re_registers regs
;
8166 regex_t private_preg
;
8167 int len
= strlen (string
);
8168 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
8170 private_preg
= *preg
;
8172 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
8173 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
8175 /* The user has told us exactly how many registers to return
8176 information about, via `nmatch'. We have to pass that on to the
8177 matching routines. */
8178 private_preg
.regs_allocated
= REGS_FIXED
;
8182 regs
.num_regs
= nmatch
;
8183 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
8184 if (regs
.start
== NULL
)
8185 return (int) REG_NOMATCH
;
8186 regs
.end
= regs
.start
+ nmatch
;
8189 /* Perform the searching operation. */
8190 ret
= re_search (&private_preg
, string
, len
,
8191 /* start: */ 0, /* range: */ len
,
8192 want_reg_info
? ®s
: (struct re_registers
*) 0);
8194 /* Copy the register information to the POSIX structure. */
8201 for (r
= 0; r
< nmatch
; r
++)
8203 pmatch
[r
].rm_so
= regs
.start
[r
];
8204 pmatch
[r
].rm_eo
= regs
.end
[r
];
8208 /* If we needed the temporary register info, free the space now. */
8212 /* We want zero return to mean success, unlike `re_search'. */
8213 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
8216 weak_alias (__regexec
, regexec
)
8220 /* Returns a message corresponding to an error code, ERRCODE, returned
8221 from either regcomp or regexec. We don't use PREG here. */
8224 regerror (errcode
, preg
, errbuf
, errbuf_size
)
8226 const regex_t
*preg
;
8236 || errcode
>= (int) (sizeof (re_error_msgid_idx
)
8237 / sizeof (re_error_msgid_idx
[0])))
8238 /* Only error codes returned by the rest of the code should be passed
8239 to this routine. If we are given anything else, or if other regex
8240 code generates an invalid error code, then the program has a bug.
8241 Dump core so we can fix it. */
8244 msg
= gettext (re_error_msgid
+ re_error_msgid_idx
[errcode
]);
8246 msg_size
= strlen (msg
) + 1; /* Includes the null. */
8248 if (errbuf_size
!= 0)
8250 if (msg_size
> errbuf_size
)
8252 #if 0 && (defined HAVE_MEMPCPY || defined _LIBC)
8253 *((char *) __mempcpy (errbuf
, msg
, errbuf_size
- 1)) = '\0';
8255 memcpy (errbuf
, msg
, errbuf_size
- 1);
8256 errbuf
[errbuf_size
- 1] = 0;
8260 memcpy (errbuf
, msg
, msg_size
);
8266 weak_alias (__regerror
, regerror
)
8270 /* Free dynamically allocated space used by PREG. */
8276 if (preg
->buffer
!= NULL
)
8277 free (preg
->buffer
);
8278 preg
->buffer
= NULL
;
8280 preg
->allocated
= 0;
8283 if (preg
->fastmap
!= NULL
)
8284 free (preg
->fastmap
);
8285 preg
->fastmap
= NULL
;
8286 preg
->fastmap_accurate
= 0;
8288 if (preg
->translate
!= NULL
)
8289 free (preg
->translate
);
8290 preg
->translate
= NULL
;
8293 weak_alias (__regfree
, regfree
)
8296 #endif /* not emacs */
8298 #endif /* not INSIDE_RECURSION */
8302 #undef STORE_NUMBER_AND_INCR
8303 #undef EXTRACT_NUMBER
8304 #undef EXTRACT_NUMBER_AND_INCR
8306 #undef DEBUG_PRINT_COMPILED_PATTERN
8307 #undef DEBUG_PRINT_DOUBLE_STRING
8309 #undef INIT_FAIL_STACK
8310 #undef RESET_FAIL_STACK
8311 #undef DOUBLE_FAIL_STACK
8312 #undef PUSH_PATTERN_OP
8313 #undef PUSH_FAILURE_POINTER
8314 #undef PUSH_FAILURE_INT
8315 #undef PUSH_FAILURE_ELT
8316 #undef POP_FAILURE_POINTER
8317 #undef POP_FAILURE_INT
8318 #undef POP_FAILURE_ELT
8321 #undef PUSH_FAILURE_POINT
8322 #undef POP_FAILURE_POINT
8324 #undef REG_UNSET_VALUE
8332 #undef INIT_BUF_SIZE
8333 #undef GET_BUFFER_SPACE
8341 #undef EXTEND_BUFFER
8342 #undef GET_UNSIGNED_NUMBER
8343 #undef FREE_STACK_RETURN
8345 # undef POINTER_TO_OFFSET
8346 # undef MATCHING_IN_FRST_STRING
8348 # undef AT_STRINGS_BEG
8349 # undef AT_STRINGS_END
8352 # undef FREE_VARIABLES
8353 # undef NO_HIGHEST_ACTIVE_REG
8354 # undef NO_LOWEST_ACTIVE_REG
8358 # undef COMPILED_BUFFER_VAR
8359 # undef OFFSET_ADDRESS_SIZE
8360 # undef CHAR_CLASS_SIZE
8367 # define DEFINED_ONCE
8368 #endif /* USE_INCLUDED_REGEX */