1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
6 Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
7 2002, 2003, 2004 Free Software Foundation, Inc.
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 2, or (at your option)
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License along
20 with this program; if not, write to the Free Software Foundation,
21 Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
23 /* AIX requires this to be the first thing in the file. */
24 #if defined _AIX && !defined REGEX_MALLOC
35 #ifndef INSIDE_RECURSION
39 # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
41 /* For platform which support the ISO C amendement 1 functionality we
42 support user defined character classes. */
43 # if defined _LIBC || WIDE_CHAR_SUPPORT
44 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
50 /* We have to keep the namespace clean. */
51 # define regfree(preg) __regfree (preg)
52 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
53 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
54 # define regerror(errcode, preg, errbuf, errbuf_size) \
55 __regerror(errcode, preg, errbuf, errbuf_size)
56 # define re_set_registers(bu, re, nu, st, en) \
57 __re_set_registers (bu, re, nu, st, en)
58 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
59 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
60 # define re_match(bufp, string, size, pos, regs) \
61 __re_match (bufp, string, size, pos, regs)
62 # define re_search(bufp, string, size, startpos, range, regs) \
63 __re_search (bufp, string, size, startpos, range, regs)
64 # define re_compile_pattern(pattern, length, bufp) \
65 __re_compile_pattern (pattern, length, bufp)
66 # define re_set_syntax(syntax) __re_set_syntax (syntax)
67 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
68 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
69 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
71 # define btowc __btowc
72 # define iswctype __iswctype
73 # define mbrtowc __mbrtowc
74 # define wcslen __wcslen
75 # define wcscoll __wcscoll
76 # define wcrtomb __wcrtomb
78 /* We are also using some library internals. */
79 # include <locale/localeinfo.h>
80 # include <locale/elem-hash.h>
81 # include <langinfo.h>
82 # include <locale/coll-lookup.h>
88 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
89 /* This define is so xgettext can find the internationalizable strings. */
90 # define gettext_noop(msgid) msgid
92 /* This is for other GNU distributions with internationalized messages. */
96 /* Support for bounded pointers. */
97 # if !defined _LIBC && !defined __BOUNDED_POINTERS__
98 # define __bounded /* nothing */
99 # define __unbounded /* nothing */
100 # define __ptrvalue /* nothing */
103 /* The `emacs' switch turns on certain matching commands
104 that make sense only in Emacs. */
111 # else /* not emacs */
113 /* If we are not linking with Emacs proper,
114 we can't use the relocating allocator
115 even if config.h says that we can. */
120 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
121 If nothing else has been done, use the method below. */
122 # ifdef INHIBIT_STRING_HEADER
123 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
124 # if !defined bzero && !defined bcopy
125 # undef INHIBIT_STRING_HEADER
130 /* This is the normal way of making sure we have a bcopy and a bzero.
131 This is used in most programs--a few other programs avoid this
132 by defining INHIBIT_STRING_HEADER. */
133 # ifndef INHIBIT_STRING_HEADER
137 # define bzero(s, n) (memset (s, '\0', n), (s))
139 # define bzero(s, n) __bzero (s, n)
144 /* Define the syntax stuff for \<, \>, etc. */
146 /* This must be nonzero for the wordchar and notwordchar pattern
147 commands in re_match_2. */
152 # ifdef SWITCH_ENUM_BUG
153 # define SWITCH_ENUM_CAST(x) ((int)(x))
155 # define SWITCH_ENUM_CAST(x) (x)
158 # endif /* not emacs */
163 # define MB_LEN_MAX 1
166 /* Get the interface, including the syntax bits. */
169 /* isalpha etc. are used for the character classes. */
172 /* Jim Meyering writes:
174 "... Some ctype macros are valid only for character codes that
175 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
176 using /bin/cc or gcc but without giving an ansi option). So, all
177 ctype uses should be through macros like ISPRINT... If
178 STDC_HEADERS is defined, then autoconf has verified that the ctype
179 macros don't need to be guarded with references to isascii. ...
180 Defining isascii to 1 should let any compiler worth its salt
181 eliminate the && through constant folding."
182 Solaris defines some of these symbols so we must undefine them first. */
184 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
185 # define IN_CTYPE_DOMAIN(c) 1
187 # define IN_CTYPE_DOMAIN(c) isascii(c)
191 # define ISBLANK(c) (IN_CTYPE_DOMAIN (c) && isblank (c))
193 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
196 # define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isgraph (c))
198 # define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isprint (c) && !isspace (c))
202 # define ISPRINT(c) (IN_CTYPE_DOMAIN (c) && isprint (c))
203 # define ISDIGIT(c) (IN_CTYPE_DOMAIN (c) && isdigit (c))
204 # define ISALNUM(c) (IN_CTYPE_DOMAIN (c) && isalnum (c))
205 # define ISALPHA(c) (IN_CTYPE_DOMAIN (c) && isalpha (c))
206 # define ISCNTRL(c) (IN_CTYPE_DOMAIN (c) && iscntrl (c))
207 # define ISLOWER(c) (IN_CTYPE_DOMAIN (c) && islower (c))
208 # define ISPUNCT(c) (IN_CTYPE_DOMAIN (c) && ispunct (c))
209 # define ISSPACE(c) (IN_CTYPE_DOMAIN (c) && isspace (c))
210 # define ISUPPER(c) (IN_CTYPE_DOMAIN (c) && isupper (c))
211 # define ISXDIGIT(c) (IN_CTYPE_DOMAIN (c) && isxdigit (c))
214 # define TOLOWER(c) _tolower(c)
216 # define TOLOWER(c) tolower(c)
220 /* How many characters in the character set. */
221 # define CHAR_SET_SIZE 256
225 extern char *re_syntax_table;
227 # else /* not SYNTAX_TABLE */
229 static char re_syntax_table[CHAR_SET_SIZE];
232 init_syntax_once (void)
239 bzero (re_syntax_table, sizeof re_syntax_table);
241 for (c = 0; c < CHAR_SET_SIZE; ++c)
243 re_syntax_table[c] = Sword;
245 re_syntax_table['_'] = Sword;
250 # endif /* not SYNTAX_TABLE */
252 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
256 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
257 use `alloca' instead of `malloc'. This is because using malloc in
258 re_search* or re_match* could cause memory leaks when C-g is used in
259 Emacs; also, malloc is slower and causes storage fragmentation. On
260 the other hand, malloc is more portable, and easier to debug.
262 Because we sometimes use alloca, some routines have to be macros,
263 not functions -- `alloca'-allocated space disappears at the end of the
264 function it is called in. */
268 # define REGEX_ALLOCATE malloc
269 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
270 # define REGEX_FREE free
272 # else /* not REGEX_MALLOC */
274 /* Emacs already defines alloca, sometimes. */
277 /* Make alloca work the best possible way. */
280 # endif /* not alloca */
282 # define REGEX_ALLOCATE alloca
284 /* Assumes a `char *destination' variable. */
285 # define REGEX_REALLOCATE(source, osize, nsize) \
286 (destination = (char *) alloca (nsize), \
287 memcpy (destination, source, osize))
289 /* No need to do anything to free, after alloca. */
290 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
292 # endif /* not REGEX_MALLOC */
294 /* Define how to allocate the failure stack. */
296 # if defined REL_ALLOC && defined REGEX_MALLOC
298 # define REGEX_ALLOCATE_STACK(size) \
299 r_alloc (&failure_stack_ptr, (size))
300 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
301 r_re_alloc (&failure_stack_ptr, (nsize))
302 # define REGEX_FREE_STACK(ptr) \
303 r_alloc_free (&failure_stack_ptr)
305 # else /* not using relocating allocator */
309 # define REGEX_ALLOCATE_STACK malloc
310 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
311 # define REGEX_FREE_STACK free
313 # else /* not REGEX_MALLOC */
315 # define REGEX_ALLOCATE_STACK alloca
317 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
318 REGEX_REALLOCATE (source, osize, nsize)
319 /* No need to explicitly free anything. */
320 # define REGEX_FREE_STACK(arg)
322 # endif /* not REGEX_MALLOC */
323 # endif /* not using relocating allocator */
326 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
327 `string1' or just past its end. This works if PTR is NULL, which is
329 # define FIRST_STRING_P(ptr) \
330 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
332 /* (Re)Allocate N items of type T using malloc, or fail. */
333 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
334 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
335 # define RETALLOC_IF(addr, n, t) \
336 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
337 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
339 # define BYTEWIDTH 8 /* In bits. */
341 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
345 # define MAX(a, b) ((a) > (b) ? (a) : (b))
346 # define MIN(a, b) ((a) < (b) ? (a) : (b))
348 typedef char boolean;
352 static reg_errcode_t byte_regex_compile (const char *pattern, size_t size,
354 struct re_pattern_buffer *bufp);
356 static int byte_re_match_2_internal (struct re_pattern_buffer *bufp,
357 const char *string1, int size1,
358 const char *string2, int size2,
360 struct re_registers *regs,
362 static int byte_re_search_2 (struct re_pattern_buffer *bufp,
363 const char *string1, int size1,
364 const char *string2, int size2,
365 int startpos, int range,
366 struct re_registers *regs, int stop);
367 static int byte_re_compile_fastmap (struct re_pattern_buffer *bufp);
370 static reg_errcode_t wcs_regex_compile (const char *pattern, size_t size,
372 struct re_pattern_buffer *bufp);
375 static int wcs_re_match_2_internal (struct re_pattern_buffer *bufp,
376 const char *cstring1, int csize1,
377 const char *cstring2, int csize2,
379 struct re_registers *regs,
381 wchar_t *string1, int size1,
382 wchar_t *string2, int size2,
383 int *mbs_offset1, int *mbs_offset2);
384 static int wcs_re_search_2 (struct re_pattern_buffer *bufp,
385 const char *string1, int size1,
386 const char *string2, int size2,
387 int startpos, int range,
388 struct re_registers *regs, int stop);
389 static int wcs_re_compile_fastmap (struct re_pattern_buffer *bufp);
392 /* These are the command codes that appear in compiled regular
393 expressions. Some opcodes are followed by argument bytes. A
394 command code can specify any interpretation whatsoever for its
395 arguments. Zero bytes may appear in the compiled regular expression. */
401 /* Succeed right away--no more backtracking. */
404 /* Followed by one byte giving n, then by n literal bytes. */
408 /* Same as exactn, but contains binary data. */
412 /* Matches any (more or less) character. */
415 /* Matches any one char belonging to specified set. First
416 following byte is number of bitmap bytes. Then come bytes
417 for a bitmap saying which chars are in. Bits in each byte
418 are ordered low-bit-first. A character is in the set if its
419 bit is 1. A character too large to have a bit in the map is
420 automatically not in the set. */
421 /* ifdef MBS_SUPPORT, following element is length of character
422 classes, length of collating symbols, length of equivalence
423 classes, length of character ranges, and length of characters.
424 Next, character class element, collating symbols elements,
425 equivalence class elements, range elements, and character
427 See regex_compile function. */
430 /* Same parameters as charset, but match any character that is
431 not one of those specified. */
434 /* Start remembering the text that is matched, for storing in a
435 register. Followed by one byte with the register number, in
436 the range 0 to one less than the pattern buffer's re_nsub
437 field. Then followed by one byte with the number of groups
438 inner to this one. (This last has to be part of the
439 start_memory only because we need it in the on_failure_jump
443 /* Stop remembering the text that is matched and store it in a
444 memory register. Followed by one byte with the register
445 number, in the range 0 to one less than `re_nsub' in the
446 pattern buffer, and one byte with the number of inner groups,
447 just like `start_memory'. (We need the number of inner
448 groups here because we don't have any easy way of finding the
449 corresponding start_memory when we're at a stop_memory.) */
452 /* Match a duplicate of something remembered. Followed by one
453 byte containing the register number. */
456 /* Fail unless at beginning of line. */
459 /* Fail unless at end of line. */
462 /* Succeeds if at beginning of buffer (if emacs) or at beginning
463 of string to be matched (if not). */
466 /* Analogously, for end of buffer/string. */
469 /* Followed by two byte relative address to which to jump. */
472 /* Same as jump, but marks the end of an alternative. */
475 /* Followed by two-byte relative address of place to resume at
476 in case of failure. */
477 /* ifdef MBS_SUPPORT, the size of address is 1. */
480 /* Like on_failure_jump, but pushes a placeholder instead of the
481 current string position when executed. */
482 on_failure_keep_string_jump,
484 /* Throw away latest failure point and then jump to following
485 two-byte relative address. */
486 /* ifdef MBS_SUPPORT, the size of address is 1. */
489 /* Change to pop_failure_jump if know won't have to backtrack to
490 match; otherwise change to jump. This is used to jump
491 back to the beginning of a repeat. If what follows this jump
492 clearly won't match what the repeat does, such that we can be
493 sure that there is no use backtracking out of repetitions
494 already matched, then we change it to a pop_failure_jump.
495 Followed by two-byte address. */
496 /* ifdef MBS_SUPPORT, the size of address is 1. */
499 /* Jump to following two-byte address, and push a dummy failure
500 point. This failure point will be thrown away if an attempt
501 is made to use it for a failure. A `+' construct makes this
502 before the first repeat. Also used as an intermediary kind
503 of jump when compiling an alternative. */
504 /* ifdef MBS_SUPPORT, the size of address is 1. */
507 /* Push a dummy failure point and continue. Used at the end of
511 /* Followed by two-byte relative address and two-byte number n.
512 After matching N times, jump to the address upon failure. */
513 /* ifdef MBS_SUPPORT, the size of address is 1. */
516 /* Followed by two-byte relative address, and two-byte number n.
517 Jump to the address N times, then fail. */
518 /* ifdef MBS_SUPPORT, the size of address is 1. */
521 /* Set the following two-byte relative address to the
522 subsequent two-byte number. The address *includes* the two
524 /* ifdef MBS_SUPPORT, the size of address is 1. */
527 wordchar, /* Matches any word-constituent character. */
528 notwordchar, /* Matches any char that is not a word-constituent. */
530 wordbeg, /* Succeeds if at word beginning. */
531 wordend, /* Succeeds if at word end. */
533 wordbound, /* Succeeds if at a word boundary. */
534 notwordbound /* Succeeds if not at a word boundary. */
537 ,before_dot, /* Succeeds if before point. */
538 at_dot, /* Succeeds if at point. */
539 after_dot, /* Succeeds if after point. */
541 /* Matches any character whose syntax is specified. Followed by
542 a byte which contains a syntax code, e.g., Sword. */
545 /* Matches any character whose syntax is not that specified. */
549 #endif /* not INSIDE_RECURSION */
554 # define UCHAR_T unsigned char
555 # define COMPILED_BUFFER_VAR bufp->buffer
556 # define OFFSET_ADDRESS_SIZE 2
557 # define PREFIX(name) byte_##name
558 # define ARG_PREFIX(name) name
559 # define PUT_CHAR(c) putchar (c)
562 # define CHAR_T wchar_t
563 # define UCHAR_T wchar_t
564 # define COMPILED_BUFFER_VAR wc_buffer
565 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
566 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
567 # define PREFIX(name) wcs_##name
568 # define ARG_PREFIX(name) c##name
569 /* Should we use wide stream?? */
570 # define PUT_CHAR(c) printf ("%C", c);
576 # define INSIDE_RECURSION
578 # undef INSIDE_RECURSION
581 # define INSIDE_RECURSION
583 # undef INSIDE_RECURSION
586 #include "unlocked-io.h"
588 #ifdef INSIDE_RECURSION
589 /* Common operations on the compiled pattern. */
591 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
592 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
595 # define STORE_NUMBER(destination, number) \
597 *(destination) = (UCHAR_T)(number); \
600 # define STORE_NUMBER(destination, number) \
602 (destination)[0] = (number) & 0377; \
603 (destination)[1] = (number) >> 8; \
607 /* Same as STORE_NUMBER, except increment DESTINATION to
608 the byte after where the number is stored. Therefore, DESTINATION
609 must be an lvalue. */
610 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
612 # define STORE_NUMBER_AND_INCR(destination, number) \
614 STORE_NUMBER (destination, number); \
615 (destination) += OFFSET_ADDRESS_SIZE; \
618 /* Put into DESTINATION a number stored in two contiguous bytes starting
620 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
623 # define EXTRACT_NUMBER(destination, source) \
625 (destination) = *(source); \
628 # define EXTRACT_NUMBER(destination, source) \
630 (destination) = *(source) & 0377; \
631 (destination) += (signed char) (*((source) + 1)) << 8; \
637 PREFIX(extract_number) (int *dest, UCHAR_T *source)
642 signed char temp = source[1];
643 *dest = *source & 0377;
648 # ifndef EXTRACT_MACROS /* To debug the macros. */
649 # undef EXTRACT_NUMBER
650 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
651 # endif /* not EXTRACT_MACROS */
655 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
656 SOURCE must be an lvalue. */
658 # define EXTRACT_NUMBER_AND_INCR(destination, source) \
660 EXTRACT_NUMBER (destination, source); \
661 (source) += OFFSET_ADDRESS_SIZE; \
666 PREFIX(extract_number_and_incr) (int *destination, UCHAR_T **source)
668 PREFIX(extract_number) (destination, *source);
669 *source += OFFSET_ADDRESS_SIZE;
672 # ifndef EXTRACT_MACROS
673 # undef EXTRACT_NUMBER_AND_INCR
674 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
675 PREFIX(extract_number_and_incr) (&dest, &src)
676 # endif /* not EXTRACT_MACROS */
682 /* If DEBUG is defined, Regex prints many voluminous messages about what
683 it is doing (if the variable `debug' is nonzero). If linked with the
684 main program in `iregex.c', you can enter patterns and strings
685 interactively. And if linked with the main program in `main.c' and
686 the other test files, you can run the already-written tests. */
690 # ifndef DEFINED_ONCE
692 /* We use standard I/O for debugging. */
695 /* It is useful to test things that ``must'' be true when debugging. */
700 # define DEBUG_STATEMENT(e) e
701 # define DEBUG_PRINT1(x) if (debug) printf (x)
702 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
703 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
704 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
705 # endif /* not DEFINED_ONCE */
707 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
708 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
709 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
710 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
713 /* Print the fastmap in human-readable form. */
715 # ifndef DEFINED_ONCE
717 print_fastmap (char *fastmap)
719 unsigned was_a_range = 0;
722 while (i < (1 << BYTEWIDTH))
728 while (i < (1 << BYTEWIDTH) && fastmap[i])
742 # endif /* not DEFINED_ONCE */
745 /* Print a compiled pattern string in human-readable form, starting at
746 the START pointer into it and ending just before the pointer END. */
749 PREFIX(print_partial_compiled_pattern) (UCHAR_T *start, UCHAR_T *end)
762 /* Loop over pattern commands. */
766 printf ("%td:\t", p - start);
768 printf ("%ld:\t", (long int) (p - start));
771 switch ((re_opcode_t) *p++)
779 printf ("/exactn/%d", mcnt);
791 printf ("/exactn_bin/%d", mcnt);
794 printf("/%lx", (long int) *p++);
798 # endif /* MBS_SUPPORT */
802 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++);
807 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++);
811 printf ("/duplicate/%ld", (long int) *p++);
824 printf ("/charset [%s",
825 (re_opcode_t) *(workp - 1) == charset_not ? "^" : "");
827 length = *workp++; /* the length of char_classes */
828 for (i=0 ; i<length ; i++)
829 printf("[:%lx:]", (long int) *p++);
830 length = *workp++; /* the length of collating_symbol */
831 for (i=0 ; i<length ;)
835 PUT_CHAR((i++,*p++));
839 length = *workp++; /* the length of equivalence_class */
840 for (i=0 ; i<length ;)
844 PUT_CHAR((i++,*p++));
848 length = *workp++; /* the length of char_range */
849 for (i=0 ; i<length ; i++)
851 wchar_t range_start = *p++;
852 wchar_t range_end = *p++;
853 printf("%C-%C", range_start, range_end);
855 length = *workp++; /* the length of char */
856 for (i=0 ; i<length ; i++)
860 register int c, last = -100;
861 register int in_range = 0;
863 printf ("/charset [%s",
864 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
866 assert (p + *p < pend);
868 for (c = 0; c < 256; c++)
870 && (p[1 + (c/8)] & (1 << (c % 8))))
872 /* Are we starting a range? */
873 if (last + 1 == c && ! in_range)
878 /* Have we broken a range? */
879 else if (last + 1 != c && in_range)
909 case on_failure_jump:
910 PREFIX(extract_number_and_incr) (&mcnt, &p);
912 printf ("/on_failure_jump to %td", p + mcnt - start);
914 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
918 case on_failure_keep_string_jump:
919 PREFIX(extract_number_and_incr) (&mcnt, &p);
921 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start);
923 printf ("/on_failure_keep_string_jump to %ld",
924 (long int) (p + mcnt - start));
928 case dummy_failure_jump:
929 PREFIX(extract_number_and_incr) (&mcnt, &p);
931 printf ("/dummy_failure_jump to %td", p + mcnt - start);
933 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
937 case push_dummy_failure:
938 printf ("/push_dummy_failure");
942 PREFIX(extract_number_and_incr) (&mcnt, &p);
944 printf ("/maybe_pop_jump to %td", p + mcnt - start);
946 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
950 case pop_failure_jump:
951 PREFIX(extract_number_and_incr) (&mcnt, &p);
953 printf ("/pop_failure_jump to %td", p + mcnt - start);
955 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
960 PREFIX(extract_number_and_incr) (&mcnt, &p);
962 printf ("/jump_past_alt to %td", p + mcnt - start);
964 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
969 PREFIX(extract_number_and_incr) (&mcnt, &p);
971 printf ("/jump to %td", p + mcnt - start);
973 printf ("/jump to %ld", (long int) (p + mcnt - start));
978 PREFIX(extract_number_and_incr) (&mcnt, &p);
980 PREFIX(extract_number_and_incr) (&mcnt2, &p);
982 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2);
984 printf ("/succeed_n to %ld, %d times",
985 (long int) (p1 - start), mcnt2);
990 PREFIX(extract_number_and_incr) (&mcnt, &p);
992 PREFIX(extract_number_and_incr) (&mcnt2, &p);
993 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
997 PREFIX(extract_number_and_incr) (&mcnt, &p);
999 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1001 printf ("/set_number_at location %td to %d", p1 - start, mcnt2);
1003 printf ("/set_number_at location %ld to %d",
1004 (long int) (p1 - start), mcnt2);
1009 printf ("/wordbound");
1013 printf ("/notwordbound");
1017 printf ("/wordbeg");
1021 printf ("/wordend");
1026 printf ("/before_dot");
1034 printf ("/after_dot");
1038 printf ("/syntaxspec");
1040 printf ("/%d", mcnt);
1044 printf ("/notsyntaxspec");
1046 printf ("/%d", mcnt);
1051 printf ("/wordchar");
1055 printf ("/notwordchar");
1067 printf ("?%ld", (long int) *(p-1));
1074 printf ("%td:\tend of pattern.\n", p - start);
1076 printf ("%ld:\tend of pattern.\n", (long int) (p - start));
1082 PREFIX(print_compiled_pattern) (struct re_pattern_buffer *bufp)
1084 UCHAR_T *buffer = (UCHAR_T*) bufp->buffer;
1086 PREFIX(print_partial_compiled_pattern) (buffer, buffer
1087 + bufp->used / sizeof(UCHAR_T));
1088 printf ("%ld bytes used/%ld bytes allocated.\n",
1089 bufp->used, bufp->allocated);
1091 if (bufp->fastmap_accurate && bufp->fastmap)
1093 printf ("fastmap: ");
1094 print_fastmap (bufp->fastmap);
1098 printf ("re_nsub: %Zd\t", bufp->re_nsub);
1100 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
1102 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1103 printf ("can_be_null: %d\t", bufp->can_be_null);
1104 printf ("newline_anchor: %d\n", bufp->newline_anchor);
1105 printf ("no_sub: %d\t", bufp->no_sub);
1106 printf ("not_bol: %d\t", bufp->not_bol);
1107 printf ("not_eol: %d\t", bufp->not_eol);
1108 printf ("syntax: %lx\n", bufp->syntax);
1109 /* Perhaps we should print the translate table? */
1114 PREFIX(print_double_string) (const CHAR_T *where,
1115 const CHAR_T *string1,
1116 const CHAR_T *string2,
1128 if (FIRST_STRING_P (where))
1130 for (this_char = where - string1; this_char < size1; this_char++)
1131 PUT_CHAR (string1[this_char]);
1137 for (this_char = where - string2; this_char < size2; this_char++)
1139 PUT_CHAR (string2[this_char]);
1142 fputs ("...", stdout);
1149 # ifndef DEFINED_ONCE
1158 # else /* not DEBUG */
1160 # ifndef DEFINED_ONCE
1164 # define DEBUG_STATEMENT(e)
1165 # define DEBUG_PRINT1(x)
1166 # define DEBUG_PRINT2(x1, x2)
1167 # define DEBUG_PRINT3(x1, x2, x3)
1168 # define DEBUG_PRINT4(x1, x2, x3, x4)
1169 # endif /* not DEFINED_ONCE */
1170 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1171 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1173 # endif /* not DEBUG */
1178 /* This convert a multibyte string to a wide character string.
1179 And write their correspondances to offset_buffer(see below)
1180 and write whether each wchar_t is binary data to is_binary.
1181 This assume invalid multibyte sequences as binary data.
1182 We assume offset_buffer and is_binary is already allocated
1186 convert_mbs_to_wcs (CHAR_T *dest,
1187 const unsigned char* src,
1189 /* The length of multibyte string. */
1192 /* Correspondences between src(char string) and
1193 dest(wchar_t string) for optimization. E.g.:
1195 dest = {'X', 'Y', 'Z'}
1196 (each "xxx", "y" and "zz" represent one
1197 multibyte character corresponding to 'X',
1199 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"),
1206 wchar_t *pdest = dest;
1207 const unsigned char *psrc = src;
1208 size_t wc_count = 0;
1212 size_t mb_remain = len;
1213 size_t mb_count = 0;
1215 /* Initialize the conversion state. */
1216 memset (&mbs, 0, sizeof (mbstate_t));
1218 offset_buffer[0] = 0;
1219 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed,
1222 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs);
1225 /* failed to convert. maybe src contains binary data.
1226 So we consume 1 byte manualy. */
1230 is_binary[wc_count] = TRUE;
1233 is_binary[wc_count] = FALSE;
1234 /* In sjis encoding, we use yen sign as escape character in
1235 place of reverse solidus. So we convert 0x5c(yen sign in
1236 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1237 solidus in UCS2). */
1238 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5)
1239 *pdest = (wchar_t) *psrc;
1241 offset_buffer[wc_count + 1] = mb_count += consumed;
1244 /* Fill remain of the buffer with sentinel. */
1245 for (i = wc_count + 1 ; i <= len ; i++)
1246 offset_buffer[i] = mb_count + 1;
1253 #else /* not INSIDE_RECURSION */
1255 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1256 also be assigned to arbitrarily: each pattern buffer stores its own
1257 syntax, so it can be changed between regex compilations. */
1258 /* This has no initializer because initialized variables in Emacs
1259 become read-only after dumping. */
1260 reg_syntax_t re_syntax_options;
1263 /* Specify the precise syntax of regexps for compilation. This provides
1264 for compatibility for various utilities which historically have
1265 different, incompatible syntaxes.
1267 The argument SYNTAX is a bit mask comprised of the various bits
1268 defined in regex.h. We return the old syntax. */
1271 re_set_syntax (reg_syntax_t syntax)
1273 reg_syntax_t ret = re_syntax_options;
1275 re_syntax_options = syntax;
1277 if (syntax & RE_DEBUG)
1279 else if (debug) /* was on but now is not */
1285 weak_alias (__re_set_syntax, re_set_syntax)
1288 /* This table gives an error message for each of the error codes listed
1289 in regex.h. Obviously the order here has to be same as there.
1290 POSIX doesn't require that we do anything for REG_NOERROR,
1291 but why not be nice? */
1293 static const char re_error_msgid[] =
1295 # define REG_NOERROR_IDX 0
1296 gettext_noop ("Success") /* REG_NOERROR */
1298 # define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1299 gettext_noop ("No match") /* REG_NOMATCH */
1301 # define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1302 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1304 # define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1305 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1307 # define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1308 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1310 # define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1311 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1313 # define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1314 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1316 # define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1317 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1319 # define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1320 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1322 # define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1323 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1325 # define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1326 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1328 # define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1329 gettext_noop ("Invalid range end") /* REG_ERANGE */
1331 # define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1332 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1334 # define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1335 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1337 # define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1338 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1340 # define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1341 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1343 # define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1344 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1347 static const size_t re_error_msgid_idx[] =
1368 #endif /* INSIDE_RECURSION */
1370 #ifndef DEFINED_ONCE
1371 /* Avoiding alloca during matching, to placate r_alloc. */
1373 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1374 searching and matching functions should not call alloca. On some
1375 systems, alloca is implemented in terms of malloc, and if we're
1376 using the relocating allocator routines, then malloc could cause a
1377 relocation, which might (if the strings being searched are in the
1378 ralloc heap) shift the data out from underneath the regexp
1381 Here's another reason to avoid allocation: Emacs
1382 processes input from X in a signal handler; processing X input may
1383 call malloc; if input arrives while a matching routine is calling
1384 malloc, then we're scrod. But Emacs can't just block input while
1385 calling matching routines; then we don't notice interrupts when
1386 they come in. So, Emacs blocks input around all regexp calls
1387 except the matching calls, which it leaves unprotected, in the
1388 faith that they will not malloc. */
1390 /* Normally, this is fine. */
1391 # define MATCH_MAY_ALLOCATE
1393 /* When using GNU C, we are not REALLY using the C alloca, no matter
1394 what config.h may say. So don't take precautions for it. */
1399 /* The match routines may not allocate if (1) they would do it with malloc
1400 and (2) it's not safe for them to use malloc.
1401 Note that if REL_ALLOC is defined, matching would not use malloc for the
1402 failure stack, but we would still use it for the register vectors;
1403 so REL_ALLOC should not affect this. */
1404 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1405 # undef MATCH_MAY_ALLOCATE
1407 #endif /* not DEFINED_ONCE */
1409 #ifdef INSIDE_RECURSION
1410 /* Failure stack declarations and macros; both re_compile_fastmap and
1411 re_match_2 use a failure stack. These have to be macros because of
1412 REGEX_ALLOCATE_STACK. */
1415 /* Number of failure points for which to initially allocate space
1416 when matching. If this number is exceeded, we allocate more
1417 space, so it is not a hard limit. */
1418 # ifndef INIT_FAILURE_ALLOC
1419 # define INIT_FAILURE_ALLOC 5
1422 /* Roughly the maximum number of failure points on the stack. Would be
1423 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1424 This is a variable only so users of regex can assign to it; we never
1425 change it ourselves. */
1427 # ifdef INT_IS_16BIT
1429 # ifndef DEFINED_ONCE
1430 # if defined MATCH_MAY_ALLOCATE
1431 /* 4400 was enough to cause a crash on Alpha OSF/1,
1432 whose default stack limit is 2mb. */
1433 long int re_max_failures = 4000;
1435 long int re_max_failures = 2000;
1439 union PREFIX(fail_stack_elt)
1445 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1449 PREFIX(fail_stack_elt_t) *stack;
1450 unsigned long int size;
1451 unsigned long int avail; /* Offset of next open position. */
1452 } PREFIX(fail_stack_type);
1454 # else /* not INT_IS_16BIT */
1456 # ifndef DEFINED_ONCE
1457 # if defined MATCH_MAY_ALLOCATE
1458 /* 4400 was enough to cause a crash on Alpha OSF/1,
1459 whose default stack limit is 2mb. */
1460 int re_max_failures = 4000;
1462 int re_max_failures = 2000;
1466 union PREFIX(fail_stack_elt)
1472 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1476 PREFIX(fail_stack_elt_t) *stack;
1478 unsigned avail; /* Offset of next open position. */
1479 } PREFIX(fail_stack_type);
1481 # endif /* INT_IS_16BIT */
1483 # ifndef DEFINED_ONCE
1484 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1485 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1486 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1490 /* Define macros to initialize and free the failure stack.
1491 Do `return -2' if the alloc fails. */
1493 # ifdef MATCH_MAY_ALLOCATE
1494 # define INIT_FAIL_STACK() \
1496 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1497 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1499 if (fail_stack.stack == NULL) \
1502 fail_stack.size = INIT_FAILURE_ALLOC; \
1503 fail_stack.avail = 0; \
1506 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1508 # define INIT_FAIL_STACK() \
1510 fail_stack.avail = 0; \
1513 # define RESET_FAIL_STACK()
1517 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1519 Return 1 if succeeds, and 0 if either ran out of memory
1520 allocating space for it or it was already too large.
1522 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1524 # define DOUBLE_FAIL_STACK(fail_stack) \
1525 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1527 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1528 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1529 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1530 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1532 (fail_stack).stack == NULL \
1534 : ((fail_stack).size <<= 1, \
1538 /* Push pointer POINTER on FAIL_STACK.
1539 Return 1 if was able to do so and 0 if ran out of memory allocating
1541 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1542 ((FAIL_STACK_FULL () \
1543 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1545 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1548 /* Push a pointer value onto the failure stack.
1549 Assumes the variable `fail_stack'. Probably should only
1550 be called from within `PUSH_FAILURE_POINT'. */
1551 # define PUSH_FAILURE_POINTER(item) \
1552 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1554 /* This pushes an integer-valued item onto the failure stack.
1555 Assumes the variable `fail_stack'. Probably should only
1556 be called from within `PUSH_FAILURE_POINT'. */
1557 # define PUSH_FAILURE_INT(item) \
1558 fail_stack.stack[fail_stack.avail++].integer = (item)
1560 /* Push a fail_stack_elt_t value onto the failure stack.
1561 Assumes the variable `fail_stack'. Probably should only
1562 be called from within `PUSH_FAILURE_POINT'. */
1563 # define PUSH_FAILURE_ELT(item) \
1564 fail_stack.stack[fail_stack.avail++] = (item)
1566 /* These three POP... operations complement the three PUSH... operations.
1567 All assume that `fail_stack' is nonempty. */
1568 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1569 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1570 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1572 /* Used to omit pushing failure point id's when we're not debugging. */
1574 # define DEBUG_PUSH PUSH_FAILURE_INT
1575 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1577 # define DEBUG_PUSH(item)
1578 # define DEBUG_POP(item_addr)
1582 /* Push the information about the state we will need
1583 if we ever fail back to it.
1585 Requires variables fail_stack, regstart, regend, reg_info, and
1586 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1589 Does `return FAILURE_CODE' if runs out of memory. */
1591 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1593 char *destination; \
1594 /* Must be int, so when we don't save any registers, the arithmetic \
1595 of 0 + -1 isn't done as unsigned. */ \
1596 /* Can't be int, since there is not a shred of a guarantee that int \
1597 is wide enough to hold a value of something to which pointer can \
1599 active_reg_t this_reg; \
1601 DEBUG_STATEMENT (failure_id++); \
1602 DEBUG_STATEMENT (nfailure_points_pushed++); \
1603 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1604 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1605 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1607 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1608 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1610 /* Ensure we have enough space allocated for what we will push. */ \
1611 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1613 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1614 return failure_code; \
1616 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1617 (fail_stack).size); \
1618 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1621 /* Push the info, starting with the registers. */ \
1622 DEBUG_PRINT1 ("\n"); \
1625 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1628 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1629 DEBUG_STATEMENT (num_regs_pushed++); \
1631 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1632 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1634 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1635 PUSH_FAILURE_POINTER (regend[this_reg]); \
1637 DEBUG_PRINT2 (" info: %p\n ", \
1638 reg_info[this_reg].word.pointer); \
1639 DEBUG_PRINT2 (" match_null=%d", \
1640 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1641 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1642 DEBUG_PRINT2 (" matched_something=%d", \
1643 MATCHED_SOMETHING (reg_info[this_reg])); \
1644 DEBUG_PRINT2 (" ever_matched=%d", \
1645 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1646 DEBUG_PRINT1 ("\n"); \
1647 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1650 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1651 PUSH_FAILURE_INT (lowest_active_reg); \
1653 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1654 PUSH_FAILURE_INT (highest_active_reg); \
1656 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1657 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1658 PUSH_FAILURE_POINTER (pattern_place); \
1660 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1661 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1663 DEBUG_PRINT1 ("'\n"); \
1664 PUSH_FAILURE_POINTER (string_place); \
1666 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1667 DEBUG_PUSH (failure_id); \
1670 # ifndef DEFINED_ONCE
1671 /* This is the number of items that are pushed and popped on the stack
1672 for each register. */
1673 # define NUM_REG_ITEMS 3
1675 /* Individual items aside from the registers. */
1677 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1679 # define NUM_NONREG_ITEMS 4
1682 /* We push at most this many items on the stack. */
1683 /* We used to use (num_regs - 1), which is the number of registers
1684 this regexp will save; but that was changed to 5
1685 to avoid stack overflow for a regexp with lots of parens. */
1686 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1688 /* We actually push this many items. */
1689 # define NUM_FAILURE_ITEMS \
1691 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1695 /* How many items can still be added to the stack without overflowing it. */
1696 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1697 # endif /* not DEFINED_ONCE */
1700 /* Pops what PUSH_FAIL_STACK pushes.
1702 We restore into the parameters, all of which should be lvalues:
1703 STR -- the saved data position.
1704 PAT -- the saved pattern position.
1705 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1706 REGSTART, REGEND -- arrays of string positions.
1707 REG_INFO -- array of information about each subexpression.
1709 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1710 `pend', `string1', `size1', `string2', and `size2'. */
1711 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1713 DEBUG_STATEMENT (unsigned failure_id;) \
1714 active_reg_t this_reg; \
1715 const UCHAR_T *string_temp; \
1717 assert (!FAIL_STACK_EMPTY ()); \
1719 /* Remove failure points and point to how many regs pushed. */ \
1720 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1721 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1722 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1724 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1726 DEBUG_POP (&failure_id); \
1727 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1729 /* If the saved string location is NULL, it came from an \
1730 on_failure_keep_string_jump opcode, and we want to throw away the \
1731 saved NULL, thus retaining our current position in the string. */ \
1732 string_temp = POP_FAILURE_POINTER (); \
1733 if (string_temp != NULL) \
1734 str = (const CHAR_T *) string_temp; \
1736 DEBUG_PRINT2 (" Popping string %p: `", str); \
1737 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1738 DEBUG_PRINT1 ("'\n"); \
1740 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1741 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1742 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1744 /* Restore register info. */ \
1745 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1746 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1748 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1749 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1752 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1754 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1756 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1757 DEBUG_PRINT2 (" info: %p\n", \
1758 reg_info[this_reg].word.pointer); \
1760 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1761 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1763 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1764 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1768 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1770 reg_info[this_reg].word.integer = 0; \
1771 regend[this_reg] = 0; \
1772 regstart[this_reg] = 0; \
1774 highest_active_reg = high_reg; \
1777 set_regs_matched_done = 0; \
1778 DEBUG_STATEMENT (nfailure_points_popped++); \
1779 } /* POP_FAILURE_POINT */
1781 /* Structure for per-register (a.k.a. per-group) information.
1782 Other register information, such as the
1783 starting and ending positions (which are addresses), and the list of
1784 inner groups (which is a bits list) are maintained in separate
1787 We are making a (strictly speaking) nonportable assumption here: that
1788 the compiler will pack our bit fields into something that fits into
1789 the type of `word', i.e., is something that fits into one item on the
1793 /* Declarations and macros for re_match_2. */
1797 PREFIX(fail_stack_elt_t) word;
1800 /* This field is one if this group can match the empty string,
1801 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1802 # define MATCH_NULL_UNSET_VALUE 3
1803 unsigned match_null_string_p : 2;
1804 unsigned is_active : 1;
1805 unsigned matched_something : 1;
1806 unsigned ever_matched_something : 1;
1808 } PREFIX(register_info_type);
1810 # ifndef DEFINED_ONCE
1811 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1812 # define IS_ACTIVE(R) ((R).bits.is_active)
1813 # define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1814 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1817 /* Call this when have matched a real character; it sets `matched' flags
1818 for the subexpressions which we are currently inside. Also records
1819 that those subexprs have matched. */
1820 # define SET_REGS_MATCHED() \
1823 if (!set_regs_matched_done) \
1826 set_regs_matched_done = 1; \
1827 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1829 MATCHED_SOMETHING (reg_info[r]) \
1830 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1836 # endif /* not DEFINED_ONCE */
1838 /* Registers are set to a sentinel when they haven't yet matched. */
1839 static CHAR_T PREFIX(reg_unset_dummy);
1840 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1841 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1843 /* Subroutine declarations and macros for regex_compile. */
1844 static void PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg);
1845 static void PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc,
1846 int arg1, int arg2);
1847 static void PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc,
1848 int arg, UCHAR_T *end);
1849 static void PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc,
1850 int arg1, int arg2, UCHAR_T *end);
1851 static boolean PREFIX(at_begline_loc_p) (const CHAR_T *pattern,
1853 reg_syntax_t syntax);
1854 static boolean PREFIX(at_endline_loc_p) (const CHAR_T *p,
1856 reg_syntax_t syntax);
1858 static reg_errcode_t wcs_compile_range (CHAR_T range_start,
1859 const CHAR_T **p_ptr,
1862 reg_syntax_t syntax,
1865 static void insert_space (int num, CHAR_T *loc, CHAR_T *end);
1867 static reg_errcode_t byte_compile_range (unsigned int range_start,
1871 reg_syntax_t syntax,
1875 /* Fetch the next character in the uncompiled pattern---translating it
1876 if necessary. Also cast from a signed character in the constant
1877 string passed to us by the user to an unsigned char that we can use
1878 as an array index (in, e.g., `translate'). */
1879 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1880 because it is impossible to allocate 4GB array for some encodings
1881 which have 4 byte character_set like UCS4. */
1884 # define PATFETCH(c) \
1885 do {if (p == pend) return REG_EEND; \
1886 c = (UCHAR_T) *p++; \
1887 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1890 # define PATFETCH(c) \
1891 do {if (p == pend) return REG_EEND; \
1892 c = (unsigned char) *p++; \
1893 if (translate) c = (unsigned char) translate[c]; \
1898 /* Fetch the next character in the uncompiled pattern, with no
1900 # define PATFETCH_RAW(c) \
1901 do {if (p == pend) return REG_EEND; \
1902 c = (UCHAR_T) *p++; \
1905 /* Go backwards one character in the pattern. */
1906 # define PATUNFETCH p--
1909 /* If `translate' is non-null, return translate[D], else just D. We
1910 cast the subscript to translate because some data is declared as
1911 `char *', to avoid warnings when a string constant is passed. But
1912 when we use a character as a subscript we must make it unsigned. */
1913 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1914 because it is impossible to allocate 4GB array for some encodings
1915 which have 4 byte character_set like UCS4. */
1919 # define TRANSLATE(d) \
1920 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1921 ? (char) translate[(unsigned char) (d)] : (d))
1923 # define TRANSLATE(d) \
1924 (translate ? (char) translate[(unsigned char) (d)] : (d))
1929 /* Macros for outputting the compiled pattern into `buffer'. */
1931 /* If the buffer isn't allocated when it comes in, use this. */
1932 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
1934 /* Make sure we have at least N more bytes of space in buffer. */
1936 # define GET_BUFFER_SPACE(n) \
1937 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
1938 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
1941 # define GET_BUFFER_SPACE(n) \
1942 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1946 /* Make sure we have one more byte of buffer space and then add C to it. */
1947 # define BUF_PUSH(c) \
1949 GET_BUFFER_SPACE (1); \
1950 *b++ = (UCHAR_T) (c); \
1954 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1955 # define BUF_PUSH_2(c1, c2) \
1957 GET_BUFFER_SPACE (2); \
1958 *b++ = (UCHAR_T) (c1); \
1959 *b++ = (UCHAR_T) (c2); \
1963 /* As with BUF_PUSH_2, except for three bytes. */
1964 # define BUF_PUSH_3(c1, c2, c3) \
1966 GET_BUFFER_SPACE (3); \
1967 *b++ = (UCHAR_T) (c1); \
1968 *b++ = (UCHAR_T) (c2); \
1969 *b++ = (UCHAR_T) (c3); \
1972 /* Store a jump with opcode OP at LOC to location TO. We store a
1973 relative address offset by the three bytes the jump itself occupies. */
1974 # define STORE_JUMP(op, loc, to) \
1975 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
1977 /* Likewise, for a two-argument jump. */
1978 # define STORE_JUMP2(op, loc, to, arg) \
1979 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
1981 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1982 # define INSERT_JUMP(op, loc, to) \
1983 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
1985 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1986 # define INSERT_JUMP2(op, loc, to, arg) \
1987 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
1990 /* This is not an arbitrary limit: the arguments which represent offsets
1991 into the pattern are two bytes long. So if 2^16 bytes turns out to
1992 be too small, many things would have to change. */
1993 /* Any other compiler which, like MSC, has allocation limit below 2^16
1994 bytes will have to use approach similar to what was done below for
1995 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1996 reallocating to 0 bytes. Such thing is not going to work too well.
1997 You have been warned!! */
1998 # ifndef DEFINED_ONCE
1999 # if defined _MSC_VER && !defined WIN32
2000 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2001 The REALLOC define eliminates a flurry of conversion warnings,
2002 but is not required. */
2003 # define MAX_BUF_SIZE 65500L
2004 # define REALLOC(p,s) realloc ((p), (size_t) (s))
2006 # define MAX_BUF_SIZE (1L << 16)
2007 # define REALLOC(p,s) realloc ((p), (s))
2010 /* Extend the buffer by twice its current size via realloc and
2011 reset the pointers that pointed into the old block to point to the
2012 correct places in the new one. If extending the buffer results in it
2013 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2014 # if __BOUNDED_POINTERS__
2015 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2016 # define MOVE_BUFFER_POINTER(P) \
2017 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2018 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2021 SET_HIGH_BOUND (b); \
2022 SET_HIGH_BOUND (begalt); \
2023 if (fixup_alt_jump) \
2024 SET_HIGH_BOUND (fixup_alt_jump); \
2026 SET_HIGH_BOUND (laststart); \
2027 if (pending_exact) \
2028 SET_HIGH_BOUND (pending_exact); \
2031 # define MOVE_BUFFER_POINTER(P) (P) += incr
2032 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
2034 # endif /* not DEFINED_ONCE */
2037 # define EXTEND_BUFFER() \
2039 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2041 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2043 bufp->allocated <<= 1; \
2044 if (bufp->allocated > MAX_BUF_SIZE) \
2045 bufp->allocated = MAX_BUF_SIZE; \
2046 /* How many characters the new buffer can have? */ \
2047 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2048 if (wchar_count == 0) wchar_count = 1; \
2049 /* Truncate the buffer to CHAR_T align. */ \
2050 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2051 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2052 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2053 if (COMPILED_BUFFER_VAR == NULL) \
2054 return REG_ESPACE; \
2055 /* If the buffer moved, move all the pointers into it. */ \
2056 if (old_buffer != COMPILED_BUFFER_VAR) \
2058 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2059 MOVE_BUFFER_POINTER (b); \
2060 MOVE_BUFFER_POINTER (begalt); \
2061 if (fixup_alt_jump) \
2062 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2064 MOVE_BUFFER_POINTER (laststart); \
2065 if (pending_exact) \
2066 MOVE_BUFFER_POINTER (pending_exact); \
2068 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2071 # define EXTEND_BUFFER() \
2073 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2074 if (bufp->allocated == MAX_BUF_SIZE) \
2076 bufp->allocated <<= 1; \
2077 if (bufp->allocated > MAX_BUF_SIZE) \
2078 bufp->allocated = MAX_BUF_SIZE; \
2079 bufp->buffer = REALLOC (COMPILED_BUFFER_VAR, bufp->allocated); \
2080 if (COMPILED_BUFFER_VAR == NULL) \
2081 return REG_ESPACE; \
2082 /* If the buffer moved, move all the pointers into it. */ \
2083 if (old_buffer != COMPILED_BUFFER_VAR) \
2085 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2086 MOVE_BUFFER_POINTER (b); \
2087 MOVE_BUFFER_POINTER (begalt); \
2088 if (fixup_alt_jump) \
2089 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2091 MOVE_BUFFER_POINTER (laststart); \
2092 if (pending_exact) \
2093 MOVE_BUFFER_POINTER (pending_exact); \
2095 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2099 # ifndef DEFINED_ONCE
2100 /* Since we have one byte reserved for the register number argument to
2101 {start,stop}_memory, the maximum number of groups we can report
2102 things about is what fits in that byte. */
2103 # define MAX_REGNUM 255
2105 /* But patterns can have more than `MAX_REGNUM' registers. We just
2106 ignore the excess. */
2107 typedef unsigned regnum_t;
2110 /* Macros for the compile stack. */
2112 /* Since offsets can go either forwards or backwards, this type needs to
2113 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2114 /* int may be not enough when sizeof(int) == 2. */
2115 typedef long pattern_offset_t;
2119 pattern_offset_t begalt_offset;
2120 pattern_offset_t fixup_alt_jump;
2121 pattern_offset_t inner_group_offset;
2122 pattern_offset_t laststart_offset;
2124 } compile_stack_elt_t;
2129 compile_stack_elt_t *stack;
2131 unsigned avail; /* Offset of next open position. */
2132 } compile_stack_type;
2135 # define INIT_COMPILE_STACK_SIZE 32
2137 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2138 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2140 /* The next available element. */
2141 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2143 # endif /* not DEFINED_ONCE */
2145 /* Set the bit for character C in a list. */
2146 # ifndef DEFINED_ONCE
2147 # define SET_LIST_BIT(c) \
2148 (b[((unsigned char) (c)) / BYTEWIDTH] \
2149 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2150 # endif /* DEFINED_ONCE */
2152 /* Get the next unsigned number in the uncompiled pattern. */
2153 # define GET_UNSIGNED_NUMBER(num) \
2158 if (c < '0' || c > '9') \
2160 if (num <= RE_DUP_MAX) \
2164 num = num * 10 + c - '0'; \
2169 # ifndef DEFINED_ONCE
2170 # if defined _LIBC || WIDE_CHAR_SUPPORT
2171 /* The GNU C library provides support for user-defined character classes
2172 and the functions from ISO C amendement 1. */
2173 # ifdef CHARCLASS_NAME_MAX
2174 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2176 /* This shouldn't happen but some implementation might still have this
2177 problem. Use a reasonable default value. */
2178 # define CHAR_CLASS_MAX_LENGTH 256
2182 # define IS_CHAR_CLASS(string) __wctype (string)
2184 # define IS_CHAR_CLASS(string) wctype (string)
2187 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2189 # define IS_CHAR_CLASS(string) \
2190 (STREQ (string, "alpha") || STREQ (string, "upper") \
2191 || STREQ (string, "lower") || STREQ (string, "digit") \
2192 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2193 || STREQ (string, "space") || STREQ (string, "print") \
2194 || STREQ (string, "punct") || STREQ (string, "graph") \
2195 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2197 # endif /* DEFINED_ONCE */
2199 # ifndef MATCH_MAY_ALLOCATE
2201 /* If we cannot allocate large objects within re_match_2_internal,
2202 we make the fail stack and register vectors global.
2203 The fail stack, we grow to the maximum size when a regexp
2205 The register vectors, we adjust in size each time we
2206 compile a regexp, according to the number of registers it needs. */
2208 static PREFIX(fail_stack_type) fail_stack;
2210 /* Size with which the following vectors are currently allocated.
2211 That is so we can make them bigger as needed,
2212 but never make them smaller. */
2213 # ifdef DEFINED_ONCE
2214 static int regs_allocated_size;
2216 static const char ** regstart, ** regend;
2217 static const char ** old_regstart, ** old_regend;
2218 static const char **best_regstart, **best_regend;
2219 static const char **reg_dummy;
2220 # endif /* DEFINED_ONCE */
2222 static PREFIX(register_info_type) *PREFIX(reg_info);
2223 static PREFIX(register_info_type) *PREFIX(reg_info_dummy);
2225 /* Make the register vectors big enough for NUM_REGS registers,
2226 but don't make them smaller. */
2229 PREFIX(regex_grow_registers) (int num_regs)
2231 if (num_regs > regs_allocated_size)
2233 RETALLOC_IF (regstart, num_regs, const char *);
2234 RETALLOC_IF (regend, num_regs, const char *);
2235 RETALLOC_IF (old_regstart, num_regs, const char *);
2236 RETALLOC_IF (old_regend, num_regs, const char *);
2237 RETALLOC_IF (best_regstart, num_regs, const char *);
2238 RETALLOC_IF (best_regend, num_regs, const char *);
2239 RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type));
2240 RETALLOC_IF (reg_dummy, num_regs, const char *);
2241 RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type));
2243 regs_allocated_size = num_regs;
2247 # endif /* not MATCH_MAY_ALLOCATE */
2249 # ifndef DEFINED_ONCE
2250 static boolean group_in_compile_stack (compile_stack_type
2253 # endif /* not DEFINED_ONCE */
2255 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2256 Returns one of error codes defined in `regex.h', or zero for success.
2258 Assumes the `allocated' (and perhaps `buffer') and `translate'
2259 fields are set in BUFP on entry.
2261 If it succeeds, results are put in BUFP (if it returns an error, the
2262 contents of BUFP are undefined):
2263 `buffer' is the compiled pattern;
2264 `syntax' is set to SYNTAX;
2265 `used' is set to the length of the compiled pattern;
2266 `fastmap_accurate' is zero;
2267 `re_nsub' is the number of subexpressions in PATTERN;
2268 `not_bol' and `not_eol' are zero;
2270 The `fastmap' and `newline_anchor' fields are neither
2271 examined nor set. */
2273 /* Return, freeing storage we allocated. */
2275 # define FREE_STACK_RETURN(value) \
2276 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2278 # define FREE_STACK_RETURN(value) \
2279 return (free (compile_stack.stack), value)
2282 static reg_errcode_t
2283 PREFIX(regex_compile) (const char *ARG_PREFIX(pattern),
2284 size_t ARG_PREFIX(size),
2285 reg_syntax_t syntax,
2286 struct re_pattern_buffer *bufp)
2288 /* We fetch characters from PATTERN here. Even though PATTERN is
2289 `char *' (i.e., signed), we declare these variables as unsigned, so
2290 they can be reliably used as array indices. */
2291 register UCHAR_T c, c1;
2294 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2295 CHAR_T *pattern, *COMPILED_BUFFER_VAR;
2297 /* offset buffer for optimization. See convert_mbs_to_wc. */
2298 int *mbs_offset = NULL;
2299 /* It hold whether each wchar_t is binary data or not. */
2300 char *is_binary = NULL;
2301 /* A flag whether exactn is handling binary data or not. */
2302 char is_exactn_bin = FALSE;
2305 /* A random temporary spot in PATTERN. */
2308 /* Points to the end of the buffer, where we should append. */
2309 register UCHAR_T *b;
2311 /* Keeps track of unclosed groups. */
2312 compile_stack_type compile_stack;
2314 /* Points to the current (ending) position in the pattern. */
2319 const CHAR_T *p = pattern;
2320 const CHAR_T *pend = pattern + size;
2323 /* How to translate the characters in the pattern. */
2324 RE_TRANSLATE_TYPE translate = bufp->translate;
2326 /* Address of the count-byte of the most recently inserted `exactn'
2327 command. This makes it possible to tell if a new exact-match
2328 character can be added to that command or if the character requires
2329 a new `exactn' command. */
2330 UCHAR_T *pending_exact = 0;
2332 /* Address of start of the most recently finished expression.
2333 This tells, e.g., postfix * where to find the start of its
2334 operand. Reset at the beginning of groups and alternatives. */
2335 UCHAR_T *laststart = 0;
2337 /* Address of beginning of regexp, or inside of last group. */
2340 /* Address of the place where a forward jump should go to the end of
2341 the containing expression. Each alternative of an `or' -- except the
2342 last -- ends with a forward jump of this sort. */
2343 UCHAR_T *fixup_alt_jump = 0;
2345 /* Counts open-groups as they are encountered. Remembered for the
2346 matching close-group on the compile stack, so the same register
2347 number is put in the stop_memory as the start_memory. */
2348 regnum_t regnum = 0;
2351 /* Initialize the wchar_t PATTERN and offset_buffer. */
2352 p = pend = pattern = TALLOC(csize + 1, CHAR_T);
2353 mbs_offset = TALLOC(csize + 1, int);
2354 is_binary = TALLOC(csize + 1, char);
2355 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL)
2362 pattern[csize] = L'\0'; /* sentinel */
2363 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary);
2375 DEBUG_PRINT1 ("\nCompiling pattern: ");
2378 unsigned debug_count;
2380 for (debug_count = 0; debug_count < size; debug_count++)
2381 PUT_CHAR (pattern[debug_count]);
2386 /* Initialize the compile stack. */
2387 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2388 if (compile_stack.stack == NULL)
2398 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2399 compile_stack.avail = 0;
2401 /* Initialize the pattern buffer. */
2402 bufp->syntax = syntax;
2403 bufp->fastmap_accurate = 0;
2404 bufp->not_bol = bufp->not_eol = 0;
2406 /* Set `used' to zero, so that if we return an error, the pattern
2407 printer (for debugging) will think there's no pattern. We reset it
2411 /* Always count groups, whether or not bufp->no_sub is set. */
2414 #if !defined emacs && !defined SYNTAX_TABLE
2415 /* Initialize the syntax table. */
2416 init_syntax_once ();
2419 if (bufp->allocated == 0)
2422 { /* If zero allocated, but buffer is non-null, try to realloc
2423 enough space. This loses if buffer's address is bogus, but
2424 that is the user's responsibility. */
2426 /* Free bufp->buffer and allocate an array for wchar_t pattern
2429 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T),
2432 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T);
2436 { /* Caller did not allocate a buffer. Do it for them. */
2437 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T),
2441 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE);
2443 bufp->buffer = (char*)COMPILED_BUFFER_VAR;
2445 bufp->allocated = INIT_BUF_SIZE;
2449 COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer;
2452 begalt = b = COMPILED_BUFFER_VAR;
2454 /* Loop through the uncompiled pattern until we're at the end. */
2463 if ( /* If at start of pattern, it's an operator. */
2465 /* If context independent, it's an operator. */
2466 || syntax & RE_CONTEXT_INDEP_ANCHORS
2467 /* Otherwise, depends on what's come before. */
2468 || PREFIX(at_begline_loc_p) (pattern, p, syntax))
2478 if ( /* If at end of pattern, it's an operator. */
2480 /* If context independent, it's an operator. */
2481 || syntax & RE_CONTEXT_INDEP_ANCHORS
2482 /* Otherwise, depends on what's next. */
2483 || PREFIX(at_endline_loc_p) (p, pend, syntax))
2493 if ((syntax & RE_BK_PLUS_QM)
2494 || (syntax & RE_LIMITED_OPS))
2498 /* If there is no previous pattern... */
2501 if (syntax & RE_CONTEXT_INVALID_OPS)
2502 FREE_STACK_RETURN (REG_BADRPT);
2503 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2508 /* Are we optimizing this jump? */
2509 boolean keep_string_p = false;
2511 /* 1 means zero (many) matches is allowed. */
2512 char zero_times_ok = 0, many_times_ok = 0;
2514 /* If there is a sequence of repetition chars, collapse it
2515 down to just one (the right one). We can't combine
2516 interval operators with these because of, e.g., `a{2}*',
2517 which should only match an even number of `a's. */
2521 zero_times_ok |= c != '+';
2522 many_times_ok |= c != '?';
2530 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2533 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2535 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2538 if (!(c1 == '+' || c1 == '?'))
2553 /* If we get here, we found another repeat character. */
2556 /* Star, etc. applied to an empty pattern is equivalent
2557 to an empty pattern. */
2561 /* Now we know whether or not zero matches is allowed
2562 and also whether or not two or more matches is allowed. */
2564 { /* More than one repetition is allowed, so put in at the
2565 end a backward relative jump from `b' to before the next
2566 jump we're going to put in below (which jumps from
2567 laststart to after this jump).
2569 But if we are at the `*' in the exact sequence `.*\n',
2570 insert an unconditional jump backwards to the .,
2571 instead of the beginning of the loop. This way we only
2572 push a failure point once, instead of every time
2573 through the loop. */
2574 assert (p - 1 > pattern);
2576 /* Allocate the space for the jump. */
2577 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2579 /* We know we are not at the first character of the pattern,
2580 because laststart was nonzero. And we've already
2581 incremented `p', by the way, to be the character after
2582 the `*'. Do we have to do something analogous here
2583 for null bytes, because of RE_DOT_NOT_NULL? */
2584 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2586 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2587 && !(syntax & RE_DOT_NEWLINE))
2588 { /* We have .*\n. */
2589 STORE_JUMP (jump, b, laststart);
2590 keep_string_p = true;
2593 /* Anything else. */
2594 STORE_JUMP (maybe_pop_jump, b, laststart -
2595 (1 + OFFSET_ADDRESS_SIZE));
2597 /* We've added more stuff to the buffer. */
2598 b += 1 + OFFSET_ADDRESS_SIZE;
2601 /* On failure, jump from laststart to b + 3, which will be the
2602 end of the buffer after this jump is inserted. */
2603 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2605 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2606 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2608 laststart, b + 1 + OFFSET_ADDRESS_SIZE);
2610 b += 1 + OFFSET_ADDRESS_SIZE;
2614 /* At least one repetition is required, so insert a
2615 `dummy_failure_jump' before the initial
2616 `on_failure_jump' instruction of the loop. This
2617 effects a skip over that instruction the first time
2618 we hit that loop. */
2619 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2620 INSERT_JUMP (dummy_failure_jump, laststart, laststart +
2621 2 + 2 * OFFSET_ADDRESS_SIZE);
2622 b += 1 + OFFSET_ADDRESS_SIZE;
2636 boolean had_char_class = false;
2638 CHAR_T range_start = 0xffffffff;
2640 unsigned int range_start = 0xffffffff;
2642 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2645 /* We assume a charset(_not) structure as a wchar_t array.
2646 charset[0] = (re_opcode_t) charset(_not)
2647 charset[1] = l (= length of char_classes)
2648 charset[2] = m (= length of collating_symbols)
2649 charset[3] = n (= length of equivalence_classes)
2650 charset[4] = o (= length of char_ranges)
2651 charset[5] = p (= length of chars)
2653 charset[6] = char_class (wctype_t)
2654 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2656 charset[l+5] = char_class (wctype_t)
2658 charset[l+6] = collating_symbol (wchar_t)
2660 charset[l+m+5] = collating_symbol (wchar_t)
2661 ifdef _LIBC we use the index if
2662 _NL_COLLATE_SYMB_EXTRAMB instead of
2665 charset[l+m+6] = equivalence_classes (wchar_t)
2667 charset[l+m+n+5] = equivalence_classes (wchar_t)
2668 ifdef _LIBC we use the index in
2669 _NL_COLLATE_WEIGHT instead of
2672 charset[l+m+n+6] = range_start
2673 charset[l+m+n+7] = range_end
2675 charset[l+m+n+2o+4] = range_start
2676 charset[l+m+n+2o+5] = range_end
2677 ifdef _LIBC we use the value looked up
2678 in _NL_COLLATE_COLLSEQ instead of
2681 charset[l+m+n+2o+6] = char
2683 charset[l+m+n+2o+p+5] = char
2687 /* We need at least 6 spaces: the opcode, the length of
2688 char_classes, the length of collating_symbols, the length of
2689 equivalence_classes, the length of char_ranges, the length of
2691 GET_BUFFER_SPACE (6);
2693 /* Save b as laststart. And We use laststart as the pointer
2694 to the first element of the charset here.
2695 In other words, laststart[i] indicates charset[i]. */
2698 /* We test `*p == '^' twice, instead of using an if
2699 statement, so we only need one BUF_PUSH. */
2700 BUF_PUSH (*p == '^' ? charset_not : charset);
2704 /* Push the length of char_classes, the length of
2705 collating_symbols, the length of equivalence_classes, the
2706 length of char_ranges and the length of chars. */
2707 BUF_PUSH_3 (0, 0, 0);
2710 /* Remember the first position in the bracket expression. */
2713 /* charset_not matches newline according to a syntax bit. */
2714 if ((re_opcode_t) b[-6] == charset_not
2715 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2718 laststart[5]++; /* Update the length of characters */
2721 /* Read in characters and ranges, setting map bits. */
2724 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2728 /* \ might escape characters inside [...] and [^...]. */
2729 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2731 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2735 laststart[5]++; /* Update the length of chars */
2740 /* Could be the end of the bracket expression. If it's
2741 not (i.e., when the bracket expression is `[]' so
2742 far), the ']' character bit gets set way below. */
2743 if (c == ']' && p != p1 + 1)
2746 /* Look ahead to see if it's a range when the last thing
2747 was a character class. */
2748 if (had_char_class && c == '-' && *p != ']')
2749 FREE_STACK_RETURN (REG_ERANGE);
2751 /* Look ahead to see if it's a range when the last thing
2752 was a character: if this is a hyphen not at the
2753 beginning or the end of a list, then it's the range
2756 && !(p - 2 >= pattern && p[-2] == '[')
2757 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2761 /* Allocate the space for range_start and range_end. */
2762 GET_BUFFER_SPACE (2);
2763 /* Update the pointer to indicate end of buffer. */
2765 ret = wcs_compile_range (range_start, &p, pend, translate,
2766 syntax, b, laststart);
2767 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2768 range_start = 0xffffffff;
2770 else if (p[0] == '-' && p[1] != ']')
2771 { /* This handles ranges made up of characters only. */
2774 /* Move past the `-'. */
2776 /* Allocate the space for range_start and range_end. */
2777 GET_BUFFER_SPACE (2);
2778 /* Update the pointer to indicate end of buffer. */
2780 ret = wcs_compile_range (c, &p, pend, translate, syntax, b,
2782 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2783 range_start = 0xffffffff;
2786 /* See if we're at the beginning of a possible character
2788 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2789 { /* Leave room for the null. */
2790 char str[CHAR_CLASS_MAX_LENGTH + 1];
2795 /* If pattern is `[[:'. */
2796 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2801 if ((c == ':' && *p == ']') || p == pend)
2803 if (c1 < CHAR_CLASS_MAX_LENGTH)
2806 /* This is in any case an invalid class name. */
2811 /* If isn't a word bracketed by `[:' and `:]':
2812 undo the ending character, the letters, and leave
2813 the leading `:' and `[' (but store them as character). */
2814 if (c == ':' && *p == ']')
2819 /* Query the character class as wctype_t. */
2820 wt = IS_CHAR_CLASS (str);
2822 FREE_STACK_RETURN (REG_ECTYPE);
2824 /* Throw away the ] at the end of the character
2828 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2830 /* Allocate the space for character class. */
2831 GET_BUFFER_SPACE(CHAR_CLASS_SIZE);
2832 /* Update the pointer to indicate end of buffer. */
2833 b += CHAR_CLASS_SIZE;
2834 /* Move data which follow character classes
2835 not to violate the data. */
2836 insert_space(CHAR_CLASS_SIZE,
2837 laststart + 6 + laststart[1],
2839 alignedp = ((uintptr_t)(laststart + 6 + laststart[1])
2840 + __alignof__(wctype_t) - 1)
2841 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2842 /* Store the character class. */
2843 *((wctype_t*)alignedp) = wt;
2844 /* Update length of char_classes */
2845 laststart[1] += CHAR_CLASS_SIZE;
2847 had_char_class = true;
2856 laststart[5] += 2; /* Update the length of characters */
2858 had_char_class = false;
2861 else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '='
2864 CHAR_T str[128]; /* Should be large enough. */
2865 CHAR_T delim = *p; /* '=' or '.' */
2868 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2873 /* If pattern is `[[=' or '[[.'. */
2874 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2879 if ((c == delim && *p == ']') || p == pend)
2881 if (c1 < sizeof (str) - 1)
2884 /* This is in any case an invalid class name. */
2889 if (c == delim && *p == ']' && str[0] != '\0')
2891 unsigned int i, offset;
2892 /* If we have no collation data we use the default
2893 collation in which each character is in a class
2894 by itself. It also means that ASCII is the
2895 character set and therefore we cannot have character
2896 with more than one byte in the multibyte
2899 /* If not defined _LIBC, we push the name and
2900 `\0' for the sake of matching performance. */
2901 int datasize = c1 + 1;
2909 FREE_STACK_RETURN (REG_ECOLLATE);
2914 const int32_t *table;
2915 const int32_t *weights;
2916 const int32_t *extra;
2917 const int32_t *indirect;
2920 /* This #include defines a local function! */
2921 # include <locale/weightwc.h>
2925 /* We push the index for equivalence class. */
2928 table = (const int32_t *)
2929 _NL_CURRENT (LC_COLLATE,
2930 _NL_COLLATE_TABLEWC);
2931 weights = (const int32_t *)
2932 _NL_CURRENT (LC_COLLATE,
2933 _NL_COLLATE_WEIGHTWC);
2934 extra = (const int32_t *)
2935 _NL_CURRENT (LC_COLLATE,
2936 _NL_COLLATE_EXTRAWC);
2937 indirect = (const int32_t *)
2938 _NL_CURRENT (LC_COLLATE,
2939 _NL_COLLATE_INDIRECTWC);
2941 idx = findidx ((const wint_t**)&cp);
2942 if (idx == 0 || cp < (wint_t*) str + c1)
2943 /* This is no valid character. */
2944 FREE_STACK_RETURN (REG_ECOLLATE);
2946 str[0] = (wchar_t)idx;
2948 else /* delim == '.' */
2950 /* We push collation sequence value
2951 for collating symbol. */
2953 const int32_t *symb_table;
2954 const unsigned char *extra;
2961 /* We have to convert the name to a single-byte
2962 string. This is possible since the names
2963 consist of ASCII characters and the internal
2964 representation is UCS4. */
2965 for (i = 0; i < c1; ++i)
2966 char_str[i] = str[i];
2969 _NL_CURRENT_WORD (LC_COLLATE,
2970 _NL_COLLATE_SYMB_HASH_SIZEMB);
2971 symb_table = (const int32_t *)
2972 _NL_CURRENT (LC_COLLATE,
2973 _NL_COLLATE_SYMB_TABLEMB);
2974 extra = (const unsigned char *)
2975 _NL_CURRENT (LC_COLLATE,
2976 _NL_COLLATE_SYMB_EXTRAMB);
2978 /* Locate the character in the hashing table. */
2979 hash = elem_hash (char_str, c1);
2982 elem = hash % table_size;
2983 second = hash % (table_size - 2);
2984 while (symb_table[2 * elem] != 0)
2986 /* First compare the hashing value. */
2987 if (symb_table[2 * elem] == hash
2988 && c1 == extra[symb_table[2 * elem + 1]]
2989 && memcmp (char_str,
2990 &extra[symb_table[2 * elem + 1]
2993 /* Yep, this is the entry. */
2994 idx = symb_table[2 * elem + 1];
2995 idx += 1 + extra[idx];
3003 if (symb_table[2 * elem] != 0)
3005 /* Compute the index of the byte sequence
3007 idx += 1 + extra[idx];
3008 /* Adjust for the alignment. */
3009 idx = (idx + 3) & ~3;
3011 str[0] = (wchar_t) idx + 4;
3013 else if (symb_table[2 * elem] == 0 && c1 == 1)
3015 /* No valid character. Match it as a
3016 single byte character. */
3017 had_char_class = false;
3019 /* Update the length of characters */
3021 range_start = str[0];
3023 /* Throw away the ] at the end of the
3024 collating symbol. */
3026 /* exit from the switch block. */
3030 FREE_STACK_RETURN (REG_ECOLLATE);
3035 /* Throw away the ] at the end of the equivalence
3036 class (or collating symbol). */
3039 /* Allocate the space for the equivalence class
3040 (or collating symbol) (and '\0' if needed). */
3041 GET_BUFFER_SPACE(datasize);
3042 /* Update the pointer to indicate end of buffer. */
3046 { /* equivalence class */
3047 /* Calculate the offset of char_ranges,
3048 which is next to equivalence_classes. */
3049 offset = laststart[1] + laststart[2]
3052 insert_space(datasize, laststart + offset, b - 1);
3054 /* Write the equivalence_class and \0. */
3055 for (i = 0 ; i < datasize ; i++)
3056 laststart[offset + i] = str[i];
3058 /* Update the length of equivalence_classes. */
3059 laststart[3] += datasize;
3060 had_char_class = true;
3062 else /* delim == '.' */
3063 { /* collating symbol */
3064 /* Calculate the offset of the equivalence_classes,
3065 which is next to collating_symbols. */
3066 offset = laststart[1] + laststart[2] + 6;
3067 /* Insert space and write the collationg_symbol
3069 insert_space(datasize, laststart + offset, b-1);
3070 for (i = 0 ; i < datasize ; i++)
3071 laststart[offset + i] = str[i];
3073 /* In re_match_2_internal if range_start < -1, we
3074 assume -range_start is the offset of the
3075 collating symbol which is specified as
3076 the character of the range start. So we assign
3077 -(laststart[1] + laststart[2] + 6) to
3079 range_start = -(laststart[1] + laststart[2] + 6);
3080 /* Update the length of collating_symbol. */
3081 laststart[2] += datasize;
3082 had_char_class = false;
3092 laststart[5] += 2; /* Update the length of characters */
3093 range_start = delim;
3094 had_char_class = false;
3099 had_char_class = false;
3101 laststart[5]++; /* Update the length of characters */
3107 /* Ensure that we have enough space to push a charset: the
3108 opcode, the length count, and the bitset; 34 bytes in all. */
3109 GET_BUFFER_SPACE (34);
3113 /* We test `*p == '^' twice, instead of using an if
3114 statement, so we only need one BUF_PUSH. */
3115 BUF_PUSH (*p == '^' ? charset_not : charset);
3119 /* Remember the first position in the bracket expression. */
3122 /* Push the number of bytes in the bitmap. */
3123 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
3125 /* Clear the whole map. */
3126 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
3128 /* charset_not matches newline according to a syntax bit. */
3129 if ((re_opcode_t) b[-2] == charset_not
3130 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
3131 SET_LIST_BIT ('\n');
3133 /* Read in characters and ranges, setting map bits. */
3136 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3140 /* \ might escape characters inside [...] and [^...]. */
3141 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
3143 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3151 /* Could be the end of the bracket expression. If it's
3152 not (i.e., when the bracket expression is `[]' so
3153 far), the ']' character bit gets set way below. */
3154 if (c == ']' && p != p1 + 1)
3157 /* Look ahead to see if it's a range when the last thing
3158 was a character class. */
3159 if (had_char_class && c == '-' && *p != ']')
3160 FREE_STACK_RETURN (REG_ERANGE);
3162 /* Look ahead to see if it's a range when the last thing
3163 was a character: if this is a hyphen not at the
3164 beginning or the end of a list, then it's the range
3167 && !(p - 2 >= pattern && p[-2] == '[')
3168 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
3172 = byte_compile_range (range_start, &p, pend, translate,
3174 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3175 range_start = 0xffffffff;
3178 else if (p[0] == '-' && p[1] != ']')
3179 { /* This handles ranges made up of characters only. */
3182 /* Move past the `-'. */
3185 ret = byte_compile_range (c, &p, pend, translate, syntax, b);
3186 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3187 range_start = 0xffffffff;
3190 /* See if we're at the beginning of a possible character
3193 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3194 { /* Leave room for the null. */
3195 char str[CHAR_CLASS_MAX_LENGTH + 1];
3200 /* If pattern is `[[:'. */
3201 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3206 if ((c == ':' && *p == ']') || p == pend)
3208 if (c1 < CHAR_CLASS_MAX_LENGTH)
3211 /* This is in any case an invalid class name. */
3216 /* If isn't a word bracketed by `[:' and `:]':
3217 undo the ending character, the letters, and leave
3218 the leading `:' and `[' (but set bits for them). */
3219 if (c == ':' && *p == ']')
3221 # if defined _LIBC || WIDE_CHAR_SUPPORT
3222 boolean is_lower = STREQ (str, "lower");
3223 boolean is_upper = STREQ (str, "upper");
3227 wt = IS_CHAR_CLASS (str);
3229 FREE_STACK_RETURN (REG_ECTYPE);
3231 /* Throw away the ] at the end of the character
3235 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3237 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
3239 if (iswctype (btowc (ch), wt))
3242 if (translate && (is_upper || is_lower)
3243 && (ISUPPER (ch) || ISLOWER (ch)))
3247 had_char_class = true;
3250 boolean is_alnum = STREQ (str, "alnum");
3251 boolean is_alpha = STREQ (str, "alpha");
3252 boolean is_blank = STREQ (str, "blank");
3253 boolean is_cntrl = STREQ (str, "cntrl");
3254 boolean is_digit = STREQ (str, "digit");
3255 boolean is_graph = STREQ (str, "graph");
3256 boolean is_lower = STREQ (str, "lower");
3257 boolean is_print = STREQ (str, "print");
3258 boolean is_punct = STREQ (str, "punct");
3259 boolean is_space = STREQ (str, "space");
3260 boolean is_upper = STREQ (str, "upper");
3261 boolean is_xdigit = STREQ (str, "xdigit");
3263 if (!IS_CHAR_CLASS (str))
3264 FREE_STACK_RETURN (REG_ECTYPE);
3266 /* Throw away the ] at the end of the character
3270 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3272 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
3274 /* This was split into 3 if's to
3275 avoid an arbitrary limit in some compiler. */
3276 if ( (is_alnum && ISALNUM (ch))
3277 || (is_alpha && ISALPHA (ch))
3278 || (is_blank && ISBLANK (ch))
3279 || (is_cntrl && ISCNTRL (ch)))
3281 if ( (is_digit && ISDIGIT (ch))
3282 || (is_graph && ISGRAPH (ch))
3283 || (is_lower && ISLOWER (ch))
3284 || (is_print && ISPRINT (ch)))
3286 if ( (is_punct && ISPUNCT (ch))
3287 || (is_space && ISSPACE (ch))
3288 || (is_upper && ISUPPER (ch))
3289 || (is_xdigit && ISXDIGIT (ch)))
3291 if ( translate && (is_upper || is_lower)
3292 && (ISUPPER (ch) || ISLOWER (ch)))
3295 had_char_class = true;
3296 # endif /* libc || wctype.h */
3306 had_char_class = false;
3309 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
3311 unsigned char str[MB_LEN_MAX + 1];
3314 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3320 /* If pattern is `[[='. */
3321 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3326 if ((c == '=' && *p == ']') || p == pend)
3328 if (c1 < MB_LEN_MAX)
3331 /* This is in any case an invalid class name. */
3336 if (c == '=' && *p == ']' && str[0] != '\0')
3338 /* If we have no collation data we use the default
3339 collation in which each character is in a class
3340 by itself. It also means that ASCII is the
3341 character set and therefore we cannot have character
3342 with more than one byte in the multibyte
3349 FREE_STACK_RETURN (REG_ECOLLATE);
3351 /* Throw away the ] at the end of the equivalence
3355 /* Set the bit for the character. */
3356 SET_LIST_BIT (str[0]);
3361 /* Try to match the byte sequence in `str' against
3362 those known to the collate implementation.
3363 First find out whether the bytes in `str' are
3364 actually from exactly one character. */
3365 const int32_t *table;
3366 const unsigned char *weights;
3367 const unsigned char *extra;
3368 const int32_t *indirect;
3370 const unsigned char *cp = str;
3373 /* This #include defines a local function! */
3374 # include <locale/weight.h>
3376 table = (const int32_t *)
3377 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3378 weights = (const unsigned char *)
3379 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3380 extra = (const unsigned char *)
3381 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3382 indirect = (const int32_t *)
3383 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3385 idx = findidx (&cp);
3386 if (idx == 0 || cp < str + c1)
3387 /* This is no valid character. */
3388 FREE_STACK_RETURN (REG_ECOLLATE);
3390 /* Throw away the ] at the end of the equivalence
3394 /* Now we have to go throught the whole table
3395 and find all characters which have the same
3398 XXX Note that this is not entirely correct.
3399 we would have to match multibyte sequences
3400 but this is not possible with the current
3402 for (ch = 1; ch < 256; ++ch)
3403 /* XXX This test would have to be changed if we
3404 would allow matching multibyte sequences. */
3407 int32_t idx2 = table[ch];
3408 size_t len = weights[idx2];
3410 /* Test whether the lenghts match. */
3411 if (weights[idx] == len)
3413 /* They do. New compare the bytes of
3418 && (weights[idx + 1 + cnt]
3419 == weights[idx2 + 1 + cnt]))
3423 /* They match. Mark the character as
3430 had_char_class = true;
3440 had_char_class = false;
3443 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
3445 unsigned char str[128]; /* Should be large enough. */
3448 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3454 /* If pattern is `[[.'. */
3455 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3460 if ((c == '.' && *p == ']') || p == pend)
3462 if (c1 < sizeof (str))
3465 /* This is in any case an invalid class name. */
3470 if (c == '.' && *p == ']' && str[0] != '\0')
3472 /* If we have no collation data we use the default
3473 collation in which each character is the name
3474 for its own class which contains only the one
3475 character. It also means that ASCII is the
3476 character set and therefore we cannot have character
3477 with more than one byte in the multibyte
3484 FREE_STACK_RETURN (REG_ECOLLATE);
3486 /* Throw away the ] at the end of the equivalence
3490 /* Set the bit for the character. */
3491 SET_LIST_BIT (str[0]);
3492 range_start = ((const unsigned char *) str)[0];
3497 /* Try to match the byte sequence in `str' against
3498 those known to the collate implementation.
3499 First find out whether the bytes in `str' are
3500 actually from exactly one character. */
3502 const int32_t *symb_table;
3503 const unsigned char *extra;
3510 _NL_CURRENT_WORD (LC_COLLATE,
3511 _NL_COLLATE_SYMB_HASH_SIZEMB);
3512 symb_table = (const int32_t *)
3513 _NL_CURRENT (LC_COLLATE,
3514 _NL_COLLATE_SYMB_TABLEMB);
3515 extra = (const unsigned char *)
3516 _NL_CURRENT (LC_COLLATE,
3517 _NL_COLLATE_SYMB_EXTRAMB);
3519 /* Locate the character in the hashing table. */
3520 hash = elem_hash (str, c1);
3523 elem = hash % table_size;
3524 second = hash % (table_size - 2);
3525 while (symb_table[2 * elem] != 0)
3527 /* First compare the hashing value. */
3528 if (symb_table[2 * elem] == hash
3529 && c1 == extra[symb_table[2 * elem + 1]]
3531 &extra[symb_table[2 * elem + 1]
3535 /* Yep, this is the entry. */
3536 idx = symb_table[2 * elem + 1];
3537 idx += 1 + extra[idx];
3545 if (symb_table[2 * elem] == 0)
3546 /* This is no valid character. */
3547 FREE_STACK_RETURN (REG_ECOLLATE);
3549 /* Throw away the ] at the end of the equivalence
3553 /* Now add the multibyte character(s) we found
3556 XXX Note that this is not entirely correct.
3557 we would have to match multibyte sequences
3558 but this is not possible with the current
3559 implementation. Also, we have to match
3560 collating symbols, which expand to more than
3561 one file, as a whole and not allow the
3562 individual bytes. */
3565 range_start = extra[idx];
3568 SET_LIST_BIT (extra[idx]);
3573 had_char_class = false;
3583 had_char_class = false;
3588 had_char_class = false;
3594 /* Discard any (non)matching list bytes that are all 0 at the
3595 end of the map. Decrease the map-length byte too. */
3596 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3605 if (syntax & RE_NO_BK_PARENS)
3612 if (syntax & RE_NO_BK_PARENS)
3619 if (syntax & RE_NEWLINE_ALT)
3626 if (syntax & RE_NO_BK_VBAR)
3633 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3634 goto handle_interval;
3640 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3642 /* Do not translate the character after the \, so that we can
3643 distinguish, e.g., \B from \b, even if we normally would
3644 translate, e.g., B to b. */
3650 if (syntax & RE_NO_BK_PARENS)
3651 goto normal_backslash;
3657 if (COMPILE_STACK_FULL)
3659 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3660 compile_stack_elt_t);
3661 if (compile_stack.stack == NULL) return REG_ESPACE;
3663 compile_stack.size <<= 1;
3666 /* These are the values to restore when we hit end of this
3667 group. They are all relative offsets, so that if the
3668 whole pattern moves because of realloc, they will still
3670 COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR;
3671 COMPILE_STACK_TOP.fixup_alt_jump
3672 = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0;
3673 COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR;
3674 COMPILE_STACK_TOP.regnum = regnum;
3676 /* We will eventually replace the 0 with the number of
3677 groups inner to this one. But do not push a
3678 start_memory for groups beyond the last one we can
3679 represent in the compiled pattern. */
3680 if (regnum <= MAX_REGNUM)
3682 COMPILE_STACK_TOP.inner_group_offset = b
3683 - COMPILED_BUFFER_VAR + 2;
3684 BUF_PUSH_3 (start_memory, regnum, 0);
3687 compile_stack.avail++;
3692 /* If we've reached MAX_REGNUM groups, then this open
3693 won't actually generate any code, so we'll have to
3694 clear pending_exact explicitly. */
3700 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3702 if (COMPILE_STACK_EMPTY)
3704 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3705 goto normal_backslash;
3707 FREE_STACK_RETURN (REG_ERPAREN);
3712 { /* Push a dummy failure point at the end of the
3713 alternative for a possible future
3714 `pop_failure_jump' to pop. See comments at
3715 `push_dummy_failure' in `re_match_2'. */
3716 BUF_PUSH (push_dummy_failure);
3718 /* We allocated space for this jump when we assigned
3719 to `fixup_alt_jump', in the `handle_alt' case below. */
3720 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
3723 /* See similar code for backslashed left paren above. */
3724 if (COMPILE_STACK_EMPTY)
3726 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3729 FREE_STACK_RETURN (REG_ERPAREN);
3732 /* Since we just checked for an empty stack above, this
3733 ``can't happen''. */
3734 assert (compile_stack.avail != 0);
3736 /* We don't just want to restore into `regnum', because
3737 later groups should continue to be numbered higher,
3738 as in `(ab)c(de)' -- the second group is #2. */
3739 regnum_t this_group_regnum;
3741 compile_stack.avail--;
3742 begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset;
3744 = COMPILE_STACK_TOP.fixup_alt_jump
3745 ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1
3747 laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset;
3748 this_group_regnum = COMPILE_STACK_TOP.regnum;
3749 /* If we've reached MAX_REGNUM groups, then this open
3750 won't actually generate any code, so we'll have to
3751 clear pending_exact explicitly. */
3754 /* We're at the end of the group, so now we know how many
3755 groups were inside this one. */
3756 if (this_group_regnum <= MAX_REGNUM)
3758 UCHAR_T *inner_group_loc
3759 = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset;
3761 *inner_group_loc = regnum - this_group_regnum;
3762 BUF_PUSH_3 (stop_memory, this_group_regnum,
3763 regnum - this_group_regnum);
3769 case '|': /* `\|'. */
3770 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3771 goto normal_backslash;
3773 if (syntax & RE_LIMITED_OPS)
3776 /* Insert before the previous alternative a jump which
3777 jumps to this alternative if the former fails. */
3778 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3779 INSERT_JUMP (on_failure_jump, begalt,
3780 b + 2 + 2 * OFFSET_ADDRESS_SIZE);
3782 b += 1 + OFFSET_ADDRESS_SIZE;
3784 /* The alternative before this one has a jump after it
3785 which gets executed if it gets matched. Adjust that
3786 jump so it will jump to this alternative's analogous
3787 jump (put in below, which in turn will jump to the next
3788 (if any) alternative's such jump, etc.). The last such
3789 jump jumps to the correct final destination. A picture:
3795 If we are at `b', then fixup_alt_jump right now points to a
3796 three-byte space after `a'. We'll put in the jump, set
3797 fixup_alt_jump to right after `b', and leave behind three
3798 bytes which we'll fill in when we get to after `c'. */
3801 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3803 /* Mark and leave space for a jump after this alternative,
3804 to be filled in later either by next alternative or
3805 when know we're at the end of a series of alternatives. */
3807 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3808 b += 1 + OFFSET_ADDRESS_SIZE;
3816 /* If \{ is a literal. */
3817 if (!(syntax & RE_INTERVALS)
3818 /* If we're at `\{' and it's not the open-interval
3820 || (syntax & RE_NO_BK_BRACES))
3821 goto normal_backslash;
3825 /* If got here, then the syntax allows intervals. */
3827 /* At least (most) this many matches must be made. */
3828 int lower_bound = -1, upper_bound = -1;
3830 /* Place in the uncompiled pattern (i.e., just after
3831 the '{') to go back to if the interval is invalid. */
3832 const CHAR_T *beg_interval = p;
3835 goto invalid_interval;
3837 GET_UNSIGNED_NUMBER (lower_bound);
3841 GET_UNSIGNED_NUMBER (upper_bound);
3842 if (upper_bound < 0)
3843 upper_bound = RE_DUP_MAX;
3846 /* Interval such as `{1}' => match exactly once. */
3847 upper_bound = lower_bound;
3849 if (! (0 <= lower_bound && lower_bound <= upper_bound))
3850 goto invalid_interval;
3852 if (!(syntax & RE_NO_BK_BRACES))
3854 if (c != '\\' || p == pend)
3855 goto invalid_interval;
3860 goto invalid_interval;
3862 /* If it's invalid to have no preceding re. */
3865 if (syntax & RE_CONTEXT_INVALID_OPS
3866 && !(syntax & RE_INVALID_INTERVAL_ORD))
3867 FREE_STACK_RETURN (REG_BADRPT);
3868 else if (syntax & RE_CONTEXT_INDEP_OPS)
3871 goto unfetch_interval;
3874 /* We just parsed a valid interval. */
3876 if (RE_DUP_MAX < upper_bound)
3877 FREE_STACK_RETURN (REG_BADBR);
3879 /* If the upper bound is zero, don't want to succeed at
3880 all; jump from `laststart' to `b + 3', which will be
3881 the end of the buffer after we insert the jump. */
3882 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3883 instead of 'b + 3'. */
3884 if (upper_bound == 0)
3886 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3887 INSERT_JUMP (jump, laststart, b + 1
3888 + OFFSET_ADDRESS_SIZE);
3889 b += 1 + OFFSET_ADDRESS_SIZE;
3892 /* Otherwise, we have a nontrivial interval. When
3893 we're all done, the pattern will look like:
3894 set_number_at <jump count> <upper bound>
3895 set_number_at <succeed_n count> <lower bound>
3896 succeed_n <after jump addr> <succeed_n count>
3898 jump_n <succeed_n addr> <jump count>
3899 (The upper bound and `jump_n' are omitted if
3900 `upper_bound' is 1, though.) */
3902 { /* If the upper bound is > 1, we need to insert
3903 more at the end of the loop. */
3904 unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE +
3905 (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE);
3907 GET_BUFFER_SPACE (nbytes);
3909 /* Initialize lower bound of the `succeed_n', even
3910 though it will be set during matching by its
3911 attendant `set_number_at' (inserted next),
3912 because `re_compile_fastmap' needs to know.
3913 Jump to the `jump_n' we might insert below. */
3914 INSERT_JUMP2 (succeed_n, laststart,
3915 b + 1 + 2 * OFFSET_ADDRESS_SIZE
3916 + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE)
3918 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3920 /* Code to initialize the lower bound. Insert
3921 before the `succeed_n'. The `5' is the last two
3922 bytes of this `set_number_at', plus 3 bytes of
3923 the following `succeed_n'. */
3924 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
3925 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
3926 of the following `succeed_n'. */
3927 PREFIX(insert_op2) (set_number_at, laststart, 1
3928 + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b);
3929 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3931 if (upper_bound > 1)
3932 { /* More than one repetition is allowed, so
3933 append a backward jump to the `succeed_n'
3934 that starts this interval.
3936 When we've reached this during matching,
3937 we'll have matched the interval once, so
3938 jump back only `upper_bound - 1' times. */
3939 STORE_JUMP2 (jump_n, b, laststart
3940 + 2 * OFFSET_ADDRESS_SIZE + 1,
3942 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3944 /* The location we want to set is the second
3945 parameter of the `jump_n'; that is `b-2' as
3946 an absolute address. `laststart' will be
3947 the `set_number_at' we're about to insert;
3948 `laststart+3' the number to set, the source
3949 for the relative address. But we are
3950 inserting into the middle of the pattern --
3951 so everything is getting moved up by 5.
3952 Conclusion: (b - 2) - (laststart + 3) + 5,
3953 i.e., b - laststart.
3955 We insert this at the beginning of the loop
3956 so that if we fail during matching, we'll
3957 reinitialize the bounds. */
3958 PREFIX(insert_op2) (set_number_at, laststart,
3960 upper_bound - 1, b);
3961 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3968 if (!(syntax & RE_INVALID_INTERVAL_ORD))
3969 FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR);
3971 /* Match the characters as literals. */
3974 if (syntax & RE_NO_BK_BRACES)
3977 goto normal_backslash;
3981 /* There is no way to specify the before_dot and after_dot
3982 operators. rms says this is ok. --karl */
3990 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3996 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
4002 if (syntax & RE_NO_GNU_OPS)
4005 BUF_PUSH (wordchar);
4010 if (syntax & RE_NO_GNU_OPS)
4013 BUF_PUSH (notwordchar);
4018 if (syntax & RE_NO_GNU_OPS)
4024 if (syntax & RE_NO_GNU_OPS)
4030 if (syntax & RE_NO_GNU_OPS)
4032 BUF_PUSH (wordbound);
4036 if (syntax & RE_NO_GNU_OPS)
4038 BUF_PUSH (notwordbound);
4042 if (syntax & RE_NO_GNU_OPS)
4048 if (syntax & RE_NO_GNU_OPS)
4053 case '1': case '2': case '3': case '4': case '5':
4054 case '6': case '7': case '8': case '9':
4055 if (syntax & RE_NO_BK_REFS)
4061 FREE_STACK_RETURN (REG_ESUBREG);
4063 /* Can't back reference to a subexpression if inside of it. */
4064 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
4068 BUF_PUSH_2 (duplicate, c1);
4074 if (syntax & RE_BK_PLUS_QM)
4077 goto normal_backslash;
4081 /* You might think it would be useful for \ to mean
4082 not to translate; but if we don't translate it
4083 it will never match anything. */
4091 /* Expects the character in `c'. */
4093 /* If no exactn currently being built. */
4096 /* If last exactn handle binary(or character) and
4097 new exactn handle character(or binary). */
4098 || is_exactn_bin != is_binary[p - 1 - pattern]
4101 /* If last exactn not at current position. */
4102 || pending_exact + *pending_exact + 1 != b
4104 /* We have only one byte following the exactn for the count. */
4105 || *pending_exact == (1 << BYTEWIDTH) - 1
4107 /* If followed by a repetition operator. */
4108 || *p == '*' || *p == '^'
4109 || ((syntax & RE_BK_PLUS_QM)
4110 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
4111 : (*p == '+' || *p == '?'))
4112 || ((syntax & RE_INTERVALS)
4113 && ((syntax & RE_NO_BK_BRACES)
4115 : (p[0] == '\\' && p[1] == '{'))))
4117 /* Start building a new exactn. */
4122 /* Is this exactn binary data or character? */
4123 is_exactn_bin = is_binary[p - 1 - pattern];
4125 BUF_PUSH_2 (exactn_bin, 0);
4127 BUF_PUSH_2 (exactn, 0);
4129 BUF_PUSH_2 (exactn, 0);
4131 pending_exact = b - 1;
4138 } /* while p != pend */
4141 /* Through the pattern now. */
4144 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
4146 if (!COMPILE_STACK_EMPTY)
4147 FREE_STACK_RETURN (REG_EPAREN);
4149 /* If we don't want backtracking, force success
4150 the first time we reach the end of the compiled pattern. */
4151 if (syntax & RE_NO_POSIX_BACKTRACKING)
4159 free (compile_stack.stack);
4161 /* We have succeeded; set the length of the buffer. */
4163 bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR;
4165 bufp->used = b - bufp->buffer;
4171 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4172 PREFIX(print_compiled_pattern) (bufp);
4176 #ifndef MATCH_MAY_ALLOCATE
4177 /* Initialize the failure stack to the largest possible stack. This
4178 isn't necessary unless we're trying to avoid calling alloca in
4179 the search and match routines. */
4181 int num_regs = bufp->re_nsub + 1;
4183 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4184 is strictly greater than re_max_failures, the largest possible stack
4185 is 2 * re_max_failures failure points. */
4186 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
4188 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
4191 if (! fail_stack.stack)
4193 = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size
4194 * sizeof (PREFIX(fail_stack_elt_t)));
4197 = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack,
4199 * sizeof (PREFIX(fail_stack_elt_t))));
4200 # else /* not emacs */
4201 if (! fail_stack.stack)
4203 = malloc (fail_stack.size * sizeof (PREFIX(fail_stack_elt_t)));
4206 = realloc (fail_stack.stack,
4207 fail_stack.size * sizeof (PREFIX(fail_stack_elt_t)));
4208 # endif /* not emacs */
4211 PREFIX(regex_grow_registers) (num_regs);
4213 #endif /* not MATCH_MAY_ALLOCATE */
4216 } /* regex_compile */
4218 /* Subroutines for `regex_compile'. */
4220 /* Store OP at LOC followed by two-byte integer parameter ARG. */
4221 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4224 PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg)
4226 *loc = (UCHAR_T) op;
4227 STORE_NUMBER (loc + 1, arg);
4231 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4232 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4235 PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc, int arg1, int arg2)
4237 *loc = (UCHAR_T) op;
4238 STORE_NUMBER (loc + 1, arg1);
4239 STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2);
4243 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
4244 for OP followed by two-byte integer parameter ARG. */
4245 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4248 PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc, int arg, UCHAR_T *end)
4250 register UCHAR_T *pfrom = end;
4251 register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE;
4253 while (pfrom != loc)
4256 PREFIX(store_op1) (op, loc, arg);
4260 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4261 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4264 PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc, int arg1, int arg2,
4267 register UCHAR_T *pfrom = end;
4268 register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE;
4270 while (pfrom != loc)
4273 PREFIX(store_op2) (op, loc, arg1, arg2);
4277 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
4278 after an alternative or a begin-subexpression. We assume there is at
4279 least one character before the ^. */
4282 PREFIX(at_begline_loc_p) (const CHAR_T *pattern, const CHAR_T *p,
4283 reg_syntax_t syntax)
4285 const CHAR_T *prev = p - 2;
4286 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
4289 /* After a subexpression? */
4290 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
4291 /* After an alternative? */
4292 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
4296 /* The dual of at_begline_loc_p. This one is for $. We assume there is
4297 at least one character after the $, i.e., `P < PEND'. */
4300 PREFIX(at_endline_loc_p) (const CHAR_T *p, const CHAR_T *pend,
4301 reg_syntax_t syntax)
4303 const CHAR_T *next = p;
4304 boolean next_backslash = *next == '\\';
4305 const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0;
4308 /* Before a subexpression? */
4309 (syntax & RE_NO_BK_PARENS ? *next == ')'
4310 : next_backslash && next_next && *next_next == ')')
4311 /* Before an alternative? */
4312 || (syntax & RE_NO_BK_VBAR ? *next == '|'
4313 : next_backslash && next_next && *next_next == '|');
4316 #else /* not INSIDE_RECURSION */
4318 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4319 false if it's not. */
4322 group_in_compile_stack (compile_stack_type compile_stack,
4327 for (this_element = compile_stack.avail - 1;
4330 if (compile_stack.stack[this_element].regnum == regnum)
4335 #endif /* not INSIDE_RECURSION */
4337 #ifdef INSIDE_RECURSION
4340 /* This insert space, which size is "num", into the pattern at "loc".
4341 "end" must point the end of the allocated buffer. */
4343 insert_space (int num, CHAR_T *loc, CHAR_T *end)
4345 register CHAR_T *pto = end;
4346 register CHAR_T *pfrom = end - num;
4348 while (pfrom >= loc)
4354 static reg_errcode_t
4355 wcs_compile_range (CHAR_T range_start_char,
4356 const CHAR_T **p_ptr, const CHAR_T *pend,
4357 RE_TRANSLATE_TYPE translate, reg_syntax_t syntax,
4358 CHAR_T *b, CHAR_T *char_set)
4360 const CHAR_T *p = *p_ptr;
4361 CHAR_T range_start, range_end;
4365 uint32_t start_val, end_val;
4371 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4374 const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE,
4375 _NL_COLLATE_COLLSEQWC);
4376 const unsigned char *extra = (const unsigned char *)
4377 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4379 if (range_start_char < -1)
4381 /* range_start is a collating symbol. */
4383 /* Retreive the index and get collation sequence value. */
4384 wextra = (int32_t*)(extra + char_set[-range_start_char]);
4385 start_val = wextra[1 + *wextra];
4388 start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char));
4390 end_val = collseq_table_lookup (collseq, TRANSLATE (p[0]));
4392 /* Report an error if the range is empty and the syntax prohibits
4394 ret = ((syntax & RE_NO_EMPTY_RANGES)
4395 && (start_val > end_val))? REG_ERANGE : REG_NOERROR;
4397 /* Insert space to the end of the char_ranges. */
4398 insert_space(2, b - char_set[5] - 2, b - 1);
4399 *(b - char_set[5] - 2) = (wchar_t)start_val;
4400 *(b - char_set[5] - 1) = (wchar_t)end_val;
4401 char_set[4]++; /* ranges_index */
4406 range_start = (range_start_char >= 0)? TRANSLATE (range_start_char):
4408 range_end = TRANSLATE (p[0]);
4409 /* Report an error if the range is empty and the syntax prohibits
4411 ret = ((syntax & RE_NO_EMPTY_RANGES)
4412 && (range_start > range_end))? REG_ERANGE : REG_NOERROR;
4414 /* Insert space to the end of the char_ranges. */
4415 insert_space(2, b - char_set[5] - 2, b - 1);
4416 *(b - char_set[5] - 2) = range_start;
4417 *(b - char_set[5] - 1) = range_end;
4418 char_set[4]++; /* ranges_index */
4420 /* Have to increment the pointer into the pattern string, so the
4421 caller isn't still at the ending character. */
4427 /* Read the ending character of a range (in a bracket expression) from the
4428 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4429 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4430 Then we set the translation of all bits between the starting and
4431 ending characters (inclusive) in the compiled pattern B.
4433 Return an error code.
4435 We use these short variable names so we can use the same macros as
4436 `regex_compile' itself. */
4438 static reg_errcode_t
4439 byte_compile_range (unsigned int range_start_char,
4440 const char **p_ptr, const char *pend,
4441 RE_TRANSLATE_TYPE translate, reg_syntax_t syntax,
4445 const char *p = *p_ptr;
4448 const unsigned char *collseq;
4449 unsigned int start_colseq;
4450 unsigned int end_colseq;
4458 /* Have to increment the pointer into the pattern string, so the
4459 caller isn't still at the ending character. */
4462 /* Report an error if the range is empty and the syntax prohibits this. */
4463 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
4466 collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
4467 _NL_COLLATE_COLLSEQMB);
4469 start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)];
4470 end_colseq = collseq[(unsigned char) TRANSLATE (p[0])];
4471 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
4473 unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)];
4475 if (start_colseq <= this_colseq && this_colseq <= end_colseq)
4477 SET_LIST_BIT (TRANSLATE (this_char));
4482 /* Here we see why `this_char' has to be larger than an `unsigned
4483 char' -- we would otherwise go into an infinite loop, since all
4484 characters <= 0xff. */
4485 range_start_char = TRANSLATE (range_start_char);
4486 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4487 and some compilers cast it to int implicitly, so following for_loop
4488 may fall to (almost) infinite loop.
4489 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4490 To avoid this, we cast p[0] to unsigned int and truncate it. */
4491 end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1));
4493 for (this_char = range_start_char; this_char <= end_char; ++this_char)
4495 SET_LIST_BIT (TRANSLATE (this_char));
4504 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4505 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4506 characters can start a string that matches the pattern. This fastmap
4507 is used by re_search to skip quickly over impossible starting points.
4509 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4510 area as BUFP->fastmap.
4512 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4515 Returns 0 if we succeed, -2 if an internal error. */
4518 /* local function for re_compile_fastmap.
4519 truncate wchar_t character to char. */
4521 static unsigned char
4522 truncate_wchar (CHAR_T c)
4524 unsigned char buf[MB_CUR_MAX];
4527 memset (&state, '\0', sizeof (state));
4528 retval = wcrtomb (buf, c, &state);
4529 return retval > 0 ? buf[0] : (unsigned char) c;
4534 PREFIX(re_compile_fastmap) (struct re_pattern_buffer *bufp)
4537 #ifdef MATCH_MAY_ALLOCATE
4538 PREFIX(fail_stack_type) fail_stack;
4540 #ifndef REGEX_MALLOC
4544 register char *fastmap = bufp->fastmap;
4547 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4548 pattern to (char*) in regex_compile. */
4549 UCHAR_T *pattern = (UCHAR_T*)bufp->buffer;
4550 register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used);
4552 UCHAR_T *pattern = bufp->buffer;
4553 register UCHAR_T *pend = pattern + bufp->used;
4555 UCHAR_T *p = pattern;
4558 /* This holds the pointer to the failure stack, when
4559 it is allocated relocatably. */
4560 fail_stack_elt_t *failure_stack_ptr;
4563 /* Assume that each path through the pattern can be null until
4564 proven otherwise. We set this false at the bottom of switch
4565 statement, to which we get only if a particular path doesn't
4566 match the empty string. */
4567 boolean path_can_be_null = true;
4569 /* We aren't doing a `succeed_n' to begin with. */
4570 boolean succeed_n_p = false;
4572 assert (fastmap != NULL && p != NULL);
4575 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4576 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4577 bufp->can_be_null = 0;
4581 if (p == pend || *p == succeed)
4583 /* We have reached the (effective) end of pattern. */
4584 if (!FAIL_STACK_EMPTY ())
4586 bufp->can_be_null |= path_can_be_null;
4588 /* Reset for next path. */
4589 path_can_be_null = true;
4591 p = fail_stack.stack[--fail_stack.avail].pointer;
4599 /* We should never be about to go beyond the end of the pattern. */
4602 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4605 /* I guess the idea here is to simply not bother with a fastmap
4606 if a backreference is used, since it's too hard to figure out
4607 the fastmap for the corresponding group. Setting
4608 `can_be_null' stops `re_search_2' from using the fastmap, so
4609 that is all we do. */
4611 bufp->can_be_null = 1;
4615 /* Following are the cases which match a character. These end
4620 fastmap[truncate_wchar(p[1])] = 1;
4634 /* It is hard to distinguish fastmap from (multi byte) characters
4635 which depends on current locale. */
4640 bufp->can_be_null = 1;
4644 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4645 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
4651 /* Chars beyond end of map must be allowed. */
4652 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
4655 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4656 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
4662 for (j = 0; j < (1 << BYTEWIDTH); j++)
4663 if (SYNTAX (j) == Sword)
4669 for (j = 0; j < (1 << BYTEWIDTH); j++)
4670 if (SYNTAX (j) != Sword)
4677 int fastmap_newline = fastmap['\n'];
4679 /* `.' matches anything ... */
4680 for (j = 0; j < (1 << BYTEWIDTH); j++)
4683 /* ... except perhaps newline. */
4684 if (!(bufp->syntax & RE_DOT_NEWLINE))
4685 fastmap['\n'] = fastmap_newline;
4687 /* Return if we have already set `can_be_null'; if we have,
4688 then the fastmap is irrelevant. Something's wrong here. */
4689 else if (bufp->can_be_null)
4692 /* Otherwise, have to check alternative paths. */
4699 for (j = 0; j < (1 << BYTEWIDTH); j++)
4700 if (SYNTAX (j) == (enum syntaxcode) k)
4707 for (j = 0; j < (1 << BYTEWIDTH); j++)
4708 if (SYNTAX (j) != (enum syntaxcode) k)
4713 /* All cases after this match the empty string. These end with
4733 case push_dummy_failure:
4738 case pop_failure_jump:
4739 case maybe_pop_jump:
4742 case dummy_failure_jump:
4743 EXTRACT_NUMBER_AND_INCR (j, p);
4748 /* Jump backward implies we just went through the body of a
4749 loop and matched nothing. Opcode jumped to should be
4750 `on_failure_jump' or `succeed_n'. Just treat it like an
4751 ordinary jump. For a * loop, it has pushed its failure
4752 point already; if so, discard that as redundant. */
4753 if ((re_opcode_t) *p != on_failure_jump
4754 && (re_opcode_t) *p != succeed_n)
4758 EXTRACT_NUMBER_AND_INCR (j, p);
4761 /* If what's on the stack is where we are now, pop it. */
4762 if (!FAIL_STACK_EMPTY ()
4763 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
4769 case on_failure_jump:
4770 case on_failure_keep_string_jump:
4771 handle_on_failure_jump:
4772 EXTRACT_NUMBER_AND_INCR (j, p);
4774 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4775 end of the pattern. We don't want to push such a point,
4776 since when we restore it above, entering the switch will
4777 increment `p' past the end of the pattern. We don't need
4778 to push such a point since we obviously won't find any more
4779 fastmap entries beyond `pend'. Such a pattern can match
4780 the null string, though. */
4783 if (!PUSH_PATTERN_OP (p + j, fail_stack))
4785 RESET_FAIL_STACK ();
4790 bufp->can_be_null = 1;
4794 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
4795 succeed_n_p = false;
4802 /* Get to the number of times to succeed. */
4803 p += OFFSET_ADDRESS_SIZE;
4805 /* Increment p past the n for when k != 0. */
4806 EXTRACT_NUMBER_AND_INCR (k, p);
4809 p -= 2 * OFFSET_ADDRESS_SIZE;
4810 succeed_n_p = true; /* Spaghetti code alert. */
4811 goto handle_on_failure_jump;
4817 p += 2 * OFFSET_ADDRESS_SIZE;
4828 abort (); /* We have listed all the cases. */
4831 /* Getting here means we have found the possible starting
4832 characters for one path of the pattern -- and that the empty
4833 string does not match. We need not follow this path further.
4834 Instead, look at the next alternative (remembered on the
4835 stack), or quit if no more. The test at the top of the loop
4836 does these things. */
4837 path_can_be_null = false;
4841 /* Set `can_be_null' for the last path (also the first path, if the
4842 pattern is empty). */
4843 bufp->can_be_null |= path_can_be_null;
4846 RESET_FAIL_STACK ();
4850 #else /* not INSIDE_RECURSION */
4853 re_compile_fastmap (struct re_pattern_buffer *bufp)
4856 if (MB_CUR_MAX != 1)
4857 return wcs_re_compile_fastmap(bufp);
4860 return byte_re_compile_fastmap(bufp);
4861 } /* re_compile_fastmap */
4863 weak_alias (__re_compile_fastmap, re_compile_fastmap)
4867 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4868 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4869 this memory for recording register information. STARTS and ENDS
4870 must be allocated using the malloc library routine, and must each
4871 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4873 If NUM_REGS == 0, then subsequent matches should allocate their own
4876 Unless this function is called, the first search or match using
4877 PATTERN_BUFFER will allocate its own register data, without
4878 freeing the old data. */
4881 re_set_registers (struct re_pattern_buffer *bufp,
4882 struct re_registers *regs,
4883 unsigned int num_regs,
4884 regoff_t *starts, regoff_t *ends)
4888 bufp->regs_allocated = REGS_REALLOCATE;
4889 regs->num_regs = num_regs;
4890 regs->start = starts;
4895 bufp->regs_allocated = REGS_UNALLOCATED;
4897 regs->start = regs->end = (regoff_t *) 0;
4901 weak_alias (__re_set_registers, re_set_registers)
4904 /* Searching routines. */
4906 /* Like re_search_2, below, but only one string is specified, and
4907 doesn't let you say where to stop matching. */
4910 re_search (struct re_pattern_buffer *bufp,
4912 int size, int startpos, int range,
4913 struct re_registers *regs)
4915 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4919 weak_alias (__re_search, re_search)
4923 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4924 virtual concatenation of STRING1 and STRING2, starting first at index
4925 STARTPOS, then at STARTPOS + 1, and so on.
4927 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4929 RANGE is how far to scan while trying to match. RANGE = 0 means try
4930 only at STARTPOS; in general, the last start tried is STARTPOS +
4933 In REGS, return the indices of the virtual concatenation of STRING1
4934 and STRING2 that matched the entire BUFP->buffer and its contained
4937 Do not consider matching one past the index STOP in the virtual
4938 concatenation of STRING1 and STRING2.
4940 We return either the position in the strings at which the match was
4941 found, -1 if no match, or -2 if error (such as failure
4945 re_search_2 (struct re_pattern_buffer *bufp,
4946 const char *string1, int size1,
4947 const char *string2, int size2,
4948 int startpos, int range,
4949 struct re_registers *regs,
4953 if (MB_CUR_MAX != 1)
4954 return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos,
4958 return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos,
4962 weak_alias (__re_search_2, re_search_2)
4965 #endif /* not INSIDE_RECURSION */
4967 #ifdef INSIDE_RECURSION
4969 #ifdef MATCH_MAY_ALLOCATE
4970 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
4972 # define FREE_VAR(var) if (var) free (var); var = NULL
4976 # define MAX_ALLOCA_SIZE 2000
4978 # define FREE_WCS_BUFFERS() \
4980 if (size1 > MAX_ALLOCA_SIZE) \
4982 free (wcs_string1); \
4983 free (mbs_offset1); \
4987 FREE_VAR (wcs_string1); \
4988 FREE_VAR (mbs_offset1); \
4990 if (size2 > MAX_ALLOCA_SIZE) \
4992 free (wcs_string2); \
4993 free (mbs_offset2); \
4997 FREE_VAR (wcs_string2); \
4998 FREE_VAR (mbs_offset2); \
5006 PREFIX(re_search_2) (struct re_pattern_buffer *bufp,
5007 const char *string1, int size1,
5008 const char *string2, int size2,
5009 int startpos, int range,
5010 struct re_registers *regs,
5014 register char *fastmap = bufp->fastmap;
5015 register RE_TRANSLATE_TYPE translate = bufp->translate;
5016 int total_size = size1 + size2;
5017 int endpos = startpos + range;
5019 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5020 wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL;
5021 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5022 int wcs_size1 = 0, wcs_size2 = 0;
5023 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5024 int *mbs_offset1 = NULL, *mbs_offset2 = NULL;
5025 /* They hold whether each wchar_t is binary data or not. */
5026 char *is_binary = NULL;
5029 /* Check for out-of-range STARTPOS. */
5030 if (startpos < 0 || startpos > total_size)
5033 /* Fix up RANGE if it might eventually take us outside
5034 the virtual concatenation of STRING1 and STRING2.
5035 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5037 range = 0 - startpos;
5038 else if (endpos > total_size)
5039 range = total_size - startpos;
5041 /* If the search isn't to be a backwards one, don't waste time in a
5042 search for a pattern that must be anchored. */
5043 if (bufp->used > 0 && range > 0
5044 && ((re_opcode_t) bufp->buffer[0] == begbuf
5045 /* `begline' is like `begbuf' if it cannot match at newlines. */
5046 || ((re_opcode_t) bufp->buffer[0] == begline
5047 && !bufp->newline_anchor)))
5056 /* In a forward search for something that starts with \=.
5057 don't keep searching past point. */
5058 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
5060 range = PT - startpos;
5066 /* Update the fastmap now if not correct already. */
5067 if (fastmap && !bufp->fastmap_accurate)
5068 if (re_compile_fastmap (bufp) == -2)
5072 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5073 fill them with converted string. */
5076 if (size1 > MAX_ALLOCA_SIZE)
5078 wcs_string1 = TALLOC (size1 + 1, CHAR_T);
5079 mbs_offset1 = TALLOC (size1 + 1, int);
5080 is_binary = TALLOC (size1 + 1, char);
5084 wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T);
5085 mbs_offset1 = REGEX_TALLOC (size1 + 1, int);
5086 is_binary = REGEX_TALLOC (size1 + 1, char);
5088 if (!wcs_string1 || !mbs_offset1 || !is_binary)
5090 if (size1 > MAX_ALLOCA_SIZE)
5098 FREE_VAR (wcs_string1);
5099 FREE_VAR (mbs_offset1);
5100 FREE_VAR (is_binary);
5104 wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1,
5105 mbs_offset1, is_binary);
5106 wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */
5107 if (size1 > MAX_ALLOCA_SIZE)
5110 FREE_VAR (is_binary);
5114 if (size2 > MAX_ALLOCA_SIZE)
5116 wcs_string2 = TALLOC (size2 + 1, CHAR_T);
5117 mbs_offset2 = TALLOC (size2 + 1, int);
5118 is_binary = TALLOC (size2 + 1, char);
5122 wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T);
5123 mbs_offset2 = REGEX_TALLOC (size2 + 1, int);
5124 is_binary = REGEX_TALLOC (size2 + 1, char);
5126 if (!wcs_string2 || !mbs_offset2 || !is_binary)
5128 FREE_WCS_BUFFERS ();
5129 if (size2 > MAX_ALLOCA_SIZE)
5132 FREE_VAR (is_binary);
5135 wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2,
5136 mbs_offset2, is_binary);
5137 wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */
5138 if (size2 > MAX_ALLOCA_SIZE)
5141 FREE_VAR (is_binary);
5146 /* Loop through the string, looking for a place to start matching. */
5149 /* If a fastmap is supplied, skip quickly over characters that
5150 cannot be the start of a match. If the pattern can match the
5151 null string, however, we don't need to skip characters; we want
5152 the first null string. */
5153 if (fastmap && startpos < total_size && !bufp->can_be_null)
5155 if (range > 0) /* Searching forwards. */
5157 register const char *d;
5158 register int lim = 0;
5161 if (startpos < size1 && startpos + range >= size1)
5162 lim = range - (size1 - startpos);
5164 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
5166 /* Written out as an if-else to avoid testing `translate'
5170 && !fastmap[(unsigned char)
5171 translate[(unsigned char) *d++]])
5174 while (range > lim && !fastmap[(unsigned char) *d++])
5177 startpos += irange - range;
5179 else /* Searching backwards. */
5181 register CHAR_T c = (size1 == 0 || startpos >= size1
5182 ? string2[startpos - size1]
5183 : string1[startpos]);
5185 if (!fastmap[(unsigned char) TRANSLATE (c)])
5190 /* If can't match the null string, and that's all we have left, fail. */
5191 if (range >= 0 && startpos == total_size && fastmap
5192 && !bufp->can_be_null)
5195 FREE_WCS_BUFFERS ();
5201 val = wcs_re_match_2_internal (bufp, string1, size1, string2,
5202 size2, startpos, regs, stop,
5203 wcs_string1, wcs_size1,
5204 wcs_string2, wcs_size2,
5205 mbs_offset1, mbs_offset2);
5207 val = byte_re_match_2_internal (bufp, string1, size1, string2,
5208 size2, startpos, regs, stop);
5211 #ifndef REGEX_MALLOC
5220 FREE_WCS_BUFFERS ();
5228 FREE_WCS_BUFFERS ();
5248 FREE_WCS_BUFFERS ();
5254 /* This converts PTR, a pointer into one of the search wchar_t strings
5255 `string1' and `string2' into an multibyte string offset from the
5256 beginning of that string. We use mbs_offset to optimize.
5257 See convert_mbs_to_wcs. */
5258 # define POINTER_TO_OFFSET(ptr) \
5259 (FIRST_STRING_P (ptr) \
5260 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5261 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5264 /* This converts PTR, a pointer into one of the search strings `string1'
5265 and `string2' into an offset from the beginning of that string. */
5266 # define POINTER_TO_OFFSET(ptr) \
5267 (FIRST_STRING_P (ptr) \
5268 ? ((regoff_t) ((ptr) - string1)) \
5269 : ((regoff_t) ((ptr) - string2 + size1)))
5272 /* Macros for dealing with the split strings in re_match_2. */
5274 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5276 /* Call before fetching a character with *d. This switches over to
5277 string2 if necessary. */
5278 #define PREFETCH() \
5281 /* End of string2 => fail. */ \
5282 if (dend == end_match_2) \
5284 /* End of string1 => advance to string2. */ \
5286 dend = end_match_2; \
5289 /* Test if at very beginning or at very end of the virtual concatenation
5290 of `string1' and `string2'. If only one string, it's `string2'. */
5291 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5292 #define AT_STRINGS_END(d) ((d) == end2)
5295 /* Test if D points to a character which is word-constituent. We have
5296 two special cases to check for: if past the end of string1, look at
5297 the first character in string2; and if before the beginning of
5298 string2, look at the last character in string1. */
5300 /* Use internationalized API instead of SYNTAX. */
5301 # define WORDCHAR_P(d) \
5302 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5303 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5304 || ((d) == end1 ? *string2 \
5305 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5307 # define WORDCHAR_P(d) \
5308 (SYNTAX ((d) == end1 ? *string2 \
5309 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5313 /* Disabled due to a compiler bug -- see comment at case wordbound */
5315 /* Test if the character before D and the one at D differ with respect
5316 to being word-constituent. */
5317 #define AT_WORD_BOUNDARY(d) \
5318 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5319 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5322 /* Free everything we malloc. */
5323 #ifdef MATCH_MAY_ALLOCATE
5325 # define FREE_VARIABLES() \
5327 REGEX_FREE_STACK (fail_stack.stack); \
5328 FREE_VAR (regstart); \
5329 FREE_VAR (regend); \
5330 FREE_VAR (old_regstart); \
5331 FREE_VAR (old_regend); \
5332 FREE_VAR (best_regstart); \
5333 FREE_VAR (best_regend); \
5334 FREE_VAR (reg_info); \
5335 FREE_VAR (reg_dummy); \
5336 FREE_VAR (reg_info_dummy); \
5337 if (!cant_free_wcs_buf) \
5339 FREE_VAR (string1); \
5340 FREE_VAR (string2); \
5341 FREE_VAR (mbs_offset1); \
5342 FREE_VAR (mbs_offset2); \
5346 # define FREE_VARIABLES() \
5348 REGEX_FREE_STACK (fail_stack.stack); \
5349 FREE_VAR (regstart); \
5350 FREE_VAR (regend); \
5351 FREE_VAR (old_regstart); \
5352 FREE_VAR (old_regend); \
5353 FREE_VAR (best_regstart); \
5354 FREE_VAR (best_regend); \
5355 FREE_VAR (reg_info); \
5356 FREE_VAR (reg_dummy); \
5357 FREE_VAR (reg_info_dummy); \
5362 # define FREE_VARIABLES() \
5364 if (!cant_free_wcs_buf) \
5366 FREE_VAR (string1); \
5367 FREE_VAR (string2); \
5368 FREE_VAR (mbs_offset1); \
5369 FREE_VAR (mbs_offset2); \
5373 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5375 #endif /* not MATCH_MAY_ALLOCATE */
5377 /* These values must meet several constraints. They must not be valid
5378 register values; since we have a limit of 255 registers (because
5379 we use only one byte in the pattern for the register number), we can
5380 use numbers larger than 255. They must differ by 1, because of
5381 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5382 be larger than the value for the highest register, so we do not try
5383 to actually save any registers when none are active. */
5384 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5385 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5387 #else /* not INSIDE_RECURSION */
5388 /* Matching routines. */
5390 #ifndef emacs /* Emacs never uses this. */
5391 /* re_match is like re_match_2 except it takes only a single string. */
5394 re_match (struct re_pattern_buffer *bufp,
5397 struct re_registers *regs)
5401 if (MB_CUR_MAX != 1)
5402 result = wcs_re_match_2_internal (bufp, NULL, 0, string, size,
5404 NULL, 0, NULL, 0, NULL, NULL);
5407 result = byte_re_match_2_internal (bufp, NULL, 0, string, size,
5409 # ifndef REGEX_MALLOC
5417 weak_alias (__re_match, re_match)
5419 #endif /* not emacs */
5421 #endif /* not INSIDE_RECURSION */
5423 #ifdef INSIDE_RECURSION
5424 static boolean PREFIX(group_match_null_string_p) (UCHAR_T **p,
5426 PREFIX(register_info_type) *reg_info);
5427 static boolean PREFIX(alt_match_null_string_p) (UCHAR_T *p,
5429 PREFIX(register_info_type) *reg_info);
5430 static boolean PREFIX(common_op_match_null_string_p) (UCHAR_T **p,
5432 PREFIX(register_info_type) *reg_info);
5433 static int PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2,
5434 int len, char *translate);
5435 #else /* not INSIDE_RECURSION */
5437 /* re_match_2 matches the compiled pattern in BUFP against the
5438 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5439 and SIZE2, respectively). We start matching at POS, and stop
5442 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5443 store offsets for the substring each group matched in REGS. See the
5444 documentation for exactly how many groups we fill.
5446 We return -1 if no match, -2 if an internal error (such as the
5447 failure stack overflowing). Otherwise, we return the length of the
5448 matched substring. */
5451 re_match_2 (struct re_pattern_buffer *bufp,
5452 const char *string1, int size1,
5453 const char *string2, int size2,
5454 int pos, struct re_registers *regs,
5459 if (MB_CUR_MAX != 1)
5460 result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2,
5462 NULL, 0, NULL, 0, NULL, NULL);
5465 result = byte_re_match_2_internal (bufp, string1, size1, string2, size2,
5468 #ifndef REGEX_MALLOC
5476 weak_alias (__re_match_2, re_match_2)
5479 #endif /* not INSIDE_RECURSION */
5481 #ifdef INSIDE_RECURSION
5485 /* This check the substring (from 0, to length) of the multibyte string,
5486 to which offset_buffer correspond. And count how many wchar_t_characters
5487 the substring occupy. We use offset_buffer to optimization.
5488 See convert_mbs_to_wcs. */
5491 count_mbs_length (int *offset_buffer, int length)
5495 /* Check whether the size is valid. */
5499 if (offset_buffer == NULL)
5502 /* If there are no multibyte character, offset_buffer[i] == i.
5503 Optmize for this case. */
5504 if (offset_buffer[length] == length)
5507 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5513 int middle = (lower + upper) / 2;
5514 if (middle == lower || middle == upper)
5516 if (offset_buffer[middle] > length)
5518 else if (offset_buffer[middle] < length)
5528 /* This is a separate function so that we can force an alloca cleanup
5532 wcs_re_match_2_internal (struct re_pattern_buffer *bufp,
5533 const char *cstring1, int csize1,
5534 const char *cstring2, int csize2,
5536 struct re_registers *regs,
5538 /* string1 == string2 == NULL means
5539 string1/2, size1/2 and mbs_offset1/2 need
5540 setting up in this function. */
5541 /* We need wchar_t * buffers corresponding to
5542 cstring1, cstring2. */
5543 wchar_t *string1, int size1,
5544 wchar_t *string2, int size2,
5545 /* Offset buffer for optimization. See
5546 convert_mbs_to_wc. */
5551 byte_re_match_2_internal (struct re_pattern_buffer *bufp,
5552 const char *string1, int size1,
5553 const char *string2, int size2,
5555 struct re_registers *regs,
5559 /* General temporaries. */
5563 /* They hold whether each wchar_t is binary data or not. */
5564 char *is_binary = NULL;
5565 /* If true, we can't free string1/2, mbs_offset1/2. */
5566 int cant_free_wcs_buf = 1;
5569 /* Just past the end of the corresponding string. */
5570 const CHAR_T *end1, *end2;
5572 /* Pointers into string1 and string2, just past the last characters in
5573 each to consider matching. */
5574 const CHAR_T *end_match_1, *end_match_2;
5576 /* Where we are in the data, and the end of the current string. */
5577 const CHAR_T *d, *dend;
5579 /* Where we are in the pattern, and the end of the pattern. */
5581 UCHAR_T *pattern, *p;
5582 register UCHAR_T *pend;
5584 UCHAR_T *p = bufp->buffer;
5585 register UCHAR_T *pend = p + bufp->used;
5588 /* Mark the opcode just after a start_memory, so we can test for an
5589 empty subpattern when we get to the stop_memory. */
5590 UCHAR_T *just_past_start_mem = 0;
5592 /* We use this to map every character in the string. */
5593 RE_TRANSLATE_TYPE translate = bufp->translate;
5595 /* Failure point stack. Each place that can handle a failure further
5596 down the line pushes a failure point on this stack. It consists of
5597 restart, regend, and reg_info for all registers corresponding to
5598 the subexpressions we're currently inside, plus the number of such
5599 registers, and, finally, two char *'s. The first char * is where
5600 to resume scanning the pattern; the second one is where to resume
5601 scanning the strings. If the latter is zero, the failure point is
5602 a ``dummy''; if a failure happens and the failure point is a dummy,
5603 it gets discarded and the next next one is tried. */
5604 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5605 PREFIX(fail_stack_type) fail_stack;
5608 static unsigned failure_id;
5609 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5613 /* This holds the pointer to the failure stack, when
5614 it is allocated relocatably. */
5615 fail_stack_elt_t *failure_stack_ptr;
5618 /* We fill all the registers internally, independent of what we
5619 return, for use in backreferences. The number here includes
5620 an element for register zero. */
5621 size_t num_regs = bufp->re_nsub + 1;
5623 /* The currently active registers. */
5624 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
5625 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
5627 /* Information on the contents of registers. These are pointers into
5628 the input strings; they record just what was matched (on this
5629 attempt) by a subexpression part of the pattern, that is, the
5630 regnum-th regstart pointer points to where in the pattern we began
5631 matching and the regnum-th regend points to right after where we
5632 stopped matching the regnum-th subexpression. (The zeroth register
5633 keeps track of what the whole pattern matches.) */
5634 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5635 const CHAR_T **regstart, **regend;
5638 /* If a group that's operated upon by a repetition operator fails to
5639 match anything, then the register for its start will need to be
5640 restored because it will have been set to wherever in the string we
5641 are when we last see its open-group operator. Similarly for a
5643 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5644 const CHAR_T **old_regstart, **old_regend;
5647 /* The is_active field of reg_info helps us keep track of which (possibly
5648 nested) subexpressions we are currently in. The matched_something
5649 field of reg_info[reg_num] helps us tell whether or not we have
5650 matched any of the pattern so far this time through the reg_num-th
5651 subexpression. These two fields get reset each time through any
5652 loop their register is in. */
5653 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5654 PREFIX(register_info_type) *reg_info;
5657 /* The following record the register info as found in the above
5658 variables when we find a match better than any we've seen before.
5659 This happens as we backtrack through the failure points, which in
5660 turn happens only if we have not yet matched the entire string. */
5661 unsigned best_regs_set = false;
5662 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5663 const CHAR_T **best_regstart, **best_regend;
5666 /* Logically, this is `best_regend[0]'. But we don't want to have to
5667 allocate space for that if we're not allocating space for anything
5668 else (see below). Also, we never need info about register 0 for
5669 any of the other register vectors, and it seems rather a kludge to
5670 treat `best_regend' differently than the rest. So we keep track of
5671 the end of the best match so far in a separate variable. We
5672 initialize this to NULL so that when we backtrack the first time
5673 and need to test it, it's not garbage. */
5674 const CHAR_T *match_end = NULL;
5676 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5677 int set_regs_matched_done = 0;
5679 /* Used when we pop values we don't care about. */
5680 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5681 const CHAR_T **reg_dummy;
5682 PREFIX(register_info_type) *reg_info_dummy;
5686 /* Counts the total number of registers pushed. */
5687 unsigned num_regs_pushed = 0;
5690 /* Definitions for state transitions. More efficiently for gcc. */
5692 # if defined HAVE_SUBTRACT_LOCAL_LABELS && defined SHARED
5697 const void *__unbounded ptr; \
5698 offset = (p == pend \
5699 ? 0 : jmptable[SWITCH_ENUM_CAST ((re_opcode_t) *p++)]); \
5700 ptr = &&end_of_pattern + offset; \
5705 &&label_##x - &&end_of_pattern
5706 # define JUMP_TABLE_TYPE const int
5711 const void *__unbounded ptr; \
5712 ptr = (p == pend ? &&end_of_pattern \
5713 : jmptable[SWITCH_ENUM_CAST ((re_opcode_t) *p++)]); \
5719 # define JUMP_TABLE_TYPE const void *const
5721 # define CASE(x) label_##x
5722 static JUMP_TABLE_TYPE jmptable[] =
5741 REF (jump_past_alt),
5742 REF (on_failure_jump),
5743 REF (on_failure_keep_string_jump),
5744 REF (pop_failure_jump),
5745 REF (maybe_pop_jump),
5746 REF (dummy_failure_jump),
5747 REF (push_dummy_failure),
5750 REF (set_number_at),
5772 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5776 #ifdef MATCH_MAY_ALLOCATE
5777 /* Do not bother to initialize all the register variables if there are
5778 no groups in the pattern, as it takes a fair amount of time. If
5779 there are groups, we include space for register 0 (the whole
5780 pattern), even though we never use it, since it simplifies the
5781 array indexing. We should fix this. */
5784 regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5785 regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5786 old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5787 old_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5788 best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5789 best_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5790 reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5791 reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *);
5792 reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5794 if (!(regstart && regend && old_regstart && old_regend && reg_info
5795 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
5803 /* We must initialize all our variables to NULL, so that
5804 `FREE_VARIABLES' doesn't try to free them. */
5805 regstart = regend = old_regstart = old_regend = best_regstart
5806 = best_regend = reg_dummy = NULL;
5807 reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL;
5809 #endif /* MATCH_MAY_ALLOCATE */
5811 /* The starting position is bogus. */
5813 if (pos < 0 || pos > csize1 + csize2)
5815 if (pos < 0 || pos > size1 + size2)
5823 /* Allocate wchar_t array for string1 and string2 and
5824 fill them with converted string. */
5825 if (string1 == NULL && string2 == NULL)
5827 /* We need seting up buffers here. */
5829 /* We must free wcs buffers in this function. */
5830 cant_free_wcs_buf = 0;
5834 string1 = REGEX_TALLOC (csize1 + 1, CHAR_T);
5835 mbs_offset1 = REGEX_TALLOC (csize1 + 1, int);
5836 is_binary = REGEX_TALLOC (csize1 + 1, char);
5837 if (!string1 || !mbs_offset1 || !is_binary)
5840 FREE_VAR (mbs_offset1);
5841 FREE_VAR (is_binary);
5847 string2 = REGEX_TALLOC (csize2 + 1, CHAR_T);
5848 mbs_offset2 = REGEX_TALLOC (csize2 + 1, int);
5849 is_binary = REGEX_TALLOC (csize2 + 1, char);
5850 if (!string2 || !mbs_offset2 || !is_binary)
5853 FREE_VAR (mbs_offset1);
5855 FREE_VAR (mbs_offset2);
5856 FREE_VAR (is_binary);
5859 size2 = convert_mbs_to_wcs(string2, cstring2, csize2,
5860 mbs_offset2, is_binary);
5861 string2[size2] = L'\0'; /* for a sentinel */
5862 FREE_VAR (is_binary);
5866 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5867 pattern to (char*) in regex_compile. */
5868 p = pattern = (CHAR_T*)bufp->buffer;
5869 pend = (CHAR_T*)(bufp->buffer + bufp->used);
5873 /* Initialize subexpression text positions to -1 to mark ones that no
5874 start_memory/stop_memory has been seen for. Also initialize the
5875 register information struct. */
5876 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5878 regstart[mcnt] = regend[mcnt]
5879 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
5881 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
5882 IS_ACTIVE (reg_info[mcnt]) = 0;
5883 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5884 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5887 /* We move `string1' into `string2' if the latter's empty -- but not if
5888 `string1' is null. */
5889 if (size2 == 0 && string1 != NULL)
5896 mbs_offset2 = mbs_offset1;
5902 end1 = string1 + size1;
5903 end2 = string2 + size2;
5905 /* Compute where to stop matching, within the two strings. */
5909 mcnt = count_mbs_length(mbs_offset1, stop);
5910 end_match_1 = string1 + mcnt;
5911 end_match_2 = string2;
5915 if (stop > csize1 + csize2)
5916 stop = csize1 + csize2;
5918 mcnt = count_mbs_length(mbs_offset2, stop-csize1);
5919 end_match_2 = string2 + mcnt;
5922 { /* count_mbs_length return error. */
5929 end_match_1 = string1 + stop;
5930 end_match_2 = string2;
5935 end_match_2 = string2 + stop - size1;
5939 /* `p' scans through the pattern as `d' scans through the data.
5940 `dend' is the end of the input string that `d' points within. `d'
5941 is advanced into the following input string whenever necessary, but
5942 this happens before fetching; therefore, at the beginning of the
5943 loop, `d' can be pointing at the end of a string, but it cannot
5946 if (size1 > 0 && pos <= csize1)
5948 mcnt = count_mbs_length(mbs_offset1, pos);
5954 mcnt = count_mbs_length(mbs_offset2, pos-csize1);
5960 { /* count_mbs_length return error. */
5965 if (size1 > 0 && pos <= size1)
5972 d = string2 + pos - size1;
5977 DEBUG_PRINT1 ("The compiled pattern is:\n");
5978 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5979 DEBUG_PRINT1 ("The string to match is: `");
5980 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5981 DEBUG_PRINT1 ("'\n");
5983 /* This loops over pattern commands. It exits by returning from the
5984 function if the match is complete, or it drops through if the match
5985 fails at this starting point in the input data. */
5989 DEBUG_PRINT2 ("\n%p: ", p);
5991 DEBUG_PRINT2 ("\n0x%x: ", p);
6003 /* End of pattern means we might have succeeded. */
6004 DEBUG_PRINT1 ("end of pattern ... ");
6006 /* If we haven't matched the entire string, and we want the
6007 longest match, try backtracking. */
6008 if (d != end_match_2)
6010 /* 1 if this match ends in the same string (string1 or string2)
6011 as the best previous match. */
6012 boolean same_str_p = (FIRST_STRING_P (match_end)
6013 == MATCHING_IN_FIRST_STRING);
6014 /* 1 if this match is the best seen so far. */
6015 boolean best_match_p;
6017 /* AIX compiler got confused when this was combined
6018 with the previous declaration. */
6020 best_match_p = d > match_end;
6022 best_match_p = !MATCHING_IN_FIRST_STRING;
6024 DEBUG_PRINT1 ("backtracking.\n");
6026 if (!FAIL_STACK_EMPTY ())
6027 { /* More failure points to try. */
6029 /* If exceeds best match so far, save it. */
6030 if (!best_regs_set || best_match_p)
6032 best_regs_set = true;
6035 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6037 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6039 best_regstart[mcnt] = regstart[mcnt];
6040 best_regend[mcnt] = regend[mcnt];
6046 /* If no failure points, don't restore garbage. And if
6047 last match is real best match, don't restore second
6049 else if (best_regs_set && !best_match_p)
6052 /* Restore best match. It may happen that `dend ==
6053 end_match_1' while the restored d is in string2.
6054 For example, the pattern `x.*y.*z' against the
6055 strings `x-' and `y-z-', if the two strings are
6056 not consecutive in memory. */
6057 DEBUG_PRINT1 ("Restoring best registers.\n");
6060 dend = ((d >= string1 && d <= end1)
6061 ? end_match_1 : end_match_2);
6063 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6065 regstart[mcnt] = best_regstart[mcnt];
6066 regend[mcnt] = best_regend[mcnt];
6069 } /* d != end_match_2 */
6072 DEBUG_PRINT1 ("Accepting match.\n");
6073 /* If caller wants register contents data back, do it. */
6074 if (regs && !bufp->no_sub)
6076 /* Have the register data arrays been allocated? */
6077 if (bufp->regs_allocated == REGS_UNALLOCATED)
6078 { /* No. So allocate them with malloc. We need one
6079 extra element beyond `num_regs' for the `-1' marker
6081 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
6082 regs->start = TALLOC (regs->num_regs, regoff_t);
6083 regs->end = TALLOC (regs->num_regs, regoff_t);
6084 if (regs->start == NULL || regs->end == NULL)
6089 bufp->regs_allocated = REGS_REALLOCATE;
6091 else if (bufp->regs_allocated == REGS_REALLOCATE)
6092 { /* Yes. If we need more elements than were already
6093 allocated, reallocate them. If we need fewer, just
6095 if (regs->num_regs < num_regs + 1)
6097 regs->num_regs = num_regs + 1;
6098 RETALLOC (regs->start, regs->num_regs, regoff_t);
6099 RETALLOC (regs->end, regs->num_regs, regoff_t);
6100 if (regs->start == NULL || regs->end == NULL)
6109 /* These braces fend off a "empty body in an else-statement"
6110 warning under GCC when assert expands to nothing. */
6111 assert (bufp->regs_allocated == REGS_FIXED);
6114 /* Convert the pointer data in `regstart' and `regend' to
6115 indices. Register zero has to be set differently,
6116 since we haven't kept track of any info for it. */
6117 if (regs->num_regs > 0)
6119 regs->start[0] = pos;
6121 if (MATCHING_IN_FIRST_STRING)
6122 regs->end[0] = (mbs_offset1 != NULL ?
6123 mbs_offset1[d-string1] : 0);
6125 regs->end[0] = csize1 + (mbs_offset2 != NULL
6126 ? mbs_offset2[d-string2] : 0);
6128 regs->end[0] = (MATCHING_IN_FIRST_STRING
6129 ? ((regoff_t) (d - string1))
6130 : ((regoff_t) (d - string2 + size1)));
6134 /* Go through the first `min (num_regs, regs->num_regs)'
6135 registers, since that is all we initialized. */
6136 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
6139 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
6140 regs->start[mcnt] = regs->end[mcnt] = -1;
6144 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
6146 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
6150 /* If the regs structure we return has more elements than
6151 were in the pattern, set the extra elements to -1. If
6152 we (re)allocated the registers, this is the case,
6153 because we always allocate enough to have at least one
6155 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
6156 regs->start[mcnt] = regs->end[mcnt] = -1;
6157 } /* regs && !bufp->no_sub */
6159 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6160 nfailure_points_pushed, nfailure_points_popped,
6161 nfailure_points_pushed - nfailure_points_popped);
6162 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
6165 if (MATCHING_IN_FIRST_STRING)
6166 mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0;
6168 mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) +
6172 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
6173 ? string1 : string2 - size1);
6176 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
6183 /* Otherwise match next pattern command. */
6184 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
6187 /* Ignore these. Used to ignore the n of succeed_n's which
6188 currently have n == 0. */
6190 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6194 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6197 /* Match the next n pattern characters exactly. The following
6198 byte in the pattern defines n, and the n bytes after that
6199 are the characters to match. */
6205 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
6207 /* This is written out as an if-else so we don't waste time
6208 testing `translate' inside the loop. */
6217 if ((UCHAR_T) translate[(unsigned char) *d++]
6223 if (*d++ != (CHAR_T) *p++)
6227 if ((UCHAR_T) translate[(unsigned char) *d++]
6239 if (*d++ != (CHAR_T) *p++) goto fail;
6243 SET_REGS_MATCHED ();
6247 /* Match any character except possibly a newline or a null. */
6249 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6253 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
6254 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
6257 SET_REGS_MATCHED ();
6258 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d);
6268 unsigned int i, char_class_length, coll_symbol_length,
6269 equiv_class_length, ranges_length, chars_length, length;
6270 CHAR_T *workp, *workp2, *charset_top;
6271 #define WORK_BUFFER_SIZE 128
6272 CHAR_T str_buf[WORK_BUFFER_SIZE];
6277 boolean not = (re_opcode_t) *(p - 1) == charset_not;
6279 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6281 c = TRANSLATE (*d); /* The character to match. */
6284 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
6286 charset_top = p - 1;
6287 char_class_length = *p++;
6288 coll_symbol_length = *p++;
6289 equiv_class_length = *p++;
6290 ranges_length = *p++;
6291 chars_length = *p++;
6292 /* p points charset[6], so the address of the next instruction
6293 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6294 where l=length of char_classes, m=length of collating_symbol,
6295 n=equivalence_class, o=length of char_range,
6296 p'=length of character. */
6298 /* Update p to indicate the next instruction. */
6299 p += char_class_length + coll_symbol_length+ equiv_class_length +
6300 2*ranges_length + chars_length;
6302 /* match with char_class? */
6303 for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE)
6306 uintptr_t alignedp = ((uintptr_t)workp
6307 + __alignof__(wctype_t) - 1)
6308 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6309 wctype = *((wctype_t*)alignedp);
6310 workp += CHAR_CLASS_SIZE;
6311 if (iswctype((wint_t)c, wctype))
6312 goto char_set_matched;
6315 /* match with collating_symbol? */
6319 const unsigned char *extra = (const unsigned char *)
6320 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
6322 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;
6326 wextra = (int32_t*)(extra + *workp++);
6327 for (i = 0; i < *wextra; ++i)
6328 if (TRANSLATE(d[i]) != wextra[1 + i])
6333 /* Update d, however d will be incremented at
6334 char_set_matched:, we decrement d here. */
6336 goto char_set_matched;
6340 else /* (nrules == 0) */
6342 /* If we can't look up collation data, we use wcscoll
6345 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;)
6347 const CHAR_T *backup_d = d, *backup_dend = dend;
6348 length = wcslen (workp);
6350 /* If wcscoll(the collating symbol, whole string) > 0,
6351 any substring of the string never match with the
6352 collating symbol. */
6353 if (wcscoll (workp, d) > 0)
6355 workp += length + 1;
6359 /* First, we compare the collating symbol with
6360 the first character of the string.
6361 If it don't match, we add the next character to
6362 the compare buffer in turn. */
6363 for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++)
6368 if (dend == end_match_2)
6374 /* add next character to the compare buffer. */
6375 str_buf[i] = TRANSLATE(*d);
6376 str_buf[i+1] = '\0';
6378 match = wcscoll (workp, str_buf);
6380 goto char_set_matched;
6383 /* (str_buf > workp) indicate (str_buf + X > workp),
6384 because for all X (str_buf + X > str_buf).
6385 So we don't need continue this loop. */
6388 /* Otherwise(str_buf < workp),
6389 (str_buf+next_character) may equals (workp).
6390 So we continue this loop. */
6395 workp += length + 1;
6398 /* match with equivalence_class? */
6402 const CHAR_T *backup_d = d, *backup_dend = dend;
6403 /* Try to match the equivalence class against
6404 those known to the collate implementation. */
6405 const int32_t *table;
6406 const int32_t *weights;
6407 const int32_t *extra;
6408 const int32_t *indirect;
6413 /* This #include defines a local function! */
6414 # include <locale/weightwc.h>
6416 table = (const int32_t *)
6417 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
6418 weights = (const wint_t *)
6419 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
6420 extra = (const wint_t *)
6421 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
6422 indirect = (const int32_t *)
6423 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
6425 /* Write 1 collating element to str_buf, and
6429 for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++)
6431 cp = (wint_t*)str_buf;
6434 if (dend == end_match_2)
6439 str_buf[i] = TRANSLATE(*(d+i));
6440 str_buf[i+1] = '\0'; /* sentinel */
6441 idx2 = findidx ((const wint_t**)&cp);
6444 /* Update d, however d will be incremented at
6445 char_set_matched:, we decrement d here. */
6446 d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1);
6449 if (dend == end_match_2)
6458 len = weights[idx2];
6460 for (workp2 = workp + equiv_class_length ; workp < workp2 ;
6463 idx = (int32_t)*workp;
6464 /* We already checked idx != 0 in regex_compile. */
6466 if (idx2 != 0 && len == weights[idx])
6469 while (cnt < len && (weights[idx + 1 + cnt]
6470 == weights[idx2 + 1 + cnt]))
6474 goto char_set_matched;
6481 else /* (nrules == 0) */
6483 /* If we can't look up collation data, we use wcscoll
6486 for (workp2 = workp + equiv_class_length ; workp < workp2 ;)
6488 const CHAR_T *backup_d = d, *backup_dend = dend;
6489 length = wcslen (workp);
6491 /* If wcscoll(the collating symbol, whole string) > 0,
6492 any substring of the string never match with the
6493 collating symbol. */
6494 if (wcscoll (workp, d) > 0)
6496 workp += length + 1;
6500 /* First, we compare the equivalence class with
6501 the first character of the string.
6502 If it don't match, we add the next character to
6503 the compare buffer in turn. */
6504 for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++)
6509 if (dend == end_match_2)
6515 /* add next character to the compare buffer. */
6516 str_buf[i] = TRANSLATE(*d);
6517 str_buf[i+1] = '\0';
6519 match = wcscoll (workp, str_buf);
6522 goto char_set_matched;
6525 /* (str_buf > workp) indicate (str_buf + X > workp),
6526 because for all X (str_buf + X > str_buf).
6527 So we don't need continue this loop. */
6530 /* Otherwise(str_buf < workp),
6531 (str_buf+next_character) may equals (workp).
6532 So we continue this loop. */
6537 workp += length + 1;
6541 /* match with char_range? */
6545 uint32_t collseqval;
6546 const char *collseq = (const char *)
6547 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
6549 collseqval = collseq_table_lookup (collseq, c);
6551 for (; workp < p - chars_length ;)
6553 uint32_t start_val, end_val;
6555 /* We already compute the collation sequence value
6556 of the characters (or collating symbols). */
6557 start_val = (uint32_t) *workp++; /* range_start */
6558 end_val = (uint32_t) *workp++; /* range_end */
6560 if (start_val <= collseqval && collseqval <= end_val)
6561 goto char_set_matched;
6567 /* We set range_start_char at str_buf[0], range_end_char
6568 at str_buf[4], and compared char at str_buf[2]. */
6573 for (; workp < p - chars_length ;)
6575 wchar_t *range_start_char, *range_end_char;
6577 /* match if (range_start_char <= c <= range_end_char). */
6579 /* If range_start(or end) < 0, we assume -range_start(end)
6580 is the offset of the collating symbol which is specified
6581 as the character of the range start(end). */
6585 range_start_char = charset_top - (*workp++);
6588 str_buf[0] = *workp++;
6589 range_start_char = str_buf;
6594 range_end_char = charset_top - (*workp++);
6597 str_buf[4] = *workp++;
6598 range_end_char = str_buf + 4;
6601 if (wcscoll (range_start_char, str_buf+2) <= 0
6602 && wcscoll (str_buf+2, range_end_char) <= 0)
6603 goto char_set_matched;
6607 /* match with char? */
6608 for (; workp < p ; workp++)
6610 goto char_set_matched;
6617 /* Cast to `unsigned' instead of `unsigned char' in case the
6618 bit list is a full 32 bytes long. */
6619 if (c < (unsigned) (*p * BYTEWIDTH)
6620 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6625 if (!not) goto fail;
6626 #undef WORK_BUFFER_SIZE
6628 SET_REGS_MATCHED ();
6634 /* The beginning of a group is represented by start_memory.
6635 The arguments are the register number in the next byte, and the
6636 number of groups inner to this one in the next. The text
6637 matched within the group is recorded (in the internal
6638 registers data structure) under the register number. */
6639 CASE (start_memory):
6640 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6641 (long int) *p, (long int) p[1]);
6643 /* Find out if this group can match the empty string. */
6644 p1 = p; /* To send to group_match_null_string_p. */
6646 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
6647 REG_MATCH_NULL_STRING_P (reg_info[*p])
6648 = PREFIX(group_match_null_string_p) (&p1, pend, reg_info);
6650 /* Save the position in the string where we were the last time
6651 we were at this open-group operator in case the group is
6652 operated upon by a repetition operator, e.g., with `(a*)*b'
6653 against `ab'; then we want to ignore where we are now in
6654 the string in case this attempt to match fails. */
6655 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6656 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
6658 DEBUG_PRINT2 (" old_regstart: %d\n",
6659 POINTER_TO_OFFSET (old_regstart[*p]));
6662 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
6664 IS_ACTIVE (reg_info[*p]) = 1;
6665 MATCHED_SOMETHING (reg_info[*p]) = 0;
6667 /* Clear this whenever we change the register activity status. */
6668 set_regs_matched_done = 0;
6670 /* This is the new highest active register. */
6671 highest_active_reg = *p;
6673 /* If nothing was active before, this is the new lowest active
6675 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6676 lowest_active_reg = *p;
6678 /* Move past the register number and inner group count. */
6680 just_past_start_mem = p;
6685 /* The stop_memory opcode represents the end of a group. Its
6686 arguments are the same as start_memory's: the register
6687 number, and the number of inner groups. */
6689 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6690 (long int) *p, (long int) p[1]);
6692 /* We need to save the string position the last time we were at
6693 this close-group operator in case the group is operated
6694 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6695 against `aba'; then we want to ignore where we are now in
6696 the string in case this attempt to match fails. */
6697 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6698 ? REG_UNSET (regend[*p]) ? d : regend[*p]
6700 DEBUG_PRINT2 (" old_regend: %d\n",
6701 POINTER_TO_OFFSET (old_regend[*p]));
6704 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
6706 /* This register isn't active anymore. */
6707 IS_ACTIVE (reg_info[*p]) = 0;
6709 /* Clear this whenever we change the register activity status. */
6710 set_regs_matched_done = 0;
6712 /* If this was the only register active, nothing is active
6714 if (lowest_active_reg == highest_active_reg)
6716 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6717 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6720 { /* We must scan for the new highest active register, since
6721 it isn't necessarily one less than now: consider
6722 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6723 new highest active register is 1. */
6725 while (r > 0 && !IS_ACTIVE (reg_info[r]))
6728 /* If we end up at register zero, that means that we saved
6729 the registers as the result of an `on_failure_jump', not
6730 a `start_memory', and we jumped to past the innermost
6731 `stop_memory'. For example, in ((.)*) we save
6732 registers 1 and 2 as a result of the *, but when we pop
6733 back to the second ), we are at the stop_memory 1.
6734 Thus, nothing is active. */
6737 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6738 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6741 highest_active_reg = r;
6744 /* If just failed to match something this time around with a
6745 group that's operated on by a repetition operator, try to
6746 force exit from the ``loop'', and restore the register
6747 information for this group that we had before trying this
6749 if ((!MATCHED_SOMETHING (reg_info[*p])
6750 || just_past_start_mem == p - 1)
6753 boolean is_a_jump_n = false;
6757 switch ((re_opcode_t) *p1++)
6761 case pop_failure_jump:
6762 case maybe_pop_jump:
6764 case dummy_failure_jump:
6765 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6767 p1 += OFFSET_ADDRESS_SIZE;
6775 /* If the next operation is a jump backwards in the pattern
6776 to an on_failure_jump right before the start_memory
6777 corresponding to this stop_memory, exit from the loop
6778 by forcing a failure after pushing on the stack the
6779 on_failure_jump's jump in the pattern, and d. */
6780 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
6781 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory
6782 && p1[2+OFFSET_ADDRESS_SIZE] == *p)
6784 /* If this group ever matched anything, then restore
6785 what its registers were before trying this last
6786 failed match, e.g., with `(a*)*b' against `ab' for
6787 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6788 against `aba' for regend[3].
6790 Also restore the registers for inner groups for,
6791 e.g., `((a*)(b*))*' against `aba' (register 3 would
6792 otherwise get trashed). */
6794 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
6798 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
6800 /* Restore this and inner groups' (if any) registers. */
6801 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
6804 regstart[r] = old_regstart[r];
6806 /* xx why this test? */
6807 if (old_regend[r] >= regstart[r])
6808 regend[r] = old_regend[r];
6812 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6813 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
6819 /* Move past the register number and the inner group count. */
6824 /* \<digit> has been turned into a `duplicate' command which is
6825 followed by the numeric value of <digit> as the register number. */
6828 register const CHAR_T *d2, *dend2;
6829 int regno = *p++; /* Get which register to match against. */
6830 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
6832 /* Can't back reference a group which we've never matched. */
6833 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
6836 /* Where in input to try to start matching. */
6837 d2 = regstart[regno];
6839 /* Where to stop matching; if both the place to start and
6840 the place to stop matching are in the same string, then
6841 set to the place to stop, otherwise, for now have to use
6842 the end of the first string. */
6844 dend2 = ((FIRST_STRING_P (regstart[regno])
6845 == FIRST_STRING_P (regend[regno]))
6846 ? regend[regno] : end_match_1);
6849 /* If necessary, advance to next segment in register
6853 if (dend2 == end_match_2) break;
6854 if (dend2 == regend[regno]) break;
6856 /* End of string1 => advance to string2. */
6858 dend2 = regend[regno];
6860 /* At end of register contents => success */
6861 if (d2 == dend2) break;
6863 /* If necessary, advance to next segment in data. */
6866 /* How many characters left in this segment to match. */
6869 /* Want how many consecutive characters we can match in
6870 one shot, so, if necessary, adjust the count. */
6871 if (mcnt > dend2 - d2)
6874 /* Compare that many; failure if mismatch, else move
6877 ? PREFIX(bcmp_translate) (d, d2, mcnt, translate)
6878 : memcmp (d, d2, mcnt*sizeof(UCHAR_T)))
6880 d += mcnt, d2 += mcnt;
6882 /* Do this because we've match some characters. */
6883 SET_REGS_MATCHED ();
6889 /* begline matches the empty string at the beginning of the string
6890 (unless `not_bol' is set in `bufp'), and, if
6891 `newline_anchor' is set, after newlines. */
6893 DEBUG_PRINT1 ("EXECUTING begline.\n");
6895 if (AT_STRINGS_BEG (d))
6902 else if (d[-1] == '\n' && bufp->newline_anchor)
6906 /* In all other cases, we fail. */
6910 /* endline is the dual of begline. */
6912 DEBUG_PRINT1 ("EXECUTING endline.\n");
6914 if (AT_STRINGS_END (d))
6922 /* We have to ``prefetch'' the next character. */
6923 else if ((d == end1 ? *string2 : *d) == '\n'
6924 && bufp->newline_anchor)
6931 /* Match at the very beginning of the data. */
6933 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
6934 if (AT_STRINGS_BEG (d))
6941 /* Match at the very end of the data. */
6943 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
6944 if (AT_STRINGS_END (d))
6951 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
6952 pushes NULL as the value for the string on the stack. Then
6953 `pop_failure_point' will keep the current value for the
6954 string, instead of restoring it. To see why, consider
6955 matching `foo\nbar' against `.*\n'. The .* matches the foo;
6956 then the . fails against the \n. But the next thing we want
6957 to do is match the \n against the \n; if we restored the
6958 string value, we would be back at the foo.
6960 Because this is used only in specific cases, we don't need to
6961 check all the things that `on_failure_jump' does, to make
6962 sure the right things get saved on the stack. Hence we don't
6963 share its code. The only reason to push anything on the
6964 stack at all is that otherwise we would have to change
6965 `anychar's code to do something besides goto fail in this
6966 case; that seems worse than this. */
6967 CASE (on_failure_keep_string_jump):
6968 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
6970 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6972 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
6974 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
6977 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
6981 /* Uses of on_failure_jump:
6983 Each alternative starts with an on_failure_jump that points
6984 to the beginning of the next alternative. Each alternative
6985 except the last ends with a jump that in effect jumps past
6986 the rest of the alternatives. (They really jump to the
6987 ending jump of the following alternative, because tensioning
6988 these jumps is a hassle.)
6990 Repeats start with an on_failure_jump that points past both
6991 the repetition text and either the following jump or
6992 pop_failure_jump back to this on_failure_jump. */
6993 CASE (on_failure_jump):
6995 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
6997 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6999 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
7001 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
7004 /* If this on_failure_jump comes right before a group (i.e.,
7005 the original * applied to a group), save the information
7006 for that group and all inner ones, so that if we fail back
7007 to this point, the group's information will be correct.
7008 For example, in \(a*\)*\1, we need the preceding group,
7009 and in \(zz\(a*\)b*\)\2, we need the inner group. */
7011 /* We can't use `p' to check ahead because we push
7012 a failure point to `p + mcnt' after we do this. */
7015 /* We need to skip no_op's before we look for the
7016 start_memory in case this on_failure_jump is happening as
7017 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7019 while (p1 < pend && (re_opcode_t) *p1 == no_op)
7022 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
7024 /* We have a new highest active register now. This will
7025 get reset at the start_memory we are about to get to,
7026 but we will have saved all the registers relevant to
7027 this repetition op, as described above. */
7028 highest_active_reg = *(p1 + 1) + *(p1 + 2);
7029 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
7030 lowest_active_reg = *(p1 + 1);
7033 DEBUG_PRINT1 (":\n");
7034 PUSH_FAILURE_POINT (p + mcnt, d, -2);
7038 /* A smart repeat ends with `maybe_pop_jump'.
7039 We change it to either `pop_failure_jump' or `jump'. */
7040 CASE (maybe_pop_jump):
7041 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7042 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
7044 register UCHAR_T *p2 = p;
7046 /* Compare the beginning of the repeat with what in the
7047 pattern follows its end. If we can establish that there
7048 is nothing that they would both match, i.e., that we
7049 would have to backtrack because of (as in, e.g., `a*a')
7050 then we can change to pop_failure_jump, because we'll
7051 never have to backtrack.
7053 This is not true in the case of alternatives: in
7054 `(a|ab)*' we do need to backtrack to the `ab' alternative
7055 (e.g., if the string was `ab'). But instead of trying to
7056 detect that here, the alternative has put on a dummy
7057 failure point which is what we will end up popping. */
7059 /* Skip over open/close-group commands.
7060 If what follows this loop is a ...+ construct,
7061 look at what begins its body, since we will have to
7062 match at least one of that. */
7066 && ((re_opcode_t) *p2 == stop_memory
7067 || (re_opcode_t) *p2 == start_memory))
7069 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend
7070 && (re_opcode_t) *p2 == dummy_failure_jump)
7071 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE;
7077 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7078 to the `maybe_finalize_jump' of this case. Examine what
7081 /* If we're at the end of the pattern, we can change. */
7084 /* Consider what happens when matching ":\(.*\)"
7085 against ":/". I don't really understand this code
7087 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7090 (" End of pattern: change to `pop_failure_jump'.\n");
7093 else if ((re_opcode_t) *p2 == exactn
7095 || (re_opcode_t) *p2 == exactn_bin
7097 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
7100 = *p2 == (UCHAR_T) endline ? '\n' : p2[2];
7102 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn
7104 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin
7106 ) && p1[3+OFFSET_ADDRESS_SIZE] != c)
7108 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7111 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7113 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]);
7115 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7117 (char) p1[3+OFFSET_ADDRESS_SIZE]);
7122 else if ((re_opcode_t) p1[3] == charset
7123 || (re_opcode_t) p1[3] == charset_not)
7125 int not = (re_opcode_t) p1[3] == charset_not;
7127 if (c < (unsigned) (p1[4] * BYTEWIDTH)
7128 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
7131 /* `not' is equal to 1 if c would match, which means
7132 that we can't change to pop_failure_jump. */
7135 p[-3] = (unsigned char) pop_failure_jump;
7136 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7139 #endif /* not WCHAR */
7142 else if ((re_opcode_t) *p2 == charset)
7144 /* We win if the first character of the loop is not part
7146 if ((re_opcode_t) p1[3] == exactn
7147 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
7148 && (p2[2 + p1[5] / BYTEWIDTH]
7149 & (1 << (p1[5] % BYTEWIDTH)))))
7151 p[-3] = (unsigned char) pop_failure_jump;
7152 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7155 else if ((re_opcode_t) p1[3] == charset_not)
7158 /* We win if the charset_not inside the loop
7159 lists every character listed in the charset after. */
7160 for (idx = 0; idx < (int) p2[1]; idx++)
7161 if (! (p2[2 + idx] == 0
7162 || (idx < (int) p1[4]
7163 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
7168 p[-3] = (unsigned char) pop_failure_jump;
7169 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7172 else if ((re_opcode_t) p1[3] == charset)
7175 /* We win if the charset inside the loop
7176 has no overlap with the one after the loop. */
7178 idx < (int) p2[1] && idx < (int) p1[4];
7180 if ((p2[2 + idx] & p1[5 + idx]) != 0)
7183 if (idx == p2[1] || idx == p1[4])
7185 p[-3] = (unsigned char) pop_failure_jump;
7186 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7190 #endif /* not WCHAR */
7192 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */
7193 if ((re_opcode_t) p[-1] != pop_failure_jump)
7195 p[-1] = (UCHAR_T) jump;
7196 DEBUG_PRINT1 (" Match => jump.\n");
7197 goto unconditional_jump;
7199 /* Note fall through. */
7202 /* The end of a simple repeat has a pop_failure_jump back to
7203 its matching on_failure_jump, where the latter will push a
7204 failure point. The pop_failure_jump takes off failure
7205 points put on by this pop_failure_jump's matching
7206 on_failure_jump; we got through the pattern to here from the
7207 matching on_failure_jump, so didn't fail. */
7208 CASE (pop_failure_jump):
7210 /* We need to pass separate storage for the lowest and
7211 highest registers, even though we don't care about the
7212 actual values. Otherwise, we will restore only one
7213 register from the stack, since lowest will == highest in
7214 `pop_failure_point'. */
7215 active_reg_t dummy_low_reg, dummy_high_reg;
7216 UCHAR_T *pdummy = NULL;
7217 const CHAR_T *sdummy = NULL;
7219 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7220 POP_FAILURE_POINT (sdummy, pdummy,
7221 dummy_low_reg, dummy_high_reg,
7222 reg_dummy, reg_dummy, reg_info_dummy);
7224 /* Note fall through. */
7228 DEBUG_PRINT2 ("\n%p: ", p);
7230 DEBUG_PRINT2 ("\n0x%x: ", p);
7232 /* Note fall through. */
7234 /* Unconditionally jump (without popping any failure points). */
7236 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
7237 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
7238 p += mcnt; /* Do the jump. */
7240 DEBUG_PRINT2 ("(to %p).\n", p);
7242 DEBUG_PRINT2 ("(to 0x%x).\n", p);
7247 /* We need this opcode so we can detect where alternatives end
7248 in `group_match_null_string_p' et al. */
7249 CASE (jump_past_alt):
7250 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7251 goto unconditional_jump;
7254 /* Normally, the on_failure_jump pushes a failure point, which
7255 then gets popped at pop_failure_jump. We will end up at
7256 pop_failure_jump, also, and with a pattern of, say, `a+', we
7257 are skipping over the on_failure_jump, so we have to push
7258 something meaningless for pop_failure_jump to pop. */
7259 CASE (dummy_failure_jump):
7260 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7261 /* It doesn't matter what we push for the string here. What
7262 the code at `fail' tests is the value for the pattern. */
7263 PUSH_FAILURE_POINT (NULL, NULL, -2);
7264 goto unconditional_jump;
7267 /* At the end of an alternative, we need to push a dummy failure
7268 point in case we are followed by a `pop_failure_jump', because
7269 we don't want the failure point for the alternative to be
7270 popped. For example, matching `(a|ab)*' against `aab'
7271 requires that we match the `ab' alternative. */
7272 CASE (push_dummy_failure):
7273 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7274 /* See comments just above at `dummy_failure_jump' about the
7276 PUSH_FAILURE_POINT (NULL, NULL, -2);
7279 /* Have to succeed matching what follows at least n times.
7280 After that, handle like `on_failure_jump'. */
7282 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7283 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
7286 /* Originally, this is how many times we HAVE to succeed. */
7290 p += OFFSET_ADDRESS_SIZE;
7291 STORE_NUMBER_AND_INCR (p, mcnt);
7293 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE
7296 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE
7303 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7304 p + OFFSET_ADDRESS_SIZE);
7306 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7307 p + OFFSET_ADDRESS_SIZE);
7311 p[1] = (UCHAR_T) no_op;
7313 p[2] = (UCHAR_T) no_op;
7314 p[3] = (UCHAR_T) no_op;
7321 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7322 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
7324 /* Originally, this is how many times we CAN jump. */
7328 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt);
7331 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE,
7334 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE,
7337 goto unconditional_jump;
7339 /* If don't have to jump any more, skip over the rest of command. */
7341 p += 2 * OFFSET_ADDRESS_SIZE;
7344 CASE (set_number_at):
7346 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7348 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7350 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7352 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
7354 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
7356 STORE_NUMBER (p1, mcnt);
7361 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7362 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7363 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7364 macro and introducing temporary variables works around the bug. */
7367 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7368 if (AT_WORD_BOUNDARY (d))
7374 CASE (notwordbound):
7375 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7376 if (AT_WORD_BOUNDARY (d))
7382 boolean prevchar, thischar;
7384 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7385 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7390 prevchar = WORDCHAR_P (d - 1);
7391 thischar = WORDCHAR_P (d);
7392 if (prevchar != thischar)
7399 CASE (notwordbound):
7401 boolean prevchar, thischar;
7403 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7404 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7407 prevchar = WORDCHAR_P (d - 1);
7408 thischar = WORDCHAR_P (d);
7409 if (prevchar != thischar)
7416 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7417 if (!AT_STRINGS_END (d) && WORDCHAR_P (d)
7418 && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
7425 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7426 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
7427 && (AT_STRINGS_END (d) || !WORDCHAR_P (d)))
7435 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7436 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
7441 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7442 if (PTR_CHAR_POS ((unsigned char *) d) != point)
7447 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7448 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
7453 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
7458 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7462 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7464 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
7466 SET_REGS_MATCHED ();
7469 CASE (notsyntaxspec):
7470 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
7472 goto matchnotsyntax;
7475 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7479 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7481 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
7483 SET_REGS_MATCHED ();
7486 #else /* not emacs */
7488 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7490 if (!WORDCHAR_P (d))
7492 SET_REGS_MATCHED ();
7497 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7501 SET_REGS_MATCHED ();
7504 #endif /* not emacs */
7510 continue; /* Successfully executed one pattern command; keep going. */
7514 /* We goto here if a matching operation fails. */
7516 if (!FAIL_STACK_EMPTY ())
7517 { /* A restart point is known. Restore to that state. */
7518 DEBUG_PRINT1 ("\nFAIL:\n");
7519 POP_FAILURE_POINT (d, p,
7520 lowest_active_reg, highest_active_reg,
7521 regstart, regend, reg_info);
7523 /* If this failure point is a dummy, try the next one. */
7527 /* If we failed to the end of the pattern, don't examine *p. */
7531 boolean is_a_jump_n = false;
7533 /* If failed to a backwards jump that's part of a repetition
7534 loop, need to pop this failure point and use the next one. */
7535 switch ((re_opcode_t) *p)
7539 case maybe_pop_jump:
7540 case pop_failure_jump:
7543 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7546 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
7548 && (re_opcode_t) *p1 == on_failure_jump))
7556 if (d >= string1 && d <= end1)
7560 break; /* Matching at this starting point really fails. */
7564 goto restore_best_regs;
7568 return -1; /* Failure to match. */
7571 /* Subroutine definitions for re_match_2. */
7574 /* We are passed P pointing to a register number after a start_memory.
7576 Return true if the pattern up to the corresponding stop_memory can
7577 match the empty string, and false otherwise.
7579 If we find the matching stop_memory, sets P to point to one past its number.
7580 Otherwise, sets P to an undefined byte less than or equal to END.
7582 We don't handle duplicates properly (yet). */
7585 PREFIX(group_match_null_string_p) (UCHAR_T **p, UCHAR_T *end,
7586 PREFIX(register_info_type) *reg_info)
7589 /* Point to after the args to the start_memory. */
7590 UCHAR_T *p1 = *p + 2;
7594 /* Skip over opcodes that can match nothing, and return true or
7595 false, as appropriate, when we get to one that can't, or to the
7596 matching stop_memory. */
7598 switch ((re_opcode_t) *p1)
7600 /* Could be either a loop or a series of alternatives. */
7601 case on_failure_jump:
7603 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7605 /* If the next operation is not a jump backwards in the
7610 /* Go through the on_failure_jumps of the alternatives,
7611 seeing if any of the alternatives cannot match nothing.
7612 The last alternative starts with only a jump,
7613 whereas the rest start with on_failure_jump and end
7614 with a jump, e.g., here is the pattern for `a|b|c':
7616 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7617 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7620 So, we have to first go through the first (n-1)
7621 alternatives and then deal with the last one separately. */
7624 /* Deal with the first (n-1) alternatives, which start
7625 with an on_failure_jump (see above) that jumps to right
7626 past a jump_past_alt. */
7628 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] ==
7631 /* `mcnt' holds how many bytes long the alternative
7632 is, including the ending `jump_past_alt' and
7635 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt -
7636 (1 + OFFSET_ADDRESS_SIZE),
7640 /* Move to right after this alternative, including the
7644 /* Break if it's the beginning of an n-th alternative
7645 that doesn't begin with an on_failure_jump. */
7646 if ((re_opcode_t) *p1 != on_failure_jump)
7649 /* Still have to check that it's not an n-th
7650 alternative that starts with an on_failure_jump. */
7652 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7653 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] !=
7656 /* Get to the beginning of the n-th alternative. */
7657 p1 -= 1 + OFFSET_ADDRESS_SIZE;
7662 /* Deal with the last alternative: go back and get number
7663 of the `jump_past_alt' just before it. `mcnt' contains
7664 the length of the alternative. */
7665 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE);
7667 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info))
7670 p1 += mcnt; /* Get past the n-th alternative. */
7676 assert (p1[1] == **p);
7682 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7685 } /* while p1 < end */
7688 } /* group_match_null_string_p */
7691 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7692 It expects P to be the first byte of a single alternative and END one
7693 byte past the last. The alternative can contain groups. */
7696 PREFIX(alt_match_null_string_p) (UCHAR_T *p, UCHAR_T *end,
7697 PREFIX(register_info_type) *reg_info)
7704 /* Skip over opcodes that can match nothing, and break when we get
7705 to one that can't. */
7707 switch ((re_opcode_t) *p1)
7710 case on_failure_jump:
7712 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7717 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7720 } /* while p1 < end */
7723 } /* alt_match_null_string_p */
7726 /* Deals with the ops common to group_match_null_string_p and
7727 alt_match_null_string_p.
7729 Sets P to one after the op and its arguments, if any. */
7732 PREFIX(common_op_match_null_string_p) (UCHAR_T **p, UCHAR_T *end,
7733 PREFIX(register_info_type) *reg_info)
7740 switch ((re_opcode_t) *p1++)
7760 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
7761 ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info);
7763 /* Have to set this here in case we're checking a group which
7764 contains a group and a back reference to it. */
7766 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
7767 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
7773 /* If this is an optimized succeed_n for zero times, make the jump. */
7775 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7783 /* Get to the number of times to succeed. */
7784 p1 += OFFSET_ADDRESS_SIZE;
7785 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7789 p1 -= 2 * OFFSET_ADDRESS_SIZE;
7790 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7798 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
7803 p1 += 2 * OFFSET_ADDRESS_SIZE;
7806 /* All other opcodes mean we cannot match the empty string. */
7812 } /* common_op_match_null_string_p */
7815 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7816 bytes; nonzero otherwise. */
7819 PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2,
7821 RE_TRANSLATE_TYPE translate)
7823 register const UCHAR_T *p1 = (const UCHAR_T *) s1;
7824 register const UCHAR_T *p2 = (const UCHAR_T *) s2;
7828 if (((*p1<=0xff)?translate[*p1++]:*p1++)
7829 != ((*p2<=0xff)?translate[*p2++]:*p2++))
7832 if (translate[*p1++] != translate[*p2++]) return 1;
7840 #else /* not INSIDE_RECURSION */
7842 /* Entry points for GNU code. */
7844 /* re_compile_pattern is the GNU regular expression compiler: it
7845 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7846 Returns 0 if the pattern was valid, otherwise an error string.
7848 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7849 are set in BUFP on entry.
7851 We call regex_compile to do the actual compilation. */
7854 re_compile_pattern (const char *pattern,
7856 struct re_pattern_buffer *bufp)
7860 /* GNU code is written to assume at least RE_NREGS registers will be set
7861 (and at least one extra will be -1). */
7862 bufp->regs_allocated = REGS_UNALLOCATED;
7864 /* And GNU code determines whether or not to get register information
7865 by passing null for the REGS argument to re_match, etc., not by
7869 /* Match anchors at newline. */
7870 bufp->newline_anchor = 1;
7873 if (MB_CUR_MAX != 1)
7874 ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp);
7877 ret = byte_regex_compile (pattern, length, re_syntax_options, bufp);
7881 return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7884 weak_alias (__re_compile_pattern, re_compile_pattern)
7887 /* Entry points compatible with 4.2 BSD regex library. We don't define
7888 them unless specifically requested. */
7890 #if defined _REGEX_RE_COMP || defined _LIBC
7892 /* BSD has one and only one pattern buffer. */
7893 static struct re_pattern_buffer re_comp_buf;
7897 /* Make these definitions weak in libc, so POSIX programs can redefine
7898 these names if they don't use our functions, and still use
7899 regcomp/regexec below without link errors. */
7902 re_comp (const char *s)
7908 if (!re_comp_buf.buffer)
7909 return gettext ("No previous regular expression");
7913 if (!re_comp_buf.buffer)
7915 re_comp_buf.buffer = malloc (200);
7916 if (re_comp_buf.buffer == NULL)
7917 return (char *) gettext (re_error_msgid
7918 + re_error_msgid_idx[(int) REG_ESPACE]);
7919 re_comp_buf.allocated = 200;
7921 re_comp_buf.fastmap = malloc (1 << BYTEWIDTH);
7922 if (re_comp_buf.fastmap == NULL)
7923 return (char *) gettext (re_error_msgid
7924 + re_error_msgid_idx[(int) REG_ESPACE]);
7927 /* Since `re_exec' always passes NULL for the `regs' argument, we
7928 don't need to initialize the pattern buffer fields which affect it. */
7930 /* Match anchors at newlines. */
7931 re_comp_buf.newline_anchor = 1;
7934 if (MB_CUR_MAX != 1)
7935 ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7938 ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7943 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
7944 return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7952 re_exec (const char *s)
7954 const int len = strlen (s);
7956 0 <= re_search (&re_comp_buf, s, len, 0, len, 0);
7959 #endif /* _REGEX_RE_COMP */
7961 /* POSIX.2 functions. Don't define these for Emacs. */
7965 /* regcomp takes a regular expression as a string and compiles it.
7967 PREG is a regex_t *. We do not expect any fields to be initialized,
7968 since POSIX says we shouldn't. Thus, we set
7970 `buffer' to the compiled pattern;
7971 `used' to the length of the compiled pattern;
7972 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
7973 REG_EXTENDED bit in CFLAGS is set; otherwise, to
7974 RE_SYNTAX_POSIX_BASIC;
7975 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
7976 `fastmap' to an allocated space for the fastmap;
7977 `fastmap_accurate' to zero;
7978 `re_nsub' to the number of subexpressions in PATTERN.
7980 PATTERN is the address of the pattern string.
7982 CFLAGS is a series of bits which affect compilation.
7984 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
7985 use POSIX basic syntax.
7987 If REG_NEWLINE is set, then . and [^...] don't match newline.
7988 Also, regexec will try a match beginning after every newline.
7990 If REG_ICASE is set, then we considers upper- and lowercase
7991 versions of letters to be equivalent when matching.
7993 If REG_NOSUB is set, then when PREG is passed to regexec, that
7994 routine will report only success or failure, and nothing about the
7997 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
7998 the return codes and their meanings.) */
8001 regcomp (regex_t *preg, const char *pattern, int cflags)
8005 = (cflags & REG_EXTENDED) ?
8006 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
8008 /* regex_compile will allocate the space for the compiled pattern. */
8010 preg->allocated = 0;
8013 /* Try to allocate space for the fastmap. */
8014 preg->fastmap = malloc (1 << BYTEWIDTH);
8016 if (cflags & REG_ICASE)
8020 preg->translate = malloc (CHAR_SET_SIZE
8021 * sizeof (*(RE_TRANSLATE_TYPE)0));
8022 if (preg->translate == NULL)
8023 return (int) REG_ESPACE;
8025 /* Map uppercase characters to corresponding lowercase ones. */
8026 for (i = 0; i < CHAR_SET_SIZE; i++)
8027 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
8030 preg->translate = NULL;
8032 /* If REG_NEWLINE is set, newlines are treated differently. */
8033 if (cflags & REG_NEWLINE)
8034 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
8035 syntax &= ~RE_DOT_NEWLINE;
8036 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
8037 /* It also changes the matching behavior. */
8038 preg->newline_anchor = 1;
8041 preg->newline_anchor = 0;
8043 preg->no_sub = !!(cflags & REG_NOSUB);
8045 /* POSIX says a null character in the pattern terminates it, so we
8046 can use strlen here in compiling the pattern. */
8048 if (MB_CUR_MAX != 1)
8049 ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg);
8052 ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg);
8054 /* POSIX doesn't distinguish between an unmatched open-group and an
8055 unmatched close-group: both are REG_EPAREN. */
8056 if (ret == REG_ERPAREN) ret = REG_EPAREN;
8058 if (ret == REG_NOERROR && preg->fastmap)
8060 /* Compute the fastmap now, since regexec cannot modify the pattern
8062 if (re_compile_fastmap (preg) == -2)
8064 /* Some error occurred while computing the fastmap, just forget
8066 free (preg->fastmap);
8067 preg->fastmap = NULL;
8074 weak_alias (__regcomp, regcomp)
8078 /* regexec searches for a given pattern, specified by PREG, in the
8081 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8082 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8083 least NMATCH elements, and we set them to the offsets of the
8084 corresponding matched substrings.
8086 EFLAGS specifies `execution flags' which affect matching: if
8087 REG_NOTBOL is set, then ^ does not match at the beginning of the
8088 string; if REG_NOTEOL is set, then $ does not match at the end.
8090 We return 0 if we find a match and REG_NOMATCH if not. */
8093 regexec (const regex_t *preg, const char *string,
8094 size_t nmatch, regmatch_t pmatch[], int eflags)
8097 struct re_registers regs;
8098 regex_t private_preg;
8099 int len = strlen (string);
8100 boolean want_reg_info = !preg->no_sub && nmatch > 0;
8102 private_preg = *preg;
8104 private_preg.not_bol = !!(eflags & REG_NOTBOL);
8105 private_preg.not_eol = !!(eflags & REG_NOTEOL);
8107 /* The user has told us exactly how many registers to return
8108 information about, via `nmatch'. We have to pass that on to the
8109 matching routines. */
8110 private_preg.regs_allocated = REGS_FIXED;
8114 regs.num_regs = nmatch;
8115 regs.start = TALLOC (nmatch * 2, regoff_t);
8116 if (regs.start == NULL)
8117 return (int) REG_NOMATCH;
8118 regs.end = regs.start + nmatch;
8121 /* Perform the searching operation. */
8122 ret = re_search (&private_preg, string, len,
8123 /* start: */ 0, /* range: */ len,
8124 want_reg_info ? ®s : 0);
8126 /* Copy the register information to the POSIX structure. */
8133 for (r = 0; r < nmatch; r++)
8135 pmatch[r].rm_so = regs.start[r];
8136 pmatch[r].rm_eo = regs.end[r];
8140 /* If we needed the temporary register info, free the space now. */
8144 /* We want zero return to mean success, unlike `re_search'. */
8145 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
8148 weak_alias (__regexec, regexec)
8152 /* Returns a message corresponding to an error code, ERRCODE, returned
8153 from either regcomp or regexec. We don't use PREG here. */
8156 regerror (int errcode, const regex_t *preg, char *errbuf, size_t errbuf_size)
8162 || errcode >= (int) (sizeof (re_error_msgid_idx)
8163 / sizeof (re_error_msgid_idx[0])))
8164 /* Only error codes returned by the rest of the code should be passed
8165 to this routine. If we are given anything else, or if other regex
8166 code generates an invalid error code, then the program has a bug.
8167 Dump core so we can fix it. */
8170 msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
8172 msg_size = strlen (msg) + 1; /* Includes the null. */
8174 if (errbuf_size != 0)
8176 if (msg_size > errbuf_size)
8178 #if defined HAVE_MEMPCPY || defined _LIBC
8179 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
8181 memcpy (errbuf, msg, errbuf_size - 1);
8182 errbuf[errbuf_size - 1] = 0;
8186 memcpy (errbuf, msg, msg_size);
8192 weak_alias (__regerror, regerror)
8196 /* Free dynamically allocated space used by PREG. */
8199 regfree (regex_t *preg)
8201 if (preg->buffer != NULL)
8202 free (preg->buffer);
8203 preg->buffer = NULL;
8205 preg->allocated = 0;
8208 if (preg->fastmap != NULL)
8209 free (preg->fastmap);
8210 preg->fastmap = NULL;
8211 preg->fastmap_accurate = 0;
8213 if (preg->translate != NULL)
8214 free (preg->translate);
8215 preg->translate = NULL;
8218 weak_alias (__regfree, regfree)
8221 #endif /* not emacs */
8223 #endif /* not INSIDE_RECURSION */
8227 #undef STORE_NUMBER_AND_INCR
8228 #undef EXTRACT_NUMBER
8229 #undef EXTRACT_NUMBER_AND_INCR
8231 #undef DEBUG_PRINT_COMPILED_PATTERN
8232 #undef DEBUG_PRINT_DOUBLE_STRING
8234 #undef INIT_FAIL_STACK
8235 #undef RESET_FAIL_STACK
8236 #undef DOUBLE_FAIL_STACK
8237 #undef PUSH_PATTERN_OP
8238 #undef PUSH_FAILURE_POINTER
8239 #undef PUSH_FAILURE_INT
8240 #undef PUSH_FAILURE_ELT
8241 #undef POP_FAILURE_POINTER
8242 #undef POP_FAILURE_INT
8243 #undef POP_FAILURE_ELT
8246 #undef PUSH_FAILURE_POINT
8247 #undef POP_FAILURE_POINT
8249 #undef REG_UNSET_VALUE
8257 #undef INIT_BUF_SIZE
8258 #undef GET_BUFFER_SPACE
8266 #undef EXTEND_BUFFER
8267 #undef GET_UNSIGNED_NUMBER
8268 #undef FREE_STACK_RETURN
8270 # undef POINTER_TO_OFFSET
8271 # undef MATCHING_IN_FRST_STRING
8273 # undef AT_STRINGS_BEG
8274 # undef AT_STRINGS_END
8277 # undef FREE_VARIABLES
8278 # undef NO_HIGHEST_ACTIVE_REG
8279 # undef NO_LOWEST_ACTIVE_REG
8283 # undef COMPILED_BUFFER_VAR
8284 # undef OFFSET_ADDRESS_SIZE
8285 # undef CHAR_CLASS_SIZE
8292 # define DEFINED_ONCE