1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
5 Copyright (C) 1993-1999, 2000, 2001 Free Software Foundation, Inc.
6 This file is part of the GNU C Library.
8 The GNU C Library is free software; you can redistribute it and/or
9 modify it under the terms of the GNU Lesser General Public
10 License as published by the Free Software Foundation; either
11 version 2.1 of the License, or (at your option) any later version.
13 The GNU C Library is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public
19 License along with the GNU C Library; if not, write to the Free
20 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
23 /* AIX requires this to be the first thing in the file. */
24 #if defined _AIX && !defined REGEX_MALLOC
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>
53 # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
55 /* For platform which support the ISO C amendement 1 functionality we
56 support user defined character classes. */
57 # if defined _LIBC || WIDE_CHAR_SUPPORT
58 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
64 /* We have to keep the namespace clean. */
65 # define regfree(preg) __regfree (preg)
66 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
67 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
68 # define regerror(errcode, preg, errbuf, errbuf_size) \
69 __regerror(errcode, preg, errbuf, errbuf_size)
70 # define re_set_registers(bu, re, nu, st, en) \
71 __re_set_registers (bu, re, nu, st, en)
72 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
73 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
74 # define re_match(bufp, string, size, pos, regs) \
75 __re_match (bufp, string, size, pos, regs)
76 # define re_search(bufp, string, size, startpos, range, regs) \
77 __re_search (bufp, string, size, startpos, range, regs)
78 # define re_compile_pattern(pattern, length, bufp) \
79 __re_compile_pattern (pattern, length, bufp)
80 # define re_set_syntax(syntax) __re_set_syntax (syntax)
81 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
82 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
83 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
85 # define btowc __btowc
87 /* We are also using some library internals. */
88 # include <locale/localeinfo.h>
89 # include <locale/elem-hash.h>
90 # include <langinfo.h>
91 # include <locale/coll-lookup.h>
94 /* This is for other GNU distributions with internationalized messages. */
95 # if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC
99 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
102 # define gettext(msgid) (msgid)
105 # ifndef gettext_noop
106 /* This define is so xgettext can find the internationalizable
108 # define gettext_noop(String) String
111 /* The `emacs' switch turns on certain matching commands
112 that make sense only in Emacs. */
119 # else /* not emacs */
121 /* If we are not linking with Emacs proper,
122 we can't use the relocating allocator
123 even if config.h says that we can. */
126 # if defined STDC_HEADERS || defined _LIBC
133 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
134 If nothing else has been done, use the method below. */
135 # ifdef INHIBIT_STRING_HEADER
136 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
137 # if !defined bzero && !defined bcopy
138 # undef INHIBIT_STRING_HEADER
143 /* This is the normal way of making sure we have a bcopy and a bzero.
144 This is used in most programs--a few other programs avoid this
145 by defining INHIBIT_STRING_HEADER. */
146 # ifndef INHIBIT_STRING_HEADER
147 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
151 # define bzero(s, n) (memset (s, '\0', n), (s))
153 # define bzero(s, n) __bzero (s, n)
157 # include <strings.h>
159 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
162 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
167 /* Define the syntax stuff for \<, \>, etc. */
169 /* This must be nonzero for the wordchar and notwordchar pattern
170 commands in re_match_2. */
175 # ifdef SWITCH_ENUM_BUG
176 # define SWITCH_ENUM_CAST(x) ((int)(x))
178 # define SWITCH_ENUM_CAST(x) (x)
181 # endif /* not emacs */
183 # if defined _LIBC || HAVE_LIMITS_H
188 # define MB_LEN_MAX 1
191 /* Get the interface, including the syntax bits. */
194 /* isalpha etc. are used for the character classes. */
197 /* Jim Meyering writes:
199 "... Some ctype macros are valid only for character codes that
200 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
201 using /bin/cc or gcc but without giving an ansi option). So, all
202 ctype uses should be through macros like ISPRINT... If
203 STDC_HEADERS is defined, then autoconf has verified that the ctype
204 macros don't need to be guarded with references to isascii. ...
205 Defining isascii to 1 should let any compiler worth its salt
206 eliminate the && through constant folding."
207 Solaris defines some of these symbols so we must undefine them first. */
210 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
211 # define ISASCII(c) 1
213 # define ISASCII(c) isascii(c)
217 # define ISBLANK(c) (ISASCII (c) && isblank (c))
219 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
222 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
224 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
228 # define ISPRINT(c) (ISASCII (c) && isprint (c))
229 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
230 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
231 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
232 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
233 # define ISLOWER(c) (ISASCII (c) && islower (c))
234 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
235 # define ISSPACE(c) (ISASCII (c) && isspace (c))
236 # define ISUPPER(c) (ISASCII (c) && isupper (c))
237 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
240 # define TOLOWER(c) _tolower(c)
242 # define TOLOWER(c) tolower(c)
246 # define NULL (void *)0
249 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
250 since ours (we hope) works properly with all combinations of
251 machines, compilers, `char' and `unsigned char' argument types.
252 (Per Bothner suggested the basic approach.) */
253 # undef SIGN_EXTEND_CHAR
255 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
256 # else /* not __STDC__ */
257 /* As in Harbison and Steele. */
258 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
262 /* How many characters in the character set. */
263 # define CHAR_SET_SIZE 256
267 extern char *re_syntax_table
;
269 # else /* not SYNTAX_TABLE */
271 static char re_syntax_table
[CHAR_SET_SIZE
];
273 static void init_syntax_once
PARAMS ((void));
283 bzero (re_syntax_table
, sizeof re_syntax_table
);
285 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
287 re_syntax_table
[c
] = Sword
;
289 re_syntax_table
['_'] = Sword
;
294 # endif /* not SYNTAX_TABLE */
296 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
300 /* Integer type for pointers. */
301 # if !defined _LIBC && !defined HAVE_UINTPTR_T
302 typedef unsigned long int uintptr_t;
305 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
306 use `alloca' instead of `malloc'. This is because using malloc in
307 re_search* or re_match* could cause memory leaks when C-g is used in
308 Emacs; also, malloc is slower and causes storage fragmentation. On
309 the other hand, malloc is more portable, and easier to debug.
311 Because we sometimes use alloca, some routines have to be macros,
312 not functions -- `alloca'-allocated space disappears at the end of the
313 function it is called in. */
317 # define REGEX_ALLOCATE malloc
318 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
319 # define REGEX_FREE free
321 # else /* not REGEX_MALLOC */
323 /* Emacs already defines alloca, sometimes. */
326 /* Make alloca work the best possible way. */
328 # define alloca __builtin_alloca
329 # else /* not __GNUC__ */
332 # endif /* HAVE_ALLOCA_H */
333 # endif /* not __GNUC__ */
335 # endif /* not alloca */
337 # define REGEX_ALLOCATE alloca
339 /* Assumes a `char *destination' variable. */
340 # define REGEX_REALLOCATE(source, osize, nsize) \
341 (destination = (char *) alloca (nsize), \
342 memcpy (destination, source, osize))
344 /* No need to do anything to free, after alloca. */
345 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
347 # endif /* not REGEX_MALLOC */
349 /* Define how to allocate the failure stack. */
351 # if defined REL_ALLOC && defined REGEX_MALLOC
353 # define REGEX_ALLOCATE_STACK(size) \
354 r_alloc (&failure_stack_ptr, (size))
355 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
356 r_re_alloc (&failure_stack_ptr, (nsize))
357 # define REGEX_FREE_STACK(ptr) \
358 r_alloc_free (&failure_stack_ptr)
360 # else /* not using relocating allocator */
364 # define REGEX_ALLOCATE_STACK malloc
365 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
366 # define REGEX_FREE_STACK free
368 # else /* not REGEX_MALLOC */
370 # define REGEX_ALLOCATE_STACK alloca
372 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
373 REGEX_REALLOCATE (source, osize, nsize)
374 /* No need to explicitly free anything. */
375 # define REGEX_FREE_STACK(arg)
377 # endif /* not REGEX_MALLOC */
378 # endif /* not using relocating allocator */
381 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
382 `string1' or just past its end. This works if PTR is NULL, which is
384 # define FIRST_STRING_P(ptr) \
385 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
387 /* (Re)Allocate N items of type T using malloc, or fail. */
388 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
389 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
390 # define RETALLOC_IF(addr, n, t) \
391 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
392 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
394 # define BYTEWIDTH 8 /* In bits. */
396 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
400 # define MAX(a, b) ((a) > (b) ? (a) : (b))
401 # define MIN(a, b) ((a) < (b) ? (a) : (b))
403 typedef char boolean
;
407 static reg_errcode_t byte_regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
409 struct re_pattern_buffer
*bufp
));
411 static int byte_re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
412 const char *string1
, int size1
,
413 const char *string2
, int size2
,
415 struct re_registers
*regs
,
417 static int byte_re_search_2
PARAMS ((struct re_pattern_buffer
*bufp
,
418 const char *string1
, int size1
,
419 const char *string2
, int size2
,
420 int startpos
, int range
,
421 struct re_registers
*regs
, int stop
));
422 static int byte_re_compile_fastmap
PARAMS ((struct re_pattern_buffer
*bufp
));
425 static reg_errcode_t wcs_regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
427 struct re_pattern_buffer
*bufp
));
430 static int wcs_re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
431 const char *cstring1
, int csize1
,
432 const char *cstring2
, int csize2
,
434 struct re_registers
*regs
,
436 wchar_t *string1
, int size1
,
437 wchar_t *string2
, int size2
,
438 int *mbs_offset1
, int *mbs_offset2
));
439 static int wcs_re_search_2
PARAMS ((struct re_pattern_buffer
*bufp
,
440 const char *string1
, int size1
,
441 const char *string2
, int size2
,
442 int startpos
, int range
,
443 struct re_registers
*regs
, int stop
));
444 static int wcs_re_compile_fastmap
PARAMS ((struct re_pattern_buffer
*bufp
));
447 /* These are the command codes that appear in compiled regular
448 expressions. Some opcodes are followed by argument bytes. A
449 command code can specify any interpretation whatsoever for its
450 arguments. Zero bytes may appear in the compiled regular expression. */
456 /* Succeed right away--no more backtracking. */
459 /* Followed by one byte giving n, then by n literal bytes. */
463 /* Same as exactn, but contains binary data. */
467 /* Matches any (more or less) character. */
470 /* Matches any one char belonging to specified set. First
471 following byte is number of bitmap bytes. Then come bytes
472 for a bitmap saying which chars are in. Bits in each byte
473 are ordered low-bit-first. A character is in the set if its
474 bit is 1. A character too large to have a bit in the map is
475 automatically not in the set. */
476 /* ifdef MBS_SUPPORT, following element is length of character
477 classes, length of collating symbols, length of equivalence
478 classes, length of character ranges, and length of characters.
479 Next, character class element, collating symbols elements,
480 equivalence class elements, range elements, and character
482 See regex_compile function. */
485 /* Same parameters as charset, but match any character that is
486 not one of those specified. */
489 /* Start remembering the text that is matched, for storing in a
490 register. Followed by one byte with the register number, in
491 the range 0 to one less than the pattern buffer's re_nsub
492 field. Then followed by one byte with the number of groups
493 inner to this one. (This last has to be part of the
494 start_memory only because we need it in the on_failure_jump
498 /* Stop remembering the text that is matched and store it in a
499 memory register. Followed by one byte with the register
500 number, in the range 0 to one less than `re_nsub' in the
501 pattern buffer, and one byte with the number of inner groups,
502 just like `start_memory'. (We need the number of inner
503 groups here because we don't have any easy way of finding the
504 corresponding start_memory when we're at a stop_memory.) */
507 /* Match a duplicate of something remembered. Followed by one
508 byte containing the register number. */
511 /* Fail unless at beginning of line. */
514 /* Fail unless at end of line. */
517 /* Succeeds if at beginning of buffer (if emacs) or at beginning
518 of string to be matched (if not). */
521 /* Analogously, for end of buffer/string. */
524 /* Followed by two byte relative address to which to jump. */
527 /* Same as jump, but marks the end of an alternative. */
530 /* Followed by two-byte relative address of place to resume at
531 in case of failure. */
532 /* ifdef MBS_SUPPORT, the size of address is 1. */
535 /* Like on_failure_jump, but pushes a placeholder instead of the
536 current string position when executed. */
537 on_failure_keep_string_jump
,
539 /* Throw away latest failure point and then jump to following
540 two-byte relative address. */
541 /* ifdef MBS_SUPPORT, the size of address is 1. */
544 /* Change to pop_failure_jump if know won't have to backtrack to
545 match; otherwise change to jump. This is used to jump
546 back to the beginning of a repeat. If what follows this jump
547 clearly won't match what the repeat does, such that we can be
548 sure that there is no use backtracking out of repetitions
549 already matched, then we change it to a pop_failure_jump.
550 Followed by two-byte address. */
551 /* ifdef MBS_SUPPORT, the size of address is 1. */
554 /* Jump to following two-byte address, and push a dummy failure
555 point. This failure point will be thrown away if an attempt
556 is made to use it for a failure. A `+' construct makes this
557 before the first repeat. Also used as an intermediary kind
558 of jump when compiling an alternative. */
559 /* ifdef MBS_SUPPORT, the size of address is 1. */
562 /* Push a dummy failure point and continue. Used at the end of
566 /* Followed by two-byte relative address and two-byte number n.
567 After matching N times, jump to the address upon failure. */
568 /* ifdef MBS_SUPPORT, the size of address is 1. */
571 /* Followed by two-byte relative address, and two-byte number n.
572 Jump to the address N times, then fail. */
573 /* ifdef MBS_SUPPORT, the size of address is 1. */
576 /* Set the following two-byte relative address to the
577 subsequent two-byte number. The address *includes* the two
579 /* ifdef MBS_SUPPORT, the size of address is 1. */
582 wordchar
, /* Matches any word-constituent character. */
583 notwordchar
, /* Matches any char that is not a word-constituent. */
585 wordbeg
, /* Succeeds if at word beginning. */
586 wordend
, /* Succeeds if at word end. */
588 wordbound
, /* Succeeds if at a word boundary. */
589 notwordbound
/* Succeeds if not at a word boundary. */
592 ,before_dot
, /* Succeeds if before point. */
593 at_dot
, /* Succeeds if at point. */
594 after_dot
, /* Succeeds if after point. */
596 /* Matches any character whose syntax is specified. Followed by
597 a byte which contains a syntax code, e.g., Sword. */
600 /* Matches any character whose syntax is not that specified. */
604 #endif /* not INSIDE_RECURSION */
609 # define UCHAR_T unsigned char
610 # define COMPILED_BUFFER_VAR bufp->buffer
611 # define OFFSET_ADDRESS_SIZE 2
612 # define PREFIX(name) byte_##name
613 # define ARG_PREFIX(name) name
614 # define PUT_CHAR(c) putchar (c)
617 # define CHAR_T wchar_t
618 # define UCHAR_T wchar_t
619 # define COMPILED_BUFFER_VAR wc_buffer
620 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
621 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
622 # define PREFIX(name) wcs_##name
623 # define ARG_PREFIX(name) c##name
624 /* Should we use wide stream?? */
625 # define PUT_CHAR(c) printf ("%C", c);
631 # define INSIDE_RECURSION
633 # undef INSIDE_RECURSION
636 # define INSIDE_RECURSION
638 # undef INSIDE_RECURSION
642 #ifdef INSIDE_RECURSION
643 /* Common operations on the compiled pattern. */
645 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
646 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
649 # define STORE_NUMBER(destination, number) \
651 *(destination) = (UCHAR_T)(number); \
654 # define STORE_NUMBER(destination, number) \
656 (destination)[0] = (number) & 0377; \
657 (destination)[1] = (number) >> 8; \
661 /* Same as STORE_NUMBER, except increment DESTINATION to
662 the byte after where the number is stored. Therefore, DESTINATION
663 must be an lvalue. */
664 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
666 # define STORE_NUMBER_AND_INCR(destination, number) \
668 STORE_NUMBER (destination, number); \
669 (destination) += OFFSET_ADDRESS_SIZE; \
672 /* Put into DESTINATION a number stored in two contiguous bytes starting
674 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
677 # define EXTRACT_NUMBER(destination, source) \
679 (destination) = *(source); \
682 # define EXTRACT_NUMBER(destination, source) \
684 (destination) = *(source) & 0377; \
685 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
690 static void PREFIX(extract_number
) _RE_ARGS ((int *dest
, UCHAR_T
*source
));
692 PREFIX(extract_number
) (dest
, source
)
699 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
700 *dest
= *source
& 0377;
705 # ifndef EXTRACT_MACROS /* To debug the macros. */
706 # undef EXTRACT_NUMBER
707 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
708 # endif /* not EXTRACT_MACROS */
712 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
713 SOURCE must be an lvalue. */
715 # define EXTRACT_NUMBER_AND_INCR(destination, source) \
717 EXTRACT_NUMBER (destination, source); \
718 (source) += OFFSET_ADDRESS_SIZE; \
722 static void PREFIX(extract_number_and_incr
) _RE_ARGS ((int *destination
,
725 PREFIX(extract_number_and_incr
) (destination
, source
)
729 PREFIX(extract_number
) (destination
, *source
);
730 *source
+= OFFSET_ADDRESS_SIZE
;
733 # ifndef EXTRACT_MACROS
734 # undef EXTRACT_NUMBER_AND_INCR
735 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
736 PREFIX(extract_number_and_incr) (&dest, &src)
737 # endif /* not EXTRACT_MACROS */
743 /* If DEBUG is defined, Regex prints many voluminous messages about what
744 it is doing (if the variable `debug' is nonzero). If linked with the
745 main program in `iregex.c', you can enter patterns and strings
746 interactively. And if linked with the main program in `main.c' and
747 the other test files, you can run the already-written tests. */
751 # ifndef DEFINED_ONCE
753 /* We use standard I/O for debugging. */
756 /* It is useful to test things that ``must'' be true when debugging. */
761 # define DEBUG_STATEMENT(e) e
762 # define DEBUG_PRINT1(x) if (debug) printf (x)
763 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
764 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
765 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
766 # endif /* not DEFINED_ONCE */
768 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
769 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
770 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
771 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
774 /* Print the fastmap in human-readable form. */
776 # ifndef DEFINED_ONCE
778 print_fastmap (fastmap
)
781 unsigned was_a_range
= 0;
784 while (i
< (1 << BYTEWIDTH
))
790 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
804 # endif /* not DEFINED_ONCE */
807 /* Print a compiled pattern string in human-readable form, starting at
808 the START pointer into it and ending just before the pointer END. */
811 PREFIX(print_partial_compiled_pattern
) (start
, end
)
826 /* Loop over pattern commands. */
830 printf ("%td:\t", p
- start
);
832 printf ("%ld:\t", (long int) (p
- start
));
835 switch ((re_opcode_t
) *p
++)
843 printf ("/exactn/%d", mcnt
);
855 printf ("/exactn_bin/%d", mcnt
);
858 printf("/%lx", (long int) *p
++);
862 # endif /* MBS_SUPPORT */
866 printf ("/start_memory/%d/%ld", mcnt
, (long int) *p
++);
871 printf ("/stop_memory/%d/%ld", mcnt
, (long int) *p
++);
875 printf ("/duplicate/%ld", (long int) *p
++);
888 printf ("/charset [%s",
889 (re_opcode_t
) *(workp
- 1) == charset_not
? "^" : "");
891 length
= *workp
++; /* the length of char_classes */
892 for (i
=0 ; i
<length
; i
++)
893 printf("[:%lx:]", (long int) *p
++);
894 length
= *workp
++; /* the length of collating_symbol */
895 for (i
=0 ; i
<length
;)
899 PUT_CHAR((i
++,*p
++));
903 length
= *workp
++; /* the length of equivalence_class */
904 for (i
=0 ; i
<length
;)
908 PUT_CHAR((i
++,*p
++));
912 length
= *workp
++; /* the length of char_range */
913 for (i
=0 ; i
<length
; i
++)
915 wchar_t range_start
= *p
++;
916 wchar_t range_end
= *p
++;
917 printf("%C-%C", range_start
, range_end
);
919 length
= *workp
++; /* the length of char */
920 for (i
=0 ; i
<length
; i
++)
924 register int c
, last
= -100;
925 register int in_range
= 0;
927 printf ("/charset [%s",
928 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
930 assert (p
+ *p
< pend
);
932 for (c
= 0; c
< 256; c
++)
934 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
936 /* Are we starting a range? */
937 if (last
+ 1 == c
&& ! in_range
)
942 /* Have we broken a range? */
943 else if (last
+ 1 != c
&& in_range
)
973 case on_failure_jump
:
974 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
976 printf ("/on_failure_jump to %td", p
+ mcnt
- start
);
978 printf ("/on_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
982 case on_failure_keep_string_jump
:
983 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
985 printf ("/on_failure_keep_string_jump to %td", p
+ mcnt
- start
);
987 printf ("/on_failure_keep_string_jump to %ld",
988 (long int) (p
+ mcnt
- start
));
992 case dummy_failure_jump
:
993 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
995 printf ("/dummy_failure_jump to %td", p
+ mcnt
- start
);
997 printf ("/dummy_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
1001 case push_dummy_failure
:
1002 printf ("/push_dummy_failure");
1005 case maybe_pop_jump
:
1006 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1008 printf ("/maybe_pop_jump to %td", p
+ mcnt
- start
);
1010 printf ("/maybe_pop_jump to %ld", (long int) (p
+ mcnt
- start
));
1014 case pop_failure_jump
:
1015 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1017 printf ("/pop_failure_jump to %td", p
+ mcnt
- start
);
1019 printf ("/pop_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
1024 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1026 printf ("/jump_past_alt to %td", p
+ mcnt
- start
);
1028 printf ("/jump_past_alt to %ld", (long int) (p
+ mcnt
- start
));
1033 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1035 printf ("/jump to %td", p
+ mcnt
- start
);
1037 printf ("/jump to %ld", (long int) (p
+ mcnt
- start
));
1042 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1044 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1046 printf ("/succeed_n to %td, %d times", p1
- start
, mcnt2
);
1048 printf ("/succeed_n to %ld, %d times",
1049 (long int) (p1
- start
), mcnt2
);
1054 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1056 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1057 printf ("/jump_n to %d, %d times", p1
- start
, mcnt2
);
1061 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1063 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1065 printf ("/set_number_at location %td to %d", p1
- start
, mcnt2
);
1067 printf ("/set_number_at location %ld to %d",
1068 (long int) (p1
- start
), mcnt2
);
1073 printf ("/wordbound");
1077 printf ("/notwordbound");
1081 printf ("/wordbeg");
1085 printf ("/wordend");
1090 printf ("/before_dot");
1098 printf ("/after_dot");
1102 printf ("/syntaxspec");
1104 printf ("/%d", mcnt
);
1108 printf ("/notsyntaxspec");
1110 printf ("/%d", mcnt
);
1115 printf ("/wordchar");
1119 printf ("/notwordchar");
1131 printf ("?%ld", (long int) *(p
-1));
1138 printf ("%td:\tend of pattern.\n", p
- start
);
1140 printf ("%ld:\tend of pattern.\n", (long int) (p
- start
));
1146 PREFIX(print_compiled_pattern
) (bufp
)
1147 struct re_pattern_buffer
*bufp
;
1149 UCHAR_T
*buffer
= (UCHAR_T
*) bufp
->buffer
;
1151 PREFIX(print_partial_compiled_pattern
) (buffer
, buffer
1152 + bufp
->used
/ sizeof(UCHAR_T
));
1153 printf ("%ld bytes used/%ld bytes allocated.\n",
1154 bufp
->used
, bufp
->allocated
);
1156 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1158 printf ("fastmap: ");
1159 print_fastmap (bufp
->fastmap
);
1163 printf ("re_nsub: %Zd\t", bufp
->re_nsub
);
1165 printf ("re_nsub: %ld\t", (long int) bufp
->re_nsub
);
1167 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1168 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1169 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
1170 printf ("no_sub: %d\t", bufp
->no_sub
);
1171 printf ("not_bol: %d\t", bufp
->not_bol
);
1172 printf ("not_eol: %d\t", bufp
->not_eol
);
1173 printf ("syntax: %lx\n", bufp
->syntax
);
1174 /* Perhaps we should print the translate table? */
1179 PREFIX(print_double_string
) (where
, string1
, size1
, string2
, size2
)
1180 const CHAR_T
*where
;
1181 const CHAR_T
*string1
;
1182 const CHAR_T
*string2
;
1194 if (FIRST_STRING_P (where
))
1196 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1197 PUT_CHAR (string1
[this_char
]);
1203 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1205 PUT_CHAR (string2
[this_char
]);
1208 fputs ("...", stdout
);
1215 # ifndef DEFINED_ONCE
1224 # else /* not DEBUG */
1226 # ifndef DEFINED_ONCE
1230 # define DEBUG_STATEMENT(e)
1231 # define DEBUG_PRINT1(x)
1232 # define DEBUG_PRINT2(x1, x2)
1233 # define DEBUG_PRINT3(x1, x2, x3)
1234 # define DEBUG_PRINT4(x1, x2, x3, x4)
1235 # endif /* not DEFINED_ONCE */
1236 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1237 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1239 # endif /* not DEBUG */
1244 /* This convert a multibyte string to a wide character string.
1245 And write their correspondances to offset_buffer(see below)
1246 and write whether each wchar_t is binary data to is_binary.
1247 This assume invalid multibyte sequences as binary data.
1248 We assume offset_buffer and is_binary is already allocated
1251 static size_t convert_mbs_to_wcs (CHAR_T
*dest
, const unsigned char* src
,
1252 size_t len
, int *offset_buffer
,
1255 convert_mbs_to_wcs (dest
, src
, len
, offset_buffer
, is_binary
)
1257 const unsigned char* src
;
1258 size_t len
; /* the length of multibyte string. */
1260 /* It hold correspondances between src(char string) and
1261 dest(wchar_t string) for optimization.
1263 dest = {'X', 'Y', 'Z'}
1264 (each "xxx", "y" and "zz" represent one multibyte character
1265 corresponding to 'X', 'Y' and 'Z'.)
1266 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1272 wchar_t *pdest
= dest
;
1273 const unsigned char *psrc
= src
;
1274 size_t wc_count
= 0;
1278 size_t mb_remain
= len
;
1279 size_t mb_count
= 0;
1281 /* Initialize the conversion state. */
1282 memset (&mbs
, 0, sizeof (mbstate_t));
1284 offset_buffer
[0] = 0;
1285 for( ; mb_remain
> 0 ; ++wc_count
, ++pdest
, mb_remain
-= consumed
,
1289 consumed
= __mbrtowc (pdest
, psrc
, mb_remain
, &mbs
);
1291 consumed
= mbrtowc (pdest
, psrc
, mb_remain
, &mbs
);
1295 /* failed to convert. maybe src contains binary data.
1296 So we consume 1 byte manualy. */
1300 is_binary
[wc_count
] = TRUE
;
1303 is_binary
[wc_count
] = FALSE
;
1304 /* In sjis encoding, we use yen sign as escape character in
1305 place of reverse solidus. So we convert 0x5c(yen sign in
1306 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1307 solidus in UCS2). */
1308 if (consumed
== 1 && (int) *psrc
== 0x5c && (int) *pdest
== 0xa5)
1309 *pdest
= (wchar_t) *psrc
;
1311 offset_buffer
[wc_count
+ 1] = mb_count
+= consumed
;
1314 /* Fill remain of the buffer with sentinel. */
1315 for (i
= wc_count
+ 1 ; i
<= len
; i
++)
1316 offset_buffer
[i
] = mb_count
+ 1;
1323 #else /* not INSIDE_RECURSION */
1325 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1326 also be assigned to arbitrarily: each pattern buffer stores its own
1327 syntax, so it can be changed between regex compilations. */
1328 /* This has no initializer because initialized variables in Emacs
1329 become read-only after dumping. */
1330 reg_syntax_t re_syntax_options
;
1333 /* Specify the precise syntax of regexps for compilation. This provides
1334 for compatibility for various utilities which historically have
1335 different, incompatible syntaxes.
1337 The argument SYNTAX is a bit mask comprised of the various bits
1338 defined in regex.h. We return the old syntax. */
1341 re_set_syntax (syntax
)
1342 reg_syntax_t syntax
;
1344 reg_syntax_t ret
= re_syntax_options
;
1346 re_syntax_options
= syntax
;
1348 if (syntax
& RE_DEBUG
)
1350 else if (debug
) /* was on but now is not */
1356 weak_alias (__re_set_syntax
, re_set_syntax
)
1359 /* This table gives an error message for each of the error codes listed
1360 in regex.h. Obviously the order here has to be same as there.
1361 POSIX doesn't require that we do anything for REG_NOERROR,
1362 but why not be nice? */
1364 static const char re_error_msgid
[] =
1366 # define REG_NOERROR_IDX 0
1367 gettext_noop ("Success") /* REG_NOERROR */
1369 # define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1370 gettext_noop ("No match") /* REG_NOMATCH */
1372 # define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1373 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1375 # define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1376 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1378 # define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1379 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1381 # define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1382 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1384 # define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1385 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1387 # define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1388 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1390 # define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1391 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1393 # define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1394 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1396 # define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1397 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1399 # define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1400 gettext_noop ("Invalid range end") /* REG_ERANGE */
1402 # define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1403 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1405 # define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1406 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1408 # define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1409 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1411 # define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1412 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1414 # define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1415 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1418 static const size_t re_error_msgid_idx
[] =
1439 #endif /* INSIDE_RECURSION */
1441 #ifndef DEFINED_ONCE
1442 /* Avoiding alloca during matching, to placate r_alloc. */
1444 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1445 searching and matching functions should not call alloca. On some
1446 systems, alloca is implemented in terms of malloc, and if we're
1447 using the relocating allocator routines, then malloc could cause a
1448 relocation, which might (if the strings being searched are in the
1449 ralloc heap) shift the data out from underneath the regexp
1452 Here's another reason to avoid allocation: Emacs
1453 processes input from X in a signal handler; processing X input may
1454 call malloc; if input arrives while a matching routine is calling
1455 malloc, then we're scrod. But Emacs can't just block input while
1456 calling matching routines; then we don't notice interrupts when
1457 they come in. So, Emacs blocks input around all regexp calls
1458 except the matching calls, which it leaves unprotected, in the
1459 faith that they will not malloc. */
1461 /* Normally, this is fine. */
1462 # define MATCH_MAY_ALLOCATE
1464 /* When using GNU C, we are not REALLY using the C alloca, no matter
1465 what config.h may say. So don't take precautions for it. */
1470 /* The match routines may not allocate if (1) they would do it with malloc
1471 and (2) it's not safe for them to use malloc.
1472 Note that if REL_ALLOC is defined, matching would not use malloc for the
1473 failure stack, but we would still use it for the register vectors;
1474 so REL_ALLOC should not affect this. */
1475 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1476 # undef MATCH_MAY_ALLOCATE
1478 #endif /* not DEFINED_ONCE */
1480 #ifdef INSIDE_RECURSION
1481 /* Failure stack declarations and macros; both re_compile_fastmap and
1482 re_match_2 use a failure stack. These have to be macros because of
1483 REGEX_ALLOCATE_STACK. */
1486 /* Number of failure points for which to initially allocate space
1487 when matching. If this number is exceeded, we allocate more
1488 space, so it is not a hard limit. */
1489 # ifndef INIT_FAILURE_ALLOC
1490 # define INIT_FAILURE_ALLOC 5
1493 /* Roughly the maximum number of failure points on the stack. Would be
1494 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1495 This is a variable only so users of regex can assign to it; we never
1496 change it ourselves. */
1498 # ifdef INT_IS_16BIT
1500 # ifndef DEFINED_ONCE
1501 # if defined MATCH_MAY_ALLOCATE
1502 /* 4400 was enough to cause a crash on Alpha OSF/1,
1503 whose default stack limit is 2mb. */
1504 long int re_max_failures
= 4000;
1506 long int re_max_failures
= 2000;
1510 union PREFIX(fail_stack_elt
)
1516 typedef union PREFIX(fail_stack_elt
) PREFIX(fail_stack_elt_t
);
1520 PREFIX(fail_stack_elt_t
) *stack
;
1521 unsigned long int size
;
1522 unsigned long int avail
; /* Offset of next open position. */
1523 } PREFIX(fail_stack_type
);
1525 # else /* not INT_IS_16BIT */
1527 # ifndef DEFINED_ONCE
1528 # if defined MATCH_MAY_ALLOCATE
1529 /* 4400 was enough to cause a crash on Alpha OSF/1,
1530 whose default stack limit is 2mb. */
1531 int re_max_failures
= 4000;
1533 int re_max_failures
= 2000;
1537 union PREFIX(fail_stack_elt
)
1543 typedef union PREFIX(fail_stack_elt
) PREFIX(fail_stack_elt_t
);
1547 PREFIX(fail_stack_elt_t
) *stack
;
1549 unsigned avail
; /* Offset of next open position. */
1550 } PREFIX(fail_stack_type
);
1552 # endif /* INT_IS_16BIT */
1554 # ifndef DEFINED_ONCE
1555 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1556 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1557 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1561 /* Define macros to initialize and free the failure stack.
1562 Do `return -2' if the alloc fails. */
1564 # ifdef MATCH_MAY_ALLOCATE
1565 # define INIT_FAIL_STACK() \
1567 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1568 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1570 if (fail_stack.stack == NULL) \
1573 fail_stack.size = INIT_FAILURE_ALLOC; \
1574 fail_stack.avail = 0; \
1577 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1579 # define INIT_FAIL_STACK() \
1581 fail_stack.avail = 0; \
1584 # define RESET_FAIL_STACK()
1588 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1590 Return 1 if succeeds, and 0 if either ran out of memory
1591 allocating space for it or it was already too large.
1593 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1595 # define DOUBLE_FAIL_STACK(fail_stack) \
1596 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1598 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1599 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1600 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1601 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1603 (fail_stack).stack == NULL \
1605 : ((fail_stack).size <<= 1, \
1609 /* Push pointer POINTER on FAIL_STACK.
1610 Return 1 if was able to do so and 0 if ran out of memory allocating
1612 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1613 ((FAIL_STACK_FULL () \
1614 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1616 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1619 /* Push a pointer value onto the failure stack.
1620 Assumes the variable `fail_stack'. Probably should only
1621 be called from within `PUSH_FAILURE_POINT'. */
1622 # define PUSH_FAILURE_POINTER(item) \
1623 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1625 /* This pushes an integer-valued item onto the failure stack.
1626 Assumes the variable `fail_stack'. Probably should only
1627 be called from within `PUSH_FAILURE_POINT'. */
1628 # define PUSH_FAILURE_INT(item) \
1629 fail_stack.stack[fail_stack.avail++].integer = (item)
1631 /* Push a fail_stack_elt_t value onto the failure stack.
1632 Assumes the variable `fail_stack'. Probably should only
1633 be called from within `PUSH_FAILURE_POINT'. */
1634 # define PUSH_FAILURE_ELT(item) \
1635 fail_stack.stack[fail_stack.avail++] = (item)
1637 /* These three POP... operations complement the three PUSH... operations.
1638 All assume that `fail_stack' is nonempty. */
1639 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1640 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1641 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1643 /* Used to omit pushing failure point id's when we're not debugging. */
1645 # define DEBUG_PUSH PUSH_FAILURE_INT
1646 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1648 # define DEBUG_PUSH(item)
1649 # define DEBUG_POP(item_addr)
1653 /* Push the information about the state we will need
1654 if we ever fail back to it.
1656 Requires variables fail_stack, regstart, regend, reg_info, and
1657 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1660 Does `return FAILURE_CODE' if runs out of memory. */
1662 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1664 char *destination; \
1665 /* Must be int, so when we don't save any registers, the arithmetic \
1666 of 0 + -1 isn't done as unsigned. */ \
1667 /* Can't be int, since there is not a shred of a guarantee that int \
1668 is wide enough to hold a value of something to which pointer can \
1670 active_reg_t this_reg; \
1672 DEBUG_STATEMENT (failure_id++); \
1673 DEBUG_STATEMENT (nfailure_points_pushed++); \
1674 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1675 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1676 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1678 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1679 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1681 /* Ensure we have enough space allocated for what we will push. */ \
1682 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1684 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1685 return failure_code; \
1687 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1688 (fail_stack).size); \
1689 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1692 /* Push the info, starting with the registers. */ \
1693 DEBUG_PRINT1 ("\n"); \
1696 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1699 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1700 DEBUG_STATEMENT (num_regs_pushed++); \
1702 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1703 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1705 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1706 PUSH_FAILURE_POINTER (regend[this_reg]); \
1708 DEBUG_PRINT2 (" info: %p\n ", \
1709 reg_info[this_reg].word.pointer); \
1710 DEBUG_PRINT2 (" match_null=%d", \
1711 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1712 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1713 DEBUG_PRINT2 (" matched_something=%d", \
1714 MATCHED_SOMETHING (reg_info[this_reg])); \
1715 DEBUG_PRINT2 (" ever_matched=%d", \
1716 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1717 DEBUG_PRINT1 ("\n"); \
1718 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1721 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1722 PUSH_FAILURE_INT (lowest_active_reg); \
1724 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1725 PUSH_FAILURE_INT (highest_active_reg); \
1727 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1728 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1729 PUSH_FAILURE_POINTER (pattern_place); \
1731 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1732 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1734 DEBUG_PRINT1 ("'\n"); \
1735 PUSH_FAILURE_POINTER (string_place); \
1737 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1738 DEBUG_PUSH (failure_id); \
1741 # ifndef DEFINED_ONCE
1742 /* This is the number of items that are pushed and popped on the stack
1743 for each register. */
1744 # define NUM_REG_ITEMS 3
1746 /* Individual items aside from the registers. */
1748 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1750 # define NUM_NONREG_ITEMS 4
1753 /* We push at most this many items on the stack. */
1754 /* We used to use (num_regs - 1), which is the number of registers
1755 this regexp will save; but that was changed to 5
1756 to avoid stack overflow for a regexp with lots of parens. */
1757 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1759 /* We actually push this many items. */
1760 # define NUM_FAILURE_ITEMS \
1762 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1766 /* How many items can still be added to the stack without overflowing it. */
1767 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1768 # endif /* not DEFINED_ONCE */
1771 /* Pops what PUSH_FAIL_STACK pushes.
1773 We restore into the parameters, all of which should be lvalues:
1774 STR -- the saved data position.
1775 PAT -- the saved pattern position.
1776 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1777 REGSTART, REGEND -- arrays of string positions.
1778 REG_INFO -- array of information about each subexpression.
1780 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1781 `pend', `string1', `size1', `string2', and `size2'. */
1782 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1784 DEBUG_STATEMENT (unsigned failure_id;) \
1785 active_reg_t this_reg; \
1786 const UCHAR_T *string_temp; \
1788 assert (!FAIL_STACK_EMPTY ()); \
1790 /* Remove failure points and point to how many regs pushed. */ \
1791 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1792 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1793 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1795 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1797 DEBUG_POP (&failure_id); \
1798 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1800 /* If the saved string location is NULL, it came from an \
1801 on_failure_keep_string_jump opcode, and we want to throw away the \
1802 saved NULL, thus retaining our current position in the string. */ \
1803 string_temp = POP_FAILURE_POINTER (); \
1804 if (string_temp != NULL) \
1805 str = (const CHAR_T *) string_temp; \
1807 DEBUG_PRINT2 (" Popping string %p: `", str); \
1808 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1809 DEBUG_PRINT1 ("'\n"); \
1811 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1812 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1813 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1815 /* Restore register info. */ \
1816 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1817 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1819 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1820 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1823 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1825 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1827 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1828 DEBUG_PRINT2 (" info: %p\n", \
1829 reg_info[this_reg].word.pointer); \
1831 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1832 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1834 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1835 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1839 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1841 reg_info[this_reg].word.integer = 0; \
1842 regend[this_reg] = 0; \
1843 regstart[this_reg] = 0; \
1845 highest_active_reg = high_reg; \
1848 set_regs_matched_done = 0; \
1849 DEBUG_STATEMENT (nfailure_points_popped++); \
1850 } /* POP_FAILURE_POINT */
1852 /* Structure for per-register (a.k.a. per-group) information.
1853 Other register information, such as the
1854 starting and ending positions (which are addresses), and the list of
1855 inner groups (which is a bits list) are maintained in separate
1858 We are making a (strictly speaking) nonportable assumption here: that
1859 the compiler will pack our bit fields into something that fits into
1860 the type of `word', i.e., is something that fits into one item on the
1864 /* Declarations and macros for re_match_2. */
1868 PREFIX(fail_stack_elt_t
) word
;
1871 /* This field is one if this group can match the empty string,
1872 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1873 # define MATCH_NULL_UNSET_VALUE 3
1874 unsigned match_null_string_p
: 2;
1875 unsigned is_active
: 1;
1876 unsigned matched_something
: 1;
1877 unsigned ever_matched_something
: 1;
1879 } PREFIX(register_info_type
);
1881 # ifndef DEFINED_ONCE
1882 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1883 # define IS_ACTIVE(R) ((R).bits.is_active)
1884 # define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1885 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1888 /* Call this when have matched a real character; it sets `matched' flags
1889 for the subexpressions which we are currently inside. Also records
1890 that those subexprs have matched. */
1891 # define SET_REGS_MATCHED() \
1894 if (!set_regs_matched_done) \
1897 set_regs_matched_done = 1; \
1898 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1900 MATCHED_SOMETHING (reg_info[r]) \
1901 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1907 # endif /* not DEFINED_ONCE */
1909 /* Registers are set to a sentinel when they haven't yet matched. */
1910 static CHAR_T
PREFIX(reg_unset_dummy
);
1911 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1912 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1914 /* Subroutine declarations and macros for regex_compile. */
1915 static void PREFIX(store_op1
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
, int arg
));
1916 static void PREFIX(store_op2
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
,
1917 int arg1
, int arg2
));
1918 static void PREFIX(insert_op1
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
,
1919 int arg
, UCHAR_T
*end
));
1920 static void PREFIX(insert_op2
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
,
1921 int arg1
, int arg2
, UCHAR_T
*end
));
1922 static boolean
PREFIX(at_begline_loc_p
) _RE_ARGS ((const CHAR_T
*pattern
,
1924 reg_syntax_t syntax
));
1925 static boolean
PREFIX(at_endline_loc_p
) _RE_ARGS ((const CHAR_T
*p
,
1927 reg_syntax_t syntax
));
1929 static reg_errcode_t wcs_compile_range
_RE_ARGS ((CHAR_T range_start
,
1930 const CHAR_T
**p_ptr
,
1933 reg_syntax_t syntax
,
1936 static void insert_space
_RE_ARGS ((int num
, CHAR_T
*loc
, CHAR_T
*end
));
1938 static reg_errcode_t byte_compile_range
_RE_ARGS ((unsigned int range_start
,
1942 reg_syntax_t syntax
,
1946 /* Fetch the next character in the uncompiled pattern---translating it
1947 if necessary. Also cast from a signed character in the constant
1948 string passed to us by the user to an unsigned char that we can use
1949 as an array index (in, e.g., `translate'). */
1950 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1951 because it is impossible to allocate 4GB array for some encodings
1952 which have 4 byte character_set like UCS4. */
1955 # define PATFETCH(c) \
1956 do {if (p == pend) return REG_EEND; \
1957 c = (UCHAR_T) *p++; \
1958 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1961 # define PATFETCH(c) \
1962 do {if (p == pend) return REG_EEND; \
1963 c = (unsigned char) *p++; \
1964 if (translate) c = (unsigned char) translate[c]; \
1969 /* Fetch the next character in the uncompiled pattern, with no
1971 # define PATFETCH_RAW(c) \
1972 do {if (p == pend) return REG_EEND; \
1973 c = (UCHAR_T) *p++; \
1976 /* Go backwards one character in the pattern. */
1977 # define PATUNFETCH p--
1980 /* If `translate' is non-null, return translate[D], else just D. We
1981 cast the subscript to translate because some data is declared as
1982 `char *', to avoid warnings when a string constant is passed. But
1983 when we use a character as a subscript we must make it unsigned. */
1984 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1985 because it is impossible to allocate 4GB array for some encodings
1986 which have 4 byte character_set like UCS4. */
1990 # define TRANSLATE(d) \
1991 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1992 ? (char) translate[(unsigned char) (d)] : (d))
1994 # define TRANSLATE(d) \
1995 (translate ? (char) translate[(unsigned char) (d)] : (d))
2000 /* Macros for outputting the compiled pattern into `buffer'. */
2002 /* If the buffer isn't allocated when it comes in, use this. */
2003 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
2005 /* Make sure we have at least N more bytes of space in buffer. */
2007 # define GET_BUFFER_SPACE(n) \
2008 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
2009 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
2012 # define GET_BUFFER_SPACE(n) \
2013 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
2017 /* Make sure we have one more byte of buffer space and then add C to it. */
2018 # define BUF_PUSH(c) \
2020 GET_BUFFER_SPACE (1); \
2021 *b++ = (UCHAR_T) (c); \
2025 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
2026 # define BUF_PUSH_2(c1, c2) \
2028 GET_BUFFER_SPACE (2); \
2029 *b++ = (UCHAR_T) (c1); \
2030 *b++ = (UCHAR_T) (c2); \
2034 /* As with BUF_PUSH_2, except for three bytes. */
2035 # define BUF_PUSH_3(c1, c2, c3) \
2037 GET_BUFFER_SPACE (3); \
2038 *b++ = (UCHAR_T) (c1); \
2039 *b++ = (UCHAR_T) (c2); \
2040 *b++ = (UCHAR_T) (c3); \
2043 /* Store a jump with opcode OP at LOC to location TO. We store a
2044 relative address offset by the three bytes the jump itself occupies. */
2045 # define STORE_JUMP(op, loc, to) \
2046 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
2048 /* Likewise, for a two-argument jump. */
2049 # define STORE_JUMP2(op, loc, to, arg) \
2050 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
2052 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
2053 # define INSERT_JUMP(op, loc, to) \
2054 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
2056 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
2057 # define INSERT_JUMP2(op, loc, to, arg) \
2058 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
2061 /* This is not an arbitrary limit: the arguments which represent offsets
2062 into the pattern are two bytes long. So if 2^16 bytes turns out to
2063 be too small, many things would have to change. */
2064 /* Any other compiler which, like MSC, has allocation limit below 2^16
2065 bytes will have to use approach similar to what was done below for
2066 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2067 reallocating to 0 bytes. Such thing is not going to work too well.
2068 You have been warned!! */
2069 # ifndef DEFINED_ONCE
2070 # if defined _MSC_VER && !defined WIN32
2071 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2072 The REALLOC define eliminates a flurry of conversion warnings,
2073 but is not required. */
2074 # define MAX_BUF_SIZE 65500L
2075 # define REALLOC(p,s) realloc ((p), (size_t) (s))
2077 # define MAX_BUF_SIZE (1L << 16)
2078 # define REALLOC(p,s) realloc ((p), (s))
2081 /* Extend the buffer by twice its current size via realloc and
2082 reset the pointers that pointed into the old block to point to the
2083 correct places in the new one. If extending the buffer results in it
2084 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2085 # if __BOUNDED_POINTERS__
2086 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2087 # define MOVE_BUFFER_POINTER(P) \
2088 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2089 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2092 SET_HIGH_BOUND (b); \
2093 SET_HIGH_BOUND (begalt); \
2094 if (fixup_alt_jump) \
2095 SET_HIGH_BOUND (fixup_alt_jump); \
2097 SET_HIGH_BOUND (laststart); \
2098 if (pending_exact) \
2099 SET_HIGH_BOUND (pending_exact); \
2102 # define MOVE_BUFFER_POINTER(P) (P) += incr
2103 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
2105 # endif /* not DEFINED_ONCE */
2108 # define EXTEND_BUFFER() \
2110 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2112 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2114 bufp->allocated <<= 1; \
2115 if (bufp->allocated > MAX_BUF_SIZE) \
2116 bufp->allocated = MAX_BUF_SIZE; \
2117 /* How many characters the new buffer can have? */ \
2118 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2119 if (wchar_count == 0) wchar_count = 1; \
2120 /* Truncate the buffer to CHAR_T align. */ \
2121 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2122 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2123 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2124 if (COMPILED_BUFFER_VAR == NULL) \
2125 return REG_ESPACE; \
2126 /* If the buffer moved, move all the pointers into it. */ \
2127 if (old_buffer != COMPILED_BUFFER_VAR) \
2129 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2130 MOVE_BUFFER_POINTER (b); \
2131 MOVE_BUFFER_POINTER (begalt); \
2132 if (fixup_alt_jump) \
2133 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2135 MOVE_BUFFER_POINTER (laststart); \
2136 if (pending_exact) \
2137 MOVE_BUFFER_POINTER (pending_exact); \
2139 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2142 # define EXTEND_BUFFER() \
2144 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2145 if (bufp->allocated == MAX_BUF_SIZE) \
2147 bufp->allocated <<= 1; \
2148 if (bufp->allocated > MAX_BUF_SIZE) \
2149 bufp->allocated = MAX_BUF_SIZE; \
2150 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2152 if (COMPILED_BUFFER_VAR == NULL) \
2153 return REG_ESPACE; \
2154 /* If the buffer moved, move all the pointers into it. */ \
2155 if (old_buffer != COMPILED_BUFFER_VAR) \
2157 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2158 MOVE_BUFFER_POINTER (b); \
2159 MOVE_BUFFER_POINTER (begalt); \
2160 if (fixup_alt_jump) \
2161 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2163 MOVE_BUFFER_POINTER (laststart); \
2164 if (pending_exact) \
2165 MOVE_BUFFER_POINTER (pending_exact); \
2167 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2171 # ifndef DEFINED_ONCE
2172 /* Since we have one byte reserved for the register number argument to
2173 {start,stop}_memory, the maximum number of groups we can report
2174 things about is what fits in that byte. */
2175 # define MAX_REGNUM 255
2177 /* But patterns can have more than `MAX_REGNUM' registers. We just
2178 ignore the excess. */
2179 typedef unsigned regnum_t
;
2182 /* Macros for the compile stack. */
2184 /* Since offsets can go either forwards or backwards, this type needs to
2185 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2186 /* int may be not enough when sizeof(int) == 2. */
2187 typedef long pattern_offset_t
;
2191 pattern_offset_t begalt_offset
;
2192 pattern_offset_t fixup_alt_jump
;
2193 pattern_offset_t inner_group_offset
;
2194 pattern_offset_t laststart_offset
;
2196 } compile_stack_elt_t
;
2201 compile_stack_elt_t
*stack
;
2203 unsigned avail
; /* Offset of next open position. */
2204 } compile_stack_type
;
2207 # define INIT_COMPILE_STACK_SIZE 32
2209 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2210 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2212 /* The next available element. */
2213 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2215 # endif /* not DEFINED_ONCE */
2217 /* Set the bit for character C in a list. */
2218 # ifndef DEFINED_ONCE
2219 # define SET_LIST_BIT(c) \
2220 (b[((unsigned char) (c)) / BYTEWIDTH] \
2221 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2222 # endif /* DEFINED_ONCE */
2224 /* Get the next unsigned number in the uncompiled pattern. */
2225 # define GET_UNSIGNED_NUMBER(num) \
2230 if (c < '0' || c > '9') \
2232 if (num <= RE_DUP_MAX) \
2236 num = num * 10 + c - '0'; \
2241 # ifndef DEFINED_ONCE
2242 # if defined _LIBC || WIDE_CHAR_SUPPORT
2243 /* The GNU C library provides support for user-defined character classes
2244 and the functions from ISO C amendement 1. */
2245 # ifdef CHARCLASS_NAME_MAX
2246 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2248 /* This shouldn't happen but some implementation might still have this
2249 problem. Use a reasonable default value. */
2250 # define CHAR_CLASS_MAX_LENGTH 256
2254 # define IS_CHAR_CLASS(string) __wctype (string)
2256 # define IS_CHAR_CLASS(string) wctype (string)
2259 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2261 # define IS_CHAR_CLASS(string) \
2262 (STREQ (string, "alpha") || STREQ (string, "upper") \
2263 || STREQ (string, "lower") || STREQ (string, "digit") \
2264 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2265 || STREQ (string, "space") || STREQ (string, "print") \
2266 || STREQ (string, "punct") || STREQ (string, "graph") \
2267 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2269 # endif /* DEFINED_ONCE */
2271 # ifndef MATCH_MAY_ALLOCATE
2273 /* If we cannot allocate large objects within re_match_2_internal,
2274 we make the fail stack and register vectors global.
2275 The fail stack, we grow to the maximum size when a regexp
2277 The register vectors, we adjust in size each time we
2278 compile a regexp, according to the number of registers it needs. */
2280 static PREFIX(fail_stack_type
) fail_stack
;
2282 /* Size with which the following vectors are currently allocated.
2283 That is so we can make them bigger as needed,
2284 but never make them smaller. */
2285 # ifdef DEFINED_ONCE
2286 static int regs_allocated_size
;
2288 static const char ** regstart
, ** regend
;
2289 static const char ** old_regstart
, ** old_regend
;
2290 static const char **best_regstart
, **best_regend
;
2291 static const char **reg_dummy
;
2292 # endif /* DEFINED_ONCE */
2294 static PREFIX(register_info_type
) *PREFIX(reg_info
);
2295 static PREFIX(register_info_type
) *PREFIX(reg_info_dummy
);
2297 /* Make the register vectors big enough for NUM_REGS registers,
2298 but don't make them smaller. */
2301 PREFIX(regex_grow_registers
) (num_regs
)
2304 if (num_regs
> regs_allocated_size
)
2306 RETALLOC_IF (regstart
, num_regs
, const char *);
2307 RETALLOC_IF (regend
, num_regs
, const char *);
2308 RETALLOC_IF (old_regstart
, num_regs
, const char *);
2309 RETALLOC_IF (old_regend
, num_regs
, const char *);
2310 RETALLOC_IF (best_regstart
, num_regs
, const char *);
2311 RETALLOC_IF (best_regend
, num_regs
, const char *);
2312 RETALLOC_IF (PREFIX(reg_info
), num_regs
, PREFIX(register_info_type
));
2313 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
2314 RETALLOC_IF (PREFIX(reg_info_dummy
), num_regs
, PREFIX(register_info_type
));
2316 regs_allocated_size
= num_regs
;
2320 # endif /* not MATCH_MAY_ALLOCATE */
2322 # ifndef DEFINED_ONCE
2323 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2326 # endif /* not DEFINED_ONCE */
2328 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2329 Returns one of error codes defined in `regex.h', or zero for success.
2331 Assumes the `allocated' (and perhaps `buffer') and `translate'
2332 fields are set in BUFP on entry.
2334 If it succeeds, results are put in BUFP (if it returns an error, the
2335 contents of BUFP are undefined):
2336 `buffer' is the compiled pattern;
2337 `syntax' is set to SYNTAX;
2338 `used' is set to the length of the compiled pattern;
2339 `fastmap_accurate' is zero;
2340 `re_nsub' is the number of subexpressions in PATTERN;
2341 `not_bol' and `not_eol' are zero;
2343 The `fastmap' and `newline_anchor' fields are neither
2344 examined nor set. */
2346 /* Return, freeing storage we allocated. */
2348 # define FREE_STACK_RETURN(value) \
2349 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2351 # define FREE_STACK_RETURN(value) \
2352 return (free (compile_stack.stack), value)
2355 static reg_errcode_t
2356 PREFIX(regex_compile
) (ARG_PREFIX(pattern
), ARG_PREFIX(size
), syntax
, bufp
)
2357 const char *ARG_PREFIX(pattern
);
2358 size_t ARG_PREFIX(size
);
2359 reg_syntax_t syntax
;
2360 struct re_pattern_buffer
*bufp
;
2362 /* We fetch characters from PATTERN here. Even though PATTERN is
2363 `char *' (i.e., signed), we declare these variables as unsigned, so
2364 they can be reliably used as array indices. */
2365 register UCHAR_T c
, c1
;
2368 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2369 CHAR_T
*pattern
, *COMPILED_BUFFER_VAR
;
2371 /* offset buffer for optimization. See convert_mbs_to_wc. */
2372 int *mbs_offset
= NULL
;
2373 /* It hold whether each wchar_t is binary data or not. */
2374 char *is_binary
= NULL
;
2375 /* A flag whether exactn is handling binary data or not. */
2376 char is_exactn_bin
= FALSE
;
2379 /* A random temporary spot in PATTERN. */
2382 /* Points to the end of the buffer, where we should append. */
2383 register UCHAR_T
*b
;
2385 /* Keeps track of unclosed groups. */
2386 compile_stack_type compile_stack
;
2388 /* Points to the current (ending) position in the pattern. */
2393 const CHAR_T
*p
= pattern
;
2394 const CHAR_T
*pend
= pattern
+ size
;
2397 /* How to translate the characters in the pattern. */
2398 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2400 /* Address of the count-byte of the most recently inserted `exactn'
2401 command. This makes it possible to tell if a new exact-match
2402 character can be added to that command or if the character requires
2403 a new `exactn' command. */
2404 UCHAR_T
*pending_exact
= 0;
2406 /* Address of start of the most recently finished expression.
2407 This tells, e.g., postfix * where to find the start of its
2408 operand. Reset at the beginning of groups and alternatives. */
2409 UCHAR_T
*laststart
= 0;
2411 /* Address of beginning of regexp, or inside of last group. */
2414 /* Address of the place where a forward jump should go to the end of
2415 the containing expression. Each alternative of an `or' -- except the
2416 last -- ends with a forward jump of this sort. */
2417 UCHAR_T
*fixup_alt_jump
= 0;
2419 /* Counts open-groups as they are encountered. Remembered for the
2420 matching close-group on the compile stack, so the same register
2421 number is put in the stop_memory as the start_memory. */
2422 regnum_t regnum
= 0;
2425 /* Initialize the wchar_t PATTERN and offset_buffer. */
2426 p
= pend
= pattern
= TALLOC(csize
+ 1, CHAR_T
);
2427 mbs_offset
= TALLOC(csize
+ 1, int);
2428 is_binary
= TALLOC(csize
+ 1, char);
2429 if (pattern
== NULL
|| mbs_offset
== NULL
|| is_binary
== NULL
)
2436 pattern
[csize
] = L
'\0'; /* sentinel */
2437 size
= convert_mbs_to_wcs(pattern
, cpattern
, csize
, mbs_offset
, is_binary
);
2449 DEBUG_PRINT1 ("\nCompiling pattern: ");
2452 unsigned debug_count
;
2454 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2455 PUT_CHAR (pattern
[debug_count
]);
2460 /* Initialize the compile stack. */
2461 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2462 if (compile_stack
.stack
== NULL
)
2472 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2473 compile_stack
.avail
= 0;
2475 /* Initialize the pattern buffer. */
2476 bufp
->syntax
= syntax
;
2477 bufp
->fastmap_accurate
= 0;
2478 bufp
->not_bol
= bufp
->not_eol
= 0;
2480 /* Set `used' to zero, so that if we return an error, the pattern
2481 printer (for debugging) will think there's no pattern. We reset it
2485 /* Always count groups, whether or not bufp->no_sub is set. */
2488 #if !defined emacs && !defined SYNTAX_TABLE
2489 /* Initialize the syntax table. */
2490 init_syntax_once ();
2493 if (bufp
->allocated
== 0)
2496 { /* If zero allocated, but buffer is non-null, try to realloc
2497 enough space. This loses if buffer's address is bogus, but
2498 that is the user's responsibility. */
2500 /* Free bufp->buffer and allocate an array for wchar_t pattern
2503 COMPILED_BUFFER_VAR
= TALLOC (INIT_BUF_SIZE
/sizeof(UCHAR_T
),
2506 RETALLOC (COMPILED_BUFFER_VAR
, INIT_BUF_SIZE
, UCHAR_T
);
2510 { /* Caller did not allocate a buffer. Do it for them. */
2511 COMPILED_BUFFER_VAR
= TALLOC (INIT_BUF_SIZE
/ sizeof(UCHAR_T
),
2515 if (!COMPILED_BUFFER_VAR
) FREE_STACK_RETURN (REG_ESPACE
);
2517 bufp
->buffer
= (char*)COMPILED_BUFFER_VAR
;
2519 bufp
->allocated
= INIT_BUF_SIZE
;
2523 COMPILED_BUFFER_VAR
= (UCHAR_T
*) bufp
->buffer
;
2526 begalt
= b
= COMPILED_BUFFER_VAR
;
2528 /* Loop through the uncompiled pattern until we're at the end. */
2537 if ( /* If at start of pattern, it's an operator. */
2539 /* If context independent, it's an operator. */
2540 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2541 /* Otherwise, depends on what's come before. */
2542 || PREFIX(at_begline_loc_p
) (pattern
, p
, syntax
))
2552 if ( /* If at end of pattern, it's an operator. */
2554 /* If context independent, it's an operator. */
2555 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2556 /* Otherwise, depends on what's next. */
2557 || PREFIX(at_endline_loc_p
) (p
, pend
, syntax
))
2567 if ((syntax
& RE_BK_PLUS_QM
)
2568 || (syntax
& RE_LIMITED_OPS
))
2572 /* If there is no previous pattern... */
2575 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2576 FREE_STACK_RETURN (REG_BADRPT
);
2577 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2582 /* Are we optimizing this jump? */
2583 boolean keep_string_p
= false;
2585 /* 1 means zero (many) matches is allowed. */
2586 char zero_times_ok
= 0, many_times_ok
= 0;
2588 /* If there is a sequence of repetition chars, collapse it
2589 down to just one (the right one). We can't combine
2590 interval operators with these because of, e.g., `a{2}*',
2591 which should only match an even number of `a's. */
2595 zero_times_ok
|= c
!= '+';
2596 many_times_ok
|= c
!= '?';
2604 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
2607 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
2609 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2612 if (!(c1
== '+' || c1
== '?'))
2627 /* If we get here, we found another repeat character. */
2630 /* Star, etc. applied to an empty pattern is equivalent
2631 to an empty pattern. */
2635 /* Now we know whether or not zero matches is allowed
2636 and also whether or not two or more matches is allowed. */
2638 { /* More than one repetition is allowed, so put in at the
2639 end a backward relative jump from `b' to before the next
2640 jump we're going to put in below (which jumps from
2641 laststart to after this jump).
2643 But if we are at the `*' in the exact sequence `.*\n',
2644 insert an unconditional jump backwards to the .,
2645 instead of the beginning of the loop. This way we only
2646 push a failure point once, instead of every time
2647 through the loop. */
2648 assert (p
- 1 > pattern
);
2650 /* Allocate the space for the jump. */
2651 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2653 /* We know we are not at the first character of the pattern,
2654 because laststart was nonzero. And we've already
2655 incremented `p', by the way, to be the character after
2656 the `*'. Do we have to do something analogous here
2657 for null bytes, because of RE_DOT_NOT_NULL? */
2658 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2660 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2661 && !(syntax
& RE_DOT_NEWLINE
))
2662 { /* We have .*\n. */
2663 STORE_JUMP (jump
, b
, laststart
);
2664 keep_string_p
= true;
2667 /* Anything else. */
2668 STORE_JUMP (maybe_pop_jump
, b
, laststart
-
2669 (1 + OFFSET_ADDRESS_SIZE
));
2671 /* We've added more stuff to the buffer. */
2672 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2675 /* On failure, jump from laststart to b + 3, which will be the
2676 end of the buffer after this jump is inserted. */
2677 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2679 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2680 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2682 laststart
, b
+ 1 + OFFSET_ADDRESS_SIZE
);
2684 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2688 /* At least one repetition is required, so insert a
2689 `dummy_failure_jump' before the initial
2690 `on_failure_jump' instruction of the loop. This
2691 effects a skip over that instruction the first time
2692 we hit that loop. */
2693 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2694 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+
2695 2 + 2 * OFFSET_ADDRESS_SIZE
);
2696 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2710 boolean had_char_class
= false;
2712 CHAR_T range_start
= 0xffffffff;
2714 unsigned int range_start
= 0xffffffff;
2716 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2719 /* We assume a charset(_not) structure as a wchar_t array.
2720 charset[0] = (re_opcode_t) charset(_not)
2721 charset[1] = l (= length of char_classes)
2722 charset[2] = m (= length of collating_symbols)
2723 charset[3] = n (= length of equivalence_classes)
2724 charset[4] = o (= length of char_ranges)
2725 charset[5] = p (= length of chars)
2727 charset[6] = char_class (wctype_t)
2728 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2730 charset[l+5] = char_class (wctype_t)
2732 charset[l+6] = collating_symbol (wchar_t)
2734 charset[l+m+5] = collating_symbol (wchar_t)
2735 ifdef _LIBC we use the index if
2736 _NL_COLLATE_SYMB_EXTRAMB instead of
2739 charset[l+m+6] = equivalence_classes (wchar_t)
2741 charset[l+m+n+5] = equivalence_classes (wchar_t)
2742 ifdef _LIBC we use the index in
2743 _NL_COLLATE_WEIGHT instead of
2746 charset[l+m+n+6] = range_start
2747 charset[l+m+n+7] = range_end
2749 charset[l+m+n+2o+4] = range_start
2750 charset[l+m+n+2o+5] = range_end
2751 ifdef _LIBC we use the value looked up
2752 in _NL_COLLATE_COLLSEQ instead of
2755 charset[l+m+n+2o+6] = char
2757 charset[l+m+n+2o+p+5] = char
2761 /* We need at least 6 spaces: the opcode, the length of
2762 char_classes, the length of collating_symbols, the length of
2763 equivalence_classes, the length of char_ranges, the length of
2765 GET_BUFFER_SPACE (6);
2767 /* Save b as laststart. And We use laststart as the pointer
2768 to the first element of the charset here.
2769 In other words, laststart[i] indicates charset[i]. */
2772 /* We test `*p == '^' twice, instead of using an if
2773 statement, so we only need one BUF_PUSH. */
2774 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2778 /* Push the length of char_classes, the length of
2779 collating_symbols, the length of equivalence_classes, the
2780 length of char_ranges and the length of chars. */
2781 BUF_PUSH_3 (0, 0, 0);
2784 /* Remember the first position in the bracket expression. */
2787 /* charset_not matches newline according to a syntax bit. */
2788 if ((re_opcode_t
) b
[-6] == charset_not
2789 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2792 laststart
[5]++; /* Update the length of characters */
2795 /* Read in characters and ranges, setting map bits. */
2798 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2802 /* \ might escape characters inside [...] and [^...]. */
2803 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2805 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2809 laststart
[5]++; /* Update the length of chars */
2814 /* Could be the end of the bracket expression. If it's
2815 not (i.e., when the bracket expression is `[]' so
2816 far), the ']' character bit gets set way below. */
2817 if (c
== ']' && p
!= p1
+ 1)
2820 /* Look ahead to see if it's a range when the last thing
2821 was a character class. */
2822 if (had_char_class
&& c
== '-' && *p
!= ']')
2823 FREE_STACK_RETURN (REG_ERANGE
);
2825 /* Look ahead to see if it's a range when the last thing
2826 was a character: if this is a hyphen not at the
2827 beginning or the end of a list, then it's the range
2830 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2831 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2835 /* Allocate the space for range_start and range_end. */
2836 GET_BUFFER_SPACE (2);
2837 /* Update the pointer to indicate end of buffer. */
2839 ret
= wcs_compile_range (range_start
, &p
, pend
, translate
,
2840 syntax
, b
, laststart
);
2841 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2842 range_start
= 0xffffffff;
2844 else if (p
[0] == '-' && p
[1] != ']')
2845 { /* This handles ranges made up of characters only. */
2848 /* Move past the `-'. */
2850 /* Allocate the space for range_start and range_end. */
2851 GET_BUFFER_SPACE (2);
2852 /* Update the pointer to indicate end of buffer. */
2854 ret
= wcs_compile_range (c
, &p
, pend
, translate
, syntax
, b
,
2856 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2857 range_start
= 0xffffffff;
2860 /* See if we're at the beginning of a possible character
2862 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2863 { /* Leave room for the null. */
2864 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2869 /* If pattern is `[[:'. */
2870 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2875 if ((c
== ':' && *p
== ']') || p
== pend
)
2877 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2880 /* This is in any case an invalid class name. */
2885 /* If isn't a word bracketed by `[:' and `:]':
2886 undo the ending character, the letters, and leave
2887 the leading `:' and `[' (but store them as character). */
2888 if (c
== ':' && *p
== ']')
2893 /* Query the character class as wctype_t. */
2894 wt
= IS_CHAR_CLASS (str
);
2896 FREE_STACK_RETURN (REG_ECTYPE
);
2898 /* Throw away the ] at the end of the character
2902 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2904 /* Allocate the space for character class. */
2905 GET_BUFFER_SPACE(CHAR_CLASS_SIZE
);
2906 /* Update the pointer to indicate end of buffer. */
2907 b
+= CHAR_CLASS_SIZE
;
2908 /* Move data which follow character classes
2909 not to violate the data. */
2910 insert_space(CHAR_CLASS_SIZE
,
2911 laststart
+ 6 + laststart
[1],
2913 alignedp
= ((uintptr_t)(laststart
+ 6 + laststart
[1])
2914 + __alignof__(wctype_t) - 1)
2915 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2916 /* Store the character class. */
2917 *((wctype_t*)alignedp
) = wt
;
2918 /* Update length of char_classes */
2919 laststart
[1] += CHAR_CLASS_SIZE
;
2921 had_char_class
= true;
2930 laststart
[5] += 2; /* Update the length of characters */
2932 had_char_class
= false;
2935 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && (*p
== '='
2938 CHAR_T str
[128]; /* Should be large enough. */
2939 CHAR_T delim
= *p
; /* '=' or '.' */
2942 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
2947 /* If pattern is `[[=' or '[[.'. */
2948 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2953 if ((c
== delim
&& *p
== ']') || p
== pend
)
2955 if (c1
< sizeof (str
) - 1)
2958 /* This is in any case an invalid class name. */
2963 if (c
== delim
&& *p
== ']' && str
[0] != '\0')
2965 unsigned int i
, offset
;
2966 /* If we have no collation data we use the default
2967 collation in which each character is in a class
2968 by itself. It also means that ASCII is the
2969 character set and therefore we cannot have character
2970 with more than one byte in the multibyte
2973 /* If not defined _LIBC, we push the name and
2974 `\0' for the sake of matching performance. */
2975 int datasize
= c1
+ 1;
2983 FREE_STACK_RETURN (REG_ECOLLATE
);
2988 const int32_t *table
;
2989 const int32_t *weights
;
2990 const int32_t *extra
;
2991 const int32_t *indirect
;
2994 /* This #include defines a local function! */
2995 # include <locale/weightwc.h>
2999 /* We push the index for equivalence class. */
3002 table
= (const int32_t *)
3003 _NL_CURRENT (LC_COLLATE
,
3004 _NL_COLLATE_TABLEWC
);
3005 weights
= (const int32_t *)
3006 _NL_CURRENT (LC_COLLATE
,
3007 _NL_COLLATE_WEIGHTWC
);
3008 extra
= (const int32_t *)
3009 _NL_CURRENT (LC_COLLATE
,
3010 _NL_COLLATE_EXTRAWC
);
3011 indirect
= (const int32_t *)
3012 _NL_CURRENT (LC_COLLATE
,
3013 _NL_COLLATE_INDIRECTWC
);
3015 idx
= findidx ((const wint_t**)&cp
);
3016 if (idx
== 0 || cp
< (wint_t*) str
+ c1
)
3017 /* This is no valid character. */
3018 FREE_STACK_RETURN (REG_ECOLLATE
);
3020 str
[0] = (wchar_t)idx
;
3022 else /* delim == '.' */
3024 /* We push collation sequence value
3025 for collating symbol. */
3027 const int32_t *symb_table
;
3028 const unsigned char *extra
;
3035 /* We have to convert the name to a single-byte
3036 string. This is possible since the names
3037 consist of ASCII characters and the internal
3038 representation is UCS4. */
3039 for (i
= 0; i
< c1
; ++i
)
3040 char_str
[i
] = str
[i
];
3043 _NL_CURRENT_WORD (LC_COLLATE
,
3044 _NL_COLLATE_SYMB_HASH_SIZEMB
);
3045 symb_table
= (const int32_t *)
3046 _NL_CURRENT (LC_COLLATE
,
3047 _NL_COLLATE_SYMB_TABLEMB
);
3048 extra
= (const unsigned char *)
3049 _NL_CURRENT (LC_COLLATE
,
3050 _NL_COLLATE_SYMB_EXTRAMB
);
3052 /* Locate the character in the hashing table. */
3053 hash
= elem_hash (char_str
, c1
);
3056 elem
= hash
% table_size
;
3057 second
= hash
% (table_size
- 2);
3058 while (symb_table
[2 * elem
] != 0)
3060 /* First compare the hashing value. */
3061 if (symb_table
[2 * elem
] == hash
3062 && c1
== extra
[symb_table
[2 * elem
+ 1]]
3063 && memcmp (char_str
,
3064 &extra
[symb_table
[2 * elem
+ 1]
3067 /* Yep, this is the entry. */
3068 idx
= symb_table
[2 * elem
+ 1];
3069 idx
+= 1 + extra
[idx
];
3077 if (symb_table
[2 * elem
] != 0)
3079 /* Compute the index of the byte sequence
3081 idx
+= 1 + extra
[idx
];
3082 /* Adjust for the alignment. */
3083 idx
= (idx
+ 3) & ~3;
3085 str
[0] = (wchar_t) idx
+ 4;
3087 else if (symb_table
[2 * elem
] == 0 && c1
== 1)
3089 /* No valid character. Match it as a
3090 single byte character. */
3091 had_char_class
= false;
3093 /* Update the length of characters */
3095 range_start
= str
[0];
3097 /* Throw away the ] at the end of the
3098 collating symbol. */
3100 /* exit from the switch block. */
3104 FREE_STACK_RETURN (REG_ECOLLATE
);
3109 /* Throw away the ] at the end of the equivalence
3110 class (or collating symbol). */
3113 /* Allocate the space for the equivalence class
3114 (or collating symbol) (and '\0' if needed). */
3115 GET_BUFFER_SPACE(datasize
);
3116 /* Update the pointer to indicate end of buffer. */
3120 { /* equivalence class */
3121 /* Calculate the offset of char_ranges,
3122 which is next to equivalence_classes. */
3123 offset
= laststart
[1] + laststart
[2]
3126 insert_space(datasize
, laststart
+ offset
, b
- 1);
3128 /* Write the equivalence_class and \0. */
3129 for (i
= 0 ; i
< datasize
; i
++)
3130 laststart
[offset
+ i
] = str
[i
];
3132 /* Update the length of equivalence_classes. */
3133 laststart
[3] += datasize
;
3134 had_char_class
= true;
3136 else /* delim == '.' */
3137 { /* collating symbol */
3138 /* Calculate the offset of the equivalence_classes,
3139 which is next to collating_symbols. */
3140 offset
= laststart
[1] + laststart
[2] + 6;
3141 /* Insert space and write the collationg_symbol
3143 insert_space(datasize
, laststart
+ offset
, b
-1);
3144 for (i
= 0 ; i
< datasize
; i
++)
3145 laststart
[offset
+ i
] = str
[i
];
3147 /* In re_match_2_internal if range_start < -1, we
3148 assume -range_start is the offset of the
3149 collating symbol which is specified as
3150 the character of the range start. So we assign
3151 -(laststart[1] + laststart[2] + 6) to
3153 range_start
= -(laststart
[1] + laststart
[2] + 6);
3154 /* Update the length of collating_symbol. */
3155 laststart
[2] += datasize
;
3156 had_char_class
= false;
3166 laststart
[5] += 2; /* Update the length of characters */
3167 range_start
= delim
;
3168 had_char_class
= false;
3173 had_char_class
= false;
3175 laststart
[5]++; /* Update the length of characters */
3181 /* Ensure that we have enough space to push a charset: the
3182 opcode, the length count, and the bitset; 34 bytes in all. */
3183 GET_BUFFER_SPACE (34);
3187 /* We test `*p == '^' twice, instead of using an if
3188 statement, so we only need one BUF_PUSH. */
3189 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
3193 /* Remember the first position in the bracket expression. */
3196 /* Push the number of bytes in the bitmap. */
3197 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
3199 /* Clear the whole map. */
3200 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
3202 /* charset_not matches newline according to a syntax bit. */
3203 if ((re_opcode_t
) b
[-2] == charset_not
3204 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
3205 SET_LIST_BIT ('\n');
3207 /* Read in characters and ranges, setting map bits. */
3210 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3214 /* \ might escape characters inside [...] and [^...]. */
3215 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
3217 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3225 /* Could be the end of the bracket expression. If it's
3226 not (i.e., when the bracket expression is `[]' so
3227 far), the ']' character bit gets set way below. */
3228 if (c
== ']' && p
!= p1
+ 1)
3231 /* Look ahead to see if it's a range when the last thing
3232 was a character class. */
3233 if (had_char_class
&& c
== '-' && *p
!= ']')
3234 FREE_STACK_RETURN (REG_ERANGE
);
3236 /* Look ahead to see if it's a range when the last thing
3237 was a character: if this is a hyphen not at the
3238 beginning or the end of a list, then it's the range
3241 && !(p
- 2 >= pattern
&& p
[-2] == '[')
3242 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
3246 = byte_compile_range (range_start
, &p
, pend
, translate
,
3248 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
3249 range_start
= 0xffffffff;
3252 else if (p
[0] == '-' && p
[1] != ']')
3253 { /* This handles ranges made up of characters only. */
3256 /* Move past the `-'. */
3259 ret
= byte_compile_range (c
, &p
, pend
, translate
, syntax
, b
);
3260 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
3261 range_start
= 0xffffffff;
3264 /* See if we're at the beginning of a possible character
3267 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
3268 { /* Leave room for the null. */
3269 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
3274 /* If pattern is `[[:'. */
3275 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3280 if ((c
== ':' && *p
== ']') || p
== pend
)
3282 if (c1
< CHAR_CLASS_MAX_LENGTH
)
3285 /* This is in any case an invalid class name. */
3290 /* If isn't a word bracketed by `[:' and `:]':
3291 undo the ending character, the letters, and leave
3292 the leading `:' and `[' (but set bits for them). */
3293 if (c
== ':' && *p
== ']')
3295 # if defined _LIBC || WIDE_CHAR_SUPPORT
3296 boolean is_lower
= STREQ (str
, "lower");
3297 boolean is_upper
= STREQ (str
, "upper");
3301 wt
= IS_CHAR_CLASS (str
);
3303 FREE_STACK_RETURN (REG_ECTYPE
);
3305 /* Throw away the ] at the end of the character
3309 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3311 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
3314 if (__iswctype (__btowc (ch
), wt
))
3317 if (iswctype (btowc (ch
), wt
))
3321 if (translate
&& (is_upper
|| is_lower
)
3322 && (ISUPPER (ch
) || ISLOWER (ch
)))
3326 had_char_class
= true;
3329 boolean is_alnum
= STREQ (str
, "alnum");
3330 boolean is_alpha
= STREQ (str
, "alpha");
3331 boolean is_blank
= STREQ (str
, "blank");
3332 boolean is_cntrl
= STREQ (str
, "cntrl");
3333 boolean is_digit
= STREQ (str
, "digit");
3334 boolean is_graph
= STREQ (str
, "graph");
3335 boolean is_lower
= STREQ (str
, "lower");
3336 boolean is_print
= STREQ (str
, "print");
3337 boolean is_punct
= STREQ (str
, "punct");
3338 boolean is_space
= STREQ (str
, "space");
3339 boolean is_upper
= STREQ (str
, "upper");
3340 boolean is_xdigit
= STREQ (str
, "xdigit");
3342 if (!IS_CHAR_CLASS (str
))
3343 FREE_STACK_RETURN (REG_ECTYPE
);
3345 /* Throw away the ] at the end of the character
3349 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3351 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
3353 /* This was split into 3 if's to
3354 avoid an arbitrary limit in some compiler. */
3355 if ( (is_alnum
&& ISALNUM (ch
))
3356 || (is_alpha
&& ISALPHA (ch
))
3357 || (is_blank
&& ISBLANK (ch
))
3358 || (is_cntrl
&& ISCNTRL (ch
)))
3360 if ( (is_digit
&& ISDIGIT (ch
))
3361 || (is_graph
&& ISGRAPH (ch
))
3362 || (is_lower
&& ISLOWER (ch
))
3363 || (is_print
&& ISPRINT (ch
)))
3365 if ( (is_punct
&& ISPUNCT (ch
))
3366 || (is_space
&& ISSPACE (ch
))
3367 || (is_upper
&& ISUPPER (ch
))
3368 || (is_xdigit
&& ISXDIGIT (ch
)))
3370 if ( translate
&& (is_upper
|| is_lower
)
3371 && (ISUPPER (ch
) || ISLOWER (ch
)))
3374 had_char_class
= true;
3375 # endif /* libc || wctype.h */
3385 had_char_class
= false;
3388 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '=')
3390 unsigned char str
[MB_LEN_MAX
+ 1];
3393 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3399 /* If pattern is `[[='. */
3400 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3405 if ((c
== '=' && *p
== ']') || p
== pend
)
3407 if (c1
< MB_LEN_MAX
)
3410 /* This is in any case an invalid class name. */
3415 if (c
== '=' && *p
== ']' && str
[0] != '\0')
3417 /* If we have no collation data we use the default
3418 collation in which each character is in a class
3419 by itself. It also means that ASCII is the
3420 character set and therefore we cannot have character
3421 with more than one byte in the multibyte
3428 FREE_STACK_RETURN (REG_ECOLLATE
);
3430 /* Throw away the ] at the end of the equivalence
3434 /* Set the bit for the character. */
3435 SET_LIST_BIT (str
[0]);
3440 /* Try to match the byte sequence in `str' against
3441 those known to the collate implementation.
3442 First find out whether the bytes in `str' are
3443 actually from exactly one character. */
3444 const int32_t *table
;
3445 const unsigned char *weights
;
3446 const unsigned char *extra
;
3447 const int32_t *indirect
;
3449 const unsigned char *cp
= str
;
3452 /* This #include defines a local function! */
3453 # include <locale/weight.h>
3455 table
= (const int32_t *)
3456 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEMB
);
3457 weights
= (const unsigned char *)
3458 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTMB
);
3459 extra
= (const unsigned char *)
3460 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAMB
);
3461 indirect
= (const int32_t *)
3462 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTMB
);
3464 idx
= findidx (&cp
);
3465 if (idx
== 0 || cp
< str
+ c1
)
3466 /* This is no valid character. */
3467 FREE_STACK_RETURN (REG_ECOLLATE
);
3469 /* Throw away the ] at the end of the equivalence
3473 /* Now we have to go throught the whole table
3474 and find all characters which have the same
3477 XXX Note that this is not entirely correct.
3478 we would have to match multibyte sequences
3479 but this is not possible with the current
3481 for (ch
= 1; ch
< 256; ++ch
)
3482 /* XXX This test would have to be changed if we
3483 would allow matching multibyte sequences. */
3486 int32_t idx2
= table
[ch
];
3487 size_t len
= weights
[idx2
];
3489 /* Test whether the lenghts match. */
3490 if (weights
[idx
] == len
)
3492 /* They do. New compare the bytes of
3497 && (weights
[idx
+ 1 + cnt
]
3498 == weights
[idx2
+ 1 + cnt
]))
3502 /* They match. Mark the character as
3509 had_char_class
= true;
3519 had_char_class
= false;
3522 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '.')
3524 unsigned char str
[128]; /* Should be large enough. */
3527 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3533 /* If pattern is `[[.'. */
3534 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3539 if ((c
== '.' && *p
== ']') || p
== pend
)
3541 if (c1
< sizeof (str
))
3544 /* This is in any case an invalid class name. */
3549 if (c
== '.' && *p
== ']' && str
[0] != '\0')
3551 /* If we have no collation data we use the default
3552 collation in which each character is the name
3553 for its own class which contains only the one
3554 character. It also means that ASCII is the
3555 character set and therefore we cannot have character
3556 with more than one byte in the multibyte
3563 FREE_STACK_RETURN (REG_ECOLLATE
);
3565 /* Throw away the ] at the end of the equivalence
3569 /* Set the bit for the character. */
3570 SET_LIST_BIT (str
[0]);
3571 range_start
= ((const unsigned char *) str
)[0];
3576 /* Try to match the byte sequence in `str' against
3577 those known to the collate implementation.
3578 First find out whether the bytes in `str' are
3579 actually from exactly one character. */
3581 const int32_t *symb_table
;
3582 const unsigned char *extra
;
3589 _NL_CURRENT_WORD (LC_COLLATE
,
3590 _NL_COLLATE_SYMB_HASH_SIZEMB
);
3591 symb_table
= (const int32_t *)
3592 _NL_CURRENT (LC_COLLATE
,
3593 _NL_COLLATE_SYMB_TABLEMB
);
3594 extra
= (const unsigned char *)
3595 _NL_CURRENT (LC_COLLATE
,
3596 _NL_COLLATE_SYMB_EXTRAMB
);
3598 /* Locate the character in the hashing table. */
3599 hash
= elem_hash (str
, c1
);
3602 elem
= hash
% table_size
;
3603 second
= hash
% (table_size
- 2);
3604 while (symb_table
[2 * elem
] != 0)
3606 /* First compare the hashing value. */
3607 if (symb_table
[2 * elem
] == hash
3608 && c1
== extra
[symb_table
[2 * elem
+ 1]]
3610 &extra
[symb_table
[2 * elem
+ 1]
3614 /* Yep, this is the entry. */
3615 idx
= symb_table
[2 * elem
+ 1];
3616 idx
+= 1 + extra
[idx
];
3624 if (symb_table
[2 * elem
] == 0)
3625 /* This is no valid character. */
3626 FREE_STACK_RETURN (REG_ECOLLATE
);
3628 /* Throw away the ] at the end of the equivalence
3632 /* Now add the multibyte character(s) we found
3635 XXX Note that this is not entirely correct.
3636 we would have to match multibyte sequences
3637 but this is not possible with the current
3638 implementation. Also, we have to match
3639 collating symbols, which expand to more than
3640 one file, as a whole and not allow the
3641 individual bytes. */
3644 range_start
= extra
[idx
];
3647 SET_LIST_BIT (extra
[idx
]);
3652 had_char_class
= false;
3662 had_char_class
= false;
3667 had_char_class
= false;
3673 /* Discard any (non)matching list bytes that are all 0 at the
3674 end of the map. Decrease the map-length byte too. */
3675 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3684 if (syntax
& RE_NO_BK_PARENS
)
3691 if (syntax
& RE_NO_BK_PARENS
)
3698 if (syntax
& RE_NEWLINE_ALT
)
3705 if (syntax
& RE_NO_BK_VBAR
)
3712 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3713 goto handle_interval
;
3719 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3721 /* Do not translate the character after the \, so that we can
3722 distinguish, e.g., \B from \b, even if we normally would
3723 translate, e.g., B to b. */
3729 if (syntax
& RE_NO_BK_PARENS
)
3730 goto normal_backslash
;
3736 if (COMPILE_STACK_FULL
)
3738 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3739 compile_stack_elt_t
);
3740 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3742 compile_stack
.size
<<= 1;
3745 /* These are the values to restore when we hit end of this
3746 group. They are all relative offsets, so that if the
3747 whole pattern moves because of realloc, they will still
3749 COMPILE_STACK_TOP
.begalt_offset
= begalt
- COMPILED_BUFFER_VAR
;
3750 COMPILE_STACK_TOP
.fixup_alt_jump
3751 = fixup_alt_jump
? fixup_alt_jump
- COMPILED_BUFFER_VAR
+ 1 : 0;
3752 COMPILE_STACK_TOP
.laststart_offset
= b
- COMPILED_BUFFER_VAR
;
3753 COMPILE_STACK_TOP
.regnum
= regnum
;
3755 /* We will eventually replace the 0 with the number of
3756 groups inner to this one. But do not push a
3757 start_memory for groups beyond the last one we can
3758 represent in the compiled pattern. */
3759 if (regnum
<= MAX_REGNUM
)
3761 COMPILE_STACK_TOP
.inner_group_offset
= b
3762 - COMPILED_BUFFER_VAR
+ 2;
3763 BUF_PUSH_3 (start_memory
, regnum
, 0);
3766 compile_stack
.avail
++;
3771 /* If we've reached MAX_REGNUM groups, then this open
3772 won't actually generate any code, so we'll have to
3773 clear pending_exact explicitly. */
3779 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3781 if (COMPILE_STACK_EMPTY
)
3783 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3784 goto normal_backslash
;
3786 FREE_STACK_RETURN (REG_ERPAREN
);
3791 { /* Push a dummy failure point at the end of the
3792 alternative for a possible future
3793 `pop_failure_jump' to pop. See comments at
3794 `push_dummy_failure' in `re_match_2'. */
3795 BUF_PUSH (push_dummy_failure
);
3797 /* We allocated space for this jump when we assigned
3798 to `fixup_alt_jump', in the `handle_alt' case below. */
3799 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
3802 /* See similar code for backslashed left paren above. */
3803 if (COMPILE_STACK_EMPTY
)
3805 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3808 FREE_STACK_RETURN (REG_ERPAREN
);
3811 /* Since we just checked for an empty stack above, this
3812 ``can't happen''. */
3813 assert (compile_stack
.avail
!= 0);
3815 /* We don't just want to restore into `regnum', because
3816 later groups should continue to be numbered higher,
3817 as in `(ab)c(de)' -- the second group is #2. */
3818 regnum_t this_group_regnum
;
3820 compile_stack
.avail
--;
3821 begalt
= COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.begalt_offset
;
3823 = COMPILE_STACK_TOP
.fixup_alt_jump
3824 ? COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3826 laststart
= COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.laststart_offset
;
3827 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
3828 /* If we've reached MAX_REGNUM groups, then this open
3829 won't actually generate any code, so we'll have to
3830 clear pending_exact explicitly. */
3833 /* We're at the end of the group, so now we know how many
3834 groups were inside this one. */
3835 if (this_group_regnum
<= MAX_REGNUM
)
3837 UCHAR_T
*inner_group_loc
3838 = COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.inner_group_offset
;
3840 *inner_group_loc
= regnum
- this_group_regnum
;
3841 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
3842 regnum
- this_group_regnum
);
3848 case '|': /* `\|'. */
3849 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3850 goto normal_backslash
;
3852 if (syntax
& RE_LIMITED_OPS
)
3855 /* Insert before the previous alternative a jump which
3856 jumps to this alternative if the former fails. */
3857 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3858 INSERT_JUMP (on_failure_jump
, begalt
,
3859 b
+ 2 + 2 * OFFSET_ADDRESS_SIZE
);
3861 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3863 /* The alternative before this one has a jump after it
3864 which gets executed if it gets matched. Adjust that
3865 jump so it will jump to this alternative's analogous
3866 jump (put in below, which in turn will jump to the next
3867 (if any) alternative's such jump, etc.). The last such
3868 jump jumps to the correct final destination. A picture:
3874 If we are at `b', then fixup_alt_jump right now points to a
3875 three-byte space after `a'. We'll put in the jump, set
3876 fixup_alt_jump to right after `b', and leave behind three
3877 bytes which we'll fill in when we get to after `c'. */
3880 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
3882 /* Mark and leave space for a jump after this alternative,
3883 to be filled in later either by next alternative or
3884 when know we're at the end of a series of alternatives. */
3886 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3887 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3895 /* If \{ is a literal. */
3896 if (!(syntax
& RE_INTERVALS
)
3897 /* If we're at `\{' and it's not the open-interval
3899 || (syntax
& RE_NO_BK_BRACES
))
3900 goto normal_backslash
;
3904 /* If got here, then the syntax allows intervals. */
3906 /* At least (most) this many matches must be made. */
3907 int lower_bound
= -1, upper_bound
= -1;
3909 /* Place in the uncompiled pattern (i.e., just after
3910 the '{') to go back to if the interval is invalid. */
3911 const CHAR_T
*beg_interval
= p
;
3914 goto invalid_interval
;
3916 GET_UNSIGNED_NUMBER (lower_bound
);
3920 GET_UNSIGNED_NUMBER (upper_bound
);
3921 if (upper_bound
< 0)
3922 upper_bound
= RE_DUP_MAX
;
3925 /* Interval such as `{1}' => match exactly once. */
3926 upper_bound
= lower_bound
;
3928 if (! (0 <= lower_bound
&& lower_bound
<= upper_bound
))
3929 goto invalid_interval
;
3931 if (!(syntax
& RE_NO_BK_BRACES
))
3933 if (c
!= '\\' || p
== pend
)
3934 goto invalid_interval
;
3939 goto invalid_interval
;
3941 /* If it's invalid to have no preceding re. */
3944 if (syntax
& RE_CONTEXT_INVALID_OPS
3945 && !(syntax
& RE_INVALID_INTERVAL_ORD
))
3946 FREE_STACK_RETURN (REG_BADRPT
);
3947 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3950 goto unfetch_interval
;
3953 /* We just parsed a valid interval. */
3955 if (RE_DUP_MAX
< upper_bound
)
3956 FREE_STACK_RETURN (REG_BADBR
);
3958 /* If the upper bound is zero, don't want to succeed at
3959 all; jump from `laststart' to `b + 3', which will be
3960 the end of the buffer after we insert the jump. */
3961 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3962 instead of 'b + 3'. */
3963 if (upper_bound
== 0)
3965 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3966 INSERT_JUMP (jump
, laststart
, b
+ 1
3967 + OFFSET_ADDRESS_SIZE
);
3968 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3971 /* Otherwise, we have a nontrivial interval. When
3972 we're all done, the pattern will look like:
3973 set_number_at <jump count> <upper bound>
3974 set_number_at <succeed_n count> <lower bound>
3975 succeed_n <after jump addr> <succeed_n count>
3977 jump_n <succeed_n addr> <jump count>
3978 (The upper bound and `jump_n' are omitted if
3979 `upper_bound' is 1, though.) */
3981 { /* If the upper bound is > 1, we need to insert
3982 more at the end of the loop. */
3983 unsigned nbytes
= 2 + 4 * OFFSET_ADDRESS_SIZE
+
3984 (upper_bound
> 1) * (2 + 4 * OFFSET_ADDRESS_SIZE
);
3986 GET_BUFFER_SPACE (nbytes
);
3988 /* Initialize lower bound of the `succeed_n', even
3989 though it will be set during matching by its
3990 attendant `set_number_at' (inserted next),
3991 because `re_compile_fastmap' needs to know.
3992 Jump to the `jump_n' we might insert below. */
3993 INSERT_JUMP2 (succeed_n
, laststart
,
3994 b
+ 1 + 2 * OFFSET_ADDRESS_SIZE
3995 + (upper_bound
> 1) * (1 + 2 * OFFSET_ADDRESS_SIZE
)
3997 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
3999 /* Code to initialize the lower bound. Insert
4000 before the `succeed_n'. The `5' is the last two
4001 bytes of this `set_number_at', plus 3 bytes of
4002 the following `succeed_n'. */
4003 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
4004 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
4005 of the following `succeed_n'. */
4006 PREFIX(insert_op2
) (set_number_at
, laststart
, 1
4007 + 2 * OFFSET_ADDRESS_SIZE
, lower_bound
, b
);
4008 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
4010 if (upper_bound
> 1)
4011 { /* More than one repetition is allowed, so
4012 append a backward jump to the `succeed_n'
4013 that starts this interval.
4015 When we've reached this during matching,
4016 we'll have matched the interval once, so
4017 jump back only `upper_bound - 1' times. */
4018 STORE_JUMP2 (jump_n
, b
, laststart
4019 + 2 * OFFSET_ADDRESS_SIZE
+ 1,
4021 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
4023 /* The location we want to set is the second
4024 parameter of the `jump_n'; that is `b-2' as
4025 an absolute address. `laststart' will be
4026 the `set_number_at' we're about to insert;
4027 `laststart+3' the number to set, the source
4028 for the relative address. But we are
4029 inserting into the middle of the pattern --
4030 so everything is getting moved up by 5.
4031 Conclusion: (b - 2) - (laststart + 3) + 5,
4032 i.e., b - laststart.
4034 We insert this at the beginning of the loop
4035 so that if we fail during matching, we'll
4036 reinitialize the bounds. */
4037 PREFIX(insert_op2
) (set_number_at
, laststart
,
4039 upper_bound
- 1, b
);
4040 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
4047 if (!(syntax
& RE_INVALID_INTERVAL_ORD
))
4048 FREE_STACK_RETURN (p
== pend
? REG_EBRACE
: REG_BADBR
);
4050 /* Match the characters as literals. */
4053 if (syntax
& RE_NO_BK_BRACES
)
4056 goto normal_backslash
;
4060 /* There is no way to specify the before_dot and after_dot
4061 operators. rms says this is ok. --karl */
4069 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
4075 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
4081 if (syntax
& RE_NO_GNU_OPS
)
4084 BUF_PUSH (wordchar
);
4089 if (syntax
& RE_NO_GNU_OPS
)
4092 BUF_PUSH (notwordchar
);
4097 if (syntax
& RE_NO_GNU_OPS
)
4103 if (syntax
& RE_NO_GNU_OPS
)
4109 if (syntax
& RE_NO_GNU_OPS
)
4111 BUF_PUSH (wordbound
);
4115 if (syntax
& RE_NO_GNU_OPS
)
4117 BUF_PUSH (notwordbound
);
4121 if (syntax
& RE_NO_GNU_OPS
)
4127 if (syntax
& RE_NO_GNU_OPS
)
4132 case '1': case '2': case '3': case '4': case '5':
4133 case '6': case '7': case '8': case '9':
4134 if (syntax
& RE_NO_BK_REFS
)
4140 FREE_STACK_RETURN (REG_ESUBREG
);
4142 /* Can't back reference to a subexpression if inside of it. */
4143 if (group_in_compile_stack (compile_stack
, (regnum_t
) c1
))
4147 BUF_PUSH_2 (duplicate
, c1
);
4153 if (syntax
& RE_BK_PLUS_QM
)
4156 goto normal_backslash
;
4160 /* You might think it would be useful for \ to mean
4161 not to translate; but if we don't translate it
4162 it will never match anything. */
4170 /* Expects the character in `c'. */
4172 /* If no exactn currently being built. */
4175 /* If last exactn handle binary(or character) and
4176 new exactn handle character(or binary). */
4177 || is_exactn_bin
!= is_binary
[p
- 1 - pattern
]
4180 /* If last exactn not at current position. */
4181 || pending_exact
+ *pending_exact
+ 1 != b
4183 /* We have only one byte following the exactn for the count. */
4184 || *pending_exact
== (1 << BYTEWIDTH
) - 1
4186 /* If followed by a repetition operator. */
4187 || *p
== '*' || *p
== '^'
4188 || ((syntax
& RE_BK_PLUS_QM
)
4189 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
4190 : (*p
== '+' || *p
== '?'))
4191 || ((syntax
& RE_INTERVALS
)
4192 && ((syntax
& RE_NO_BK_BRACES
)
4194 : (p
[0] == '\\' && p
[1] == '{'))))
4196 /* Start building a new exactn. */
4201 /* Is this exactn binary data or character? */
4202 is_exactn_bin
= is_binary
[p
- 1 - pattern
];
4204 BUF_PUSH_2 (exactn_bin
, 0);
4206 BUF_PUSH_2 (exactn
, 0);
4208 BUF_PUSH_2 (exactn
, 0);
4210 pending_exact
= b
- 1;
4217 } /* while p != pend */
4220 /* Through the pattern now. */
4223 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
4225 if (!COMPILE_STACK_EMPTY
)
4226 FREE_STACK_RETURN (REG_EPAREN
);
4228 /* If we don't want backtracking, force success
4229 the first time we reach the end of the compiled pattern. */
4230 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
4238 free (compile_stack
.stack
);
4240 /* We have succeeded; set the length of the buffer. */
4242 bufp
->used
= (uintptr_t) b
- (uintptr_t) COMPILED_BUFFER_VAR
;
4244 bufp
->used
= b
- bufp
->buffer
;
4250 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4251 PREFIX(print_compiled_pattern
) (bufp
);
4255 #ifndef MATCH_MAY_ALLOCATE
4256 /* Initialize the failure stack to the largest possible stack. This
4257 isn't necessary unless we're trying to avoid calling alloca in
4258 the search and match routines. */
4260 int num_regs
= bufp
->re_nsub
+ 1;
4262 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4263 is strictly greater than re_max_failures, the largest possible stack
4264 is 2 * re_max_failures failure points. */
4265 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
4267 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
4270 if (! fail_stack
.stack
)
4272 = (PREFIX(fail_stack_elt_t
) *) xmalloc (fail_stack
.size
4273 * sizeof (PREFIX(fail_stack_elt_t
)));
4276 = (PREFIX(fail_stack_elt_t
) *) xrealloc (fail_stack
.stack
,
4278 * sizeof (PREFIX(fail_stack_elt_t
))));
4279 # else /* not emacs */
4280 if (! fail_stack
.stack
)
4282 = (PREFIX(fail_stack_elt_t
) *) malloc (fail_stack
.size
4283 * sizeof (PREFIX(fail_stack_elt_t
)));
4286 = (PREFIX(fail_stack_elt_t
) *) realloc (fail_stack
.stack
,
4288 * sizeof (PREFIX(fail_stack_elt_t
))));
4289 # endif /* not emacs */
4292 PREFIX(regex_grow_registers
) (num_regs
);
4294 #endif /* not MATCH_MAY_ALLOCATE */
4297 } /* regex_compile */
4299 /* Subroutines for `regex_compile'. */
4301 /* Store OP at LOC followed by two-byte integer parameter ARG. */
4302 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4305 PREFIX(store_op1
) (op
, loc
, arg
)
4310 *loc
= (UCHAR_T
) op
;
4311 STORE_NUMBER (loc
+ 1, arg
);
4315 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4316 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4319 PREFIX(store_op2
) (op
, loc
, arg1
, arg2
)
4324 *loc
= (UCHAR_T
) op
;
4325 STORE_NUMBER (loc
+ 1, arg1
);
4326 STORE_NUMBER (loc
+ 1 + OFFSET_ADDRESS_SIZE
, arg2
);
4330 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
4331 for OP followed by two-byte integer parameter ARG. */
4332 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4335 PREFIX(insert_op1
) (op
, loc
, arg
, end
)
4341 register UCHAR_T
*pfrom
= end
;
4342 register UCHAR_T
*pto
= end
+ 1 + OFFSET_ADDRESS_SIZE
;
4344 while (pfrom
!= loc
)
4347 PREFIX(store_op1
) (op
, loc
, arg
);
4351 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4352 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4355 PREFIX(insert_op2
) (op
, loc
, arg1
, arg2
, end
)
4361 register UCHAR_T
*pfrom
= end
;
4362 register UCHAR_T
*pto
= end
+ 1 + 2 * OFFSET_ADDRESS_SIZE
;
4364 while (pfrom
!= loc
)
4367 PREFIX(store_op2
) (op
, loc
, arg1
, arg2
);
4371 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
4372 after an alternative or a begin-subexpression. We assume there is at
4373 least one character before the ^. */
4376 PREFIX(at_begline_loc_p
) (pattern
, p
, syntax
)
4377 const CHAR_T
*pattern
, *p
;
4378 reg_syntax_t syntax
;
4380 const CHAR_T
*prev
= p
- 2;
4381 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
4384 /* After a subexpression? */
4385 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
4386 /* After an alternative? */
4387 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
4391 /* The dual of at_begline_loc_p. This one is for $. We assume there is
4392 at least one character after the $, i.e., `P < PEND'. */
4395 PREFIX(at_endline_loc_p
) (p
, pend
, syntax
)
4396 const CHAR_T
*p
, *pend
;
4397 reg_syntax_t syntax
;
4399 const CHAR_T
*next
= p
;
4400 boolean next_backslash
= *next
== '\\';
4401 const CHAR_T
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
4404 /* Before a subexpression? */
4405 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
4406 : next_backslash
&& next_next
&& *next_next
== ')')
4407 /* Before an alternative? */
4408 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
4409 : next_backslash
&& next_next
&& *next_next
== '|');
4412 #else /* not INSIDE_RECURSION */
4414 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4415 false if it's not. */
4418 group_in_compile_stack (compile_stack
, regnum
)
4419 compile_stack_type compile_stack
;
4424 for (this_element
= compile_stack
.avail
- 1;
4427 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
4432 #endif /* not INSIDE_RECURSION */
4434 #ifdef INSIDE_RECURSION
4437 /* This insert space, which size is "num", into the pattern at "loc".
4438 "end" must point the end of the allocated buffer. */
4440 insert_space (num
, loc
, end
)
4445 register CHAR_T
*pto
= end
;
4446 register CHAR_T
*pfrom
= end
- num
;
4448 while (pfrom
>= loc
)
4454 static reg_errcode_t
4455 wcs_compile_range (range_start_char
, p_ptr
, pend
, translate
, syntax
, b
,
4457 CHAR_T range_start_char
;
4458 const CHAR_T
**p_ptr
, *pend
;
4459 CHAR_T
*char_set
, *b
;
4460 RE_TRANSLATE_TYPE translate
;
4461 reg_syntax_t syntax
;
4463 const CHAR_T
*p
= *p_ptr
;
4464 CHAR_T range_start
, range_end
;
4468 uint32_t start_val
, end_val
;
4474 nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
4477 const char *collseq
= (const char *) _NL_CURRENT(LC_COLLATE
,
4478 _NL_COLLATE_COLLSEQWC
);
4479 const unsigned char *extra
= (const unsigned char *)
4480 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
4482 if (range_start_char
< -1)
4484 /* range_start is a collating symbol. */
4486 /* Retreive the index and get collation sequence value. */
4487 wextra
= (int32_t*)(extra
+ char_set
[-range_start_char
]);
4488 start_val
= wextra
[1 + *wextra
];
4491 start_val
= collseq_table_lookup(collseq
, TRANSLATE(range_start_char
));
4493 end_val
= collseq_table_lookup (collseq
, TRANSLATE (p
[0]));
4495 /* Report an error if the range is empty and the syntax prohibits
4497 ret
= ((syntax
& RE_NO_EMPTY_RANGES
)
4498 && (start_val
> end_val
))? REG_ERANGE
: REG_NOERROR
;
4500 /* Insert space to the end of the char_ranges. */
4501 insert_space(2, b
- char_set
[5] - 2, b
- 1);
4502 *(b
- char_set
[5] - 2) = (wchar_t)start_val
;
4503 *(b
- char_set
[5] - 1) = (wchar_t)end_val
;
4504 char_set
[4]++; /* ranges_index */
4509 range_start
= (range_start_char
>= 0)? TRANSLATE (range_start_char
):
4511 range_end
= TRANSLATE (p
[0]);
4512 /* Report an error if the range is empty and the syntax prohibits
4514 ret
= ((syntax
& RE_NO_EMPTY_RANGES
)
4515 && (range_start
> range_end
))? REG_ERANGE
: REG_NOERROR
;
4517 /* Insert space to the end of the char_ranges. */
4518 insert_space(2, b
- char_set
[5] - 2, b
- 1);
4519 *(b
- char_set
[5] - 2) = range_start
;
4520 *(b
- char_set
[5] - 1) = range_end
;
4521 char_set
[4]++; /* ranges_index */
4523 /* Have to increment the pointer into the pattern string, so the
4524 caller isn't still at the ending character. */
4530 /* Read the ending character of a range (in a bracket expression) from the
4531 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4532 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4533 Then we set the translation of all bits between the starting and
4534 ending characters (inclusive) in the compiled pattern B.
4536 Return an error code.
4538 We use these short variable names so we can use the same macros as
4539 `regex_compile' itself. */
4541 static reg_errcode_t
4542 byte_compile_range (range_start_char
, p_ptr
, pend
, translate
, syntax
, b
)
4543 unsigned int range_start_char
;
4544 const char **p_ptr
, *pend
;
4545 RE_TRANSLATE_TYPE translate
;
4546 reg_syntax_t syntax
;
4550 const char *p
= *p_ptr
;
4553 const unsigned char *collseq
;
4554 unsigned int start_colseq
;
4555 unsigned int end_colseq
;
4563 /* Have to increment the pointer into the pattern string, so the
4564 caller isn't still at the ending character. */
4567 /* Report an error if the range is empty and the syntax prohibits this. */
4568 ret
= syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
4571 collseq
= (const unsigned char *) _NL_CURRENT (LC_COLLATE
,
4572 _NL_COLLATE_COLLSEQMB
);
4574 start_colseq
= collseq
[(unsigned char) TRANSLATE (range_start_char
)];
4575 end_colseq
= collseq
[(unsigned char) TRANSLATE (p
[0])];
4576 for (this_char
= 0; this_char
<= (unsigned char) -1; ++this_char
)
4578 unsigned int this_colseq
= collseq
[(unsigned char) TRANSLATE (this_char
)];
4580 if (start_colseq
<= this_colseq
&& this_colseq
<= end_colseq
)
4582 SET_LIST_BIT (TRANSLATE (this_char
));
4587 /* Here we see why `this_char' has to be larger than an `unsigned
4588 char' -- we would otherwise go into an infinite loop, since all
4589 characters <= 0xff. */
4590 range_start_char
= TRANSLATE (range_start_char
);
4591 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4592 and some compilers cast it to int implicitly, so following for_loop
4593 may fall to (almost) infinite loop.
4594 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4595 To avoid this, we cast p[0] to unsigned int and truncate it. */
4596 end_char
= ((unsigned)TRANSLATE(p
[0]) & ((1 << BYTEWIDTH
) - 1));
4598 for (this_char
= range_start_char
; this_char
<= end_char
; ++this_char
)
4600 SET_LIST_BIT (TRANSLATE (this_char
));
4609 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4610 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4611 characters can start a string that matches the pattern. This fastmap
4612 is used by re_search to skip quickly over impossible starting points.
4614 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4615 area as BUFP->fastmap.
4617 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4620 Returns 0 if we succeed, -2 if an internal error. */
4623 /* local function for re_compile_fastmap.
4624 truncate wchar_t character to char. */
4625 static unsigned char truncate_wchar (CHAR_T c
);
4627 static unsigned char
4631 unsigned char buf
[MB_CUR_MAX
];
4634 memset (&state
, '\0', sizeof (state
));
4636 retval
= __wcrtomb (buf
, c
, &state
);
4638 retval
= wcrtomb (buf
, c
, &state
);
4640 return retval
> 0 ? buf
[0] : (unsigned char) c
;
4645 PREFIX(re_compile_fastmap
) (bufp
)
4646 struct re_pattern_buffer
*bufp
;
4649 #ifdef MATCH_MAY_ALLOCATE
4650 PREFIX(fail_stack_type
) fail_stack
;
4652 #ifndef REGEX_MALLOC
4656 register char *fastmap
= bufp
->fastmap
;
4659 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4660 pattern to (char*) in regex_compile. */
4661 UCHAR_T
*pattern
= (UCHAR_T
*)bufp
->buffer
;
4662 register UCHAR_T
*pend
= (UCHAR_T
*) (bufp
->buffer
+ bufp
->used
);
4664 UCHAR_T
*pattern
= bufp
->buffer
;
4665 register UCHAR_T
*pend
= pattern
+ bufp
->used
;
4667 UCHAR_T
*p
= pattern
;
4670 /* This holds the pointer to the failure stack, when
4671 it is allocated relocatably. */
4672 fail_stack_elt_t
*failure_stack_ptr
;
4675 /* Assume that each path through the pattern can be null until
4676 proven otherwise. We set this false at the bottom of switch
4677 statement, to which we get only if a particular path doesn't
4678 match the empty string. */
4679 boolean path_can_be_null
= true;
4681 /* We aren't doing a `succeed_n' to begin with. */
4682 boolean succeed_n_p
= false;
4684 assert (fastmap
!= NULL
&& p
!= NULL
);
4687 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4688 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4689 bufp
->can_be_null
= 0;
4693 if (p
== pend
|| *p
== succeed
)
4695 /* We have reached the (effective) end of pattern. */
4696 if (!FAIL_STACK_EMPTY ())
4698 bufp
->can_be_null
|= path_can_be_null
;
4700 /* Reset for next path. */
4701 path_can_be_null
= true;
4703 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
4711 /* We should never be about to go beyond the end of the pattern. */
4714 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4717 /* I guess the idea here is to simply not bother with a fastmap
4718 if a backreference is used, since it's too hard to figure out
4719 the fastmap for the corresponding group. Setting
4720 `can_be_null' stops `re_search_2' from using the fastmap, so
4721 that is all we do. */
4723 bufp
->can_be_null
= 1;
4727 /* Following are the cases which match a character. These end
4732 fastmap
[truncate_wchar(p
[1])] = 1;
4746 /* It is hard to distinguish fastmap from (multi byte) characters
4747 which depends on current locale. */
4752 bufp
->can_be_null
= 1;
4756 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
4757 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
4763 /* Chars beyond end of map must be allowed. */
4764 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
4767 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
4768 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
4774 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4775 if (SYNTAX (j
) == Sword
)
4781 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4782 if (SYNTAX (j
) != Sword
)
4789 int fastmap_newline
= fastmap
['\n'];
4791 /* `.' matches anything ... */
4792 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4795 /* ... except perhaps newline. */
4796 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
4797 fastmap
['\n'] = fastmap_newline
;
4799 /* Return if we have already set `can_be_null'; if we have,
4800 then the fastmap is irrelevant. Something's wrong here. */
4801 else if (bufp
->can_be_null
)
4804 /* Otherwise, have to check alternative paths. */
4811 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4812 if (SYNTAX (j
) == (enum syntaxcode
) k
)
4819 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4820 if (SYNTAX (j
) != (enum syntaxcode
) k
)
4825 /* All cases after this match the empty string. These end with
4845 case push_dummy_failure
:
4850 case pop_failure_jump
:
4851 case maybe_pop_jump
:
4854 case dummy_failure_jump
:
4855 EXTRACT_NUMBER_AND_INCR (j
, p
);
4860 /* Jump backward implies we just went through the body of a
4861 loop and matched nothing. Opcode jumped to should be
4862 `on_failure_jump' or `succeed_n'. Just treat it like an
4863 ordinary jump. For a * loop, it has pushed its failure
4864 point already; if so, discard that as redundant. */
4865 if ((re_opcode_t
) *p
!= on_failure_jump
4866 && (re_opcode_t
) *p
!= succeed_n
)
4870 EXTRACT_NUMBER_AND_INCR (j
, p
);
4873 /* If what's on the stack is where we are now, pop it. */
4874 if (!FAIL_STACK_EMPTY ()
4875 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
4881 case on_failure_jump
:
4882 case on_failure_keep_string_jump
:
4883 handle_on_failure_jump
:
4884 EXTRACT_NUMBER_AND_INCR (j
, p
);
4886 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4887 end of the pattern. We don't want to push such a point,
4888 since when we restore it above, entering the switch will
4889 increment `p' past the end of the pattern. We don't need
4890 to push such a point since we obviously won't find any more
4891 fastmap entries beyond `pend'. Such a pattern can match
4892 the null string, though. */
4895 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
4897 RESET_FAIL_STACK ();
4902 bufp
->can_be_null
= 1;
4906 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
4907 succeed_n_p
= false;
4914 /* Get to the number of times to succeed. */
4915 p
+= OFFSET_ADDRESS_SIZE
;
4917 /* Increment p past the n for when k != 0. */
4918 EXTRACT_NUMBER_AND_INCR (k
, p
);
4921 p
-= 2 * OFFSET_ADDRESS_SIZE
;
4922 succeed_n_p
= true; /* Spaghetti code alert. */
4923 goto handle_on_failure_jump
;
4929 p
+= 2 * OFFSET_ADDRESS_SIZE
;
4940 abort (); /* We have listed all the cases. */
4943 /* Getting here means we have found the possible starting
4944 characters for one path of the pattern -- and that the empty
4945 string does not match. We need not follow this path further.
4946 Instead, look at the next alternative (remembered on the
4947 stack), or quit if no more. The test at the top of the loop
4948 does these things. */
4949 path_can_be_null
= false;
4953 /* Set `can_be_null' for the last path (also the first path, if the
4954 pattern is empty). */
4955 bufp
->can_be_null
|= path_can_be_null
;
4958 RESET_FAIL_STACK ();
4962 #else /* not INSIDE_RECURSION */
4965 re_compile_fastmap (bufp
)
4966 struct re_pattern_buffer
*bufp
;
4969 if (MB_CUR_MAX
!= 1)
4970 return wcs_re_compile_fastmap(bufp
);
4973 return byte_re_compile_fastmap(bufp
);
4974 } /* re_compile_fastmap */
4976 weak_alias (__re_compile_fastmap
, re_compile_fastmap
)
4980 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4981 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4982 this memory for recording register information. STARTS and ENDS
4983 must be allocated using the malloc library routine, and must each
4984 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4986 If NUM_REGS == 0, then subsequent matches should allocate their own
4989 Unless this function is called, the first search or match using
4990 PATTERN_BUFFER will allocate its own register data, without
4991 freeing the old data. */
4994 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4995 struct re_pattern_buffer
*bufp
;
4996 struct re_registers
*regs
;
4998 regoff_t
*starts
, *ends
;
5002 bufp
->regs_allocated
= REGS_REALLOCATE
;
5003 regs
->num_regs
= num_regs
;
5004 regs
->start
= starts
;
5009 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5011 regs
->start
= regs
->end
= (regoff_t
*) 0;
5015 weak_alias (__re_set_registers
, re_set_registers
)
5018 /* Searching routines. */
5020 /* Like re_search_2, below, but only one string is specified, and
5021 doesn't let you say where to stop matching. */
5024 re_search (bufp
, string
, size
, startpos
, range
, regs
)
5025 struct re_pattern_buffer
*bufp
;
5027 int size
, startpos
, range
;
5028 struct re_registers
*regs
;
5030 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
5034 weak_alias (__re_search
, re_search
)
5038 /* Using the compiled pattern in BUFP->buffer, first tries to match the
5039 virtual concatenation of STRING1 and STRING2, starting first at index
5040 STARTPOS, then at STARTPOS + 1, and so on.
5042 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
5044 RANGE is how far to scan while trying to match. RANGE = 0 means try
5045 only at STARTPOS; in general, the last start tried is STARTPOS +
5048 In REGS, return the indices of the virtual concatenation of STRING1
5049 and STRING2 that matched the entire BUFP->buffer and its contained
5052 Do not consider matching one past the index STOP in the virtual
5053 concatenation of STRING1 and STRING2.
5055 We return either the position in the strings at which the match was
5056 found, -1 if no match, or -2 if error (such as failure
5060 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
5061 struct re_pattern_buffer
*bufp
;
5062 const char *string1
, *string2
;
5066 struct re_registers
*regs
;
5070 if (MB_CUR_MAX
!= 1)
5071 return wcs_re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
,
5075 return byte_re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
,
5079 weak_alias (__re_search_2
, re_search_2
)
5082 #endif /* not INSIDE_RECURSION */
5084 #ifdef INSIDE_RECURSION
5086 #ifdef MATCH_MAY_ALLOCATE
5087 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5089 # define FREE_VAR(var) if (var) free (var); var = NULL
5093 # define MAX_ALLOCA_SIZE 2000
5095 # define FREE_WCS_BUFFERS() \
5097 if (size1 > MAX_ALLOCA_SIZE) \
5099 free (wcs_string1); \
5100 free (mbs_offset1); \
5104 FREE_VAR (wcs_string1); \
5105 FREE_VAR (mbs_offset1); \
5107 if (size2 > MAX_ALLOCA_SIZE) \
5109 free (wcs_string2); \
5110 free (mbs_offset2); \
5114 FREE_VAR (wcs_string2); \
5115 FREE_VAR (mbs_offset2); \
5123 PREFIX(re_search_2
) (bufp
, string1
, size1
, string2
, size2
, startpos
, range
,
5125 struct re_pattern_buffer
*bufp
;
5126 const char *string1
, *string2
;
5130 struct re_registers
*regs
;
5134 register char *fastmap
= bufp
->fastmap
;
5135 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5136 int total_size
= size1
+ size2
;
5137 int endpos
= startpos
+ range
;
5139 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5140 wchar_t *wcs_string1
= NULL
, *wcs_string2
= NULL
;
5141 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5142 int wcs_size1
= 0, wcs_size2
= 0;
5143 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5144 int *mbs_offset1
= NULL
, *mbs_offset2
= NULL
;
5145 /* They hold whether each wchar_t is binary data or not. */
5146 char *is_binary
= NULL
;
5149 /* Check for out-of-range STARTPOS. */
5150 if (startpos
< 0 || startpos
> total_size
)
5153 /* Fix up RANGE if it might eventually take us outside
5154 the virtual concatenation of STRING1 and STRING2.
5155 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5157 range
= 0 - startpos
;
5158 else if (endpos
> total_size
)
5159 range
= total_size
- startpos
;
5161 /* If the search isn't to be a backwards one, don't waste time in a
5162 search for a pattern that must be anchored. */
5163 if (bufp
->used
> 0 && range
> 0
5164 && ((re_opcode_t
) bufp
->buffer
[0] == begbuf
5165 /* `begline' is like `begbuf' if it cannot match at newlines. */
5166 || ((re_opcode_t
) bufp
->buffer
[0] == begline
5167 && !bufp
->newline_anchor
)))
5176 /* In a forward search for something that starts with \=.
5177 don't keep searching past point. */
5178 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
5180 range
= PT
- startpos
;
5186 /* Update the fastmap now if not correct already. */
5187 if (fastmap
&& !bufp
->fastmap_accurate
)
5188 if (re_compile_fastmap (bufp
) == -2)
5192 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5193 fill them with converted string. */
5196 if (size1
> MAX_ALLOCA_SIZE
)
5198 wcs_string1
= TALLOC (size1
+ 1, CHAR_T
);
5199 mbs_offset1
= TALLOC (size1
+ 1, int);
5200 is_binary
= TALLOC (size1
+ 1, char);
5204 wcs_string1
= REGEX_TALLOC (size1
+ 1, CHAR_T
);
5205 mbs_offset1
= REGEX_TALLOC (size1
+ 1, int);
5206 is_binary
= REGEX_TALLOC (size1
+ 1, char);
5208 if (!wcs_string1
|| !mbs_offset1
|| !is_binary
)
5210 if (size1
> MAX_ALLOCA_SIZE
)
5218 FREE_VAR (wcs_string1
);
5219 FREE_VAR (mbs_offset1
);
5220 FREE_VAR (is_binary
);
5224 wcs_size1
= convert_mbs_to_wcs(wcs_string1
, string1
, size1
,
5225 mbs_offset1
, is_binary
);
5226 wcs_string1
[wcs_size1
] = L
'\0'; /* for a sentinel */
5227 if (size1
> MAX_ALLOCA_SIZE
)
5230 FREE_VAR (is_binary
);
5234 if (size2
> MAX_ALLOCA_SIZE
)
5236 wcs_string2
= TALLOC (size2
+ 1, CHAR_T
);
5237 mbs_offset2
= TALLOC (size2
+ 1, int);
5238 is_binary
= TALLOC (size2
+ 1, char);
5242 wcs_string2
= REGEX_TALLOC (size2
+ 1, CHAR_T
);
5243 mbs_offset2
= REGEX_TALLOC (size2
+ 1, int);
5244 is_binary
= REGEX_TALLOC (size2
+ 1, char);
5246 if (!wcs_string2
|| !mbs_offset2
|| !is_binary
)
5248 FREE_WCS_BUFFERS ();
5249 if (size2
> MAX_ALLOCA_SIZE
)
5252 FREE_VAR (is_binary
);
5255 wcs_size2
= convert_mbs_to_wcs(wcs_string2
, string2
, size2
,
5256 mbs_offset2
, is_binary
);
5257 wcs_string2
[wcs_size2
] = L
'\0'; /* for a sentinel */
5258 if (size2
> MAX_ALLOCA_SIZE
)
5261 FREE_VAR (is_binary
);
5266 /* Loop through the string, looking for a place to start matching. */
5269 /* If a fastmap is supplied, skip quickly over characters that
5270 cannot be the start of a match. If the pattern can match the
5271 null string, however, we don't need to skip characters; we want
5272 the first null string. */
5273 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
5275 if (range
> 0) /* Searching forwards. */
5277 register const char *d
;
5278 register int lim
= 0;
5281 if (startpos
< size1
&& startpos
+ range
>= size1
)
5282 lim
= range
- (size1
- startpos
);
5284 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
5286 /* Written out as an if-else to avoid testing `translate'
5290 && !fastmap
[(unsigned char)
5291 translate
[(unsigned char) *d
++]])
5294 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
5297 startpos
+= irange
- range
;
5299 else /* Searching backwards. */
5301 register CHAR_T c
= (size1
== 0 || startpos
>= size1
5302 ? string2
[startpos
- size1
]
5303 : string1
[startpos
]);
5305 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
5310 /* If can't match the null string, and that's all we have left, fail. */
5311 if (range
>= 0 && startpos
== total_size
&& fastmap
5312 && !bufp
->can_be_null
)
5315 FREE_WCS_BUFFERS ();
5321 val
= wcs_re_match_2_internal (bufp
, string1
, size1
, string2
,
5322 size2
, startpos
, regs
, stop
,
5323 wcs_string1
, wcs_size1
,
5324 wcs_string2
, wcs_size2
,
5325 mbs_offset1
, mbs_offset2
);
5327 val
= byte_re_match_2_internal (bufp
, string1
, size1
, string2
,
5328 size2
, startpos
, regs
, stop
);
5331 #ifndef REGEX_MALLOC
5340 FREE_WCS_BUFFERS ();
5348 FREE_WCS_BUFFERS ();
5368 FREE_WCS_BUFFERS ();
5374 /* This converts PTR, a pointer into one of the search wchar_t strings
5375 `string1' and `string2' into an multibyte string offset from the
5376 beginning of that string. We use mbs_offset to optimize.
5377 See convert_mbs_to_wcs. */
5378 # define POINTER_TO_OFFSET(ptr) \
5379 (FIRST_STRING_P (ptr) \
5380 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5381 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5384 /* This converts PTR, a pointer into one of the search strings `string1'
5385 and `string2' into an offset from the beginning of that string. */
5386 # define POINTER_TO_OFFSET(ptr) \
5387 (FIRST_STRING_P (ptr) \
5388 ? ((regoff_t) ((ptr) - string1)) \
5389 : ((regoff_t) ((ptr) - string2 + size1)))
5392 /* Macros for dealing with the split strings in re_match_2. */
5394 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5396 /* Call before fetching a character with *d. This switches over to
5397 string2 if necessary. */
5398 #define PREFETCH() \
5401 /* End of string2 => fail. */ \
5402 if (dend == end_match_2) \
5404 /* End of string1 => advance to string2. */ \
5406 dend = end_match_2; \
5409 /* Test if at very beginning or at very end of the virtual concatenation
5410 of `string1' and `string2'. If only one string, it's `string2'. */
5411 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5412 #define AT_STRINGS_END(d) ((d) == end2)
5415 /* Test if D points to a character which is word-constituent. We have
5416 two special cases to check for: if past the end of string1, look at
5417 the first character in string2; and if before the beginning of
5418 string2, look at the last character in string1. */
5420 /* Use internationalized API instead of SYNTAX. */
5421 # define WORDCHAR_P(d) \
5422 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5423 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5424 || ((d) == end1 ? *string2 \
5425 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5427 # define WORDCHAR_P(d) \
5428 (SYNTAX ((d) == end1 ? *string2 \
5429 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5433 /* Disabled due to a compiler bug -- see comment at case wordbound */
5435 /* Test if the character before D and the one at D differ with respect
5436 to being word-constituent. */
5437 #define AT_WORD_BOUNDARY(d) \
5438 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5439 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5442 /* Free everything we malloc. */
5443 #ifdef MATCH_MAY_ALLOCATE
5445 # define FREE_VARIABLES() \
5447 REGEX_FREE_STACK (fail_stack.stack); \
5448 FREE_VAR (regstart); \
5449 FREE_VAR (regend); \
5450 FREE_VAR (old_regstart); \
5451 FREE_VAR (old_regend); \
5452 FREE_VAR (best_regstart); \
5453 FREE_VAR (best_regend); \
5454 FREE_VAR (reg_info); \
5455 FREE_VAR (reg_dummy); \
5456 FREE_VAR (reg_info_dummy); \
5457 if (!cant_free_wcs_buf) \
5459 FREE_VAR (string1); \
5460 FREE_VAR (string2); \
5461 FREE_VAR (mbs_offset1); \
5462 FREE_VAR (mbs_offset2); \
5466 # define FREE_VARIABLES() \
5468 REGEX_FREE_STACK (fail_stack.stack); \
5469 FREE_VAR (regstart); \
5470 FREE_VAR (regend); \
5471 FREE_VAR (old_regstart); \
5472 FREE_VAR (old_regend); \
5473 FREE_VAR (best_regstart); \
5474 FREE_VAR (best_regend); \
5475 FREE_VAR (reg_info); \
5476 FREE_VAR (reg_dummy); \
5477 FREE_VAR (reg_info_dummy); \
5482 # define FREE_VARIABLES() \
5484 if (!cant_free_wcs_buf) \
5486 FREE_VAR (string1); \
5487 FREE_VAR (string2); \
5488 FREE_VAR (mbs_offset1); \
5489 FREE_VAR (mbs_offset2); \
5493 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5495 #endif /* not MATCH_MAY_ALLOCATE */
5497 /* These values must meet several constraints. They must not be valid
5498 register values; since we have a limit of 255 registers (because
5499 we use only one byte in the pattern for the register number), we can
5500 use numbers larger than 255. They must differ by 1, because of
5501 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5502 be larger than the value for the highest register, so we do not try
5503 to actually save any registers when none are active. */
5504 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5505 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5507 #else /* not INSIDE_RECURSION */
5508 /* Matching routines. */
5510 #ifndef emacs /* Emacs never uses this. */
5511 /* re_match is like re_match_2 except it takes only a single string. */
5514 re_match (bufp
, string
, size
, pos
, regs
)
5515 struct re_pattern_buffer
*bufp
;
5518 struct re_registers
*regs
;
5522 if (MB_CUR_MAX
!= 1)
5523 result
= wcs_re_match_2_internal (bufp
, NULL
, 0, string
, size
,
5525 NULL
, 0, NULL
, 0, NULL
, NULL
);
5528 result
= byte_re_match_2_internal (bufp
, NULL
, 0, string
, size
,
5530 # ifndef REGEX_MALLOC
5538 weak_alias (__re_match
, re_match
)
5540 #endif /* not emacs */
5542 #endif /* not INSIDE_RECURSION */
5544 #ifdef INSIDE_RECURSION
5545 static boolean
PREFIX(group_match_null_string_p
) _RE_ARGS ((UCHAR_T
**p
,
5547 PREFIX(register_info_type
) *reg_info
));
5548 static boolean
PREFIX(alt_match_null_string_p
) _RE_ARGS ((UCHAR_T
*p
,
5550 PREFIX(register_info_type
) *reg_info
));
5551 static boolean
PREFIX(common_op_match_null_string_p
) _RE_ARGS ((UCHAR_T
**p
,
5553 PREFIX(register_info_type
) *reg_info
));
5554 static int PREFIX(bcmp_translate
) _RE_ARGS ((const CHAR_T
*s1
, const CHAR_T
*s2
,
5555 int len
, char *translate
));
5556 #else /* not INSIDE_RECURSION */
5558 /* re_match_2 matches the compiled pattern in BUFP against the
5559 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5560 and SIZE2, respectively). We start matching at POS, and stop
5563 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5564 store offsets for the substring each group matched in REGS. See the
5565 documentation for exactly how many groups we fill.
5567 We return -1 if no match, -2 if an internal error (such as the
5568 failure stack overflowing). Otherwise, we return the length of the
5569 matched substring. */
5572 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
5573 struct re_pattern_buffer
*bufp
;
5574 const char *string1
, *string2
;
5577 struct re_registers
*regs
;
5582 if (MB_CUR_MAX
!= 1)
5583 result
= wcs_re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
5585 NULL
, 0, NULL
, 0, NULL
, NULL
);
5588 result
= byte_re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
5591 #ifndef REGEX_MALLOC
5599 weak_alias (__re_match_2
, re_match_2
)
5602 #endif /* not INSIDE_RECURSION */
5604 #ifdef INSIDE_RECURSION
5607 static int count_mbs_length
PARAMS ((int *, int));
5609 /* This check the substring (from 0, to length) of the multibyte string,
5610 to which offset_buffer correspond. And count how many wchar_t_characters
5611 the substring occupy. We use offset_buffer to optimization.
5612 See convert_mbs_to_wcs. */
5615 count_mbs_length(offset_buffer
, length
)
5621 /* Check whether the size is valid. */
5625 if (offset_buffer
== NULL
)
5628 /* If there are no multibyte character, offset_buffer[i] == i.
5629 Optmize for this case. */
5630 if (offset_buffer
[length
] == length
)
5633 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5639 int middle
= (lower
+ upper
) / 2;
5640 if (middle
== lower
|| middle
== upper
)
5642 if (offset_buffer
[middle
] > length
)
5644 else if (offset_buffer
[middle
] < length
)
5654 /* This is a separate function so that we can force an alloca cleanup
5658 wcs_re_match_2_internal (bufp
, cstring1
, csize1
, cstring2
, csize2
, pos
,
5659 regs
, stop
, string1
, size1
, string2
, size2
,
5660 mbs_offset1
, mbs_offset2
)
5661 struct re_pattern_buffer
*bufp
;
5662 const char *cstring1
, *cstring2
;
5665 struct re_registers
*regs
;
5667 /* string1 == string2 == NULL means string1/2, size1/2 and
5668 mbs_offset1/2 need seting up in this function. */
5669 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5670 wchar_t *string1
, *string2
;
5671 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5673 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5674 int *mbs_offset1
, *mbs_offset2
;
5677 byte_re_match_2_internal (bufp
, string1
, size1
,string2
, size2
, pos
,
5679 struct re_pattern_buffer
*bufp
;
5680 const char *string1
, *string2
;
5683 struct re_registers
*regs
;
5687 /* General temporaries. */
5691 /* They hold whether each wchar_t is binary data or not. */
5692 char *is_binary
= NULL
;
5693 /* If true, we can't free string1/2, mbs_offset1/2. */
5694 int cant_free_wcs_buf
= 1;
5697 /* Just past the end of the corresponding string. */
5698 const CHAR_T
*end1
, *end2
;
5700 /* Pointers into string1 and string2, just past the last characters in
5701 each to consider matching. */
5702 const CHAR_T
*end_match_1
, *end_match_2
;
5704 /* Where we are in the data, and the end of the current string. */
5705 const CHAR_T
*d
, *dend
;
5707 /* Where we are in the pattern, and the end of the pattern. */
5709 UCHAR_T
*pattern
, *p
;
5710 register UCHAR_T
*pend
;
5712 UCHAR_T
*p
= bufp
->buffer
;
5713 register UCHAR_T
*pend
= p
+ bufp
->used
;
5716 /* Mark the opcode just after a start_memory, so we can test for an
5717 empty subpattern when we get to the stop_memory. */
5718 UCHAR_T
*just_past_start_mem
= 0;
5720 /* We use this to map every character in the string. */
5721 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5723 /* Failure point stack. Each place that can handle a failure further
5724 down the line pushes a failure point on this stack. It consists of
5725 restart, regend, and reg_info for all registers corresponding to
5726 the subexpressions we're currently inside, plus the number of such
5727 registers, and, finally, two char *'s. The first char * is where
5728 to resume scanning the pattern; the second one is where to resume
5729 scanning the strings. If the latter is zero, the failure point is
5730 a ``dummy''; if a failure happens and the failure point is a dummy,
5731 it gets discarded and the next next one is tried. */
5732 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5733 PREFIX(fail_stack_type
) fail_stack
;
5736 static unsigned failure_id
;
5737 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5741 /* This holds the pointer to the failure stack, when
5742 it is allocated relocatably. */
5743 fail_stack_elt_t
*failure_stack_ptr
;
5746 /* We fill all the registers internally, independent of what we
5747 return, for use in backreferences. The number here includes
5748 an element for register zero. */
5749 size_t num_regs
= bufp
->re_nsub
+ 1;
5751 /* The currently active registers. */
5752 active_reg_t lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
5753 active_reg_t highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
5755 /* Information on the contents of registers. These are pointers into
5756 the input strings; they record just what was matched (on this
5757 attempt) by a subexpression part of the pattern, that is, the
5758 regnum-th regstart pointer points to where in the pattern we began
5759 matching and the regnum-th regend points to right after where we
5760 stopped matching the regnum-th subexpression. (The zeroth register
5761 keeps track of what the whole pattern matches.) */
5762 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5763 const CHAR_T
**regstart
, **regend
;
5766 /* If a group that's operated upon by a repetition operator fails to
5767 match anything, then the register for its start will need to be
5768 restored because it will have been set to wherever in the string we
5769 are when we last see its open-group operator. Similarly for a
5771 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5772 const CHAR_T
**old_regstart
, **old_regend
;
5775 /* The is_active field of reg_info helps us keep track of which (possibly
5776 nested) subexpressions we are currently in. The matched_something
5777 field of reg_info[reg_num] helps us tell whether or not we have
5778 matched any of the pattern so far this time through the reg_num-th
5779 subexpression. These two fields get reset each time through any
5780 loop their register is in. */
5781 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5782 PREFIX(register_info_type
) *reg_info
;
5785 /* The following record the register info as found in the above
5786 variables when we find a match better than any we've seen before.
5787 This happens as we backtrack through the failure points, which in
5788 turn happens only if we have not yet matched the entire string. */
5789 unsigned best_regs_set
= false;
5790 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5791 const CHAR_T
**best_regstart
, **best_regend
;
5794 /* Logically, this is `best_regend[0]'. But we don't want to have to
5795 allocate space for that if we're not allocating space for anything
5796 else (see below). Also, we never need info about register 0 for
5797 any of the other register vectors, and it seems rather a kludge to
5798 treat `best_regend' differently than the rest. So we keep track of
5799 the end of the best match so far in a separate variable. We
5800 initialize this to NULL so that when we backtrack the first time
5801 and need to test it, it's not garbage. */
5802 const CHAR_T
*match_end
= NULL
;
5804 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5805 int set_regs_matched_done
= 0;
5807 /* Used when we pop values we don't care about. */
5808 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5809 const CHAR_T
**reg_dummy
;
5810 PREFIX(register_info_type
) *reg_info_dummy
;
5814 /* Counts the total number of registers pushed. */
5815 unsigned num_regs_pushed
= 0;
5818 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5822 #ifdef MATCH_MAY_ALLOCATE
5823 /* Do not bother to initialize all the register variables if there are
5824 no groups in the pattern, as it takes a fair amount of time. If
5825 there are groups, we include space for register 0 (the whole
5826 pattern), even though we never use it, since it simplifies the
5827 array indexing. We should fix this. */
5830 regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5831 regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5832 old_regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5833 old_regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5834 best_regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5835 best_regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5836 reg_info
= REGEX_TALLOC (num_regs
, PREFIX(register_info_type
));
5837 reg_dummy
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5838 reg_info_dummy
= REGEX_TALLOC (num_regs
, PREFIX(register_info_type
));
5840 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
5841 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
5849 /* We must initialize all our variables to NULL, so that
5850 `FREE_VARIABLES' doesn't try to free them. */
5851 regstart
= regend
= old_regstart
= old_regend
= best_regstart
5852 = best_regend
= reg_dummy
= NULL
;
5853 reg_info
= reg_info_dummy
= (PREFIX(register_info_type
) *) NULL
;
5855 #endif /* MATCH_MAY_ALLOCATE */
5857 /* The starting position is bogus. */
5859 if (pos
< 0 || pos
> csize1
+ csize2
)
5861 if (pos
< 0 || pos
> size1
+ size2
)
5869 /* Allocate wchar_t array for string1 and string2 and
5870 fill them with converted string. */
5871 if (string1
== NULL
&& string2
== NULL
)
5873 /* We need seting up buffers here. */
5875 /* We must free wcs buffers in this function. */
5876 cant_free_wcs_buf
= 0;
5880 string1
= REGEX_TALLOC (csize1
+ 1, CHAR_T
);
5881 mbs_offset1
= REGEX_TALLOC (csize1
+ 1, int);
5882 is_binary
= REGEX_TALLOC (csize1
+ 1, char);
5883 if (!string1
|| !mbs_offset1
|| !is_binary
)
5886 FREE_VAR (mbs_offset1
);
5887 FREE_VAR (is_binary
);
5893 string2
= REGEX_TALLOC (csize2
+ 1, CHAR_T
);
5894 mbs_offset2
= REGEX_TALLOC (csize2
+ 1, int);
5895 is_binary
= REGEX_TALLOC (csize2
+ 1, char);
5896 if (!string2
|| !mbs_offset2
|| !is_binary
)
5899 FREE_VAR (mbs_offset1
);
5901 FREE_VAR (mbs_offset2
);
5902 FREE_VAR (is_binary
);
5905 size2
= convert_mbs_to_wcs(string2
, cstring2
, csize2
,
5906 mbs_offset2
, is_binary
);
5907 string2
[size2
] = L
'\0'; /* for a sentinel */
5908 FREE_VAR (is_binary
);
5912 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5913 pattern to (char*) in regex_compile. */
5914 p
= pattern
= (CHAR_T
*)bufp
->buffer
;
5915 pend
= (CHAR_T
*)(bufp
->buffer
+ bufp
->used
);
5919 /* Initialize subexpression text positions to -1 to mark ones that no
5920 start_memory/stop_memory has been seen for. Also initialize the
5921 register information struct. */
5922 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
5924 regstart
[mcnt
] = regend
[mcnt
]
5925 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
5927 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
5928 IS_ACTIVE (reg_info
[mcnt
]) = 0;
5929 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
5930 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
5933 /* We move `string1' into `string2' if the latter's empty -- but not if
5934 `string1' is null. */
5935 if (size2
== 0 && string1
!= NULL
)
5942 mbs_offset2
= mbs_offset1
;
5948 end1
= string1
+ size1
;
5949 end2
= string2
+ size2
;
5951 /* Compute where to stop matching, within the two strings. */
5955 mcnt
= count_mbs_length(mbs_offset1
, stop
);
5956 end_match_1
= string1
+ mcnt
;
5957 end_match_2
= string2
;
5961 if (stop
> csize1
+ csize2
)
5962 stop
= csize1
+ csize2
;
5964 mcnt
= count_mbs_length(mbs_offset2
, stop
-csize1
);
5965 end_match_2
= string2
+ mcnt
;
5968 { /* count_mbs_length return error. */
5975 end_match_1
= string1
+ stop
;
5976 end_match_2
= string2
;
5981 end_match_2
= string2
+ stop
- size1
;
5985 /* `p' scans through the pattern as `d' scans through the data.
5986 `dend' is the end of the input string that `d' points within. `d'
5987 is advanced into the following input string whenever necessary, but
5988 this happens before fetching; therefore, at the beginning of the
5989 loop, `d' can be pointing at the end of a string, but it cannot
5992 if (size1
> 0 && pos
<= csize1
)
5994 mcnt
= count_mbs_length(mbs_offset1
, pos
);
6000 mcnt
= count_mbs_length(mbs_offset2
, pos
-csize1
);
6006 { /* count_mbs_length return error. */
6011 if (size1
> 0 && pos
<= size1
)
6018 d
= string2
+ pos
- size1
;
6023 DEBUG_PRINT1 ("The compiled pattern is:\n");
6024 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
6025 DEBUG_PRINT1 ("The string to match is: `");
6026 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
6027 DEBUG_PRINT1 ("'\n");
6029 /* This loops over pattern commands. It exits by returning from the
6030 function if the match is complete, or it drops through if the match
6031 fails at this starting point in the input data. */
6035 DEBUG_PRINT2 ("\n%p: ", p
);
6037 DEBUG_PRINT2 ("\n0x%x: ", p
);
6041 { /* End of pattern means we might have succeeded. */
6042 DEBUG_PRINT1 ("end of pattern ... ");
6044 /* If we haven't matched the entire string, and we want the
6045 longest match, try backtracking. */
6046 if (d
!= end_match_2
)
6048 /* 1 if this match ends in the same string (string1 or string2)
6049 as the best previous match. */
6050 boolean same_str_p
= (FIRST_STRING_P (match_end
)
6051 == MATCHING_IN_FIRST_STRING
);
6052 /* 1 if this match is the best seen so far. */
6053 boolean best_match_p
;
6055 /* AIX compiler got confused when this was combined
6056 with the previous declaration. */
6058 best_match_p
= d
> match_end
;
6060 best_match_p
= !MATCHING_IN_FIRST_STRING
;
6062 DEBUG_PRINT1 ("backtracking.\n");
6064 if (!FAIL_STACK_EMPTY ())
6065 { /* More failure points to try. */
6067 /* If exceeds best match so far, save it. */
6068 if (!best_regs_set
|| best_match_p
)
6070 best_regs_set
= true;
6073 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6075 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
6077 best_regstart
[mcnt
] = regstart
[mcnt
];
6078 best_regend
[mcnt
] = regend
[mcnt
];
6084 /* If no failure points, don't restore garbage. And if
6085 last match is real best match, don't restore second
6087 else if (best_regs_set
&& !best_match_p
)
6090 /* Restore best match. It may happen that `dend ==
6091 end_match_1' while the restored d is in string2.
6092 For example, the pattern `x.*y.*z' against the
6093 strings `x-' and `y-z-', if the two strings are
6094 not consecutive in memory. */
6095 DEBUG_PRINT1 ("Restoring best registers.\n");
6098 dend
= ((d
>= string1
&& d
<= end1
)
6099 ? end_match_1
: end_match_2
);
6101 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
6103 regstart
[mcnt
] = best_regstart
[mcnt
];
6104 regend
[mcnt
] = best_regend
[mcnt
];
6107 } /* d != end_match_2 */
6110 DEBUG_PRINT1 ("Accepting match.\n");
6111 /* If caller wants register contents data back, do it. */
6112 if (regs
&& !bufp
->no_sub
)
6114 /* Have the register data arrays been allocated? */
6115 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
6116 { /* No. So allocate them with malloc. We need one
6117 extra element beyond `num_regs' for the `-1' marker
6119 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
6120 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
6121 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
6122 if (regs
->start
== NULL
|| regs
->end
== NULL
)
6127 bufp
->regs_allocated
= REGS_REALLOCATE
;
6129 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
6130 { /* Yes. If we need more elements than were already
6131 allocated, reallocate them. If we need fewer, just
6133 if (regs
->num_regs
< num_regs
+ 1)
6135 regs
->num_regs
= num_regs
+ 1;
6136 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
6137 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
6138 if (regs
->start
== NULL
|| regs
->end
== NULL
)
6147 /* These braces fend off a "empty body in an else-statement"
6148 warning under GCC when assert expands to nothing. */
6149 assert (bufp
->regs_allocated
== REGS_FIXED
);
6152 /* Convert the pointer data in `regstart' and `regend' to
6153 indices. Register zero has to be set differently,
6154 since we haven't kept track of any info for it. */
6155 if (regs
->num_regs
> 0)
6157 regs
->start
[0] = pos
;
6159 if (MATCHING_IN_FIRST_STRING
)
6160 regs
->end
[0] = mbs_offset1
!= NULL
?
6161 mbs_offset1
[d
-string1
] : 0;
6163 regs
->end
[0] = csize1
+ (mbs_offset2
!= NULL
?
6164 mbs_offset2
[d
-string2
] : 0);
6166 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
6167 ? ((regoff_t
) (d
- string1
))
6168 : ((regoff_t
) (d
- string2
+ size1
)));
6172 /* Go through the first `min (num_regs, regs->num_regs)'
6173 registers, since that is all we initialized. */
6174 for (mcnt
= 1; (unsigned) mcnt
< MIN (num_regs
, regs
->num_regs
);
6177 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
6178 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
6182 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
6184 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
6188 /* If the regs structure we return has more elements than
6189 were in the pattern, set the extra elements to -1. If
6190 we (re)allocated the registers, this is the case,
6191 because we always allocate enough to have at least one
6193 for (mcnt
= num_regs
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
6194 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
6195 } /* regs && !bufp->no_sub */
6197 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6198 nfailure_points_pushed
, nfailure_points_popped
,
6199 nfailure_points_pushed
- nfailure_points_popped
);
6200 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
6203 if (MATCHING_IN_FIRST_STRING
)
6204 mcnt
= mbs_offset1
!= NULL
? mbs_offset1
[d
-string1
] : 0;
6206 mcnt
= (mbs_offset2
!= NULL
? mbs_offset2
[d
-string2
] : 0) +
6210 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
6215 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
6221 /* Otherwise match next pattern command. */
6222 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
6224 /* Ignore these. Used to ignore the n of succeed_n's which
6225 currently have n == 0. */
6227 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6231 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6234 /* Match the next n pattern characters exactly. The following
6235 byte in the pattern defines n, and the n bytes after that
6236 are the characters to match. */
6242 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
6244 /* This is written out as an if-else so we don't waste time
6245 testing `translate' inside the loop. */
6254 if ((UCHAR_T
) translate
[(unsigned char) *d
++]
6260 if (*d
++ != (CHAR_T
) *p
++)
6264 if ((UCHAR_T
) translate
[(unsigned char) *d
++]
6276 if (*d
++ != (CHAR_T
) *p
++) goto fail
;
6280 SET_REGS_MATCHED ();
6284 /* Match any character except possibly a newline or a null. */
6286 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6290 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
6291 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
6294 SET_REGS_MATCHED ();
6295 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d
);
6305 unsigned int i
, char_class_length
, coll_symbol_length
,
6306 equiv_class_length
, ranges_length
, chars_length
, length
;
6307 CHAR_T
*workp
, *workp2
, *charset_top
;
6308 #define WORK_BUFFER_SIZE 128
6309 CHAR_T str_buf
[WORK_BUFFER_SIZE
];
6314 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
6316 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6318 c
= TRANSLATE (*d
); /* The character to match. */
6321 nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
6323 charset_top
= p
- 1;
6324 char_class_length
= *p
++;
6325 coll_symbol_length
= *p
++;
6326 equiv_class_length
= *p
++;
6327 ranges_length
= *p
++;
6328 chars_length
= *p
++;
6329 /* p points charset[6], so the address of the next instruction
6330 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6331 where l=length of char_classes, m=length of collating_symbol,
6332 n=equivalence_class, o=length of char_range,
6333 p'=length of character. */
6335 /* Update p to indicate the next instruction. */
6336 p
+= char_class_length
+ coll_symbol_length
+ equiv_class_length
+
6337 2*ranges_length
+ chars_length
;
6339 /* match with char_class? */
6340 for (i
= 0; i
< char_class_length
; i
+= CHAR_CLASS_SIZE
)
6343 uintptr_t alignedp
= ((uintptr_t)workp
6344 + __alignof__(wctype_t) - 1)
6345 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6346 wctype
= *((wctype_t*)alignedp
);
6347 workp
+= CHAR_CLASS_SIZE
;
6349 if (__iswctype((wint_t)c
, wctype
))
6350 goto char_set_matched
;
6352 if (iswctype((wint_t)c
, wctype
))
6353 goto char_set_matched
;
6357 /* match with collating_symbol? */
6361 const unsigned char *extra
= (const unsigned char *)
6362 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
6364 for (workp2
= workp
+ coll_symbol_length
; workp
< workp2
;
6368 wextra
= (int32_t*)(extra
+ *workp
++);
6369 for (i
= 0; i
< *wextra
; ++i
)
6370 if (TRANSLATE(d
[i
]) != wextra
[1 + i
])
6375 /* Update d, however d will be incremented at
6376 char_set_matched:, we decrement d here. */
6378 goto char_set_matched
;
6382 else /* (nrules == 0) */
6384 /* If we can't look up collation data, we use wcscoll
6387 for (workp2
= workp
+ coll_symbol_length
; workp
< workp2
;)
6389 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6391 length
= __wcslen (workp
);
6393 length
= wcslen (workp
);
6396 /* If wcscoll(the collating symbol, whole string) > 0,
6397 any substring of the string never match with the
6398 collating symbol. */
6400 if (__wcscoll (workp
, d
) > 0)
6402 if (wcscoll (workp
, d
) > 0)
6405 workp
+= length
+ 1;
6409 /* First, we compare the collating symbol with
6410 the first character of the string.
6411 If it don't match, we add the next character to
6412 the compare buffer in turn. */
6413 for (i
= 0 ; i
< WORK_BUFFER_SIZE
-1 ; i
++, d
++)
6418 if (dend
== end_match_2
)
6424 /* add next character to the compare buffer. */
6425 str_buf
[i
] = TRANSLATE(*d
);
6426 str_buf
[i
+1] = '\0';
6429 match
= __wcscoll (workp
, str_buf
);
6431 match
= wcscoll (workp
, str_buf
);
6434 goto char_set_matched
;
6437 /* (str_buf > workp) indicate (str_buf + X > workp),
6438 because for all X (str_buf + X > str_buf).
6439 So we don't need continue this loop. */
6442 /* Otherwise(str_buf < workp),
6443 (str_buf+next_character) may equals (workp).
6444 So we continue this loop. */
6449 workp
+= length
+ 1;
6452 /* match with equivalence_class? */
6456 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6457 /* Try to match the equivalence class against
6458 those known to the collate implementation. */
6459 const int32_t *table
;
6460 const int32_t *weights
;
6461 const int32_t *extra
;
6462 const int32_t *indirect
;
6467 /* This #include defines a local function! */
6468 # include <locale/weightwc.h>
6470 table
= (const int32_t *)
6471 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEWC
);
6472 weights
= (const wint_t *)
6473 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTWC
);
6474 extra
= (const wint_t *)
6475 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAWC
);
6476 indirect
= (const int32_t *)
6477 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTWC
);
6479 /* Write 1 collating element to str_buf, and
6483 for (i
= 0 ; idx2
== 0 && i
< WORK_BUFFER_SIZE
- 1; i
++)
6485 cp
= (wint_t*)str_buf
;
6488 if (dend
== end_match_2
)
6493 str_buf
[i
] = TRANSLATE(*(d
+i
));
6494 str_buf
[i
+1] = '\0'; /* sentinel */
6495 idx2
= findidx ((const wint_t**)&cp
);
6498 /* Update d, however d will be incremented at
6499 char_set_matched:, we decrement d here. */
6500 d
= backup_d
+ ((wchar_t*)cp
- (wchar_t*)str_buf
- 1);
6503 if (dend
== end_match_2
)
6512 len
= weights
[idx2
];
6514 for (workp2
= workp
+ equiv_class_length
; workp
< workp2
;
6517 idx
= (int32_t)*workp
;
6518 /* We already checked idx != 0 in regex_compile. */
6520 if (idx2
!= 0 && len
== weights
[idx
])
6523 while (cnt
< len
&& (weights
[idx
+ 1 + cnt
]
6524 == weights
[idx2
+ 1 + cnt
]))
6528 goto char_set_matched
;
6535 else /* (nrules == 0) */
6537 /* If we can't look up collation data, we use wcscoll
6540 for (workp2
= workp
+ equiv_class_length
; workp
< workp2
;)
6542 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6544 length
= __wcslen (workp
);
6546 length
= wcslen (workp
);
6549 /* If wcscoll(the collating symbol, whole string) > 0,
6550 any substring of the string never match with the
6551 collating symbol. */
6553 if (__wcscoll (workp
, d
) > 0)
6555 if (wcscoll (workp
, d
) > 0)
6558 workp
+= length
+ 1;
6562 /* First, we compare the equivalence class with
6563 the first character of the string.
6564 If it don't match, we add the next character to
6565 the compare buffer in turn. */
6566 for (i
= 0 ; i
< WORK_BUFFER_SIZE
- 1 ; i
++, d
++)
6571 if (dend
== end_match_2
)
6577 /* add next character to the compare buffer. */
6578 str_buf
[i
] = TRANSLATE(*d
);
6579 str_buf
[i
+1] = '\0';
6582 match
= __wcscoll (workp
, str_buf
);
6584 match
= wcscoll (workp
, str_buf
);
6588 goto char_set_matched
;
6591 /* (str_buf > workp) indicate (str_buf + X > workp),
6592 because for all X (str_buf + X > str_buf).
6593 So we don't need continue this loop. */
6596 /* Otherwise(str_buf < workp),
6597 (str_buf+next_character) may equals (workp).
6598 So we continue this loop. */
6603 workp
+= length
+ 1;
6607 /* match with char_range? */
6611 uint32_t collseqval
;
6612 const char *collseq
= (const char *)
6613 _NL_CURRENT(LC_COLLATE
, _NL_COLLATE_COLLSEQWC
);
6615 collseqval
= collseq_table_lookup (collseq
, c
);
6617 for (; workp
< p
- chars_length
;)
6619 uint32_t start_val
, end_val
;
6621 /* We already compute the collation sequence value
6622 of the characters (or collating symbols). */
6623 start_val
= (uint32_t) *workp
++; /* range_start */
6624 end_val
= (uint32_t) *workp
++; /* range_end */
6626 if (start_val
<= collseqval
&& collseqval
<= end_val
)
6627 goto char_set_matched
;
6633 /* We set range_start_char at str_buf[0], range_end_char
6634 at str_buf[4], and compared char at str_buf[2]. */
6639 for (; workp
< p
- chars_length
;)
6641 wchar_t *range_start_char
, *range_end_char
;
6643 /* match if (range_start_char <= c <= range_end_char). */
6645 /* If range_start(or end) < 0, we assume -range_start(end)
6646 is the offset of the collating symbol which is specified
6647 as the character of the range start(end). */
6651 range_start_char
= charset_top
- (*workp
++);
6654 str_buf
[0] = *workp
++;
6655 range_start_char
= str_buf
;
6660 range_end_char
= charset_top
- (*workp
++);
6663 str_buf
[4] = *workp
++;
6664 range_end_char
= str_buf
+ 4;
6668 if (__wcscoll (range_start_char
, str_buf
+2) <= 0
6669 && __wcscoll (str_buf
+2, range_end_char
) <= 0)
6671 if (wcscoll (range_start_char
, str_buf
+2) <= 0
6672 && wcscoll (str_buf
+2, range_end_char
) <= 0)
6674 goto char_set_matched
;
6678 /* match with char? */
6679 for (; workp
< p
; workp
++)
6681 goto char_set_matched
;
6688 /* Cast to `unsigned' instead of `unsigned char' in case the
6689 bit list is a full 32 bytes long. */
6690 if (c
< (unsigned) (*p
* BYTEWIDTH
)
6691 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
6696 if (!not) goto fail
;
6697 #undef WORK_BUFFER_SIZE
6699 SET_REGS_MATCHED ();
6705 /* The beginning of a group is represented by start_memory.
6706 The arguments are the register number in the next byte, and the
6707 number of groups inner to this one in the next. The text
6708 matched within the group is recorded (in the internal
6709 registers data structure) under the register number. */
6711 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6712 (long int) *p
, (long int) p
[1]);
6714 /* Find out if this group can match the empty string. */
6715 p1
= p
; /* To send to group_match_null_string_p. */
6717 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
6718 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6719 = PREFIX(group_match_null_string_p
) (&p1
, pend
, reg_info
);
6721 /* Save the position in the string where we were the last time
6722 we were at this open-group operator in case the group is
6723 operated upon by a repetition operator, e.g., with `(a*)*b'
6724 against `ab'; then we want to ignore where we are now in
6725 the string in case this attempt to match fails. */
6726 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6727 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
6729 DEBUG_PRINT2 (" old_regstart: %d\n",
6730 POINTER_TO_OFFSET (old_regstart
[*p
]));
6733 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
6735 IS_ACTIVE (reg_info
[*p
]) = 1;
6736 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
6738 /* Clear this whenever we change the register activity status. */
6739 set_regs_matched_done
= 0;
6741 /* This is the new highest active register. */
6742 highest_active_reg
= *p
;
6744 /* If nothing was active before, this is the new lowest active
6746 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
6747 lowest_active_reg
= *p
;
6749 /* Move past the register number and inner group count. */
6751 just_past_start_mem
= p
;
6756 /* The stop_memory opcode represents the end of a group. Its
6757 arguments are the same as start_memory's: the register
6758 number, and the number of inner groups. */
6760 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6761 (long int) *p
, (long int) p
[1]);
6763 /* We need to save the string position the last time we were at
6764 this close-group operator in case the group is operated
6765 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6766 against `aba'; then we want to ignore where we are now in
6767 the string in case this attempt to match fails. */
6768 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6769 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
6771 DEBUG_PRINT2 (" old_regend: %d\n",
6772 POINTER_TO_OFFSET (old_regend
[*p
]));
6775 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
6777 /* This register isn't active anymore. */
6778 IS_ACTIVE (reg_info
[*p
]) = 0;
6780 /* Clear this whenever we change the register activity status. */
6781 set_regs_matched_done
= 0;
6783 /* If this was the only register active, nothing is active
6785 if (lowest_active_reg
== highest_active_reg
)
6787 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
6788 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
6791 { /* We must scan for the new highest active register, since
6792 it isn't necessarily one less than now: consider
6793 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6794 new highest active register is 1. */
6796 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
6799 /* If we end up at register zero, that means that we saved
6800 the registers as the result of an `on_failure_jump', not
6801 a `start_memory', and we jumped to past the innermost
6802 `stop_memory'. For example, in ((.)*) we save
6803 registers 1 and 2 as a result of the *, but when we pop
6804 back to the second ), we are at the stop_memory 1.
6805 Thus, nothing is active. */
6808 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
6809 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
6812 highest_active_reg
= r
;
6815 /* If just failed to match something this time around with a
6816 group that's operated on by a repetition operator, try to
6817 force exit from the ``loop'', and restore the register
6818 information for this group that we had before trying this
6820 if ((!MATCHED_SOMETHING (reg_info
[*p
])
6821 || just_past_start_mem
== p
- 1)
6824 boolean is_a_jump_n
= false;
6828 switch ((re_opcode_t
) *p1
++)
6832 case pop_failure_jump
:
6833 case maybe_pop_jump
:
6835 case dummy_failure_jump
:
6836 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
6838 p1
+= OFFSET_ADDRESS_SIZE
;
6846 /* If the next operation is a jump backwards in the pattern
6847 to an on_failure_jump right before the start_memory
6848 corresponding to this stop_memory, exit from the loop
6849 by forcing a failure after pushing on the stack the
6850 on_failure_jump's jump in the pattern, and d. */
6851 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
6852 && (re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == start_memory
6853 && p1
[2+OFFSET_ADDRESS_SIZE
] == *p
)
6855 /* If this group ever matched anything, then restore
6856 what its registers were before trying this last
6857 failed match, e.g., with `(a*)*b' against `ab' for
6858 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6859 against `aba' for regend[3].
6861 Also restore the registers for inner groups for,
6862 e.g., `((a*)(b*))*' against `aba' (register 3 would
6863 otherwise get trashed). */
6865 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
6869 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
6871 /* Restore this and inner groups' (if any) registers. */
6872 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
6875 regstart
[r
] = old_regstart
[r
];
6877 /* xx why this test? */
6878 if (old_regend
[r
] >= regstart
[r
])
6879 regend
[r
] = old_regend
[r
];
6883 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
6884 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
6890 /* Move past the register number and the inner group count. */
6895 /* \<digit> has been turned into a `duplicate' command which is
6896 followed by the numeric value of <digit> as the register number. */
6899 register const CHAR_T
*d2
, *dend2
;
6900 int regno
= *p
++; /* Get which register to match against. */
6901 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
6903 /* Can't back reference a group which we've never matched. */
6904 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
6907 /* Where in input to try to start matching. */
6908 d2
= regstart
[regno
];
6910 /* Where to stop matching; if both the place to start and
6911 the place to stop matching are in the same string, then
6912 set to the place to stop, otherwise, for now have to use
6913 the end of the first string. */
6915 dend2
= ((FIRST_STRING_P (regstart
[regno
])
6916 == FIRST_STRING_P (regend
[regno
]))
6917 ? regend
[regno
] : end_match_1
);
6920 /* If necessary, advance to next segment in register
6924 if (dend2
== end_match_2
) break;
6925 if (dend2
== regend
[regno
]) break;
6927 /* End of string1 => advance to string2. */
6929 dend2
= regend
[regno
];
6931 /* At end of register contents => success */
6932 if (d2
== dend2
) break;
6934 /* If necessary, advance to next segment in data. */
6937 /* How many characters left in this segment to match. */
6940 /* Want how many consecutive characters we can match in
6941 one shot, so, if necessary, adjust the count. */
6942 if (mcnt
> dend2
- d2
)
6945 /* Compare that many; failure if mismatch, else move
6948 ? PREFIX(bcmp_translate
) (d
, d2
, mcnt
, translate
)
6949 : memcmp (d
, d2
, mcnt
*sizeof(UCHAR_T
)))
6951 d
+= mcnt
, d2
+= mcnt
;
6953 /* Do this because we've match some characters. */
6954 SET_REGS_MATCHED ();
6960 /* begline matches the empty string at the beginning of the string
6961 (unless `not_bol' is set in `bufp'), and, if
6962 `newline_anchor' is set, after newlines. */
6964 DEBUG_PRINT1 ("EXECUTING begline.\n");
6966 if (AT_STRINGS_BEG (d
))
6968 if (!bufp
->not_bol
) break;
6970 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
6974 /* In all other cases, we fail. */
6978 /* endline is the dual of begline. */
6980 DEBUG_PRINT1 ("EXECUTING endline.\n");
6982 if (AT_STRINGS_END (d
))
6984 if (!bufp
->not_eol
) break;
6987 /* We have to ``prefetch'' the next character. */
6988 else if ((d
== end1
? *string2
: *d
) == '\n'
6989 && bufp
->newline_anchor
)
6996 /* Match at the very beginning of the data. */
6998 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
6999 if (AT_STRINGS_BEG (d
))
7004 /* Match at the very end of the data. */
7006 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
7007 if (AT_STRINGS_END (d
))
7012 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
7013 pushes NULL as the value for the string on the stack. Then
7014 `pop_failure_point' will keep the current value for the
7015 string, instead of restoring it. To see why, consider
7016 matching `foo\nbar' against `.*\n'. The .* matches the foo;
7017 then the . fails against the \n. But the next thing we want
7018 to do is match the \n against the \n; if we restored the
7019 string value, we would be back at the foo.
7021 Because this is used only in specific cases, we don't need to
7022 check all the things that `on_failure_jump' does, to make
7023 sure the right things get saved on the stack. Hence we don't
7024 share its code. The only reason to push anything on the
7025 stack at all is that otherwise we would have to change
7026 `anychar's code to do something besides goto fail in this
7027 case; that seems worse than this. */
7028 case on_failure_keep_string_jump
:
7029 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
7031 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7033 DEBUG_PRINT3 (" %d (to %p):\n", mcnt
, p
+ mcnt
);
7035 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
7038 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
7042 /* Uses of on_failure_jump:
7044 Each alternative starts with an on_failure_jump that points
7045 to the beginning of the next alternative. Each alternative
7046 except the last ends with a jump that in effect jumps past
7047 the rest of the alternatives. (They really jump to the
7048 ending jump of the following alternative, because tensioning
7049 these jumps is a hassle.)
7051 Repeats start with an on_failure_jump that points past both
7052 the repetition text and either the following jump or
7053 pop_failure_jump back to this on_failure_jump. */
7054 case on_failure_jump
:
7056 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
7058 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7060 DEBUG_PRINT3 (" %d (to %p)", mcnt
, p
+ mcnt
);
7062 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
7065 /* If this on_failure_jump comes right before a group (i.e.,
7066 the original * applied to a group), save the information
7067 for that group and all inner ones, so that if we fail back
7068 to this point, the group's information will be correct.
7069 For example, in \(a*\)*\1, we need the preceding group,
7070 and in \(zz\(a*\)b*\)\2, we need the inner group. */
7072 /* We can't use `p' to check ahead because we push
7073 a failure point to `p + mcnt' after we do this. */
7076 /* We need to skip no_op's before we look for the
7077 start_memory in case this on_failure_jump is happening as
7078 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7080 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
7083 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
7085 /* We have a new highest active register now. This will
7086 get reset at the start_memory we are about to get to,
7087 but we will have saved all the registers relevant to
7088 this repetition op, as described above. */
7089 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
7090 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
7091 lowest_active_reg
= *(p1
+ 1);
7094 DEBUG_PRINT1 (":\n");
7095 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
7099 /* A smart repeat ends with `maybe_pop_jump'.
7100 We change it to either `pop_failure_jump' or `jump'. */
7101 case maybe_pop_jump
:
7102 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7103 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
7105 register UCHAR_T
*p2
= p
;
7107 /* Compare the beginning of the repeat with what in the
7108 pattern follows its end. If we can establish that there
7109 is nothing that they would both match, i.e., that we
7110 would have to backtrack because of (as in, e.g., `a*a')
7111 then we can change to pop_failure_jump, because we'll
7112 never have to backtrack.
7114 This is not true in the case of alternatives: in
7115 `(a|ab)*' we do need to backtrack to the `ab' alternative
7116 (e.g., if the string was `ab'). But instead of trying to
7117 detect that here, the alternative has put on a dummy
7118 failure point which is what we will end up popping. */
7120 /* Skip over open/close-group commands.
7121 If what follows this loop is a ...+ construct,
7122 look at what begins its body, since we will have to
7123 match at least one of that. */
7127 && ((re_opcode_t
) *p2
== stop_memory
7128 || (re_opcode_t
) *p2
== start_memory
))
7130 else if (p2
+ 2 + 2 * OFFSET_ADDRESS_SIZE
< pend
7131 && (re_opcode_t
) *p2
== dummy_failure_jump
)
7132 p2
+= 2 + 2 * OFFSET_ADDRESS_SIZE
;
7138 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7139 to the `maybe_finalize_jump' of this case. Examine what
7142 /* If we're at the end of the pattern, we can change. */
7145 /* Consider what happens when matching ":\(.*\)"
7146 against ":/". I don't really understand this code
7148 p
[-(1+OFFSET_ADDRESS_SIZE
)] = (UCHAR_T
)
7151 (" End of pattern: change to `pop_failure_jump'.\n");
7154 else if ((re_opcode_t
) *p2
== exactn
7156 || (re_opcode_t
) *p2
== exactn_bin
7158 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
7161 = *p2
== (UCHAR_T
) endline
? '\n' : p2
[2];
7163 if (((re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == exactn
7165 || (re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == exactn_bin
7167 ) && p1
[3+OFFSET_ADDRESS_SIZE
] != c
)
7169 p
[-(1+OFFSET_ADDRESS_SIZE
)] = (UCHAR_T
)
7172 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7174 (wint_t) p1
[3+OFFSET_ADDRESS_SIZE
]);
7176 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7178 (char) p1
[3+OFFSET_ADDRESS_SIZE
]);
7183 else if ((re_opcode_t
) p1
[3] == charset
7184 || (re_opcode_t
) p1
[3] == charset_not
)
7186 int not = (re_opcode_t
) p1
[3] == charset_not
;
7188 if (c
< (unsigned) (p1
[4] * BYTEWIDTH
)
7189 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
7192 /* `not' is equal to 1 if c would match, which means
7193 that we can't change to pop_failure_jump. */
7196 p
[-3] = (unsigned char) pop_failure_jump
;
7197 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7200 #endif /* not WCHAR */
7203 else if ((re_opcode_t
) *p2
== charset
)
7205 /* We win if the first character of the loop is not part
7207 if ((re_opcode_t
) p1
[3] == exactn
7208 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
7209 && (p2
[2 + p1
[5] / BYTEWIDTH
]
7210 & (1 << (p1
[5] % BYTEWIDTH
)))))
7212 p
[-3] = (unsigned char) pop_failure_jump
;
7213 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7216 else if ((re_opcode_t
) p1
[3] == charset_not
)
7219 /* We win if the charset_not inside the loop
7220 lists every character listed in the charset after. */
7221 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
7222 if (! (p2
[2 + idx
] == 0
7223 || (idx
< (int) p1
[4]
7224 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
7229 p
[-3] = (unsigned char) pop_failure_jump
;
7230 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7233 else if ((re_opcode_t
) p1
[3] == charset
)
7236 /* We win if the charset inside the loop
7237 has no overlap with the one after the loop. */
7239 idx
< (int) p2
[1] && idx
< (int) p1
[4];
7241 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
7244 if (idx
== p2
[1] || idx
== p1
[4])
7246 p
[-3] = (unsigned char) pop_failure_jump
;
7247 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7251 #endif /* not WCHAR */
7253 p
-= OFFSET_ADDRESS_SIZE
; /* Point at relative address again. */
7254 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
7256 p
[-1] = (UCHAR_T
) jump
;
7257 DEBUG_PRINT1 (" Match => jump.\n");
7258 goto unconditional_jump
;
7260 /* Note fall through. */
7263 /* The end of a simple repeat has a pop_failure_jump back to
7264 its matching on_failure_jump, where the latter will push a
7265 failure point. The pop_failure_jump takes off failure
7266 points put on by this pop_failure_jump's matching
7267 on_failure_jump; we got through the pattern to here from the
7268 matching on_failure_jump, so didn't fail. */
7269 case pop_failure_jump
:
7271 /* We need to pass separate storage for the lowest and
7272 highest registers, even though we don't care about the
7273 actual values. Otherwise, we will restore only one
7274 register from the stack, since lowest will == highest in
7275 `pop_failure_point'. */
7276 active_reg_t dummy_low_reg
, dummy_high_reg
;
7277 UCHAR_T
*pdummy
= NULL
;
7278 const CHAR_T
*sdummy
= NULL
;
7280 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7281 POP_FAILURE_POINT (sdummy
, pdummy
,
7282 dummy_low_reg
, dummy_high_reg
,
7283 reg_dummy
, reg_dummy
, reg_info_dummy
);
7285 /* Note fall through. */
7289 DEBUG_PRINT2 ("\n%p: ", p
);
7291 DEBUG_PRINT2 ("\n0x%x: ", p
);
7293 /* Note fall through. */
7295 /* Unconditionally jump (without popping any failure points). */
7297 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
7298 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
7299 p
+= mcnt
; /* Do the jump. */
7301 DEBUG_PRINT2 ("(to %p).\n", p
);
7303 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
7308 /* We need this opcode so we can detect where alternatives end
7309 in `group_match_null_string_p' et al. */
7311 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7312 goto unconditional_jump
;
7315 /* Normally, the on_failure_jump pushes a failure point, which
7316 then gets popped at pop_failure_jump. We will end up at
7317 pop_failure_jump, also, and with a pattern of, say, `a+', we
7318 are skipping over the on_failure_jump, so we have to push
7319 something meaningless for pop_failure_jump to pop. */
7320 case dummy_failure_jump
:
7321 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7322 /* It doesn't matter what we push for the string here. What
7323 the code at `fail' tests is the value for the pattern. */
7324 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
7325 goto unconditional_jump
;
7328 /* At the end of an alternative, we need to push a dummy failure
7329 point in case we are followed by a `pop_failure_jump', because
7330 we don't want the failure point for the alternative to be
7331 popped. For example, matching `(a|ab)*' against `aab'
7332 requires that we match the `ab' alternative. */
7333 case push_dummy_failure
:
7334 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7335 /* See comments just above at `dummy_failure_jump' about the
7337 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
7340 /* Have to succeed matching what follows at least n times.
7341 After that, handle like `on_failure_jump'. */
7343 EXTRACT_NUMBER (mcnt
, p
+ OFFSET_ADDRESS_SIZE
);
7344 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
7347 /* Originally, this is how many times we HAVE to succeed. */
7351 p
+= OFFSET_ADDRESS_SIZE
;
7352 STORE_NUMBER_AND_INCR (p
, mcnt
);
7354 DEBUG_PRINT3 (" Setting %p to %d.\n", p
- OFFSET_ADDRESS_SIZE
7357 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
- OFFSET_ADDRESS_SIZE
7364 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7365 p
+ OFFSET_ADDRESS_SIZE
);
7367 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7368 p
+ OFFSET_ADDRESS_SIZE
);
7372 p
[1] = (UCHAR_T
) no_op
;
7374 p
[2] = (UCHAR_T
) no_op
;
7375 p
[3] = (UCHAR_T
) no_op
;
7382 EXTRACT_NUMBER (mcnt
, p
+ OFFSET_ADDRESS_SIZE
);
7383 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
7385 /* Originally, this is how many times we CAN jump. */
7389 STORE_NUMBER (p
+ OFFSET_ADDRESS_SIZE
, mcnt
);
7392 DEBUG_PRINT3 (" Setting %p to %d.\n", p
+ OFFSET_ADDRESS_SIZE
,
7395 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
+ OFFSET_ADDRESS_SIZE
,
7398 goto unconditional_jump
;
7400 /* If don't have to jump any more, skip over the rest of command. */
7402 p
+= 2 * OFFSET_ADDRESS_SIZE
;
7407 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7409 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7411 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7413 DEBUG_PRINT3 (" Setting %p to %d.\n", p1
, mcnt
);
7415 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
7417 STORE_NUMBER (p1
, mcnt
);
7422 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7423 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7424 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7425 macro and introducing temporary variables works around the bug. */
7428 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7429 if (AT_WORD_BOUNDARY (d
))
7434 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7435 if (AT_WORD_BOUNDARY (d
))
7441 boolean prevchar
, thischar
;
7443 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7444 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
7447 prevchar
= WORDCHAR_P (d
- 1);
7448 thischar
= WORDCHAR_P (d
);
7449 if (prevchar
!= thischar
)
7456 boolean prevchar
, thischar
;
7458 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7459 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
7462 prevchar
= WORDCHAR_P (d
- 1);
7463 thischar
= WORDCHAR_P (d
);
7464 if (prevchar
!= thischar
)
7471 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7472 if (!AT_STRINGS_END (d
) && WORDCHAR_P (d
)
7473 && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
7478 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7479 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
7480 && (AT_STRINGS_END (d
) || !WORDCHAR_P (d
)))
7486 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7487 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
7492 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7493 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
7498 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7499 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
7504 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
7509 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7513 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7515 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
7517 SET_REGS_MATCHED ();
7521 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
7523 goto matchnotsyntax
;
7526 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7530 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7532 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
7534 SET_REGS_MATCHED ();
7537 #else /* not emacs */
7539 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7541 if (!WORDCHAR_P (d
))
7543 SET_REGS_MATCHED ();
7548 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7552 SET_REGS_MATCHED ();
7555 #endif /* not emacs */
7560 continue; /* Successfully executed one pattern command; keep going. */
7563 /* We goto here if a matching operation fails. */
7565 if (!FAIL_STACK_EMPTY ())
7566 { /* A restart point is known. Restore to that state. */
7567 DEBUG_PRINT1 ("\nFAIL:\n");
7568 POP_FAILURE_POINT (d
, p
,
7569 lowest_active_reg
, highest_active_reg
,
7570 regstart
, regend
, reg_info
);
7572 /* If this failure point is a dummy, try the next one. */
7576 /* If we failed to the end of the pattern, don't examine *p. */
7580 boolean is_a_jump_n
= false;
7582 /* If failed to a backwards jump that's part of a repetition
7583 loop, need to pop this failure point and use the next one. */
7584 switch ((re_opcode_t
) *p
)
7588 case maybe_pop_jump
:
7589 case pop_failure_jump
:
7592 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7595 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
7597 && (re_opcode_t
) *p1
== on_failure_jump
))
7605 if (d
>= string1
&& d
<= end1
)
7609 break; /* Matching at this starting point really fails. */
7613 goto restore_best_regs
;
7617 return -1; /* Failure to match. */
7620 /* Subroutine definitions for re_match_2. */
7623 /* We are passed P pointing to a register number after a start_memory.
7625 Return true if the pattern up to the corresponding stop_memory can
7626 match the empty string, and false otherwise.
7628 If we find the matching stop_memory, sets P to point to one past its number.
7629 Otherwise, sets P to an undefined byte less than or equal to END.
7631 We don't handle duplicates properly (yet). */
7634 PREFIX(group_match_null_string_p
) (p
, end
, reg_info
)
7636 PREFIX(register_info_type
) *reg_info
;
7639 /* Point to after the args to the start_memory. */
7640 UCHAR_T
*p1
= *p
+ 2;
7644 /* Skip over opcodes that can match nothing, and return true or
7645 false, as appropriate, when we get to one that can't, or to the
7646 matching stop_memory. */
7648 switch ((re_opcode_t
) *p1
)
7650 /* Could be either a loop or a series of alternatives. */
7651 case on_failure_jump
:
7653 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7655 /* If the next operation is not a jump backwards in the
7660 /* Go through the on_failure_jumps of the alternatives,
7661 seeing if any of the alternatives cannot match nothing.
7662 The last alternative starts with only a jump,
7663 whereas the rest start with on_failure_jump and end
7664 with a jump, e.g., here is the pattern for `a|b|c':
7666 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7667 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7670 So, we have to first go through the first (n-1)
7671 alternatives and then deal with the last one separately. */
7674 /* Deal with the first (n-1) alternatives, which start
7675 with an on_failure_jump (see above) that jumps to right
7676 past a jump_past_alt. */
7678 while ((re_opcode_t
) p1
[mcnt
-(1+OFFSET_ADDRESS_SIZE
)] ==
7681 /* `mcnt' holds how many bytes long the alternative
7682 is, including the ending `jump_past_alt' and
7685 if (!PREFIX(alt_match_null_string_p
) (p1
, p1
+ mcnt
-
7686 (1 + OFFSET_ADDRESS_SIZE
),
7690 /* Move to right after this alternative, including the
7694 /* Break if it's the beginning of an n-th alternative
7695 that doesn't begin with an on_failure_jump. */
7696 if ((re_opcode_t
) *p1
!= on_failure_jump
)
7699 /* Still have to check that it's not an n-th
7700 alternative that starts with an on_failure_jump. */
7702 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7703 if ((re_opcode_t
) p1
[mcnt
-(1+OFFSET_ADDRESS_SIZE
)] !=
7706 /* Get to the beginning of the n-th alternative. */
7707 p1
-= 1 + OFFSET_ADDRESS_SIZE
;
7712 /* Deal with the last alternative: go back and get number
7713 of the `jump_past_alt' just before it. `mcnt' contains
7714 the length of the alternative. */
7715 EXTRACT_NUMBER (mcnt
, p1
- OFFSET_ADDRESS_SIZE
);
7717 if (!PREFIX(alt_match_null_string_p
) (p1
, p1
+ mcnt
, reg_info
))
7720 p1
+= mcnt
; /* Get past the n-th alternative. */
7726 assert (p1
[1] == **p
);
7732 if (!PREFIX(common_op_match_null_string_p
) (&p1
, end
, reg_info
))
7735 } /* while p1 < end */
7738 } /* group_match_null_string_p */
7741 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7742 It expects P to be the first byte of a single alternative and END one
7743 byte past the last. The alternative can contain groups. */
7746 PREFIX(alt_match_null_string_p
) (p
, end
, reg_info
)
7748 PREFIX(register_info_type
) *reg_info
;
7755 /* Skip over opcodes that can match nothing, and break when we get
7756 to one that can't. */
7758 switch ((re_opcode_t
) *p1
)
7761 case on_failure_jump
:
7763 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7768 if (!PREFIX(common_op_match_null_string_p
) (&p1
, end
, reg_info
))
7771 } /* while p1 < end */
7774 } /* alt_match_null_string_p */
7777 /* Deals with the ops common to group_match_null_string_p and
7778 alt_match_null_string_p.
7780 Sets P to one after the op and its arguments, if any. */
7783 PREFIX(common_op_match_null_string_p
) (p
, end
, reg_info
)
7785 PREFIX(register_info_type
) *reg_info
;
7792 switch ((re_opcode_t
) *p1
++)
7812 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
7813 ret
= PREFIX(group_match_null_string_p
) (&p1
, end
, reg_info
);
7815 /* Have to set this here in case we're checking a group which
7816 contains a group and a back reference to it. */
7818 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
7819 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
7825 /* If this is an optimized succeed_n for zero times, make the jump. */
7827 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7835 /* Get to the number of times to succeed. */
7836 p1
+= OFFSET_ADDRESS_SIZE
;
7837 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7841 p1
-= 2 * OFFSET_ADDRESS_SIZE
;
7842 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7850 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
7855 p1
+= 2 * OFFSET_ADDRESS_SIZE
;
7858 /* All other opcodes mean we cannot match the empty string. */
7864 } /* common_op_match_null_string_p */
7867 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7868 bytes; nonzero otherwise. */
7871 PREFIX(bcmp_translate
) (s1
, s2
, len
, translate
)
7872 const CHAR_T
*s1
, *s2
;
7874 RE_TRANSLATE_TYPE translate
;
7876 register const UCHAR_T
*p1
= (const UCHAR_T
*) s1
;
7877 register const UCHAR_T
*p2
= (const UCHAR_T
*) s2
;
7881 if (((*p1
<=0xff)?translate
[*p1
++]:*p1
++)
7882 != ((*p2
<=0xff)?translate
[*p2
++]:*p2
++))
7885 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
7893 #else /* not INSIDE_RECURSION */
7895 /* Entry points for GNU code. */
7897 /* re_compile_pattern is the GNU regular expression compiler: it
7898 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7899 Returns 0 if the pattern was valid, otherwise an error string.
7901 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7902 are set in BUFP on entry.
7904 We call regex_compile to do the actual compilation. */
7907 re_compile_pattern (pattern
, length
, bufp
)
7908 const char *pattern
;
7910 struct re_pattern_buffer
*bufp
;
7914 /* GNU code is written to assume at least RE_NREGS registers will be set
7915 (and at least one extra will be -1). */
7916 bufp
->regs_allocated
= REGS_UNALLOCATED
;
7918 /* And GNU code determines whether or not to get register information
7919 by passing null for the REGS argument to re_match, etc., not by
7923 /* Match anchors at newline. */
7924 bufp
->newline_anchor
= 1;
7927 if (MB_CUR_MAX
!= 1)
7928 ret
= wcs_regex_compile (pattern
, length
, re_syntax_options
, bufp
);
7931 ret
= byte_regex_compile (pattern
, length
, re_syntax_options
, bufp
);
7935 return gettext (re_error_msgid
+ re_error_msgid_idx
[(int) ret
]);
7938 weak_alias (__re_compile_pattern
, re_compile_pattern
)
7941 /* Entry points compatible with 4.2 BSD regex library. We don't define
7942 them unless specifically requested. */
7944 #if defined _REGEX_RE_COMP || defined _LIBC
7946 /* BSD has one and only one pattern buffer. */
7947 static struct re_pattern_buffer re_comp_buf
;
7951 /* Make these definitions weak in libc, so POSIX programs can redefine
7952 these names if they don't use our functions, and still use
7953 regcomp/regexec below without link errors. */
7963 if (!re_comp_buf
.buffer
)
7964 return gettext ("No previous regular expression");
7968 if (!re_comp_buf
.buffer
)
7970 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
7971 if (re_comp_buf
.buffer
== NULL
)
7972 return (char *) gettext (re_error_msgid
7973 + re_error_msgid_idx
[(int) REG_ESPACE
]);
7974 re_comp_buf
.allocated
= 200;
7976 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
7977 if (re_comp_buf
.fastmap
== NULL
)
7978 return (char *) gettext (re_error_msgid
7979 + re_error_msgid_idx
[(int) REG_ESPACE
]);
7982 /* Since `re_exec' always passes NULL for the `regs' argument, we
7983 don't need to initialize the pattern buffer fields which affect it. */
7985 /* Match anchors at newlines. */
7986 re_comp_buf
.newline_anchor
= 1;
7989 if (MB_CUR_MAX
!= 1)
7990 ret
= wcs_regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
7993 ret
= byte_regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
7998 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
7999 return (char *) gettext (re_error_msgid
+ re_error_msgid_idx
[(int) ret
]);
8010 const int len
= strlen (s
);
8012 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
8015 #endif /* _REGEX_RE_COMP */
8017 /* POSIX.2 functions. Don't define these for Emacs. */
8021 /* regcomp takes a regular expression as a string and compiles it.
8023 PREG is a regex_t *. We do not expect any fields to be initialized,
8024 since POSIX says we shouldn't. Thus, we set
8026 `buffer' to the compiled pattern;
8027 `used' to the length of the compiled pattern;
8028 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
8029 REG_EXTENDED bit in CFLAGS is set; otherwise, to
8030 RE_SYNTAX_POSIX_BASIC;
8031 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
8032 `fastmap' to an allocated space for the fastmap;
8033 `fastmap_accurate' to zero;
8034 `re_nsub' to the number of subexpressions in PATTERN.
8036 PATTERN is the address of the pattern string.
8038 CFLAGS is a series of bits which affect compilation.
8040 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
8041 use POSIX basic syntax.
8043 If REG_NEWLINE is set, then . and [^...] don't match newline.
8044 Also, regexec will try a match beginning after every newline.
8046 If REG_ICASE is set, then we considers upper- and lowercase
8047 versions of letters to be equivalent when matching.
8049 If REG_NOSUB is set, then when PREG is passed to regexec, that
8050 routine will report only success or failure, and nothing about the
8053 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
8054 the return codes and their meanings.) */
8057 regcomp (preg
, pattern
, cflags
)
8059 const char *pattern
;
8064 = (cflags
& REG_EXTENDED
) ?
8065 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
8067 /* regex_compile will allocate the space for the compiled pattern. */
8069 preg
->allocated
= 0;
8072 /* Try to allocate space for the fastmap. */
8073 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
8075 if (cflags
& REG_ICASE
)
8080 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
8081 * sizeof (*(RE_TRANSLATE_TYPE
)0));
8082 if (preg
->translate
== NULL
)
8083 return (int) REG_ESPACE
;
8085 /* Map uppercase characters to corresponding lowercase ones. */
8086 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
8087 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
8090 preg
->translate
= NULL
;
8092 /* If REG_NEWLINE is set, newlines are treated differently. */
8093 if (cflags
& REG_NEWLINE
)
8094 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
8095 syntax
&= ~RE_DOT_NEWLINE
;
8096 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
8097 /* It also changes the matching behavior. */
8098 preg
->newline_anchor
= 1;
8101 preg
->newline_anchor
= 0;
8103 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
8105 /* POSIX says a null character in the pattern terminates it, so we
8106 can use strlen here in compiling the pattern. */
8108 if (MB_CUR_MAX
!= 1)
8109 ret
= wcs_regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
8112 ret
= byte_regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
8114 /* POSIX doesn't distinguish between an unmatched open-group and an
8115 unmatched close-group: both are REG_EPAREN. */
8116 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
8118 if (ret
== REG_NOERROR
&& preg
->fastmap
)
8120 /* Compute the fastmap now, since regexec cannot modify the pattern
8122 if (re_compile_fastmap (preg
) == -2)
8124 /* Some error occurred while computing the fastmap, just forget
8126 free (preg
->fastmap
);
8127 preg
->fastmap
= NULL
;
8134 weak_alias (__regcomp
, regcomp
)
8138 /* regexec searches for a given pattern, specified by PREG, in the
8141 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8142 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8143 least NMATCH elements, and we set them to the offsets of the
8144 corresponding matched substrings.
8146 EFLAGS specifies `execution flags' which affect matching: if
8147 REG_NOTBOL is set, then ^ does not match at the beginning of the
8148 string; if REG_NOTEOL is set, then $ does not match at the end.
8150 We return 0 if we find a match and REG_NOMATCH if not. */
8153 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
8154 const regex_t
*preg
;
8157 regmatch_t pmatch
[];
8161 struct re_registers regs
;
8162 regex_t private_preg
;
8163 int len
= strlen (string
);
8164 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
8166 private_preg
= *preg
;
8168 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
8169 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
8171 /* The user has told us exactly how many registers to return
8172 information about, via `nmatch'. We have to pass that on to the
8173 matching routines. */
8174 private_preg
.regs_allocated
= REGS_FIXED
;
8178 regs
.num_regs
= nmatch
;
8179 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
8180 if (regs
.start
== NULL
)
8181 return (int) REG_NOMATCH
;
8182 regs
.end
= regs
.start
+ nmatch
;
8185 /* Perform the searching operation. */
8186 ret
= re_search (&private_preg
, string
, len
,
8187 /* start: */ 0, /* range: */ len
,
8188 want_reg_info
? ®s
: (struct re_registers
*) 0);
8190 /* Copy the register information to the POSIX structure. */
8197 for (r
= 0; r
< nmatch
; r
++)
8199 pmatch
[r
].rm_so
= regs
.start
[r
];
8200 pmatch
[r
].rm_eo
= regs
.end
[r
];
8204 /* If we needed the temporary register info, free the space now. */
8208 /* We want zero return to mean success, unlike `re_search'. */
8209 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
8212 weak_alias (__regexec
, regexec
)
8216 /* Returns a message corresponding to an error code, ERRCODE, returned
8217 from either regcomp or regexec. We don't use PREG here. */
8220 regerror (errcode
, preg
, errbuf
, errbuf_size
)
8222 const regex_t
*preg
;
8230 || errcode
>= (int) (sizeof (re_error_msgid_idx
)
8231 / sizeof (re_error_msgid_idx
[0])))
8232 /* Only error codes returned by the rest of the code should be passed
8233 to this routine. If we are given anything else, or if other regex
8234 code generates an invalid error code, then the program has a bug.
8235 Dump core so we can fix it. */
8238 msg
= gettext (re_error_msgid
+ re_error_msgid_idx
[errcode
]);
8240 msg_size
= strlen (msg
) + 1; /* Includes the null. */
8242 if (errbuf_size
!= 0)
8244 if (msg_size
> errbuf_size
)
8246 #if defined HAVE_MEMPCPY || defined _LIBC
8247 *((char *) __mempcpy (errbuf
, msg
, errbuf_size
- 1)) = '\0';
8249 memcpy (errbuf
, msg
, errbuf_size
- 1);
8250 errbuf
[errbuf_size
- 1] = 0;
8254 memcpy (errbuf
, msg
, msg_size
);
8260 weak_alias (__regerror
, regerror
)
8264 /* Free dynamically allocated space used by PREG. */
8270 if (preg
->buffer
!= NULL
)
8271 free (preg
->buffer
);
8272 preg
->buffer
= NULL
;
8274 preg
->allocated
= 0;
8277 if (preg
->fastmap
!= NULL
)
8278 free (preg
->fastmap
);
8279 preg
->fastmap
= NULL
;
8280 preg
->fastmap_accurate
= 0;
8282 if (preg
->translate
!= NULL
)
8283 free (preg
->translate
);
8284 preg
->translate
= NULL
;
8287 weak_alias (__regfree
, regfree
)
8290 #endif /* not emacs */
8292 #endif /* not INSIDE_RECURSION */
8296 #undef STORE_NUMBER_AND_INCR
8297 #undef EXTRACT_NUMBER
8298 #undef EXTRACT_NUMBER_AND_INCR
8300 #undef DEBUG_PRINT_COMPILED_PATTERN
8301 #undef DEBUG_PRINT_DOUBLE_STRING
8303 #undef INIT_FAIL_STACK
8304 #undef RESET_FAIL_STACK
8305 #undef DOUBLE_FAIL_STACK
8306 #undef PUSH_PATTERN_OP
8307 #undef PUSH_FAILURE_POINTER
8308 #undef PUSH_FAILURE_INT
8309 #undef PUSH_FAILURE_ELT
8310 #undef POP_FAILURE_POINTER
8311 #undef POP_FAILURE_INT
8312 #undef POP_FAILURE_ELT
8315 #undef PUSH_FAILURE_POINT
8316 #undef POP_FAILURE_POINT
8318 #undef REG_UNSET_VALUE
8326 #undef INIT_BUF_SIZE
8327 #undef GET_BUFFER_SPACE
8335 #undef EXTEND_BUFFER
8336 #undef GET_UNSIGNED_NUMBER
8337 #undef FREE_STACK_RETURN
8339 # undef POINTER_TO_OFFSET
8340 # undef MATCHING_IN_FRST_STRING
8342 # undef AT_STRINGS_BEG
8343 # undef AT_STRINGS_END
8346 # undef FREE_VARIABLES
8347 # undef NO_HIGHEST_ACTIVE_REG
8348 # undef NO_LOWEST_ACTIVE_REG
8352 # undef COMPILED_BUFFER_VAR
8353 # undef OFFSET_ADDRESS_SIZE
8354 # undef CHAR_CLASS_SIZE
8361 # define DEFINED_ONCE
8362 #endif /* USE_INCLUDED_REGEX */