1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
6 Copyright (C) 1993-2023 Free Software Foundation, Inc.
7 This file is part of the GNU C Library.
9 The GNU C Library is free software; you can redistribute it and/or
10 modify it under the terms of the GNU Lesser General Public
11 License as published by the Free Software Foundation; either
12 version 2.1 of the License, or (at your option) any later version.
14 The GNU C Library is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 Lesser General Public License for more details.
19 You should have received a copy of the GNU Lesser General Public
20 License along with the GNU C Library; if not, write to the Free
21 Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
24 /* This file has been modified for usage in libiberty. It includes "xregex.h"
25 instead of <regex.h>. The "xregex.h" header file renames all external
26 routines with an "x" prefix so they do not collide with the native regex
27 routines or with other components regex routines. */
28 /* AIX requires this to be the first thing in the file. */
29 #if defined _AIX && !defined __GNUC__ && !defined REGEX_MALLOC
34 # pragma GCC diagnostic ignored "-Wuse-after-free"
40 #ifndef INSIDE_RECURSION
48 #ifndef INSIDE_RECURSION
50 # if defined STDC_HEADERS && !defined emacs
52 # define PTR_INT_TYPE ptrdiff_t
54 /* We need this for `regex.h', and perhaps for the Emacs include files. */
55 # include <sys/types.h>
56 # define PTR_INT_TYPE long
59 # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
61 /* For platform which support the ISO C amendement 1 functionality we
62 support user defined character classes. */
63 # if defined _LIBC || WIDE_CHAR_SUPPORT
64 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
70 /* We have to keep the namespace clean. */
71 # define regfree(preg) __regfree (preg)
72 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
73 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
74 # define regerror(errcode, preg, errbuf, errbuf_size) \
75 __regerror(errcode, preg, errbuf, errbuf_size)
76 # define re_set_registers(bu, re, nu, st, en) \
77 __re_set_registers (bu, re, nu, st, en)
78 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
79 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
80 # define re_match(bufp, string, size, pos, regs) \
81 __re_match (bufp, string, size, pos, regs)
82 # define re_search(bufp, string, size, startpos, range, regs) \
83 __re_search (bufp, string, size, startpos, range, regs)
84 # define re_compile_pattern(pattern, length, bufp) \
85 __re_compile_pattern (pattern, length, bufp)
86 # define re_set_syntax(syntax) __re_set_syntax (syntax)
87 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
88 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
89 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
91 # define btowc __btowc
93 /* We are also using some library internals. */
94 # include <locale/localeinfo.h>
95 # include <locale/elem-hash.h>
96 # include <langinfo.h>
97 # include <locale/coll-lookup.h>
100 /* This is for other GNU distributions with internationalized messages. */
101 # if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC
102 # include <libintl.h>
105 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
108 # define gettext(msgid) (msgid)
111 # ifndef gettext_noop
112 /* This define is so xgettext can find the internationalizable
114 # define gettext_noop(String) String
117 /* The `emacs' switch turns on certain matching commands
118 that make sense only in Emacs. */
125 # else /* not emacs */
127 /* If we are not linking with Emacs proper,
128 we can't use the relocating allocator
129 even if config.h says that we can. */
132 # if defined STDC_HEADERS || defined _LIBC
139 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
140 If nothing else has been done, use the method below. */
141 # ifdef INHIBIT_STRING_HEADER
142 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
143 # if !defined bzero && !defined bcopy
144 # undef INHIBIT_STRING_HEADER
149 /* This is the normal way of making sure we have a bcopy and a bzero.
150 This is used in most programs--a few other programs avoid this
151 by defining INHIBIT_STRING_HEADER. */
152 # ifndef INHIBIT_STRING_HEADER
153 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
157 # define bzero(s, n) ((void) memset (s, '\0', n))
159 # define bzero(s, n) __bzero (s, n)
163 # include <strings.h>
165 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
168 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
173 /* Define the syntax stuff for \<, \>, etc. */
175 /* This must be nonzero for the wordchar and notwordchar pattern
176 commands in re_match_2. */
181 # ifdef SWITCH_ENUM_BUG
182 # define SWITCH_ENUM_CAST(x) ((int)(x))
184 # define SWITCH_ENUM_CAST(x) (x)
187 # endif /* not emacs */
189 # if defined _LIBC || HAVE_LIMITS_H
194 # define MB_LEN_MAX 1
197 /* Get the interface, including the syntax bits. */
198 # include "xregex.h" /* change for libiberty */
200 /* isalpha etc. are used for the character classes. */
203 /* Jim Meyering writes:
205 "... Some ctype macros are valid only for character codes that
206 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
207 using /bin/cc or gcc but without giving an ansi option). So, all
208 ctype uses should be through macros like ISPRINT... If
209 STDC_HEADERS is defined, then autoconf has verified that the ctype
210 macros don't need to be guarded with references to isascii. ...
211 Defining isascii to 1 should let any compiler worth its salt
212 eliminate the && through constant folding."
213 Solaris defines some of these symbols so we must undefine them first. */
216 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
217 # define ISASCII(c) 1
219 # define ISASCII(c) isascii(c)
223 # define ISBLANK(c) (ISASCII (c) && isblank (c))
225 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
228 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
230 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
234 # define ISPRINT(c) (ISASCII (c) && isprint (c))
235 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
236 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
237 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
238 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
239 # define ISLOWER(c) (ISASCII (c) && islower (c))
240 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
241 # define ISSPACE(c) (ISASCII (c) && isspace (c))
242 # define ISUPPER(c) (ISASCII (c) && isupper (c))
243 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
246 # define TOLOWER(c) _tolower(c)
248 # define TOLOWER(c) tolower(c)
252 # define NULL (void *)0
255 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
256 since ours (we hope) works properly with all combinations of
257 machines, compilers, `char' and `unsigned char' argument types.
258 (Per Bothner suggested the basic approach.) */
259 # undef SIGN_EXTEND_CHAR
261 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
262 # else /* not __STDC__ */
263 /* As in Harbison and Steele. */
264 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
268 /* How many characters in the character set. */
269 # define CHAR_SET_SIZE 256
273 extern char *re_syntax_table
;
275 # else /* not SYNTAX_TABLE */
277 static char re_syntax_table
[CHAR_SET_SIZE
];
279 static void init_syntax_once (void);
282 init_syntax_once (void)
289 bzero (re_syntax_table
, sizeof re_syntax_table
);
291 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
293 re_syntax_table
[c
] = Sword
;
295 re_syntax_table
['_'] = Sword
;
300 # endif /* not SYNTAX_TABLE */
302 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
306 /* Integer type for pointers. */
307 # if !defined _LIBC && !defined HAVE_UINTPTR_T
308 typedef unsigned long int uintptr_t;
311 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
312 use `alloca' instead of `malloc'. This is because using malloc in
313 re_search* or re_match* could cause memory leaks when C-g is used in
314 Emacs; also, malloc is slower and causes storage fragmentation. On
315 the other hand, malloc is more portable, and easier to debug.
317 Because we sometimes use alloca, some routines have to be macros,
318 not functions -- `alloca'-allocated space disappears at the end of the
319 function it is called in. */
323 # define REGEX_ALLOCATE malloc
324 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
325 # define REGEX_FREE free
327 # else /* not REGEX_MALLOC */
329 /* Emacs already defines alloca, sometimes. */
332 /* Make alloca work the best possible way. */
334 # define alloca __builtin_alloca
335 # else /* not __GNUC__ */
338 # endif /* HAVE_ALLOCA_H */
339 # endif /* not __GNUC__ */
341 # endif /* not alloca */
343 # define REGEX_ALLOCATE alloca
345 /* Assumes a `char *destination' variable. */
346 # define REGEX_REALLOCATE(source, osize, nsize) \
347 (destination = (char *) alloca (nsize), \
348 memcpy (destination, source, osize))
350 /* No need to do anything to free, after alloca. */
351 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
353 # endif /* not REGEX_MALLOC */
355 /* Define how to allocate the failure stack. */
357 # if defined REL_ALLOC && defined REGEX_MALLOC
359 # define REGEX_ALLOCATE_STACK(size) \
360 r_alloc (&failure_stack_ptr, (size))
361 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
362 r_re_alloc (&failure_stack_ptr, (nsize))
363 # define REGEX_FREE_STACK(ptr) \
364 r_alloc_free (&failure_stack_ptr)
366 # else /* not using relocating allocator */
370 # define REGEX_ALLOCATE_STACK malloc
371 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
372 # define REGEX_FREE_STACK free
374 # else /* not REGEX_MALLOC */
376 # define REGEX_ALLOCATE_STACK alloca
378 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
379 REGEX_REALLOCATE (source, osize, nsize)
380 /* No need to explicitly free anything. */
381 # define REGEX_FREE_STACK(arg)
383 # endif /* not REGEX_MALLOC */
384 # endif /* not using relocating allocator */
387 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
388 `string1' or just past its end. This works if PTR is NULL, which is
390 # define FIRST_STRING_P(ptr) \
391 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
393 /* (Re)Allocate N items of type T using malloc, or fail. */
394 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
395 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
396 # define RETALLOC_IF(addr, n, t) \
397 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
398 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
400 # define BYTEWIDTH 8 /* In bits. */
402 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
406 # define MAX(a, b) ((a) > (b) ? (a) : (b))
407 # define MIN(a, b) ((a) < (b) ? (a) : (b))
409 typedef char boolean
;
413 static reg_errcode_t
byte_regex_compile (const char *pattern
, size_t size
,
415 struct re_pattern_buffer
*bufp
);
417 static int byte_re_match_2_internal (struct re_pattern_buffer
*bufp
,
418 const char *string1
, int size1
,
419 const char *string2
, int size2
,
421 struct re_registers
*regs
,
423 static int byte_re_search_2 (struct re_pattern_buffer
*bufp
,
424 const char *string1
, int size1
,
425 const char *string2
, int size2
,
426 int startpos
, int range
,
427 struct re_registers
*regs
, int stop
);
428 static int byte_re_compile_fastmap (struct re_pattern_buffer
*bufp
);
431 static reg_errcode_t
wcs_regex_compile (const char *pattern
, size_t size
,
433 struct re_pattern_buffer
*bufp
);
436 static int wcs_re_match_2_internal (struct re_pattern_buffer
*bufp
,
437 const char *cstring1
, int csize1
,
438 const char *cstring2
, int csize2
,
440 struct re_registers
*regs
,
442 wchar_t *string1
, int size1
,
443 wchar_t *string2
, int size2
,
444 int *mbs_offset1
, int *mbs_offset2
);
445 static int wcs_re_search_2 (struct re_pattern_buffer
*bufp
,
446 const char *string1
, int size1
,
447 const char *string2
, int size2
,
448 int startpos
, int range
,
449 struct re_registers
*regs
, int stop
);
450 static int wcs_re_compile_fastmap (struct re_pattern_buffer
*bufp
);
453 /* These are the command codes that appear in compiled regular
454 expressions. Some opcodes are followed by argument bytes. A
455 command code can specify any interpretation whatsoever for its
456 arguments. Zero bytes may appear in the compiled regular expression. */
462 /* Succeed right away--no more backtracking. */
465 /* Followed by one byte giving n, then by n literal bytes. */
469 /* Same as exactn, but contains binary data. */
473 /* Matches any (more or less) character. */
476 /* Matches any one char belonging to specified set. First
477 following byte is number of bitmap bytes. Then come bytes
478 for a bitmap saying which chars are in. Bits in each byte
479 are ordered low-bit-first. A character is in the set if its
480 bit is 1. A character too large to have a bit in the map is
481 automatically not in the set. */
482 /* ifdef MBS_SUPPORT, following element is length of character
483 classes, length of collating symbols, length of equivalence
484 classes, length of character ranges, and length of characters.
485 Next, character class element, collating symbols elements,
486 equivalence class elements, range elements, and character
488 See regex_compile function. */
491 /* Same parameters as charset, but match any character that is
492 not one of those specified. */
495 /* Start remembering the text that is matched, for storing in a
496 register. Followed by one byte with the register number, in
497 the range 0 to one less than the pattern buffer's re_nsub
498 field. Then followed by one byte with the number of groups
499 inner to this one. (This last has to be part of the
500 start_memory only because we need it in the on_failure_jump
504 /* Stop remembering the text that is matched and store it in a
505 memory register. Followed by one byte with the register
506 number, in the range 0 to one less than `re_nsub' in the
507 pattern buffer, and one byte with the number of inner groups,
508 just like `start_memory'. (We need the number of inner
509 groups here because we don't have any easy way of finding the
510 corresponding start_memory when we're at a stop_memory.) */
513 /* Match a duplicate of something remembered. Followed by one
514 byte containing the register number. */
517 /* Fail unless at beginning of line. */
520 /* Fail unless at end of line. */
523 /* Succeeds if at beginning of buffer (if emacs) or at beginning
524 of string to be matched (if not). */
527 /* Analogously, for end of buffer/string. */
530 /* Followed by two byte relative address to which to jump. */
533 /* Same as jump, but marks the end of an alternative. */
536 /* Followed by two-byte relative address of place to resume at
537 in case of failure. */
538 /* ifdef MBS_SUPPORT, the size of address is 1. */
541 /* Like on_failure_jump, but pushes a placeholder instead of the
542 current string position when executed. */
543 on_failure_keep_string_jump
,
545 /* Throw away latest failure point and then jump to following
546 two-byte relative address. */
547 /* ifdef MBS_SUPPORT, the size of address is 1. */
550 /* Change to pop_failure_jump if know won't have to backtrack to
551 match; otherwise change to jump. This is used to jump
552 back to the beginning of a repeat. If what follows this jump
553 clearly won't match what the repeat does, such that we can be
554 sure that there is no use backtracking out of repetitions
555 already matched, then we change it to a pop_failure_jump.
556 Followed by two-byte address. */
557 /* ifdef MBS_SUPPORT, the size of address is 1. */
560 /* Jump to following two-byte address, and push a dummy failure
561 point. This failure point will be thrown away if an attempt
562 is made to use it for a failure. A `+' construct makes this
563 before the first repeat. Also used as an intermediary kind
564 of jump when compiling an alternative. */
565 /* ifdef MBS_SUPPORT, the size of address is 1. */
568 /* Push a dummy failure point and continue. Used at the end of
572 /* Followed by two-byte relative address and two-byte number n.
573 After matching N times, jump to the address upon failure. */
574 /* ifdef MBS_SUPPORT, the size of address is 1. */
577 /* Followed by two-byte relative address, and two-byte number n.
578 Jump to the address N times, then fail. */
579 /* ifdef MBS_SUPPORT, the size of address is 1. */
582 /* Set the following two-byte relative address to the
583 subsequent two-byte number. The address *includes* the two
585 /* ifdef MBS_SUPPORT, the size of address is 1. */
588 wordchar
, /* Matches any word-constituent character. */
589 notwordchar
, /* Matches any char that is not a word-constituent. */
591 wordbeg
, /* Succeeds if at word beginning. */
592 wordend
, /* Succeeds if at word end. */
594 wordbound
, /* Succeeds if at a word boundary. */
595 notwordbound
/* Succeeds if not at a word boundary. */
598 ,before_dot
, /* Succeeds if before point. */
599 at_dot
, /* Succeeds if at point. */
600 after_dot
, /* Succeeds if after point. */
602 /* Matches any character whose syntax is specified. Followed by
603 a byte which contains a syntax code, e.g., Sword. */
606 /* Matches any character whose syntax is not that specified. */
610 #endif /* not INSIDE_RECURSION */
615 # define UCHAR_T unsigned char
616 # define COMPILED_BUFFER_VAR bufp->buffer
617 # define OFFSET_ADDRESS_SIZE 2
618 # define PREFIX(name) byte_##name
619 # define ARG_PREFIX(name) name
620 # define PUT_CHAR(c) putchar (c)
623 # define CHAR_T wchar_t
624 # define UCHAR_T wchar_t
625 # define COMPILED_BUFFER_VAR wc_buffer
626 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
627 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
628 # define PREFIX(name) wcs_##name
629 # define ARG_PREFIX(name) c##name
630 /* Should we use wide stream?? */
631 # define PUT_CHAR(c) printf ("%C", c);
637 # define INSIDE_RECURSION
639 # undef INSIDE_RECURSION
642 # define INSIDE_RECURSION
644 # undef INSIDE_RECURSION
648 #ifdef INSIDE_RECURSION
649 /* Common operations on the compiled pattern. */
651 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
652 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
655 # define STORE_NUMBER(destination, number) \
657 *(destination) = (UCHAR_T)(number); \
660 # define STORE_NUMBER(destination, number) \
662 (destination)[0] = (number) & 0377; \
663 (destination)[1] = (number) >> 8; \
667 /* Same as STORE_NUMBER, except increment DESTINATION to
668 the byte after where the number is stored. Therefore, DESTINATION
669 must be an lvalue. */
670 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
672 # define STORE_NUMBER_AND_INCR(destination, number) \
674 STORE_NUMBER (destination, number); \
675 (destination) += OFFSET_ADDRESS_SIZE; \
678 /* Put into DESTINATION a number stored in two contiguous bytes starting
680 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
683 # define EXTRACT_NUMBER(destination, source) \
685 (destination) = *(source); \
688 # define EXTRACT_NUMBER(destination, source) \
690 (destination) = *(source) & 0377; \
691 (destination) += ((unsigned) SIGN_EXTEND_CHAR (*((source) + 1))) << 8; \
696 static void PREFIX(extract_number
) (int *dest
, UCHAR_T
*source
);
698 PREFIX(extract_number
) (int *dest
, UCHAR_T
*source
)
703 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
704 *dest
= *source
& 0377;
709 # ifndef EXTRACT_MACROS /* To debug the macros. */
710 # undef EXTRACT_NUMBER
711 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
712 # endif /* not EXTRACT_MACROS */
716 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
717 SOURCE must be an lvalue. */
719 # define EXTRACT_NUMBER_AND_INCR(destination, source) \
721 EXTRACT_NUMBER (destination, source); \
722 (source) += OFFSET_ADDRESS_SIZE; \
726 static void PREFIX(extract_number_and_incr
) (int *destination
,
729 PREFIX(extract_number_and_incr
) (int *destination
, UCHAR_T
**source
)
731 PREFIX(extract_number
) (destination
, *source
);
732 *source
+= OFFSET_ADDRESS_SIZE
;
735 # ifndef EXTRACT_MACROS
736 # undef EXTRACT_NUMBER_AND_INCR
737 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
738 PREFIX(extract_number_and_incr) (&dest, &src)
739 # endif /* not EXTRACT_MACROS */
745 /* If DEBUG is defined, Regex prints many voluminous messages about what
746 it is doing (if the variable `debug' is nonzero). If linked with the
747 main program in `iregex.c', you can enter patterns and strings
748 interactively. And if linked with the main program in `main.c' and
749 the other test files, you can run the already-written tests. */
753 # ifndef DEFINED_ONCE
755 /* We use standard I/O for debugging. */
758 /* It is useful to test things that ``must'' be true when debugging. */
763 # define DEBUG_STATEMENT(e) e
764 # define DEBUG_PRINT1(x) if (debug) printf (x)
765 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
766 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
767 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
768 # endif /* not DEFINED_ONCE */
770 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
771 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
772 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
773 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
776 /* Print the fastmap in human-readable form. */
778 # ifndef DEFINED_ONCE
780 print_fastmap (char *fastmap
)
782 unsigned was_a_range
= 0;
785 while (i
< (1 << BYTEWIDTH
))
791 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
805 # endif /* not DEFINED_ONCE */
808 /* Print a compiled pattern string in human-readable form, starting at
809 the START pointer into it and ending just before the pointer END. */
812 PREFIX(print_partial_compiled_pattern
) (UCHAR_T
*start
, UCHAR_T
*end
)
825 /* Loop over pattern commands. */
829 printf ("%td:\t", p
- start
);
831 printf ("%ld:\t", (long int) (p
- start
));
834 switch ((re_opcode_t
) *p
++)
842 printf ("/exactn/%d", mcnt
);
854 printf ("/exactn_bin/%d", mcnt
);
857 printf("/%lx", (long int) *p
++);
861 # endif /* MBS_SUPPORT */
865 printf ("/start_memory/%d/%ld", mcnt
, (long int) *p
++);
870 printf ("/stop_memory/%d/%ld", mcnt
, (long int) *p
++);
874 printf ("/duplicate/%ld", (long int) *p
++);
887 printf ("/charset [%s",
888 (re_opcode_t
) *(workp
- 1) == charset_not
? "^" : "");
890 length
= *workp
++; /* the length of char_classes */
891 for (i
=0 ; i
<length
; i
++)
892 printf("[:%lx:]", (long int) *p
++);
893 length
= *workp
++; /* the length of collating_symbol */
894 for (i
=0 ; i
<length
;)
898 PUT_CHAR((i
++,*p
++));
902 length
= *workp
++; /* the length of equivalence_class */
903 for (i
=0 ; i
<length
;)
907 PUT_CHAR((i
++,*p
++));
911 length
= *workp
++; /* the length of char_range */
912 for (i
=0 ; i
<length
; i
++)
914 wchar_t range_start
= *p
++;
915 wchar_t range_end
= *p
++;
916 printf("%C-%C", range_start
, range_end
);
918 length
= *workp
++; /* the length of char */
919 for (i
=0 ; i
<length
; i
++)
923 register int c
, last
= -100;
924 register int in_range
= 0;
926 printf ("/charset [%s",
927 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
929 assert (p
+ *p
< pend
);
931 for (c
= 0; c
< 256; c
++)
933 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
935 /* Are we starting a range? */
936 if (last
+ 1 == c
&& ! in_range
)
941 /* Have we broken a range? */
942 else if (last
+ 1 != c
&& in_range
)
972 case on_failure_jump
:
973 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
975 printf ("/on_failure_jump to %td", p
+ mcnt
- start
);
977 printf ("/on_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
981 case on_failure_keep_string_jump
:
982 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
984 printf ("/on_failure_keep_string_jump to %td", p
+ mcnt
- start
);
986 printf ("/on_failure_keep_string_jump to %ld",
987 (long int) (p
+ mcnt
- start
));
991 case dummy_failure_jump
:
992 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
994 printf ("/dummy_failure_jump to %td", p
+ mcnt
- start
);
996 printf ("/dummy_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
1000 case push_dummy_failure
:
1001 printf ("/push_dummy_failure");
1004 case maybe_pop_jump
:
1005 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1007 printf ("/maybe_pop_jump to %td", p
+ mcnt
- start
);
1009 printf ("/maybe_pop_jump to %ld", (long int) (p
+ mcnt
- start
));
1013 case pop_failure_jump
:
1014 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1016 printf ("/pop_failure_jump to %td", p
+ mcnt
- start
);
1018 printf ("/pop_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
1023 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1025 printf ("/jump_past_alt to %td", p
+ mcnt
- start
);
1027 printf ("/jump_past_alt to %ld", (long int) (p
+ mcnt
- start
));
1032 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1034 printf ("/jump to %td", p
+ mcnt
- start
);
1036 printf ("/jump to %ld", (long int) (p
+ mcnt
- start
));
1041 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1043 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1045 printf ("/succeed_n to %td, %d times", p1
- start
, mcnt2
);
1047 printf ("/succeed_n to %ld, %d times",
1048 (long int) (p1
- start
), mcnt2
);
1053 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1055 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1056 printf ("/jump_n to %d, %d times", p1
- start
, mcnt2
);
1060 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1062 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1064 printf ("/set_number_at location %td to %d", p1
- start
, mcnt2
);
1066 printf ("/set_number_at location %ld to %d",
1067 (long int) (p1
- start
), mcnt2
);
1072 printf ("/wordbound");
1076 printf ("/notwordbound");
1080 printf ("/wordbeg");
1084 printf ("/wordend");
1089 printf ("/before_dot");
1097 printf ("/after_dot");
1101 printf ("/syntaxspec");
1103 printf ("/%d", mcnt
);
1107 printf ("/notsyntaxspec");
1109 printf ("/%d", mcnt
);
1114 printf ("/wordchar");
1118 printf ("/notwordchar");
1130 printf ("?%ld", (long int) *(p
-1));
1137 printf ("%td:\tend of pattern.\n", p
- start
);
1139 printf ("%ld:\tend of pattern.\n", (long int) (p
- start
));
1145 PREFIX(print_compiled_pattern
) (struct re_pattern_buffer
*bufp
)
1147 UCHAR_T
*buffer
= (UCHAR_T
*) bufp
->buffer
;
1149 PREFIX(print_partial_compiled_pattern
) (buffer
, buffer
1150 + bufp
->used
/ sizeof(UCHAR_T
));
1151 printf ("%ld bytes used/%ld bytes allocated.\n",
1152 bufp
->used
, bufp
->allocated
);
1154 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1156 printf ("fastmap: ");
1157 print_fastmap (bufp
->fastmap
);
1161 printf ("re_nsub: %Zd\t", bufp
->re_nsub
);
1163 printf ("re_nsub: %ld\t", (long int) bufp
->re_nsub
);
1165 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1166 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1167 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
1168 printf ("no_sub: %d\t", bufp
->no_sub
);
1169 printf ("not_bol: %d\t", bufp
->not_bol
);
1170 printf ("not_eol: %d\t", bufp
->not_eol
);
1171 printf ("syntax: %lx\n", bufp
->syntax
);
1172 /* Perhaps we should print the translate table? */
1177 PREFIX(print_double_string
) (const CHAR_T
*where
, const CHAR_T
*string1
,
1178 int size1
, const CHAR_T
*string2
, int size2
)
1188 if (FIRST_STRING_P (where
))
1190 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1191 PUT_CHAR (string1
[this_char
]);
1197 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1199 PUT_CHAR (string2
[this_char
]);
1202 fputs ("...", stdout
);
1209 # ifndef DEFINED_ONCE
1217 # else /* not DEBUG */
1219 # ifndef DEFINED_ONCE
1223 # define DEBUG_STATEMENT(e)
1224 # define DEBUG_PRINT1(x)
1225 # define DEBUG_PRINT2(x1, x2)
1226 # define DEBUG_PRINT3(x1, x2, x3)
1227 # define DEBUG_PRINT4(x1, x2, x3, x4)
1228 # endif /* not DEFINED_ONCE */
1229 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1230 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1232 # endif /* not DEBUG */
1237 /* This convert a multibyte string to a wide character string.
1238 And write their correspondances to offset_buffer(see below)
1239 and write whether each wchar_t is binary data to is_binary.
1240 This assume invalid multibyte sequences as binary data.
1241 We assume offset_buffer and is_binary is already allocated
1244 static size_t convert_mbs_to_wcs (CHAR_T
*dest
, const unsigned char* src
,
1245 size_t len
, int *offset_buffer
,
1248 convert_mbs_to_wcs (CHAR_T
*dest
, const unsigned char*src
, size_t len
,
1249 int *offset_buffer
, char *is_binary
)
1250 /* It hold correspondances between src(char string) and
1251 dest(wchar_t string) for optimization.
1253 dest = {'X', 'Y', 'Z'}
1254 (each "xxx", "y" and "zz" represent one multibyte character
1255 corresponding to 'X', 'Y' and 'Z'.)
1256 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1260 wchar_t *pdest
= dest
;
1261 const unsigned char *psrc
= src
;
1262 size_t wc_count
= 0;
1266 size_t mb_remain
= len
;
1267 size_t mb_count
= 0;
1269 /* Initialize the conversion state. */
1270 memset (&mbs
, 0, sizeof (mbstate_t));
1272 offset_buffer
[0] = 0;
1273 for( ; mb_remain
> 0 ; ++wc_count
, ++pdest
, mb_remain
-= consumed
,
1277 consumed
= __mbrtowc (pdest
, psrc
, mb_remain
, &mbs
);
1279 consumed
= mbrtowc (pdest
, psrc
, mb_remain
, &mbs
);
1283 /* failed to convert. maybe src contains binary data.
1284 So we consume 1 byte manualy. */
1288 is_binary
[wc_count
] = TRUE
;
1291 is_binary
[wc_count
] = FALSE
;
1292 /* In sjis encoding, we use yen sign as escape character in
1293 place of reverse solidus. So we convert 0x5c(yen sign in
1294 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1295 solidus in UCS2). */
1296 if (consumed
== 1 && (int) *psrc
== 0x5c && (int) *pdest
== 0xa5)
1297 *pdest
= (wchar_t) *psrc
;
1299 offset_buffer
[wc_count
+ 1] = mb_count
+= consumed
;
1302 /* Fill remain of the buffer with sentinel. */
1303 for (i
= wc_count
+ 1 ; i
<= len
; i
++)
1304 offset_buffer
[i
] = mb_count
+ 1;
1311 #else /* not INSIDE_RECURSION */
1313 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1314 also be assigned to arbitrarily: each pattern buffer stores its own
1315 syntax, so it can be changed between regex compilations. */
1316 /* This has no initializer because initialized variables in Emacs
1317 become read-only after dumping. */
1318 reg_syntax_t re_syntax_options
;
1321 /* Specify the precise syntax of regexps for compilation. This provides
1322 for compatibility for various utilities which historically have
1323 different, incompatible syntaxes.
1325 The argument SYNTAX is a bit mask comprised of the various bits
1326 defined in regex.h. We return the old syntax. */
1329 re_set_syntax (reg_syntax_t syntax
)
1331 reg_syntax_t ret
= re_syntax_options
;
1333 re_syntax_options
= syntax
;
1335 if (syntax
& RE_DEBUG
)
1337 else if (debug
) /* was on but now is not */
1343 weak_alias (__re_set_syntax
, re_set_syntax
)
1346 /* This table gives an error message for each of the error codes listed
1347 in regex.h. Obviously the order here has to be same as there.
1348 POSIX doesn't require that we do anything for REG_NOERROR,
1349 but why not be nice? */
1351 static const char *re_error_msgid
[] =
1353 gettext_noop ("Success"), /* REG_NOERROR */
1354 gettext_noop ("No match"), /* REG_NOMATCH */
1355 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1356 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1357 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1358 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1359 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1360 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1361 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1362 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1363 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1364 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1365 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1366 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1367 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1368 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1369 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1372 #endif /* INSIDE_RECURSION */
1374 #ifndef DEFINED_ONCE
1375 /* Avoiding alloca during matching, to placate r_alloc. */
1377 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1378 searching and matching functions should not call alloca. On some
1379 systems, alloca is implemented in terms of malloc, and if we're
1380 using the relocating allocator routines, then malloc could cause a
1381 relocation, which might (if the strings being searched are in the
1382 ralloc heap) shift the data out from underneath the regexp
1385 Here's another reason to avoid allocation: Emacs
1386 processes input from X in a signal handler; processing X input may
1387 call malloc; if input arrives while a matching routine is calling
1388 malloc, then we're scrod. But Emacs can't just block input while
1389 calling matching routines; then we don't notice interrupts when
1390 they come in. So, Emacs blocks input around all regexp calls
1391 except the matching calls, which it leaves unprotected, in the
1392 faith that they will not malloc. */
1394 /* Normally, this is fine. */
1395 # define MATCH_MAY_ALLOCATE
1397 /* When using GNU C, we are not REALLY using the C alloca, no matter
1398 what config.h may say. So don't take precautions for it. */
1403 /* The match routines may not allocate if (1) they would do it with malloc
1404 and (2) it's not safe for them to use malloc.
1405 Note that if REL_ALLOC is defined, matching would not use malloc for the
1406 failure stack, but we would still use it for the register vectors;
1407 so REL_ALLOC should not affect this. */
1408 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1409 # undef MATCH_MAY_ALLOCATE
1411 #endif /* not DEFINED_ONCE */
1413 #ifdef INSIDE_RECURSION
1414 /* Failure stack declarations and macros; both re_compile_fastmap and
1415 re_match_2 use a failure stack. These have to be macros because of
1416 REGEX_ALLOCATE_STACK. */
1419 /* Number of failure points for which to initially allocate space
1420 when matching. If this number is exceeded, we allocate more
1421 space, so it is not a hard limit. */
1422 # ifndef INIT_FAILURE_ALLOC
1423 # define INIT_FAILURE_ALLOC 5
1426 /* Roughly the maximum number of failure points on the stack. Would be
1427 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1428 This is a variable only so users of regex can assign to it; we never
1429 change it ourselves. */
1431 # ifdef INT_IS_16BIT
1433 # ifndef DEFINED_ONCE
1434 # if defined MATCH_MAY_ALLOCATE
1435 /* 4400 was enough to cause a crash on Alpha OSF/1,
1436 whose default stack limit is 2mb. */
1437 long int re_max_failures
= 4000;
1439 long int re_max_failures
= 2000;
1443 union PREFIX(fail_stack_elt
)
1449 typedef union PREFIX(fail_stack_elt
) PREFIX(fail_stack_elt_t
);
1453 PREFIX(fail_stack_elt_t
) *stack
;
1454 unsigned long int size
;
1455 unsigned long int avail
; /* Offset of next open position. */
1456 } PREFIX(fail_stack_type
);
1458 # else /* not INT_IS_16BIT */
1460 # ifndef DEFINED_ONCE
1461 # if defined MATCH_MAY_ALLOCATE
1462 /* 4400 was enough to cause a crash on Alpha OSF/1,
1463 whose default stack limit is 2mb. */
1464 int re_max_failures
= 4000;
1466 int re_max_failures
= 2000;
1470 union PREFIX(fail_stack_elt
)
1476 typedef union PREFIX(fail_stack_elt
) PREFIX(fail_stack_elt_t
);
1480 PREFIX(fail_stack_elt_t
) *stack
;
1482 unsigned avail
; /* Offset of next open position. */
1483 } PREFIX(fail_stack_type
);
1485 # endif /* INT_IS_16BIT */
1487 # ifndef DEFINED_ONCE
1488 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1489 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1490 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1494 /* Define macros to initialize and free the failure stack.
1495 Do `return -2' if the alloc fails. */
1497 # ifdef MATCH_MAY_ALLOCATE
1498 # define INIT_FAIL_STACK() \
1500 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1501 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1503 if (fail_stack.stack == NULL) \
1506 fail_stack.size = INIT_FAILURE_ALLOC; \
1507 fail_stack.avail = 0; \
1510 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1512 # define INIT_FAIL_STACK() \
1514 fail_stack.avail = 0; \
1517 # define RESET_FAIL_STACK()
1521 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1523 Return 1 if succeeds, and 0 if either ran out of memory
1524 allocating space for it or it was already too large.
1526 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1528 # define DOUBLE_FAIL_STACK(fail_stack) \
1529 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1531 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1532 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1533 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1534 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1536 (fail_stack).stack == NULL \
1538 : ((fail_stack).size <<= 1, \
1542 /* Push pointer POINTER on FAIL_STACK.
1543 Return 1 if was able to do so and 0 if ran out of memory allocating
1545 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1546 ((FAIL_STACK_FULL () \
1547 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1549 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1552 /* Push a pointer value onto the failure stack.
1553 Assumes the variable `fail_stack'. Probably should only
1554 be called from within `PUSH_FAILURE_POINT'. */
1555 # define PUSH_FAILURE_POINTER(item) \
1556 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1558 /* This pushes an integer-valued item onto the failure stack.
1559 Assumes the variable `fail_stack'. Probably should only
1560 be called from within `PUSH_FAILURE_POINT'. */
1561 # define PUSH_FAILURE_INT(item) \
1562 fail_stack.stack[fail_stack.avail++].integer = (item)
1564 /* Push a fail_stack_elt_t value onto the failure stack.
1565 Assumes the variable `fail_stack'. Probably should only
1566 be called from within `PUSH_FAILURE_POINT'. */
1567 # define PUSH_FAILURE_ELT(item) \
1568 fail_stack.stack[fail_stack.avail++] = (item)
1570 /* These three POP... operations complement the three PUSH... operations.
1571 All assume that `fail_stack' is nonempty. */
1572 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1573 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1574 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1576 /* Used to omit pushing failure point id's when we're not debugging. */
1578 # define DEBUG_PUSH PUSH_FAILURE_INT
1579 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1581 # define DEBUG_PUSH(item)
1582 # define DEBUG_POP(item_addr)
1586 /* Push the information about the state we will need
1587 if we ever fail back to it.
1589 Requires variables fail_stack, regstart, regend, reg_info, and
1590 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1593 Does `return FAILURE_CODE' if runs out of memory. */
1595 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1597 char *destination; \
1598 /* Must be int, so when we don't save any registers, the arithmetic \
1599 of 0 + -1 isn't done as unsigned. */ \
1600 /* Can't be int, since there is not a shred of a guarantee that int \
1601 is wide enough to hold a value of something to which pointer can \
1603 active_reg_t this_reg; \
1605 DEBUG_STATEMENT (failure_id++); \
1606 DEBUG_STATEMENT (nfailure_points_pushed++); \
1607 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1608 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1609 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1611 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1612 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1614 /* Ensure we have enough space allocated for what we will push. */ \
1615 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1617 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1618 return failure_code; \
1620 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1621 (fail_stack).size); \
1622 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1625 /* Push the info, starting with the registers. */ \
1626 DEBUG_PRINT1 ("\n"); \
1629 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1632 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1633 DEBUG_STATEMENT (num_regs_pushed++); \
1635 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1636 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1638 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1639 PUSH_FAILURE_POINTER (regend[this_reg]); \
1641 DEBUG_PRINT2 (" info: %p\n ", \
1642 reg_info[this_reg].word.pointer); \
1643 DEBUG_PRINT2 (" match_null=%d", \
1644 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1645 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1646 DEBUG_PRINT2 (" matched_something=%d", \
1647 MATCHED_SOMETHING (reg_info[this_reg])); \
1648 DEBUG_PRINT2 (" ever_matched=%d", \
1649 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1650 DEBUG_PRINT1 ("\n"); \
1651 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1654 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1655 PUSH_FAILURE_INT (lowest_active_reg); \
1657 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1658 PUSH_FAILURE_INT (highest_active_reg); \
1660 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1661 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1662 PUSH_FAILURE_POINTER (pattern_place); \
1664 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1665 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1667 DEBUG_PRINT1 ("'\n"); \
1668 PUSH_FAILURE_POINTER (string_place); \
1670 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1671 DEBUG_PUSH (failure_id); \
1674 # ifndef DEFINED_ONCE
1675 /* This is the number of items that are pushed and popped on the stack
1676 for each register. */
1677 # define NUM_REG_ITEMS 3
1679 /* Individual items aside from the registers. */
1681 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1683 # define NUM_NONREG_ITEMS 4
1686 /* We push at most this many items on the stack. */
1687 /* We used to use (num_regs - 1), which is the number of registers
1688 this regexp will save; but that was changed to 5
1689 to avoid stack overflow for a regexp with lots of parens. */
1690 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1692 /* We actually push this many items. */
1693 # define NUM_FAILURE_ITEMS \
1695 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1699 /* How many items can still be added to the stack without overflowing it. */
1700 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1701 # endif /* not DEFINED_ONCE */
1704 /* Pops what PUSH_FAIL_STACK pushes.
1706 We restore into the parameters, all of which should be lvalues:
1707 STR -- the saved data position.
1708 PAT -- the saved pattern position.
1709 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1710 REGSTART, REGEND -- arrays of string positions.
1711 REG_INFO -- array of information about each subexpression.
1713 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1714 `pend', `string1', `size1', `string2', and `size2'. */
1715 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1717 DEBUG_STATEMENT (unsigned failure_id;) \
1718 active_reg_t this_reg; \
1719 const UCHAR_T *string_temp; \
1721 assert (!FAIL_STACK_EMPTY ()); \
1723 /* Remove failure points and point to how many regs pushed. */ \
1724 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1725 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1726 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1728 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1730 DEBUG_POP (&failure_id); \
1731 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1733 /* If the saved string location is NULL, it came from an \
1734 on_failure_keep_string_jump opcode, and we want to throw away the \
1735 saved NULL, thus retaining our current position in the string. */ \
1736 string_temp = POP_FAILURE_POINTER (); \
1737 if (string_temp != NULL) \
1738 str = (const CHAR_T *) string_temp; \
1740 DEBUG_PRINT2 (" Popping string %p: `", str); \
1741 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1742 DEBUG_PRINT1 ("'\n"); \
1744 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1745 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1746 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1748 /* Restore register info. */ \
1749 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1750 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1752 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1753 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1756 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1758 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1760 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1761 DEBUG_PRINT2 (" info: %p\n", \
1762 reg_info[this_reg].word.pointer); \
1764 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1765 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1767 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1768 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1772 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1774 reg_info[this_reg].word.integer = 0; \
1775 regend[this_reg] = 0; \
1776 regstart[this_reg] = 0; \
1778 highest_active_reg = high_reg; \
1781 set_regs_matched_done = 0; \
1782 DEBUG_STATEMENT (nfailure_points_popped++); \
1783 } /* POP_FAILURE_POINT */
1785 /* Structure for per-register (a.k.a. per-group) information.
1786 Other register information, such as the
1787 starting and ending positions (which are addresses), and the list of
1788 inner groups (which is a bits list) are maintained in separate
1791 We are making a (strictly speaking) nonportable assumption here: that
1792 the compiler will pack our bit fields into something that fits into
1793 the type of `word', i.e., is something that fits into one item on the
1797 /* Declarations and macros for re_match_2. */
1801 PREFIX(fail_stack_elt_t
) word
;
1804 /* This field is one if this group can match the empty string,
1805 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1806 # define MATCH_NULL_UNSET_VALUE 3
1807 unsigned match_null_string_p
: 2;
1808 unsigned is_active
: 1;
1809 unsigned matched_something
: 1;
1810 unsigned ever_matched_something
: 1;
1812 } PREFIX(register_info_type
);
1814 # ifndef DEFINED_ONCE
1815 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1816 # define IS_ACTIVE(R) ((R).bits.is_active)
1817 # define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1818 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1821 /* Call this when have matched a real character; it sets `matched' flags
1822 for the subexpressions which we are currently inside. Also records
1823 that those subexprs have matched. */
1824 # define SET_REGS_MATCHED() \
1827 if (!set_regs_matched_done) \
1830 set_regs_matched_done = 1; \
1831 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1833 MATCHED_SOMETHING (reg_info[r]) \
1834 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1840 # endif /* not DEFINED_ONCE */
1842 /* Registers are set to a sentinel when they haven't yet matched. */
1843 static CHAR_T
PREFIX(reg_unset_dummy
);
1844 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1845 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1847 /* Subroutine declarations and macros for regex_compile. */
1848 static void PREFIX(store_op1
) (re_opcode_t op
, UCHAR_T
*loc
, int arg
);
1849 static void PREFIX(store_op2
) (re_opcode_t op
, UCHAR_T
*loc
,
1850 int arg1
, int arg2
);
1851 static void PREFIX(insert_op1
) (re_opcode_t op
, UCHAR_T
*loc
,
1852 int arg
, UCHAR_T
*end
);
1853 static void PREFIX(insert_op2
) (re_opcode_t op
, UCHAR_T
*loc
,
1854 int arg1
, int arg2
, UCHAR_T
*end
);
1855 static boolean
PREFIX(at_begline_loc_p
) (const CHAR_T
*pattern
,
1857 reg_syntax_t syntax
);
1858 static boolean
PREFIX(at_endline_loc_p
) (const CHAR_T
*p
,
1860 reg_syntax_t syntax
);
1862 static reg_errcode_t
wcs_compile_range (CHAR_T range_start
,
1863 const CHAR_T
**p_ptr
,
1866 reg_syntax_t syntax
,
1869 static void insert_space (int num
, CHAR_T
*loc
, CHAR_T
*end
);
1871 static reg_errcode_t
byte_compile_range (unsigned int range_start
,
1875 reg_syntax_t syntax
,
1879 /* Fetch the next character in the uncompiled pattern---translating it
1880 if necessary. Also cast from a signed character in the constant
1881 string passed to us by the user to an unsigned char that we can use
1882 as an array index (in, e.g., `translate'). */
1883 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1884 because it is impossible to allocate 4GB array for some encodings
1885 which have 4 byte character_set like UCS4. */
1888 # define PATFETCH(c) \
1889 do {if (p == pend) return REG_EEND; \
1890 c = (UCHAR_T) *p++; \
1891 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1894 # define PATFETCH(c) \
1895 do {if (p == pend) return REG_EEND; \
1896 c = (unsigned char) *p++; \
1897 if (translate) c = (unsigned char) translate[c]; \
1902 /* Fetch the next character in the uncompiled pattern, with no
1904 # define PATFETCH_RAW(c) \
1905 do {if (p == pend) return REG_EEND; \
1906 c = (UCHAR_T) *p++; \
1909 /* Go backwards one character in the pattern. */
1910 # define PATUNFETCH p--
1913 /* If `translate' is non-null, return translate[D], else just D. We
1914 cast the subscript to translate because some data is declared as
1915 `char *', to avoid warnings when a string constant is passed. But
1916 when we use a character as a subscript we must make it unsigned. */
1917 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1918 because it is impossible to allocate 4GB array for some encodings
1919 which have 4 byte character_set like UCS4. */
1923 # define TRANSLATE(d) \
1924 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1925 ? (char) translate[(unsigned char) (d)] : (d))
1927 # define TRANSLATE(d) \
1928 (translate ? (char) translate[(unsigned char) (d)] : (char) (d))
1933 /* Macros for outputting the compiled pattern into `buffer'. */
1935 /* If the buffer isn't allocated when it comes in, use this. */
1936 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
1938 /* Make sure we have at least N more bytes of space in buffer. */
1940 # define GET_BUFFER_SPACE(n) \
1941 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
1942 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
1945 # define GET_BUFFER_SPACE(n) \
1946 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1950 /* Make sure we have one more byte of buffer space and then add C to it. */
1951 # define BUF_PUSH(c) \
1953 GET_BUFFER_SPACE (1); \
1954 *b++ = (UCHAR_T) (c); \
1958 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1959 # define BUF_PUSH_2(c1, c2) \
1961 GET_BUFFER_SPACE (2); \
1962 *b++ = (UCHAR_T) (c1); \
1963 *b++ = (UCHAR_T) (c2); \
1967 /* As with BUF_PUSH_2, except for three bytes. */
1968 # define BUF_PUSH_3(c1, c2, c3) \
1970 GET_BUFFER_SPACE (3); \
1971 *b++ = (UCHAR_T) (c1); \
1972 *b++ = (UCHAR_T) (c2); \
1973 *b++ = (UCHAR_T) (c3); \
1976 /* Store a jump with opcode OP at LOC to location TO. We store a
1977 relative address offset by the three bytes the jump itself occupies. */
1978 # define STORE_JUMP(op, loc, to) \
1979 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
1981 /* Likewise, for a two-argument jump. */
1982 # define STORE_JUMP2(op, loc, to, arg) \
1983 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
1985 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1986 # define INSERT_JUMP(op, loc, to) \
1987 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
1989 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1990 # define INSERT_JUMP2(op, loc, to, arg) \
1991 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
1994 /* This is not an arbitrary limit: the arguments which represent offsets
1995 into the pattern are two bytes long. So if 2^16 bytes turns out to
1996 be too small, many things would have to change. */
1997 /* Any other compiler which, like MSC, has allocation limit below 2^16
1998 bytes will have to use approach similar to what was done below for
1999 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2000 reallocating to 0 bytes. Such thing is not going to work too well.
2001 You have been warned!! */
2002 # ifndef DEFINED_ONCE
2003 # if defined _MSC_VER && !defined WIN32
2004 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2005 The REALLOC define eliminates a flurry of conversion warnings,
2006 but is not required. */
2007 # define MAX_BUF_SIZE 65500L
2008 # define REALLOC(p,s) realloc ((p), (size_t) (s))
2010 # define MAX_BUF_SIZE (1L << 16)
2011 # define REALLOC(p,s) realloc ((p), (s))
2014 /* Extend the buffer by twice its current size via realloc and
2015 reset the pointers that pointed into the old block to point to the
2016 correct places in the new one. If extending the buffer results in it
2017 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2018 # if __BOUNDED_POINTERS__
2019 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2020 # define MOVE_BUFFER_POINTER(P) \
2021 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2022 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2025 SET_HIGH_BOUND (b); \
2026 SET_HIGH_BOUND (begalt); \
2027 if (fixup_alt_jump) \
2028 SET_HIGH_BOUND (fixup_alt_jump); \
2030 SET_HIGH_BOUND (laststart); \
2031 if (pending_exact) \
2032 SET_HIGH_BOUND (pending_exact); \
2035 # define MOVE_BUFFER_POINTER(P) (P) += incr
2036 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
2038 # endif /* not DEFINED_ONCE */
2041 # define EXTEND_BUFFER() \
2043 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2045 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2047 bufp->allocated <<= 1; \
2048 if (bufp->allocated > MAX_BUF_SIZE) \
2049 bufp->allocated = MAX_BUF_SIZE; \
2050 /* How many characters the new buffer can have? */ \
2051 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2052 if (wchar_count == 0) wchar_count = 1; \
2053 /* Truncate the buffer to CHAR_T align. */ \
2054 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2055 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2056 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2057 if (COMPILED_BUFFER_VAR == NULL) \
2058 return REG_ESPACE; \
2059 /* If the buffer moved, move all the pointers into it. */ \
2060 if (old_buffer != COMPILED_BUFFER_VAR) \
2062 PTR_INT_TYPE incr = COMPILED_BUFFER_VAR - old_buffer; \
2063 MOVE_BUFFER_POINTER (b); \
2064 MOVE_BUFFER_POINTER (begalt); \
2065 if (fixup_alt_jump) \
2066 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2068 MOVE_BUFFER_POINTER (laststart); \
2069 if (pending_exact) \
2070 MOVE_BUFFER_POINTER (pending_exact); \
2072 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2075 # define EXTEND_BUFFER() \
2077 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2078 if (bufp->allocated == MAX_BUF_SIZE) \
2080 bufp->allocated <<= 1; \
2081 if (bufp->allocated > MAX_BUF_SIZE) \
2082 bufp->allocated = MAX_BUF_SIZE; \
2083 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2085 if (COMPILED_BUFFER_VAR == NULL) \
2086 return REG_ESPACE; \
2087 /* If the buffer moved, move all the pointers into it. */ \
2088 if (old_buffer != COMPILED_BUFFER_VAR) \
2090 PTR_INT_TYPE incr = COMPILED_BUFFER_VAR - old_buffer; \
2091 MOVE_BUFFER_POINTER (b); \
2092 MOVE_BUFFER_POINTER (begalt); \
2093 if (fixup_alt_jump) \
2094 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2096 MOVE_BUFFER_POINTER (laststart); \
2097 if (pending_exact) \
2098 MOVE_BUFFER_POINTER (pending_exact); \
2100 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2104 # ifndef DEFINED_ONCE
2105 /* Since we have one byte reserved for the register number argument to
2106 {start,stop}_memory, the maximum number of groups we can report
2107 things about is what fits in that byte. */
2108 # define MAX_REGNUM 255
2110 /* But patterns can have more than `MAX_REGNUM' registers. We just
2111 ignore the excess. */
2112 typedef unsigned regnum_t
;
2115 /* Macros for the compile stack. */
2117 /* Since offsets can go either forwards or backwards, this type needs to
2118 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2119 /* int may be not enough when sizeof(int) == 2. */
2120 typedef long pattern_offset_t
;
2124 pattern_offset_t begalt_offset
;
2125 pattern_offset_t fixup_alt_jump
;
2126 pattern_offset_t inner_group_offset
;
2127 pattern_offset_t laststart_offset
;
2129 } compile_stack_elt_t
;
2134 compile_stack_elt_t
*stack
;
2136 unsigned avail
; /* Offset of next open position. */
2137 } compile_stack_type
;
2140 # define INIT_COMPILE_STACK_SIZE 32
2142 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2143 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2145 /* The next available element. */
2146 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2148 # endif /* not DEFINED_ONCE */
2150 /* Set the bit for character C in a list. */
2151 # ifndef DEFINED_ONCE
2152 # define SET_LIST_BIT(c) \
2153 (b[((unsigned char) (c)) / BYTEWIDTH] \
2154 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2155 # endif /* DEFINED_ONCE */
2157 /* Get the next unsigned number in the uncompiled pattern. */
2158 # define GET_UNSIGNED_NUMBER(num) \
2163 if (c < '0' || c > '9') \
2165 if (num <= RE_DUP_MAX) \
2169 num = num * 10 + c - '0'; \
2174 # ifndef DEFINED_ONCE
2175 # if defined _LIBC || WIDE_CHAR_SUPPORT
2176 /* The GNU C library provides support for user-defined character classes
2177 and the functions from ISO C amendement 1. */
2178 # ifdef CHARCLASS_NAME_MAX
2179 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2181 /* This shouldn't happen but some implementation might still have this
2182 problem. Use a reasonable default value. */
2183 # define CHAR_CLASS_MAX_LENGTH 256
2187 # define IS_CHAR_CLASS(string) __wctype (string)
2189 # define IS_CHAR_CLASS(string) wctype (string)
2192 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2194 # define IS_CHAR_CLASS(string) \
2195 (STREQ (string, "alpha") || STREQ (string, "upper") \
2196 || STREQ (string, "lower") || STREQ (string, "digit") \
2197 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2198 || STREQ (string, "space") || STREQ (string, "print") \
2199 || STREQ (string, "punct") || STREQ (string, "graph") \
2200 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2202 # endif /* DEFINED_ONCE */
2204 # ifndef MATCH_MAY_ALLOCATE
2206 /* If we cannot allocate large objects within re_match_2_internal,
2207 we make the fail stack and register vectors global.
2208 The fail stack, we grow to the maximum size when a regexp
2210 The register vectors, we adjust in size each time we
2211 compile a regexp, according to the number of registers it needs. */
2213 static PREFIX(fail_stack_type
) fail_stack
;
2215 /* Size with which the following vectors are currently allocated.
2216 That is so we can make them bigger as needed,
2217 but never make them smaller. */
2218 # ifdef DEFINED_ONCE
2219 static int regs_allocated_size
;
2221 static const char ** regstart
, ** regend
;
2222 static const char ** old_regstart
, ** old_regend
;
2223 static const char **best_regstart
, **best_regend
;
2224 static const char **reg_dummy
;
2225 # endif /* DEFINED_ONCE */
2227 static PREFIX(register_info_type
) *PREFIX(reg_info
);
2228 static PREFIX(register_info_type
) *PREFIX(reg_info_dummy
);
2230 /* Make the register vectors big enough for NUM_REGS registers,
2231 but don't make them smaller. */
2234 PREFIX(regex_grow_registers
) (int num_regs
)
2236 if (num_regs
> regs_allocated_size
)
2238 RETALLOC_IF (regstart
, num_regs
, const char *);
2239 RETALLOC_IF (regend
, num_regs
, const char *);
2240 RETALLOC_IF (old_regstart
, num_regs
, const char *);
2241 RETALLOC_IF (old_regend
, num_regs
, const char *);
2242 RETALLOC_IF (best_regstart
, num_regs
, const char *);
2243 RETALLOC_IF (best_regend
, num_regs
, const char *);
2244 RETALLOC_IF (PREFIX(reg_info
), num_regs
, PREFIX(register_info_type
));
2245 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
2246 RETALLOC_IF (PREFIX(reg_info_dummy
), num_regs
, PREFIX(register_info_type
));
2248 regs_allocated_size
= num_regs
;
2252 # endif /* not MATCH_MAY_ALLOCATE */
2254 # ifndef DEFINED_ONCE
2255 static boolean
group_in_compile_stack (compile_stack_type compile_stack
,
2257 # endif /* not DEFINED_ONCE */
2259 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2260 Returns one of error codes defined in `regex.h', or zero for success.
2262 Assumes the `allocated' (and perhaps `buffer') and `translate'
2263 fields are set in BUFP on entry.
2265 If it succeeds, results are put in BUFP (if it returns an error, the
2266 contents of BUFP are undefined):
2267 `buffer' is the compiled pattern;
2268 `syntax' is set to SYNTAX;
2269 `used' is set to the length of the compiled pattern;
2270 `fastmap_accurate' is zero;
2271 `re_nsub' is the number of subexpressions in PATTERN;
2272 `not_bol' and `not_eol' are zero;
2274 The `fastmap' and `newline_anchor' fields are neither
2275 examined nor set. */
2277 /* Return, freeing storage we allocated. */
2279 # define FREE_STACK_RETURN(value) \
2280 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2282 # define FREE_STACK_RETURN(value) \
2283 return (free (compile_stack.stack), value)
2286 static reg_errcode_t
2287 PREFIX(regex_compile
) (const char *ARG_PREFIX(pattern
),
2288 size_t ARG_PREFIX(size
), reg_syntax_t syntax
,
2289 struct re_pattern_buffer
*bufp
)
2291 /* We fetch characters from PATTERN here. Even though PATTERN is
2292 `char *' (i.e., signed), we declare these variables as unsigned, so
2293 they can be reliably used as array indices. */
2294 register UCHAR_T c
, c1
;
2297 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2298 CHAR_T
*pattern
, *COMPILED_BUFFER_VAR
;
2300 /* offset buffer for optimization. See convert_mbs_to_wc. */
2301 int *mbs_offset
= NULL
;
2302 /* It hold whether each wchar_t is binary data or not. */
2303 char *is_binary
= NULL
;
2304 /* A flag whether exactn is handling binary data or not. */
2305 char is_exactn_bin
= FALSE
;
2308 /* A random temporary spot in PATTERN. */
2311 /* Points to the end of the buffer, where we should append. */
2312 register UCHAR_T
*b
;
2314 /* Keeps track of unclosed groups. */
2315 compile_stack_type compile_stack
;
2317 /* Points to the current (ending) position in the pattern. */
2322 const CHAR_T
*p
= pattern
;
2323 const CHAR_T
*pend
= pattern
+ size
;
2326 /* How to translate the characters in the pattern. */
2327 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2329 /* Address of the count-byte of the most recently inserted `exactn'
2330 command. This makes it possible to tell if a new exact-match
2331 character can be added to that command or if the character requires
2332 a new `exactn' command. */
2333 UCHAR_T
*pending_exact
= 0;
2335 /* Address of start of the most recently finished expression.
2336 This tells, e.g., postfix * where to find the start of its
2337 operand. Reset at the beginning of groups and alternatives. */
2338 UCHAR_T
*laststart
= 0;
2340 /* Address of beginning of regexp, or inside of last group. */
2343 /* Address of the place where a forward jump should go to the end of
2344 the containing expression. Each alternative of an `or' -- except the
2345 last -- ends with a forward jump of this sort. */
2346 UCHAR_T
*fixup_alt_jump
= 0;
2348 /* Counts open-groups as they are encountered. Remembered for the
2349 matching close-group on the compile stack, so the same register
2350 number is put in the stop_memory as the start_memory. */
2351 regnum_t regnum
= 0;
2354 /* Initialize the wchar_t PATTERN and offset_buffer. */
2355 p
= pend
= pattern
= TALLOC(csize
+ 1, CHAR_T
);
2356 mbs_offset
= TALLOC(csize
+ 1, int);
2357 is_binary
= TALLOC(csize
+ 1, char);
2358 if (pattern
== NULL
|| mbs_offset
== NULL
|| is_binary
== NULL
)
2365 pattern
[csize
] = L
'\0'; /* sentinel */
2366 size
= convert_mbs_to_wcs(pattern
, cpattern
, csize
, mbs_offset
, is_binary
);
2378 DEBUG_PRINT1 ("\nCompiling pattern: ");
2381 unsigned debug_count
;
2383 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2384 PUT_CHAR (pattern
[debug_count
]);
2389 /* Initialize the compile stack. */
2390 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2391 if (compile_stack
.stack
== NULL
)
2401 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2402 compile_stack
.avail
= 0;
2404 /* Initialize the pattern buffer. */
2405 bufp
->syntax
= syntax
;
2406 bufp
->fastmap_accurate
= 0;
2407 bufp
->not_bol
= bufp
->not_eol
= 0;
2409 /* Set `used' to zero, so that if we return an error, the pattern
2410 printer (for debugging) will think there's no pattern. We reset it
2414 /* Always count groups, whether or not bufp->no_sub is set. */
2417 #if !defined emacs && !defined SYNTAX_TABLE
2418 /* Initialize the syntax table. */
2419 init_syntax_once ();
2422 if (bufp
->allocated
== 0)
2425 { /* If zero allocated, but buffer is non-null, try to realloc
2426 enough space. This loses if buffer's address is bogus, but
2427 that is the user's responsibility. */
2429 /* Free bufp->buffer and allocate an array for wchar_t pattern
2432 COMPILED_BUFFER_VAR
= TALLOC (INIT_BUF_SIZE
/sizeof(UCHAR_T
),
2435 RETALLOC (COMPILED_BUFFER_VAR
, INIT_BUF_SIZE
, UCHAR_T
);
2439 { /* Caller did not allocate a buffer. Do it for them. */
2440 COMPILED_BUFFER_VAR
= TALLOC (INIT_BUF_SIZE
/ sizeof(UCHAR_T
),
2444 if (!COMPILED_BUFFER_VAR
) FREE_STACK_RETURN (REG_ESPACE
);
2446 bufp
->buffer
= (char*)COMPILED_BUFFER_VAR
;
2448 bufp
->allocated
= INIT_BUF_SIZE
;
2452 COMPILED_BUFFER_VAR
= (UCHAR_T
*) bufp
->buffer
;
2455 begalt
= b
= COMPILED_BUFFER_VAR
;
2457 /* Loop through the uncompiled pattern until we're at the end. */
2466 if ( /* If at start of pattern, it's an operator. */
2468 /* If context independent, it's an operator. */
2469 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2470 /* Otherwise, depends on what's come before. */
2471 || PREFIX(at_begline_loc_p
) (pattern
, p
, syntax
))
2481 if ( /* If at end of pattern, it's an operator. */
2483 /* If context independent, it's an operator. */
2484 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2485 /* Otherwise, depends on what's next. */
2486 || PREFIX(at_endline_loc_p
) (p
, pend
, syntax
))
2496 if ((syntax
& RE_BK_PLUS_QM
)
2497 || (syntax
& RE_LIMITED_OPS
))
2502 /* If there is no previous pattern... */
2505 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2506 FREE_STACK_RETURN (REG_BADRPT
);
2507 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2512 /* Are we optimizing this jump? */
2513 boolean keep_string_p
= false;
2515 /* 1 means zero (many) matches is allowed. */
2516 char zero_times_ok
= 0, many_times_ok
= 0;
2518 /* If there is a sequence of repetition chars, collapse it
2519 down to just one (the right one). We can't combine
2520 interval operators with these because of, e.g., `a{2}*',
2521 which should only match an even number of `a's. */
2525 zero_times_ok
|= c
!= '+';
2526 many_times_ok
|= c
!= '?';
2534 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
2537 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
2539 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2542 if (!(c1
== '+' || c1
== '?'))
2557 /* If we get here, we found another repeat character. */
2560 /* Star, etc. applied to an empty pattern is equivalent
2561 to an empty pattern. */
2565 /* Now we know whether or not zero matches is allowed
2566 and also whether or not two or more matches is allowed. */
2568 { /* More than one repetition is allowed, so put in at the
2569 end a backward relative jump from `b' to before the next
2570 jump we're going to put in below (which jumps from
2571 laststart to after this jump).
2573 But if we are at the `*' in the exact sequence `.*\n',
2574 insert an unconditional jump backwards to the .,
2575 instead of the beginning of the loop. This way we only
2576 push a failure point once, instead of every time
2577 through the loop. */
2578 assert (p
- 1 > pattern
);
2580 /* Allocate the space for the jump. */
2581 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2583 /* We know we are not at the first character of the pattern,
2584 because laststart was nonzero. And we've already
2585 incremented `p', by the way, to be the character after
2586 the `*'. Do we have to do something analogous here
2587 for null bytes, because of RE_DOT_NOT_NULL? */
2588 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2590 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2591 && !(syntax
& RE_DOT_NEWLINE
))
2592 { /* We have .*\n. */
2593 STORE_JUMP (jump
, b
, laststart
);
2594 keep_string_p
= true;
2597 /* Anything else. */
2598 STORE_JUMP (maybe_pop_jump
, b
, laststart
-
2599 (1 + OFFSET_ADDRESS_SIZE
));
2601 /* We've added more stuff to the buffer. */
2602 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2605 /* On failure, jump from laststart to b + 3, which will be the
2606 end of the buffer after this jump is inserted. */
2607 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2609 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2610 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2612 laststart
, b
+ 1 + OFFSET_ADDRESS_SIZE
);
2614 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2618 /* At least one repetition is required, so insert a
2619 `dummy_failure_jump' before the initial
2620 `on_failure_jump' instruction of the loop. This
2621 effects a skip over that instruction the first time
2622 we hit that loop. */
2623 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2624 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+
2625 2 + 2 * OFFSET_ADDRESS_SIZE
);
2626 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2640 boolean had_char_class
= false;
2642 CHAR_T range_start
= 0xffffffff;
2644 unsigned int range_start
= 0xffffffff;
2646 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2649 /* We assume a charset(_not) structure as a wchar_t array.
2650 charset[0] = (re_opcode_t) charset(_not)
2651 charset[1] = l (= length of char_classes)
2652 charset[2] = m (= length of collating_symbols)
2653 charset[3] = n (= length of equivalence_classes)
2654 charset[4] = o (= length of char_ranges)
2655 charset[5] = p (= length of chars)
2657 charset[6] = char_class (wctype_t)
2658 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2660 charset[l+5] = char_class (wctype_t)
2662 charset[l+6] = collating_symbol (wchar_t)
2664 charset[l+m+5] = collating_symbol (wchar_t)
2665 ifdef _LIBC we use the index if
2666 _NL_COLLATE_SYMB_EXTRAMB instead of
2669 charset[l+m+6] = equivalence_classes (wchar_t)
2671 charset[l+m+n+5] = equivalence_classes (wchar_t)
2672 ifdef _LIBC we use the index in
2673 _NL_COLLATE_WEIGHT instead of
2676 charset[l+m+n+6] = range_start
2677 charset[l+m+n+7] = range_end
2679 charset[l+m+n+2o+4] = range_start
2680 charset[l+m+n+2o+5] = range_end
2681 ifdef _LIBC we use the value looked up
2682 in _NL_COLLATE_COLLSEQ instead of
2685 charset[l+m+n+2o+6] = char
2687 charset[l+m+n+2o+p+5] = char
2691 /* We need at least 6 spaces: the opcode, the length of
2692 char_classes, the length of collating_symbols, the length of
2693 equivalence_classes, the length of char_ranges, the length of
2695 GET_BUFFER_SPACE (6);
2697 /* Save b as laststart. And We use laststart as the pointer
2698 to the first element of the charset here.
2699 In other words, laststart[i] indicates charset[i]. */
2702 /* We test `*p == '^' twice, instead of using an if
2703 statement, so we only need one BUF_PUSH. */
2704 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2708 /* Push the length of char_classes, the length of
2709 collating_symbols, the length of equivalence_classes, the
2710 length of char_ranges and the length of chars. */
2711 BUF_PUSH_3 (0, 0, 0);
2714 /* Remember the first position in the bracket expression. */
2717 /* charset_not matches newline according to a syntax bit. */
2718 if ((re_opcode_t
) b
[-6] == charset_not
2719 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2722 laststart
[5]++; /* Update the length of characters */
2725 /* Read in characters and ranges, setting map bits. */
2728 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2732 /* \ might escape characters inside [...] and [^...]. */
2733 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2735 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2739 laststart
[5]++; /* Update the length of chars */
2744 /* Could be the end of the bracket expression. If it's
2745 not (i.e., when the bracket expression is `[]' so
2746 far), the ']' character bit gets set way below. */
2747 if (c
== ']' && p
!= p1
+ 1)
2750 /* Look ahead to see if it's a range when the last thing
2751 was a character class. */
2752 if (had_char_class
&& c
== '-' && *p
!= ']')
2753 FREE_STACK_RETURN (REG_ERANGE
);
2755 /* Look ahead to see if it's a range when the last thing
2756 was a character: if this is a hyphen not at the
2757 beginning or the end of a list, then it's the range
2760 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2761 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2765 /* Allocate the space for range_start and range_end. */
2766 GET_BUFFER_SPACE (2);
2767 /* Update the pointer to indicate end of buffer. */
2769 ret
= wcs_compile_range (range_start
, &p
, pend
, translate
,
2770 syntax
, b
, laststart
);
2771 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2772 range_start
= 0xffffffff;
2774 else if (p
[0] == '-' && p
[1] != ']')
2775 { /* This handles ranges made up of characters only. */
2778 /* Move past the `-'. */
2780 /* Allocate the space for range_start and range_end. */
2781 GET_BUFFER_SPACE (2);
2782 /* Update the pointer to indicate end of buffer. */
2784 ret
= wcs_compile_range (c
, &p
, pend
, translate
, syntax
, b
,
2786 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2787 range_start
= 0xffffffff;
2790 /* See if we're at the beginning of a possible character
2792 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2793 { /* Leave room for the null. */
2794 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2799 /* If pattern is `[[:'. */
2800 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2805 if ((c
== ':' && *p
== ']') || p
== pend
)
2807 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2810 /* This is in any case an invalid class name. */
2815 /* If isn't a word bracketed by `[:' and `:]':
2816 undo the ending character, the letters, and leave
2817 the leading `:' and `[' (but store them as character). */
2818 if (c
== ':' && *p
== ']')
2823 /* Query the character class as wctype_t. */
2824 wt
= IS_CHAR_CLASS (str
);
2826 FREE_STACK_RETURN (REG_ECTYPE
);
2828 /* Throw away the ] at the end of the character
2832 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2834 /* Allocate the space for character class. */
2835 GET_BUFFER_SPACE(CHAR_CLASS_SIZE
);
2836 /* Update the pointer to indicate end of buffer. */
2837 b
+= CHAR_CLASS_SIZE
;
2838 /* Move data which follow character classes
2839 not to violate the data. */
2840 insert_space(CHAR_CLASS_SIZE
,
2841 laststart
+ 6 + laststart
[1],
2843 alignedp
= ((uintptr_t)(laststart
+ 6 + laststart
[1])
2844 + __alignof__(wctype_t) - 1)
2845 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2846 /* Store the character class. */
2847 *((wctype_t*)alignedp
) = wt
;
2848 /* Update length of char_classes */
2849 laststart
[1] += CHAR_CLASS_SIZE
;
2851 had_char_class
= true;
2860 laststart
[5] += 2; /* Update the length of characters */
2862 had_char_class
= false;
2865 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && (*p
== '='
2868 CHAR_T str
[128]; /* Should be large enough. */
2869 CHAR_T delim
= *p
; /* '=' or '.' */
2872 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
2877 /* If pattern is `[[=' or '[[.'. */
2878 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2883 if ((c
== delim
&& *p
== ']') || p
== pend
)
2885 if (c1
< sizeof (str
) - 1)
2888 /* This is in any case an invalid class name. */
2893 if (c
== delim
&& *p
== ']' && str
[0] != '\0')
2895 unsigned int i
, offset
;
2896 /* If we have no collation data we use the default
2897 collation in which each character is in a class
2898 by itself. It also means that ASCII is the
2899 character set and therefore we cannot have character
2900 with more than one byte in the multibyte
2903 /* If not defined _LIBC, we push the name and
2904 `\0' for the sake of matching performance. */
2905 int datasize
= c1
+ 1;
2913 FREE_STACK_RETURN (REG_ECOLLATE
);
2918 const int32_t *table
;
2919 const int32_t *weights
;
2920 const int32_t *extra
;
2921 const int32_t *indirect
;
2924 /* This #include defines a local function! */
2925 # include <locale/weightwc.h>
2929 /* We push the index for equivalence class. */
2932 table
= (const int32_t *)
2933 _NL_CURRENT (LC_COLLATE
,
2934 _NL_COLLATE_TABLEWC
);
2935 weights
= (const int32_t *)
2936 _NL_CURRENT (LC_COLLATE
,
2937 _NL_COLLATE_WEIGHTWC
);
2938 extra
= (const int32_t *)
2939 _NL_CURRENT (LC_COLLATE
,
2940 _NL_COLLATE_EXTRAWC
);
2941 indirect
= (const int32_t *)
2942 _NL_CURRENT (LC_COLLATE
,
2943 _NL_COLLATE_INDIRECTWC
);
2945 idx
= findidx ((const wint_t**)&cp
);
2946 if (idx
== 0 || cp
< (wint_t*) str
+ c1
)
2947 /* This is no valid character. */
2948 FREE_STACK_RETURN (REG_ECOLLATE
);
2950 str
[0] = (wchar_t)idx
;
2952 else /* delim == '.' */
2954 /* We push collation sequence value
2955 for collating symbol. */
2957 const int32_t *symb_table
;
2958 const unsigned char *extra
;
2965 /* We have to convert the name to a single-byte
2966 string. This is possible since the names
2967 consist of ASCII characters and the internal
2968 representation is UCS4. */
2969 for (i
= 0; i
< c1
; ++i
)
2970 char_str
[i
] = str
[i
];
2973 _NL_CURRENT_WORD (LC_COLLATE
,
2974 _NL_COLLATE_SYMB_HASH_SIZEMB
);
2975 symb_table
= (const int32_t *)
2976 _NL_CURRENT (LC_COLLATE
,
2977 _NL_COLLATE_SYMB_TABLEMB
);
2978 extra
= (const unsigned char *)
2979 _NL_CURRENT (LC_COLLATE
,
2980 _NL_COLLATE_SYMB_EXTRAMB
);
2982 /* Locate the character in the hashing table. */
2983 hash
= elem_hash (char_str
, c1
);
2986 elem
= hash
% table_size
;
2987 second
= hash
% (table_size
- 2);
2988 while (symb_table
[2 * elem
] != 0)
2990 /* First compare the hashing value. */
2991 if (symb_table
[2 * elem
] == hash
2992 && c1
== extra
[symb_table
[2 * elem
+ 1]]
2993 && memcmp (char_str
,
2994 &extra
[symb_table
[2 * elem
+ 1]
2997 /* Yep, this is the entry. */
2998 idx
= symb_table
[2 * elem
+ 1];
2999 idx
+= 1 + extra
[idx
];
3007 if (symb_table
[2 * elem
] != 0)
3009 /* Compute the index of the byte sequence
3011 idx
+= 1 + extra
[idx
];
3012 /* Adjust for the alignment. */
3013 idx
= (idx
+ 3) & ~3;
3015 str
[0] = (wchar_t) idx
+ 4;
3017 else if (symb_table
[2 * elem
] == 0 && c1
== 1)
3019 /* No valid character. Match it as a
3020 single byte character. */
3021 had_char_class
= false;
3023 /* Update the length of characters */
3025 range_start
= str
[0];
3027 /* Throw away the ] at the end of the
3028 collating symbol. */
3030 /* exit from the switch block. */
3034 FREE_STACK_RETURN (REG_ECOLLATE
);
3039 /* Throw away the ] at the end of the equivalence
3040 class (or collating symbol). */
3043 /* Allocate the space for the equivalence class
3044 (or collating symbol) (and '\0' if needed). */
3045 GET_BUFFER_SPACE(datasize
);
3046 /* Update the pointer to indicate end of buffer. */
3050 { /* equivalence class */
3051 /* Calculate the offset of char_ranges,
3052 which is next to equivalence_classes. */
3053 offset
= laststart
[1] + laststart
[2]
3056 insert_space(datasize
, laststart
+ offset
, b
- 1);
3058 /* Write the equivalence_class and \0. */
3059 for (i
= 0 ; i
< datasize
; i
++)
3060 laststart
[offset
+ i
] = str
[i
];
3062 /* Update the length of equivalence_classes. */
3063 laststart
[3] += datasize
;
3064 had_char_class
= true;
3066 else /* delim == '.' */
3067 { /* collating symbol */
3068 /* Calculate the offset of the equivalence_classes,
3069 which is next to collating_symbols. */
3070 offset
= laststart
[1] + laststart
[2] + 6;
3071 /* Insert space and write the collationg_symbol
3073 insert_space(datasize
, laststart
+ offset
, b
-1);
3074 for (i
= 0 ; i
< datasize
; i
++)
3075 laststart
[offset
+ i
] = str
[i
];
3077 /* In re_match_2_internal if range_start < -1, we
3078 assume -range_start is the offset of the
3079 collating symbol which is specified as
3080 the character of the range start. So we assign
3081 -(laststart[1] + laststart[2] + 6) to
3083 range_start
= -(laststart
[1] + laststart
[2] + 6);
3084 /* Update the length of collating_symbol. */
3085 laststart
[2] += datasize
;
3086 had_char_class
= false;
3096 laststart
[5] += 2; /* Update the length of characters */
3097 range_start
= delim
;
3098 had_char_class
= false;
3103 had_char_class
= false;
3105 laststart
[5]++; /* Update the length of characters */
3111 /* Ensure that we have enough space to push a charset: the
3112 opcode, the length count, and the bitset; 34 bytes in all. */
3113 GET_BUFFER_SPACE (34);
3117 /* We test `*p == '^' twice, instead of using an if
3118 statement, so we only need one BUF_PUSH. */
3119 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
3123 /* Remember the first position in the bracket expression. */
3126 /* Push the number of bytes in the bitmap. */
3127 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
3129 /* Clear the whole map. */
3130 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
3132 /* charset_not matches newline according to a syntax bit. */
3133 if ((re_opcode_t
) b
[-2] == charset_not
3134 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
3135 SET_LIST_BIT ('\n');
3137 /* Read in characters and ranges, setting map bits. */
3140 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3144 /* \ might escape characters inside [...] and [^...]. */
3145 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
3147 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3155 /* Could be the end of the bracket expression. If it's
3156 not (i.e., when the bracket expression is `[]' so
3157 far), the ']' character bit gets set way below. */
3158 if (c
== ']' && p
!= p1
+ 1)
3161 /* Look ahead to see if it's a range when the last thing
3162 was a character class. */
3163 if (had_char_class
&& c
== '-' && *p
!= ']')
3164 FREE_STACK_RETURN (REG_ERANGE
);
3166 /* Look ahead to see if it's a range when the last thing
3167 was a character: if this is a hyphen not at the
3168 beginning or the end of a list, then it's the range
3171 && !(p
- 2 >= pattern
&& p
[-2] == '[')
3172 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
3176 = byte_compile_range (range_start
, &p
, pend
, translate
,
3178 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
3179 range_start
= 0xffffffff;
3182 else if (p
[0] == '-' && p
[1] != ']')
3183 { /* This handles ranges made up of characters only. */
3186 /* Move past the `-'. */
3189 ret
= byte_compile_range (c
, &p
, pend
, translate
, syntax
, b
);
3190 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
3191 range_start
= 0xffffffff;
3194 /* See if we're at the beginning of a possible character
3197 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
3198 { /* Leave room for the null. */
3199 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
3204 /* If pattern is `[[:'. */
3205 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3210 if ((c
== ':' && *p
== ']') || p
== pend
)
3212 if (c1
< CHAR_CLASS_MAX_LENGTH
)
3215 /* This is in any case an invalid class name. */
3220 /* If isn't a word bracketed by `[:' and `:]':
3221 undo the ending character, the letters, and leave
3222 the leading `:' and `[' (but set bits for them). */
3223 if (c
== ':' && *p
== ']')
3225 # if defined _LIBC || WIDE_CHAR_SUPPORT
3226 boolean is_lower
= STREQ (str
, "lower");
3227 boolean is_upper
= STREQ (str
, "upper");
3231 wt
= IS_CHAR_CLASS (str
);
3233 FREE_STACK_RETURN (REG_ECTYPE
);
3235 /* Throw away the ] at the end of the character
3239 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3241 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
3244 if (__iswctype (__btowc (ch
), wt
))
3247 if (iswctype (btowc (ch
), wt
))
3251 if (translate
&& (is_upper
|| is_lower
)
3252 && (ISUPPER (ch
) || ISLOWER (ch
)))
3256 had_char_class
= true;
3259 boolean is_alnum
= STREQ (str
, "alnum");
3260 boolean is_alpha
= STREQ (str
, "alpha");
3261 boolean is_blank
= STREQ (str
, "blank");
3262 boolean is_cntrl
= STREQ (str
, "cntrl");
3263 boolean is_digit
= STREQ (str
, "digit");
3264 boolean is_graph
= STREQ (str
, "graph");
3265 boolean is_lower
= STREQ (str
, "lower");
3266 boolean is_print
= STREQ (str
, "print");
3267 boolean is_punct
= STREQ (str
, "punct");
3268 boolean is_space
= STREQ (str
, "space");
3269 boolean is_upper
= STREQ (str
, "upper");
3270 boolean is_xdigit
= STREQ (str
, "xdigit");
3272 if (!IS_CHAR_CLASS (str
))
3273 FREE_STACK_RETURN (REG_ECTYPE
);
3275 /* Throw away the ] at the end of the character
3279 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3281 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
3283 /* This was split into 3 if's to
3284 avoid an arbitrary limit in some compiler. */
3285 if ( (is_alnum
&& ISALNUM (ch
))
3286 || (is_alpha
&& ISALPHA (ch
))
3287 || (is_blank
&& ISBLANK (ch
))
3288 || (is_cntrl
&& ISCNTRL (ch
)))
3290 if ( (is_digit
&& ISDIGIT (ch
))
3291 || (is_graph
&& ISGRAPH (ch
))
3292 || (is_lower
&& ISLOWER (ch
))
3293 || (is_print
&& ISPRINT (ch
)))
3295 if ( (is_punct
&& ISPUNCT (ch
))
3296 || (is_space
&& ISSPACE (ch
))
3297 || (is_upper
&& ISUPPER (ch
))
3298 || (is_xdigit
&& ISXDIGIT (ch
)))
3300 if ( translate
&& (is_upper
|| is_lower
)
3301 && (ISUPPER (ch
) || ISLOWER (ch
)))
3304 had_char_class
= true;
3305 # endif /* libc || wctype.h */
3315 had_char_class
= false;
3318 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '=')
3320 unsigned char str
[MB_LEN_MAX
+ 1];
3323 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3329 /* If pattern is `[[='. */
3330 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3335 if ((c
== '=' && *p
== ']') || p
== pend
)
3337 if (c1
< MB_LEN_MAX
)
3340 /* This is in any case an invalid class name. */
3345 if (c
== '=' && *p
== ']' && str
[0] != '\0')
3347 /* If we have no collation data we use the default
3348 collation in which each character is in a class
3349 by itself. It also means that ASCII is the
3350 character set and therefore we cannot have character
3351 with more than one byte in the multibyte
3358 FREE_STACK_RETURN (REG_ECOLLATE
);
3360 /* Throw away the ] at the end of the equivalence
3364 /* Set the bit for the character. */
3365 SET_LIST_BIT (str
[0]);
3370 /* Try to match the byte sequence in `str' against
3371 those known to the collate implementation.
3372 First find out whether the bytes in `str' are
3373 actually from exactly one character. */
3374 const int32_t *table
;
3375 const unsigned char *weights
;
3376 const unsigned char *extra
;
3377 const int32_t *indirect
;
3379 const unsigned char *cp
= str
;
3382 /* This #include defines a local function! */
3383 # include <locale/weight.h>
3385 table
= (const int32_t *)
3386 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEMB
);
3387 weights
= (const unsigned char *)
3388 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTMB
);
3389 extra
= (const unsigned char *)
3390 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAMB
);
3391 indirect
= (const int32_t *)
3392 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTMB
);
3394 idx
= findidx (&cp
);
3395 if (idx
== 0 || cp
< str
+ c1
)
3396 /* This is no valid character. */
3397 FREE_STACK_RETURN (REG_ECOLLATE
);
3399 /* Throw away the ] at the end of the equivalence
3403 /* Now we have to go through the whole table
3404 and find all characters which have the same
3407 XXX Note that this is not entirely correct.
3408 we would have to match multibyte sequences
3409 but this is not possible with the current
3411 for (ch
= 1; ch
< 256; ++ch
)
3412 /* XXX This test would have to be changed if we
3413 would allow matching multibyte sequences. */
3416 int32_t idx2
= table
[ch
];
3417 size_t len
= weights
[idx2
];
3419 /* Test whether the lenghts match. */
3420 if (weights
[idx
] == len
)
3422 /* They do. New compare the bytes of
3427 && (weights
[idx
+ 1 + cnt
]
3428 == weights
[idx2
+ 1 + cnt
]))
3432 /* They match. Mark the character as
3439 had_char_class
= true;
3449 had_char_class
= false;
3452 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '.')
3454 unsigned char str
[128]; /* Should be large enough. */
3457 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3463 /* If pattern is `[[.'. */
3464 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3469 if ((c
== '.' && *p
== ']') || p
== pend
)
3471 if (c1
< sizeof (str
))
3474 /* This is in any case an invalid class name. */
3479 if (c
== '.' && *p
== ']' && str
[0] != '\0')
3481 /* If we have no collation data we use the default
3482 collation in which each character is the name
3483 for its own class which contains only the one
3484 character. It also means that ASCII is the
3485 character set and therefore we cannot have character
3486 with more than one byte in the multibyte
3493 FREE_STACK_RETURN (REG_ECOLLATE
);
3495 /* Throw away the ] at the end of the equivalence
3499 /* Set the bit for the character. */
3500 SET_LIST_BIT (str
[0]);
3501 range_start
= ((const unsigned char *) str
)[0];
3506 /* Try to match the byte sequence in `str' against
3507 those known to the collate implementation.
3508 First find out whether the bytes in `str' are
3509 actually from exactly one character. */
3511 const int32_t *symb_table
;
3512 const unsigned char *extra
;
3519 _NL_CURRENT_WORD (LC_COLLATE
,
3520 _NL_COLLATE_SYMB_HASH_SIZEMB
);
3521 symb_table
= (const int32_t *)
3522 _NL_CURRENT (LC_COLLATE
,
3523 _NL_COLLATE_SYMB_TABLEMB
);
3524 extra
= (const unsigned char *)
3525 _NL_CURRENT (LC_COLLATE
,
3526 _NL_COLLATE_SYMB_EXTRAMB
);
3528 /* Locate the character in the hashing table. */
3529 hash
= elem_hash (str
, c1
);
3532 elem
= hash
% table_size
;
3533 second
= hash
% (table_size
- 2);
3534 while (symb_table
[2 * elem
] != 0)
3536 /* First compare the hashing value. */
3537 if (symb_table
[2 * elem
] == hash
3538 && c1
== extra
[symb_table
[2 * elem
+ 1]]
3540 &extra
[symb_table
[2 * elem
+ 1]
3544 /* Yep, this is the entry. */
3545 idx
= symb_table
[2 * elem
+ 1];
3546 idx
+= 1 + extra
[idx
];
3554 if (symb_table
[2 * elem
] == 0)
3555 /* This is no valid character. */
3556 FREE_STACK_RETURN (REG_ECOLLATE
);
3558 /* Throw away the ] at the end of the equivalence
3562 /* Now add the multibyte character(s) we found
3565 XXX Note that this is not entirely correct.
3566 we would have to match multibyte sequences
3567 but this is not possible with the current
3568 implementation. Also, we have to match
3569 collating symbols, which expand to more than
3570 one file, as a whole and not allow the
3571 individual bytes. */
3574 range_start
= extra
[idx
];
3577 SET_LIST_BIT (extra
[idx
]);
3582 had_char_class
= false;
3592 had_char_class
= false;
3597 had_char_class
= false;
3603 /* Discard any (non)matching list bytes that are all 0 at the
3604 end of the map. Decrease the map-length byte too. */
3605 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3614 if (syntax
& RE_NO_BK_PARENS
)
3621 if (syntax
& RE_NO_BK_PARENS
)
3628 if (syntax
& RE_NEWLINE_ALT
)
3635 if (syntax
& RE_NO_BK_VBAR
)
3642 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3643 goto handle_interval
;
3649 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3651 /* Do not translate the character after the \, so that we can
3652 distinguish, e.g., \B from \b, even if we normally would
3653 translate, e.g., B to b. */
3659 if (syntax
& RE_NO_BK_PARENS
)
3660 goto normal_backslash
;
3666 if (COMPILE_STACK_FULL
)
3668 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3669 compile_stack_elt_t
);
3670 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3672 compile_stack
.size
<<= 1;
3675 /* These are the values to restore when we hit end of this
3676 group. They are all relative offsets, so that if the
3677 whole pattern moves because of realloc, they will still
3679 COMPILE_STACK_TOP
.begalt_offset
= begalt
- COMPILED_BUFFER_VAR
;
3680 COMPILE_STACK_TOP
.fixup_alt_jump
3681 = fixup_alt_jump
? fixup_alt_jump
- COMPILED_BUFFER_VAR
+ 1 : 0;
3682 COMPILE_STACK_TOP
.laststart_offset
= b
- COMPILED_BUFFER_VAR
;
3683 COMPILE_STACK_TOP
.regnum
= regnum
;
3685 /* We will eventually replace the 0 with the number of
3686 groups inner to this one. But do not push a
3687 start_memory for groups beyond the last one we can
3688 represent in the compiled pattern. */
3689 if (regnum
<= MAX_REGNUM
)
3691 COMPILE_STACK_TOP
.inner_group_offset
= b
3692 - COMPILED_BUFFER_VAR
+ 2;
3693 BUF_PUSH_3 (start_memory
, regnum
, 0);
3696 compile_stack
.avail
++;
3701 /* If we've reached MAX_REGNUM groups, then this open
3702 won't actually generate any code, so we'll have to
3703 clear pending_exact explicitly. */
3709 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3711 if (COMPILE_STACK_EMPTY
)
3713 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3714 goto normal_backslash
;
3716 FREE_STACK_RETURN (REG_ERPAREN
);
3721 { /* Push a dummy failure point at the end of the
3722 alternative for a possible future
3723 `pop_failure_jump' to pop. See comments at
3724 `push_dummy_failure' in `re_match_2'. */
3725 BUF_PUSH (push_dummy_failure
);
3727 /* We allocated space for this jump when we assigned
3728 to `fixup_alt_jump', in the `handle_alt' case below. */
3729 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
3732 /* See similar code for backslashed left paren above. */
3733 if (COMPILE_STACK_EMPTY
)
3735 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3738 FREE_STACK_RETURN (REG_ERPAREN
);
3741 /* Since we just checked for an empty stack above, this
3742 ``can't happen''. */
3743 assert (compile_stack
.avail
!= 0);
3745 /* We don't just want to restore into `regnum', because
3746 later groups should continue to be numbered higher,
3747 as in `(ab)c(de)' -- the second group is #2. */
3748 regnum_t this_group_regnum
;
3750 compile_stack
.avail
--;
3751 begalt
= COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.begalt_offset
;
3753 = COMPILE_STACK_TOP
.fixup_alt_jump
3754 ? COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3756 laststart
= COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.laststart_offset
;
3757 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
3758 /* If we've reached MAX_REGNUM groups, then this open
3759 won't actually generate any code, so we'll have to
3760 clear pending_exact explicitly. */
3763 /* We're at the end of the group, so now we know how many
3764 groups were inside this one. */
3765 if (this_group_regnum
<= MAX_REGNUM
)
3767 UCHAR_T
*inner_group_loc
3768 = COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.inner_group_offset
;
3770 *inner_group_loc
= regnum
- this_group_regnum
;
3771 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
3772 regnum
- this_group_regnum
);
3778 case '|': /* `\|'. */
3779 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3780 goto normal_backslash
;
3782 if (syntax
& RE_LIMITED_OPS
)
3785 /* Insert before the previous alternative a jump which
3786 jumps to this alternative if the former fails. */
3787 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3788 INSERT_JUMP (on_failure_jump
, begalt
,
3789 b
+ 2 + 2 * OFFSET_ADDRESS_SIZE
);
3791 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3793 /* The alternative before this one has a jump after it
3794 which gets executed if it gets matched. Adjust that
3795 jump so it will jump to this alternative's analogous
3796 jump (put in below, which in turn will jump to the next
3797 (if any) alternative's such jump, etc.). The last such
3798 jump jumps to the correct final destination. A picture:
3804 If we are at `b', then fixup_alt_jump right now points to a
3805 three-byte space after `a'. We'll put in the jump, set
3806 fixup_alt_jump to right after `b', and leave behind three
3807 bytes which we'll fill in when we get to after `c'. */
3810 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
3812 /* Mark and leave space for a jump after this alternative,
3813 to be filled in later either by next alternative or
3814 when know we're at the end of a series of alternatives. */
3816 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3817 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3825 /* If \{ is a literal. */
3826 if (!(syntax
& RE_INTERVALS
)
3827 /* If we're at `\{' and it's not the open-interval
3829 || (syntax
& RE_NO_BK_BRACES
))
3830 goto normal_backslash
;
3834 /* If got here, then the syntax allows intervals. */
3836 /* At least (most) this many matches must be made. */
3837 int lower_bound
= -1, upper_bound
= -1;
3839 /* Place in the uncompiled pattern (i.e., just after
3840 the '{') to go back to if the interval is invalid. */
3841 const CHAR_T
*beg_interval
= p
;
3844 goto invalid_interval
;
3846 GET_UNSIGNED_NUMBER (lower_bound
);
3850 GET_UNSIGNED_NUMBER (upper_bound
);
3851 if (upper_bound
< 0)
3852 upper_bound
= RE_DUP_MAX
;
3855 /* Interval such as `{1}' => match exactly once. */
3856 upper_bound
= lower_bound
;
3858 if (! (0 <= lower_bound
&& lower_bound
<= upper_bound
))
3859 goto invalid_interval
;
3861 if (!(syntax
& RE_NO_BK_BRACES
))
3863 if (c
!= '\\' || p
== pend
)
3864 goto invalid_interval
;
3869 goto invalid_interval
;
3871 /* If it's invalid to have no preceding re. */
3874 if (syntax
& RE_CONTEXT_INVALID_OPS
3875 && !(syntax
& RE_INVALID_INTERVAL_ORD
))
3876 FREE_STACK_RETURN (REG_BADRPT
);
3877 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3880 goto unfetch_interval
;
3883 /* We just parsed a valid interval. */
3885 if (RE_DUP_MAX
< upper_bound
)
3886 FREE_STACK_RETURN (REG_BADBR
);
3888 /* If the upper bound is zero, don't want to succeed at
3889 all; jump from `laststart' to `b + 3', which will be
3890 the end of the buffer after we insert the jump. */
3891 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3892 instead of 'b + 3'. */
3893 if (upper_bound
== 0)
3895 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3896 INSERT_JUMP (jump
, laststart
, b
+ 1
3897 + OFFSET_ADDRESS_SIZE
);
3898 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3901 /* Otherwise, we have a nontrivial interval. When
3902 we're all done, the pattern will look like:
3903 set_number_at <jump count> <upper bound>
3904 set_number_at <succeed_n count> <lower bound>
3905 succeed_n <after jump addr> <succeed_n count>
3907 jump_n <succeed_n addr> <jump count>
3908 (The upper bound and `jump_n' are omitted if
3909 `upper_bound' is 1, though.) */
3911 { /* If the upper bound is > 1, we need to insert
3912 more at the end of the loop. */
3913 unsigned nbytes
= 2 + 4 * OFFSET_ADDRESS_SIZE
+
3914 (upper_bound
> 1) * (2 + 4 * OFFSET_ADDRESS_SIZE
);
3916 GET_BUFFER_SPACE (nbytes
);
3918 /* Initialize lower bound of the `succeed_n', even
3919 though it will be set during matching by its
3920 attendant `set_number_at' (inserted next),
3921 because `re_compile_fastmap' needs to know.
3922 Jump to the `jump_n' we might insert below. */
3923 INSERT_JUMP2 (succeed_n
, laststart
,
3924 b
+ 1 + 2 * OFFSET_ADDRESS_SIZE
3925 + (upper_bound
> 1) * (1 + 2 * OFFSET_ADDRESS_SIZE
)
3927 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
3929 /* Code to initialize the lower bound. Insert
3930 before the `succeed_n'. The `5' is the last two
3931 bytes of this `set_number_at', plus 3 bytes of
3932 the following `succeed_n'. */
3933 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
3934 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
3935 of the following `succeed_n'. */
3936 PREFIX(insert_op2
) (set_number_at
, laststart
, 1
3937 + 2 * OFFSET_ADDRESS_SIZE
, lower_bound
, b
);
3938 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
3940 if (upper_bound
> 1)
3941 { /* More than one repetition is allowed, so
3942 append a backward jump to the `succeed_n'
3943 that starts this interval.
3945 When we've reached this during matching,
3946 we'll have matched the interval once, so
3947 jump back only `upper_bound - 1' times. */
3948 STORE_JUMP2 (jump_n
, b
, laststart
3949 + 2 * OFFSET_ADDRESS_SIZE
+ 1,
3951 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
3953 /* The location we want to set is the second
3954 parameter of the `jump_n'; that is `b-2' as
3955 an absolute address. `laststart' will be
3956 the `set_number_at' we're about to insert;
3957 `laststart+3' the number to set, the source
3958 for the relative address. But we are
3959 inserting into the middle of the pattern --
3960 so everything is getting moved up by 5.
3961 Conclusion: (b - 2) - (laststart + 3) + 5,
3962 i.e., b - laststart.
3964 We insert this at the beginning of the loop
3965 so that if we fail during matching, we'll
3966 reinitialize the bounds. */
3967 PREFIX(insert_op2
) (set_number_at
, laststart
,
3969 upper_bound
- 1, b
);
3970 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
3977 if (!(syntax
& RE_INVALID_INTERVAL_ORD
))
3978 FREE_STACK_RETURN (p
== pend
? REG_EBRACE
: REG_BADBR
);
3980 /* Match the characters as literals. */
3983 if (syntax
& RE_NO_BK_BRACES
)
3986 goto normal_backslash
;
3990 /* There is no way to specify the before_dot and after_dot
3991 operators. rms says this is ok. --karl */
3999 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
4005 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
4011 if (syntax
& RE_NO_GNU_OPS
)
4014 BUF_PUSH (wordchar
);
4019 if (syntax
& RE_NO_GNU_OPS
)
4022 BUF_PUSH (notwordchar
);
4027 if (syntax
& RE_NO_GNU_OPS
)
4033 if (syntax
& RE_NO_GNU_OPS
)
4039 if (syntax
& RE_NO_GNU_OPS
)
4041 BUF_PUSH (wordbound
);
4045 if (syntax
& RE_NO_GNU_OPS
)
4047 BUF_PUSH (notwordbound
);
4051 if (syntax
& RE_NO_GNU_OPS
)
4057 if (syntax
& RE_NO_GNU_OPS
)
4062 case '1': case '2': case '3': case '4': case '5':
4063 case '6': case '7': case '8': case '9':
4064 if (syntax
& RE_NO_BK_REFS
)
4070 FREE_STACK_RETURN (REG_ESUBREG
);
4072 /* Can't back reference to a subexpression if inside of it. */
4073 if (group_in_compile_stack (compile_stack
, (regnum_t
) c1
))
4077 BUF_PUSH_2 (duplicate
, c1
);
4083 if (syntax
& RE_BK_PLUS_QM
)
4086 goto normal_backslash
;
4090 /* You might think it would be useful for \ to mean
4091 not to translate; but if we don't translate it
4092 it will never match anything. */
4100 /* Expects the character in `c'. */
4102 /* If no exactn currently being built. */
4105 /* If last exactn handle binary(or character) and
4106 new exactn handle character(or binary). */
4107 || is_exactn_bin
!= is_binary
[p
- 1 - pattern
]
4110 /* If last exactn not at current position. */
4111 || pending_exact
+ *pending_exact
+ 1 != b
4113 /* We have only one byte following the exactn for the count. */
4114 || *pending_exact
== (1 << BYTEWIDTH
) - 1
4116 /* If followed by a repetition operator. */
4117 || *p
== '*' || *p
== '^'
4118 || ((syntax
& RE_BK_PLUS_QM
)
4119 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
4120 : (*p
== '+' || *p
== '?'))
4121 || ((syntax
& RE_INTERVALS
)
4122 && ((syntax
& RE_NO_BK_BRACES
)
4124 : (p
[0] == '\\' && p
[1] == '{'))))
4126 /* Start building a new exactn. */
4131 /* Is this exactn binary data or character? */
4132 is_exactn_bin
= is_binary
[p
- 1 - pattern
];
4134 BUF_PUSH_2 (exactn_bin
, 0);
4136 BUF_PUSH_2 (exactn
, 0);
4138 BUF_PUSH_2 (exactn
, 0);
4140 pending_exact
= b
- 1;
4147 } /* while p != pend */
4150 /* Through the pattern now. */
4153 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
4155 if (!COMPILE_STACK_EMPTY
)
4156 FREE_STACK_RETURN (REG_EPAREN
);
4158 /* If we don't want backtracking, force success
4159 the first time we reach the end of the compiled pattern. */
4160 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
4168 free (compile_stack
.stack
);
4170 /* We have succeeded; set the length of the buffer. */
4172 bufp
->used
= (uintptr_t) b
- (uintptr_t) COMPILED_BUFFER_VAR
;
4174 bufp
->used
= b
- bufp
->buffer
;
4180 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4181 PREFIX(print_compiled_pattern
) (bufp
);
4185 #ifndef MATCH_MAY_ALLOCATE
4186 /* Initialize the failure stack to the largest possible stack. This
4187 isn't necessary unless we're trying to avoid calling alloca in
4188 the search and match routines. */
4190 int num_regs
= bufp
->re_nsub
+ 1;
4192 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4193 is strictly greater than re_max_failures, the largest possible stack
4194 is 2 * re_max_failures failure points. */
4195 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
4197 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
4200 if (! fail_stack
.stack
)
4202 = (PREFIX(fail_stack_elt_t
) *) xmalloc (fail_stack
.size
4203 * sizeof (PREFIX(fail_stack_elt_t
)));
4206 = (PREFIX(fail_stack_elt_t
) *) xrealloc (fail_stack
.stack
,
4208 * sizeof (PREFIX(fail_stack_elt_t
))));
4209 # else /* not emacs */
4210 if (! fail_stack
.stack
)
4212 = (PREFIX(fail_stack_elt_t
) *) malloc (fail_stack
.size
4213 * sizeof (PREFIX(fail_stack_elt_t
)));
4216 = (PREFIX(fail_stack_elt_t
) *) realloc (fail_stack
.stack
,
4218 * sizeof (PREFIX(fail_stack_elt_t
))));
4219 # endif /* not emacs */
4222 PREFIX(regex_grow_registers
) (num_regs
);
4224 #endif /* not MATCH_MAY_ALLOCATE */
4227 } /* regex_compile */
4229 /* Subroutines for `regex_compile'. */
4231 /* Store OP at LOC followed by two-byte integer parameter ARG. */
4232 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4235 PREFIX(store_op1
) (re_opcode_t op
, UCHAR_T
*loc
, int arg
)
4237 *loc
= (UCHAR_T
) op
;
4238 STORE_NUMBER (loc
+ 1, arg
);
4242 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4243 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4246 PREFIX(store_op2
) (re_opcode_t op
, UCHAR_T
*loc
, int arg1
, int arg2
)
4248 *loc
= (UCHAR_T
) op
;
4249 STORE_NUMBER (loc
+ 1, arg1
);
4250 STORE_NUMBER (loc
+ 1 + OFFSET_ADDRESS_SIZE
, arg2
);
4254 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
4255 for OP followed by two-byte integer parameter ARG. */
4256 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4259 PREFIX(insert_op1
) (re_opcode_t op
, UCHAR_T
*loc
, int arg
, UCHAR_T
*end
)
4261 register UCHAR_T
*pfrom
= end
;
4262 register UCHAR_T
*pto
= end
+ 1 + OFFSET_ADDRESS_SIZE
;
4264 while (pfrom
!= loc
)
4267 PREFIX(store_op1
) (op
, loc
, arg
);
4271 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4272 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4275 PREFIX(insert_op2
) (re_opcode_t op
, UCHAR_T
*loc
, int arg1
,
4276 int arg2
, UCHAR_T
*end
)
4278 register UCHAR_T
*pfrom
= end
;
4279 register UCHAR_T
*pto
= end
+ 1 + 2 * OFFSET_ADDRESS_SIZE
;
4281 while (pfrom
!= loc
)
4284 PREFIX(store_op2
) (op
, loc
, arg1
, arg2
);
4288 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
4289 after an alternative or a begin-subexpression. We assume there is at
4290 least one character before the ^. */
4293 PREFIX(at_begline_loc_p
) (const CHAR_T
*pattern
, const CHAR_T
*p
,
4294 reg_syntax_t syntax
)
4296 const CHAR_T
*prev
= p
- 2;
4297 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
4300 /* After a subexpression? */
4301 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
4302 /* After an alternative? */
4303 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
4307 /* The dual of at_begline_loc_p. This one is for $. We assume there is
4308 at least one character after the $, i.e., `P < PEND'. */
4311 PREFIX(at_endline_loc_p
) (const CHAR_T
*p
, const CHAR_T
*pend
,
4312 reg_syntax_t syntax
)
4314 const CHAR_T
*next
= p
;
4315 boolean next_backslash
= *next
== '\\';
4316 const CHAR_T
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
4319 /* Before a subexpression? */
4320 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
4321 : next_backslash
&& next_next
&& *next_next
== ')')
4322 /* Before an alternative? */
4323 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
4324 : next_backslash
&& next_next
&& *next_next
== '|');
4327 #else /* not INSIDE_RECURSION */
4329 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4330 false if it's not. */
4333 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
4337 for (this_element
= compile_stack
.avail
- 1;
4340 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
4345 #endif /* not INSIDE_RECURSION */
4347 #ifdef INSIDE_RECURSION
4350 /* This insert space, which size is "num", into the pattern at "loc".
4351 "end" must point the end of the allocated buffer. */
4353 insert_space (int num
, CHAR_T
*loc
, CHAR_T
*end
)
4355 register CHAR_T
*pto
= end
;
4356 register CHAR_T
*pfrom
= end
- num
;
4358 while (pfrom
>= loc
)
4364 static reg_errcode_t
4365 wcs_compile_range (CHAR_T range_start_char
, const CHAR_T
**p_ptr
,
4366 const CHAR_T
*pend
, RE_TRANSLATE_TYPE translate
,
4367 reg_syntax_t syntax
, CHAR_T
*b
, CHAR_T
*char_set
)
4369 const CHAR_T
*p
= *p_ptr
;
4370 CHAR_T range_start
, range_end
;
4374 uint32_t start_val
, end_val
;
4380 nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
4383 const char *collseq
= (const char *) _NL_CURRENT(LC_COLLATE
,
4384 _NL_COLLATE_COLLSEQWC
);
4385 const unsigned char *extra
= (const unsigned char *)
4386 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
4388 if (range_start_char
< -1)
4390 /* range_start is a collating symbol. */
4392 /* Retreive the index and get collation sequence value. */
4393 wextra
= (int32_t*)(extra
+ char_set
[-range_start_char
]);
4394 start_val
= wextra
[1 + *wextra
];
4397 start_val
= collseq_table_lookup(collseq
, TRANSLATE(range_start_char
));
4399 end_val
= collseq_table_lookup (collseq
, TRANSLATE (p
[0]));
4401 /* Report an error if the range is empty and the syntax prohibits
4403 ret
= ((syntax
& RE_NO_EMPTY_RANGES
)
4404 && (start_val
> end_val
))? REG_ERANGE
: REG_NOERROR
;
4406 /* Insert space to the end of the char_ranges. */
4407 insert_space(2, b
- char_set
[5] - 2, b
- 1);
4408 *(b
- char_set
[5] - 2) = (wchar_t)start_val
;
4409 *(b
- char_set
[5] - 1) = (wchar_t)end_val
;
4410 char_set
[4]++; /* ranges_index */
4415 range_start
= (range_start_char
>= 0)? TRANSLATE (range_start_char
):
4417 range_end
= TRANSLATE (p
[0]);
4418 /* Report an error if the range is empty and the syntax prohibits
4420 ret
= ((syntax
& RE_NO_EMPTY_RANGES
)
4421 && (range_start
> range_end
))? REG_ERANGE
: REG_NOERROR
;
4423 /* Insert space to the end of the char_ranges. */
4424 insert_space(2, b
- char_set
[5] - 2, b
- 1);
4425 *(b
- char_set
[5] - 2) = range_start
;
4426 *(b
- char_set
[5] - 1) = range_end
;
4427 char_set
[4]++; /* ranges_index */
4429 /* Have to increment the pointer into the pattern string, so the
4430 caller isn't still at the ending character. */
4436 /* Read the ending character of a range (in a bracket expression) from the
4437 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4438 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4439 Then we set the translation of all bits between the starting and
4440 ending characters (inclusive) in the compiled pattern B.
4442 Return an error code.
4444 We use these short variable names so we can use the same macros as
4445 `regex_compile' itself. */
4447 static reg_errcode_t
4448 byte_compile_range (unsigned int range_start_char
, const char **p_ptr
,
4449 const char *pend
, RE_TRANSLATE_TYPE translate
,
4450 reg_syntax_t syntax
, unsigned char *b
)
4453 const char *p
= *p_ptr
;
4456 const unsigned char *collseq
;
4457 unsigned int start_colseq
;
4458 unsigned int end_colseq
;
4466 /* Have to increment the pointer into the pattern string, so the
4467 caller isn't still at the ending character. */
4470 /* Report an error if the range is empty and the syntax prohibits this. */
4471 ret
= syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
4474 collseq
= (const unsigned char *) _NL_CURRENT (LC_COLLATE
,
4475 _NL_COLLATE_COLLSEQMB
);
4477 start_colseq
= collseq
[(unsigned char) TRANSLATE (range_start_char
)];
4478 end_colseq
= collseq
[(unsigned char) TRANSLATE (p
[0])];
4479 for (this_char
= 0; this_char
<= (unsigned char) -1; ++this_char
)
4481 unsigned int this_colseq
= collseq
[(unsigned char) TRANSLATE (this_char
)];
4483 if (start_colseq
<= this_colseq
&& this_colseq
<= end_colseq
)
4485 SET_LIST_BIT (TRANSLATE (this_char
));
4490 /* Here we see why `this_char' has to be larger than an `unsigned
4491 char' -- we would otherwise go into an infinite loop, since all
4492 characters <= 0xff. */
4493 range_start_char
= TRANSLATE (range_start_char
);
4494 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4495 and some compilers cast it to int implicitly, so following for_loop
4496 may fall to (almost) infinite loop.
4497 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4498 To avoid this, we cast p[0] to unsigned int and truncate it. */
4499 end_char
= ((unsigned)TRANSLATE(p
[0]) & ((1 << BYTEWIDTH
) - 1));
4501 for (this_char
= range_start_char
; this_char
<= end_char
; ++this_char
)
4503 SET_LIST_BIT (TRANSLATE (this_char
));
4512 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4513 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4514 characters can start a string that matches the pattern. This fastmap
4515 is used by re_search to skip quickly over impossible starting points.
4517 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4518 area as BUFP->fastmap.
4520 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4523 Returns 0 if we succeed, -2 if an internal error. */
4526 /* local function for re_compile_fastmap.
4527 truncate wchar_t character to char. */
4528 static unsigned char truncate_wchar (CHAR_T c
);
4530 static unsigned char
4531 truncate_wchar (CHAR_T c
)
4533 unsigned char buf
[MB_CUR_MAX
];
4536 memset (&state
, '\0', sizeof (state
));
4538 retval
= __wcrtomb (buf
, c
, &state
);
4540 retval
= wcrtomb (buf
, c
, &state
);
4542 return retval
> 0 ? buf
[0] : (unsigned char) c
;
4547 PREFIX(re_compile_fastmap
) (struct re_pattern_buffer
*bufp
)
4550 #ifdef MATCH_MAY_ALLOCATE
4551 PREFIX(fail_stack_type
) fail_stack
;
4553 #ifndef REGEX_MALLOC
4557 register char *fastmap
= bufp
->fastmap
;
4560 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4561 pattern to (char*) in regex_compile. */
4562 UCHAR_T
*pattern
= (UCHAR_T
*)bufp
->buffer
;
4563 register UCHAR_T
*pend
= (UCHAR_T
*) (bufp
->buffer
+ bufp
->used
);
4565 UCHAR_T
*pattern
= bufp
->buffer
;
4566 register UCHAR_T
*pend
= pattern
+ bufp
->used
;
4568 UCHAR_T
*p
= pattern
;
4571 /* This holds the pointer to the failure stack, when
4572 it is allocated relocatably. */
4573 fail_stack_elt_t
*failure_stack_ptr
;
4576 /* Assume that each path through the pattern can be null until
4577 proven otherwise. We set this false at the bottom of switch
4578 statement, to which we get only if a particular path doesn't
4579 match the empty string. */
4580 boolean path_can_be_null
= true;
4582 /* We aren't doing a `succeed_n' to begin with. */
4583 boolean succeed_n_p
= false;
4585 assert (fastmap
!= NULL
&& p
!= NULL
);
4588 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4589 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4590 bufp
->can_be_null
= 0;
4594 if (p
== pend
|| *p
== (UCHAR_T
) succeed
)
4596 /* We have reached the (effective) end of pattern. */
4597 if (!FAIL_STACK_EMPTY ())
4599 bufp
->can_be_null
|= path_can_be_null
;
4601 /* Reset for next path. */
4602 path_can_be_null
= true;
4604 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
4612 /* We should never be about to go beyond the end of the pattern. */
4615 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4618 /* I guess the idea here is to simply not bother with a fastmap
4619 if a backreference is used, since it's too hard to figure out
4620 the fastmap for the corresponding group. Setting
4621 `can_be_null' stops `re_search_2' from using the fastmap, so
4622 that is all we do. */
4624 bufp
->can_be_null
= 1;
4628 /* Following are the cases which match a character. These end
4633 fastmap
[truncate_wchar(p
[1])] = 1;
4647 /* It is hard to distinguish fastmap from (multi byte) characters
4648 which depends on current locale. */
4653 bufp
->can_be_null
= 1;
4657 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
4658 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
4664 /* Chars beyond end of map must be allowed. */
4665 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
4668 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
4669 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
4675 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4676 if (SYNTAX (j
) == Sword
)
4682 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4683 if (SYNTAX (j
) != Sword
)
4690 int fastmap_newline
= fastmap
['\n'];
4692 /* `.' matches anything ... */
4693 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4696 /* ... except perhaps newline. */
4697 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
4698 fastmap
['\n'] = fastmap_newline
;
4700 /* Return if we have already set `can_be_null'; if we have,
4701 then the fastmap is irrelevant. Something's wrong here. */
4702 else if (bufp
->can_be_null
)
4705 /* Otherwise, have to check alternative paths. */
4712 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4713 if (SYNTAX (j
) == (enum syntaxcode
) k
)
4720 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4721 if (SYNTAX (j
) != (enum syntaxcode
) k
)
4726 /* All cases after this match the empty string. These end with
4746 case push_dummy_failure
:
4751 case pop_failure_jump
:
4752 case maybe_pop_jump
:
4755 case dummy_failure_jump
:
4756 EXTRACT_NUMBER_AND_INCR (j
, p
);
4761 /* Jump backward implies we just went through the body of a
4762 loop and matched nothing. Opcode jumped to should be
4763 `on_failure_jump' or `succeed_n'. Just treat it like an
4764 ordinary jump. For a * loop, it has pushed its failure
4765 point already; if so, discard that as redundant. */
4766 if ((re_opcode_t
) *p
!= on_failure_jump
4767 && (re_opcode_t
) *p
!= succeed_n
)
4771 EXTRACT_NUMBER_AND_INCR (j
, p
);
4774 /* If what's on the stack is where we are now, pop it. */
4775 if (!FAIL_STACK_EMPTY ()
4776 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
4782 case on_failure_jump
:
4783 case on_failure_keep_string_jump
:
4784 handle_on_failure_jump
:
4785 EXTRACT_NUMBER_AND_INCR (j
, p
);
4787 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4788 end of the pattern. We don't want to push such a point,
4789 since when we restore it above, entering the switch will
4790 increment `p' past the end of the pattern. We don't need
4791 to push such a point since we obviously won't find any more
4792 fastmap entries beyond `pend'. Such a pattern can match
4793 the null string, though. */
4796 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
4798 RESET_FAIL_STACK ();
4803 bufp
->can_be_null
= 1;
4807 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
4808 succeed_n_p
= false;
4815 /* Get to the number of times to succeed. */
4816 p
+= OFFSET_ADDRESS_SIZE
;
4818 /* Increment p past the n for when k != 0. */
4819 EXTRACT_NUMBER_AND_INCR (k
, p
);
4822 p
-= 2 * OFFSET_ADDRESS_SIZE
;
4823 succeed_n_p
= true; /* Spaghetti code alert. */
4824 goto handle_on_failure_jump
;
4830 p
+= 2 * OFFSET_ADDRESS_SIZE
;
4841 abort (); /* We have listed all the cases. */
4844 /* Getting here means we have found the possible starting
4845 characters for one path of the pattern -- and that the empty
4846 string does not match. We need not follow this path further.
4847 Instead, look at the next alternative (remembered on the
4848 stack), or quit if no more. The test at the top of the loop
4849 does these things. */
4850 path_can_be_null
= false;
4854 /* Set `can_be_null' for the last path (also the first path, if the
4855 pattern is empty). */
4856 bufp
->can_be_null
|= path_can_be_null
;
4859 RESET_FAIL_STACK ();
4863 #else /* not INSIDE_RECURSION */
4866 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4869 if (MB_CUR_MAX
!= 1)
4870 return wcs_re_compile_fastmap(bufp
);
4873 return byte_re_compile_fastmap(bufp
);
4874 } /* re_compile_fastmap */
4876 weak_alias (__re_compile_fastmap
, re_compile_fastmap
)
4880 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4881 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4882 this memory for recording register information. STARTS and ENDS
4883 must be allocated using the malloc library routine, and must each
4884 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4886 If NUM_REGS == 0, then subsequent matches should allocate their own
4889 Unless this function is called, the first search or match using
4890 PATTERN_BUFFER will allocate its own register data, without
4891 freeing the old data. */
4894 re_set_registers (struct re_pattern_buffer
*bufp
,
4895 struct re_registers
*regs
, unsigned num_regs
,
4896 regoff_t
*starts
, regoff_t
*ends
)
4900 bufp
->regs_allocated
= REGS_REALLOCATE
;
4901 regs
->num_regs
= num_regs
;
4902 regs
->start
= starts
;
4907 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4909 regs
->start
= regs
->end
= (regoff_t
*) 0;
4913 weak_alias (__re_set_registers
, re_set_registers
)
4916 /* Searching routines. */
4918 /* Like re_search_2, below, but only one string is specified, and
4919 doesn't let you say where to stop matching. */
4922 re_search (struct re_pattern_buffer
*bufp
, const char *string
, int size
,
4923 int startpos
, int range
, struct re_registers
*regs
)
4925 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4929 weak_alias (__re_search
, re_search
)
4933 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4934 virtual concatenation of STRING1 and STRING2, starting first at index
4935 STARTPOS, then at STARTPOS + 1, and so on.
4937 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4939 RANGE is how far to scan while trying to match. RANGE = 0 means try
4940 only at STARTPOS; in general, the last start tried is STARTPOS +
4943 In REGS, return the indices of the virtual concatenation of STRING1
4944 and STRING2 that matched the entire BUFP->buffer and its contained
4947 Do not consider matching one past the index STOP in the virtual
4948 concatenation of STRING1 and STRING2.
4950 We return either the position in the strings at which the match was
4951 found, -1 if no match, or -2 if error (such as failure
4955 re_search_2 (struct re_pattern_buffer
*bufp
, const char *string1
, int size1
,
4956 const char *string2
, int size2
, int startpos
, int range
,
4957 struct re_registers
*regs
, int stop
)
4960 if (MB_CUR_MAX
!= 1)
4961 return wcs_re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
,
4965 return byte_re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
,
4969 weak_alias (__re_search_2
, re_search_2
)
4972 #endif /* not INSIDE_RECURSION */
4974 #ifdef INSIDE_RECURSION
4976 #ifdef MATCH_MAY_ALLOCATE
4977 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
4979 # define FREE_VAR(var) free (var); var = NULL
4983 # define MAX_ALLOCA_SIZE 2000
4985 # define FREE_WCS_BUFFERS() \
4987 if (size1 > MAX_ALLOCA_SIZE) \
4989 free (wcs_string1); \
4990 free (mbs_offset1); \
4994 FREE_VAR (wcs_string1); \
4995 FREE_VAR (mbs_offset1); \
4997 if (size2 > MAX_ALLOCA_SIZE) \
4999 free (wcs_string2); \
5000 free (mbs_offset2); \
5004 FREE_VAR (wcs_string2); \
5005 FREE_VAR (mbs_offset2); \
5013 PREFIX(re_search_2
) (struct re_pattern_buffer
*bufp
, const char *string1
,
5014 int size1
, const char *string2
, int size2
,
5015 int startpos
, int range
,
5016 struct re_registers
*regs
, int stop
)
5019 register char *fastmap
= bufp
->fastmap
;
5020 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5021 int total_size
= size1
+ size2
;
5022 int endpos
= startpos
+ range
;
5024 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5025 wchar_t *wcs_string1
= NULL
, *wcs_string2
= NULL
;
5026 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5027 int wcs_size1
= 0, wcs_size2
= 0;
5028 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5029 int *mbs_offset1
= NULL
, *mbs_offset2
= NULL
;
5030 /* They hold whether each wchar_t is binary data or not. */
5031 char *is_binary
= NULL
;
5034 /* Check for out-of-range STARTPOS. */
5035 if (startpos
< 0 || startpos
> total_size
)
5038 /* Fix up RANGE if it might eventually take us outside
5039 the virtual concatenation of STRING1 and STRING2.
5040 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5042 range
= 0 - startpos
;
5043 else if (endpos
> total_size
)
5044 range
= total_size
- startpos
;
5046 /* If the search isn't to be a backwards one, don't waste time in a
5047 search for a pattern that must be anchored. */
5048 if (bufp
->used
> 0 && range
> 0
5049 && ((re_opcode_t
) bufp
->buffer
[0] == begbuf
5050 /* `begline' is like `begbuf' if it cannot match at newlines. */
5051 || ((re_opcode_t
) bufp
->buffer
[0] == begline
5052 && !bufp
->newline_anchor
)))
5061 /* In a forward search for something that starts with \=.
5062 don't keep searching past point. */
5063 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
5065 range
= PT
- startpos
;
5071 /* Update the fastmap now if not correct already. */
5072 if (fastmap
&& !bufp
->fastmap_accurate
)
5073 if (re_compile_fastmap (bufp
) == -2)
5077 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5078 fill them with converted string. */
5081 if (size1
> MAX_ALLOCA_SIZE
)
5083 wcs_string1
= TALLOC (size1
+ 1, CHAR_T
);
5084 mbs_offset1
= TALLOC (size1
+ 1, int);
5085 is_binary
= TALLOC (size1
+ 1, char);
5089 wcs_string1
= REGEX_TALLOC (size1
+ 1, CHAR_T
);
5090 mbs_offset1
= REGEX_TALLOC (size1
+ 1, int);
5091 is_binary
= REGEX_TALLOC (size1
+ 1, char);
5093 if (!wcs_string1
|| !mbs_offset1
|| !is_binary
)
5095 if (size1
> MAX_ALLOCA_SIZE
)
5103 FREE_VAR (wcs_string1
);
5104 FREE_VAR (mbs_offset1
);
5105 FREE_VAR (is_binary
);
5109 wcs_size1
= convert_mbs_to_wcs(wcs_string1
, string1
, size1
,
5110 mbs_offset1
, is_binary
);
5111 wcs_string1
[wcs_size1
] = L
'\0'; /* for a sentinel */
5112 if (size1
> MAX_ALLOCA_SIZE
)
5115 FREE_VAR (is_binary
);
5119 if (size2
> MAX_ALLOCA_SIZE
)
5121 wcs_string2
= TALLOC (size2
+ 1, CHAR_T
);
5122 mbs_offset2
= TALLOC (size2
+ 1, int);
5123 is_binary
= TALLOC (size2
+ 1, char);
5127 wcs_string2
= REGEX_TALLOC (size2
+ 1, CHAR_T
);
5128 mbs_offset2
= REGEX_TALLOC (size2
+ 1, int);
5129 is_binary
= REGEX_TALLOC (size2
+ 1, char);
5131 if (!wcs_string2
|| !mbs_offset2
|| !is_binary
)
5133 FREE_WCS_BUFFERS ();
5134 if (size2
> MAX_ALLOCA_SIZE
)
5137 FREE_VAR (is_binary
);
5140 wcs_size2
= convert_mbs_to_wcs(wcs_string2
, string2
, size2
,
5141 mbs_offset2
, is_binary
);
5142 wcs_string2
[wcs_size2
] = L
'\0'; /* for a sentinel */
5143 if (size2
> MAX_ALLOCA_SIZE
)
5146 FREE_VAR (is_binary
);
5151 /* Loop through the string, looking for a place to start matching. */
5154 /* If a fastmap is supplied, skip quickly over characters that
5155 cannot be the start of a match. If the pattern can match the
5156 null string, however, we don't need to skip characters; we want
5157 the first null string. */
5158 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
5160 if (range
> 0) /* Searching forwards. */
5162 register const char *d
;
5163 register int lim
= 0;
5166 if (startpos
< size1
&& startpos
+ range
>= size1
)
5167 lim
= range
- (size1
- startpos
);
5169 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
5171 /* Written out as an if-else to avoid testing `translate'
5175 && !fastmap
[(unsigned char)
5176 translate
[(unsigned char) *d
++]])
5179 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
5182 startpos
+= irange
- range
;
5184 else /* Searching backwards. */
5186 register CHAR_T c
= (size1
== 0 || startpos
>= size1
5187 ? string2
[startpos
- size1
]
5188 : string1
[startpos
]);
5190 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
5195 /* If can't match the null string, and that's all we have left, fail. */
5196 if (range
>= 0 && startpos
== total_size
&& fastmap
5197 && !bufp
->can_be_null
)
5200 FREE_WCS_BUFFERS ();
5206 val
= wcs_re_match_2_internal (bufp
, string1
, size1
, string2
,
5207 size2
, startpos
, regs
, stop
,
5208 wcs_string1
, wcs_size1
,
5209 wcs_string2
, wcs_size2
,
5210 mbs_offset1
, mbs_offset2
);
5212 val
= byte_re_match_2_internal (bufp
, string1
, size1
, string2
,
5213 size2
, startpos
, regs
, stop
);
5216 #ifndef REGEX_MALLOC
5225 FREE_WCS_BUFFERS ();
5233 FREE_WCS_BUFFERS ();
5253 FREE_WCS_BUFFERS ();
5259 /* This converts PTR, a pointer into one of the search wchar_t strings
5260 `string1' and `string2' into an multibyte string offset from the
5261 beginning of that string. We use mbs_offset to optimize.
5262 See convert_mbs_to_wcs. */
5263 # define POINTER_TO_OFFSET(ptr) \
5264 (FIRST_STRING_P (ptr) \
5265 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5266 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5269 /* This converts PTR, a pointer into one of the search strings `string1'
5270 and `string2' into an offset from the beginning of that string. */
5271 # define POINTER_TO_OFFSET(ptr) \
5272 (FIRST_STRING_P (ptr) \
5273 ? ((regoff_t) ((ptr) - string1)) \
5274 : ((regoff_t) ((ptr) - string2 + size1)))
5277 /* Macros for dealing with the split strings in re_match_2. */
5279 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5281 /* Call before fetching a character with *d. This switches over to
5282 string2 if necessary. */
5283 #define PREFETCH() \
5286 /* End of string2 => fail. */ \
5287 if (dend == end_match_2) \
5289 /* End of string1 => advance to string2. */ \
5291 dend = end_match_2; \
5294 /* Test if at very beginning or at very end of the virtual concatenation
5295 of `string1' and `string2'. If only one string, it's `string2'. */
5296 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5297 #define AT_STRINGS_END(d) ((d) == end2)
5300 /* Test if D points to a character which is word-constituent. We have
5301 two special cases to check for: if past the end of string1, look at
5302 the first character in string2; and if before the beginning of
5303 string2, look at the last character in string1. */
5305 /* Use internationalized API instead of SYNTAX. */
5306 # define WORDCHAR_P(d) \
5307 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5308 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5309 || ((d) == end1 ? *string2 \
5310 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5312 # define WORDCHAR_P(d) \
5313 (SYNTAX ((d) == end1 ? *string2 \
5314 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5318 /* Disabled due to a compiler bug -- see comment at case wordbound */
5320 /* Test if the character before D and the one at D differ with respect
5321 to being word-constituent. */
5322 #define AT_WORD_BOUNDARY(d) \
5323 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5324 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5327 /* Free everything we malloc. */
5328 #ifdef MATCH_MAY_ALLOCATE
5330 # define FREE_VARIABLES() \
5332 REGEX_FREE_STACK (fail_stack.stack); \
5333 FREE_VAR (regstart); \
5334 FREE_VAR (regend); \
5335 FREE_VAR (old_regstart); \
5336 FREE_VAR (old_regend); \
5337 FREE_VAR (best_regstart); \
5338 FREE_VAR (best_regend); \
5339 FREE_VAR (reg_info); \
5340 FREE_VAR (reg_dummy); \
5341 FREE_VAR (reg_info_dummy); \
5342 if (!cant_free_wcs_buf) \
5344 FREE_VAR (string1); \
5345 FREE_VAR (string2); \
5346 FREE_VAR (mbs_offset1); \
5347 FREE_VAR (mbs_offset2); \
5351 # define FREE_VARIABLES() \
5353 REGEX_FREE_STACK (fail_stack.stack); \
5354 FREE_VAR (regstart); \
5355 FREE_VAR (regend); \
5356 FREE_VAR (old_regstart); \
5357 FREE_VAR (old_regend); \
5358 FREE_VAR (best_regstart); \
5359 FREE_VAR (best_regend); \
5360 FREE_VAR (reg_info); \
5361 FREE_VAR (reg_dummy); \
5362 FREE_VAR (reg_info_dummy); \
5367 # define FREE_VARIABLES() \
5369 if (!cant_free_wcs_buf) \
5371 FREE_VAR (string1); \
5372 FREE_VAR (string2); \
5373 FREE_VAR (mbs_offset1); \
5374 FREE_VAR (mbs_offset2); \
5378 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5380 #endif /* not MATCH_MAY_ALLOCATE */
5382 /* These values must meet several constraints. They must not be valid
5383 register values; since we have a limit of 255 registers (because
5384 we use only one byte in the pattern for the register number), we can
5385 use numbers larger than 255. They must differ by 1, because of
5386 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5387 be larger than the value for the highest register, so we do not try
5388 to actually save any registers when none are active. */
5389 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5390 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5392 #else /* not INSIDE_RECURSION */
5393 /* Matching routines. */
5395 #ifndef emacs /* Emacs never uses this. */
5396 /* re_match is like re_match_2 except it takes only a single string. */
5399 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
5400 int size
, int pos
, struct re_registers
*regs
)
5404 if (MB_CUR_MAX
!= 1)
5405 result
= wcs_re_match_2_internal (bufp
, NULL
, 0, string
, size
,
5407 NULL
, 0, NULL
, 0, NULL
, NULL
);
5410 result
= byte_re_match_2_internal (bufp
, NULL
, 0, string
, size
,
5412 # ifndef REGEX_MALLOC
5420 weak_alias (__re_match
, re_match
)
5422 #endif /* not emacs */
5424 #endif /* not INSIDE_RECURSION */
5426 #ifdef INSIDE_RECURSION
5427 static boolean
PREFIX(group_match_null_string_p
) (UCHAR_T
**p
,
5429 PREFIX(register_info_type
) *reg_info
);
5430 static boolean
PREFIX(alt_match_null_string_p
) (UCHAR_T
*p
,
5432 PREFIX(register_info_type
) *reg_info
);
5433 static boolean
PREFIX(common_op_match_null_string_p
) (UCHAR_T
**p
,
5435 PREFIX(register_info_type
) *reg_info
);
5436 static int PREFIX(bcmp_translate
) (const CHAR_T
*s1
, const CHAR_T
*s2
,
5437 int len
, char *translate
);
5438 #else /* not INSIDE_RECURSION */
5440 /* re_match_2 matches the compiled pattern in BUFP against the
5441 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5442 and SIZE2, respectively). We start matching at POS, and stop
5445 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5446 store offsets for the substring each group matched in REGS. See the
5447 documentation for exactly how many groups we fill.
5449 We return -1 if no match, -2 if an internal error (such as the
5450 failure stack overflowing). Otherwise, we return the length of the
5451 matched substring. */
5454 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
, int size1
,
5455 const char *string2
, int size2
, int pos
,
5456 struct re_registers
*regs
, int stop
)
5460 if (MB_CUR_MAX
!= 1)
5461 result
= wcs_re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
5463 NULL
, 0, NULL
, 0, NULL
, NULL
);
5466 result
= byte_re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
5469 #ifndef REGEX_MALLOC
5477 weak_alias (__re_match_2
, re_match_2
)
5480 #endif /* not INSIDE_RECURSION */
5482 #ifdef INSIDE_RECURSION
5485 static int count_mbs_length (int *, int);
5487 /* This check the substring (from 0, to length) of the multibyte string,
5488 to which offset_buffer correspond. And count how many wchar_t_characters
5489 the substring occupy. We use offset_buffer to optimization.
5490 See convert_mbs_to_wcs. */
5493 count_mbs_length(int *offset_buffer
, int length
)
5497 /* Check whether the size is valid. */
5501 if (offset_buffer
== NULL
)
5504 /* If there are no multibyte character, offset_buffer[i] == i.
5505 Optmize for this case. */
5506 if (offset_buffer
[length
] == length
)
5509 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5515 int middle
= (lower
+ upper
) / 2;
5516 if (middle
== lower
|| middle
== upper
)
5518 if (offset_buffer
[middle
] > length
)
5520 else if (offset_buffer
[middle
] < length
)
5530 /* This is a separate function so that we can force an alloca cleanup
5534 wcs_re_match_2_internal (struct re_pattern_buffer
*bufp
,
5535 const char *cstring1
, int csize1
,
5536 const char *cstring2
, int csize2
,
5538 struct re_registers
*regs
,
5540 /* string1 == string2 == NULL means string1/2, size1/2 and
5541 mbs_offset1/2 need seting up in this function. */
5542 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5543 wchar_t *string1
, int size1
,
5544 wchar_t *string2
, int size2
,
5545 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5546 int *mbs_offset1
, int *mbs_offset2
)
5549 byte_re_match_2_internal (struct re_pattern_buffer
*bufp
,
5550 const char *string1
, int size1
,
5551 const char *string2
, int size2
,
5553 struct re_registers
*regs
, int stop
)
5556 /* General temporaries. */
5560 /* They hold whether each wchar_t is binary data or not. */
5561 char *is_binary
= NULL
;
5562 /* If true, we can't free string1/2, mbs_offset1/2. */
5563 int cant_free_wcs_buf
= 1;
5566 /* Just past the end of the corresponding string. */
5567 const CHAR_T
*end1
, *end2
;
5569 /* Pointers into string1 and string2, just past the last characters in
5570 each to consider matching. */
5571 const CHAR_T
*end_match_1
, *end_match_2
;
5573 /* Where we are in the data, and the end of the current string. */
5574 const CHAR_T
*d
, *dend
;
5576 /* Where we are in the pattern, and the end of the pattern. */
5578 UCHAR_T
*pattern
, *p
;
5579 register UCHAR_T
*pend
;
5581 UCHAR_T
*p
= bufp
->buffer
;
5582 register UCHAR_T
*pend
= p
+ bufp
->used
;
5585 /* Mark the opcode just after a start_memory, so we can test for an
5586 empty subpattern when we get to the stop_memory. */
5587 UCHAR_T
*just_past_start_mem
= 0;
5589 /* We use this to map every character in the string. */
5590 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5592 /* Failure point stack. Each place that can handle a failure further
5593 down the line pushes a failure point on this stack. It consists of
5594 restart, regend, and reg_info for all registers corresponding to
5595 the subexpressions we're currently inside, plus the number of such
5596 registers, and, finally, two char *'s. The first char * is where
5597 to resume scanning the pattern; the second one is where to resume
5598 scanning the strings. If the latter is zero, the failure point is
5599 a ``dummy''; if a failure happens and the failure point is a dummy,
5600 it gets discarded and the next next one is tried. */
5601 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5602 PREFIX(fail_stack_type
) fail_stack
;
5605 static unsigned failure_id
;
5606 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5610 /* This holds the pointer to the failure stack, when
5611 it is allocated relocatably. */
5612 fail_stack_elt_t
*failure_stack_ptr
;
5615 /* We fill all the registers internally, independent of what we
5616 return, for use in backreferences. The number here includes
5617 an element for register zero. */
5618 size_t num_regs
= bufp
->re_nsub
+ 1;
5620 /* The currently active registers. */
5621 active_reg_t lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
5622 active_reg_t highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
5624 /* Information on the contents of registers. These are pointers into
5625 the input strings; they record just what was matched (on this
5626 attempt) by a subexpression part of the pattern, that is, the
5627 regnum-th regstart pointer points to where in the pattern we began
5628 matching and the regnum-th regend points to right after where we
5629 stopped matching the regnum-th subexpression. (The zeroth register
5630 keeps track of what the whole pattern matches.) */
5631 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5632 const CHAR_T
**regstart
, **regend
;
5635 /* If a group that's operated upon by a repetition operator fails to
5636 match anything, then the register for its start will need to be
5637 restored because it will have been set to wherever in the string we
5638 are when we last see its open-group operator. Similarly for a
5640 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5641 const CHAR_T
**old_regstart
, **old_regend
;
5644 /* The is_active field of reg_info helps us keep track of which (possibly
5645 nested) subexpressions we are currently in. The matched_something
5646 field of reg_info[reg_num] helps us tell whether or not we have
5647 matched any of the pattern so far this time through the reg_num-th
5648 subexpression. These two fields get reset each time through any
5649 loop their register is in. */
5650 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5651 PREFIX(register_info_type
) *reg_info
;
5654 /* The following record the register info as found in the above
5655 variables when we find a match better than any we've seen before.
5656 This happens as we backtrack through the failure points, which in
5657 turn happens only if we have not yet matched the entire string. */
5658 unsigned best_regs_set
= false;
5659 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5660 const CHAR_T
**best_regstart
, **best_regend
;
5663 /* Logically, this is `best_regend[0]'. But we don't want to have to
5664 allocate space for that if we're not allocating space for anything
5665 else (see below). Also, we never need info about register 0 for
5666 any of the other register vectors, and it seems rather a kludge to
5667 treat `best_regend' differently than the rest. So we keep track of
5668 the end of the best match so far in a separate variable. We
5669 initialize this to NULL so that when we backtrack the first time
5670 and need to test it, it's not garbage. */
5671 const CHAR_T
*match_end
= NULL
;
5673 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5674 int set_regs_matched_done
= 0;
5676 /* Used when we pop values we don't care about. */
5677 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5678 const CHAR_T
**reg_dummy
;
5679 PREFIX(register_info_type
) *reg_info_dummy
;
5683 /* Counts the total number of registers pushed. */
5684 unsigned num_regs_pushed
= 0;
5687 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5691 #ifdef MATCH_MAY_ALLOCATE
5692 /* Do not bother to initialize all the register variables if there are
5693 no groups in the pattern, as it takes a fair amount of time. If
5694 there are groups, we include space for register 0 (the whole
5695 pattern), even though we never use it, since it simplifies the
5696 array indexing. We should fix this. */
5699 regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5700 regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5701 old_regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5702 old_regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5703 best_regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5704 best_regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5705 reg_info
= REGEX_TALLOC (num_regs
, PREFIX(register_info_type
));
5706 reg_dummy
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5707 reg_info_dummy
= REGEX_TALLOC (num_regs
, PREFIX(register_info_type
));
5709 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
5710 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
5718 /* We must initialize all our variables to NULL, so that
5719 `FREE_VARIABLES' doesn't try to free them. */
5720 regstart
= regend
= old_regstart
= old_regend
= best_regstart
5721 = best_regend
= reg_dummy
= NULL
;
5722 reg_info
= reg_info_dummy
= (PREFIX(register_info_type
) *) NULL
;
5724 #endif /* MATCH_MAY_ALLOCATE */
5726 /* The starting position is bogus. */
5728 if (pos
< 0 || pos
> csize1
+ csize2
)
5730 if (pos
< 0 || pos
> size1
+ size2
)
5738 /* Allocate wchar_t array for string1 and string2 and
5739 fill them with converted string. */
5740 if (string1
== NULL
&& string2
== NULL
)
5742 /* We need seting up buffers here. */
5744 /* We must free wcs buffers in this function. */
5745 cant_free_wcs_buf
= 0;
5749 string1
= REGEX_TALLOC (csize1
+ 1, CHAR_T
);
5750 mbs_offset1
= REGEX_TALLOC (csize1
+ 1, int);
5751 is_binary
= REGEX_TALLOC (csize1
+ 1, char);
5752 if (!string1
|| !mbs_offset1
|| !is_binary
)
5755 FREE_VAR (mbs_offset1
);
5756 FREE_VAR (is_binary
);
5762 string2
= REGEX_TALLOC (csize2
+ 1, CHAR_T
);
5763 mbs_offset2
= REGEX_TALLOC (csize2
+ 1, int);
5764 is_binary
= REGEX_TALLOC (csize2
+ 1, char);
5765 if (!string2
|| !mbs_offset2
|| !is_binary
)
5768 FREE_VAR (mbs_offset1
);
5770 FREE_VAR (mbs_offset2
);
5771 FREE_VAR (is_binary
);
5774 size2
= convert_mbs_to_wcs(string2
, cstring2
, csize2
,
5775 mbs_offset2
, is_binary
);
5776 string2
[size2
] = L
'\0'; /* for a sentinel */
5777 FREE_VAR (is_binary
);
5781 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5782 pattern to (char*) in regex_compile. */
5783 p
= pattern
= (CHAR_T
*)bufp
->buffer
;
5784 pend
= (CHAR_T
*)(bufp
->buffer
+ bufp
->used
);
5788 /* Initialize subexpression text positions to -1 to mark ones that no
5789 start_memory/stop_memory has been seen for. Also initialize the
5790 register information struct. */
5791 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
5793 regstart
[mcnt
] = regend
[mcnt
]
5794 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
5796 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
5797 IS_ACTIVE (reg_info
[mcnt
]) = 0;
5798 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
5799 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
5802 /* We move `string1' into `string2' if the latter's empty -- but not if
5803 `string1' is null. */
5804 if (size2
== 0 && string1
!= NULL
)
5811 mbs_offset2
= mbs_offset1
;
5817 end1
= string1
+ size1
;
5818 end2
= string2
+ size2
;
5820 /* Compute where to stop matching, within the two strings. */
5824 mcnt
= count_mbs_length(mbs_offset1
, stop
);
5825 end_match_1
= string1
+ mcnt
;
5826 end_match_2
= string2
;
5830 if (stop
> csize1
+ csize2
)
5831 stop
= csize1
+ csize2
;
5833 mcnt
= count_mbs_length(mbs_offset2
, stop
-csize1
);
5834 end_match_2
= string2
+ mcnt
;
5837 { /* count_mbs_length return error. */
5844 end_match_1
= string1
+ stop
;
5845 end_match_2
= string2
;
5850 end_match_2
= string2
+ stop
- size1
;
5854 /* `p' scans through the pattern as `d' scans through the data.
5855 `dend' is the end of the input string that `d' points within. `d'
5856 is advanced into the following input string whenever necessary, but
5857 this happens before fetching; therefore, at the beginning of the
5858 loop, `d' can be pointing at the end of a string, but it cannot
5861 if (size1
> 0 && pos
<= csize1
)
5863 mcnt
= count_mbs_length(mbs_offset1
, pos
);
5869 mcnt
= count_mbs_length(mbs_offset2
, pos
-csize1
);
5875 { /* count_mbs_length return error. */
5880 if (size1
> 0 && pos
<= size1
)
5887 d
= string2
+ pos
- size1
;
5892 DEBUG_PRINT1 ("The compiled pattern is:\n");
5893 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5894 DEBUG_PRINT1 ("The string to match is: `");
5895 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5896 DEBUG_PRINT1 ("'\n");
5898 /* This loops over pattern commands. It exits by returning from the
5899 function if the match is complete, or it drops through if the match
5900 fails at this starting point in the input data. */
5904 DEBUG_PRINT2 ("\n%p: ", p
);
5906 DEBUG_PRINT2 ("\n0x%x: ", p
);
5910 { /* End of pattern means we might have succeeded. */
5911 DEBUG_PRINT1 ("end of pattern ... ");
5913 /* If we haven't matched the entire string, and we want the
5914 longest match, try backtracking. */
5915 if (d
!= end_match_2
)
5917 /* 1 if this match ends in the same string (string1 or string2)
5918 as the best previous match. */
5921 /* 1 if this match is the best seen so far. */
5922 boolean best_match_p
;
5924 same_str_p
= (FIRST_STRING_P (match_end
)
5925 == MATCHING_IN_FIRST_STRING
);
5927 /* AIX compiler got confused when this was combined
5928 with the previous declaration. */
5930 best_match_p
= d
> match_end
;
5932 best_match_p
= !MATCHING_IN_FIRST_STRING
;
5934 DEBUG_PRINT1 ("backtracking.\n");
5936 if (!FAIL_STACK_EMPTY ())
5937 { /* More failure points to try. */
5939 /* If exceeds best match so far, save it. */
5940 if (!best_regs_set
|| best_match_p
)
5942 best_regs_set
= true;
5945 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5947 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
5949 best_regstart
[mcnt
] = regstart
[mcnt
];
5950 best_regend
[mcnt
] = regend
[mcnt
];
5956 /* If no failure points, don't restore garbage. And if
5957 last match is real best match, don't restore second
5959 else if (best_regs_set
&& !best_match_p
)
5962 /* Restore best match. It may happen that `dend ==
5963 end_match_1' while the restored d is in string2.
5964 For example, the pattern `x.*y.*z' against the
5965 strings `x-' and `y-z-', if the two strings are
5966 not consecutive in memory. */
5967 DEBUG_PRINT1 ("Restoring best registers.\n");
5970 dend
= ((d
>= string1
&& d
<= end1
)
5971 ? end_match_1
: end_match_2
);
5973 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
5975 regstart
[mcnt
] = best_regstart
[mcnt
];
5976 regend
[mcnt
] = best_regend
[mcnt
];
5979 } /* d != end_match_2 */
5982 DEBUG_PRINT1 ("Accepting match.\n");
5983 /* If caller wants register contents data back, do it. */
5984 if (regs
&& !bufp
->no_sub
)
5986 /* Have the register data arrays been allocated? */
5987 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5988 { /* No. So allocate them with malloc. We need one
5989 extra element beyond `num_regs' for the `-1' marker
5991 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5992 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5993 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5994 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5999 bufp
->regs_allocated
= REGS_REALLOCATE
;
6001 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
6002 { /* Yes. If we need more elements than were already
6003 allocated, reallocate them. If we need fewer, just
6005 if (regs
->num_regs
< num_regs
+ 1)
6007 regs
->num_regs
= num_regs
+ 1;
6008 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
6009 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
6010 if (regs
->start
== NULL
|| regs
->end
== NULL
)
6019 /* These braces fend off a "empty body in an else-statement"
6020 warning under GCC when assert expands to nothing. */
6021 assert (bufp
->regs_allocated
== REGS_FIXED
);
6024 /* Convert the pointer data in `regstart' and `regend' to
6025 indices. Register zero has to be set differently,
6026 since we haven't kept track of any info for it. */
6027 if (regs
->num_regs
> 0)
6029 regs
->start
[0] = pos
;
6031 if (MATCHING_IN_FIRST_STRING
)
6032 regs
->end
[0] = mbs_offset1
!= NULL
?
6033 mbs_offset1
[d
-string1
] : 0;
6035 regs
->end
[0] = csize1
+ (mbs_offset2
!= NULL
?
6036 mbs_offset2
[d
-string2
] : 0);
6038 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
6039 ? ((regoff_t
) (d
- string1
))
6040 : ((regoff_t
) (d
- string2
+ size1
)));
6044 /* Go through the first `min (num_regs, regs->num_regs)'
6045 registers, since that is all we initialized. */
6046 for (mcnt
= 1; (unsigned) mcnt
< MIN (num_regs
, regs
->num_regs
);
6049 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
6050 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
6054 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
6056 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
6060 /* If the regs structure we return has more elements than
6061 were in the pattern, set the extra elements to -1. If
6062 we (re)allocated the registers, this is the case,
6063 because we always allocate enough to have at least one
6065 for (mcnt
= num_regs
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
6066 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
6067 } /* regs && !bufp->no_sub */
6069 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6070 nfailure_points_pushed
, nfailure_points_popped
,
6071 nfailure_points_pushed
- nfailure_points_popped
);
6072 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
6075 if (MATCHING_IN_FIRST_STRING
)
6076 mcnt
= mbs_offset1
!= NULL
? mbs_offset1
[d
-string1
] : 0;
6078 mcnt
= (mbs_offset2
!= NULL
? mbs_offset2
[d
-string2
] : 0) +
6082 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
6087 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
6093 /* Otherwise match next pattern command. */
6094 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
6096 /* Ignore these. Used to ignore the n of succeed_n's which
6097 currently have n == 0. */
6099 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6103 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6106 /* Match the next n pattern characters exactly. The following
6107 byte in the pattern defines n, and the n bytes after that
6108 are the characters to match. */
6114 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
6116 /* This is written out as an if-else so we don't waste time
6117 testing `translate' inside the loop. */
6126 if ((UCHAR_T
) translate
[(unsigned char) *d
++]
6132 if (*d
++ != (CHAR_T
) *p
++)
6136 if ((UCHAR_T
) translate
[(unsigned char) *d
++]
6148 if (*d
++ != (CHAR_T
) *p
++) goto fail
;
6152 SET_REGS_MATCHED ();
6156 /* Match any character except possibly a newline or a null. */
6158 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6162 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
6163 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
6166 SET_REGS_MATCHED ();
6167 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d
);
6177 unsigned int i
, char_class_length
, coll_symbol_length
,
6178 equiv_class_length
, ranges_length
, chars_length
, length
;
6179 CHAR_T
*workp
, *workp2
, *charset_top
;
6180 #define WORK_BUFFER_SIZE 128
6181 CHAR_T str_buf
[WORK_BUFFER_SIZE
];
6186 boolean negate
= (re_opcode_t
) *(p
- 1) == charset_not
;
6188 DEBUG_PRINT2 ("EXECUTING charset%s.\n", negate
? "_not" : "");
6190 c
= TRANSLATE (*d
); /* The character to match. */
6193 nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
6195 charset_top
= p
- 1;
6196 char_class_length
= *p
++;
6197 coll_symbol_length
= *p
++;
6198 equiv_class_length
= *p
++;
6199 ranges_length
= *p
++;
6200 chars_length
= *p
++;
6201 /* p points charset[6], so the address of the next instruction
6202 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6203 where l=length of char_classes, m=length of collating_symbol,
6204 n=equivalence_class, o=length of char_range,
6205 p'=length of character. */
6207 /* Update p to indicate the next instruction. */
6208 p
+= char_class_length
+ coll_symbol_length
+ equiv_class_length
+
6209 2*ranges_length
+ chars_length
;
6211 /* match with char_class? */
6212 for (i
= 0; i
< char_class_length
; i
+= CHAR_CLASS_SIZE
)
6215 uintptr_t alignedp
= ((uintptr_t)workp
6216 + __alignof__(wctype_t) - 1)
6217 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6218 wctype
= *((wctype_t*)alignedp
);
6219 workp
+= CHAR_CLASS_SIZE
;
6221 if (__iswctype((wint_t)c
, wctype
))
6222 goto char_set_matched
;
6224 if (iswctype((wint_t)c
, wctype
))
6225 goto char_set_matched
;
6229 /* match with collating_symbol? */
6233 const unsigned char *extra
= (const unsigned char *)
6234 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
6236 for (workp2
= workp
+ coll_symbol_length
; workp
< workp2
;
6240 wextra
= (int32_t*)(extra
+ *workp
++);
6241 for (i
= 0; i
< *wextra
; ++i
)
6242 if (TRANSLATE(d
[i
]) != wextra
[1 + i
])
6247 /* Update d, however d will be incremented at
6248 char_set_matched:, we decrement d here. */
6250 goto char_set_matched
;
6254 else /* (nrules == 0) */
6256 /* If we can't look up collation data, we use wcscoll
6259 for (workp2
= workp
+ coll_symbol_length
; workp
< workp2
;)
6261 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6263 length
= __wcslen (workp
);
6265 length
= wcslen (workp
);
6268 /* If wcscoll(the collating symbol, whole string) > 0,
6269 any substring of the string never match with the
6270 collating symbol. */
6272 if (__wcscoll (workp
, d
) > 0)
6274 if (wcscoll (workp
, d
) > 0)
6277 workp
+= length
+ 1;
6281 /* First, we compare the collating symbol with
6282 the first character of the string.
6283 If it don't match, we add the next character to
6284 the compare buffer in turn. */
6285 for (i
= 0 ; i
< WORK_BUFFER_SIZE
-1 ; i
++, d
++)
6290 if (dend
== end_match_2
)
6296 /* add next character to the compare buffer. */
6297 str_buf
[i
] = TRANSLATE(*d
);
6298 str_buf
[i
+1] = '\0';
6301 match
= __wcscoll (workp
, str_buf
);
6303 match
= wcscoll (workp
, str_buf
);
6306 goto char_set_matched
;
6309 /* (str_buf > workp) indicate (str_buf + X > workp),
6310 because for all X (str_buf + X > str_buf).
6311 So we don't need continue this loop. */
6314 /* Otherwise(str_buf < workp),
6315 (str_buf+next_character) may equals (workp).
6316 So we continue this loop. */
6321 workp
+= length
+ 1;
6324 /* match with equivalence_class? */
6328 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6329 /* Try to match the equivalence class against
6330 those known to the collate implementation. */
6331 const int32_t *table
;
6332 const int32_t *weights
;
6333 const int32_t *extra
;
6334 const int32_t *indirect
;
6339 /* This #include defines a local function! */
6340 # include <locale/weightwc.h>
6342 table
= (const int32_t *)
6343 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEWC
);
6344 weights
= (const wint_t *)
6345 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTWC
);
6346 extra
= (const wint_t *)
6347 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAWC
);
6348 indirect
= (const int32_t *)
6349 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTWC
);
6351 /* Write 1 collating element to str_buf, and
6355 for (i
= 0 ; idx2
== 0 && i
< WORK_BUFFER_SIZE
- 1; i
++)
6357 cp
= (wint_t*)str_buf
;
6360 if (dend
== end_match_2
)
6365 str_buf
[i
] = TRANSLATE(*(d
+i
));
6366 str_buf
[i
+1] = '\0'; /* sentinel */
6367 idx2
= findidx ((const wint_t**)&cp
);
6370 /* Update d, however d will be incremented at
6371 char_set_matched:, we decrement d here. */
6372 d
= backup_d
+ ((wchar_t*)cp
- (wchar_t*)str_buf
- 1);
6375 if (dend
== end_match_2
)
6384 len
= weights
[idx2
];
6386 for (workp2
= workp
+ equiv_class_length
; workp
< workp2
;
6389 idx
= (int32_t)*workp
;
6390 /* We already checked idx != 0 in regex_compile. */
6392 if (idx2
!= 0 && len
== weights
[idx
])
6395 while (cnt
< len
&& (weights
[idx
+ 1 + cnt
]
6396 == weights
[idx2
+ 1 + cnt
]))
6400 goto char_set_matched
;
6407 else /* (nrules == 0) */
6409 /* If we can't look up collation data, we use wcscoll
6412 for (workp2
= workp
+ equiv_class_length
; workp
< workp2
;)
6414 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6416 length
= __wcslen (workp
);
6418 length
= wcslen (workp
);
6421 /* If wcscoll(the collating symbol, whole string) > 0,
6422 any substring of the string never match with the
6423 collating symbol. */
6425 if (__wcscoll (workp
, d
) > 0)
6427 if (wcscoll (workp
, d
) > 0)
6430 workp
+= length
+ 1;
6434 /* First, we compare the equivalence class with
6435 the first character of the string.
6436 If it don't match, we add the next character to
6437 the compare buffer in turn. */
6438 for (i
= 0 ; i
< WORK_BUFFER_SIZE
- 1 ; i
++, d
++)
6443 if (dend
== end_match_2
)
6449 /* add next character to the compare buffer. */
6450 str_buf
[i
] = TRANSLATE(*d
);
6451 str_buf
[i
+1] = '\0';
6454 match
= __wcscoll (workp
, str_buf
);
6456 match
= wcscoll (workp
, str_buf
);
6460 goto char_set_matched
;
6463 /* (str_buf > workp) indicate (str_buf + X > workp),
6464 because for all X (str_buf + X > str_buf).
6465 So we don't need continue this loop. */
6468 /* Otherwise(str_buf < workp),
6469 (str_buf+next_character) may equals (workp).
6470 So we continue this loop. */
6475 workp
+= length
+ 1;
6479 /* match with char_range? */
6483 uint32_t collseqval
;
6484 const char *collseq
= (const char *)
6485 _NL_CURRENT(LC_COLLATE
, _NL_COLLATE_COLLSEQWC
);
6487 collseqval
= collseq_table_lookup (collseq
, c
);
6489 for (; workp
< p
- chars_length
;)
6491 uint32_t start_val
, end_val
;
6493 /* We already compute the collation sequence value
6494 of the characters (or collating symbols). */
6495 start_val
= (uint32_t) *workp
++; /* range_start */
6496 end_val
= (uint32_t) *workp
++; /* range_end */
6498 if (start_val
<= collseqval
&& collseqval
<= end_val
)
6499 goto char_set_matched
;
6505 /* We set range_start_char at str_buf[0], range_end_char
6506 at str_buf[4], and compared char at str_buf[2]. */
6511 for (; workp
< p
- chars_length
;)
6513 wchar_t *range_start_char
, *range_end_char
;
6515 /* match if (range_start_char <= c <= range_end_char). */
6517 /* If range_start(or end) < 0, we assume -range_start(end)
6518 is the offset of the collating symbol which is specified
6519 as the character of the range start(end). */
6523 range_start_char
= charset_top
- (*workp
++);
6526 str_buf
[0] = *workp
++;
6527 range_start_char
= str_buf
;
6532 range_end_char
= charset_top
- (*workp
++);
6535 str_buf
[4] = *workp
++;
6536 range_end_char
= str_buf
+ 4;
6540 if (__wcscoll (range_start_char
, str_buf
+2) <= 0
6541 && __wcscoll (str_buf
+2, range_end_char
) <= 0)
6543 if (wcscoll (range_start_char
, str_buf
+2) <= 0
6544 && wcscoll (str_buf
+2, range_end_char
) <= 0)
6546 goto char_set_matched
;
6550 /* match with char? */
6551 for (; workp
< p
; workp
++)
6553 goto char_set_matched
;
6558 if (negate
) goto fail
;
6560 /* Cast to `unsigned' instead of `unsigned char' in case the
6561 bit list is a full 32 bytes long. */
6562 if (c
< (unsigned) (*p
* BYTEWIDTH
)
6563 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
6568 if (!negate
) goto fail
;
6569 #undef WORK_BUFFER_SIZE
6571 SET_REGS_MATCHED ();
6577 /* The beginning of a group is represented by start_memory.
6578 The arguments are the register number in the next byte, and the
6579 number of groups inner to this one in the next. The text
6580 matched within the group is recorded (in the internal
6581 registers data structure) under the register number. */
6583 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6584 (long int) *p
, (long int) p
[1]);
6586 /* Find out if this group can match the empty string. */
6587 p1
= p
; /* To send to group_match_null_string_p. */
6589 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
6590 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6591 = PREFIX(group_match_null_string_p
) (&p1
, pend
, reg_info
);
6593 /* Save the position in the string where we were the last time
6594 we were at this open-group operator in case the group is
6595 operated upon by a repetition operator, e.g., with `(a*)*b'
6596 against `ab'; then we want to ignore where we are now in
6597 the string in case this attempt to match fails. */
6598 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6599 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
6601 DEBUG_PRINT2 (" old_regstart: %d\n",
6602 POINTER_TO_OFFSET (old_regstart
[*p
]));
6605 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
6607 IS_ACTIVE (reg_info
[*p
]) = 1;
6608 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
6610 /* Clear this whenever we change the register activity status. */
6611 set_regs_matched_done
= 0;
6613 /* This is the new highest active register. */
6614 highest_active_reg
= *p
;
6616 /* If nothing was active before, this is the new lowest active
6618 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
6619 lowest_active_reg
= *p
;
6621 /* Move past the register number and inner group count. */
6623 just_past_start_mem
= p
;
6628 /* The stop_memory opcode represents the end of a group. Its
6629 arguments are the same as start_memory's: the register
6630 number, and the number of inner groups. */
6632 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6633 (long int) *p
, (long int) p
[1]);
6635 /* We need to save the string position the last time we were at
6636 this close-group operator in case the group is operated
6637 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6638 against `aba'; then we want to ignore where we are now in
6639 the string in case this attempt to match fails. */
6640 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6641 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
6643 DEBUG_PRINT2 (" old_regend: %d\n",
6644 POINTER_TO_OFFSET (old_regend
[*p
]));
6647 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
6649 /* This register isn't active anymore. */
6650 IS_ACTIVE (reg_info
[*p
]) = 0;
6652 /* Clear this whenever we change the register activity status. */
6653 set_regs_matched_done
= 0;
6655 /* If this was the only register active, nothing is active
6657 if (lowest_active_reg
== highest_active_reg
)
6659 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
6660 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
6663 { /* We must scan for the new highest active register, since
6664 it isn't necessarily one less than now: consider
6665 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6666 new highest active register is 1. */
6668 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
6671 /* If we end up at register zero, that means that we saved
6672 the registers as the result of an `on_failure_jump', not
6673 a `start_memory', and we jumped to past the innermost
6674 `stop_memory'. For example, in ((.)*) we save
6675 registers 1 and 2 as a result of the *, but when we pop
6676 back to the second ), we are at the stop_memory 1.
6677 Thus, nothing is active. */
6680 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
6681 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
6684 highest_active_reg
= r
;
6687 /* If just failed to match something this time around with a
6688 group that's operated on by a repetition operator, try to
6689 force exit from the ``loop'', and restore the register
6690 information for this group that we had before trying this
6692 if ((!MATCHED_SOMETHING (reg_info
[*p
])
6693 || just_past_start_mem
== p
- 1)
6696 boolean is_a_jump_n
= false;
6700 switch ((re_opcode_t
) *p1
++)
6705 case pop_failure_jump
:
6706 case maybe_pop_jump
:
6708 case dummy_failure_jump
:
6709 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
6711 p1
+= OFFSET_ADDRESS_SIZE
;
6719 /* If the next operation is a jump backwards in the pattern
6720 to an on_failure_jump right before the start_memory
6721 corresponding to this stop_memory, exit from the loop
6722 by forcing a failure after pushing on the stack the
6723 on_failure_jump's jump in the pattern, and d. */
6724 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
6725 && (re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == start_memory
6726 && p1
[2+OFFSET_ADDRESS_SIZE
] == *p
)
6728 /* If this group ever matched anything, then restore
6729 what its registers were before trying this last
6730 failed match, e.g., with `(a*)*b' against `ab' for
6731 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6732 against `aba' for regend[3].
6734 Also restore the registers for inner groups for,
6735 e.g., `((a*)(b*))*' against `aba' (register 3 would
6736 otherwise get trashed). */
6738 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
6742 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
6744 /* Restore this and inner groups' (if any) registers. */
6745 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
6748 regstart
[r
] = old_regstart
[r
];
6750 /* xx why this test? */
6751 if (old_regend
[r
] >= regstart
[r
])
6752 regend
[r
] = old_regend
[r
];
6756 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
6757 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
6763 /* Move past the register number and the inner group count. */
6768 /* \<digit> has been turned into a `duplicate' command which is
6769 followed by the numeric value of <digit> as the register number. */
6772 register const CHAR_T
*d2
, *dend2
;
6773 int regno
= *p
++; /* Get which register to match against. */
6774 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
6776 /* Can't back reference a group which we've never matched. */
6777 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
6780 /* Where in input to try to start matching. */
6781 d2
= regstart
[regno
];
6783 /* Where to stop matching; if both the place to start and
6784 the place to stop matching are in the same string, then
6785 set to the place to stop, otherwise, for now have to use
6786 the end of the first string. */
6788 dend2
= ((FIRST_STRING_P (regstart
[regno
])
6789 == FIRST_STRING_P (regend
[regno
]))
6790 ? regend
[regno
] : end_match_1
);
6793 /* If necessary, advance to next segment in register
6797 if (dend2
== end_match_2
) break;
6798 if (dend2
== regend
[regno
]) break;
6800 /* End of string1 => advance to string2. */
6802 dend2
= regend
[regno
];
6804 /* At end of register contents => success */
6805 if (d2
== dend2
) break;
6807 /* If necessary, advance to next segment in data. */
6810 /* How many characters left in this segment to match. */
6813 /* Want how many consecutive characters we can match in
6814 one shot, so, if necessary, adjust the count. */
6815 if (mcnt
> dend2
- d2
)
6818 /* Compare that many; failure if mismatch, else move
6821 ? PREFIX(bcmp_translate
) (d
, d2
, mcnt
, translate
)
6822 : memcmp (d
, d2
, mcnt
*sizeof(UCHAR_T
)))
6824 d
+= mcnt
, d2
+= mcnt
;
6826 /* Do this because we've match some characters. */
6827 SET_REGS_MATCHED ();
6833 /* begline matches the empty string at the beginning of the string
6834 (unless `not_bol' is set in `bufp'), and, if
6835 `newline_anchor' is set, after newlines. */
6837 DEBUG_PRINT1 ("EXECUTING begline.\n");
6839 if (AT_STRINGS_BEG (d
))
6841 if (!bufp
->not_bol
) break;
6843 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
6847 /* In all other cases, we fail. */
6851 /* endline is the dual of begline. */
6853 DEBUG_PRINT1 ("EXECUTING endline.\n");
6855 if (AT_STRINGS_END (d
))
6857 if (!bufp
->not_eol
) break;
6860 /* We have to ``prefetch'' the next character. */
6861 else if ((d
== end1
? *string2
: *d
) == '\n'
6862 && bufp
->newline_anchor
)
6869 /* Match at the very beginning of the data. */
6871 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
6872 if (AT_STRINGS_BEG (d
))
6877 /* Match at the very end of the data. */
6879 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
6880 if (AT_STRINGS_END (d
))
6885 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
6886 pushes NULL as the value for the string on the stack. Then
6887 `pop_failure_point' will keep the current value for the
6888 string, instead of restoring it. To see why, consider
6889 matching `foo\nbar' against `.*\n'. The .* matches the foo;
6890 then the . fails against the \n. But the next thing we want
6891 to do is match the \n against the \n; if we restored the
6892 string value, we would be back at the foo.
6894 Because this is used only in specific cases, we don't need to
6895 check all the things that `on_failure_jump' does, to make
6896 sure the right things get saved on the stack. Hence we don't
6897 share its code. The only reason to push anything on the
6898 stack at all is that otherwise we would have to change
6899 `anychar's code to do something besides goto fail in this
6900 case; that seems worse than this. */
6901 case on_failure_keep_string_jump
:
6902 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
6904 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6906 DEBUG_PRINT3 (" %d (to %p):\n", mcnt
, p
+ mcnt
);
6908 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
6911 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
6915 /* Uses of on_failure_jump:
6917 Each alternative starts with an on_failure_jump that points
6918 to the beginning of the next alternative. Each alternative
6919 except the last ends with a jump that in effect jumps past
6920 the rest of the alternatives. (They really jump to the
6921 ending jump of the following alternative, because tensioning
6922 these jumps is a hassle.)
6924 Repeats start with an on_failure_jump that points past both
6925 the repetition text and either the following jump or
6926 pop_failure_jump back to this on_failure_jump. */
6927 case on_failure_jump
:
6929 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
6931 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6933 DEBUG_PRINT3 (" %d (to %p)", mcnt
, p
+ mcnt
);
6935 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
6938 /* If this on_failure_jump comes right before a group (i.e.,
6939 the original * applied to a group), save the information
6940 for that group and all inner ones, so that if we fail back
6941 to this point, the group's information will be correct.
6942 For example, in \(a*\)*\1, we need the preceding group,
6943 and in \(zz\(a*\)b*\)\2, we need the inner group. */
6945 /* We can't use `p' to check ahead because we push
6946 a failure point to `p + mcnt' after we do this. */
6949 /* We need to skip no_op's before we look for the
6950 start_memory in case this on_failure_jump is happening as
6951 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
6953 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
6956 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
6958 /* We have a new highest active register now. This will
6959 get reset at the start_memory we are about to get to,
6960 but we will have saved all the registers relevant to
6961 this repetition op, as described above. */
6962 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
6963 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
6964 lowest_active_reg
= *(p1
+ 1);
6967 DEBUG_PRINT1 (":\n");
6968 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
6972 /* A smart repeat ends with `maybe_pop_jump'.
6973 We change it to either `pop_failure_jump' or `jump'. */
6974 case maybe_pop_jump
:
6975 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6976 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
6978 register UCHAR_T
*p2
= p
;
6980 /* Compare the beginning of the repeat with what in the
6981 pattern follows its end. If we can establish that there
6982 is nothing that they would both match, i.e., that we
6983 would have to backtrack because of (as in, e.g., `a*a')
6984 then we can change to pop_failure_jump, because we'll
6985 never have to backtrack.
6987 This is not true in the case of alternatives: in
6988 `(a|ab)*' we do need to backtrack to the `ab' alternative
6989 (e.g., if the string was `ab'). But instead of trying to
6990 detect that here, the alternative has put on a dummy
6991 failure point which is what we will end up popping. */
6993 /* Skip over open/close-group commands.
6994 If what follows this loop is a ...+ construct,
6995 look at what begins its body, since we will have to
6996 match at least one of that. */
7000 && ((re_opcode_t
) *p2
== stop_memory
7001 || (re_opcode_t
) *p2
== start_memory
))
7003 else if (p2
+ 2 + 2 * OFFSET_ADDRESS_SIZE
< pend
7004 && (re_opcode_t
) *p2
== dummy_failure_jump
)
7005 p2
+= 2 + 2 * OFFSET_ADDRESS_SIZE
;
7011 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7012 to the `maybe_finalize_jump' of this case. Examine what
7015 /* If we're at the end of the pattern, we can change. */
7018 /* Consider what happens when matching ":\(.*\)"
7019 against ":/". I don't really understand this code
7021 p
[-(1+OFFSET_ADDRESS_SIZE
)] = (UCHAR_T
)
7024 (" End of pattern: change to `pop_failure_jump'.\n");
7027 else if ((re_opcode_t
) *p2
== exactn
7029 || (re_opcode_t
) *p2
== exactn_bin
7031 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
7034 = *p2
== (UCHAR_T
) endline
? '\n' : p2
[2];
7036 if (((re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == exactn
7038 || (re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == exactn_bin
7040 ) && p1
[3+OFFSET_ADDRESS_SIZE
] != c
)
7042 p
[-(1+OFFSET_ADDRESS_SIZE
)] = (UCHAR_T
)
7045 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7047 (wint_t) p1
[3+OFFSET_ADDRESS_SIZE
]);
7049 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7051 (char) p1
[3+OFFSET_ADDRESS_SIZE
]);
7056 else if ((re_opcode_t
) p1
[3] == charset
7057 || (re_opcode_t
) p1
[3] == charset_not
)
7059 int negate
= (re_opcode_t
) p1
[3] == charset_not
;
7061 if (c
< (unsigned) (p1
[4] * BYTEWIDTH
)
7062 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
7065 /* `negate' is equal to 1 if c would match, which means
7066 that we can't change to pop_failure_jump. */
7069 p
[-3] = (unsigned char) pop_failure_jump
;
7070 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7073 #endif /* not WCHAR */
7076 else if ((re_opcode_t
) *p2
== charset
)
7078 /* We win if the first character of the loop is not part
7080 if ((re_opcode_t
) p1
[3] == exactn
7081 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
7082 && (p2
[2 + p1
[5] / BYTEWIDTH
]
7083 & (1 << (p1
[5] % BYTEWIDTH
)))))
7085 p
[-3] = (unsigned char) pop_failure_jump
;
7086 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7089 else if ((re_opcode_t
) p1
[3] == charset_not
)
7092 /* We win if the charset_not inside the loop
7093 lists every character listed in the charset after. */
7094 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
7095 if (! (p2
[2 + idx
] == 0
7096 || (idx
< (int) p1
[4]
7097 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
7102 p
[-3] = (unsigned char) pop_failure_jump
;
7103 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7106 else if ((re_opcode_t
) p1
[3] == charset
)
7109 /* We win if the charset inside the loop
7110 has no overlap with the one after the loop. */
7112 idx
< (int) p2
[1] && idx
< (int) p1
[4];
7114 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
7117 if (idx
== p2
[1] || idx
== p1
[4])
7119 p
[-3] = (unsigned char) pop_failure_jump
;
7120 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7124 #endif /* not WCHAR */
7126 p
-= OFFSET_ADDRESS_SIZE
; /* Point at relative address again. */
7127 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
7129 p
[-1] = (UCHAR_T
) jump
;
7130 DEBUG_PRINT1 (" Match => jump.\n");
7131 goto unconditional_jump
;
7136 /* The end of a simple repeat has a pop_failure_jump back to
7137 its matching on_failure_jump, where the latter will push a
7138 failure point. The pop_failure_jump takes off failure
7139 points put on by this pop_failure_jump's matching
7140 on_failure_jump; we got through the pattern to here from the
7141 matching on_failure_jump, so didn't fail. */
7142 case pop_failure_jump
:
7144 /* We need to pass separate storage for the lowest and
7145 highest registers, even though we don't care about the
7146 actual values. Otherwise, we will restore only one
7147 register from the stack, since lowest will == highest in
7148 `pop_failure_point'. */
7149 active_reg_t dummy_low_reg
, dummy_high_reg
;
7150 UCHAR_T
*pdummy ATTRIBUTE_UNUSED
= NULL
;
7151 const CHAR_T
*sdummy ATTRIBUTE_UNUSED
= NULL
;
7153 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7154 POP_FAILURE_POINT (sdummy
, pdummy
,
7155 dummy_low_reg
, dummy_high_reg
,
7156 reg_dummy
, reg_dummy
, reg_info_dummy
);
7162 DEBUG_PRINT2 ("\n%p: ", p
);
7164 DEBUG_PRINT2 ("\n0x%x: ", p
);
7166 /* Note fall through. */
7168 /* Unconditionally jump (without popping any failure points). */
7170 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
7171 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
7172 p
+= mcnt
; /* Do the jump. */
7174 DEBUG_PRINT2 ("(to %p).\n", p
);
7176 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
7181 /* We need this opcode so we can detect where alternatives end
7182 in `group_match_null_string_p' et al. */
7184 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7185 goto unconditional_jump
;
7188 /* Normally, the on_failure_jump pushes a failure point, which
7189 then gets popped at pop_failure_jump. We will end up at
7190 pop_failure_jump, also, and with a pattern of, say, `a+', we
7191 are skipping over the on_failure_jump, so we have to push
7192 something meaningless for pop_failure_jump to pop. */
7193 case dummy_failure_jump
:
7194 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7195 /* It doesn't matter what we push for the string here. What
7196 the code at `fail' tests is the value for the pattern. */
7197 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
7198 goto unconditional_jump
;
7201 /* At the end of an alternative, we need to push a dummy failure
7202 point in case we are followed by a `pop_failure_jump', because
7203 we don't want the failure point for the alternative to be
7204 popped. For example, matching `(a|ab)*' against `aab'
7205 requires that we match the `ab' alternative. */
7206 case push_dummy_failure
:
7207 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7208 /* See comments just above at `dummy_failure_jump' about the
7210 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
7213 /* Have to succeed matching what follows at least n times.
7214 After that, handle like `on_failure_jump'. */
7216 EXTRACT_NUMBER (mcnt
, p
+ OFFSET_ADDRESS_SIZE
);
7217 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
7220 /* Originally, this is how many times we HAVE to succeed. */
7224 p
+= OFFSET_ADDRESS_SIZE
;
7225 STORE_NUMBER_AND_INCR (p
, mcnt
);
7227 DEBUG_PRINT3 (" Setting %p to %d.\n", p
- OFFSET_ADDRESS_SIZE
7230 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
- OFFSET_ADDRESS_SIZE
7237 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7238 p
+ OFFSET_ADDRESS_SIZE
);
7240 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7241 p
+ OFFSET_ADDRESS_SIZE
);
7245 p
[1] = (UCHAR_T
) no_op
;
7247 p
[2] = (UCHAR_T
) no_op
;
7248 p
[3] = (UCHAR_T
) no_op
;
7255 EXTRACT_NUMBER (mcnt
, p
+ OFFSET_ADDRESS_SIZE
);
7256 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
7258 /* Originally, this is how many times we CAN jump. */
7262 STORE_NUMBER (p
+ OFFSET_ADDRESS_SIZE
, mcnt
);
7265 DEBUG_PRINT3 (" Setting %p to %d.\n", p
+ OFFSET_ADDRESS_SIZE
,
7268 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
+ OFFSET_ADDRESS_SIZE
,
7271 goto unconditional_jump
;
7273 /* If don't have to jump any more, skip over the rest of command. */
7275 p
+= 2 * OFFSET_ADDRESS_SIZE
;
7280 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7282 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7284 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7286 DEBUG_PRINT3 (" Setting %p to %d.\n", p1
, mcnt
);
7288 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
7290 STORE_NUMBER (p1
, mcnt
);
7295 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7296 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7297 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7298 macro and introducing temporary variables works around the bug. */
7301 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7302 if (AT_WORD_BOUNDARY (d
))
7307 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7308 if (AT_WORD_BOUNDARY (d
))
7314 boolean prevchar
, thischar
;
7316 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7317 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
7320 prevchar
= WORDCHAR_P (d
- 1);
7321 thischar
= WORDCHAR_P (d
);
7322 if (prevchar
!= thischar
)
7329 boolean prevchar
, thischar
;
7331 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7332 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
7335 prevchar
= WORDCHAR_P (d
- 1);
7336 thischar
= WORDCHAR_P (d
);
7337 if (prevchar
!= thischar
)
7344 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7345 if (!AT_STRINGS_END (d
) && WORDCHAR_P (d
)
7346 && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
7351 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7352 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
7353 && (AT_STRINGS_END (d
) || !WORDCHAR_P (d
)))
7359 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7360 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
7365 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7366 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
7371 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7372 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
7377 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
7382 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7386 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7388 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
7390 SET_REGS_MATCHED ();
7394 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
7396 goto matchnotsyntax
;
7399 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7403 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7405 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
7407 SET_REGS_MATCHED ();
7410 #else /* not emacs */
7412 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7414 if (!WORDCHAR_P (d
))
7416 SET_REGS_MATCHED ();
7421 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7425 SET_REGS_MATCHED ();
7428 #endif /* not emacs */
7433 continue; /* Successfully executed one pattern command; keep going. */
7436 /* We goto here if a matching operation fails. */
7438 if (!FAIL_STACK_EMPTY ())
7439 { /* A restart point is known. Restore to that state. */
7440 DEBUG_PRINT1 ("\nFAIL:\n");
7441 POP_FAILURE_POINT (d
, p
,
7442 lowest_active_reg
, highest_active_reg
,
7443 regstart
, regend
, reg_info
);
7445 /* If this failure point is a dummy, try the next one. */
7449 /* If we failed to the end of the pattern, don't examine *p. */
7453 boolean is_a_jump_n
= false;
7455 /* If failed to a backwards jump that's part of a repetition
7456 loop, need to pop this failure point and use the next one. */
7457 switch ((re_opcode_t
) *p
)
7462 case maybe_pop_jump
:
7463 case pop_failure_jump
:
7466 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7469 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
7471 && (re_opcode_t
) *p1
== on_failure_jump
))
7479 if (d
>= string1
&& d
<= end1
)
7483 break; /* Matching at this starting point really fails. */
7487 goto restore_best_regs
;
7491 return -1; /* Failure to match. */
7494 /* Subroutine definitions for re_match_2. */
7497 /* We are passed P pointing to a register number after a start_memory.
7499 Return true if the pattern up to the corresponding stop_memory can
7500 match the empty string, and false otherwise.
7502 If we find the matching stop_memory, sets P to point to one past its number.
7503 Otherwise, sets P to an undefined byte less than or equal to END.
7505 We don't handle duplicates properly (yet). */
7508 PREFIX(group_match_null_string_p
) (UCHAR_T
**p
, UCHAR_T
*end
,
7509 PREFIX(register_info_type
) *reg_info
)
7512 /* Point to after the args to the start_memory. */
7513 UCHAR_T
*p1
= *p
+ 2;
7517 /* Skip over opcodes that can match nothing, and return true or
7518 false, as appropriate, when we get to one that can't, or to the
7519 matching stop_memory. */
7521 switch ((re_opcode_t
) *p1
)
7523 /* Could be either a loop or a series of alternatives. */
7524 case on_failure_jump
:
7526 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7528 /* If the next operation is not a jump backwards in the
7533 /* Go through the on_failure_jumps of the alternatives,
7534 seeing if any of the alternatives cannot match nothing.
7535 The last alternative starts with only a jump,
7536 whereas the rest start with on_failure_jump and end
7537 with a jump, e.g., here is the pattern for `a|b|c':
7539 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7540 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7543 So, we have to first go through the first (n-1)
7544 alternatives and then deal with the last one separately. */
7547 /* Deal with the first (n-1) alternatives, which start
7548 with an on_failure_jump (see above) that jumps to right
7549 past a jump_past_alt. */
7551 while ((re_opcode_t
) p1
[mcnt
-(1+OFFSET_ADDRESS_SIZE
)] ==
7554 /* `mcnt' holds how many bytes long the alternative
7555 is, including the ending `jump_past_alt' and
7558 if (!PREFIX(alt_match_null_string_p
) (p1
, p1
+ mcnt
-
7559 (1 + OFFSET_ADDRESS_SIZE
),
7563 /* Move to right after this alternative, including the
7567 /* Break if it's the beginning of an n-th alternative
7568 that doesn't begin with an on_failure_jump. */
7569 if ((re_opcode_t
) *p1
!= on_failure_jump
)
7572 /* Still have to check that it's not an n-th
7573 alternative that starts with an on_failure_jump. */
7575 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7576 if ((re_opcode_t
) p1
[mcnt
-(1+OFFSET_ADDRESS_SIZE
)] !=
7579 /* Get to the beginning of the n-th alternative. */
7580 p1
-= 1 + OFFSET_ADDRESS_SIZE
;
7585 /* Deal with the last alternative: go back and get number
7586 of the `jump_past_alt' just before it. `mcnt' contains
7587 the length of the alternative. */
7588 EXTRACT_NUMBER (mcnt
, p1
- OFFSET_ADDRESS_SIZE
);
7590 if (!PREFIX(alt_match_null_string_p
) (p1
, p1
+ mcnt
, reg_info
))
7593 p1
+= mcnt
; /* Get past the n-th alternative. */
7599 assert (p1
[1] == **p
);
7605 if (!PREFIX(common_op_match_null_string_p
) (&p1
, end
, reg_info
))
7608 } /* while p1 < end */
7611 } /* group_match_null_string_p */
7614 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7615 It expects P to be the first byte of a single alternative and END one
7616 byte past the last. The alternative can contain groups. */
7619 PREFIX(alt_match_null_string_p
) (UCHAR_T
*p
, UCHAR_T
*end
,
7620 PREFIX(register_info_type
) *reg_info
)
7627 /* Skip over opcodes that can match nothing, and break when we get
7628 to one that can't. */
7630 switch ((re_opcode_t
) *p1
)
7633 case on_failure_jump
:
7635 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7640 if (!PREFIX(common_op_match_null_string_p
) (&p1
, end
, reg_info
))
7643 } /* while p1 < end */
7646 } /* alt_match_null_string_p */
7649 /* Deals with the ops common to group_match_null_string_p and
7650 alt_match_null_string_p.
7652 Sets P to one after the op and its arguments, if any. */
7655 PREFIX(common_op_match_null_string_p
) (UCHAR_T
**p
, UCHAR_T
*end
,
7656 PREFIX(register_info_type
) *reg_info
)
7663 switch ((re_opcode_t
) *p1
++)
7683 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
7684 ret
= PREFIX(group_match_null_string_p
) (&p1
, end
, reg_info
);
7686 /* Have to set this here in case we're checking a group which
7687 contains a group and a back reference to it. */
7689 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
7690 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
7696 /* If this is an optimized succeed_n for zero times, make the jump. */
7698 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7706 /* Get to the number of times to succeed. */
7707 p1
+= OFFSET_ADDRESS_SIZE
;
7708 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7712 p1
-= 2 * OFFSET_ADDRESS_SIZE
;
7713 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7721 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
7726 p1
+= 2 * OFFSET_ADDRESS_SIZE
;
7730 /* All other opcodes mean we cannot match the empty string. */
7736 } /* common_op_match_null_string_p */
7739 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7740 bytes; nonzero otherwise. */
7743 PREFIX(bcmp_translate
) (const CHAR_T
*s1
, const CHAR_T
*s2
, register int len
,
7744 RE_TRANSLATE_TYPE translate
)
7746 register const UCHAR_T
*p1
= (const UCHAR_T
*) s1
;
7747 register const UCHAR_T
*p2
= (const UCHAR_T
*) s2
;
7751 if (((*p1
<=0xff)?translate
[*p1
++]:*p1
++)
7752 != ((*p2
<=0xff)?translate
[*p2
++]:*p2
++))
7755 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
7763 #else /* not INSIDE_RECURSION */
7765 /* Entry points for GNU code. */
7767 /* re_compile_pattern is the GNU regular expression compiler: it
7768 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7769 Returns 0 if the pattern was valid, otherwise an error string.
7771 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7772 are set in BUFP on entry.
7774 We call regex_compile to do the actual compilation. */
7777 re_compile_pattern (const char *pattern
, size_t length
,
7778 struct re_pattern_buffer
*bufp
)
7782 /* GNU code is written to assume at least RE_NREGS registers will be set
7783 (and at least one extra will be -1). */
7784 bufp
->regs_allocated
= REGS_UNALLOCATED
;
7786 /* And GNU code determines whether or not to get register information
7787 by passing null for the REGS argument to re_match, etc., not by
7791 /* Match anchors at newline. */
7792 bufp
->newline_anchor
= 1;
7795 if (MB_CUR_MAX
!= 1)
7796 ret
= wcs_regex_compile (pattern
, length
, re_syntax_options
, bufp
);
7799 ret
= byte_regex_compile (pattern
, length
, re_syntax_options
, bufp
);
7803 return gettext (re_error_msgid
[(int) ret
]);
7806 weak_alias (__re_compile_pattern
, re_compile_pattern
)
7809 /* Entry points compatible with 4.2 BSD regex library. We don't define
7810 them unless specifically requested. */
7812 #if defined _REGEX_RE_COMP || defined _LIBC
7814 /* BSD has one and only one pattern buffer. */
7815 static struct re_pattern_buffer re_comp_buf
;
7819 /* Make these definitions weak in libc, so POSIX programs can redefine
7820 these names if they don't use our functions, and still use
7821 regcomp/regexec below without link errors. */
7824 re_comp (const char *s
)
7830 if (!re_comp_buf
.buffer
)
7831 return (char *) gettext ("No previous regular expression");
7835 if (!re_comp_buf
.buffer
)
7837 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
7838 if (re_comp_buf
.buffer
== NULL
)
7839 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
7840 re_comp_buf
.allocated
= 200;
7842 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
7843 if (re_comp_buf
.fastmap
== NULL
)
7844 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
7847 /* Since `re_exec' always passes NULL for the `regs' argument, we
7848 don't need to initialize the pattern buffer fields which affect it. */
7850 /* Match anchors at newlines. */
7851 re_comp_buf
.newline_anchor
= 1;
7854 if (MB_CUR_MAX
!= 1)
7855 ret
= wcs_regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
7858 ret
= byte_regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
7863 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
7864 return (char *) gettext (re_error_msgid
[(int) ret
]);
7872 re_exec (const char *s
)
7874 const int len
= strlen (s
);
7876 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
7879 #endif /* _REGEX_RE_COMP */
7881 /* POSIX.2 functions. Don't define these for Emacs. */
7885 /* regcomp takes a regular expression as a string and compiles it.
7887 PREG is a regex_t *. We do not expect any fields to be initialized,
7888 since POSIX says we shouldn't. Thus, we set
7890 `buffer' to the compiled pattern;
7891 `used' to the length of the compiled pattern;
7892 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
7893 REG_EXTENDED bit in CFLAGS is set; otherwise, to
7894 RE_SYNTAX_POSIX_BASIC;
7895 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
7896 `fastmap' to an allocated space for the fastmap;
7897 `fastmap_accurate' to zero;
7898 `re_nsub' to the number of subexpressions in PATTERN.
7900 PATTERN is the address of the pattern string.
7902 CFLAGS is a series of bits which affect compilation.
7904 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
7905 use POSIX basic syntax.
7907 If REG_NEWLINE is set, then . and [^...] don't match newline.
7908 Also, regexec will try a match beginning after every newline.
7910 If REG_ICASE is set, then we considers upper- and lowercase
7911 versions of letters to be equivalent when matching.
7913 If REG_NOSUB is set, then when PREG is passed to regexec, that
7914 routine will report only success or failure, and nothing about the
7917 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
7918 the return codes and their meanings.) */
7921 regcomp (regex_t
*preg
, const char *pattern
, int cflags
)
7925 = (cflags
& REG_EXTENDED
) ?
7926 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
7928 /* regex_compile will allocate the space for the compiled pattern. */
7930 preg
->allocated
= 0;
7933 /* Try to allocate space for the fastmap. */
7934 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
7936 if (cflags
& REG_ICASE
)
7941 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
7942 * sizeof (*(RE_TRANSLATE_TYPE
)0));
7943 if (preg
->translate
== NULL
)
7944 return (int) REG_ESPACE
;
7946 /* Map uppercase characters to corresponding lowercase ones. */
7947 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
7948 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
7951 preg
->translate
= NULL
;
7953 /* If REG_NEWLINE is set, newlines are treated differently. */
7954 if (cflags
& REG_NEWLINE
)
7955 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
7956 syntax
&= ~RE_DOT_NEWLINE
;
7957 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
7958 /* It also changes the matching behavior. */
7959 preg
->newline_anchor
= 1;
7962 preg
->newline_anchor
= 0;
7964 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
7966 /* POSIX says a null character in the pattern terminates it, so we
7967 can use strlen here in compiling the pattern. */
7969 if (MB_CUR_MAX
!= 1)
7970 ret
= wcs_regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
7973 ret
= byte_regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
7975 /* POSIX doesn't distinguish between an unmatched open-group and an
7976 unmatched close-group: both are REG_EPAREN. */
7977 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
7979 if (ret
== REG_NOERROR
&& preg
->fastmap
)
7981 /* Compute the fastmap now, since regexec cannot modify the pattern
7983 if (re_compile_fastmap (preg
) == -2)
7985 /* Some error occurred while computing the fastmap, just forget
7987 free (preg
->fastmap
);
7988 preg
->fastmap
= NULL
;
7995 weak_alias (__regcomp
, regcomp
)
7999 /* regexec searches for a given pattern, specified by PREG, in the
8002 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8003 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8004 least NMATCH elements, and we set them to the offsets of the
8005 corresponding matched substrings.
8007 EFLAGS specifies `execution flags' which affect matching: if
8008 REG_NOTBOL is set, then ^ does not match at the beginning of the
8009 string; if REG_NOTEOL is set, then $ does not match at the end.
8011 We return 0 if we find a match and REG_NOMATCH if not. */
8014 regexec (const regex_t
*preg
, const char *string
, size_t nmatch
,
8015 regmatch_t pmatch
[], int eflags
)
8018 struct re_registers regs
;
8019 regex_t private_preg
;
8020 int len
= strlen (string
);
8021 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
8023 private_preg
= *preg
;
8025 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
8026 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
8028 /* The user has told us exactly how many registers to return
8029 information about, via `nmatch'. We have to pass that on to the
8030 matching routines. */
8031 private_preg
.regs_allocated
= REGS_FIXED
;
8035 regs
.num_regs
= nmatch
;
8036 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
8037 if (regs
.start
== NULL
)
8038 return (int) REG_NOMATCH
;
8039 regs
.end
= regs
.start
+ nmatch
;
8042 /* Perform the searching operation. */
8043 ret
= re_search (&private_preg
, string
, len
,
8044 /* start: */ 0, /* range: */ len
,
8045 want_reg_info
? ®s
: (struct re_registers
*) 0);
8047 /* Copy the register information to the POSIX structure. */
8054 for (r
= 0; r
< nmatch
; r
++)
8056 pmatch
[r
].rm_so
= regs
.start
[r
];
8057 pmatch
[r
].rm_eo
= regs
.end
[r
];
8061 /* If we needed the temporary register info, free the space now. */
8065 /* We want zero return to mean success, unlike `re_search'. */
8066 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
8069 weak_alias (__regexec
, regexec
)
8073 /* Returns a message corresponding to an error code, ERRCODE, returned
8074 from either regcomp or regexec. We don't use PREG here. */
8077 regerror (int errcode
, const regex_t
*preg ATTRIBUTE_UNUSED
,
8078 char *errbuf
, size_t errbuf_size
)
8084 || errcode
>= (int) (sizeof (re_error_msgid
)
8085 / sizeof (re_error_msgid
[0])))
8086 /* Only error codes returned by the rest of the code should be passed
8087 to this routine. If we are given anything else, or if other regex
8088 code generates an invalid error code, then the program has a bug.
8089 Dump core so we can fix it. */
8092 msg
= gettext (re_error_msgid
[errcode
]);
8094 msg_size
= strlen (msg
) + 1; /* Includes the null. */
8096 if (errbuf_size
!= 0)
8098 if (msg_size
> errbuf_size
)
8100 #if defined HAVE_MEMPCPY || defined _LIBC
8101 *((char *) mempcpy (errbuf
, msg
, errbuf_size
- 1)) = '\0';
8103 (void) memcpy (errbuf
, msg
, errbuf_size
- 1);
8104 errbuf
[errbuf_size
- 1] = 0;
8108 (void) memcpy (errbuf
, msg
, msg_size
);
8114 weak_alias (__regerror
, regerror
)
8118 /* Free dynamically allocated space used by PREG. */
8121 regfree (regex_t
*preg
)
8123 free (preg
->buffer
);
8124 preg
->buffer
= NULL
;
8126 preg
->allocated
= 0;
8129 free (preg
->fastmap
);
8130 preg
->fastmap
= NULL
;
8131 preg
->fastmap_accurate
= 0;
8133 free (preg
->translate
);
8134 preg
->translate
= NULL
;
8137 weak_alias (__regfree
, regfree
)
8140 #endif /* not emacs */
8142 #endif /* not INSIDE_RECURSION */
8146 #undef STORE_NUMBER_AND_INCR
8147 #undef EXTRACT_NUMBER
8148 #undef EXTRACT_NUMBER_AND_INCR
8150 #undef DEBUG_PRINT_COMPILED_PATTERN
8151 #undef DEBUG_PRINT_DOUBLE_STRING
8153 #undef INIT_FAIL_STACK
8154 #undef RESET_FAIL_STACK
8155 #undef DOUBLE_FAIL_STACK
8156 #undef PUSH_PATTERN_OP
8157 #undef PUSH_FAILURE_POINTER
8158 #undef PUSH_FAILURE_INT
8159 #undef PUSH_FAILURE_ELT
8160 #undef POP_FAILURE_POINTER
8161 #undef POP_FAILURE_INT
8162 #undef POP_FAILURE_ELT
8165 #undef PUSH_FAILURE_POINT
8166 #undef POP_FAILURE_POINT
8168 #undef REG_UNSET_VALUE
8176 #undef INIT_BUF_SIZE
8177 #undef GET_BUFFER_SPACE
8185 #undef EXTEND_BUFFER
8186 #undef GET_UNSIGNED_NUMBER
8187 #undef FREE_STACK_RETURN
8189 # undef POINTER_TO_OFFSET
8190 # undef MATCHING_IN_FRST_STRING
8192 # undef AT_STRINGS_BEG
8193 # undef AT_STRINGS_END
8196 # undef FREE_VARIABLES
8197 # undef NO_HIGHEST_ACTIVE_REG
8198 # undef NO_LOWEST_ACTIVE_REG
8202 # undef COMPILED_BUFFER_VAR
8203 # undef OFFSET_ADDRESS_SIZE
8204 # undef CHAR_CLASS_SIZE
8211 # define DEFINED_ONCE