1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
5 Copyright (C) 1993-1999, 2000, 2001 Free Software Foundation, Inc.
6 This file is part of the GNU C Library.
8 The GNU C Library is free software; you can redistribute it and/or
9 modify it under the terms of the GNU Lesser General Public
10 License as published by the Free Software Foundation; either
11 version 2.1 of the License, or (at your option) any later version.
13 The GNU C Library is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public
19 License along with the GNU C Library; if not, write to the Free
20 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
23 /* AIX requires this to be the first thing in the file. */
24 #if defined _AIX && !defined REGEX_MALLOC
36 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
37 # define PARAMS(args) args
39 # define PARAMS(args) ()
41 #endif /* Not PARAMS. */
43 #ifndef INSIDE_RECURSION
45 # if defined STDC_HEADERS && !defined emacs
48 /* We need this for `regex.h', and perhaps for the Emacs include files. */
49 # include <sys/types.h>
52 # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
54 /* For platform which support the ISO C amendement 1 functionality we
55 support user defined character classes. */
56 # if defined _LIBC || WIDE_CHAR_SUPPORT
57 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
63 /* We have to keep the namespace clean. */
64 # define regfree(preg) __regfree (preg)
65 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
66 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
67 # define regerror(errcode, preg, errbuf, errbuf_size) \
68 __regerror(errcode, preg, errbuf, errbuf_size)
69 # define re_set_registers(bu, re, nu, st, en) \
70 __re_set_registers (bu, re, nu, st, en)
71 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
72 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
73 # define re_match(bufp, string, size, pos, regs) \
74 __re_match (bufp, string, size, pos, regs)
75 # define re_search(bufp, string, size, startpos, range, regs) \
76 __re_search (bufp, string, size, startpos, range, regs)
77 # define re_compile_pattern(pattern, length, bufp) \
78 __re_compile_pattern (pattern, length, bufp)
79 # define re_set_syntax(syntax) __re_set_syntax (syntax)
80 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
81 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
82 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
84 # define btowc __btowc
86 /* We are also using some library internals. */
87 # include <locale/localeinfo.h>
88 # include <locale/elem-hash.h>
89 # include <langinfo.h>
90 # include <locale/coll-lookup.h>
93 /* This is for other GNU distributions with internationalized messages. */
94 # if HAVE_LIBINTL_H || defined _LIBC
98 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
101 # define gettext(msgid) (msgid)
104 # ifndef gettext_noop
105 /* This define is so xgettext can find the internationalizable
107 # define gettext_noop(String) String
110 /* The `emacs' switch turns on certain matching commands
111 that make sense only in Emacs. */
118 # else /* not emacs */
120 /* If we are not linking with Emacs proper,
121 we can't use the relocating allocator
122 even if config.h says that we can. */
125 # if defined STDC_HEADERS || defined _LIBC
132 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
133 If nothing else has been done, use the method below. */
134 # ifdef INHIBIT_STRING_HEADER
135 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
136 # if !defined bzero && !defined bcopy
137 # undef INHIBIT_STRING_HEADER
142 /* This is the normal way of making sure we have a bcopy and a bzero.
143 This is used in most programs--a few other programs avoid this
144 by defining INHIBIT_STRING_HEADER. */
145 # ifndef INHIBIT_STRING_HEADER
146 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
150 # define bzero(s, n) (memset (s, '\0', n), (s))
152 # define bzero(s, n) __bzero (s, n)
156 # include <strings.h>
158 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
161 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
166 /* Define the syntax stuff for \<, \>, etc. */
168 /* This must be nonzero for the wordchar and notwordchar pattern
169 commands in re_match_2. */
174 # ifdef SWITCH_ENUM_BUG
175 # define SWITCH_ENUM_CAST(x) ((int)(x))
177 # define SWITCH_ENUM_CAST(x) (x)
180 # endif /* not emacs */
182 # if defined _LIBC || HAVE_LIMITS_H
187 # define MB_LEN_MAX 1
190 /* Get the interface, including the syntax bits. */
193 /* isalpha etc. are used for the character classes. */
196 /* Jim Meyering writes:
198 "... Some ctype macros are valid only for character codes that
199 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
200 using /bin/cc or gcc but without giving an ansi option). So, all
201 ctype uses should be through macros like ISPRINT... If
202 STDC_HEADERS is defined, then autoconf has verified that the ctype
203 macros don't need to be guarded with references to isascii. ...
204 Defining isascii to 1 should let any compiler worth its salt
205 eliminate the && through constant folding."
206 Solaris defines some of these symbols so we must undefine them first. */
208 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
209 # define IN_CTYPE_DOMAIN(c) 1
211 # define IN_CTYPE_DOMAIN(c) isascii(c)
215 # define ISBLANK(c) (IN_CTYPE_DOMAIN (c) && isblank (c))
217 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
220 # define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isgraph (c))
222 # define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isprint (c) && !isspace (c))
226 # define ISPRINT(c) (IN_CTYPE_DOMAIN (c) && isprint (c))
227 # define ISDIGIT(c) (IN_CTYPE_DOMAIN (c) && isdigit (c))
228 # define ISALNUM(c) (IN_CTYPE_DOMAIN (c) && isalnum (c))
229 # define ISALPHA(c) (IN_CTYPE_DOMAIN (c) && isalpha (c))
230 # define ISCNTRL(c) (IN_CTYPE_DOMAIN (c) && iscntrl (c))
231 # define ISLOWER(c) (IN_CTYPE_DOMAIN (c) && islower (c))
232 # define ISPUNCT(c) (IN_CTYPE_DOMAIN (c) && ispunct (c))
233 # define ISSPACE(c) (IN_CTYPE_DOMAIN (c) && isspace (c))
234 # define ISUPPER(c) (IN_CTYPE_DOMAIN (c) && isupper (c))
235 # define ISXDIGIT(c) (IN_CTYPE_DOMAIN (c) && isxdigit (c))
238 # define TOLOWER(c) _tolower(c)
240 # define TOLOWER(c) tolower(c)
244 # define NULL (void *)0
247 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
248 since ours (we hope) works properly with all combinations of
249 machines, compilers, `char' and `unsigned char' argument types.
250 (Per Bothner suggested the basic approach.) */
251 # undef SIGN_EXTEND_CHAR
253 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
254 # else /* not __STDC__ */
255 /* As in Harbison and Steele. */
256 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
260 /* How many characters in the character set. */
261 # define CHAR_SET_SIZE 256
265 extern char *re_syntax_table;
267 # else /* not SYNTAX_TABLE */
269 static char re_syntax_table[CHAR_SET_SIZE];
271 static void init_syntax_once PARAMS ((void));
281 bzero (re_syntax_table, sizeof re_syntax_table);
283 for (c = 0; c < CHAR_SET_SIZE; ++c)
285 re_syntax_table[c] = Sword;
287 re_syntax_table['_'] = Sword;
292 # endif /* not SYNTAX_TABLE */
294 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
298 /* Integer type for pointers. */
300 typedef unsigned long int uintptr_t;
303 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
304 use `alloca' instead of `malloc'. This is because using malloc in
305 re_search* or re_match* could cause memory leaks when C-g is used in
306 Emacs; also, malloc is slower and causes storage fragmentation. On
307 the other hand, malloc is more portable, and easier to debug.
309 Because we sometimes use alloca, some routines have to be macros,
310 not functions -- `alloca'-allocated space disappears at the end of the
311 function it is called in. */
315 # define REGEX_ALLOCATE malloc
316 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
317 # define REGEX_FREE free
319 # else /* not REGEX_MALLOC */
321 /* Emacs already defines alloca, sometimes. */
324 /* Make alloca work the best possible way. */
326 # define alloca __builtin_alloca
327 # else /* not __GNUC__ */
330 # endif /* HAVE_ALLOCA_H */
331 # endif /* not __GNUC__ */
333 # endif /* not alloca */
335 # define REGEX_ALLOCATE alloca
337 /* Assumes a `char *destination' variable. */
338 # define REGEX_REALLOCATE(source, osize, nsize) \
339 (destination = (char *) alloca (nsize), \
340 memcpy (destination, source, osize))
342 /* No need to do anything to free, after alloca. */
343 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
345 # endif /* not REGEX_MALLOC */
347 /* Define how to allocate the failure stack. */
349 # if defined REL_ALLOC && defined REGEX_MALLOC
351 # define REGEX_ALLOCATE_STACK(size) \
352 r_alloc (&failure_stack_ptr, (size))
353 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
354 r_re_alloc (&failure_stack_ptr, (nsize))
355 # define REGEX_FREE_STACK(ptr) \
356 r_alloc_free (&failure_stack_ptr)
358 # else /* not using relocating allocator */
362 # define REGEX_ALLOCATE_STACK malloc
363 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
364 # define REGEX_FREE_STACK free
366 # else /* not REGEX_MALLOC */
368 # define REGEX_ALLOCATE_STACK alloca
370 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
371 REGEX_REALLOCATE (source, osize, nsize)
372 /* No need to explicitly free anything. */
373 # define REGEX_FREE_STACK(arg)
375 # endif /* not REGEX_MALLOC */
376 # endif /* not using relocating allocator */
379 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
380 `string1' or just past its end. This works if PTR is NULL, which is
382 # define FIRST_STRING_P(ptr) \
383 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
385 /* (Re)Allocate N items of type T using malloc, or fail. */
386 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
387 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
388 # define RETALLOC_IF(addr, n, t) \
389 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
390 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
392 # define BYTEWIDTH 8 /* In bits. */
394 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
398 # define MAX(a, b) ((a) > (b) ? (a) : (b))
399 # define MIN(a, b) ((a) < (b) ? (a) : (b))
401 typedef char boolean;
405 static reg_errcode_t byte_regex_compile _RE_ARGS ((const char *pattern, size_t size,
407 struct re_pattern_buffer *bufp));
409 static int byte_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
410 const char *string1, int size1,
411 const char *string2, int size2,
413 struct re_registers *regs,
415 static int byte_re_search_2 PARAMS ((struct re_pattern_buffer *bufp,
416 const char *string1, int size1,
417 const char *string2, int size2,
418 int startpos, int range,
419 struct re_registers *regs, int stop));
420 static int byte_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
423 static reg_errcode_t wcs_regex_compile _RE_ARGS ((const char *pattern, size_t size,
425 struct re_pattern_buffer *bufp));
428 static int wcs_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
429 const char *cstring1, int csize1,
430 const char *cstring2, int csize2,
432 struct re_registers *regs,
434 wchar_t *string1, int size1,
435 wchar_t *string2, int size2,
436 int *mbs_offset1, int *mbs_offset2));
437 static int wcs_re_search_2 PARAMS ((struct re_pattern_buffer *bufp,
438 const char *string1, int size1,
439 const char *string2, int size2,
440 int startpos, int range,
441 struct re_registers *regs, int stop));
442 static int wcs_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
445 /* These are the command codes that appear in compiled regular
446 expressions. Some opcodes are followed by argument bytes. A
447 command code can specify any interpretation whatsoever for its
448 arguments. Zero bytes may appear in the compiled regular expression. */
454 /* Succeed right away--no more backtracking. */
457 /* Followed by one byte giving n, then by n literal bytes. */
461 /* Same as exactn, but contains binary data. */
465 /* Matches any (more or less) character. */
468 /* Matches any one char belonging to specified set. First
469 following byte is number of bitmap bytes. Then come bytes
470 for a bitmap saying which chars are in. Bits in each byte
471 are ordered low-bit-first. A character is in the set if its
472 bit is 1. A character too large to have a bit in the map is
473 automatically not in the set. */
474 /* ifdef MBS_SUPPORT, following element is length of character
475 classes, length of collating symbols, length of equivalence
476 classes, length of character ranges, and length of characters.
477 Next, character class element, collating symbols elements,
478 equivalence class elements, range elements, and character
480 See regex_compile function. */
483 /* Same parameters as charset, but match any character that is
484 not one of those specified. */
487 /* Start remembering the text that is matched, for storing in a
488 register. Followed by one byte with the register number, in
489 the range 0 to one less than the pattern buffer's re_nsub
490 field. Then followed by one byte with the number of groups
491 inner to this one. (This last has to be part of the
492 start_memory only because we need it in the on_failure_jump
496 /* Stop remembering the text that is matched and store it in a
497 memory register. Followed by one byte with the register
498 number, in the range 0 to one less than `re_nsub' in the
499 pattern buffer, and one byte with the number of inner groups,
500 just like `start_memory'. (We need the number of inner
501 groups here because we don't have any easy way of finding the
502 corresponding start_memory when we're at a stop_memory.) */
505 /* Match a duplicate of something remembered. Followed by one
506 byte containing the register number. */
509 /* Fail unless at beginning of line. */
512 /* Fail unless at end of line. */
515 /* Succeeds if at beginning of buffer (if emacs) or at beginning
516 of string to be matched (if not). */
519 /* Analogously, for end of buffer/string. */
522 /* Followed by two byte relative address to which to jump. */
525 /* Same as jump, but marks the end of an alternative. */
528 /* Followed by two-byte relative address of place to resume at
529 in case of failure. */
530 /* ifdef MBS_SUPPORT, the size of address is 1. */
533 /* Like on_failure_jump, but pushes a placeholder instead of the
534 current string position when executed. */
535 on_failure_keep_string_jump,
537 /* Throw away latest failure point and then jump to following
538 two-byte relative address. */
539 /* ifdef MBS_SUPPORT, the size of address is 1. */
542 /* Change to pop_failure_jump if know won't have to backtrack to
543 match; otherwise change to jump. This is used to jump
544 back to the beginning of a repeat. If what follows this jump
545 clearly won't match what the repeat does, such that we can be
546 sure that there is no use backtracking out of repetitions
547 already matched, then we change it to a pop_failure_jump.
548 Followed by two-byte address. */
549 /* ifdef MBS_SUPPORT, the size of address is 1. */
552 /* Jump to following two-byte address, and push a dummy failure
553 point. This failure point will be thrown away if an attempt
554 is made to use it for a failure. A `+' construct makes this
555 before the first repeat. Also used as an intermediary kind
556 of jump when compiling an alternative. */
557 /* ifdef MBS_SUPPORT, the size of address is 1. */
560 /* Push a dummy failure point and continue. Used at the end of
564 /* Followed by two-byte relative address and two-byte number n.
565 After matching N times, jump to the address upon failure. */
566 /* ifdef MBS_SUPPORT, the size of address is 1. */
569 /* Followed by two-byte relative address, and two-byte number n.
570 Jump to the address N times, then fail. */
571 /* ifdef MBS_SUPPORT, the size of address is 1. */
574 /* Set the following two-byte relative address to the
575 subsequent two-byte number. The address *includes* the two
577 /* ifdef MBS_SUPPORT, the size of address is 1. */
580 wordchar, /* Matches any word-constituent character. */
581 notwordchar, /* Matches any char that is not a word-constituent. */
583 wordbeg, /* Succeeds if at word beginning. */
584 wordend, /* Succeeds if at word end. */
586 wordbound, /* Succeeds if at a word boundary. */
587 notwordbound /* Succeeds if not at a word boundary. */
590 ,before_dot, /* Succeeds if before point. */
591 at_dot, /* Succeeds if at point. */
592 after_dot, /* Succeeds if after point. */
594 /* Matches any character whose syntax is specified. Followed by
595 a byte which contains a syntax code, e.g., Sword. */
598 /* Matches any character whose syntax is not that specified. */
602 #endif /* not INSIDE_RECURSION */
607 # define UCHAR_T unsigned char
608 # define COMPILED_BUFFER_VAR bufp->buffer
609 # define OFFSET_ADDRESS_SIZE 2
610 # define PREFIX(name) byte_##name
611 # define ARG_PREFIX(name) name
612 # define PUT_CHAR(c) putchar (c)
615 # define CHAR_T wchar_t
616 # define UCHAR_T wchar_t
617 # define COMPILED_BUFFER_VAR wc_buffer
618 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
619 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
620 # define PREFIX(name) wcs_##name
621 # define ARG_PREFIX(name) c##name
622 /* Should we use wide stream?? */
623 # define PUT_CHAR(c) printf ("%C", c);
629 # define INSIDE_RECURSION
631 # undef INSIDE_RECURSION
634 # define INSIDE_RECURSION
636 # undef INSIDE_RECURSION
640 #ifdef INSIDE_RECURSION
641 /* Common operations on the compiled pattern. */
643 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
644 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
647 # define STORE_NUMBER(destination, number) \
649 *(destination) = (UCHAR_T)(number); \
652 # define STORE_NUMBER(destination, number) \
654 (destination)[0] = (number) & 0377; \
655 (destination)[1] = (number) >> 8; \
659 /* Same as STORE_NUMBER, except increment DESTINATION to
660 the byte after where the number is stored. Therefore, DESTINATION
661 must be an lvalue. */
662 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
664 # define STORE_NUMBER_AND_INCR(destination, number) \
666 STORE_NUMBER (destination, number); \
667 (destination) += OFFSET_ADDRESS_SIZE; \
670 /* Put into DESTINATION a number stored in two contiguous bytes starting
672 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
675 # define EXTRACT_NUMBER(destination, source) \
677 (destination) = *(source); \
680 # define EXTRACT_NUMBER(destination, source) \
682 (destination) = *(source) & 0377; \
683 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
688 static void PREFIX(extract_number) _RE_ARGS ((int *dest, UCHAR_T *source));
690 PREFIX(extract_number) (dest, source)
697 int temp = SIGN_EXTEND_CHAR (*(source + 1));
698 *dest = *source & 0377;
703 # ifndef EXTRACT_MACROS /* To debug the macros. */
704 # undef EXTRACT_NUMBER
705 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
706 # endif /* not EXTRACT_MACROS */
710 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
711 SOURCE must be an lvalue. */
713 # define EXTRACT_NUMBER_AND_INCR(destination, source) \
715 EXTRACT_NUMBER (destination, source); \
716 (source) += OFFSET_ADDRESS_SIZE; \
720 static void PREFIX(extract_number_and_incr) _RE_ARGS ((int *destination,
723 PREFIX(extract_number_and_incr) (destination, source)
727 PREFIX(extract_number) (destination, *source);
728 *source += OFFSET_ADDRESS_SIZE;
731 # ifndef EXTRACT_MACROS
732 # undef EXTRACT_NUMBER_AND_INCR
733 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
734 PREFIX(extract_number_and_incr) (&dest, &src)
735 # endif /* not EXTRACT_MACROS */
741 /* If DEBUG is defined, Regex prints many voluminous messages about what
742 it is doing (if the variable `debug' is nonzero). If linked with the
743 main program in `iregex.c', you can enter patterns and strings
744 interactively. And if linked with the main program in `main.c' and
745 the other test files, you can run the already-written tests. */
749 # ifndef DEFINED_ONCE
751 /* We use standard I/O for debugging. */
754 /* It is useful to test things that ``must'' be true when debugging. */
759 # define DEBUG_STATEMENT(e) e
760 # define DEBUG_PRINT1(x) if (debug) printf (x)
761 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
762 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
763 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
764 # endif /* not DEFINED_ONCE */
766 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
767 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
768 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
769 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
772 /* Print the fastmap in human-readable form. */
774 # ifndef DEFINED_ONCE
776 print_fastmap (fastmap)
779 unsigned was_a_range = 0;
782 while (i < (1 << BYTEWIDTH))
788 while (i < (1 << BYTEWIDTH) && fastmap[i])
802 # endif /* not DEFINED_ONCE */
805 /* Print a compiled pattern string in human-readable form, starting at
806 the START pointer into it and ending just before the pointer END. */
809 PREFIX(print_partial_compiled_pattern) (start, end)
824 /* Loop over pattern commands. */
828 printf ("%td:\t", p - start);
830 printf ("%ld:\t", (long int) (p - start));
833 switch ((re_opcode_t) *p++)
841 printf ("/exactn/%d", mcnt);
853 printf ("/exactn_bin/%d", mcnt);
856 printf("/%lx", (long int) *p++);
860 # endif /* MBS_SUPPORT */
864 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++);
869 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++);
873 printf ("/duplicate/%ld", (long int) *p++);
886 printf ("/charset [%s",
887 (re_opcode_t) *(workp - 1) == charset_not ? "^" : "");
889 length = *workp++; /* the length of char_classes */
890 for (i=0 ; i<length ; i++)
891 printf("[:%lx:]", (long int) *p++);
892 length = *workp++; /* the length of collating_symbol */
893 for (i=0 ; i<length ;)
897 PUT_CHAR((i++,*p++));
901 length = *workp++; /* the length of equivalence_class */
902 for (i=0 ; i<length ;)
906 PUT_CHAR((i++,*p++));
910 length = *workp++; /* the length of char_range */
911 for (i=0 ; i<length ; i++)
913 wchar_t range_start = *p++;
914 wchar_t range_end = *p++;
915 printf("%C-%C", range_start, range_end);
917 length = *workp++; /* the length of char */
918 for (i=0 ; i<length ; i++)
922 register int c, last = -100;
923 register int in_range = 0;
925 printf ("/charset [%s",
926 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
928 assert (p + *p < pend);
930 for (c = 0; c < 256; c++)
932 && (p[1 + (c/8)] & (1 << (c % 8))))
934 /* Are we starting a range? */
935 if (last + 1 == c && ! in_range)
940 /* Have we broken a range? */
941 else if (last + 1 != c && in_range)
971 case on_failure_jump:
972 PREFIX(extract_number_and_incr) (&mcnt, &p);
974 printf ("/on_failure_jump to %td", p + mcnt - start);
976 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
980 case on_failure_keep_string_jump:
981 PREFIX(extract_number_and_incr) (&mcnt, &p);
983 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start);
985 printf ("/on_failure_keep_string_jump to %ld",
986 (long int) (p + mcnt - start));
990 case dummy_failure_jump:
991 PREFIX(extract_number_and_incr) (&mcnt, &p);
993 printf ("/dummy_failure_jump to %td", p + mcnt - start);
995 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
999 case push_dummy_failure:
1000 printf ("/push_dummy_failure");
1003 case maybe_pop_jump:
1004 PREFIX(extract_number_and_incr) (&mcnt, &p);
1006 printf ("/maybe_pop_jump to %td", p + mcnt - start);
1008 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
1012 case pop_failure_jump:
1013 PREFIX(extract_number_and_incr) (&mcnt, &p);
1015 printf ("/pop_failure_jump to %td", p + mcnt - start);
1017 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
1022 PREFIX(extract_number_and_incr) (&mcnt, &p);
1024 printf ("/jump_past_alt to %td", p + mcnt - start);
1026 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
1031 PREFIX(extract_number_and_incr) (&mcnt, &p);
1033 printf ("/jump to %td", p + mcnt - start);
1035 printf ("/jump to %ld", (long int) (p + mcnt - start));
1040 PREFIX(extract_number_and_incr) (&mcnt, &p);
1042 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1044 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2);
1046 printf ("/succeed_n to %ld, %d times",
1047 (long int) (p1 - start), mcnt2);
1052 PREFIX(extract_number_and_incr) (&mcnt, &p);
1054 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1055 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
1059 PREFIX(extract_number_and_incr) (&mcnt, &p);
1061 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1063 printf ("/set_number_at location %td to %d", p1 - start, mcnt2);
1065 printf ("/set_number_at location %ld to %d",
1066 (long int) (p1 - start), mcnt2);
1071 printf ("/wordbound");
1075 printf ("/notwordbound");
1079 printf ("/wordbeg");
1083 printf ("/wordend");
1088 printf ("/before_dot");
1096 printf ("/after_dot");
1100 printf ("/syntaxspec");
1102 printf ("/%d", mcnt);
1106 printf ("/notsyntaxspec");
1108 printf ("/%d", mcnt);
1113 printf ("/wordchar");
1117 printf ("/notwordchar");
1129 printf ("?%ld", (long int) *(p-1));
1136 printf ("%td:\tend of pattern.\n", p - start);
1138 printf ("%ld:\tend of pattern.\n", (long int) (p - start));
1144 PREFIX(print_compiled_pattern) (bufp)
1145 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) (where, string1, size1, string2, size2)
1178 const CHAR_T *where;
1179 const CHAR_T *string1;
1180 const CHAR_T *string2;
1192 if (FIRST_STRING_P (where))
1194 for (this_char = where - string1; this_char < size1; this_char++)
1195 PUT_CHAR (string1[this_char]);
1201 for (this_char = where - string2; this_char < size2; this_char++)
1203 PUT_CHAR (string2[this_char]);
1206 fputs ("...", stdout);
1213 # ifndef DEFINED_ONCE
1222 # else /* not DEBUG */
1224 # ifndef DEFINED_ONCE
1228 # define DEBUG_STATEMENT(e)
1229 # define DEBUG_PRINT1(x)
1230 # define DEBUG_PRINT2(x1, x2)
1231 # define DEBUG_PRINT3(x1, x2, x3)
1232 # define DEBUG_PRINT4(x1, x2, x3, x4)
1233 # endif /* not DEFINED_ONCE */
1234 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1235 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1237 # endif /* not DEBUG */
1242 /* This convert a multibyte string to a wide character string.
1243 And write their correspondances to offset_buffer(see below)
1244 and write whether each wchar_t is binary data to is_binary.
1245 This assume invalid multibyte sequences as binary data.
1246 We assume offset_buffer and is_binary is already allocated
1249 static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src,
1250 size_t len, int *offset_buffer,
1253 convert_mbs_to_wcs (dest, src, len, offset_buffer, is_binary)
1255 const unsigned char* src;
1256 size_t len; /* the length of multibyte string. */
1258 /* It hold correspondances between src(char string) and
1259 dest(wchar_t string) for optimization.
1261 dest = {'X', 'Y', 'Z'}
1262 (each "xxx", "y" and "zz" represent one multibyte character
1263 corresponding to 'X', 'Y' and 'Z'.)
1264 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1270 wchar_t *pdest = dest;
1271 const unsigned char *psrc = src;
1272 size_t wc_count = 0;
1276 size_t mb_remain = len;
1277 size_t mb_count = 0;
1279 /* Initialize the conversion state. */
1280 memset (&mbs, 0, sizeof (mbstate_t));
1282 offset_buffer[0] = 0;
1283 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed,
1286 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs);
1289 /* failed to convert. maybe src contains binary data.
1290 So we consume 1 byte manualy. */
1294 is_binary[wc_count] = TRUE;
1297 is_binary[wc_count] = FALSE;
1298 /* In sjis encoding, we use yen sign as escape character in
1299 place of reverse solidus. So we convert 0x5c(yen sign in
1300 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1301 solidus in UCS2). */
1302 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5)
1303 *pdest = (wchar_t) *psrc;
1305 offset_buffer[wc_count + 1] = mb_count += consumed;
1308 /* Fill remain of the buffer with sentinel. */
1309 for (i = wc_count + 1 ; i <= len ; i++)
1310 offset_buffer[i] = mb_count + 1;
1317 #else /* not INSIDE_RECURSION */
1319 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1320 also be assigned to arbitrarily: each pattern buffer stores its own
1321 syntax, so it can be changed between regex compilations. */
1322 /* This has no initializer because initialized variables in Emacs
1323 become read-only after dumping. */
1324 reg_syntax_t re_syntax_options;
1327 /* Specify the precise syntax of regexps for compilation. This provides
1328 for compatibility for various utilities which historically have
1329 different, incompatible syntaxes.
1331 The argument SYNTAX is a bit mask comprised of the various bits
1332 defined in regex.h. We return the old syntax. */
1335 re_set_syntax (syntax)
1336 reg_syntax_t syntax;
1338 reg_syntax_t ret = re_syntax_options;
1340 re_syntax_options = syntax;
1342 if (syntax & RE_DEBUG)
1344 else if (debug) /* was on but now is not */
1350 weak_alias (__re_set_syntax, re_set_syntax)
1353 /* This table gives an error message for each of the error codes listed
1354 in regex.h. Obviously the order here has to be same as there.
1355 POSIX doesn't require that we do anything for REG_NOERROR,
1356 but why not be nice? */
1358 static const char re_error_msgid[] =
1360 # define REG_NOERROR_IDX 0
1361 gettext_noop ("Success") /* REG_NOERROR */
1363 # define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1364 gettext_noop ("No match") /* REG_NOMATCH */
1366 # define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1367 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1369 # define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1370 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1372 # define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1373 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1375 # define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1376 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1378 # define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1379 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1381 # define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1382 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1384 # define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1385 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1387 # define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1388 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1390 # define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1391 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1393 # define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1394 gettext_noop ("Invalid range end") /* REG_ERANGE */
1396 # define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1397 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1399 # define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1400 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1402 # define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1403 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1405 # define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1406 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1408 # define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1409 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1412 static const size_t re_error_msgid_idx[] =
1433 #endif /* INSIDE_RECURSION */
1435 #ifndef DEFINED_ONCE
1436 /* Avoiding alloca during matching, to placate r_alloc. */
1438 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1439 searching and matching functions should not call alloca. On some
1440 systems, alloca is implemented in terms of malloc, and if we're
1441 using the relocating allocator routines, then malloc could cause a
1442 relocation, which might (if the strings being searched are in the
1443 ralloc heap) shift the data out from underneath the regexp
1446 Here's another reason to avoid allocation: Emacs
1447 processes input from X in a signal handler; processing X input may
1448 call malloc; if input arrives while a matching routine is calling
1449 malloc, then we're scrod. But Emacs can't just block input while
1450 calling matching routines; then we don't notice interrupts when
1451 they come in. So, Emacs blocks input around all regexp calls
1452 except the matching calls, which it leaves unprotected, in the
1453 faith that they will not malloc. */
1455 /* Normally, this is fine. */
1456 # define MATCH_MAY_ALLOCATE
1458 /* When using GNU C, we are not REALLY using the C alloca, no matter
1459 what config.h may say. So don't take precautions for it. */
1464 /* The match routines may not allocate if (1) they would do it with malloc
1465 and (2) it's not safe for them to use malloc.
1466 Note that if REL_ALLOC is defined, matching would not use malloc for the
1467 failure stack, but we would still use it for the register vectors;
1468 so REL_ALLOC should not affect this. */
1469 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1470 # undef MATCH_MAY_ALLOCATE
1472 #endif /* not DEFINED_ONCE */
1474 #ifdef INSIDE_RECURSION
1475 /* Failure stack declarations and macros; both re_compile_fastmap and
1476 re_match_2 use a failure stack. These have to be macros because of
1477 REGEX_ALLOCATE_STACK. */
1480 /* Number of failure points for which to initially allocate space
1481 when matching. If this number is exceeded, we allocate more
1482 space, so it is not a hard limit. */
1483 # ifndef INIT_FAILURE_ALLOC
1484 # define INIT_FAILURE_ALLOC 5
1487 /* Roughly the maximum number of failure points on the stack. Would be
1488 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1489 This is a variable only so users of regex can assign to it; we never
1490 change it ourselves. */
1492 # ifdef INT_IS_16BIT
1494 # ifndef DEFINED_ONCE
1495 # if defined MATCH_MAY_ALLOCATE
1496 /* 4400 was enough to cause a crash on Alpha OSF/1,
1497 whose default stack limit is 2mb. */
1498 long int re_max_failures = 4000;
1500 long int re_max_failures = 2000;
1504 union PREFIX(fail_stack_elt)
1510 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1514 PREFIX(fail_stack_elt_t) *stack;
1515 unsigned long int size;
1516 unsigned long int avail; /* Offset of next open position. */
1517 } PREFIX(fail_stack_type);
1519 # else /* not INT_IS_16BIT */
1521 # ifndef DEFINED_ONCE
1522 # if defined MATCH_MAY_ALLOCATE
1523 /* 4400 was enough to cause a crash on Alpha OSF/1,
1524 whose default stack limit is 2mb. */
1525 int re_max_failures = 4000;
1527 int re_max_failures = 2000;
1531 union PREFIX(fail_stack_elt)
1537 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1541 PREFIX(fail_stack_elt_t) *stack;
1543 unsigned avail; /* Offset of next open position. */
1544 } PREFIX(fail_stack_type);
1546 # endif /* INT_IS_16BIT */
1548 # ifndef DEFINED_ONCE
1549 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1550 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1551 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1555 /* Define macros to initialize and free the failure stack.
1556 Do `return -2' if the alloc fails. */
1558 # ifdef MATCH_MAY_ALLOCATE
1559 # define INIT_FAIL_STACK() \
1561 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1562 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1564 if (fail_stack.stack == NULL) \
1567 fail_stack.size = INIT_FAILURE_ALLOC; \
1568 fail_stack.avail = 0; \
1571 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1573 # define INIT_FAIL_STACK() \
1575 fail_stack.avail = 0; \
1578 # define RESET_FAIL_STACK()
1582 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1584 Return 1 if succeeds, and 0 if either ran out of memory
1585 allocating space for it or it was already too large.
1587 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1589 # define DOUBLE_FAIL_STACK(fail_stack) \
1590 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1592 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1593 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1594 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1595 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1597 (fail_stack).stack == NULL \
1599 : ((fail_stack).size <<= 1, \
1603 /* Push pointer POINTER on FAIL_STACK.
1604 Return 1 if was able to do so and 0 if ran out of memory allocating
1606 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1607 ((FAIL_STACK_FULL () \
1608 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1610 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1613 /* Push a pointer value onto the failure stack.
1614 Assumes the variable `fail_stack'. Probably should only
1615 be called from within `PUSH_FAILURE_POINT'. */
1616 # define PUSH_FAILURE_POINTER(item) \
1617 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1619 /* This pushes an integer-valued item onto the failure stack.
1620 Assumes the variable `fail_stack'. Probably should only
1621 be called from within `PUSH_FAILURE_POINT'. */
1622 # define PUSH_FAILURE_INT(item) \
1623 fail_stack.stack[fail_stack.avail++].integer = (item)
1625 /* Push a fail_stack_elt_t value onto the failure stack.
1626 Assumes the variable `fail_stack'. Probably should only
1627 be called from within `PUSH_FAILURE_POINT'. */
1628 # define PUSH_FAILURE_ELT(item) \
1629 fail_stack.stack[fail_stack.avail++] = (item)
1631 /* These three POP... operations complement the three PUSH... operations.
1632 All assume that `fail_stack' is nonempty. */
1633 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1634 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1635 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1637 /* Used to omit pushing failure point id's when we're not debugging. */
1639 # define DEBUG_PUSH PUSH_FAILURE_INT
1640 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1642 # define DEBUG_PUSH(item)
1643 # define DEBUG_POP(item_addr)
1647 /* Push the information about the state we will need
1648 if we ever fail back to it.
1650 Requires variables fail_stack, regstart, regend, reg_info, and
1651 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1654 Does `return FAILURE_CODE' if runs out of memory. */
1656 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1658 char *destination; \
1659 /* Must be int, so when we don't save any registers, the arithmetic \
1660 of 0 + -1 isn't done as unsigned. */ \
1661 /* Can't be int, since there is not a shred of a guarantee that int \
1662 is wide enough to hold a value of something to which pointer can \
1664 active_reg_t this_reg; \
1666 DEBUG_STATEMENT (failure_id++); \
1667 DEBUG_STATEMENT (nfailure_points_pushed++); \
1668 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1669 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1670 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1672 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1673 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1675 /* Ensure we have enough space allocated for what we will push. */ \
1676 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1678 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1679 return failure_code; \
1681 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1682 (fail_stack).size); \
1683 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1686 /* Push the info, starting with the registers. */ \
1687 DEBUG_PRINT1 ("\n"); \
1690 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1693 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1694 DEBUG_STATEMENT (num_regs_pushed++); \
1696 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1697 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1699 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1700 PUSH_FAILURE_POINTER (regend[this_reg]); \
1702 DEBUG_PRINT2 (" info: %p\n ", \
1703 reg_info[this_reg].word.pointer); \
1704 DEBUG_PRINT2 (" match_null=%d", \
1705 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1706 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1707 DEBUG_PRINT2 (" matched_something=%d", \
1708 MATCHED_SOMETHING (reg_info[this_reg])); \
1709 DEBUG_PRINT2 (" ever_matched=%d", \
1710 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1711 DEBUG_PRINT1 ("\n"); \
1712 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1715 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1716 PUSH_FAILURE_INT (lowest_active_reg); \
1718 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1719 PUSH_FAILURE_INT (highest_active_reg); \
1721 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1722 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1723 PUSH_FAILURE_POINTER (pattern_place); \
1725 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1726 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1728 DEBUG_PRINT1 ("'\n"); \
1729 PUSH_FAILURE_POINTER (string_place); \
1731 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1732 DEBUG_PUSH (failure_id); \
1735 # ifndef DEFINED_ONCE
1736 /* This is the number of items that are pushed and popped on the stack
1737 for each register. */
1738 # define NUM_REG_ITEMS 3
1740 /* Individual items aside from the registers. */
1742 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1744 # define NUM_NONREG_ITEMS 4
1747 /* We push at most this many items on the stack. */
1748 /* We used to use (num_regs - 1), which is the number of registers
1749 this regexp will save; but that was changed to 5
1750 to avoid stack overflow for a regexp with lots of parens. */
1751 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1753 /* We actually push this many items. */
1754 # define NUM_FAILURE_ITEMS \
1756 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1760 /* How many items can still be added to the stack without overflowing it. */
1761 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1762 # endif /* not DEFINED_ONCE */
1765 /* Pops what PUSH_FAIL_STACK pushes.
1767 We restore into the parameters, all of which should be lvalues:
1768 STR -- the saved data position.
1769 PAT -- the saved pattern position.
1770 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1771 REGSTART, REGEND -- arrays of string positions.
1772 REG_INFO -- array of information about each subexpression.
1774 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1775 `pend', `string1', `size1', `string2', and `size2'. */
1776 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1778 DEBUG_STATEMENT (unsigned failure_id;) \
1779 active_reg_t this_reg; \
1780 const UCHAR_T *string_temp; \
1782 assert (!FAIL_STACK_EMPTY ()); \
1784 /* Remove failure points and point to how many regs pushed. */ \
1785 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1786 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1787 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1789 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1791 DEBUG_POP (&failure_id); \
1792 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1794 /* If the saved string location is NULL, it came from an \
1795 on_failure_keep_string_jump opcode, and we want to throw away the \
1796 saved NULL, thus retaining our current position in the string. */ \
1797 string_temp = POP_FAILURE_POINTER (); \
1798 if (string_temp != NULL) \
1799 str = (const CHAR_T *) string_temp; \
1801 DEBUG_PRINT2 (" Popping string %p: `", str); \
1802 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1803 DEBUG_PRINT1 ("'\n"); \
1805 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1806 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1807 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1809 /* Restore register info. */ \
1810 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1811 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1813 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1814 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1817 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1819 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1821 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1822 DEBUG_PRINT2 (" info: %p\n", \
1823 reg_info[this_reg].word.pointer); \
1825 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1826 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1828 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1829 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1833 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1835 reg_info[this_reg].word.integer = 0; \
1836 regend[this_reg] = 0; \
1837 regstart[this_reg] = 0; \
1839 highest_active_reg = high_reg; \
1842 set_regs_matched_done = 0; \
1843 DEBUG_STATEMENT (nfailure_points_popped++); \
1844 } /* POP_FAILURE_POINT */
1846 /* Structure for per-register (a.k.a. per-group) information.
1847 Other register information, such as the
1848 starting and ending positions (which are addresses), and the list of
1849 inner groups (which is a bits list) are maintained in separate
1852 We are making a (strictly speaking) nonportable assumption here: that
1853 the compiler will pack our bit fields into something that fits into
1854 the type of `word', i.e., is something that fits into one item on the
1858 /* Declarations and macros for re_match_2. */
1862 PREFIX(fail_stack_elt_t) word;
1865 /* This field is one if this group can match the empty string,
1866 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1867 # define MATCH_NULL_UNSET_VALUE 3
1868 unsigned match_null_string_p : 2;
1869 unsigned is_active : 1;
1870 unsigned matched_something : 1;
1871 unsigned ever_matched_something : 1;
1873 } PREFIX(register_info_type);
1875 # ifndef DEFINED_ONCE
1876 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1877 # define IS_ACTIVE(R) ((R).bits.is_active)
1878 # define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1879 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1882 /* Call this when have matched a real character; it sets `matched' flags
1883 for the subexpressions which we are currently inside. Also records
1884 that those subexprs have matched. */
1885 # define SET_REGS_MATCHED() \
1888 if (!set_regs_matched_done) \
1891 set_regs_matched_done = 1; \
1892 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1894 MATCHED_SOMETHING (reg_info[r]) \
1895 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1901 # endif /* not DEFINED_ONCE */
1903 /* Registers are set to a sentinel when they haven't yet matched. */
1904 static CHAR_T PREFIX(reg_unset_dummy);
1905 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1906 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1908 /* Subroutine declarations and macros for regex_compile. */
1909 static void PREFIX(store_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, int arg));
1910 static void PREFIX(store_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1911 int arg1, int arg2));
1912 static void PREFIX(insert_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1913 int arg, UCHAR_T *end));
1914 static void PREFIX(insert_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1915 int arg1, int arg2, UCHAR_T *end));
1916 static boolean PREFIX(at_begline_loc_p) _RE_ARGS ((const CHAR_T *pattern,
1918 reg_syntax_t syntax));
1919 static boolean PREFIX(at_endline_loc_p) _RE_ARGS ((const CHAR_T *p,
1921 reg_syntax_t syntax));
1923 static reg_errcode_t wcs_compile_range _RE_ARGS ((CHAR_T range_start,
1924 const CHAR_T **p_ptr,
1927 reg_syntax_t syntax,
1930 static void insert_space _RE_ARGS ((int num, CHAR_T *loc, CHAR_T *end));
1932 static reg_errcode_t byte_compile_range _RE_ARGS ((unsigned int range_start,
1936 reg_syntax_t syntax,
1940 /* Fetch the next character in the uncompiled pattern---translating it
1941 if necessary. Also cast from a signed character in the constant
1942 string passed to us by the user to an unsigned char that we can use
1943 as an array index (in, e.g., `translate'). */
1944 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1945 because it is impossible to allocate 4GB array for some encodings
1946 which have 4 byte character_set like UCS4. */
1949 # define PATFETCH(c) \
1950 do {if (p == pend) return REG_EEND; \
1951 c = (UCHAR_T) *p++; \
1952 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1955 # define PATFETCH(c) \
1956 do {if (p == pend) return REG_EEND; \
1957 c = (unsigned char) *p++; \
1958 if (translate) c = (unsigned char) translate[c]; \
1963 /* Fetch the next character in the uncompiled pattern, with no
1965 # define PATFETCH_RAW(c) \
1966 do {if (p == pend) return REG_EEND; \
1967 c = (UCHAR_T) *p++; \
1970 /* Go backwards one character in the pattern. */
1971 # define PATUNFETCH p--
1974 /* If `translate' is non-null, return translate[D], else just D. We
1975 cast the subscript to translate because some data is declared as
1976 `char *', to avoid warnings when a string constant is passed. But
1977 when we use a character as a subscript we must make it unsigned. */
1978 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1979 because it is impossible to allocate 4GB array for some encodings
1980 which have 4 byte character_set like UCS4. */
1984 # define TRANSLATE(d) \
1985 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1986 ? (char) translate[(unsigned char) (d)] : (d))
1988 # define TRANSLATE(d) \
1989 (translate ? (char) translate[(unsigned char) (d)] : (d))
1994 /* Macros for outputting the compiled pattern into `buffer'. */
1996 /* If the buffer isn't allocated when it comes in, use this. */
1997 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
1999 /* Make sure we have at least N more bytes of space in buffer. */
2001 # define GET_BUFFER_SPACE(n) \
2002 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
2003 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
2006 # define GET_BUFFER_SPACE(n) \
2007 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
2011 /* Make sure we have one more byte of buffer space and then add C to it. */
2012 # define BUF_PUSH(c) \
2014 GET_BUFFER_SPACE (1); \
2015 *b++ = (UCHAR_T) (c); \
2019 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
2020 # define BUF_PUSH_2(c1, c2) \
2022 GET_BUFFER_SPACE (2); \
2023 *b++ = (UCHAR_T) (c1); \
2024 *b++ = (UCHAR_T) (c2); \
2028 /* As with BUF_PUSH_2, except for three bytes. */
2029 # define BUF_PUSH_3(c1, c2, c3) \
2031 GET_BUFFER_SPACE (3); \
2032 *b++ = (UCHAR_T) (c1); \
2033 *b++ = (UCHAR_T) (c2); \
2034 *b++ = (UCHAR_T) (c3); \
2037 /* Store a jump with opcode OP at LOC to location TO. We store a
2038 relative address offset by the three bytes the jump itself occupies. */
2039 # define STORE_JUMP(op, loc, to) \
2040 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
2042 /* Likewise, for a two-argument jump. */
2043 # define STORE_JUMP2(op, loc, to, arg) \
2044 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
2046 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
2047 # define INSERT_JUMP(op, loc, to) \
2048 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
2050 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
2051 # define INSERT_JUMP2(op, loc, to, arg) \
2052 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
2055 /* This is not an arbitrary limit: the arguments which represent offsets
2056 into the pattern are two bytes long. So if 2^16 bytes turns out to
2057 be too small, many things would have to change. */
2058 /* Any other compiler which, like MSC, has allocation limit below 2^16
2059 bytes will have to use approach similar to what was done below for
2060 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2061 reallocating to 0 bytes. Such thing is not going to work too well.
2062 You have been warned!! */
2063 # ifndef DEFINED_ONCE
2064 # if defined _MSC_VER && !defined WIN32
2065 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2066 The REALLOC define eliminates a flurry of conversion warnings,
2067 but is not required. */
2068 # define MAX_BUF_SIZE 65500L
2069 # define REALLOC(p,s) realloc ((p), (size_t) (s))
2071 # define MAX_BUF_SIZE (1L << 16)
2072 # define REALLOC(p,s) realloc ((p), (s))
2075 /* Extend the buffer by twice its current size via realloc and
2076 reset the pointers that pointed into the old block to point to the
2077 correct places in the new one. If extending the buffer results in it
2078 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2079 # if __BOUNDED_POINTERS__
2080 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2081 # define MOVE_BUFFER_POINTER(P) \
2082 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2083 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2086 SET_HIGH_BOUND (b); \
2087 SET_HIGH_BOUND (begalt); \
2088 if (fixup_alt_jump) \
2089 SET_HIGH_BOUND (fixup_alt_jump); \
2091 SET_HIGH_BOUND (laststart); \
2092 if (pending_exact) \
2093 SET_HIGH_BOUND (pending_exact); \
2096 # define MOVE_BUFFER_POINTER(P) (P) += incr
2097 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
2099 # endif /* not DEFINED_ONCE */
2102 # define EXTEND_BUFFER() \
2104 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2106 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2108 bufp->allocated <<= 1; \
2109 if (bufp->allocated > MAX_BUF_SIZE) \
2110 bufp->allocated = MAX_BUF_SIZE; \
2111 /* How many characters the new buffer can have? */ \
2112 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2113 if (wchar_count == 0) wchar_count = 1; \
2114 /* Truncate the buffer to CHAR_T align. */ \
2115 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2116 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2117 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2118 if (COMPILED_BUFFER_VAR == NULL) \
2119 return REG_ESPACE; \
2120 /* If the buffer moved, move all the pointers into it. */ \
2121 if (old_buffer != COMPILED_BUFFER_VAR) \
2123 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2124 MOVE_BUFFER_POINTER (b); \
2125 MOVE_BUFFER_POINTER (begalt); \
2126 if (fixup_alt_jump) \
2127 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2129 MOVE_BUFFER_POINTER (laststart); \
2130 if (pending_exact) \
2131 MOVE_BUFFER_POINTER (pending_exact); \
2133 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2136 # define EXTEND_BUFFER() \
2138 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2139 if (bufp->allocated == MAX_BUF_SIZE) \
2141 bufp->allocated <<= 1; \
2142 if (bufp->allocated > MAX_BUF_SIZE) \
2143 bufp->allocated = MAX_BUF_SIZE; \
2144 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2146 if (COMPILED_BUFFER_VAR == NULL) \
2147 return REG_ESPACE; \
2148 /* If the buffer moved, move all the pointers into it. */ \
2149 if (old_buffer != COMPILED_BUFFER_VAR) \
2151 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2152 MOVE_BUFFER_POINTER (b); \
2153 MOVE_BUFFER_POINTER (begalt); \
2154 if (fixup_alt_jump) \
2155 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2157 MOVE_BUFFER_POINTER (laststart); \
2158 if (pending_exact) \
2159 MOVE_BUFFER_POINTER (pending_exact); \
2161 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2165 # ifndef DEFINED_ONCE
2166 /* Since we have one byte reserved for the register number argument to
2167 {start,stop}_memory, the maximum number of groups we can report
2168 things about is what fits in that byte. */
2169 # define MAX_REGNUM 255
2171 /* But patterns can have more than `MAX_REGNUM' registers. We just
2172 ignore the excess. */
2173 typedef unsigned regnum_t;
2176 /* Macros for the compile stack. */
2178 /* Since offsets can go either forwards or backwards, this type needs to
2179 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2180 /* int may be not enough when sizeof(int) == 2. */
2181 typedef long pattern_offset_t;
2185 pattern_offset_t begalt_offset;
2186 pattern_offset_t fixup_alt_jump;
2187 pattern_offset_t inner_group_offset;
2188 pattern_offset_t laststart_offset;
2190 } compile_stack_elt_t;
2195 compile_stack_elt_t *stack;
2197 unsigned avail; /* Offset of next open position. */
2198 } compile_stack_type;
2201 # define INIT_COMPILE_STACK_SIZE 32
2203 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2204 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2206 /* The next available element. */
2207 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2209 # endif /* not DEFINED_ONCE */
2211 /* Set the bit for character C in a list. */
2212 # ifndef DEFINED_ONCE
2213 # define SET_LIST_BIT(c) \
2214 (b[((unsigned char) (c)) / BYTEWIDTH] \
2215 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2216 # endif /* DEFINED_ONCE */
2218 /* Get the next unsigned number in the uncompiled pattern. */
2219 # define GET_UNSIGNED_NUMBER(num) \
2224 if (c < '0' || c > '9') \
2226 if (num <= RE_DUP_MAX) \
2230 num = num * 10 + c - '0'; \
2235 # ifndef DEFINED_ONCE
2236 # if defined _LIBC || WIDE_CHAR_SUPPORT
2237 /* The GNU C library provides support for user-defined character classes
2238 and the functions from ISO C amendement 1. */
2239 # ifdef CHARCLASS_NAME_MAX
2240 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2242 /* This shouldn't happen but some implementation might still have this
2243 problem. Use a reasonable default value. */
2244 # define CHAR_CLASS_MAX_LENGTH 256
2248 # define IS_CHAR_CLASS(string) __wctype (string)
2250 # define IS_CHAR_CLASS(string) wctype (string)
2253 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2255 # define IS_CHAR_CLASS(string) \
2256 (STREQ (string, "alpha") || STREQ (string, "upper") \
2257 || STREQ (string, "lower") || STREQ (string, "digit") \
2258 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2259 || STREQ (string, "space") || STREQ (string, "print") \
2260 || STREQ (string, "punct") || STREQ (string, "graph") \
2261 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2263 # endif /* DEFINED_ONCE */
2265 # ifndef MATCH_MAY_ALLOCATE
2267 /* If we cannot allocate large objects within re_match_2_internal,
2268 we make the fail stack and register vectors global.
2269 The fail stack, we grow to the maximum size when a regexp
2271 The register vectors, we adjust in size each time we
2272 compile a regexp, according to the number of registers it needs. */
2274 static PREFIX(fail_stack_type) fail_stack;
2276 /* Size with which the following vectors are currently allocated.
2277 That is so we can make them bigger as needed,
2278 but never make them smaller. */
2279 # ifdef DEFINED_ONCE
2280 static int regs_allocated_size;
2282 static const char ** regstart, ** regend;
2283 static const char ** old_regstart, ** old_regend;
2284 static const char **best_regstart, **best_regend;
2285 static const char **reg_dummy;
2286 # endif /* DEFINED_ONCE */
2288 static PREFIX(register_info_type) *PREFIX(reg_info);
2289 static PREFIX(register_info_type) *PREFIX(reg_info_dummy);
2291 /* Make the register vectors big enough for NUM_REGS registers,
2292 but don't make them smaller. */
2295 PREFIX(regex_grow_registers) (num_regs)
2298 if (num_regs > regs_allocated_size)
2300 RETALLOC_IF (regstart, num_regs, const char *);
2301 RETALLOC_IF (regend, num_regs, const char *);
2302 RETALLOC_IF (old_regstart, num_regs, const char *);
2303 RETALLOC_IF (old_regend, num_regs, const char *);
2304 RETALLOC_IF (best_regstart, num_regs, const char *);
2305 RETALLOC_IF (best_regend, num_regs, const char *);
2306 RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type));
2307 RETALLOC_IF (reg_dummy, num_regs, const char *);
2308 RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type));
2310 regs_allocated_size = num_regs;
2314 # endif /* not MATCH_MAY_ALLOCATE */
2316 # ifndef DEFINED_ONCE
2317 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2320 # endif /* not DEFINED_ONCE */
2322 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2323 Returns one of error codes defined in `regex.h', or zero for success.
2325 Assumes the `allocated' (and perhaps `buffer') and `translate'
2326 fields are set in BUFP on entry.
2328 If it succeeds, results are put in BUFP (if it returns an error, the
2329 contents of BUFP are undefined):
2330 `buffer' is the compiled pattern;
2331 `syntax' is set to SYNTAX;
2332 `used' is set to the length of the compiled pattern;
2333 `fastmap_accurate' is zero;
2334 `re_nsub' is the number of subexpressions in PATTERN;
2335 `not_bol' and `not_eol' are zero;
2337 The `fastmap' and `newline_anchor' fields are neither
2338 examined nor set. */
2340 /* Return, freeing storage we allocated. */
2342 # define FREE_STACK_RETURN(value) \
2343 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2345 # define FREE_STACK_RETURN(value) \
2346 return (free (compile_stack.stack), value)
2349 static reg_errcode_t
2350 PREFIX(regex_compile) (ARG_PREFIX(pattern), ARG_PREFIX(size), syntax, bufp)
2351 const char *ARG_PREFIX(pattern);
2352 size_t ARG_PREFIX(size);
2353 reg_syntax_t syntax;
2354 struct re_pattern_buffer *bufp;
2356 /* We fetch characters from PATTERN here. Even though PATTERN is
2357 `char *' (i.e., signed), we declare these variables as unsigned, so
2358 they can be reliably used as array indices. */
2359 register UCHAR_T c, c1;
2362 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2363 CHAR_T *pattern, *COMPILED_BUFFER_VAR;
2365 /* offset buffer for optimization. See convert_mbs_to_wc. */
2366 int *mbs_offset = NULL;
2367 /* It hold whether each wchar_t is binary data or not. */
2368 char *is_binary = NULL;
2369 /* A flag whether exactn is handling binary data or not. */
2370 char is_exactn_bin = FALSE;
2373 /* A random temporary spot in PATTERN. */
2376 /* Points to the end of the buffer, where we should append. */
2377 register UCHAR_T *b;
2379 /* Keeps track of unclosed groups. */
2380 compile_stack_type compile_stack;
2382 /* Points to the current (ending) position in the pattern. */
2387 const CHAR_T *p = pattern;
2388 const CHAR_T *pend = pattern + size;
2391 /* How to translate the characters in the pattern. */
2392 RE_TRANSLATE_TYPE translate = bufp->translate;
2394 /* Address of the count-byte of the most recently inserted `exactn'
2395 command. This makes it possible to tell if a new exact-match
2396 character can be added to that command or if the character requires
2397 a new `exactn' command. */
2398 UCHAR_T *pending_exact = 0;
2400 /* Address of start of the most recently finished expression.
2401 This tells, e.g., postfix * where to find the start of its
2402 operand. Reset at the beginning of groups and alternatives. */
2403 UCHAR_T *laststart = 0;
2405 /* Address of beginning of regexp, or inside of last group. */
2408 /* Address of the place where a forward jump should go to the end of
2409 the containing expression. Each alternative of an `or' -- except the
2410 last -- ends with a forward jump of this sort. */
2411 UCHAR_T *fixup_alt_jump = 0;
2413 /* Counts open-groups as they are encountered. Remembered for the
2414 matching close-group on the compile stack, so the same register
2415 number is put in the stop_memory as the start_memory. */
2416 regnum_t regnum = 0;
2419 /* Initialize the wchar_t PATTERN and offset_buffer. */
2420 p = pend = pattern = TALLOC(csize + 1, CHAR_T);
2421 mbs_offset = TALLOC(csize + 1, int);
2422 is_binary = TALLOC(csize + 1, char);
2423 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL)
2430 pattern[csize] = L'\0'; /* sentinel */
2431 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary);
2443 DEBUG_PRINT1 ("\nCompiling pattern: ");
2446 unsigned debug_count;
2448 for (debug_count = 0; debug_count < size; debug_count++)
2449 PUT_CHAR (pattern[debug_count]);
2454 /* Initialize the compile stack. */
2455 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2456 if (compile_stack.stack == NULL)
2466 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2467 compile_stack.avail = 0;
2469 /* Initialize the pattern buffer. */
2470 bufp->syntax = syntax;
2471 bufp->fastmap_accurate = 0;
2472 bufp->not_bol = bufp->not_eol = 0;
2474 /* Set `used' to zero, so that if we return an error, the pattern
2475 printer (for debugging) will think there's no pattern. We reset it
2479 /* Always count groups, whether or not bufp->no_sub is set. */
2482 #if !defined emacs && !defined SYNTAX_TABLE
2483 /* Initialize the syntax table. */
2484 init_syntax_once ();
2487 if (bufp->allocated == 0)
2490 { /* If zero allocated, but buffer is non-null, try to realloc
2491 enough space. This loses if buffer's address is bogus, but
2492 that is the user's responsibility. */
2494 /* Free bufp->buffer and allocate an array for wchar_t pattern
2497 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T),
2500 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T);
2504 { /* Caller did not allocate a buffer. Do it for them. */
2505 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T),
2509 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE);
2511 bufp->buffer = (char*)COMPILED_BUFFER_VAR;
2513 bufp->allocated = INIT_BUF_SIZE;
2517 COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer;
2520 begalt = b = COMPILED_BUFFER_VAR;
2522 /* Loop through the uncompiled pattern until we're at the end. */
2531 if ( /* If at start of pattern, it's an operator. */
2533 /* If context independent, it's an operator. */
2534 || syntax & RE_CONTEXT_INDEP_ANCHORS
2535 /* Otherwise, depends on what's come before. */
2536 || PREFIX(at_begline_loc_p) (pattern, p, syntax))
2546 if ( /* If at end of pattern, it's an operator. */
2548 /* If context independent, it's an operator. */
2549 || syntax & RE_CONTEXT_INDEP_ANCHORS
2550 /* Otherwise, depends on what's next. */
2551 || PREFIX(at_endline_loc_p) (p, pend, syntax))
2561 if ((syntax & RE_BK_PLUS_QM)
2562 || (syntax & RE_LIMITED_OPS))
2566 /* If there is no previous pattern... */
2569 if (syntax & RE_CONTEXT_INVALID_OPS)
2570 FREE_STACK_RETURN (REG_BADRPT);
2571 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2576 /* Are we optimizing this jump? */
2577 boolean keep_string_p = false;
2579 /* 1 means zero (many) matches is allowed. */
2580 char zero_times_ok = 0, many_times_ok = 0;
2582 /* If there is a sequence of repetition chars, collapse it
2583 down to just one (the right one). We can't combine
2584 interval operators with these because of, e.g., `a{2}*',
2585 which should only match an even number of `a's. */
2589 zero_times_ok |= c != '+';
2590 many_times_ok |= c != '?';
2598 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2601 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2603 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2606 if (!(c1 == '+' || c1 == '?'))
2621 /* If we get here, we found another repeat character. */
2624 /* Star, etc. applied to an empty pattern is equivalent
2625 to an empty pattern. */
2629 /* Now we know whether or not zero matches is allowed
2630 and also whether or not two or more matches is allowed. */
2632 { /* More than one repetition is allowed, so put in at the
2633 end a backward relative jump from `b' to before the next
2634 jump we're going to put in below (which jumps from
2635 laststart to after this jump).
2637 But if we are at the `*' in the exact sequence `.*\n',
2638 insert an unconditional jump backwards to the .,
2639 instead of the beginning of the loop. This way we only
2640 push a failure point once, instead of every time
2641 through the loop. */
2642 assert (p - 1 > pattern);
2644 /* Allocate the space for the jump. */
2645 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2647 /* We know we are not at the first character of the pattern,
2648 because laststart was nonzero. And we've already
2649 incremented `p', by the way, to be the character after
2650 the `*'. Do we have to do something analogous here
2651 for null bytes, because of RE_DOT_NOT_NULL? */
2652 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2654 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2655 && !(syntax & RE_DOT_NEWLINE))
2656 { /* We have .*\n. */
2657 STORE_JUMP (jump, b, laststart);
2658 keep_string_p = true;
2661 /* Anything else. */
2662 STORE_JUMP (maybe_pop_jump, b, laststart -
2663 (1 + OFFSET_ADDRESS_SIZE));
2665 /* We've added more stuff to the buffer. */
2666 b += 1 + OFFSET_ADDRESS_SIZE;
2669 /* On failure, jump from laststart to b + 3, which will be the
2670 end of the buffer after this jump is inserted. */
2671 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2673 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2674 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2676 laststart, b + 1 + OFFSET_ADDRESS_SIZE);
2678 b += 1 + OFFSET_ADDRESS_SIZE;
2682 /* At least one repetition is required, so insert a
2683 `dummy_failure_jump' before the initial
2684 `on_failure_jump' instruction of the loop. This
2685 effects a skip over that instruction the first time
2686 we hit that loop. */
2687 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2688 INSERT_JUMP (dummy_failure_jump, laststart, laststart +
2689 2 + 2 * OFFSET_ADDRESS_SIZE);
2690 b += 1 + OFFSET_ADDRESS_SIZE;
2704 boolean had_char_class = false;
2706 CHAR_T range_start = 0xffffffff;
2708 unsigned int range_start = 0xffffffff;
2710 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2713 /* We assume a charset(_not) structure as a wchar_t array.
2714 charset[0] = (re_opcode_t) charset(_not)
2715 charset[1] = l (= length of char_classes)
2716 charset[2] = m (= length of collating_symbols)
2717 charset[3] = n (= length of equivalence_classes)
2718 charset[4] = o (= length of char_ranges)
2719 charset[5] = p (= length of chars)
2721 charset[6] = char_class (wctype_t)
2722 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2724 charset[l+5] = char_class (wctype_t)
2726 charset[l+6] = collating_symbol (wchar_t)
2728 charset[l+m+5] = collating_symbol (wchar_t)
2729 ifdef _LIBC we use the index if
2730 _NL_COLLATE_SYMB_EXTRAMB instead of
2733 charset[l+m+6] = equivalence_classes (wchar_t)
2735 charset[l+m+n+5] = equivalence_classes (wchar_t)
2736 ifdef _LIBC we use the index in
2737 _NL_COLLATE_WEIGHT instead of
2740 charset[l+m+n+6] = range_start
2741 charset[l+m+n+7] = range_end
2743 charset[l+m+n+2o+4] = range_start
2744 charset[l+m+n+2o+5] = range_end
2745 ifdef _LIBC we use the value looked up
2746 in _NL_COLLATE_COLLSEQ instead of
2749 charset[l+m+n+2o+6] = char
2751 charset[l+m+n+2o+p+5] = char
2755 /* We need at least 6 spaces: the opcode, the length of
2756 char_classes, the length of collating_symbols, the length of
2757 equivalence_classes, the length of char_ranges, the length of
2759 GET_BUFFER_SPACE (6);
2761 /* Save b as laststart. And We use laststart as the pointer
2762 to the first element of the charset here.
2763 In other words, laststart[i] indicates charset[i]. */
2766 /* We test `*p == '^' twice, instead of using an if
2767 statement, so we only need one BUF_PUSH. */
2768 BUF_PUSH (*p == '^' ? charset_not : charset);
2772 /* Push the length of char_classes, the length of
2773 collating_symbols, the length of equivalence_classes, the
2774 length of char_ranges and the length of chars. */
2775 BUF_PUSH_3 (0, 0, 0);
2778 /* Remember the first position in the bracket expression. */
2781 /* charset_not matches newline according to a syntax bit. */
2782 if ((re_opcode_t) b[-6] == charset_not
2783 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2786 laststart[5]++; /* Update the length of characters */
2789 /* Read in characters and ranges, setting map bits. */
2792 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2796 /* \ might escape characters inside [...] and [^...]. */
2797 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2799 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2803 laststart[5]++; /* Update the length of chars */
2808 /* Could be the end of the bracket expression. If it's
2809 not (i.e., when the bracket expression is `[]' so
2810 far), the ']' character bit gets set way below. */
2811 if (c == ']' && p != p1 + 1)
2814 /* Look ahead to see if it's a range when the last thing
2815 was a character class. */
2816 if (had_char_class && c == '-' && *p != ']')
2817 FREE_STACK_RETURN (REG_ERANGE);
2819 /* Look ahead to see if it's a range when the last thing
2820 was a character: if this is a hyphen not at the
2821 beginning or the end of a list, then it's the range
2824 && !(p - 2 >= pattern && p[-2] == '[')
2825 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2829 /* Allocate the space for range_start and range_end. */
2830 GET_BUFFER_SPACE (2);
2831 /* Update the pointer to indicate end of buffer. */
2833 ret = wcs_compile_range (range_start, &p, pend, translate,
2834 syntax, b, laststart);
2835 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2836 range_start = 0xffffffff;
2838 else if (p[0] == '-' && p[1] != ']')
2839 { /* This handles ranges made up of characters only. */
2842 /* Move past the `-'. */
2844 /* Allocate the space for range_start and range_end. */
2845 GET_BUFFER_SPACE (2);
2846 /* Update the pointer to indicate end of buffer. */
2848 ret = wcs_compile_range (c, &p, pend, translate, syntax, b,
2850 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2851 range_start = 0xffffffff;
2854 /* See if we're at the beginning of a possible character
2856 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2857 { /* Leave room for the null. */
2858 char str[CHAR_CLASS_MAX_LENGTH + 1];
2863 /* If pattern is `[[:'. */
2864 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2869 if ((c == ':' && *p == ']') || p == pend)
2871 if (c1 < CHAR_CLASS_MAX_LENGTH)
2874 /* This is in any case an invalid class name. */
2879 /* If isn't a word bracketed by `[:' and `:]':
2880 undo the ending character, the letters, and leave
2881 the leading `:' and `[' (but store them as character). */
2882 if (c == ':' && *p == ']')
2887 /* Query the character class as wctype_t. */
2888 wt = IS_CHAR_CLASS (str);
2890 FREE_STACK_RETURN (REG_ECTYPE);
2892 /* Throw away the ] at the end of the character
2896 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2898 /* Allocate the space for character class. */
2899 GET_BUFFER_SPACE(CHAR_CLASS_SIZE);
2900 /* Update the pointer to indicate end of buffer. */
2901 b += CHAR_CLASS_SIZE;
2902 /* Move data which follow character classes
2903 not to violate the data. */
2904 insert_space(CHAR_CLASS_SIZE,
2905 laststart + 6 + laststart[1],
2907 alignedp = ((uintptr_t)(laststart + 6 + laststart[1])
2908 + __alignof__(wctype_t) - 1)
2909 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2910 /* Store the character class. */
2911 *((wctype_t*)alignedp) = wt;
2912 /* Update length of char_classes */
2913 laststart[1] += CHAR_CLASS_SIZE;
2915 had_char_class = true;
2924 laststart[5] += 2; /* Update the length of characters */
2926 had_char_class = false;
2929 else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '='
2932 CHAR_T str[128]; /* Should be large enough. */
2933 CHAR_T delim = *p; /* '=' or '.' */
2936 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2941 /* If pattern is `[[=' or '[[.'. */
2942 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2947 if ((c == delim && *p == ']') || p == pend)
2949 if (c1 < sizeof (str) - 1)
2952 /* This is in any case an invalid class name. */
2957 if (c == delim && *p == ']' && str[0] != '\0')
2959 unsigned int i, offset;
2960 /* If we have no collation data we use the default
2961 collation in which each character is in a class
2962 by itself. It also means that ASCII is the
2963 character set and therefore we cannot have character
2964 with more than one byte in the multibyte
2967 /* If not defined _LIBC, we push the name and
2968 `\0' for the sake of matching performance. */
2969 int datasize = c1 + 1;
2977 FREE_STACK_RETURN (REG_ECOLLATE);
2982 const int32_t *table;
2983 const int32_t *weights;
2984 const int32_t *extra;
2985 const int32_t *indirect;
2988 /* This #include defines a local function! */
2989 # include <locale/weightwc.h>
2993 /* We push the index for equivalence class. */
2996 table = (const int32_t *)
2997 _NL_CURRENT (LC_COLLATE,
2998 _NL_COLLATE_TABLEWC);
2999 weights = (const int32_t *)
3000 _NL_CURRENT (LC_COLLATE,
3001 _NL_COLLATE_WEIGHTWC);
3002 extra = (const int32_t *)
3003 _NL_CURRENT (LC_COLLATE,
3004 _NL_COLLATE_EXTRAWC);
3005 indirect = (const int32_t *)
3006 _NL_CURRENT (LC_COLLATE,
3007 _NL_COLLATE_INDIRECTWC);
3009 idx = findidx ((const wint_t**)&cp);
3010 if (idx == 0 || cp < (wint_t*) str + c1)
3011 /* This is no valid character. */
3012 FREE_STACK_RETURN (REG_ECOLLATE);
3014 str[0] = (wchar_t)idx;
3016 else /* delim == '.' */
3018 /* We push collation sequence value
3019 for collating symbol. */
3021 const int32_t *symb_table;
3022 const unsigned char *extra;
3029 /* We have to convert the name to a single-byte
3030 string. This is possible since the names
3031 consist of ASCII characters and the internal
3032 representation is UCS4. */
3033 for (i = 0; i < c1; ++i)
3034 char_str[i] = str[i];
3037 _NL_CURRENT_WORD (LC_COLLATE,
3038 _NL_COLLATE_SYMB_HASH_SIZEMB);
3039 symb_table = (const int32_t *)
3040 _NL_CURRENT (LC_COLLATE,
3041 _NL_COLLATE_SYMB_TABLEMB);
3042 extra = (const unsigned char *)
3043 _NL_CURRENT (LC_COLLATE,
3044 _NL_COLLATE_SYMB_EXTRAMB);
3046 /* Locate the character in the hashing table. */
3047 hash = elem_hash (char_str, c1);
3050 elem = hash % table_size;
3051 second = hash % (table_size - 2);
3052 while (symb_table[2 * elem] != 0)
3054 /* First compare the hashing value. */
3055 if (symb_table[2 * elem] == hash
3056 && c1 == extra[symb_table[2 * elem + 1]]
3057 && memcmp (char_str,
3058 &extra[symb_table[2 * elem + 1]
3061 /* Yep, this is the entry. */
3062 idx = symb_table[2 * elem + 1];
3063 idx += 1 + extra[idx];
3071 if (symb_table[2 * elem] != 0)
3073 /* Compute the index of the byte sequence
3075 idx += 1 + extra[idx];
3076 /* Adjust for the alignment. */
3077 idx = (idx + 3) & ~3;
3079 str[0] = (wchar_t) idx + 4;
3081 else if (symb_table[2 * elem] == 0 && c1 == 1)
3083 /* No valid character. Match it as a
3084 single byte character. */
3085 had_char_class = false;
3087 /* Update the length of characters */
3089 range_start = str[0];
3091 /* Throw away the ] at the end of the
3092 collating symbol. */
3094 /* exit from the switch block. */
3098 FREE_STACK_RETURN (REG_ECOLLATE);
3103 /* Throw away the ] at the end of the equivalence
3104 class (or collating symbol). */
3107 /* Allocate the space for the equivalence class
3108 (or collating symbol) (and '\0' if needed). */
3109 GET_BUFFER_SPACE(datasize);
3110 /* Update the pointer to indicate end of buffer. */
3114 { /* equivalence class */
3115 /* Calculate the offset of char_ranges,
3116 which is next to equivalence_classes. */
3117 offset = laststart[1] + laststart[2]
3120 insert_space(datasize, laststart + offset, b - 1);
3122 /* Write the equivalence_class and \0. */
3123 for (i = 0 ; i < datasize ; i++)
3124 laststart[offset + i] = str[i];
3126 /* Update the length of equivalence_classes. */
3127 laststart[3] += datasize;
3128 had_char_class = true;
3130 else /* delim == '.' */
3131 { /* collating symbol */
3132 /* Calculate the offset of the equivalence_classes,
3133 which is next to collating_symbols. */
3134 offset = laststart[1] + laststart[2] + 6;
3135 /* Insert space and write the collationg_symbol
3137 insert_space(datasize, laststart + offset, b-1);
3138 for (i = 0 ; i < datasize ; i++)
3139 laststart[offset + i] = str[i];
3141 /* In re_match_2_internal if range_start < -1, we
3142 assume -range_start is the offset of the
3143 collating symbol which is specified as
3144 the character of the range start. So we assign
3145 -(laststart[1] + laststart[2] + 6) to
3147 range_start = -(laststart[1] + laststart[2] + 6);
3148 /* Update the length of collating_symbol. */
3149 laststart[2] += datasize;
3150 had_char_class = false;
3160 laststart[5] += 2; /* Update the length of characters */
3161 range_start = delim;
3162 had_char_class = false;
3167 had_char_class = false;
3169 laststart[5]++; /* Update the length of characters */
3175 /* Ensure that we have enough space to push a charset: the
3176 opcode, the length count, and the bitset; 34 bytes in all. */
3177 GET_BUFFER_SPACE (34);
3181 /* We test `*p == '^' twice, instead of using an if
3182 statement, so we only need one BUF_PUSH. */
3183 BUF_PUSH (*p == '^' ? charset_not : charset);
3187 /* Remember the first position in the bracket expression. */
3190 /* Push the number of bytes in the bitmap. */
3191 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
3193 /* Clear the whole map. */
3194 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
3196 /* charset_not matches newline according to a syntax bit. */
3197 if ((re_opcode_t) b[-2] == charset_not
3198 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
3199 SET_LIST_BIT ('\n');
3201 /* Read in characters and ranges, setting map bits. */
3204 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3208 /* \ might escape characters inside [...] and [^...]. */
3209 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
3211 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3219 /* Could be the end of the bracket expression. If it's
3220 not (i.e., when the bracket expression is `[]' so
3221 far), the ']' character bit gets set way below. */
3222 if (c == ']' && p != p1 + 1)
3225 /* Look ahead to see if it's a range when the last thing
3226 was a character class. */
3227 if (had_char_class && c == '-' && *p != ']')
3228 FREE_STACK_RETURN (REG_ERANGE);
3230 /* Look ahead to see if it's a range when the last thing
3231 was a character: if this is a hyphen not at the
3232 beginning or the end of a list, then it's the range
3235 && !(p - 2 >= pattern && p[-2] == '[')
3236 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
3240 = byte_compile_range (range_start, &p, pend, translate,
3242 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3243 range_start = 0xffffffff;
3246 else if (p[0] == '-' && p[1] != ']')
3247 { /* This handles ranges made up of characters only. */
3250 /* Move past the `-'. */
3253 ret = byte_compile_range (c, &p, pend, translate, syntax, b);
3254 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3255 range_start = 0xffffffff;
3258 /* See if we're at the beginning of a possible character
3261 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3262 { /* Leave room for the null. */
3263 char str[CHAR_CLASS_MAX_LENGTH + 1];
3268 /* If pattern is `[[:'. */
3269 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3274 if ((c == ':' && *p == ']') || p == pend)
3276 if (c1 < CHAR_CLASS_MAX_LENGTH)
3279 /* This is in any case an invalid class name. */
3284 /* If isn't a word bracketed by `[:' and `:]':
3285 undo the ending character, the letters, and leave
3286 the leading `:' and `[' (but set bits for them). */
3287 if (c == ':' && *p == ']')
3289 # if defined _LIBC || WIDE_CHAR_SUPPORT
3290 boolean is_lower = STREQ (str, "lower");
3291 boolean is_upper = STREQ (str, "upper");
3295 wt = IS_CHAR_CLASS (str);
3297 FREE_STACK_RETURN (REG_ECTYPE);
3299 /* Throw away the ] at the end of the character
3303 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3305 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
3308 if (__iswctype (__btowc (ch), wt))
3311 if (iswctype (btowc (ch), wt))
3315 if (translate && (is_upper || is_lower)
3316 && (ISUPPER (ch) || ISLOWER (ch)))
3320 had_char_class = true;
3323 boolean is_alnum = STREQ (str, "alnum");
3324 boolean is_alpha = STREQ (str, "alpha");
3325 boolean is_blank = STREQ (str, "blank");
3326 boolean is_cntrl = STREQ (str, "cntrl");
3327 boolean is_digit = STREQ (str, "digit");
3328 boolean is_graph = STREQ (str, "graph");
3329 boolean is_lower = STREQ (str, "lower");
3330 boolean is_print = STREQ (str, "print");
3331 boolean is_punct = STREQ (str, "punct");
3332 boolean is_space = STREQ (str, "space");
3333 boolean is_upper = STREQ (str, "upper");
3334 boolean is_xdigit = STREQ (str, "xdigit");
3336 if (!IS_CHAR_CLASS (str))
3337 FREE_STACK_RETURN (REG_ECTYPE);
3339 /* Throw away the ] at the end of the character
3343 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3345 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
3347 /* This was split into 3 if's to
3348 avoid an arbitrary limit in some compiler. */
3349 if ( (is_alnum && ISALNUM (ch))
3350 || (is_alpha && ISALPHA (ch))
3351 || (is_blank && ISBLANK (ch))
3352 || (is_cntrl && ISCNTRL (ch)))
3354 if ( (is_digit && ISDIGIT (ch))
3355 || (is_graph && ISGRAPH (ch))
3356 || (is_lower && ISLOWER (ch))
3357 || (is_print && ISPRINT (ch)))
3359 if ( (is_punct && ISPUNCT (ch))
3360 || (is_space && ISSPACE (ch))
3361 || (is_upper && ISUPPER (ch))
3362 || (is_xdigit && ISXDIGIT (ch)))
3364 if ( translate && (is_upper || is_lower)
3365 && (ISUPPER (ch) || ISLOWER (ch)))
3368 had_char_class = true;
3369 # endif /* libc || wctype.h */
3379 had_char_class = false;
3382 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
3384 unsigned char str[MB_LEN_MAX + 1];
3387 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3393 /* If pattern is `[[='. */
3394 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3399 if ((c == '=' && *p == ']') || p == pend)
3401 if (c1 < MB_LEN_MAX)
3404 /* This is in any case an invalid class name. */
3409 if (c == '=' && *p == ']' && str[0] != '\0')
3411 /* If we have no collation data we use the default
3412 collation in which each character is in a class
3413 by itself. It also means that ASCII is the
3414 character set and therefore we cannot have character
3415 with more than one byte in the multibyte
3422 FREE_STACK_RETURN (REG_ECOLLATE);
3424 /* Throw away the ] at the end of the equivalence
3428 /* Set the bit for the character. */
3429 SET_LIST_BIT (str[0]);
3434 /* Try to match the byte sequence in `str' against
3435 those known to the collate implementation.
3436 First find out whether the bytes in `str' are
3437 actually from exactly one character. */
3438 const int32_t *table;
3439 const unsigned char *weights;
3440 const unsigned char *extra;
3441 const int32_t *indirect;
3443 const unsigned char *cp = str;
3446 /* This #include defines a local function! */
3447 # include <locale/weight.h>
3449 table = (const int32_t *)
3450 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3451 weights = (const unsigned char *)
3452 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3453 extra = (const unsigned char *)
3454 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3455 indirect = (const int32_t *)
3456 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3458 idx = findidx (&cp);
3459 if (idx == 0 || cp < str + c1)
3460 /* This is no valid character. */
3461 FREE_STACK_RETURN (REG_ECOLLATE);
3463 /* Throw away the ] at the end of the equivalence
3467 /* Now we have to go throught the whole table
3468 and find all characters which have the same
3471 XXX Note that this is not entirely correct.
3472 we would have to match multibyte sequences
3473 but this is not possible with the current
3475 for (ch = 1; ch < 256; ++ch)
3476 /* XXX This test would have to be changed if we
3477 would allow matching multibyte sequences. */
3480 int32_t idx2 = table[ch];
3481 size_t len = weights[idx2];
3483 /* Test whether the lenghts match. */
3484 if (weights[idx] == len)
3486 /* They do. New compare the bytes of
3491 && (weights[idx + 1 + cnt]
3492 == weights[idx2 + 1 + cnt]))
3496 /* They match. Mark the character as
3503 had_char_class = true;
3513 had_char_class = false;
3516 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
3518 unsigned char str[128]; /* Should be large enough. */
3521 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3527 /* If pattern is `[[.'. */
3528 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3533 if ((c == '.' && *p == ']') || p == pend)
3535 if (c1 < sizeof (str))
3538 /* This is in any case an invalid class name. */
3543 if (c == '.' && *p == ']' && str[0] != '\0')
3545 /* If we have no collation data we use the default
3546 collation in which each character is the name
3547 for its own class which contains only the one
3548 character. It also means that ASCII is the
3549 character set and therefore we cannot have character
3550 with more than one byte in the multibyte
3557 FREE_STACK_RETURN (REG_ECOLLATE);
3559 /* Throw away the ] at the end of the equivalence
3563 /* Set the bit for the character. */
3564 SET_LIST_BIT (str[0]);
3565 range_start = ((const unsigned char *) str)[0];
3570 /* Try to match the byte sequence in `str' against
3571 those known to the collate implementation.
3572 First find out whether the bytes in `str' are
3573 actually from exactly one character. */
3575 const int32_t *symb_table;
3576 const unsigned char *extra;
3583 _NL_CURRENT_WORD (LC_COLLATE,
3584 _NL_COLLATE_SYMB_HASH_SIZEMB);
3585 symb_table = (const int32_t *)
3586 _NL_CURRENT (LC_COLLATE,
3587 _NL_COLLATE_SYMB_TABLEMB);
3588 extra = (const unsigned char *)
3589 _NL_CURRENT (LC_COLLATE,
3590 _NL_COLLATE_SYMB_EXTRAMB);
3592 /* Locate the character in the hashing table. */
3593 hash = elem_hash (str, c1);
3596 elem = hash % table_size;
3597 second = hash % (table_size - 2);
3598 while (symb_table[2 * elem] != 0)
3600 /* First compare the hashing value. */
3601 if (symb_table[2 * elem] == hash
3602 && c1 == extra[symb_table[2 * elem + 1]]
3604 &extra[symb_table[2 * elem + 1]
3608 /* Yep, this is the entry. */
3609 idx = symb_table[2 * elem + 1];
3610 idx += 1 + extra[idx];
3618 if (symb_table[2 * elem] == 0)
3619 /* This is no valid character. */
3620 FREE_STACK_RETURN (REG_ECOLLATE);
3622 /* Throw away the ] at the end of the equivalence
3626 /* Now add the multibyte character(s) we found
3629 XXX Note that this is not entirely correct.
3630 we would have to match multibyte sequences
3631 but this is not possible with the current
3632 implementation. Also, we have to match
3633 collating symbols, which expand to more than
3634 one file, as a whole and not allow the
3635 individual bytes. */
3638 range_start = extra[idx];
3641 SET_LIST_BIT (extra[idx]);
3646 had_char_class = false;
3656 had_char_class = false;
3661 had_char_class = false;
3667 /* Discard any (non)matching list bytes that are all 0 at the
3668 end of the map. Decrease the map-length byte too. */
3669 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3678 if (syntax & RE_NO_BK_PARENS)
3685 if (syntax & RE_NO_BK_PARENS)
3692 if (syntax & RE_NEWLINE_ALT)
3699 if (syntax & RE_NO_BK_VBAR)
3706 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3707 goto handle_interval;
3713 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3715 /* Do not translate the character after the \, so that we can
3716 distinguish, e.g., \B from \b, even if we normally would
3717 translate, e.g., B to b. */
3723 if (syntax & RE_NO_BK_PARENS)
3724 goto normal_backslash;
3730 if (COMPILE_STACK_FULL)
3732 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3733 compile_stack_elt_t);
3734 if (compile_stack.stack == NULL) return REG_ESPACE;
3736 compile_stack.size <<= 1;
3739 /* These are the values to restore when we hit end of this
3740 group. They are all relative offsets, so that if the
3741 whole pattern moves because of realloc, they will still
3743 COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR;
3744 COMPILE_STACK_TOP.fixup_alt_jump
3745 = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0;
3746 COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR;
3747 COMPILE_STACK_TOP.regnum = regnum;
3749 /* We will eventually replace the 0 with the number of
3750 groups inner to this one. But do not push a
3751 start_memory for groups beyond the last one we can
3752 represent in the compiled pattern. */
3753 if (regnum <= MAX_REGNUM)
3755 COMPILE_STACK_TOP.inner_group_offset = b
3756 - COMPILED_BUFFER_VAR + 2;
3757 BUF_PUSH_3 (start_memory, regnum, 0);
3760 compile_stack.avail++;
3765 /* If we've reached MAX_REGNUM groups, then this open
3766 won't actually generate any code, so we'll have to
3767 clear pending_exact explicitly. */
3773 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3775 if (COMPILE_STACK_EMPTY)
3777 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3778 goto normal_backslash;
3780 FREE_STACK_RETURN (REG_ERPAREN);
3785 { /* Push a dummy failure point at the end of the
3786 alternative for a possible future
3787 `pop_failure_jump' to pop. See comments at
3788 `push_dummy_failure' in `re_match_2'. */
3789 BUF_PUSH (push_dummy_failure);
3791 /* We allocated space for this jump when we assigned
3792 to `fixup_alt_jump', in the `handle_alt' case below. */
3793 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
3796 /* See similar code for backslashed left paren above. */
3797 if (COMPILE_STACK_EMPTY)
3799 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3802 FREE_STACK_RETURN (REG_ERPAREN);
3805 /* Since we just checked for an empty stack above, this
3806 ``can't happen''. */
3807 assert (compile_stack.avail != 0);
3809 /* We don't just want to restore into `regnum', because
3810 later groups should continue to be numbered higher,
3811 as in `(ab)c(de)' -- the second group is #2. */
3812 regnum_t this_group_regnum;
3814 compile_stack.avail--;
3815 begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset;
3817 = COMPILE_STACK_TOP.fixup_alt_jump
3818 ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1
3820 laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset;
3821 this_group_regnum = COMPILE_STACK_TOP.regnum;
3822 /* If we've reached MAX_REGNUM groups, then this open
3823 won't actually generate any code, so we'll have to
3824 clear pending_exact explicitly. */
3827 /* We're at the end of the group, so now we know how many
3828 groups were inside this one. */
3829 if (this_group_regnum <= MAX_REGNUM)
3831 UCHAR_T *inner_group_loc
3832 = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset;
3834 *inner_group_loc = regnum - this_group_regnum;
3835 BUF_PUSH_3 (stop_memory, this_group_regnum,
3836 regnum - this_group_regnum);
3842 case '|': /* `\|'. */
3843 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3844 goto normal_backslash;
3846 if (syntax & RE_LIMITED_OPS)
3849 /* Insert before the previous alternative a jump which
3850 jumps to this alternative if the former fails. */
3851 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3852 INSERT_JUMP (on_failure_jump, begalt,
3853 b + 2 + 2 * OFFSET_ADDRESS_SIZE);
3855 b += 1 + OFFSET_ADDRESS_SIZE;
3857 /* The alternative before this one has a jump after it
3858 which gets executed if it gets matched. Adjust that
3859 jump so it will jump to this alternative's analogous
3860 jump (put in below, which in turn will jump to the next
3861 (if any) alternative's such jump, etc.). The last such
3862 jump jumps to the correct final destination. A picture:
3868 If we are at `b', then fixup_alt_jump right now points to a
3869 three-byte space after `a'. We'll put in the jump, set
3870 fixup_alt_jump to right after `b', and leave behind three
3871 bytes which we'll fill in when we get to after `c'. */
3874 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3876 /* Mark and leave space for a jump after this alternative,
3877 to be filled in later either by next alternative or
3878 when know we're at the end of a series of alternatives. */
3880 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3881 b += 1 + OFFSET_ADDRESS_SIZE;
3889 /* If \{ is a literal. */
3890 if (!(syntax & RE_INTERVALS)
3891 /* If we're at `\{' and it's not the open-interval
3893 || (syntax & RE_NO_BK_BRACES))
3894 goto normal_backslash;
3898 /* If got here, then the syntax allows intervals. */
3900 /* At least (most) this many matches must be made. */
3901 int lower_bound = -1, upper_bound = -1;
3903 /* Place in the uncompiled pattern (i.e., just after
3904 the '{') to go back to if the interval is invalid. */
3905 const CHAR_T *beg_interval = p;
3908 goto invalid_interval;
3910 GET_UNSIGNED_NUMBER (lower_bound);
3914 GET_UNSIGNED_NUMBER (upper_bound);
3915 if (upper_bound < 0)
3916 upper_bound = RE_DUP_MAX;
3919 /* Interval such as `{1}' => match exactly once. */
3920 upper_bound = lower_bound;
3922 if (! (0 <= lower_bound && lower_bound <= upper_bound))
3923 goto invalid_interval;
3925 if (!(syntax & RE_NO_BK_BRACES))
3927 if (c != '\\' || p == pend)
3928 goto invalid_interval;
3933 goto invalid_interval;
3935 /* If it's invalid to have no preceding re. */
3938 if (syntax & RE_CONTEXT_INVALID_OPS
3939 && !(syntax & RE_INVALID_INTERVAL_ORD))
3940 FREE_STACK_RETURN (REG_BADRPT);
3941 else if (syntax & RE_CONTEXT_INDEP_OPS)
3944 goto unfetch_interval;
3947 /* We just parsed a valid interval. */
3949 if (RE_DUP_MAX < upper_bound)
3950 FREE_STACK_RETURN (REG_BADBR);
3952 /* If the upper bound is zero, don't want to succeed at
3953 all; jump from `laststart' to `b + 3', which will be
3954 the end of the buffer after we insert the jump. */
3955 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3956 instead of 'b + 3'. */
3957 if (upper_bound == 0)
3959 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3960 INSERT_JUMP (jump, laststart, b + 1
3961 + OFFSET_ADDRESS_SIZE);
3962 b += 1 + OFFSET_ADDRESS_SIZE;
3965 /* Otherwise, we have a nontrivial interval. When
3966 we're all done, the pattern will look like:
3967 set_number_at <jump count> <upper bound>
3968 set_number_at <succeed_n count> <lower bound>
3969 succeed_n <after jump addr> <succeed_n count>
3971 jump_n <succeed_n addr> <jump count>
3972 (The upper bound and `jump_n' are omitted if
3973 `upper_bound' is 1, though.) */
3975 { /* If the upper bound is > 1, we need to insert
3976 more at the end of the loop. */
3977 unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE +
3978 (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE);
3980 GET_BUFFER_SPACE (nbytes);
3982 /* Initialize lower bound of the `succeed_n', even
3983 though it will be set during matching by its
3984 attendant `set_number_at' (inserted next),
3985 because `re_compile_fastmap' needs to know.
3986 Jump to the `jump_n' we might insert below. */
3987 INSERT_JUMP2 (succeed_n, laststart,
3988 b + 1 + 2 * OFFSET_ADDRESS_SIZE
3989 + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE)
3991 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3993 /* Code to initialize the lower bound. Insert
3994 before the `succeed_n'. The `5' is the last two
3995 bytes of this `set_number_at', plus 3 bytes of
3996 the following `succeed_n'. */
3997 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
3998 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
3999 of the following `succeed_n'. */
4000 PREFIX(insert_op2) (set_number_at, laststart, 1
4001 + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b);
4002 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4004 if (upper_bound > 1)
4005 { /* More than one repetition is allowed, so
4006 append a backward jump to the `succeed_n'
4007 that starts this interval.
4009 When we've reached this during matching,
4010 we'll have matched the interval once, so
4011 jump back only `upper_bound - 1' times. */
4012 STORE_JUMP2 (jump_n, b, laststart
4013 + 2 * OFFSET_ADDRESS_SIZE + 1,
4015 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4017 /* The location we want to set is the second
4018 parameter of the `jump_n'; that is `b-2' as
4019 an absolute address. `laststart' will be
4020 the `set_number_at' we're about to insert;
4021 `laststart+3' the number to set, the source
4022 for the relative address. But we are
4023 inserting into the middle of the pattern --
4024 so everything is getting moved up by 5.
4025 Conclusion: (b - 2) - (laststart + 3) + 5,
4026 i.e., b - laststart.
4028 We insert this at the beginning of the loop
4029 so that if we fail during matching, we'll
4030 reinitialize the bounds. */
4031 PREFIX(insert_op2) (set_number_at, laststart,
4033 upper_bound - 1, b);
4034 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4041 if (!(syntax & RE_INVALID_INTERVAL_ORD))
4042 FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR);
4044 /* Match the characters as literals. */
4047 if (syntax & RE_NO_BK_BRACES)
4050 goto normal_backslash;
4054 /* There is no way to specify the before_dot and after_dot
4055 operators. rms says this is ok. --karl */
4063 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
4069 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
4075 if (syntax & RE_NO_GNU_OPS)
4078 BUF_PUSH (wordchar);
4083 if (syntax & RE_NO_GNU_OPS)
4086 BUF_PUSH (notwordchar);
4091 if (syntax & RE_NO_GNU_OPS)
4097 if (syntax & RE_NO_GNU_OPS)
4103 if (syntax & RE_NO_GNU_OPS)
4105 BUF_PUSH (wordbound);
4109 if (syntax & RE_NO_GNU_OPS)
4111 BUF_PUSH (notwordbound);
4115 if (syntax & RE_NO_GNU_OPS)
4121 if (syntax & RE_NO_GNU_OPS)
4126 case '1': case '2': case '3': case '4': case '5':
4127 case '6': case '7': case '8': case '9':
4128 if (syntax & RE_NO_BK_REFS)
4134 FREE_STACK_RETURN (REG_ESUBREG);
4136 /* Can't back reference to a subexpression if inside of it. */
4137 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
4141 BUF_PUSH_2 (duplicate, c1);
4147 if (syntax & RE_BK_PLUS_QM)
4150 goto normal_backslash;
4154 /* You might think it would be useful for \ to mean
4155 not to translate; but if we don't translate it
4156 it will never match anything. */
4164 /* Expects the character in `c'. */
4166 /* If no exactn currently being built. */
4169 /* If last exactn handle binary(or character) and
4170 new exactn handle character(or binary). */
4171 || is_exactn_bin != is_binary[p - 1 - pattern]
4174 /* If last exactn not at current position. */
4175 || pending_exact + *pending_exact + 1 != b
4177 /* We have only one byte following the exactn for the count. */
4178 || *pending_exact == (1 << BYTEWIDTH) - 1
4180 /* If followed by a repetition operator. */
4181 || *p == '*' || *p == '^'
4182 || ((syntax & RE_BK_PLUS_QM)
4183 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
4184 : (*p == '+' || *p == '?'))
4185 || ((syntax & RE_INTERVALS)
4186 && ((syntax & RE_NO_BK_BRACES)
4188 : (p[0] == '\\' && p[1] == '{'))))
4190 /* Start building a new exactn. */
4195 /* Is this exactn binary data or character? */
4196 is_exactn_bin = is_binary[p - 1 - pattern];
4198 BUF_PUSH_2 (exactn_bin, 0);
4200 BUF_PUSH_2 (exactn, 0);
4202 BUF_PUSH_2 (exactn, 0);
4204 pending_exact = b - 1;
4211 } /* while p != pend */
4214 /* Through the pattern now. */
4217 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
4219 if (!COMPILE_STACK_EMPTY)
4220 FREE_STACK_RETURN (REG_EPAREN);
4222 /* If we don't want backtracking, force success
4223 the first time we reach the end of the compiled pattern. */
4224 if (syntax & RE_NO_POSIX_BACKTRACKING)
4232 free (compile_stack.stack);
4234 /* We have succeeded; set the length of the buffer. */
4236 bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR;
4238 bufp->used = b - bufp->buffer;
4244 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4245 PREFIX(print_compiled_pattern) (bufp);
4249 #ifndef MATCH_MAY_ALLOCATE
4250 /* Initialize the failure stack to the largest possible stack. This
4251 isn't necessary unless we're trying to avoid calling alloca in
4252 the search and match routines. */
4254 int num_regs = bufp->re_nsub + 1;
4256 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4257 is strictly greater than re_max_failures, the largest possible stack
4258 is 2 * re_max_failures failure points. */
4259 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
4261 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
4264 if (! fail_stack.stack)
4266 = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size
4267 * sizeof (PREFIX(fail_stack_elt_t)));
4270 = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack,
4272 * sizeof (PREFIX(fail_stack_elt_t))));
4273 # else /* not emacs */
4274 if (! fail_stack.stack)
4276 = (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size
4277 * sizeof (PREFIX(fail_stack_elt_t)));
4280 = (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack,
4282 * sizeof (PREFIX(fail_stack_elt_t))));
4283 # endif /* not emacs */
4286 PREFIX(regex_grow_registers) (num_regs);
4288 #endif /* not MATCH_MAY_ALLOCATE */
4291 } /* regex_compile */
4293 /* Subroutines for `regex_compile'. */
4295 /* Store OP at LOC followed by two-byte integer parameter ARG. */
4296 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4299 PREFIX(store_op1) (op, loc, arg)
4304 *loc = (UCHAR_T) op;
4305 STORE_NUMBER (loc + 1, arg);
4309 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4310 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4313 PREFIX(store_op2) (op, loc, arg1, arg2)
4318 *loc = (UCHAR_T) op;
4319 STORE_NUMBER (loc + 1, arg1);
4320 STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2);
4324 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
4325 for OP followed by two-byte integer parameter ARG. */
4326 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4329 PREFIX(insert_op1) (op, loc, arg, end)
4335 register UCHAR_T *pfrom = end;
4336 register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE;
4338 while (pfrom != loc)
4341 PREFIX(store_op1) (op, loc, arg);
4345 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4346 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4349 PREFIX(insert_op2) (op, loc, arg1, arg2, end)
4355 register UCHAR_T *pfrom = end;
4356 register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE;
4358 while (pfrom != loc)
4361 PREFIX(store_op2) (op, loc, arg1, arg2);
4365 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
4366 after an alternative or a begin-subexpression. We assume there is at
4367 least one character before the ^. */
4370 PREFIX(at_begline_loc_p) (pattern, p, syntax)
4371 const CHAR_T *pattern, *p;
4372 reg_syntax_t syntax;
4374 const CHAR_T *prev = p - 2;
4375 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
4378 /* After a subexpression? */
4379 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
4380 /* After an alternative? */
4381 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
4385 /* The dual of at_begline_loc_p. This one is for $. We assume there is
4386 at least one character after the $, i.e., `P < PEND'. */
4389 PREFIX(at_endline_loc_p) (p, pend, syntax)
4390 const CHAR_T *p, *pend;
4391 reg_syntax_t syntax;
4393 const CHAR_T *next = p;
4394 boolean next_backslash = *next == '\\';
4395 const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0;
4398 /* Before a subexpression? */
4399 (syntax & RE_NO_BK_PARENS ? *next == ')'
4400 : next_backslash && next_next && *next_next == ')')
4401 /* Before an alternative? */
4402 || (syntax & RE_NO_BK_VBAR ? *next == '|'
4403 : next_backslash && next_next && *next_next == '|');
4406 #else /* not INSIDE_RECURSION */
4408 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4409 false if it's not. */
4412 group_in_compile_stack (compile_stack, regnum)
4413 compile_stack_type compile_stack;
4418 for (this_element = compile_stack.avail - 1;
4421 if (compile_stack.stack[this_element].regnum == regnum)
4426 #endif /* not INSIDE_RECURSION */
4428 #ifdef INSIDE_RECURSION
4431 /* This insert space, which size is "num", into the pattern at "loc".
4432 "end" must point the end of the allocated buffer. */
4434 insert_space (num, loc, end)
4439 register CHAR_T *pto = end;
4440 register CHAR_T *pfrom = end - num;
4442 while (pfrom >= loc)
4448 static reg_errcode_t
4449 wcs_compile_range (range_start_char, p_ptr, pend, translate, syntax, b,
4451 CHAR_T range_start_char;
4452 const CHAR_T **p_ptr, *pend;
4453 CHAR_T *char_set, *b;
4454 RE_TRANSLATE_TYPE translate;
4455 reg_syntax_t syntax;
4457 const CHAR_T *p = *p_ptr;
4458 CHAR_T range_start, range_end;
4462 uint32_t start_val, end_val;
4468 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4471 const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE,
4472 _NL_COLLATE_COLLSEQWC);
4473 const unsigned char *extra = (const unsigned char *)
4474 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4476 if (range_start_char < -1)
4478 /* range_start is a collating symbol. */
4480 /* Retreive the index and get collation sequence value. */
4481 wextra = (int32_t*)(extra + char_set[-range_start_char]);
4482 start_val = wextra[1 + *wextra];
4485 start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char));
4487 end_val = collseq_table_lookup (collseq, TRANSLATE (p[0]));
4489 /* Report an error if the range is empty and the syntax prohibits
4491 ret = ((syntax & RE_NO_EMPTY_RANGES)
4492 && (start_val > end_val))? REG_ERANGE : REG_NOERROR;
4494 /* Insert space to the end of the char_ranges. */
4495 insert_space(2, b - char_set[5] - 2, b - 1);
4496 *(b - char_set[5] - 2) = (wchar_t)start_val;
4497 *(b - char_set[5] - 1) = (wchar_t)end_val;
4498 char_set[4]++; /* ranges_index */
4503 range_start = (range_start_char >= 0)? TRANSLATE (range_start_char):
4505 range_end = TRANSLATE (p[0]);
4506 /* Report an error if the range is empty and the syntax prohibits
4508 ret = ((syntax & RE_NO_EMPTY_RANGES)
4509 && (range_start > range_end))? REG_ERANGE : REG_NOERROR;
4511 /* Insert space to the end of the char_ranges. */
4512 insert_space(2, b - char_set[5] - 2, b - 1);
4513 *(b - char_set[5] - 2) = range_start;
4514 *(b - char_set[5] - 1) = range_end;
4515 char_set[4]++; /* ranges_index */
4517 /* Have to increment the pointer into the pattern string, so the
4518 caller isn't still at the ending character. */
4524 /* Read the ending character of a range (in a bracket expression) from the
4525 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4526 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4527 Then we set the translation of all bits between the starting and
4528 ending characters (inclusive) in the compiled pattern B.
4530 Return an error code.
4532 We use these short variable names so we can use the same macros as
4533 `regex_compile' itself. */
4535 static reg_errcode_t
4536 byte_compile_range (range_start_char, p_ptr, pend, translate, syntax, b)
4537 unsigned int range_start_char;
4538 const char **p_ptr, *pend;
4539 RE_TRANSLATE_TYPE translate;
4540 reg_syntax_t syntax;
4544 const char *p = *p_ptr;
4547 const unsigned char *collseq;
4548 unsigned int start_colseq;
4549 unsigned int end_colseq;
4557 /* Have to increment the pointer into the pattern string, so the
4558 caller isn't still at the ending character. */
4561 /* Report an error if the range is empty and the syntax prohibits this. */
4562 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
4565 collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
4566 _NL_COLLATE_COLLSEQMB);
4568 start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)];
4569 end_colseq = collseq[(unsigned char) TRANSLATE (p[0])];
4570 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
4572 unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)];
4574 if (start_colseq <= this_colseq && this_colseq <= end_colseq)
4576 SET_LIST_BIT (TRANSLATE (this_char));
4581 /* Here we see why `this_char' has to be larger than an `unsigned
4582 char' -- we would otherwise go into an infinite loop, since all
4583 characters <= 0xff. */
4584 range_start_char = TRANSLATE (range_start_char);
4585 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4586 and some compilers cast it to int implicitly, so following for_loop
4587 may fall to (almost) infinite loop.
4588 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4589 To avoid this, we cast p[0] to unsigned int and truncate it. */
4590 end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1));
4592 for (this_char = range_start_char; this_char <= end_char; ++this_char)
4594 SET_LIST_BIT (TRANSLATE (this_char));
4603 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4604 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4605 characters can start a string that matches the pattern. This fastmap
4606 is used by re_search to skip quickly over impossible starting points.
4608 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4609 area as BUFP->fastmap.
4611 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4614 Returns 0 if we succeed, -2 if an internal error. */
4617 /* local function for re_compile_fastmap.
4618 truncate wchar_t character to char. */
4619 static unsigned char truncate_wchar (CHAR_T c);
4621 static unsigned char
4625 unsigned char buf[MB_LEN_MAX];
4626 int retval = wctomb(buf, c);
4627 return retval > 0 ? buf[0] : (unsigned char)c;
4632 PREFIX(re_compile_fastmap) (bufp)
4633 struct re_pattern_buffer *bufp;
4636 #ifdef MATCH_MAY_ALLOCATE
4637 PREFIX(fail_stack_type) fail_stack;
4639 #ifndef REGEX_MALLOC
4643 register char *fastmap = bufp->fastmap;
4646 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4647 pattern to (char*) in regex_compile. */
4648 UCHAR_T *pattern = (UCHAR_T*)bufp->buffer;
4649 register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used);
4651 UCHAR_T *pattern = bufp->buffer;
4652 register UCHAR_T *pend = pattern + bufp->used;
4654 UCHAR_T *p = pattern;
4657 /* This holds the pointer to the failure stack, when
4658 it is allocated relocatably. */
4659 fail_stack_elt_t *failure_stack_ptr;
4662 /* Assume that each path through the pattern can be null until
4663 proven otherwise. We set this false at the bottom of switch
4664 statement, to which we get only if a particular path doesn't
4665 match the empty string. */
4666 boolean path_can_be_null = true;
4668 /* We aren't doing a `succeed_n' to begin with. */
4669 boolean succeed_n_p = false;
4671 assert (fastmap != NULL && p != NULL);
4674 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4675 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4676 bufp->can_be_null = 0;
4680 if (p == pend || *p == succeed)
4682 /* We have reached the (effective) end of pattern. */
4683 if (!FAIL_STACK_EMPTY ())
4685 bufp->can_be_null |= path_can_be_null;
4687 /* Reset for next path. */
4688 path_can_be_null = true;
4690 p = fail_stack.stack[--fail_stack.avail].pointer;
4698 /* We should never be about to go beyond the end of the pattern. */
4701 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4704 /* I guess the idea here is to simply not bother with a fastmap
4705 if a backreference is used, since it's too hard to figure out
4706 the fastmap for the corresponding group. Setting
4707 `can_be_null' stops `re_search_2' from using the fastmap, so
4708 that is all we do. */
4710 bufp->can_be_null = 1;
4714 /* Following are the cases which match a character. These end
4719 fastmap[truncate_wchar(p[1])] = 1;
4733 /* It is hard to distinguish fastmap from (multi byte) characters
4734 which depends on current locale. */
4739 bufp->can_be_null = 1;
4743 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4744 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
4750 /* Chars beyond end of map must be allowed. */
4751 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
4754 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4755 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
4761 for (j = 0; j < (1 << BYTEWIDTH); j++)
4762 if (SYNTAX (j) == Sword)
4768 for (j = 0; j < (1 << BYTEWIDTH); j++)
4769 if (SYNTAX (j) != Sword)
4776 int fastmap_newline = fastmap['\n'];
4778 /* `.' matches anything ... */
4779 for (j = 0; j < (1 << BYTEWIDTH); j++)
4782 /* ... except perhaps newline. */
4783 if (!(bufp->syntax & RE_DOT_NEWLINE))
4784 fastmap['\n'] = fastmap_newline;
4786 /* Return if we have already set `can_be_null'; if we have,
4787 then the fastmap is irrelevant. Something's wrong here. */
4788 else if (bufp->can_be_null)
4791 /* Otherwise, have to check alternative paths. */
4798 for (j = 0; j < (1 << BYTEWIDTH); j++)
4799 if (SYNTAX (j) == (enum syntaxcode) k)
4806 for (j = 0; j < (1 << BYTEWIDTH); j++)
4807 if (SYNTAX (j) != (enum syntaxcode) k)
4812 /* All cases after this match the empty string. These end with
4832 case push_dummy_failure:
4837 case pop_failure_jump:
4838 case maybe_pop_jump:
4841 case dummy_failure_jump:
4842 EXTRACT_NUMBER_AND_INCR (j, p);
4847 /* Jump backward implies we just went through the body of a
4848 loop and matched nothing. Opcode jumped to should be
4849 `on_failure_jump' or `succeed_n'. Just treat it like an
4850 ordinary jump. For a * loop, it has pushed its failure
4851 point already; if so, discard that as redundant. */
4852 if ((re_opcode_t) *p != on_failure_jump
4853 && (re_opcode_t) *p != succeed_n)
4857 EXTRACT_NUMBER_AND_INCR (j, p);
4860 /* If what's on the stack is where we are now, pop it. */
4861 if (!FAIL_STACK_EMPTY ()
4862 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
4868 case on_failure_jump:
4869 case on_failure_keep_string_jump:
4870 handle_on_failure_jump:
4871 EXTRACT_NUMBER_AND_INCR (j, p);
4873 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4874 end of the pattern. We don't want to push such a point,
4875 since when we restore it above, entering the switch will
4876 increment `p' past the end of the pattern. We don't need
4877 to push such a point since we obviously won't find any more
4878 fastmap entries beyond `pend'. Such a pattern can match
4879 the null string, though. */
4882 if (!PUSH_PATTERN_OP (p + j, fail_stack))
4884 RESET_FAIL_STACK ();
4889 bufp->can_be_null = 1;
4893 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
4894 succeed_n_p = false;
4901 /* Get to the number of times to succeed. */
4902 p += OFFSET_ADDRESS_SIZE;
4904 /* Increment p past the n for when k != 0. */
4905 EXTRACT_NUMBER_AND_INCR (k, p);
4908 p -= 2 * OFFSET_ADDRESS_SIZE;
4909 succeed_n_p = true; /* Spaghetti code alert. */
4910 goto handle_on_failure_jump;
4916 p += 2 * OFFSET_ADDRESS_SIZE;
4927 abort (); /* We have listed all the cases. */
4930 /* Getting here means we have found the possible starting
4931 characters for one path of the pattern -- and that the empty
4932 string does not match. We need not follow this path further.
4933 Instead, look at the next alternative (remembered on the
4934 stack), or quit if no more. The test at the top of the loop
4935 does these things. */
4936 path_can_be_null = false;
4940 /* Set `can_be_null' for the last path (also the first path, if the
4941 pattern is empty). */
4942 bufp->can_be_null |= path_can_be_null;
4945 RESET_FAIL_STACK ();
4949 #else /* not INSIDE_RECURSION */
4952 re_compile_fastmap (bufp)
4953 struct re_pattern_buffer *bufp;
4956 if (MB_CUR_MAX != 1)
4957 return wcs_re_compile_fastmap(bufp);
4960 return byte_re_compile_fastmap(bufp);
4961 } /* re_compile_fastmap */
4963 weak_alias (__re_compile_fastmap, re_compile_fastmap)
4967 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4968 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4969 this memory for recording register information. STARTS and ENDS
4970 must be allocated using the malloc library routine, and must each
4971 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4973 If NUM_REGS == 0, then subsequent matches should allocate their own
4976 Unless this function is called, the first search or match using
4977 PATTERN_BUFFER will allocate its own register data, without
4978 freeing the old data. */
4981 re_set_registers (bufp, regs, num_regs, starts, ends)
4982 struct re_pattern_buffer *bufp;
4983 struct re_registers *regs;
4985 regoff_t *starts, *ends;
4989 bufp->regs_allocated = REGS_REALLOCATE;
4990 regs->num_regs = num_regs;
4991 regs->start = starts;
4996 bufp->regs_allocated = REGS_UNALLOCATED;
4998 regs->start = regs->end = (regoff_t *) 0;
5002 weak_alias (__re_set_registers, re_set_registers)
5005 /* Searching routines. */
5007 /* Like re_search_2, below, but only one string is specified, and
5008 doesn't let you say where to stop matching. */
5011 re_search (bufp, string, size, startpos, range, regs)
5012 struct re_pattern_buffer *bufp;
5014 int size, startpos, range;
5015 struct re_registers *regs;
5017 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
5021 weak_alias (__re_search, re_search)
5025 /* Using the compiled pattern in BUFP->buffer, first tries to match the
5026 virtual concatenation of STRING1 and STRING2, starting first at index
5027 STARTPOS, then at STARTPOS + 1, and so on.
5029 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
5031 RANGE is how far to scan while trying to match. RANGE = 0 means try
5032 only at STARTPOS; in general, the last start tried is STARTPOS +
5035 In REGS, return the indices of the virtual concatenation of STRING1
5036 and STRING2 that matched the entire BUFP->buffer and its contained
5039 Do not consider matching one past the index STOP in the virtual
5040 concatenation of STRING1 and STRING2.
5042 We return either the position in the strings at which the match was
5043 found, -1 if no match, or -2 if error (such as failure
5047 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
5048 struct re_pattern_buffer *bufp;
5049 const char *string1, *string2;
5053 struct re_registers *regs;
5057 if (MB_CUR_MAX != 1)
5058 return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5062 return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5066 weak_alias (__re_search_2, re_search_2)
5069 #endif /* not INSIDE_RECURSION */
5071 #ifdef INSIDE_RECURSION
5073 #ifdef MATCH_MAY_ALLOCATE
5074 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5076 # define FREE_VAR(var) if (var) free (var); var = NULL
5080 # define MAX_ALLOCA_SIZE 2000
5082 # define FREE_WCS_BUFFERS() \
5084 if (size1 > MAX_ALLOCA_SIZE) \
5086 free (wcs_string1); \
5087 free (mbs_offset1); \
5091 FREE_VAR (wcs_string1); \
5092 FREE_VAR (mbs_offset1); \
5094 if (size2 > MAX_ALLOCA_SIZE) \
5096 free (wcs_string2); \
5097 free (mbs_offset2); \
5101 FREE_VAR (wcs_string2); \
5102 FREE_VAR (mbs_offset2); \
5110 PREFIX(re_search_2) (bufp, string1, size1, string2, size2, startpos, range,
5112 struct re_pattern_buffer *bufp;
5113 const char *string1, *string2;
5117 struct re_registers *regs;
5121 register char *fastmap = bufp->fastmap;
5122 register RE_TRANSLATE_TYPE translate = bufp->translate;
5123 int total_size = size1 + size2;
5124 int endpos = startpos + range;
5126 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5127 wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL;
5128 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5129 int wcs_size1 = 0, wcs_size2 = 0;
5130 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5131 int *mbs_offset1 = NULL, *mbs_offset2 = NULL;
5132 /* They hold whether each wchar_t is binary data or not. */
5133 char *is_binary = NULL;
5136 /* Check for out-of-range STARTPOS. */
5137 if (startpos < 0 || startpos > total_size)
5140 /* Fix up RANGE if it might eventually take us outside
5141 the virtual concatenation of STRING1 and STRING2.
5142 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5144 range = 0 - startpos;
5145 else if (endpos > total_size)
5146 range = total_size - startpos;
5148 /* If the search isn't to be a backwards one, don't waste time in a
5149 search for a pattern that must be anchored. */
5150 if (bufp->used > 0 && range > 0
5151 && ((re_opcode_t) bufp->buffer[0] == begbuf
5152 /* `begline' is like `begbuf' if it cannot match at newlines. */
5153 || ((re_opcode_t) bufp->buffer[0] == begline
5154 && !bufp->newline_anchor)))
5163 /* In a forward search for something that starts with \=.
5164 don't keep searching past point. */
5165 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
5167 range = PT - startpos;
5173 /* Update the fastmap now if not correct already. */
5174 if (fastmap && !bufp->fastmap_accurate)
5175 if (re_compile_fastmap (bufp) == -2)
5179 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5180 fill them with converted string. */
5183 if (size1 > MAX_ALLOCA_SIZE)
5185 wcs_string1 = TALLOC (size1 + 1, CHAR_T);
5186 mbs_offset1 = TALLOC (size1 + 1, int);
5187 is_binary = TALLOC (size1 + 1, char);
5191 wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T);
5192 mbs_offset1 = REGEX_TALLOC (size1 + 1, int);
5193 is_binary = REGEX_TALLOC (size1 + 1, char);
5195 if (!wcs_string1 || !mbs_offset1 || !is_binary)
5197 if (size1 > MAX_ALLOCA_SIZE)
5205 FREE_VAR (wcs_string1);
5206 FREE_VAR (mbs_offset1);
5207 FREE_VAR (is_binary);
5211 wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1,
5212 mbs_offset1, is_binary);
5213 wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */
5214 if (size1 > MAX_ALLOCA_SIZE)
5217 FREE_VAR (is_binary);
5221 if (size2 > MAX_ALLOCA_SIZE)
5223 wcs_string2 = TALLOC (size2 + 1, CHAR_T);
5224 mbs_offset2 = TALLOC (size2 + 1, int);
5225 is_binary = TALLOC (size2 + 1, char);
5229 wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T);
5230 mbs_offset2 = REGEX_TALLOC (size2 + 1, int);
5231 is_binary = REGEX_TALLOC (size2 + 1, char);
5233 if (!wcs_string2 || !mbs_offset2 || !is_binary)
5235 FREE_WCS_BUFFERS ();
5236 if (size2 > MAX_ALLOCA_SIZE)
5239 FREE_VAR (is_binary);
5242 wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2,
5243 mbs_offset2, is_binary);
5244 wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */
5245 if (size2 > MAX_ALLOCA_SIZE)
5248 FREE_VAR (is_binary);
5253 /* Loop through the string, looking for a place to start matching. */
5256 /* If a fastmap is supplied, skip quickly over characters that
5257 cannot be the start of a match. If the pattern can match the
5258 null string, however, we don't need to skip characters; we want
5259 the first null string. */
5260 if (fastmap && startpos < total_size && !bufp->can_be_null)
5262 if (range > 0) /* Searching forwards. */
5264 register const char *d;
5265 register int lim = 0;
5268 if (startpos < size1 && startpos + range >= size1)
5269 lim = range - (size1 - startpos);
5271 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
5273 /* Written out as an if-else to avoid testing `translate'
5277 && !fastmap[(unsigned char)
5278 translate[(unsigned char) *d++]])
5281 while (range > lim && !fastmap[(unsigned char) *d++])
5284 startpos += irange - range;
5286 else /* Searching backwards. */
5288 register CHAR_T c = (size1 == 0 || startpos >= size1
5289 ? string2[startpos - size1]
5290 : string1[startpos]);
5292 if (!fastmap[(unsigned char) TRANSLATE (c)])
5297 /* If can't match the null string, and that's all we have left, fail. */
5298 if (range >= 0 && startpos == total_size && fastmap
5299 && !bufp->can_be_null)
5302 FREE_WCS_BUFFERS ();
5308 val = wcs_re_match_2_internal (bufp, string1, size1, string2,
5309 size2, startpos, regs, stop,
5310 wcs_string1, wcs_size1,
5311 wcs_string2, wcs_size2,
5312 mbs_offset1, mbs_offset2);
5314 val = byte_re_match_2_internal (bufp, string1, size1, string2,
5315 size2, startpos, regs, stop);
5318 #ifndef REGEX_MALLOC
5327 FREE_WCS_BUFFERS ();
5335 FREE_WCS_BUFFERS ();
5355 FREE_WCS_BUFFERS ();
5361 /* This converts PTR, a pointer into one of the search wchar_t strings
5362 `string1' and `string2' into an multibyte string offset from the
5363 beginning of that string. We use mbs_offset to optimize.
5364 See convert_mbs_to_wcs. */
5365 # define POINTER_TO_OFFSET(ptr) \
5366 (FIRST_STRING_P (ptr) \
5367 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5368 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5371 /* This converts PTR, a pointer into one of the search strings `string1'
5372 and `string2' into an offset from the beginning of that string. */
5373 # define POINTER_TO_OFFSET(ptr) \
5374 (FIRST_STRING_P (ptr) \
5375 ? ((regoff_t) ((ptr) - string1)) \
5376 : ((regoff_t) ((ptr) - string2 + size1)))
5379 /* Macros for dealing with the split strings in re_match_2. */
5381 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5383 /* Call before fetching a character with *d. This switches over to
5384 string2 if necessary. */
5385 #define PREFETCH() \
5388 /* End of string2 => fail. */ \
5389 if (dend == end_match_2) \
5391 /* End of string1 => advance to string2. */ \
5393 dend = end_match_2; \
5396 /* Test if at very beginning or at very end of the virtual concatenation
5397 of `string1' and `string2'. If only one string, it's `string2'. */
5398 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5399 #define AT_STRINGS_END(d) ((d) == end2)
5402 /* Test if D points to a character which is word-constituent. We have
5403 two special cases to check for: if past the end of string1, look at
5404 the first character in string2; and if before the beginning of
5405 string2, look at the last character in string1. */
5407 /* Use internationalized API instead of SYNTAX. */
5408 # define WORDCHAR_P(d) \
5409 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5410 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5411 || ((d) == end1 ? *string2 \
5412 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5414 # define WORDCHAR_P(d) \
5415 (SYNTAX ((d) == end1 ? *string2 \
5416 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5420 /* Disabled due to a compiler bug -- see comment at case wordbound */
5422 /* Test if the character before D and the one at D differ with respect
5423 to being word-constituent. */
5424 #define AT_WORD_BOUNDARY(d) \
5425 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5426 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5429 /* Free everything we malloc. */
5430 #ifdef MATCH_MAY_ALLOCATE
5432 # define FREE_VARIABLES() \
5434 REGEX_FREE_STACK (fail_stack.stack); \
5435 FREE_VAR (regstart); \
5436 FREE_VAR (regend); \
5437 FREE_VAR (old_regstart); \
5438 FREE_VAR (old_regend); \
5439 FREE_VAR (best_regstart); \
5440 FREE_VAR (best_regend); \
5441 FREE_VAR (reg_info); \
5442 FREE_VAR (reg_dummy); \
5443 FREE_VAR (reg_info_dummy); \
5444 if (!cant_free_wcs_buf) \
5446 FREE_VAR (string1); \
5447 FREE_VAR (string2); \
5448 FREE_VAR (mbs_offset1); \
5449 FREE_VAR (mbs_offset2); \
5453 # define FREE_VARIABLES() \
5455 REGEX_FREE_STACK (fail_stack.stack); \
5456 FREE_VAR (regstart); \
5457 FREE_VAR (regend); \
5458 FREE_VAR (old_regstart); \
5459 FREE_VAR (old_regend); \
5460 FREE_VAR (best_regstart); \
5461 FREE_VAR (best_regend); \
5462 FREE_VAR (reg_info); \
5463 FREE_VAR (reg_dummy); \
5464 FREE_VAR (reg_info_dummy); \
5469 # define FREE_VARIABLES() \
5471 if (!cant_free_wcs_buf) \
5473 FREE_VAR (string1); \
5474 FREE_VAR (string2); \
5475 FREE_VAR (mbs_offset1); \
5476 FREE_VAR (mbs_offset2); \
5480 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5482 #endif /* not MATCH_MAY_ALLOCATE */
5484 /* These values must meet several constraints. They must not be valid
5485 register values; since we have a limit of 255 registers (because
5486 we use only one byte in the pattern for the register number), we can
5487 use numbers larger than 255. They must differ by 1, because of
5488 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5489 be larger than the value for the highest register, so we do not try
5490 to actually save any registers when none are active. */
5491 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5492 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5494 #else /* not INSIDE_RECURSION */
5495 /* Matching routines. */
5497 #ifndef emacs /* Emacs never uses this. */
5498 /* re_match is like re_match_2 except it takes only a single string. */
5501 re_match (bufp, string, size, pos, regs)
5502 struct re_pattern_buffer *bufp;
5505 struct re_registers *regs;
5509 if (MB_CUR_MAX != 1)
5510 result = wcs_re_match_2_internal (bufp, NULL, 0, string, size,
5512 NULL, 0, NULL, 0, NULL, NULL);
5515 result = byte_re_match_2_internal (bufp, NULL, 0, string, size,
5517 # ifndef REGEX_MALLOC
5525 weak_alias (__re_match, re_match)
5527 #endif /* not emacs */
5529 #endif /* not INSIDE_RECURSION */
5531 #ifdef INSIDE_RECURSION
5532 static boolean PREFIX(group_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5534 PREFIX(register_info_type) *reg_info));
5535 static boolean PREFIX(alt_match_null_string_p) _RE_ARGS ((UCHAR_T *p,
5537 PREFIX(register_info_type) *reg_info));
5538 static boolean PREFIX(common_op_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5540 PREFIX(register_info_type) *reg_info));
5541 static int PREFIX(bcmp_translate) _RE_ARGS ((const CHAR_T *s1, const CHAR_T *s2,
5542 int len, char *translate));
5543 #else /* not INSIDE_RECURSION */
5545 /* re_match_2 matches the compiled pattern in BUFP against the
5546 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5547 and SIZE2, respectively). We start matching at POS, and stop
5550 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5551 store offsets for the substring each group matched in REGS. See the
5552 documentation for exactly how many groups we fill.
5554 We return -1 if no match, -2 if an internal error (such as the
5555 failure stack overflowing). Otherwise, we return the length of the
5556 matched substring. */
5559 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5560 struct re_pattern_buffer *bufp;
5561 const char *string1, *string2;
5564 struct re_registers *regs;
5569 if (MB_CUR_MAX != 1)
5570 result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2,
5572 NULL, 0, NULL, 0, NULL, NULL);
5575 result = byte_re_match_2_internal (bufp, string1, size1, string2, size2,
5578 #ifndef REGEX_MALLOC
5586 weak_alias (__re_match_2, re_match_2)
5589 #endif /* not INSIDE_RECURSION */
5591 #ifdef INSIDE_RECURSION
5594 static int count_mbs_length PARAMS ((int *, int));
5596 /* This check the substring (from 0, to length) of the multibyte string,
5597 to which offset_buffer correspond. And count how many wchar_t_characters
5598 the substring occupy. We use offset_buffer to optimization.
5599 See convert_mbs_to_wcs. */
5602 count_mbs_length(offset_buffer, length)
5608 /* Check whether the size is valid. */
5612 if (offset_buffer == NULL)
5615 /* If there are no multibyte character, offset_buffer[i] == i.
5616 Optmize for this case. */
5617 if (offset_buffer[length] == length)
5620 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5626 int middle = (lower + upper) / 2;
5627 if (middle == lower || middle == upper)
5629 if (offset_buffer[middle] > length)
5631 else if (offset_buffer[middle] < length)
5641 /* This is a separate function so that we can force an alloca cleanup
5645 wcs_re_match_2_internal (bufp, cstring1, csize1, cstring2, csize2, pos,
5646 regs, stop, string1, size1, string2, size2,
5647 mbs_offset1, mbs_offset2)
5648 struct re_pattern_buffer *bufp;
5649 const char *cstring1, *cstring2;
5652 struct re_registers *regs;
5654 /* string1 == string2 == NULL means string1/2, size1/2 and
5655 mbs_offset1/2 need seting up in this function. */
5656 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5657 wchar_t *string1, *string2;
5658 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5660 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5661 int *mbs_offset1, *mbs_offset2;
5664 byte_re_match_2_internal (bufp, string1, size1,string2, size2, pos,
5666 struct re_pattern_buffer *bufp;
5667 const char *string1, *string2;
5670 struct re_registers *regs;
5674 /* General temporaries. */
5678 /* They hold whether each wchar_t is binary data or not. */
5679 char *is_binary = NULL;
5680 /* If true, we can't free string1/2, mbs_offset1/2. */
5681 int cant_free_wcs_buf = 1;
5684 /* Just past the end of the corresponding string. */
5685 const CHAR_T *end1, *end2;
5687 /* Pointers into string1 and string2, just past the last characters in
5688 each to consider matching. */
5689 const CHAR_T *end_match_1, *end_match_2;
5691 /* Where we are in the data, and the end of the current string. */
5692 const CHAR_T *d, *dend;
5694 /* Where we are in the pattern, and the end of the pattern. */
5696 UCHAR_T *pattern, *p;
5697 register UCHAR_T *pend;
5699 UCHAR_T *p = bufp->buffer;
5700 register UCHAR_T *pend = p + bufp->used;
5703 /* Mark the opcode just after a start_memory, so we can test for an
5704 empty subpattern when we get to the stop_memory. */
5705 UCHAR_T *just_past_start_mem = 0;
5707 /* We use this to map every character in the string. */
5708 RE_TRANSLATE_TYPE translate = bufp->translate;
5710 /* Failure point stack. Each place that can handle a failure further
5711 down the line pushes a failure point on this stack. It consists of
5712 restart, regend, and reg_info for all registers corresponding to
5713 the subexpressions we're currently inside, plus the number of such
5714 registers, and, finally, two char *'s. The first char * is where
5715 to resume scanning the pattern; the second one is where to resume
5716 scanning the strings. If the latter is zero, the failure point is
5717 a ``dummy''; if a failure happens and the failure point is a dummy,
5718 it gets discarded and the next next one is tried. */
5719 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5720 PREFIX(fail_stack_type) fail_stack;
5723 static unsigned failure_id;
5724 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5728 /* This holds the pointer to the failure stack, when
5729 it is allocated relocatably. */
5730 fail_stack_elt_t *failure_stack_ptr;
5733 /* We fill all the registers internally, independent of what we
5734 return, for use in backreferences. The number here includes
5735 an element for register zero. */
5736 size_t num_regs = bufp->re_nsub + 1;
5738 /* The currently active registers. */
5739 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
5740 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
5742 /* Information on the contents of registers. These are pointers into
5743 the input strings; they record just what was matched (on this
5744 attempt) by a subexpression part of the pattern, that is, the
5745 regnum-th regstart pointer points to where in the pattern we began
5746 matching and the regnum-th regend points to right after where we
5747 stopped matching the regnum-th subexpression. (The zeroth register
5748 keeps track of what the whole pattern matches.) */
5749 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5750 const CHAR_T **regstart, **regend;
5753 /* If a group that's operated upon by a repetition operator fails to
5754 match anything, then the register for its start will need to be
5755 restored because it will have been set to wherever in the string we
5756 are when we last see its open-group operator. Similarly for a
5758 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5759 const CHAR_T **old_regstart, **old_regend;
5762 /* The is_active field of reg_info helps us keep track of which (possibly
5763 nested) subexpressions we are currently in. The matched_something
5764 field of reg_info[reg_num] helps us tell whether or not we have
5765 matched any of the pattern so far this time through the reg_num-th
5766 subexpression. These two fields get reset each time through any
5767 loop their register is in. */
5768 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5769 PREFIX(register_info_type) *reg_info;
5772 /* The following record the register info as found in the above
5773 variables when we find a match better than any we've seen before.
5774 This happens as we backtrack through the failure points, which in
5775 turn happens only if we have not yet matched the entire string. */
5776 unsigned best_regs_set = false;
5777 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5778 const CHAR_T **best_regstart, **best_regend;
5781 /* Logically, this is `best_regend[0]'. But we don't want to have to
5782 allocate space for that if we're not allocating space for anything
5783 else (see below). Also, we never need info about register 0 for
5784 any of the other register vectors, and it seems rather a kludge to
5785 treat `best_regend' differently than the rest. So we keep track of
5786 the end of the best match so far in a separate variable. We
5787 initialize this to NULL so that when we backtrack the first time
5788 and need to test it, it's not garbage. */
5789 const CHAR_T *match_end = NULL;
5791 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5792 int set_regs_matched_done = 0;
5794 /* Used when we pop values we don't care about. */
5795 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5796 const CHAR_T **reg_dummy;
5797 PREFIX(register_info_type) *reg_info_dummy;
5801 /* Counts the total number of registers pushed. */
5802 unsigned num_regs_pushed = 0;
5805 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5809 #ifdef MATCH_MAY_ALLOCATE
5810 /* Do not bother to initialize all the register variables if there are
5811 no groups in the pattern, as it takes a fair amount of time. If
5812 there are groups, we include space for register 0 (the whole
5813 pattern), even though we never use it, since it simplifies the
5814 array indexing. We should fix this. */
5817 regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5818 regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5819 old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5820 old_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5821 best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5822 best_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5823 reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5824 reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *);
5825 reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5827 if (!(regstart && regend && old_regstart && old_regend && reg_info
5828 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
5836 /* We must initialize all our variables to NULL, so that
5837 `FREE_VARIABLES' doesn't try to free them. */
5838 regstart = regend = old_regstart = old_regend = best_regstart
5839 = best_regend = reg_dummy = NULL;
5840 reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL;
5842 #endif /* MATCH_MAY_ALLOCATE */
5844 /* The starting position is bogus. */
5846 if (pos < 0 || pos > csize1 + csize2)
5848 if (pos < 0 || pos > size1 + size2)
5856 /* Allocate wchar_t array for string1 and string2 and
5857 fill them with converted string. */
5858 if (string1 == NULL && string2 == NULL)
5860 /* We need seting up buffers here. */
5862 /* We must free wcs buffers in this function. */
5863 cant_free_wcs_buf = 0;
5867 string1 = REGEX_TALLOC (csize1 + 1, CHAR_T);
5868 mbs_offset1 = REGEX_TALLOC (csize1 + 1, int);
5869 is_binary = REGEX_TALLOC (csize1 + 1, char);
5870 if (!string1 || !mbs_offset1 || !is_binary)
5873 FREE_VAR (mbs_offset1);
5874 FREE_VAR (is_binary);
5880 string2 = REGEX_TALLOC (csize2 + 1, CHAR_T);
5881 mbs_offset2 = REGEX_TALLOC (csize2 + 1, int);
5882 is_binary = REGEX_TALLOC (csize2 + 1, char);
5883 if (!string2 || !mbs_offset2 || !is_binary)
5886 FREE_VAR (mbs_offset1);
5888 FREE_VAR (mbs_offset2);
5889 FREE_VAR (is_binary);
5892 size2 = convert_mbs_to_wcs(string2, cstring2, csize2,
5893 mbs_offset2, is_binary);
5894 string2[size2] = L'\0'; /* for a sentinel */
5895 FREE_VAR (is_binary);
5899 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5900 pattern to (char*) in regex_compile. */
5901 p = pattern = (CHAR_T*)bufp->buffer;
5902 pend = (CHAR_T*)(bufp->buffer + bufp->used);
5906 /* Initialize subexpression text positions to -1 to mark ones that no
5907 start_memory/stop_memory has been seen for. Also initialize the
5908 register information struct. */
5909 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5911 regstart[mcnt] = regend[mcnt]
5912 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
5914 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
5915 IS_ACTIVE (reg_info[mcnt]) = 0;
5916 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5917 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5920 /* We move `string1' into `string2' if the latter's empty -- but not if
5921 `string1' is null. */
5922 if (size2 == 0 && string1 != NULL)
5929 mbs_offset2 = mbs_offset1;
5935 end1 = string1 + size1;
5936 end2 = string2 + size2;
5938 /* Compute where to stop matching, within the two strings. */
5942 mcnt = count_mbs_length(mbs_offset1, stop);
5943 end_match_1 = string1 + mcnt;
5944 end_match_2 = string2;
5948 if (stop > csize1 + csize2)
5949 stop = csize1 + csize2;
5951 mcnt = count_mbs_length(mbs_offset2, stop-csize1);
5952 end_match_2 = string2 + mcnt;
5955 { /* count_mbs_length return error. */
5962 end_match_1 = string1 + stop;
5963 end_match_2 = string2;
5968 end_match_2 = string2 + stop - size1;
5972 /* `p' scans through the pattern as `d' scans through the data.
5973 `dend' is the end of the input string that `d' points within. `d'
5974 is advanced into the following input string whenever necessary, but
5975 this happens before fetching; therefore, at the beginning of the
5976 loop, `d' can be pointing at the end of a string, but it cannot
5979 if (size1 > 0 && pos <= csize1)
5981 mcnt = count_mbs_length(mbs_offset1, pos);
5987 mcnt = count_mbs_length(mbs_offset2, pos-csize1);
5993 { /* count_mbs_length return error. */
5998 if (size1 > 0 && pos <= size1)
6005 d = string2 + pos - size1;
6010 DEBUG_PRINT1 ("The compiled pattern is:\n");
6011 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
6012 DEBUG_PRINT1 ("The string to match is: `");
6013 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
6014 DEBUG_PRINT1 ("'\n");
6016 /* This loops over pattern commands. It exits by returning from the
6017 function if the match is complete, or it drops through if the match
6018 fails at this starting point in the input data. */
6022 DEBUG_PRINT2 ("\n%p: ", p);
6024 DEBUG_PRINT2 ("\n0x%x: ", p);
6028 { /* End of pattern means we might have succeeded. */
6029 DEBUG_PRINT1 ("end of pattern ... ");
6031 /* If we haven't matched the entire string, and we want the
6032 longest match, try backtracking. */
6033 if (d != end_match_2)
6035 /* 1 if this match ends in the same string (string1 or string2)
6036 as the best previous match. */
6037 boolean same_str_p = (FIRST_STRING_P (match_end)
6038 == MATCHING_IN_FIRST_STRING);
6039 /* 1 if this match is the best seen so far. */
6040 boolean best_match_p;
6042 /* AIX compiler got confused when this was combined
6043 with the previous declaration. */
6045 best_match_p = d > match_end;
6047 best_match_p = !MATCHING_IN_FIRST_STRING;
6049 DEBUG_PRINT1 ("backtracking.\n");
6051 if (!FAIL_STACK_EMPTY ())
6052 { /* More failure points to try. */
6054 /* If exceeds best match so far, save it. */
6055 if (!best_regs_set || best_match_p)
6057 best_regs_set = true;
6060 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6062 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6064 best_regstart[mcnt] = regstart[mcnt];
6065 best_regend[mcnt] = regend[mcnt];
6071 /* If no failure points, don't restore garbage. And if
6072 last match is real best match, don't restore second
6074 else if (best_regs_set && !best_match_p)
6077 /* Restore best match. It may happen that `dend ==
6078 end_match_1' while the restored d is in string2.
6079 For example, the pattern `x.*y.*z' against the
6080 strings `x-' and `y-z-', if the two strings are
6081 not consecutive in memory. */
6082 DEBUG_PRINT1 ("Restoring best registers.\n");
6085 dend = ((d >= string1 && d <= end1)
6086 ? end_match_1 : end_match_2);
6088 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6090 regstart[mcnt] = best_regstart[mcnt];
6091 regend[mcnt] = best_regend[mcnt];
6094 } /* d != end_match_2 */
6097 DEBUG_PRINT1 ("Accepting match.\n");
6098 /* If caller wants register contents data back, do it. */
6099 if (regs && !bufp->no_sub)
6101 /* Have the register data arrays been allocated? */
6102 if (bufp->regs_allocated == REGS_UNALLOCATED)
6103 { /* No. So allocate them with malloc. We need one
6104 extra element beyond `num_regs' for the `-1' marker
6106 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
6107 regs->start = TALLOC (regs->num_regs, regoff_t);
6108 regs->end = TALLOC (regs->num_regs, regoff_t);
6109 if (regs->start == NULL || regs->end == NULL)
6114 bufp->regs_allocated = REGS_REALLOCATE;
6116 else if (bufp->regs_allocated == REGS_REALLOCATE)
6117 { /* Yes. If we need more elements than were already
6118 allocated, reallocate them. If we need fewer, just
6120 if (regs->num_regs < num_regs + 1)
6122 regs->num_regs = num_regs + 1;
6123 RETALLOC (regs->start, regs->num_regs, regoff_t);
6124 RETALLOC (regs->end, regs->num_regs, regoff_t);
6125 if (regs->start == NULL || regs->end == NULL)
6134 /* These braces fend off a "empty body in an else-statement"
6135 warning under GCC when assert expands to nothing. */
6136 assert (bufp->regs_allocated == REGS_FIXED);
6139 /* Convert the pointer data in `regstart' and `regend' to
6140 indices. Register zero has to be set differently,
6141 since we haven't kept track of any info for it. */
6142 if (regs->num_regs > 0)
6144 regs->start[0] = pos;
6146 if (MATCHING_IN_FIRST_STRING)
6147 regs->end[0] = mbs_offset1 != NULL ?
6148 mbs_offset1[d-string1] : 0;
6150 regs->end[0] = csize1 + (mbs_offset2 != NULL ?
6151 mbs_offset2[d-string2] : 0);
6153 regs->end[0] = (MATCHING_IN_FIRST_STRING
6154 ? ((regoff_t) (d - string1))
6155 : ((regoff_t) (d - string2 + size1)));
6159 /* Go through the first `min (num_regs, regs->num_regs)'
6160 registers, since that is all we initialized. */
6161 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
6164 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
6165 regs->start[mcnt] = regs->end[mcnt] = -1;
6169 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
6171 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
6175 /* If the regs structure we return has more elements than
6176 were in the pattern, set the extra elements to -1. If
6177 we (re)allocated the registers, this is the case,
6178 because we always allocate enough to have at least one
6180 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
6181 regs->start[mcnt] = regs->end[mcnt] = -1;
6182 } /* regs && !bufp->no_sub */
6184 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6185 nfailure_points_pushed, nfailure_points_popped,
6186 nfailure_points_pushed - nfailure_points_popped);
6187 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
6190 if (MATCHING_IN_FIRST_STRING)
6191 mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0;
6193 mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) +
6197 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
6202 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
6208 /* Otherwise match next pattern command. */
6209 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
6211 /* Ignore these. Used to ignore the n of succeed_n's which
6212 currently have n == 0. */
6214 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6218 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6221 /* Match the next n pattern characters exactly. The following
6222 byte in the pattern defines n, and the n bytes after that
6223 are the characters to match. */
6229 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
6231 /* This is written out as an if-else so we don't waste time
6232 testing `translate' inside the loop. */
6241 if ((UCHAR_T) translate[(unsigned char) *d++]
6247 if (*d++ != (CHAR_T) *p++)
6251 if ((UCHAR_T) translate[(unsigned char) *d++]
6263 if (*d++ != (CHAR_T) *p++) goto fail;
6267 SET_REGS_MATCHED ();
6271 /* Match any character except possibly a newline or a null. */
6273 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6277 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
6278 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
6281 SET_REGS_MATCHED ();
6282 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d);
6292 unsigned int i, char_class_length, coll_symbol_length,
6293 equiv_class_length, ranges_length, chars_length, length;
6294 CHAR_T *workp, *workp2, *charset_top;
6295 #define WORK_BUFFER_SIZE 128
6296 CHAR_T str_buf[WORK_BUFFER_SIZE];
6301 boolean not = (re_opcode_t) *(p - 1) == charset_not;
6303 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6305 c = TRANSLATE (*d); /* The character to match. */
6308 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
6310 charset_top = p - 1;
6311 char_class_length = *p++;
6312 coll_symbol_length = *p++;
6313 equiv_class_length = *p++;
6314 ranges_length = *p++;
6315 chars_length = *p++;
6316 /* p points charset[6], so the address of the next instruction
6317 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6318 where l=length of char_classes, m=length of collating_symbol,
6319 n=equivalence_class, o=length of char_range,
6320 p'=length of character. */
6322 /* Update p to indicate the next instruction. */
6323 p += char_class_length + coll_symbol_length+ equiv_class_length +
6324 2*ranges_length + chars_length;
6326 /* match with char_class? */
6327 for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE)
6330 uintptr_t alignedp = ((uintptr_t)workp
6331 + __alignof__(wctype_t) - 1)
6332 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6333 wctype = *((wctype_t*)alignedp);
6334 workp += CHAR_CLASS_SIZE;
6335 if (iswctype((wint_t)c, wctype))
6336 goto char_set_matched;
6339 /* match with collating_symbol? */
6343 const unsigned char *extra = (const unsigned char *)
6344 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
6346 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;
6350 wextra = (int32_t*)(extra + *workp++);
6351 for (i = 0; i < *wextra; ++i)
6352 if (TRANSLATE(d[i]) != wextra[1 + i])
6357 /* Update d, however d will be incremented at
6358 char_set_matched:, we decrement d here. */
6360 goto char_set_matched;
6364 else /* (nrules == 0) */
6366 /* If we can't look up collation data, we use wcscoll
6369 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;)
6371 const CHAR_T *backup_d = d, *backup_dend = dend;
6372 length = wcslen(workp);
6374 /* If wcscoll(the collating symbol, whole string) > 0,
6375 any substring of the string never match with the
6376 collating symbol. */
6377 if (wcscoll(workp, d) > 0)
6379 workp += length + 1;
6383 /* First, we compare the collating symbol with
6384 the first character of the string.
6385 If it don't match, we add the next character to
6386 the compare buffer in turn. */
6387 for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++)
6392 if (dend == end_match_2)
6398 /* add next character to the compare buffer. */
6399 str_buf[i] = TRANSLATE(*d);
6400 str_buf[i+1] = '\0';
6402 match = wcscoll(workp, str_buf);
6404 goto char_set_matched;
6407 /* (str_buf > workp) indicate (str_buf + X > workp),
6408 because for all X (str_buf + X > str_buf).
6409 So we don't need continue this loop. */
6412 /* Otherwise(str_buf < workp),
6413 (str_buf+next_character) may equals (workp).
6414 So we continue this loop. */
6419 workp += length + 1;
6422 /* match with equivalence_class? */
6426 const CHAR_T *backup_d = d, *backup_dend = dend;
6427 /* Try to match the equivalence class against
6428 those known to the collate implementation. */
6429 const int32_t *table;
6430 const int32_t *weights;
6431 const int32_t *extra;
6432 const int32_t *indirect;
6437 /* This #include defines a local function! */
6438 # include <locale/weightwc.h>
6440 table = (const int32_t *)
6441 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
6442 weights = (const wint_t *)
6443 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
6444 extra = (const wint_t *)
6445 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
6446 indirect = (const int32_t *)
6447 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
6449 /* Write 1 collating element to str_buf, and
6453 for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++)
6455 cp = (wint_t*)str_buf;
6458 if (dend == end_match_2)
6463 str_buf[i] = TRANSLATE(*(d+i));
6464 str_buf[i+1] = '\0'; /* sentinel */
6465 idx2 = findidx ((const wint_t**)&cp);
6468 /* Update d, however d will be incremented at
6469 char_set_matched:, we decrement d here. */
6470 d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1);
6473 if (dend == end_match_2)
6482 len = weights[idx2];
6484 for (workp2 = workp + equiv_class_length ; workp < workp2 ;
6487 idx = (int32_t)*workp;
6488 /* We already checked idx != 0 in regex_compile. */
6490 if (idx2 != 0 && len == weights[idx])
6493 while (cnt < len && (weights[idx + 1 + cnt]
6494 == weights[idx2 + 1 + cnt]))
6498 goto char_set_matched;
6505 else /* (nrules == 0) */
6507 /* If we can't look up collation data, we use wcscoll
6510 for (workp2 = workp + equiv_class_length ; workp < workp2 ;)
6512 const CHAR_T *backup_d = d, *backup_dend = dend;
6513 length = wcslen(workp);
6515 /* If wcscoll(the collating symbol, whole string) > 0,
6516 any substring of the string never match with the
6517 collating symbol. */
6518 if (wcscoll(workp, d) > 0)
6520 workp += length + 1;
6524 /* First, we compare the equivalence class with
6525 the first character of the string.
6526 If it don't match, we add the next character to
6527 the compare buffer in turn. */
6528 for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++)
6533 if (dend == end_match_2)
6539 /* add next character to the compare buffer. */
6540 str_buf[i] = TRANSLATE(*d);
6541 str_buf[i+1] = '\0';
6543 match = wcscoll(workp, str_buf);
6546 goto char_set_matched;
6549 /* (str_buf > workp) indicate (str_buf + X > workp),
6550 because for all X (str_buf + X > str_buf).
6551 So we don't need continue this loop. */
6554 /* Otherwise(str_buf < workp),
6555 (str_buf+next_character) may equals (workp).
6556 So we continue this loop. */
6561 workp += length + 1;
6565 /* match with char_range? */
6569 uint32_t collseqval;
6570 const char *collseq = (const char *)
6571 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
6573 collseqval = collseq_table_lookup (collseq, c);
6575 for (; workp < p - chars_length ;)
6577 uint32_t start_val, end_val;
6579 /* We already compute the collation sequence value
6580 of the characters (or collating symbols). */
6581 start_val = (uint32_t) *workp++; /* range_start */
6582 end_val = (uint32_t) *workp++; /* range_end */
6584 if (start_val <= collseqval && collseqval <= end_val)
6585 goto char_set_matched;
6591 /* We set range_start_char at str_buf[0], range_end_char
6592 at str_buf[4], and compared char at str_buf[2]. */
6597 for (; workp < p - chars_length ;)
6599 wchar_t *range_start_char, *range_end_char;
6601 /* match if (range_start_char <= c <= range_end_char). */
6603 /* If range_start(or end) < 0, we assume -range_start(end)
6604 is the offset of the collating symbol which is specified
6605 as the character of the range start(end). */
6609 range_start_char = charset_top - (*workp++);
6612 str_buf[0] = *workp++;
6613 range_start_char = str_buf;
6618 range_end_char = charset_top - (*workp++);
6621 str_buf[4] = *workp++;
6622 range_end_char = str_buf + 4;
6625 if (wcscoll(range_start_char, str_buf+2) <= 0 &&
6626 wcscoll(str_buf+2, range_end_char) <= 0)
6628 goto char_set_matched;
6632 /* match with char? */
6633 for (; workp < p ; workp++)
6635 goto char_set_matched;
6642 /* Cast to `unsigned' instead of `unsigned char' in case the
6643 bit list is a full 32 bytes long. */
6644 if (c < (unsigned) (*p * BYTEWIDTH)
6645 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6650 if (!not) goto fail;
6651 #undef WORK_BUFFER_SIZE
6653 SET_REGS_MATCHED ();
6659 /* The beginning of a group is represented by start_memory.
6660 The arguments are the register number in the next byte, and the
6661 number of groups inner to this one in the next. The text
6662 matched within the group is recorded (in the internal
6663 registers data structure) under the register number. */
6665 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6666 (long int) *p, (long int) p[1]);
6668 /* Find out if this group can match the empty string. */
6669 p1 = p; /* To send to group_match_null_string_p. */
6671 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
6672 REG_MATCH_NULL_STRING_P (reg_info[*p])
6673 = PREFIX(group_match_null_string_p) (&p1, pend, reg_info);
6675 /* Save the position in the string where we were the last time
6676 we were at this open-group operator in case the group is
6677 operated upon by a repetition operator, e.g., with `(a*)*b'
6678 against `ab'; then we want to ignore where we are now in
6679 the string in case this attempt to match fails. */
6680 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6681 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
6683 DEBUG_PRINT2 (" old_regstart: %d\n",
6684 POINTER_TO_OFFSET (old_regstart[*p]));
6687 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
6689 IS_ACTIVE (reg_info[*p]) = 1;
6690 MATCHED_SOMETHING (reg_info[*p]) = 0;
6692 /* Clear this whenever we change the register activity status. */
6693 set_regs_matched_done = 0;
6695 /* This is the new highest active register. */
6696 highest_active_reg = *p;
6698 /* If nothing was active before, this is the new lowest active
6700 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6701 lowest_active_reg = *p;
6703 /* Move past the register number and inner group count. */
6705 just_past_start_mem = p;
6710 /* The stop_memory opcode represents the end of a group. Its
6711 arguments are the same as start_memory's: the register
6712 number, and the number of inner groups. */
6714 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6715 (long int) *p, (long int) p[1]);
6717 /* We need to save the string position the last time we were at
6718 this close-group operator in case the group is operated
6719 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6720 against `aba'; then we want to ignore where we are now in
6721 the string in case this attempt to match fails. */
6722 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6723 ? REG_UNSET (regend[*p]) ? d : regend[*p]
6725 DEBUG_PRINT2 (" old_regend: %d\n",
6726 POINTER_TO_OFFSET (old_regend[*p]));
6729 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
6731 /* This register isn't active anymore. */
6732 IS_ACTIVE (reg_info[*p]) = 0;
6734 /* Clear this whenever we change the register activity status. */
6735 set_regs_matched_done = 0;
6737 /* If this was the only register active, nothing is active
6739 if (lowest_active_reg == highest_active_reg)
6741 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6742 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6745 { /* We must scan for the new highest active register, since
6746 it isn't necessarily one less than now: consider
6747 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6748 new highest active register is 1. */
6750 while (r > 0 && !IS_ACTIVE (reg_info[r]))
6753 /* If we end up at register zero, that means that we saved
6754 the registers as the result of an `on_failure_jump', not
6755 a `start_memory', and we jumped to past the innermost
6756 `stop_memory'. For example, in ((.)*) we save
6757 registers 1 and 2 as a result of the *, but when we pop
6758 back to the second ), we are at the stop_memory 1.
6759 Thus, nothing is active. */
6762 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6763 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6766 highest_active_reg = r;
6769 /* If just failed to match something this time around with a
6770 group that's operated on by a repetition operator, try to
6771 force exit from the ``loop'', and restore the register
6772 information for this group that we had before trying this
6774 if ((!MATCHED_SOMETHING (reg_info[*p])
6775 || just_past_start_mem == p - 1)
6778 boolean is_a_jump_n = false;
6782 switch ((re_opcode_t) *p1++)
6786 case pop_failure_jump:
6787 case maybe_pop_jump:
6789 case dummy_failure_jump:
6790 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6792 p1 += OFFSET_ADDRESS_SIZE;
6800 /* If the next operation is a jump backwards in the pattern
6801 to an on_failure_jump right before the start_memory
6802 corresponding to this stop_memory, exit from the loop
6803 by forcing a failure after pushing on the stack the
6804 on_failure_jump's jump in the pattern, and d. */
6805 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
6806 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory
6807 && p1[2+OFFSET_ADDRESS_SIZE] == *p)
6809 /* If this group ever matched anything, then restore
6810 what its registers were before trying this last
6811 failed match, e.g., with `(a*)*b' against `ab' for
6812 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6813 against `aba' for regend[3].
6815 Also restore the registers for inner groups for,
6816 e.g., `((a*)(b*))*' against `aba' (register 3 would
6817 otherwise get trashed). */
6819 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
6823 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
6825 /* Restore this and inner groups' (if any) registers. */
6826 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
6829 regstart[r] = old_regstart[r];
6831 /* xx why this test? */
6832 if (old_regend[r] >= regstart[r])
6833 regend[r] = old_regend[r];
6837 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6838 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
6844 /* Move past the register number and the inner group count. */
6849 /* \<digit> has been turned into a `duplicate' command which is
6850 followed by the numeric value of <digit> as the register number. */
6853 register const CHAR_T *d2, *dend2;
6854 int regno = *p++; /* Get which register to match against. */
6855 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
6857 /* Can't back reference a group which we've never matched. */
6858 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
6861 /* Where in input to try to start matching. */
6862 d2 = regstart[regno];
6864 /* Where to stop matching; if both the place to start and
6865 the place to stop matching are in the same string, then
6866 set to the place to stop, otherwise, for now have to use
6867 the end of the first string. */
6869 dend2 = ((FIRST_STRING_P (regstart[regno])
6870 == FIRST_STRING_P (regend[regno]))
6871 ? regend[regno] : end_match_1);
6874 /* If necessary, advance to next segment in register
6878 if (dend2 == end_match_2) break;
6879 if (dend2 == regend[regno]) break;
6881 /* End of string1 => advance to string2. */
6883 dend2 = regend[regno];
6885 /* At end of register contents => success */
6886 if (d2 == dend2) break;
6888 /* If necessary, advance to next segment in data. */
6891 /* How many characters left in this segment to match. */
6894 /* Want how many consecutive characters we can match in
6895 one shot, so, if necessary, adjust the count. */
6896 if (mcnt > dend2 - d2)
6899 /* Compare that many; failure if mismatch, else move
6902 ? PREFIX(bcmp_translate) (d, d2, mcnt, translate)
6903 : memcmp (d, d2, mcnt*sizeof(UCHAR_T)))
6905 d += mcnt, d2 += mcnt;
6907 /* Do this because we've match some characters. */
6908 SET_REGS_MATCHED ();
6914 /* begline matches the empty string at the beginning of the string
6915 (unless `not_bol' is set in `bufp'), and, if
6916 `newline_anchor' is set, after newlines. */
6918 DEBUG_PRINT1 ("EXECUTING begline.\n");
6920 if (AT_STRINGS_BEG (d))
6922 if (!bufp->not_bol) break;
6924 else if (d[-1] == '\n' && bufp->newline_anchor)
6928 /* In all other cases, we fail. */
6932 /* endline is the dual of begline. */
6934 DEBUG_PRINT1 ("EXECUTING endline.\n");
6936 if (AT_STRINGS_END (d))
6938 if (!bufp->not_eol) break;
6941 /* We have to ``prefetch'' the next character. */
6942 else if ((d == end1 ? *string2 : *d) == '\n'
6943 && bufp->newline_anchor)
6950 /* Match at the very beginning of the data. */
6952 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
6953 if (AT_STRINGS_BEG (d))
6958 /* Match at the very end of the data. */
6960 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
6961 if (AT_STRINGS_END (d))
6966 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
6967 pushes NULL as the value for the string on the stack. Then
6968 `pop_failure_point' will keep the current value for the
6969 string, instead of restoring it. To see why, consider
6970 matching `foo\nbar' against `.*\n'. The .* matches the foo;
6971 then the . fails against the \n. But the next thing we want
6972 to do is match the \n against the \n; if we restored the
6973 string value, we would be back at the foo.
6975 Because this is used only in specific cases, we don't need to
6976 check all the things that `on_failure_jump' does, to make
6977 sure the right things get saved on the stack. Hence we don't
6978 share its code. The only reason to push anything on the
6979 stack at all is that otherwise we would have to change
6980 `anychar's code to do something besides goto fail in this
6981 case; that seems worse than this. */
6982 case on_failure_keep_string_jump:
6983 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
6985 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6987 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
6989 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
6992 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
6996 /* Uses of on_failure_jump:
6998 Each alternative starts with an on_failure_jump that points
6999 to the beginning of the next alternative. Each alternative
7000 except the last ends with a jump that in effect jumps past
7001 the rest of the alternatives. (They really jump to the
7002 ending jump of the following alternative, because tensioning
7003 these jumps is a hassle.)
7005 Repeats start with an on_failure_jump that points past both
7006 the repetition text and either the following jump or
7007 pop_failure_jump back to this on_failure_jump. */
7008 case on_failure_jump:
7010 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
7012 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7014 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
7016 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
7019 /* If this on_failure_jump comes right before a group (i.e.,
7020 the original * applied to a group), save the information
7021 for that group and all inner ones, so that if we fail back
7022 to this point, the group's information will be correct.
7023 For example, in \(a*\)*\1, we need the preceding group,
7024 and in \(zz\(a*\)b*\)\2, we need the inner group. */
7026 /* We can't use `p' to check ahead because we push
7027 a failure point to `p + mcnt' after we do this. */
7030 /* We need to skip no_op's before we look for the
7031 start_memory in case this on_failure_jump is happening as
7032 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7034 while (p1 < pend && (re_opcode_t) *p1 == no_op)
7037 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
7039 /* We have a new highest active register now. This will
7040 get reset at the start_memory we are about to get to,
7041 but we will have saved all the registers relevant to
7042 this repetition op, as described above. */
7043 highest_active_reg = *(p1 + 1) + *(p1 + 2);
7044 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
7045 lowest_active_reg = *(p1 + 1);
7048 DEBUG_PRINT1 (":\n");
7049 PUSH_FAILURE_POINT (p + mcnt, d, -2);
7053 /* A smart repeat ends with `maybe_pop_jump'.
7054 We change it to either `pop_failure_jump' or `jump'. */
7055 case maybe_pop_jump:
7056 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7057 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
7059 register UCHAR_T *p2 = p;
7061 /* Compare the beginning of the repeat with what in the
7062 pattern follows its end. If we can establish that there
7063 is nothing that they would both match, i.e., that we
7064 would have to backtrack because of (as in, e.g., `a*a')
7065 then we can change to pop_failure_jump, because we'll
7066 never have to backtrack.
7068 This is not true in the case of alternatives: in
7069 `(a|ab)*' we do need to backtrack to the `ab' alternative
7070 (e.g., if the string was `ab'). But instead of trying to
7071 detect that here, the alternative has put on a dummy
7072 failure point which is what we will end up popping. */
7074 /* Skip over open/close-group commands.
7075 If what follows this loop is a ...+ construct,
7076 look at what begins its body, since we will have to
7077 match at least one of that. */
7081 && ((re_opcode_t) *p2 == stop_memory
7082 || (re_opcode_t) *p2 == start_memory))
7084 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend
7085 && (re_opcode_t) *p2 == dummy_failure_jump)
7086 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE;
7092 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7093 to the `maybe_finalize_jump' of this case. Examine what
7096 /* If we're at the end of the pattern, we can change. */
7099 /* Consider what happens when matching ":\(.*\)"
7100 against ":/". I don't really understand this code
7102 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7105 (" End of pattern: change to `pop_failure_jump'.\n");
7108 else if ((re_opcode_t) *p2 == exactn
7110 || (re_opcode_t) *p2 == exactn_bin
7112 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
7115 = *p2 == (UCHAR_T) endline ? '\n' : p2[2];
7117 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn
7119 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin
7121 ) && p1[3+OFFSET_ADDRESS_SIZE] != c)
7123 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7126 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7128 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]);
7130 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7132 (char) p1[3+OFFSET_ADDRESS_SIZE]);
7137 else if ((re_opcode_t) p1[3] == charset
7138 || (re_opcode_t) p1[3] == charset_not)
7140 int not = (re_opcode_t) p1[3] == charset_not;
7142 if (c < (unsigned) (p1[4] * BYTEWIDTH)
7143 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
7146 /* `not' is equal to 1 if c would match, which means
7147 that we can't change to pop_failure_jump. */
7150 p[-3] = (unsigned char) pop_failure_jump;
7151 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7154 #endif /* not WCHAR */
7157 else if ((re_opcode_t) *p2 == charset)
7159 /* We win if the first character of the loop is not part
7161 if ((re_opcode_t) p1[3] == exactn
7162 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
7163 && (p2[2 + p1[5] / BYTEWIDTH]
7164 & (1 << (p1[5] % BYTEWIDTH)))))
7166 p[-3] = (unsigned char) pop_failure_jump;
7167 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7170 else if ((re_opcode_t) p1[3] == charset_not)
7173 /* We win if the charset_not inside the loop
7174 lists every character listed in the charset after. */
7175 for (idx = 0; idx < (int) p2[1]; idx++)
7176 if (! (p2[2 + idx] == 0
7177 || (idx < (int) p1[4]
7178 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
7183 p[-3] = (unsigned char) pop_failure_jump;
7184 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7187 else if ((re_opcode_t) p1[3] == charset)
7190 /* We win if the charset inside the loop
7191 has no overlap with the one after the loop. */
7193 idx < (int) p2[1] && idx < (int) p1[4];
7195 if ((p2[2 + idx] & p1[5 + idx]) != 0)
7198 if (idx == p2[1] || idx == p1[4])
7200 p[-3] = (unsigned char) pop_failure_jump;
7201 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7205 #endif /* not WCHAR */
7207 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */
7208 if ((re_opcode_t) p[-1] != pop_failure_jump)
7210 p[-1] = (UCHAR_T) jump;
7211 DEBUG_PRINT1 (" Match => jump.\n");
7212 goto unconditional_jump;
7214 /* Note fall through. */
7217 /* The end of a simple repeat has a pop_failure_jump back to
7218 its matching on_failure_jump, where the latter will push a
7219 failure point. The pop_failure_jump takes off failure
7220 points put on by this pop_failure_jump's matching
7221 on_failure_jump; we got through the pattern to here from the
7222 matching on_failure_jump, so didn't fail. */
7223 case pop_failure_jump:
7225 /* We need to pass separate storage for the lowest and
7226 highest registers, even though we don't care about the
7227 actual values. Otherwise, we will restore only one
7228 register from the stack, since lowest will == highest in
7229 `pop_failure_point'. */
7230 active_reg_t dummy_low_reg, dummy_high_reg;
7231 UCHAR_T *pdummy = NULL;
7232 const CHAR_T *sdummy = NULL;
7234 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7235 POP_FAILURE_POINT (sdummy, pdummy,
7236 dummy_low_reg, dummy_high_reg,
7237 reg_dummy, reg_dummy, reg_info_dummy);
7239 /* Note fall through. */
7243 DEBUG_PRINT2 ("\n%p: ", p);
7245 DEBUG_PRINT2 ("\n0x%x: ", p);
7247 /* Note fall through. */
7249 /* Unconditionally jump (without popping any failure points). */
7251 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
7252 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
7253 p += mcnt; /* Do the jump. */
7255 DEBUG_PRINT2 ("(to %p).\n", p);
7257 DEBUG_PRINT2 ("(to 0x%x).\n", p);
7262 /* We need this opcode so we can detect where alternatives end
7263 in `group_match_null_string_p' et al. */
7265 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7266 goto unconditional_jump;
7269 /* Normally, the on_failure_jump pushes a failure point, which
7270 then gets popped at pop_failure_jump. We will end up at
7271 pop_failure_jump, also, and with a pattern of, say, `a+', we
7272 are skipping over the on_failure_jump, so we have to push
7273 something meaningless for pop_failure_jump to pop. */
7274 case dummy_failure_jump:
7275 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7276 /* It doesn't matter what we push for the string here. What
7277 the code at `fail' tests is the value for the pattern. */
7278 PUSH_FAILURE_POINT (NULL, NULL, -2);
7279 goto unconditional_jump;
7282 /* At the end of an alternative, we need to push a dummy failure
7283 point in case we are followed by a `pop_failure_jump', because
7284 we don't want the failure point for the alternative to be
7285 popped. For example, matching `(a|ab)*' against `aab'
7286 requires that we match the `ab' alternative. */
7287 case push_dummy_failure:
7288 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7289 /* See comments just above at `dummy_failure_jump' about the
7291 PUSH_FAILURE_POINT (NULL, NULL, -2);
7294 /* Have to succeed matching what follows at least n times.
7295 After that, handle like `on_failure_jump'. */
7297 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7298 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
7301 /* Originally, this is how many times we HAVE to succeed. */
7305 p += OFFSET_ADDRESS_SIZE;
7306 STORE_NUMBER_AND_INCR (p, mcnt);
7308 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE
7311 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE
7318 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7319 p + OFFSET_ADDRESS_SIZE);
7321 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7322 p + OFFSET_ADDRESS_SIZE);
7326 p[1] = (UCHAR_T) no_op;
7328 p[2] = (UCHAR_T) no_op;
7329 p[3] = (UCHAR_T) no_op;
7336 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7337 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
7339 /* Originally, this is how many times we CAN jump. */
7343 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt);
7346 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE,
7349 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE,
7352 goto unconditional_jump;
7354 /* If don't have to jump any more, skip over the rest of command. */
7356 p += 2 * OFFSET_ADDRESS_SIZE;
7361 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7363 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7365 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7367 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
7369 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
7371 STORE_NUMBER (p1, mcnt);
7376 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7377 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7378 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7379 macro and introducing temporary variables works around the bug. */
7382 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7383 if (AT_WORD_BOUNDARY (d))
7388 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7389 if (AT_WORD_BOUNDARY (d))
7395 boolean prevchar, thischar;
7397 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7398 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7401 prevchar = WORDCHAR_P (d - 1);
7402 thischar = WORDCHAR_P (d);
7403 if (prevchar != thischar)
7410 boolean prevchar, thischar;
7412 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7413 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7416 prevchar = WORDCHAR_P (d - 1);
7417 thischar = WORDCHAR_P (d);
7418 if (prevchar != thischar)
7425 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7426 if (!AT_STRINGS_END (d) && WORDCHAR_P (d)
7427 && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
7432 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7433 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
7434 && (AT_STRINGS_END (d) || !WORDCHAR_P (d)))
7440 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7441 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
7446 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7447 if (PTR_CHAR_POS ((unsigned char *) d) != point)
7452 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7453 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
7458 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
7463 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7467 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7469 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
7471 SET_REGS_MATCHED ();
7475 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
7477 goto matchnotsyntax;
7480 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7484 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7486 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
7488 SET_REGS_MATCHED ();
7491 #else /* not emacs */
7493 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7495 if (!WORDCHAR_P (d))
7497 SET_REGS_MATCHED ();
7502 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7506 SET_REGS_MATCHED ();
7509 #endif /* not emacs */
7514 continue; /* Successfully executed one pattern command; keep going. */
7517 /* We goto here if a matching operation fails. */
7519 if (!FAIL_STACK_EMPTY ())
7520 { /* A restart point is known. Restore to that state. */
7521 DEBUG_PRINT1 ("\nFAIL:\n");
7522 POP_FAILURE_POINT (d, p,
7523 lowest_active_reg, highest_active_reg,
7524 regstart, regend, reg_info);
7526 /* If this failure point is a dummy, try the next one. */
7530 /* If we failed to the end of the pattern, don't examine *p. */
7534 boolean is_a_jump_n = false;
7536 /* If failed to a backwards jump that's part of a repetition
7537 loop, need to pop this failure point and use the next one. */
7538 switch ((re_opcode_t) *p)
7542 case maybe_pop_jump:
7543 case pop_failure_jump:
7546 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7549 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
7551 && (re_opcode_t) *p1 == on_failure_jump))
7559 if (d >= string1 && d <= end1)
7563 break; /* Matching at this starting point really fails. */
7567 goto restore_best_regs;
7571 return -1; /* Failure to match. */
7574 /* Subroutine definitions for re_match_2. */
7577 /* We are passed P pointing to a register number after a start_memory.
7579 Return true if the pattern up to the corresponding stop_memory can
7580 match the empty string, and false otherwise.
7582 If we find the matching stop_memory, sets P to point to one past its number.
7583 Otherwise, sets P to an undefined byte less than or equal to END.
7585 We don't handle duplicates properly (yet). */
7588 PREFIX(group_match_null_string_p) (p, end, reg_info)
7590 PREFIX(register_info_type) *reg_info;
7593 /* Point to after the args to the start_memory. */
7594 UCHAR_T *p1 = *p + 2;
7598 /* Skip over opcodes that can match nothing, and return true or
7599 false, as appropriate, when we get to one that can't, or to the
7600 matching stop_memory. */
7602 switch ((re_opcode_t) *p1)
7604 /* Could be either a loop or a series of alternatives. */
7605 case on_failure_jump:
7607 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7609 /* If the next operation is not a jump backwards in the
7614 /* Go through the on_failure_jumps of the alternatives,
7615 seeing if any of the alternatives cannot match nothing.
7616 The last alternative starts with only a jump,
7617 whereas the rest start with on_failure_jump and end
7618 with a jump, e.g., here is the pattern for `a|b|c':
7620 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7621 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7624 So, we have to first go through the first (n-1)
7625 alternatives and then deal with the last one separately. */
7628 /* Deal with the first (n-1) alternatives, which start
7629 with an on_failure_jump (see above) that jumps to right
7630 past a jump_past_alt. */
7632 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] ==
7635 /* `mcnt' holds how many bytes long the alternative
7636 is, including the ending `jump_past_alt' and
7639 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt -
7640 (1 + OFFSET_ADDRESS_SIZE),
7644 /* Move to right after this alternative, including the
7648 /* Break if it's the beginning of an n-th alternative
7649 that doesn't begin with an on_failure_jump. */
7650 if ((re_opcode_t) *p1 != on_failure_jump)
7653 /* Still have to check that it's not an n-th
7654 alternative that starts with an on_failure_jump. */
7656 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7657 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] !=
7660 /* Get to the beginning of the n-th alternative. */
7661 p1 -= 1 + OFFSET_ADDRESS_SIZE;
7666 /* Deal with the last alternative: go back and get number
7667 of the `jump_past_alt' just before it. `mcnt' contains
7668 the length of the alternative. */
7669 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE);
7671 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info))
7674 p1 += mcnt; /* Get past the n-th alternative. */
7680 assert (p1[1] == **p);
7686 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7689 } /* while p1 < end */
7692 } /* group_match_null_string_p */
7695 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7696 It expects P to be the first byte of a single alternative and END one
7697 byte past the last. The alternative can contain groups. */
7700 PREFIX(alt_match_null_string_p) (p, end, reg_info)
7702 PREFIX(register_info_type) *reg_info;
7709 /* Skip over opcodes that can match nothing, and break when we get
7710 to one that can't. */
7712 switch ((re_opcode_t) *p1)
7715 case on_failure_jump:
7717 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7722 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7725 } /* while p1 < end */
7728 } /* alt_match_null_string_p */
7731 /* Deals with the ops common to group_match_null_string_p and
7732 alt_match_null_string_p.
7734 Sets P to one after the op and its arguments, if any. */
7737 PREFIX(common_op_match_null_string_p) (p, end, reg_info)
7739 PREFIX(register_info_type) *reg_info;
7746 switch ((re_opcode_t) *p1++)
7766 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
7767 ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info);
7769 /* Have to set this here in case we're checking a group which
7770 contains a group and a back reference to it. */
7772 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
7773 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
7779 /* If this is an optimized succeed_n for zero times, make the jump. */
7781 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7789 /* Get to the number of times to succeed. */
7790 p1 += OFFSET_ADDRESS_SIZE;
7791 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7795 p1 -= 2 * OFFSET_ADDRESS_SIZE;
7796 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7804 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
7809 p1 += 2 * OFFSET_ADDRESS_SIZE;
7812 /* All other opcodes mean we cannot match the empty string. */
7818 } /* common_op_match_null_string_p */
7821 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7822 bytes; nonzero otherwise. */
7825 PREFIX(bcmp_translate) (s1, s2, len, translate)
7826 const CHAR_T *s1, *s2;
7828 RE_TRANSLATE_TYPE translate;
7830 register const UCHAR_T *p1 = (const UCHAR_T *) s1;
7831 register const UCHAR_T *p2 = (const UCHAR_T *) s2;
7835 if (((*p1<=0xff)?translate[*p1++]:*p1++)
7836 != ((*p2<=0xff)?translate[*p2++]:*p2++))
7839 if (translate[*p1++] != translate[*p2++]) return 1;
7847 #else /* not INSIDE_RECURSION */
7849 /* Entry points for GNU code. */
7851 /* re_compile_pattern is the GNU regular expression compiler: it
7852 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7853 Returns 0 if the pattern was valid, otherwise an error string.
7855 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7856 are set in BUFP on entry.
7858 We call regex_compile to do the actual compilation. */
7861 re_compile_pattern (pattern, length, bufp)
7862 const char *pattern;
7864 struct re_pattern_buffer *bufp;
7868 /* GNU code is written to assume at least RE_NREGS registers will be set
7869 (and at least one extra will be -1). */
7870 bufp->regs_allocated = REGS_UNALLOCATED;
7872 /* And GNU code determines whether or not to get register information
7873 by passing null for the REGS argument to re_match, etc., not by
7877 /* Match anchors at newline. */
7878 bufp->newline_anchor = 1;
7881 if (MB_CUR_MAX != 1)
7882 ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp);
7885 ret = byte_regex_compile (pattern, length, re_syntax_options, bufp);
7889 return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7892 weak_alias (__re_compile_pattern, re_compile_pattern)
7895 /* Entry points compatible with 4.2 BSD regex library. We don't define
7896 them unless specifically requested. */
7898 #if defined _REGEX_RE_COMP || defined _LIBC
7900 /* BSD has one and only one pattern buffer. */
7901 static struct re_pattern_buffer re_comp_buf;
7905 /* Make these definitions weak in libc, so POSIX programs can redefine
7906 these names if they don't use our functions, and still use
7907 regcomp/regexec below without link errors. */
7917 if (!re_comp_buf.buffer)
7918 return gettext ("No previous regular expression");
7922 if (!re_comp_buf.buffer)
7924 re_comp_buf.buffer = (unsigned char *) malloc (200);
7925 if (re_comp_buf.buffer == NULL)
7926 return (char *) gettext (re_error_msgid
7927 + re_error_msgid_idx[(int) REG_ESPACE]);
7928 re_comp_buf.allocated = 200;
7930 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
7931 if (re_comp_buf.fastmap == NULL)
7932 return (char *) gettext (re_error_msgid
7933 + re_error_msgid_idx[(int) REG_ESPACE]);
7936 /* Since `re_exec' always passes NULL for the `regs' argument, we
7937 don't need to initialize the pattern buffer fields which affect it. */
7939 /* Match anchors at newlines. */
7940 re_comp_buf.newline_anchor = 1;
7943 if (MB_CUR_MAX != 1)
7944 ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7947 ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7952 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
7953 return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7964 const int len = strlen (s);
7966 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
7969 #endif /* _REGEX_RE_COMP */
7971 /* POSIX.2 functions. Don't define these for Emacs. */
7975 /* regcomp takes a regular expression as a string and compiles it.
7977 PREG is a regex_t *. We do not expect any fields to be initialized,
7978 since POSIX says we shouldn't. Thus, we set
7980 `buffer' to the compiled pattern;
7981 `used' to the length of the compiled pattern;
7982 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
7983 REG_EXTENDED bit in CFLAGS is set; otherwise, to
7984 RE_SYNTAX_POSIX_BASIC;
7985 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
7986 `fastmap' to an allocated space for the fastmap;
7987 `fastmap_accurate' to zero;
7988 `re_nsub' to the number of subexpressions in PATTERN.
7990 PATTERN is the address of the pattern string.
7992 CFLAGS is a series of bits which affect compilation.
7994 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
7995 use POSIX basic syntax.
7997 If REG_NEWLINE is set, then . and [^...] don't match newline.
7998 Also, regexec will try a match beginning after every newline.
8000 If REG_ICASE is set, then we considers upper- and lowercase
8001 versions of letters to be equivalent when matching.
8003 If REG_NOSUB is set, then when PREG is passed to regexec, that
8004 routine will report only success or failure, and nothing about the
8007 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
8008 the return codes and their meanings.) */
8011 regcomp (preg, pattern, cflags)
8013 const char *pattern;
8018 = (cflags & REG_EXTENDED) ?
8019 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
8021 /* regex_compile will allocate the space for the compiled pattern. */
8023 preg->allocated = 0;
8026 /* Try to allocate space for the fastmap. */
8027 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
8029 if (cflags & REG_ICASE)
8034 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
8035 * sizeof (*(RE_TRANSLATE_TYPE)0));
8036 if (preg->translate == NULL)
8037 return (int) REG_ESPACE;
8039 /* Map uppercase characters to corresponding lowercase ones. */
8040 for (i = 0; i < CHAR_SET_SIZE; i++)
8041 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
8044 preg->translate = NULL;
8046 /* If REG_NEWLINE is set, newlines are treated differently. */
8047 if (cflags & REG_NEWLINE)
8048 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
8049 syntax &= ~RE_DOT_NEWLINE;
8050 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
8051 /* It also changes the matching behavior. */
8052 preg->newline_anchor = 1;
8055 preg->newline_anchor = 0;
8057 preg->no_sub = !!(cflags & REG_NOSUB);
8059 /* POSIX says a null character in the pattern terminates it, so we
8060 can use strlen here in compiling the pattern. */
8062 if (MB_CUR_MAX != 1)
8063 ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg);
8066 ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg);
8068 /* POSIX doesn't distinguish between an unmatched open-group and an
8069 unmatched close-group: both are REG_EPAREN. */
8070 if (ret == REG_ERPAREN) ret = REG_EPAREN;
8072 if (ret == REG_NOERROR && preg->fastmap)
8074 /* Compute the fastmap now, since regexec cannot modify the pattern
8076 if (re_compile_fastmap (preg) == -2)
8078 /* Some error occurred while computing the fastmap, just forget
8080 free (preg->fastmap);
8081 preg->fastmap = NULL;
8088 weak_alias (__regcomp, regcomp)
8092 /* regexec searches for a given pattern, specified by PREG, in the
8095 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8096 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8097 least NMATCH elements, and we set them to the offsets of the
8098 corresponding matched substrings.
8100 EFLAGS specifies `execution flags' which affect matching: if
8101 REG_NOTBOL is set, then ^ does not match at the beginning of the
8102 string; if REG_NOTEOL is set, then $ does not match at the end.
8104 We return 0 if we find a match and REG_NOMATCH if not. */
8107 regexec (preg, string, nmatch, pmatch, eflags)
8108 const regex_t *preg;
8111 regmatch_t pmatch[];
8115 struct re_registers regs;
8116 regex_t private_preg;
8117 int len = strlen (string);
8118 boolean want_reg_info = !preg->no_sub && nmatch > 0;
8120 private_preg = *preg;
8122 private_preg.not_bol = !!(eflags & REG_NOTBOL);
8123 private_preg.not_eol = !!(eflags & REG_NOTEOL);
8125 /* The user has told us exactly how many registers to return
8126 information about, via `nmatch'. We have to pass that on to the
8127 matching routines. */
8128 private_preg.regs_allocated = REGS_FIXED;
8132 regs.num_regs = nmatch;
8133 regs.start = TALLOC (nmatch * 2, regoff_t);
8134 if (regs.start == NULL)
8135 return (int) REG_NOMATCH;
8136 regs.end = regs.start + nmatch;
8139 /* Perform the searching operation. */
8140 ret = re_search (&private_preg, string, len,
8141 /* start: */ 0, /* range: */ len,
8142 want_reg_info ? ®s : (struct re_registers *) 0);
8144 /* Copy the register information to the POSIX structure. */
8151 for (r = 0; r < nmatch; r++)
8153 pmatch[r].rm_so = regs.start[r];
8154 pmatch[r].rm_eo = regs.end[r];
8158 /* If we needed the temporary register info, free the space now. */
8162 /* We want zero return to mean success, unlike `re_search'. */
8163 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
8166 weak_alias (__regexec, regexec)
8170 /* Returns a message corresponding to an error code, ERRCODE, returned
8171 from either regcomp or regexec. We don't use PREG here. */
8174 regerror (errcode, preg, errbuf, errbuf_size)
8176 const regex_t *preg;
8184 || errcode >= (int) (sizeof (re_error_msgid_idx)
8185 / sizeof (re_error_msgid_idx[0])))
8186 /* Only error codes returned by the rest of the code should be passed
8187 to this routine. If we are given anything else, or if other regex
8188 code generates an invalid error code, then the program has a bug.
8189 Dump core so we can fix it. */
8192 msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
8194 msg_size = strlen (msg) + 1; /* Includes the null. */
8196 if (errbuf_size != 0)
8198 if (msg_size > errbuf_size)
8200 #if defined HAVE_MEMPCPY || defined _LIBC
8201 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
8203 memcpy (errbuf, msg, errbuf_size - 1);
8204 errbuf[errbuf_size - 1] = 0;
8208 memcpy (errbuf, msg, msg_size);
8214 weak_alias (__regerror, regerror)
8218 /* Free dynamically allocated space used by PREG. */
8224 if (preg->buffer != NULL)
8225 free (preg->buffer);
8226 preg->buffer = NULL;
8228 preg->allocated = 0;
8231 if (preg->fastmap != NULL)
8232 free (preg->fastmap);
8233 preg->fastmap = NULL;
8234 preg->fastmap_accurate = 0;
8236 if (preg->translate != NULL)
8237 free (preg->translate);
8238 preg->translate = NULL;
8241 weak_alias (__regfree, regfree)
8244 #endif /* not emacs */
8246 #endif /* not INSIDE_RECURSION */
8250 #undef STORE_NUMBER_AND_INCR
8251 #undef EXTRACT_NUMBER
8252 #undef EXTRACT_NUMBER_AND_INCR
8254 #undef DEBUG_PRINT_COMPILED_PATTERN
8255 #undef DEBUG_PRINT_DOUBLE_STRING
8257 #undef INIT_FAIL_STACK
8258 #undef RESET_FAIL_STACK
8259 #undef DOUBLE_FAIL_STACK
8260 #undef PUSH_PATTERN_OP
8261 #undef PUSH_FAILURE_POINTER
8262 #undef PUSH_FAILURE_INT
8263 #undef PUSH_FAILURE_ELT
8264 #undef POP_FAILURE_POINTER
8265 #undef POP_FAILURE_INT
8266 #undef POP_FAILURE_ELT
8269 #undef PUSH_FAILURE_POINT
8270 #undef POP_FAILURE_POINT
8272 #undef REG_UNSET_VALUE
8280 #undef INIT_BUF_SIZE
8281 #undef GET_BUFFER_SPACE
8289 #undef EXTEND_BUFFER
8290 #undef GET_UNSIGNED_NUMBER
8291 #undef FREE_STACK_RETURN
8293 # undef POINTER_TO_OFFSET
8294 # undef MATCHING_IN_FRST_STRING
8296 # undef AT_STRINGS_BEG
8297 # undef AT_STRINGS_END
8300 # undef FREE_VARIABLES
8301 # undef NO_HIGHEST_ACTIVE_REG
8302 # undef NO_LOWEST_ACTIVE_REG
8306 # undef COMPILED_BUFFER_VAR
8307 # undef OFFSET_ADDRESS_SIZE
8308 # undef CHAR_CLASS_SIZE
8315 # define DEFINED_ONCE