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, 2005 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, 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 /* Tru64 with Desktop Toolkit C has a bug: <stdio.h> must be included
47 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
53 /* We have to keep the namespace clean. */
54 # define regfree(preg) __regfree (preg)
55 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
56 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
57 # define regerror(errcode, preg, errbuf, errbuf_size) \
58 __regerror(errcode, preg, errbuf, errbuf_size)
59 # define re_set_registers(bu, re, nu, st, en) \
60 __re_set_registers (bu, re, nu, st, en)
61 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
62 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
63 # define re_match(bufp, string, size, pos, regs) \
64 __re_match (bufp, string, size, pos, regs)
65 # define re_search(bufp, string, size, startpos, range, regs) \
66 __re_search (bufp, string, size, startpos, range, regs)
67 # define re_compile_pattern(pattern, length, bufp) \
68 __re_compile_pattern (pattern, length, bufp)
69 # define re_set_syntax(syntax) __re_set_syntax (syntax)
70 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
71 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
72 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
74 # define btowc __btowc
75 # define iswctype __iswctype
76 # define mbrtowc __mbrtowc
77 # define wcslen __wcslen
78 # define wcscoll __wcscoll
79 # define wcrtomb __wcrtomb
81 /* We are also using some library internals. */
82 # include <locale/localeinfo.h>
83 # include <locale/elem-hash.h>
84 # include <langinfo.h>
85 # include <locale/coll-lookup.h>
91 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
92 /* This define is so xgettext can find the internationalizable strings. */
93 # define gettext_noop(msgid) msgid
95 /* This is for other GNU distributions with internationalized messages. */
99 /* Support for bounded pointers. */
100 # if !defined _LIBC && !defined __BOUNDED_POINTERS__
101 # define __bounded /* nothing */
102 # define __unbounded /* nothing */
103 # define __ptrvalue /* nothing */
106 /* The `emacs' switch turns on certain matching commands
107 that make sense only in Emacs. */
114 # else /* not emacs */
116 /* If we are not linking with Emacs proper,
117 we can't use the relocating allocator
118 even if config.h says that we can. */
123 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
124 If nothing else has been done, use the method below. */
125 # ifdef INHIBIT_STRING_HEADER
126 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
127 # if !defined bzero && !defined bcopy
128 # undef INHIBIT_STRING_HEADER
133 /* This is the normal way of making sure we have a bcopy and a bzero.
134 This is used in most programs--a few other programs avoid this
135 by defining INHIBIT_STRING_HEADER. */
136 # ifndef INHIBIT_STRING_HEADER
140 # define bzero(s, n) (memset (s, '\0', n), (s))
142 # define bzero(s, n) __bzero (s, n)
147 /* Define the syntax stuff for \<, \>, etc. */
149 /* This must be nonzero for the wordchar and notwordchar pattern
150 commands in re_match_2. */
155 # ifdef SWITCH_ENUM_BUG
156 # define SWITCH_ENUM_CAST(x) ((int)(x))
158 # define SWITCH_ENUM_CAST(x) (x)
161 # endif /* not emacs */
166 # define MB_LEN_MAX 1
169 /* Get the interface, including the syntax bits. */
172 /* isalpha etc. are used for the character classes. */
175 /* Jim Meyering writes:
177 "... Some ctype macros are valid only for character codes that
178 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
179 using /bin/cc or gcc but without giving an ansi option). So, all
180 ctype uses should be through macros like ISPRINT... If
181 STDC_HEADERS is defined, then autoconf has verified that the ctype
182 macros don't need to be guarded with references to isascii. ...
183 Defining isascii to 1 should let any compiler worth its salt
184 eliminate the && through constant folding."
185 Solaris defines some of these symbols so we must undefine them first. */
187 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
188 # define IN_CTYPE_DOMAIN(c) 1
190 # define IN_CTYPE_DOMAIN(c) isascii(c)
194 # define ISBLANK(c) (IN_CTYPE_DOMAIN (c) && isblank (c))
196 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
199 # define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isgraph (c))
201 # define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isprint (c) && !isspace (c))
205 # define ISPRINT(c) (IN_CTYPE_DOMAIN (c) && isprint (c))
206 # define ISDIGIT(c) (IN_CTYPE_DOMAIN (c) && isdigit (c))
207 # define ISALNUM(c) (IN_CTYPE_DOMAIN (c) && isalnum (c))
208 # define ISALPHA(c) (IN_CTYPE_DOMAIN (c) && isalpha (c))
209 # define ISCNTRL(c) (IN_CTYPE_DOMAIN (c) && iscntrl (c))
210 # define ISLOWER(c) (IN_CTYPE_DOMAIN (c) && islower (c))
211 # define ISPUNCT(c) (IN_CTYPE_DOMAIN (c) && ispunct (c))
212 # define ISSPACE(c) (IN_CTYPE_DOMAIN (c) && isspace (c))
213 # define ISUPPER(c) (IN_CTYPE_DOMAIN (c) && isupper (c))
214 # define ISXDIGIT(c) (IN_CTYPE_DOMAIN (c) && isxdigit (c))
217 # define TOLOWER(c) _tolower(c)
219 # define TOLOWER(c) tolower(c)
223 /* How many characters in the character set. */
224 # define CHAR_SET_SIZE 256
228 extern char *re_syntax_table;
230 # else /* not SYNTAX_TABLE */
232 static char re_syntax_table[CHAR_SET_SIZE];
235 init_syntax_once (void)
242 bzero (re_syntax_table, sizeof re_syntax_table);
244 for (c = 0; c < CHAR_SET_SIZE; ++c)
246 re_syntax_table[c] = Sword;
248 re_syntax_table['_'] = Sword;
253 # endif /* not SYNTAX_TABLE */
255 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
259 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
260 use `alloca' instead of `malloc'. This is because using malloc in
261 re_search* or re_match* could cause memory leaks when C-g is used in
262 Emacs; also, malloc is slower and causes storage fragmentation. On
263 the other hand, malloc is more portable, and easier to debug.
265 Because we sometimes use alloca, some routines have to be macros,
266 not functions -- `alloca'-allocated space disappears at the end of the
267 function it is called in. */
271 # define REGEX_ALLOCATE malloc
272 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
273 # define REGEX_FREE free
275 # else /* not REGEX_MALLOC */
277 /* Emacs already defines alloca, sometimes. */
280 /* Make alloca work the best possible way. */
283 # endif /* not alloca */
285 # define REGEX_ALLOCATE alloca
287 /* Assumes a `char *destination' variable. */
288 # define REGEX_REALLOCATE(source, osize, nsize) \
289 (destination = (char *) alloca (nsize), \
290 memcpy (destination, source, osize))
292 /* No need to do anything to free, after alloca. */
293 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
295 # endif /* not REGEX_MALLOC */
297 /* Define how to allocate the failure stack. */
299 # if defined REL_ALLOC && defined REGEX_MALLOC
301 # define REGEX_ALLOCATE_STACK(size) \
302 r_alloc (&failure_stack_ptr, (size))
303 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
304 r_re_alloc (&failure_stack_ptr, (nsize))
305 # define REGEX_FREE_STACK(ptr) \
306 r_alloc_free (&failure_stack_ptr)
308 # else /* not using relocating allocator */
312 # define REGEX_ALLOCATE_STACK malloc
313 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
314 # define REGEX_FREE_STACK free
316 # else /* not REGEX_MALLOC */
318 # define REGEX_ALLOCATE_STACK alloca
320 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
321 REGEX_REALLOCATE (source, osize, nsize)
322 /* No need to explicitly free anything. */
323 # define REGEX_FREE_STACK(arg)
325 # endif /* not REGEX_MALLOC */
326 # endif /* not using relocating allocator */
329 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
330 `string1' or just past its end. This works if PTR is NULL, which is
332 # define FIRST_STRING_P(ptr) \
333 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
335 /* (Re)Allocate N items of type T using malloc, or fail. */
336 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
337 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
338 # define RETALLOC_IF(addr, n, t) \
339 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
340 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
342 # define BYTEWIDTH 8 /* In bits. */
344 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
348 # define MAX(a, b) ((a) > (b) ? (a) : (b))
349 # define MIN(a, b) ((a) < (b) ? (a) : (b))
351 typedef char boolean;
355 static reg_errcode_t byte_regex_compile (const char *pattern, size_t size,
357 struct re_pattern_buffer *bufp);
359 static int byte_re_match_2_internal (struct re_pattern_buffer *bufp,
360 const char *string1, int size1,
361 const char *string2, int size2,
363 struct re_registers *regs,
365 static int byte_re_search_2 (struct re_pattern_buffer *bufp,
366 const char *string1, int size1,
367 const char *string2, int size2,
368 int startpos, int range,
369 struct re_registers *regs, int stop);
370 static int byte_re_compile_fastmap (struct re_pattern_buffer *bufp);
373 static reg_errcode_t wcs_regex_compile (const char *pattern, size_t size,
375 struct re_pattern_buffer *bufp);
378 static int wcs_re_match_2_internal (struct re_pattern_buffer *bufp,
379 const char *cstring1, int csize1,
380 const char *cstring2, int csize2,
382 struct re_registers *regs,
384 wchar_t *string1, int size1,
385 wchar_t *string2, int size2,
386 int *mbs_offset1, int *mbs_offset2);
387 static int wcs_re_search_2 (struct re_pattern_buffer *bufp,
388 const char *string1, int size1,
389 const char *string2, int size2,
390 int startpos, int range,
391 struct re_registers *regs, int stop);
392 static int wcs_re_compile_fastmap (struct re_pattern_buffer *bufp);
395 /* These are the command codes that appear in compiled regular
396 expressions. Some opcodes are followed by argument bytes. A
397 command code can specify any interpretation whatsoever for its
398 arguments. Zero bytes may appear in the compiled regular expression. */
404 /* Succeed right away--no more backtracking. */
407 /* Followed by one byte giving n, then by n literal bytes. */
411 /* Same as exactn, but contains binary data. */
415 /* Matches any (more or less) character. */
418 /* Matches any one char belonging to specified set. First
419 following byte is number of bitmap bytes. Then come bytes
420 for a bitmap saying which chars are in. Bits in each byte
421 are ordered low-bit-first. A character is in the set if its
422 bit is 1. A character too large to have a bit in the map is
423 automatically not in the set. */
424 /* ifdef MBS_SUPPORT, following element is length of character
425 classes, length of collating symbols, length of equivalence
426 classes, length of character ranges, and length of characters.
427 Next, character class element, collating symbols elements,
428 equivalence class elements, range elements, and character
430 See regex_compile function. */
433 /* Same parameters as charset, but match any character that is
434 not one of those specified. */
437 /* Start remembering the text that is matched, for storing in a
438 register. Followed by one byte with the register number, in
439 the range 0 to one less than the pattern buffer's re_nsub
440 field. Then followed by one byte with the number of groups
441 inner to this one. (This last has to be part of the
442 start_memory only because we need it in the on_failure_jump
446 /* Stop remembering the text that is matched and store it in a
447 memory register. Followed by one byte with the register
448 number, in the range 0 to one less than `re_nsub' in the
449 pattern buffer, and one byte with the number of inner groups,
450 just like `start_memory'. (We need the number of inner
451 groups here because we don't have any easy way of finding the
452 corresponding start_memory when we're at a stop_memory.) */
455 /* Match a duplicate of something remembered. Followed by one
456 byte containing the register number. */
459 /* Fail unless at beginning of line. */
462 /* Fail unless at end of line. */
465 /* Succeeds if at beginning of buffer (if emacs) or at beginning
466 of string to be matched (if not). */
469 /* Analogously, for end of buffer/string. */
472 /* Followed by two byte relative address to which to jump. */
475 /* Same as jump, but marks the end of an alternative. */
478 /* Followed by two-byte relative address of place to resume at
479 in case of failure. */
480 /* ifdef MBS_SUPPORT, the size of address is 1. */
483 /* Like on_failure_jump, but pushes a placeholder instead of the
484 current string position when executed. */
485 on_failure_keep_string_jump,
487 /* Throw away latest failure point and then jump to following
488 two-byte relative address. */
489 /* ifdef MBS_SUPPORT, the size of address is 1. */
492 /* Change to pop_failure_jump if know won't have to backtrack to
493 match; otherwise change to jump. This is used to jump
494 back to the beginning of a repeat. If what follows this jump
495 clearly won't match what the repeat does, such that we can be
496 sure that there is no use backtracking out of repetitions
497 already matched, then we change it to a pop_failure_jump.
498 Followed by two-byte address. */
499 /* ifdef MBS_SUPPORT, the size of address is 1. */
502 /* Jump to following two-byte address, and push a dummy failure
503 point. This failure point will be thrown away if an attempt
504 is made to use it for a failure. A `+' construct makes this
505 before the first repeat. Also used as an intermediary kind
506 of jump when compiling an alternative. */
507 /* ifdef MBS_SUPPORT, the size of address is 1. */
510 /* Push a dummy failure point and continue. Used at the end of
514 /* Followed by two-byte relative address and two-byte number n.
515 After matching N times, jump to the address upon failure. */
516 /* ifdef MBS_SUPPORT, the size of address is 1. */
519 /* Followed by two-byte relative address, and two-byte number n.
520 Jump to the address N times, then fail. */
521 /* ifdef MBS_SUPPORT, the size of address is 1. */
524 /* Set the following two-byte relative address to the
525 subsequent two-byte number. The address *includes* the two
527 /* ifdef MBS_SUPPORT, the size of address is 1. */
530 wordchar, /* Matches any word-constituent character. */
531 notwordchar, /* Matches any char that is not a word-constituent. */
533 wordbeg, /* Succeeds if at word beginning. */
534 wordend, /* Succeeds if at word end. */
536 wordbound, /* Succeeds if at a word boundary. */
537 notwordbound /* Succeeds if not at a word boundary. */
540 ,before_dot, /* Succeeds if before point. */
541 at_dot, /* Succeeds if at point. */
542 after_dot, /* Succeeds if after point. */
544 /* Matches any character whose syntax is specified. Followed by
545 a byte which contains a syntax code, e.g., Sword. */
548 /* Matches any character whose syntax is not that specified. */
552 #endif /* not INSIDE_RECURSION */
557 # define UCHAR_T unsigned char
558 # define COMPILED_BUFFER_VAR bufp->buffer
559 # define OFFSET_ADDRESS_SIZE 2
560 # define PREFIX(name) byte_##name
561 # define ARG_PREFIX(name) name
562 # define PUT_CHAR(c) putchar (c)
565 # define CHAR_T wchar_t
566 # define UCHAR_T wchar_t
567 # define COMPILED_BUFFER_VAR wc_buffer
568 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
569 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
570 # define PREFIX(name) wcs_##name
571 # define ARG_PREFIX(name) c##name
572 /* Should we use wide stream?? */
573 # define PUT_CHAR(c) printf ("%C", c);
579 # define INSIDE_RECURSION
581 # undef INSIDE_RECURSION
584 # define INSIDE_RECURSION
586 # undef INSIDE_RECURSION
591 # include "unlocked-io.h"
594 #ifdef INSIDE_RECURSION
595 /* Common operations on the compiled pattern. */
597 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
598 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
601 # define STORE_NUMBER(destination, number) \
603 *(destination) = (UCHAR_T)(number); \
606 # define STORE_NUMBER(destination, number) \
608 (destination)[0] = (number) & 0377; \
609 (destination)[1] = (number) >> 8; \
613 /* Same as STORE_NUMBER, except increment DESTINATION to
614 the byte after where the number is stored. Therefore, DESTINATION
615 must be an lvalue. */
616 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
618 # define STORE_NUMBER_AND_INCR(destination, number) \
620 STORE_NUMBER (destination, number); \
621 (destination) += OFFSET_ADDRESS_SIZE; \
624 /* Put into DESTINATION a number stored in two contiguous bytes starting
626 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
629 # define EXTRACT_NUMBER(destination, source) \
631 (destination) = *(source); \
634 # define EXTRACT_NUMBER(destination, source) \
636 (destination) = *(source) & 0377; \
637 (destination) += (signed char) (*((source) + 1)) << 8; \
643 PREFIX(extract_number) (int *dest, UCHAR_T *source)
648 signed char temp = source[1];
649 *dest = *source & 0377;
654 # ifndef EXTRACT_MACROS /* To debug the macros. */
655 # undef EXTRACT_NUMBER
656 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
657 # endif /* not EXTRACT_MACROS */
661 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
662 SOURCE must be an lvalue. */
664 # define EXTRACT_NUMBER_AND_INCR(destination, source) \
666 EXTRACT_NUMBER (destination, source); \
667 (source) += OFFSET_ADDRESS_SIZE; \
672 PREFIX(extract_number_and_incr) (int *destination, UCHAR_T **source)
674 PREFIX(extract_number) (destination, *source);
675 *source += OFFSET_ADDRESS_SIZE;
678 # ifndef EXTRACT_MACROS
679 # undef EXTRACT_NUMBER_AND_INCR
680 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
681 PREFIX(extract_number_and_incr) (&dest, &src)
682 # endif /* not EXTRACT_MACROS */
688 /* If DEBUG is defined, Regex prints many voluminous messages about what
689 it is doing (if the variable `debug' is nonzero). If linked with the
690 main program in `iregex.c', you can enter patterns and strings
691 interactively. And if linked with the main program in `main.c' and
692 the other test files, you can run the already-written tests. */
696 # ifndef DEFINED_ONCE
698 /* We use standard I/O for debugging. */
701 /* It is useful to test things that ``must'' be true when debugging. */
706 # define DEBUG_STATEMENT(e) e
707 # define DEBUG_PRINT1(x) if (debug) printf (x)
708 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
709 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
710 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
711 # endif /* not DEFINED_ONCE */
713 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
714 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
715 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
716 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
719 /* Print the fastmap in human-readable form. */
721 # ifndef DEFINED_ONCE
723 print_fastmap (char *fastmap)
725 unsigned was_a_range = 0;
728 while (i < (1 << BYTEWIDTH))
734 while (i < (1 << BYTEWIDTH) && fastmap[i])
748 # endif /* not DEFINED_ONCE */
751 /* Print a compiled pattern string in human-readable form, starting at
752 the START pointer into it and ending just before the pointer END. */
755 PREFIX(print_partial_compiled_pattern) (UCHAR_T *start, UCHAR_T *end)
768 /* Loop over pattern commands. */
772 printf ("%td:\t", p - start);
774 printf ("%ld:\t", (long int) (p - start));
777 switch ((re_opcode_t) *p++)
785 printf ("/exactn/%d", mcnt);
797 printf ("/exactn_bin/%d", mcnt);
800 printf("/%lx", (long int) *p++);
804 # endif /* MBS_SUPPORT */
808 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++);
813 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++);
817 printf ("/duplicate/%ld", (long int) *p++);
830 printf ("/charset [%s",
831 (re_opcode_t) *(workp - 1) == charset_not ? "^" : "");
833 length = *workp++; /* the length of char_classes */
834 for (i=0 ; i<length ; i++)
835 printf("[:%lx:]", (long int) *p++);
836 length = *workp++; /* the length of collating_symbol */
837 for (i=0 ; i<length ;)
841 PUT_CHAR((i++,*p++));
845 length = *workp++; /* the length of equivalence_class */
846 for (i=0 ; i<length ;)
850 PUT_CHAR((i++,*p++));
854 length = *workp++; /* the length of char_range */
855 for (i=0 ; i<length ; i++)
857 wchar_t range_start = *p++;
858 wchar_t range_end = *p++;
859 printf("%C-%C", range_start, range_end);
861 length = *workp++; /* the length of char */
862 for (i=0 ; i<length ; i++)
866 register int c, last = -100;
867 register int in_range = 0;
869 printf ("/charset [%s",
870 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
872 assert (p + *p < pend);
874 for (c = 0; c < 256; c++)
876 && (p[1 + (c/8)] & (1 << (c % 8))))
878 /* Are we starting a range? */
879 if (last + 1 == c && ! in_range)
884 /* Have we broken a range? */
885 else if (last + 1 != c && in_range)
915 case on_failure_jump:
916 PREFIX(extract_number_and_incr) (&mcnt, &p);
918 printf ("/on_failure_jump to %td", p + mcnt - start);
920 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
924 case on_failure_keep_string_jump:
925 PREFIX(extract_number_and_incr) (&mcnt, &p);
927 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start);
929 printf ("/on_failure_keep_string_jump to %ld",
930 (long int) (p + mcnt - start));
934 case dummy_failure_jump:
935 PREFIX(extract_number_and_incr) (&mcnt, &p);
937 printf ("/dummy_failure_jump to %td", p + mcnt - start);
939 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
943 case push_dummy_failure:
944 printf ("/push_dummy_failure");
948 PREFIX(extract_number_and_incr) (&mcnt, &p);
950 printf ("/maybe_pop_jump to %td", p + mcnt - start);
952 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
956 case pop_failure_jump:
957 PREFIX(extract_number_and_incr) (&mcnt, &p);
959 printf ("/pop_failure_jump to %td", p + mcnt - start);
961 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
966 PREFIX(extract_number_and_incr) (&mcnt, &p);
968 printf ("/jump_past_alt to %td", p + mcnt - start);
970 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
975 PREFIX(extract_number_and_incr) (&mcnt, &p);
977 printf ("/jump to %td", p + mcnt - start);
979 printf ("/jump to %ld", (long int) (p + mcnt - start));
984 PREFIX(extract_number_and_incr) (&mcnt, &p);
986 PREFIX(extract_number_and_incr) (&mcnt2, &p);
988 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2);
990 printf ("/succeed_n to %ld, %d times",
991 (long int) (p1 - start), mcnt2);
996 PREFIX(extract_number_and_incr) (&mcnt, &p);
998 PREFIX(extract_number_and_incr) (&mcnt2, &p);
999 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
1003 PREFIX(extract_number_and_incr) (&mcnt, &p);
1005 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1007 printf ("/set_number_at location %td to %d", p1 - start, mcnt2);
1009 printf ("/set_number_at location %ld to %d",
1010 (long int) (p1 - start), mcnt2);
1015 printf ("/wordbound");
1019 printf ("/notwordbound");
1023 printf ("/wordbeg");
1027 printf ("/wordend");
1032 printf ("/before_dot");
1040 printf ("/after_dot");
1044 printf ("/syntaxspec");
1046 printf ("/%d", mcnt);
1050 printf ("/notsyntaxspec");
1052 printf ("/%d", mcnt);
1057 printf ("/wordchar");
1061 printf ("/notwordchar");
1073 printf ("?%ld", (long int) *(p-1));
1080 printf ("%td:\tend of pattern.\n", p - start);
1082 printf ("%ld:\tend of pattern.\n", (long int) (p - start));
1088 PREFIX(print_compiled_pattern) (struct re_pattern_buffer *bufp)
1090 UCHAR_T *buffer = (UCHAR_T*) bufp->buffer;
1092 PREFIX(print_partial_compiled_pattern) (buffer, buffer
1093 + bufp->used / sizeof(UCHAR_T));
1094 printf ("%ld bytes used/%ld bytes allocated.\n",
1095 bufp->used, bufp->allocated);
1097 if (bufp->fastmap_accurate && bufp->fastmap)
1099 printf ("fastmap: ");
1100 print_fastmap (bufp->fastmap);
1104 printf ("re_nsub: %Zd\t", bufp->re_nsub);
1106 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
1108 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1109 printf ("can_be_null: %d\t", bufp->can_be_null);
1110 printf ("newline_anchor: %d\n", bufp->newline_anchor);
1111 printf ("no_sub: %d\t", bufp->no_sub);
1112 printf ("not_bol: %d\t", bufp->not_bol);
1113 printf ("not_eol: %d\t", bufp->not_eol);
1114 printf ("syntax: %lx\n", bufp->syntax);
1115 /* Perhaps we should print the translate table? */
1120 PREFIX(print_double_string) (const CHAR_T *where,
1121 const CHAR_T *string1,
1122 const CHAR_T *string2,
1134 if (FIRST_STRING_P (where))
1136 for (this_char = where - string1; this_char < size1; this_char++)
1137 PUT_CHAR (string1[this_char]);
1143 for (this_char = where - string2; this_char < size2; this_char++)
1145 PUT_CHAR (string2[this_char]);
1148 fputs ("...", stdout);
1155 # ifndef DEFINED_ONCE
1164 # else /* not DEBUG */
1166 # ifndef DEFINED_ONCE
1170 # define DEBUG_STATEMENT(e)
1171 # define DEBUG_PRINT1(x)
1172 # define DEBUG_PRINT2(x1, x2)
1173 # define DEBUG_PRINT3(x1, x2, x3)
1174 # define DEBUG_PRINT4(x1, x2, x3, x4)
1175 # endif /* not DEFINED_ONCE */
1176 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1177 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1179 # endif /* not DEBUG */
1184 /* This convert a multibyte string to a wide character string.
1185 And write their correspondances to offset_buffer(see below)
1186 and write whether each wchar_t is binary data to is_binary.
1187 This assume invalid multibyte sequences as binary data.
1188 We assume offset_buffer and is_binary is already allocated
1192 convert_mbs_to_wcs (CHAR_T *dest,
1193 const unsigned char* src,
1195 /* The length of multibyte string. */
1198 /* Correspondences between src(char string) and
1199 dest(wchar_t string) for optimization. E.g.:
1201 dest = {'X', 'Y', 'Z'}
1202 (each "xxx", "y" and "zz" represent one
1203 multibyte character corresponding to 'X',
1205 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"),
1212 wchar_t *pdest = dest;
1213 const unsigned char *psrc = src;
1214 size_t wc_count = 0;
1218 size_t mb_remain = len;
1219 size_t mb_count = 0;
1221 /* Initialize the conversion state. */
1222 memset (&mbs, 0, sizeof (mbstate_t));
1224 offset_buffer[0] = 0;
1225 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed,
1228 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs);
1231 /* failed to convert. maybe src contains binary data.
1232 So we consume 1 byte manualy. */
1236 is_binary[wc_count] = TRUE;
1239 is_binary[wc_count] = FALSE;
1240 /* In sjis encoding, we use yen sign as escape character in
1241 place of reverse solidus. So we convert 0x5c(yen sign in
1242 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1243 solidus in UCS2). */
1244 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5)
1245 *pdest = (wchar_t) *psrc;
1247 offset_buffer[wc_count + 1] = mb_count += consumed;
1250 /* Fill remain of the buffer with sentinel. */
1251 for (i = wc_count + 1 ; i <= len ; i++)
1252 offset_buffer[i] = mb_count + 1;
1259 #else /* not INSIDE_RECURSION */
1261 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1262 also be assigned to arbitrarily: each pattern buffer stores its own
1263 syntax, so it can be changed between regex compilations. */
1264 /* This has no initializer because initialized variables in Emacs
1265 become read-only after dumping. */
1266 reg_syntax_t re_syntax_options;
1269 /* Specify the precise syntax of regexps for compilation. This provides
1270 for compatibility for various utilities which historically have
1271 different, incompatible syntaxes.
1273 The argument SYNTAX is a bit mask comprised of the various bits
1274 defined in regex.h. We return the old syntax. */
1277 re_set_syntax (reg_syntax_t syntax)
1279 reg_syntax_t ret = re_syntax_options;
1281 re_syntax_options = syntax;
1283 if (syntax & RE_DEBUG)
1285 else if (debug) /* was on but now is not */
1291 weak_alias (__re_set_syntax, re_set_syntax)
1294 /* This table gives an error message for each of the error codes listed
1295 in regex.h. Obviously the order here has to be same as there.
1296 POSIX doesn't require that we do anything for REG_NOERROR,
1297 but why not be nice? */
1299 static const char re_error_msgid[] =
1301 # define REG_NOERROR_IDX 0
1302 gettext_noop ("Success") /* REG_NOERROR */
1304 # define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1305 gettext_noop ("No match") /* REG_NOMATCH */
1307 # define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1308 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1310 # define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1311 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1313 # define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1314 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1316 # define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1317 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1319 # define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1320 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1322 # define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1323 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1325 # define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1326 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1328 # define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1329 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1331 # define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1332 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1334 # define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1335 gettext_noop ("Invalid range end") /* REG_ERANGE */
1337 # define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1338 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1340 # define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1341 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1343 # define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1344 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1346 # define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1347 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1349 # define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1350 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1353 static const size_t re_error_msgid_idx[] =
1374 #endif /* INSIDE_RECURSION */
1376 #ifndef DEFINED_ONCE
1377 /* Avoiding alloca during matching, to placate r_alloc. */
1379 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1380 searching and matching functions should not call alloca. On some
1381 systems, alloca is implemented in terms of malloc, and if we're
1382 using the relocating allocator routines, then malloc could cause a
1383 relocation, which might (if the strings being searched are in the
1384 ralloc heap) shift the data out from underneath the regexp
1387 Here's another reason to avoid allocation: Emacs
1388 processes input from X in a signal handler; processing X input may
1389 call malloc; if input arrives while a matching routine is calling
1390 malloc, then we're scrod. But Emacs can't just block input while
1391 calling matching routines; then we don't notice interrupts when
1392 they come in. So, Emacs blocks input around all regexp calls
1393 except the matching calls, which it leaves unprotected, in the
1394 faith that they will not malloc. */
1396 /* Normally, this is fine. */
1397 # define MATCH_MAY_ALLOCATE
1399 /* When using GNU C, we are not REALLY using the C alloca, no matter
1400 what config.h may say. So don't take precautions for it. */
1405 /* The match routines may not allocate if (1) they would do it with malloc
1406 and (2) it's not safe for them to use malloc.
1407 Note that if REL_ALLOC is defined, matching would not use malloc for the
1408 failure stack, but we would still use it for the register vectors;
1409 so REL_ALLOC should not affect this. */
1410 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1411 # undef MATCH_MAY_ALLOCATE
1413 #endif /* not DEFINED_ONCE */
1415 #ifdef INSIDE_RECURSION
1416 /* Failure stack declarations and macros; both re_compile_fastmap and
1417 re_match_2 use a failure stack. These have to be macros because of
1418 REGEX_ALLOCATE_STACK. */
1421 /* Number of failure points for which to initially allocate space
1422 when matching. If this number is exceeded, we allocate more
1423 space, so it is not a hard limit. */
1424 # ifndef INIT_FAILURE_ALLOC
1425 # define INIT_FAILURE_ALLOC 5
1428 /* Roughly the maximum number of failure points on the stack. Would be
1429 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1430 This is a variable only so users of regex can assign to it; we never
1431 change it ourselves. */
1433 # ifdef INT_IS_16BIT
1435 # ifndef DEFINED_ONCE
1436 # if defined MATCH_MAY_ALLOCATE
1437 /* 4400 was enough to cause a crash on Alpha OSF/1,
1438 whose default stack limit is 2mb. */
1439 long int re_max_failures = 4000;
1441 long int re_max_failures = 2000;
1445 union PREFIX(fail_stack_elt)
1451 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1455 PREFIX(fail_stack_elt_t) *stack;
1456 unsigned long int size;
1457 unsigned long int avail; /* Offset of next open position. */
1458 } PREFIX(fail_stack_type);
1460 # else /* not INT_IS_16BIT */
1462 # ifndef DEFINED_ONCE
1463 # if defined MATCH_MAY_ALLOCATE
1464 /* 4400 was enough to cause a crash on Alpha OSF/1,
1465 whose default stack limit is 2mb. */
1466 int re_max_failures = 4000;
1468 int re_max_failures = 2000;
1472 union PREFIX(fail_stack_elt)
1478 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1482 PREFIX(fail_stack_elt_t) *stack;
1484 unsigned avail; /* Offset of next open position. */
1485 } PREFIX(fail_stack_type);
1487 # endif /* INT_IS_16BIT */
1489 # ifndef DEFINED_ONCE
1490 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1491 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1492 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1496 /* Define macros to initialize and free the failure stack.
1497 Do `return -2' if the alloc fails. */
1499 # ifdef MATCH_MAY_ALLOCATE
1500 # define INIT_FAIL_STACK() \
1502 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1503 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1505 if (fail_stack.stack == NULL) \
1508 fail_stack.size = INIT_FAILURE_ALLOC; \
1509 fail_stack.avail = 0; \
1512 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1514 # define INIT_FAIL_STACK() \
1516 fail_stack.avail = 0; \
1519 # define RESET_FAIL_STACK()
1523 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1525 Return 1 if succeeds, and 0 if either ran out of memory
1526 allocating space for it or it was already too large.
1528 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1530 # define DOUBLE_FAIL_STACK(fail_stack) \
1531 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1533 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1534 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1535 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1536 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1538 (fail_stack).stack == NULL \
1540 : ((fail_stack).size <<= 1, \
1544 /* Push pointer POINTER on FAIL_STACK.
1545 Return 1 if was able to do so and 0 if ran out of memory allocating
1547 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1548 ((FAIL_STACK_FULL () \
1549 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1551 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1554 /* Push a pointer value 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_POINTER(item) \
1558 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1560 /* This pushes an integer-valued item 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_INT(item) \
1564 fail_stack.stack[fail_stack.avail++].integer = (item)
1566 /* Push a fail_stack_elt_t value onto the failure stack.
1567 Assumes the variable `fail_stack'. Probably should only
1568 be called from within `PUSH_FAILURE_POINT'. */
1569 # define PUSH_FAILURE_ELT(item) \
1570 fail_stack.stack[fail_stack.avail++] = (item)
1572 /* These three POP... operations complement the three PUSH... operations.
1573 All assume that `fail_stack' is nonempty. */
1574 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1575 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1576 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1578 /* Used to omit pushing failure point id's when we're not debugging. */
1580 # define DEBUG_PUSH PUSH_FAILURE_INT
1581 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1583 # define DEBUG_PUSH(item)
1584 # define DEBUG_POP(item_addr)
1588 /* Push the information about the state we will need
1589 if we ever fail back to it.
1591 Requires variables fail_stack, regstart, regend, reg_info, and
1592 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1595 Does `return FAILURE_CODE' if runs out of memory. */
1597 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1599 char *destination; \
1600 /* Must be int, so when we don't save any registers, the arithmetic \
1601 of 0 + -1 isn't done as unsigned. */ \
1602 /* Can't be int, since there is not a shred of a guarantee that int \
1603 is wide enough to hold a value of something to which pointer can \
1605 active_reg_t this_reg; \
1607 DEBUG_STATEMENT (failure_id++); \
1608 DEBUG_STATEMENT (nfailure_points_pushed++); \
1609 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1610 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1611 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1613 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1614 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1616 /* Ensure we have enough space allocated for what we will push. */ \
1617 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1619 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1620 return failure_code; \
1622 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1623 (fail_stack).size); \
1624 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1627 /* Push the info, starting with the registers. */ \
1628 DEBUG_PRINT1 ("\n"); \
1631 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1634 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1635 DEBUG_STATEMENT (num_regs_pushed++); \
1637 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1638 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1640 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1641 PUSH_FAILURE_POINTER (regend[this_reg]); \
1643 DEBUG_PRINT2 (" info: %p\n ", \
1644 reg_info[this_reg].word.pointer); \
1645 DEBUG_PRINT2 (" match_null=%d", \
1646 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1647 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1648 DEBUG_PRINT2 (" matched_something=%d", \
1649 MATCHED_SOMETHING (reg_info[this_reg])); \
1650 DEBUG_PRINT2 (" ever_matched=%d", \
1651 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1652 DEBUG_PRINT1 ("\n"); \
1653 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1656 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1657 PUSH_FAILURE_INT (lowest_active_reg); \
1659 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1660 PUSH_FAILURE_INT (highest_active_reg); \
1662 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1663 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1664 PUSH_FAILURE_POINTER (pattern_place); \
1666 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1667 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1669 DEBUG_PRINT1 ("'\n"); \
1670 PUSH_FAILURE_POINTER (string_place); \
1672 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1673 DEBUG_PUSH (failure_id); \
1676 # ifndef DEFINED_ONCE
1677 /* This is the number of items that are pushed and popped on the stack
1678 for each register. */
1679 # define NUM_REG_ITEMS 3
1681 /* Individual items aside from the registers. */
1683 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1685 # define NUM_NONREG_ITEMS 4
1688 /* We push at most this many items on the stack. */
1689 /* We used to use (num_regs - 1), which is the number of registers
1690 this regexp will save; but that was changed to 5
1691 to avoid stack overflow for a regexp with lots of parens. */
1692 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1694 /* We actually push this many items. */
1695 # define NUM_FAILURE_ITEMS \
1697 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1701 /* How many items can still be added to the stack without overflowing it. */
1702 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1703 # endif /* not DEFINED_ONCE */
1706 /* Pops what PUSH_FAIL_STACK pushes.
1708 We restore into the parameters, all of which should be lvalues:
1709 STR -- the saved data position.
1710 PAT -- the saved pattern position.
1711 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1712 REGSTART, REGEND -- arrays of string positions.
1713 REG_INFO -- array of information about each subexpression.
1715 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1716 `pend', `string1', `size1', `string2', and `size2'. */
1717 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1719 DEBUG_STATEMENT (unsigned failure_id;) \
1720 active_reg_t this_reg; \
1721 const UCHAR_T *string_temp; \
1723 assert (!FAIL_STACK_EMPTY ()); \
1725 /* Remove failure points and point to how many regs pushed. */ \
1726 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1727 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1728 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1730 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1732 DEBUG_POP (&failure_id); \
1733 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1735 /* If the saved string location is NULL, it came from an \
1736 on_failure_keep_string_jump opcode, and we want to throw away the \
1737 saved NULL, thus retaining our current position in the string. */ \
1738 string_temp = POP_FAILURE_POINTER (); \
1739 if (string_temp != NULL) \
1740 str = (const CHAR_T *) string_temp; \
1742 DEBUG_PRINT2 (" Popping string %p: `", str); \
1743 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1744 DEBUG_PRINT1 ("'\n"); \
1746 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1747 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1748 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1750 /* Restore register info. */ \
1751 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1752 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1754 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1755 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1758 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1760 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1762 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1763 DEBUG_PRINT2 (" info: %p\n", \
1764 reg_info[this_reg].word.pointer); \
1766 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1767 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1769 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1770 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1774 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1776 reg_info[this_reg].word.integer = 0; \
1777 regend[this_reg] = 0; \
1778 regstart[this_reg] = 0; \
1780 highest_active_reg = high_reg; \
1783 set_regs_matched_done = 0; \
1784 DEBUG_STATEMENT (nfailure_points_popped++); \
1785 } /* POP_FAILURE_POINT */
1787 /* Structure for per-register (a.k.a. per-group) information.
1788 Other register information, such as the
1789 starting and ending positions (which are addresses), and the list of
1790 inner groups (which is a bits list) are maintained in separate
1793 We are making a (strictly speaking) nonportable assumption here: that
1794 the compiler will pack our bit fields into something that fits into
1795 the type of `word', i.e., is something that fits into one item on the
1799 /* Declarations and macros for re_match_2. */
1803 PREFIX(fail_stack_elt_t) word;
1806 /* This field is one if this group can match the empty string,
1807 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1808 # define MATCH_NULL_UNSET_VALUE 3
1809 unsigned match_null_string_p : 2;
1810 unsigned is_active : 1;
1811 unsigned matched_something : 1;
1812 unsigned ever_matched_something : 1;
1814 } PREFIX(register_info_type);
1816 # ifndef DEFINED_ONCE
1817 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1818 # define IS_ACTIVE(R) ((R).bits.is_active)
1819 # define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1820 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1823 /* Call this when have matched a real character; it sets `matched' flags
1824 for the subexpressions which we are currently inside. Also records
1825 that those subexprs have matched. */
1826 # define SET_REGS_MATCHED() \
1829 if (!set_regs_matched_done) \
1832 set_regs_matched_done = 1; \
1833 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1835 MATCHED_SOMETHING (reg_info[r]) \
1836 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1842 # endif /* not DEFINED_ONCE */
1844 /* Registers are set to a sentinel when they haven't yet matched. */
1845 static CHAR_T PREFIX(reg_unset_dummy);
1846 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1847 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1849 /* Subroutine declarations and macros for regex_compile. */
1850 static void PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg);
1851 static void PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc,
1852 int arg1, int arg2);
1853 static void PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc,
1854 int arg, UCHAR_T *end);
1855 static void PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc,
1856 int arg1, int arg2, UCHAR_T *end);
1857 static boolean PREFIX(at_begline_loc_p) (const CHAR_T *pattern,
1859 reg_syntax_t syntax);
1860 static boolean PREFIX(at_endline_loc_p) (const CHAR_T *p,
1862 reg_syntax_t syntax);
1864 static reg_errcode_t wcs_compile_range (CHAR_T range_start,
1865 const CHAR_T **p_ptr,
1868 reg_syntax_t syntax,
1871 static void insert_space (int num, CHAR_T *loc, CHAR_T *end);
1873 static reg_errcode_t byte_compile_range (unsigned int range_start,
1877 reg_syntax_t syntax,
1881 /* Fetch the next character in the uncompiled pattern---translating it
1882 if necessary. Also cast from a signed character in the constant
1883 string passed to us by the user to an unsigned char that we can use
1884 as an array index (in, e.g., `translate'). */
1885 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1886 because it is impossible to allocate 4GB array for some encodings
1887 which have 4 byte character_set like UCS4. */
1890 # define PATFETCH(c) \
1891 do {if (p == pend) return REG_EEND; \
1892 c = (UCHAR_T) *p++; \
1893 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1896 # define PATFETCH(c) \
1897 do {if (p == pend) return REG_EEND; \
1898 c = (unsigned char) *p++; \
1899 if (translate) c = (unsigned char) translate[c]; \
1904 /* Fetch the next character in the uncompiled pattern, with no
1906 # define PATFETCH_RAW(c) \
1907 do {if (p == pend) return REG_EEND; \
1908 c = (UCHAR_T) *p++; \
1911 /* Go backwards one character in the pattern. */
1912 # define PATUNFETCH p--
1915 /* If `translate' is non-null, return translate[D], else just D. We
1916 cast the subscript to translate because some data is declared as
1917 `char *', to avoid warnings when a string constant is passed. But
1918 when we use a character as a subscript we must make it unsigned. */
1919 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1920 because it is impossible to allocate 4GB array for some encodings
1921 which have 4 byte character_set like UCS4. */
1925 # define TRANSLATE(d) \
1926 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1927 ? (char) translate[(unsigned char) (d)] : (d))
1929 # define TRANSLATE(d) \
1930 (translate ? (char) translate[(unsigned char) (d)] : (d))
1935 /* Macros for outputting the compiled pattern into `buffer'. */
1937 /* If the buffer isn't allocated when it comes in, use this. */
1938 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
1940 /* Make sure we have at least N more bytes of space in buffer. */
1942 # define GET_BUFFER_SPACE(n) \
1943 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
1944 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
1947 # define GET_BUFFER_SPACE(n) \
1948 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1952 /* Make sure we have one more byte of buffer space and then add C to it. */
1953 # define BUF_PUSH(c) \
1955 GET_BUFFER_SPACE (1); \
1956 *b++ = (UCHAR_T) (c); \
1960 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1961 # define BUF_PUSH_2(c1, c2) \
1963 GET_BUFFER_SPACE (2); \
1964 *b++ = (UCHAR_T) (c1); \
1965 *b++ = (UCHAR_T) (c2); \
1969 /* As with BUF_PUSH_2, except for three bytes. */
1970 # define BUF_PUSH_3(c1, c2, c3) \
1972 GET_BUFFER_SPACE (3); \
1973 *b++ = (UCHAR_T) (c1); \
1974 *b++ = (UCHAR_T) (c2); \
1975 *b++ = (UCHAR_T) (c3); \
1978 /* Store a jump with opcode OP at LOC to location TO. We store a
1979 relative address offset by the three bytes the jump itself occupies. */
1980 # define STORE_JUMP(op, loc, to) \
1981 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
1983 /* Likewise, for a two-argument jump. */
1984 # define STORE_JUMP2(op, loc, to, arg) \
1985 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
1987 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1988 # define INSERT_JUMP(op, loc, to) \
1989 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
1991 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1992 # define INSERT_JUMP2(op, loc, to, arg) \
1993 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
1996 /* This is not an arbitrary limit: the arguments which represent offsets
1997 into the pattern are two bytes long. So if 2^16 bytes turns out to
1998 be too small, many things would have to change. */
1999 /* Any other compiler which, like MSC, has allocation limit below 2^16
2000 bytes will have to use approach similar to what was done below for
2001 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2002 reallocating to 0 bytes. Such thing is not going to work too well.
2003 You have been warned!! */
2004 # ifndef DEFINED_ONCE
2005 # if defined _MSC_VER && !defined WIN32
2006 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2007 The REALLOC define eliminates a flurry of conversion warnings,
2008 but is not required. */
2009 # define MAX_BUF_SIZE 65500L
2010 # define REALLOC(p,s) realloc ((p), (size_t) (s))
2012 # define MAX_BUF_SIZE (1L << 16)
2013 # define REALLOC(p,s) realloc ((p), (s))
2016 /* Extend the buffer by twice its current size via realloc and
2017 reset the pointers that pointed into the old block to point to the
2018 correct places in the new one. If extending the buffer results in it
2019 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2020 # if __BOUNDED_POINTERS__
2021 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2022 # define MOVE_BUFFER_POINTER(P) \
2023 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2024 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2027 SET_HIGH_BOUND (b); \
2028 SET_HIGH_BOUND (begalt); \
2029 if (fixup_alt_jump) \
2030 SET_HIGH_BOUND (fixup_alt_jump); \
2032 SET_HIGH_BOUND (laststart); \
2033 if (pending_exact) \
2034 SET_HIGH_BOUND (pending_exact); \
2037 # define MOVE_BUFFER_POINTER(P) (P) += incr
2038 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
2040 # endif /* not DEFINED_ONCE */
2043 # define EXTEND_BUFFER() \
2045 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2047 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2049 bufp->allocated <<= 1; \
2050 if (bufp->allocated > MAX_BUF_SIZE) \
2051 bufp->allocated = MAX_BUF_SIZE; \
2052 /* How many characters the new buffer can have? */ \
2053 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2054 if (wchar_count == 0) wchar_count = 1; \
2055 /* Truncate the buffer to CHAR_T align. */ \
2056 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2057 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2058 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2059 if (COMPILED_BUFFER_VAR == NULL) \
2060 return REG_ESPACE; \
2061 /* If the buffer moved, move all the pointers into it. */ \
2062 if (old_buffer != COMPILED_BUFFER_VAR) \
2064 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2065 MOVE_BUFFER_POINTER (b); \
2066 MOVE_BUFFER_POINTER (begalt); \
2067 if (fixup_alt_jump) \
2068 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2070 MOVE_BUFFER_POINTER (laststart); \
2071 if (pending_exact) \
2072 MOVE_BUFFER_POINTER (pending_exact); \
2074 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2077 # define EXTEND_BUFFER() \
2079 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2080 if (bufp->allocated == MAX_BUF_SIZE) \
2082 bufp->allocated <<= 1; \
2083 if (bufp->allocated > MAX_BUF_SIZE) \
2084 bufp->allocated = MAX_BUF_SIZE; \
2086 = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, bufp->allocated); \
2087 if (COMPILED_BUFFER_VAR == NULL) \
2088 return REG_ESPACE; \
2089 /* If the buffer moved, move all the pointers into it. */ \
2090 if (old_buffer != COMPILED_BUFFER_VAR) \
2092 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2093 MOVE_BUFFER_POINTER (b); \
2094 MOVE_BUFFER_POINTER (begalt); \
2095 if (fixup_alt_jump) \
2096 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2098 MOVE_BUFFER_POINTER (laststart); \
2099 if (pending_exact) \
2100 MOVE_BUFFER_POINTER (pending_exact); \
2102 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2106 # ifndef DEFINED_ONCE
2107 /* Since we have one byte reserved for the register number argument to
2108 {start,stop}_memory, the maximum number of groups we can report
2109 things about is what fits in that byte. */
2110 # define MAX_REGNUM 255
2112 /* But patterns can have more than `MAX_REGNUM' registers. We just
2113 ignore the excess. */
2114 typedef unsigned regnum_t;
2117 /* Macros for the compile stack. */
2119 /* Since offsets can go either forwards or backwards, this type needs to
2120 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2121 /* int may be not enough when sizeof(int) == 2. */
2122 typedef long pattern_offset_t;
2126 pattern_offset_t begalt_offset;
2127 pattern_offset_t fixup_alt_jump;
2128 pattern_offset_t inner_group_offset;
2129 pattern_offset_t laststart_offset;
2131 } compile_stack_elt_t;
2136 compile_stack_elt_t *stack;
2138 unsigned avail; /* Offset of next open position. */
2139 } compile_stack_type;
2142 # define INIT_COMPILE_STACK_SIZE 32
2144 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2145 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2147 /* The next available element. */
2148 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2150 # endif /* not DEFINED_ONCE */
2152 /* Set the bit for character C in a list. */
2153 # ifndef DEFINED_ONCE
2154 # define SET_LIST_BIT(c) \
2155 (b[((unsigned char) (c)) / BYTEWIDTH] \
2156 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2157 # endif /* DEFINED_ONCE */
2159 /* Get the next unsigned number in the uncompiled pattern. */
2160 # define GET_UNSIGNED_NUMBER(num) \
2165 if (c < '0' || c > '9') \
2167 if (num <= RE_DUP_MAX) \
2171 num = num * 10 + c - '0'; \
2176 # ifndef DEFINED_ONCE
2177 # if defined _LIBC || WIDE_CHAR_SUPPORT
2178 /* The GNU C library provides support for user-defined character classes
2179 and the functions from ISO C amendement 1. */
2180 # ifdef CHARCLASS_NAME_MAX
2181 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2183 /* This shouldn't happen but some implementation might still have this
2184 problem. Use a reasonable default value. */
2185 # define CHAR_CLASS_MAX_LENGTH 256
2189 # define IS_CHAR_CLASS(string) __wctype (string)
2191 # define IS_CHAR_CLASS(string) wctype (string)
2194 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2196 # define IS_CHAR_CLASS(string) \
2197 (STREQ (string, "alpha") || STREQ (string, "upper") \
2198 || STREQ (string, "lower") || STREQ (string, "digit") \
2199 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2200 || STREQ (string, "space") || STREQ (string, "print") \
2201 || STREQ (string, "punct") || STREQ (string, "graph") \
2202 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2204 # endif /* DEFINED_ONCE */
2206 # ifndef MATCH_MAY_ALLOCATE
2208 /* If we cannot allocate large objects within re_match_2_internal,
2209 we make the fail stack and register vectors global.
2210 The fail stack, we grow to the maximum size when a regexp
2212 The register vectors, we adjust in size each time we
2213 compile a regexp, according to the number of registers it needs. */
2215 static PREFIX(fail_stack_type) fail_stack;
2217 /* Size with which the following vectors are currently allocated.
2218 That is so we can make them bigger as needed,
2219 but never make them smaller. */
2220 # ifdef DEFINED_ONCE
2221 static int regs_allocated_size;
2223 static const char ** regstart, ** regend;
2224 static const char ** old_regstart, ** old_regend;
2225 static const char **best_regstart, **best_regend;
2226 static const char **reg_dummy;
2227 # endif /* DEFINED_ONCE */
2229 static PREFIX(register_info_type) *PREFIX(reg_info);
2230 static PREFIX(register_info_type) *PREFIX(reg_info_dummy);
2232 /* Make the register vectors big enough for NUM_REGS registers,
2233 but don't make them smaller. */
2236 PREFIX(regex_grow_registers) (int num_regs)
2238 if (num_regs > regs_allocated_size)
2240 RETALLOC_IF (regstart, num_regs, const char *);
2241 RETALLOC_IF (regend, num_regs, const char *);
2242 RETALLOC_IF (old_regstart, num_regs, const char *);
2243 RETALLOC_IF (old_regend, num_regs, const char *);
2244 RETALLOC_IF (best_regstart, num_regs, const char *);
2245 RETALLOC_IF (best_regend, num_regs, const char *);
2246 RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type));
2247 RETALLOC_IF (reg_dummy, num_regs, const char *);
2248 RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type));
2250 regs_allocated_size = num_regs;
2254 # endif /* not MATCH_MAY_ALLOCATE */
2256 # ifndef DEFINED_ONCE
2257 static boolean group_in_compile_stack (compile_stack_type
2260 # endif /* not DEFINED_ONCE */
2262 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2263 Returns one of error codes defined in `regex.h', or zero for success.
2265 Assumes the `allocated' (and perhaps `buffer') and `translate'
2266 fields are set in BUFP on entry.
2268 If it succeeds, results are put in BUFP (if it returns an error, the
2269 contents of BUFP are undefined):
2270 `buffer' is the compiled pattern;
2271 `syntax' is set to SYNTAX;
2272 `used' is set to the length of the compiled pattern;
2273 `fastmap_accurate' is zero;
2274 `re_nsub' is the number of subexpressions in PATTERN;
2275 `not_bol' and `not_eol' are zero;
2277 The `fastmap' and `newline_anchor' fields are neither
2278 examined nor set. */
2280 /* Return, freeing storage we allocated. */
2282 # define FREE_STACK_RETURN(value) \
2283 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2285 # define FREE_STACK_RETURN(value) \
2286 return (free (compile_stack.stack), value)
2289 static reg_errcode_t
2290 PREFIX(regex_compile) (const char *ARG_PREFIX(pattern),
2291 size_t ARG_PREFIX(size),
2292 reg_syntax_t syntax,
2293 struct re_pattern_buffer *bufp)
2295 /* We fetch characters from PATTERN here. Even though PATTERN is
2296 `char *' (i.e., signed), we declare these variables as unsigned, so
2297 they can be reliably used as array indices. */
2298 register UCHAR_T c, c1;
2301 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2302 CHAR_T *pattern, *COMPILED_BUFFER_VAR;
2304 /* offset buffer for optimization. See convert_mbs_to_wc. */
2305 int *mbs_offset = NULL;
2306 /* It hold whether each wchar_t is binary data or not. */
2307 char *is_binary = NULL;
2308 /* A flag whether exactn is handling binary data or not. */
2309 char is_exactn_bin = FALSE;
2312 /* A random temporary spot in PATTERN. */
2315 /* Points to the end of the buffer, where we should append. */
2316 register UCHAR_T *b;
2318 /* Keeps track of unclosed groups. */
2319 compile_stack_type compile_stack;
2321 /* Points to the current (ending) position in the pattern. */
2326 const CHAR_T *p = pattern;
2327 const CHAR_T *pend = pattern + size;
2330 /* How to translate the characters in the pattern. */
2331 RE_TRANSLATE_TYPE translate = bufp->translate;
2333 /* Address of the count-byte of the most recently inserted `exactn'
2334 command. This makes it possible to tell if a new exact-match
2335 character can be added to that command or if the character requires
2336 a new `exactn' command. */
2337 UCHAR_T *pending_exact = 0;
2339 /* Address of start of the most recently finished expression.
2340 This tells, e.g., postfix * where to find the start of its
2341 operand. Reset at the beginning of groups and alternatives. */
2342 UCHAR_T *laststart = 0;
2344 /* Address of beginning of regexp, or inside of last group. */
2347 /* Address of the place where a forward jump should go to the end of
2348 the containing expression. Each alternative of an `or' -- except the
2349 last -- ends with a forward jump of this sort. */
2350 UCHAR_T *fixup_alt_jump = 0;
2352 /* Counts open-groups as they are encountered. Remembered for the
2353 matching close-group on the compile stack, so the same register
2354 number is put in the stop_memory as the start_memory. */
2355 regnum_t regnum = 0;
2358 /* Initialize the wchar_t PATTERN and offset_buffer. */
2359 p = pend = pattern = TALLOC(csize + 1, CHAR_T);
2360 mbs_offset = TALLOC(csize + 1, int);
2361 is_binary = TALLOC(csize + 1, char);
2362 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL)
2369 pattern[csize] = L'\0'; /* sentinel */
2370 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary);
2382 DEBUG_PRINT1 ("\nCompiling pattern: ");
2385 unsigned debug_count;
2387 for (debug_count = 0; debug_count < size; debug_count++)
2388 PUT_CHAR (pattern[debug_count]);
2393 /* Initialize the compile stack. */
2394 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2395 if (compile_stack.stack == NULL)
2405 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2406 compile_stack.avail = 0;
2408 /* Initialize the pattern buffer. */
2409 bufp->syntax = syntax;
2410 bufp->fastmap_accurate = 0;
2411 bufp->not_bol = bufp->not_eol = 0;
2413 /* Set `used' to zero, so that if we return an error, the pattern
2414 printer (for debugging) will think there's no pattern. We reset it
2418 /* Always count groups, whether or not bufp->no_sub is set. */
2421 #if !defined emacs && !defined SYNTAX_TABLE
2422 /* Initialize the syntax table. */
2423 init_syntax_once ();
2426 if (bufp->allocated == 0)
2429 { /* If zero allocated, but buffer is non-null, try to realloc
2430 enough space. This loses if buffer's address is bogus, but
2431 that is the user's responsibility. */
2433 /* Free bufp->buffer and allocate an array for wchar_t pattern
2436 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T),
2439 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T);
2443 { /* Caller did not allocate a buffer. Do it for them. */
2444 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T),
2448 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE);
2450 bufp->buffer = (char*)COMPILED_BUFFER_VAR;
2452 bufp->allocated = INIT_BUF_SIZE;
2456 COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer;
2459 begalt = b = COMPILED_BUFFER_VAR;
2461 /* Loop through the uncompiled pattern until we're at the end. */
2470 if ( /* If at start of pattern, it's an operator. */
2472 /* If context independent, it's an operator. */
2473 || syntax & RE_CONTEXT_INDEP_ANCHORS
2474 /* Otherwise, depends on what's come before. */
2475 || PREFIX(at_begline_loc_p) (pattern, p, syntax))
2485 if ( /* If at end of pattern, it's an operator. */
2487 /* If context independent, it's an operator. */
2488 || syntax & RE_CONTEXT_INDEP_ANCHORS
2489 /* Otherwise, depends on what's next. */
2490 || PREFIX(at_endline_loc_p) (p, pend, syntax))
2500 if ((syntax & RE_BK_PLUS_QM)
2501 || (syntax & RE_LIMITED_OPS))
2505 /* If there is no previous pattern... */
2508 if (syntax & RE_CONTEXT_INVALID_OPS)
2509 FREE_STACK_RETURN (REG_BADRPT);
2510 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2515 /* Are we optimizing this jump? */
2516 boolean keep_string_p = false;
2518 /* 1 means zero (many) matches is allowed. */
2519 char zero_times_ok = 0, many_times_ok = 0;
2521 /* If there is a sequence of repetition chars, collapse it
2522 down to just one (the right one). We can't combine
2523 interval operators with these because of, e.g., `a{2}*',
2524 which should only match an even number of `a's. */
2528 zero_times_ok |= c != '+';
2529 many_times_ok |= c != '?';
2537 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2540 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2542 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2545 if (!(c1 == '+' || c1 == '?'))
2560 /* If we get here, we found another repeat character. */
2563 /* Star, etc. applied to an empty pattern is equivalent
2564 to an empty pattern. */
2568 /* Now we know whether or not zero matches is allowed
2569 and also whether or not two or more matches is allowed. */
2571 { /* More than one repetition is allowed, so put in at the
2572 end a backward relative jump from `b' to before the next
2573 jump we're going to put in below (which jumps from
2574 laststart to after this jump).
2576 But if we are at the `*' in the exact sequence `.*\n',
2577 insert an unconditional jump backwards to the .,
2578 instead of the beginning of the loop. This way we only
2579 push a failure point once, instead of every time
2580 through the loop. */
2581 assert (p - 1 > pattern);
2583 /* Allocate the space for the jump. */
2584 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2586 /* We know we are not at the first character of the pattern,
2587 because laststart was nonzero. And we've already
2588 incremented `p', by the way, to be the character after
2589 the `*'. Do we have to do something analogous here
2590 for null bytes, because of RE_DOT_NOT_NULL? */
2591 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2593 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2594 && !(syntax & RE_DOT_NEWLINE))
2595 { /* We have .*\n. */
2596 STORE_JUMP (jump, b, laststart);
2597 keep_string_p = true;
2600 /* Anything else. */
2601 STORE_JUMP (maybe_pop_jump, b, laststart -
2602 (1 + OFFSET_ADDRESS_SIZE));
2604 /* We've added more stuff to the buffer. */
2605 b += 1 + OFFSET_ADDRESS_SIZE;
2608 /* On failure, jump from laststart to b + 3, which will be the
2609 end of the buffer after this jump is inserted. */
2610 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2612 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2613 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2615 laststart, b + 1 + OFFSET_ADDRESS_SIZE);
2617 b += 1 + OFFSET_ADDRESS_SIZE;
2621 /* At least one repetition is required, so insert a
2622 `dummy_failure_jump' before the initial
2623 `on_failure_jump' instruction of the loop. This
2624 effects a skip over that instruction the first time
2625 we hit that loop. */
2626 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2627 INSERT_JUMP (dummy_failure_jump, laststart, laststart +
2628 2 + 2 * OFFSET_ADDRESS_SIZE);
2629 b += 1 + OFFSET_ADDRESS_SIZE;
2643 boolean had_char_class = false;
2645 CHAR_T range_start = 0xffffffff;
2647 unsigned int range_start = 0xffffffff;
2649 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2652 /* We assume a charset(_not) structure as a wchar_t array.
2653 charset[0] = (re_opcode_t) charset(_not)
2654 charset[1] = l (= length of char_classes)
2655 charset[2] = m (= length of collating_symbols)
2656 charset[3] = n (= length of equivalence_classes)
2657 charset[4] = o (= length of char_ranges)
2658 charset[5] = p (= length of chars)
2660 charset[6] = char_class (wctype_t)
2661 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2663 charset[l+5] = char_class (wctype_t)
2665 charset[l+6] = collating_symbol (wchar_t)
2667 charset[l+m+5] = collating_symbol (wchar_t)
2668 ifdef _LIBC we use the index if
2669 _NL_COLLATE_SYMB_EXTRAMB instead of
2672 charset[l+m+6] = equivalence_classes (wchar_t)
2674 charset[l+m+n+5] = equivalence_classes (wchar_t)
2675 ifdef _LIBC we use the index in
2676 _NL_COLLATE_WEIGHT instead of
2679 charset[l+m+n+6] = range_start
2680 charset[l+m+n+7] = range_end
2682 charset[l+m+n+2o+4] = range_start
2683 charset[l+m+n+2o+5] = range_end
2684 ifdef _LIBC we use the value looked up
2685 in _NL_COLLATE_COLLSEQ instead of
2688 charset[l+m+n+2o+6] = char
2690 charset[l+m+n+2o+p+5] = char
2694 /* We need at least 6 spaces: the opcode, the length of
2695 char_classes, the length of collating_symbols, the length of
2696 equivalence_classes, the length of char_ranges, the length of
2698 GET_BUFFER_SPACE (6);
2700 /* Save b as laststart. And We use laststart as the pointer
2701 to the first element of the charset here.
2702 In other words, laststart[i] indicates charset[i]. */
2705 /* We test `*p == '^' twice, instead of using an if
2706 statement, so we only need one BUF_PUSH. */
2707 BUF_PUSH (*p == '^' ? charset_not : charset);
2711 /* Push the length of char_classes, the length of
2712 collating_symbols, the length of equivalence_classes, the
2713 length of char_ranges and the length of chars. */
2714 BUF_PUSH_3 (0, 0, 0);
2717 /* Remember the first position in the bracket expression. */
2720 /* charset_not matches newline according to a syntax bit. */
2721 if ((re_opcode_t) b[-6] == charset_not
2722 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2725 laststart[5]++; /* Update the length of characters */
2728 /* Read in characters and ranges, setting map bits. */
2731 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2735 /* \ might escape characters inside [...] and [^...]. */
2736 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2738 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2742 laststart[5]++; /* Update the length of chars */
2747 /* Could be the end of the bracket expression. If it's
2748 not (i.e., when the bracket expression is `[]' so
2749 far), the ']' character bit gets set way below. */
2750 if (c == ']' && p != p1 + 1)
2753 /* Look ahead to see if it's a range when the last thing
2754 was a character class. */
2755 if (had_char_class && c == '-' && *p != ']')
2756 FREE_STACK_RETURN (REG_ERANGE);
2758 /* Look ahead to see if it's a range when the last thing
2759 was a character: if this is a hyphen not at the
2760 beginning or the end of a list, then it's the range
2763 && !(p - 2 >= pattern && p[-2] == '[')
2764 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2768 /* Allocate the space for range_start and range_end. */
2769 GET_BUFFER_SPACE (2);
2770 /* Update the pointer to indicate end of buffer. */
2772 ret = wcs_compile_range (range_start, &p, pend, translate,
2773 syntax, b, laststart);
2774 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2775 range_start = 0xffffffff;
2777 else if (p[0] == '-' && p[1] != ']')
2778 { /* This handles ranges made up of characters only. */
2781 /* Move past the `-'. */
2783 /* Allocate the space for range_start and range_end. */
2784 GET_BUFFER_SPACE (2);
2785 /* Update the pointer to indicate end of buffer. */
2787 ret = wcs_compile_range (c, &p, pend, translate, syntax, b,
2789 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2790 range_start = 0xffffffff;
2793 /* See if we're at the beginning of a possible character
2795 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2796 { /* Leave room for the null. */
2797 char str[CHAR_CLASS_MAX_LENGTH + 1];
2802 /* If pattern is `[[:'. */
2803 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2808 if ((c == ':' && *p == ']') || p == pend)
2810 if (c1 < CHAR_CLASS_MAX_LENGTH)
2813 /* This is in any case an invalid class name. */
2818 /* If isn't a word bracketed by `[:' and `:]':
2819 undo the ending character, the letters, and leave
2820 the leading `:' and `[' (but store them as character). */
2821 if (c == ':' && *p == ']')
2826 /* Query the character class as wctype_t. */
2827 wt = IS_CHAR_CLASS (str);
2829 FREE_STACK_RETURN (REG_ECTYPE);
2831 /* Throw away the ] at the end of the character
2835 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2837 /* Allocate the space for character class. */
2838 GET_BUFFER_SPACE(CHAR_CLASS_SIZE);
2839 /* Update the pointer to indicate end of buffer. */
2840 b += CHAR_CLASS_SIZE;
2841 /* Move data which follow character classes
2842 not to violate the data. */
2843 insert_space(CHAR_CLASS_SIZE,
2844 laststart + 6 + laststart[1],
2846 alignedp = ((uintptr_t)(laststart + 6 + laststart[1])
2847 + __alignof__(wctype_t) - 1)
2848 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2849 /* Store the character class. */
2850 *((wctype_t*)alignedp) = wt;
2851 /* Update length of char_classes */
2852 laststart[1] += CHAR_CLASS_SIZE;
2854 had_char_class = true;
2863 laststart[5] += 2; /* Update the length of characters */
2865 had_char_class = false;
2868 else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '='
2871 CHAR_T str[128]; /* Should be large enough. */
2872 CHAR_T delim = *p; /* '=' or '.' */
2875 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2880 /* If pattern is `[[=' or '[[.'. */
2881 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2886 if ((c == delim && *p == ']') || p == pend)
2888 if (c1 < sizeof (str) - 1)
2891 /* This is in any case an invalid class name. */
2896 if (c == delim && *p == ']' && str[0] != '\0')
2898 unsigned int i, offset;
2899 /* If we have no collation data we use the default
2900 collation in which each character is in a class
2901 by itself. It also means that ASCII is the
2902 character set and therefore we cannot have character
2903 with more than one byte in the multibyte
2906 /* If not defined _LIBC, we push the name and
2907 `\0' for the sake of matching performance. */
2908 int datasize = c1 + 1;
2916 FREE_STACK_RETURN (REG_ECOLLATE);
2921 const int32_t *table;
2922 const int32_t *weights;
2923 const int32_t *extra;
2924 const int32_t *indirect;
2927 /* This #include defines a local function! */
2928 # include <locale/weightwc.h>
2932 /* We push the index for equivalence class. */
2935 table = (const int32_t *)
2936 _NL_CURRENT (LC_COLLATE,
2937 _NL_COLLATE_TABLEWC);
2938 weights = (const int32_t *)
2939 _NL_CURRENT (LC_COLLATE,
2940 _NL_COLLATE_WEIGHTWC);
2941 extra = (const int32_t *)
2942 _NL_CURRENT (LC_COLLATE,
2943 _NL_COLLATE_EXTRAWC);
2944 indirect = (const int32_t *)
2945 _NL_CURRENT (LC_COLLATE,
2946 _NL_COLLATE_INDIRECTWC);
2948 idx = findidx ((const wint_t**)&cp);
2949 if (idx == 0 || cp < (wint_t*) str + c1)
2950 /* This is no valid character. */
2951 FREE_STACK_RETURN (REG_ECOLLATE);
2953 str[0] = (wchar_t)idx;
2955 else /* delim == '.' */
2957 /* We push collation sequence value
2958 for collating symbol. */
2960 const int32_t *symb_table;
2961 const unsigned char *extra;
2968 /* We have to convert the name to a single-byte
2969 string. This is possible since the names
2970 consist of ASCII characters and the internal
2971 representation is UCS4. */
2972 for (i = 0; i < c1; ++i)
2973 char_str[i] = str[i];
2976 _NL_CURRENT_WORD (LC_COLLATE,
2977 _NL_COLLATE_SYMB_HASH_SIZEMB);
2978 symb_table = (const int32_t *)
2979 _NL_CURRENT (LC_COLLATE,
2980 _NL_COLLATE_SYMB_TABLEMB);
2981 extra = (const unsigned char *)
2982 _NL_CURRENT (LC_COLLATE,
2983 _NL_COLLATE_SYMB_EXTRAMB);
2985 /* Locate the character in the hashing table. */
2986 hash = elem_hash (char_str, c1);
2989 elem = hash % table_size;
2990 second = hash % (table_size - 2);
2991 while (symb_table[2 * elem] != 0)
2993 /* First compare the hashing value. */
2994 if (symb_table[2 * elem] == hash
2995 && c1 == extra[symb_table[2 * elem + 1]]
2996 && memcmp (char_str,
2997 &extra[symb_table[2 * elem + 1]
3000 /* Yep, this is the entry. */
3001 idx = symb_table[2 * elem + 1];
3002 idx += 1 + extra[idx];
3010 if (symb_table[2 * elem] != 0)
3012 /* Compute the index of the byte sequence
3014 idx += 1 + extra[idx];
3015 /* Adjust for the alignment. */
3016 idx = (idx + 3) & ~3;
3018 str[0] = (wchar_t) idx + 4;
3020 else if (symb_table[2 * elem] == 0 && c1 == 1)
3022 /* No valid character. Match it as a
3023 single byte character. */
3024 had_char_class = false;
3026 /* Update the length of characters */
3028 range_start = str[0];
3030 /* Throw away the ] at the end of the
3031 collating symbol. */
3033 /* exit from the switch block. */
3037 FREE_STACK_RETURN (REG_ECOLLATE);
3042 /* Throw away the ] at the end of the equivalence
3043 class (or collating symbol). */
3046 /* Allocate the space for the equivalence class
3047 (or collating symbol) (and '\0' if needed). */
3048 GET_BUFFER_SPACE(datasize);
3049 /* Update the pointer to indicate end of buffer. */
3053 { /* equivalence class */
3054 /* Calculate the offset of char_ranges,
3055 which is next to equivalence_classes. */
3056 offset = laststart[1] + laststart[2]
3059 insert_space(datasize, laststart + offset, b - 1);
3061 /* Write the equivalence_class and \0. */
3062 for (i = 0 ; i < datasize ; i++)
3063 laststart[offset + i] = str[i];
3065 /* Update the length of equivalence_classes. */
3066 laststart[3] += datasize;
3067 had_char_class = true;
3069 else /* delim == '.' */
3070 { /* collating symbol */
3071 /* Calculate the offset of the equivalence_classes,
3072 which is next to collating_symbols. */
3073 offset = laststart[1] + laststart[2] + 6;
3074 /* Insert space and write the collationg_symbol
3076 insert_space(datasize, laststart + offset, b-1);
3077 for (i = 0 ; i < datasize ; i++)
3078 laststart[offset + i] = str[i];
3080 /* In re_match_2_internal if range_start < -1, we
3081 assume -range_start is the offset of the
3082 collating symbol which is specified as
3083 the character of the range start. So we assign
3084 -(laststart[1] + laststart[2] + 6) to
3086 range_start = -(laststart[1] + laststart[2] + 6);
3087 /* Update the length of collating_symbol. */
3088 laststart[2] += datasize;
3089 had_char_class = false;
3099 laststart[5] += 2; /* Update the length of characters */
3100 range_start = delim;
3101 had_char_class = false;
3106 had_char_class = false;
3108 laststart[5]++; /* Update the length of characters */
3114 /* Ensure that we have enough space to push a charset: the
3115 opcode, the length count, and the bitset; 34 bytes in all. */
3116 GET_BUFFER_SPACE (34);
3120 /* We test `*p == '^' twice, instead of using an if
3121 statement, so we only need one BUF_PUSH. */
3122 BUF_PUSH (*p == '^' ? charset_not : charset);
3126 /* Remember the first position in the bracket expression. */
3129 /* Push the number of bytes in the bitmap. */
3130 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
3132 /* Clear the whole map. */
3133 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
3135 /* charset_not matches newline according to a syntax bit. */
3136 if ((re_opcode_t) b[-2] == charset_not
3137 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
3138 SET_LIST_BIT ('\n');
3140 /* Read in characters and ranges, setting map bits. */
3143 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3147 /* \ might escape characters inside [...] and [^...]. */
3148 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
3150 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3158 /* Could be the end of the bracket expression. If it's
3159 not (i.e., when the bracket expression is `[]' so
3160 far), the ']' character bit gets set way below. */
3161 if (c == ']' && p != p1 + 1)
3164 /* Look ahead to see if it's a range when the last thing
3165 was a character class. */
3166 if (had_char_class && c == '-' && *p != ']')
3167 FREE_STACK_RETURN (REG_ERANGE);
3169 /* Look ahead to see if it's a range when the last thing
3170 was a character: if this is a hyphen not at the
3171 beginning or the end of a list, then it's the range
3174 && !(p - 2 >= pattern && p[-2] == '[')
3175 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
3179 = byte_compile_range (range_start, &p, pend, translate,
3181 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3182 range_start = 0xffffffff;
3185 else if (p[0] == '-' && p[1] != ']')
3186 { /* This handles ranges made up of characters only. */
3189 /* Move past the `-'. */
3192 ret = byte_compile_range (c, &p, pend, translate, syntax, b);
3193 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3194 range_start = 0xffffffff;
3197 /* See if we're at the beginning of a possible character
3200 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3201 { /* Leave room for the null. */
3202 char str[CHAR_CLASS_MAX_LENGTH + 1];
3207 /* If pattern is `[[:'. */
3208 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3213 if ((c == ':' && *p == ']') || p == pend)
3215 if (c1 < CHAR_CLASS_MAX_LENGTH)
3218 /* This is in any case an invalid class name. */
3223 /* If isn't a word bracketed by `[:' and `:]':
3224 undo the ending character, the letters, and leave
3225 the leading `:' and `[' (but set bits for them). */
3226 if (c == ':' && *p == ']')
3228 # if defined _LIBC || WIDE_CHAR_SUPPORT
3229 boolean is_lower = STREQ (str, "lower");
3230 boolean is_upper = STREQ (str, "upper");
3234 wt = IS_CHAR_CLASS (str);
3236 FREE_STACK_RETURN (REG_ECTYPE);
3238 /* Throw away the ] at the end of the character
3242 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3244 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
3246 if (iswctype (btowc (ch), wt))
3249 if (translate && (is_upper || is_lower)
3250 && (ISUPPER (ch) || ISLOWER (ch)))
3254 had_char_class = true;
3257 boolean is_alnum = STREQ (str, "alnum");
3258 boolean is_alpha = STREQ (str, "alpha");
3259 boolean is_blank = STREQ (str, "blank");
3260 boolean is_cntrl = STREQ (str, "cntrl");
3261 boolean is_digit = STREQ (str, "digit");
3262 boolean is_graph = STREQ (str, "graph");
3263 boolean is_lower = STREQ (str, "lower");
3264 boolean is_print = STREQ (str, "print");
3265 boolean is_punct = STREQ (str, "punct");
3266 boolean is_space = STREQ (str, "space");
3267 boolean is_upper = STREQ (str, "upper");
3268 boolean is_xdigit = STREQ (str, "xdigit");
3270 if (!IS_CHAR_CLASS (str))
3271 FREE_STACK_RETURN (REG_ECTYPE);
3273 /* Throw away the ] at the end of the character
3277 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3279 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
3281 /* This was split into 3 if's to
3282 avoid an arbitrary limit in some compiler. */
3283 if ( (is_alnum && ISALNUM (ch))
3284 || (is_alpha && ISALPHA (ch))
3285 || (is_blank && ISBLANK (ch))
3286 || (is_cntrl && ISCNTRL (ch)))
3288 if ( (is_digit && ISDIGIT (ch))
3289 || (is_graph && ISGRAPH (ch))
3290 || (is_lower && ISLOWER (ch))
3291 || (is_print && ISPRINT (ch)))
3293 if ( (is_punct && ISPUNCT (ch))
3294 || (is_space && ISSPACE (ch))
3295 || (is_upper && ISUPPER (ch))
3296 || (is_xdigit && ISXDIGIT (ch)))
3298 if ( translate && (is_upper || is_lower)
3299 && (ISUPPER (ch) || ISLOWER (ch)))
3302 had_char_class = true;
3303 # endif /* libc || wctype.h */
3313 had_char_class = false;
3316 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
3318 unsigned char str[MB_LEN_MAX + 1];
3321 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3327 /* If pattern is `[[='. */
3328 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3333 if ((c == '=' && *p == ']') || p == pend)
3335 if (c1 < MB_LEN_MAX)
3338 /* This is in any case an invalid class name. */
3343 if (c == '=' && *p == ']' && str[0] != '\0')
3345 /* If we have no collation data we use the default
3346 collation in which each character is in a class
3347 by itself. It also means that ASCII is the
3348 character set and therefore we cannot have character
3349 with more than one byte in the multibyte
3356 FREE_STACK_RETURN (REG_ECOLLATE);
3358 /* Throw away the ] at the end of the equivalence
3362 /* Set the bit for the character. */
3363 SET_LIST_BIT (str[0]);
3368 /* Try to match the byte sequence in `str' against
3369 those known to the collate implementation.
3370 First find out whether the bytes in `str' are
3371 actually from exactly one character. */
3372 const int32_t *table;
3373 const unsigned char *weights;
3374 const unsigned char *extra;
3375 const int32_t *indirect;
3377 const unsigned char *cp = str;
3380 /* This #include defines a local function! */
3381 # include <locale/weight.h>
3383 table = (const int32_t *)
3384 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3385 weights = (const unsigned char *)
3386 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3387 extra = (const unsigned char *)
3388 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3389 indirect = (const int32_t *)
3390 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3392 idx = findidx (&cp);
3393 if (idx == 0 || cp < str + c1)
3394 /* This is no valid character. */
3395 FREE_STACK_RETURN (REG_ECOLLATE);
3397 /* Throw away the ] at the end of the equivalence
3401 /* Now we have to go throught the whole table
3402 and find all characters which have the same
3405 XXX Note that this is not entirely correct.
3406 we would have to match multibyte sequences
3407 but this is not possible with the current
3409 for (ch = 1; ch < 256; ++ch)
3410 /* XXX This test would have to be changed if we
3411 would allow matching multibyte sequences. */
3414 int32_t idx2 = table[ch];
3415 size_t len = weights[idx2];
3417 /* Test whether the lenghts match. */
3418 if (weights[idx] == len)
3420 /* They do. New compare the bytes of
3425 && (weights[idx + 1 + cnt]
3426 == weights[idx2 + 1 + cnt]))
3430 /* They match. Mark the character as
3437 had_char_class = true;
3447 had_char_class = false;
3450 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
3452 unsigned char str[128]; /* Should be large enough. */
3455 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3461 /* If pattern is `[[.'. */
3462 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3467 if ((c == '.' && *p == ']') || p == pend)
3469 if (c1 < sizeof (str))
3472 /* This is in any case an invalid class name. */
3477 if (c == '.' && *p == ']' && str[0] != '\0')
3479 /* If we have no collation data we use the default
3480 collation in which each character is the name
3481 for its own class which contains only the one
3482 character. It also means that ASCII is the
3483 character set and therefore we cannot have character
3484 with more than one byte in the multibyte
3491 FREE_STACK_RETURN (REG_ECOLLATE);
3493 /* Throw away the ] at the end of the equivalence
3497 /* Set the bit for the character. */
3498 SET_LIST_BIT (str[0]);
3499 range_start = ((const unsigned char *) str)[0];
3504 /* Try to match the byte sequence in `str' against
3505 those known to the collate implementation.
3506 First find out whether the bytes in `str' are
3507 actually from exactly one character. */
3509 const int32_t *symb_table;
3510 const unsigned char *extra;
3517 _NL_CURRENT_WORD (LC_COLLATE,
3518 _NL_COLLATE_SYMB_HASH_SIZEMB);
3519 symb_table = (const int32_t *)
3520 _NL_CURRENT (LC_COLLATE,
3521 _NL_COLLATE_SYMB_TABLEMB);
3522 extra = (const unsigned char *)
3523 _NL_CURRENT (LC_COLLATE,
3524 _NL_COLLATE_SYMB_EXTRAMB);
3526 /* Locate the character in the hashing table. */
3527 hash = elem_hash (str, c1);
3530 elem = hash % table_size;
3531 second = hash % (table_size - 2);
3532 while (symb_table[2 * elem] != 0)
3534 /* First compare the hashing value. */
3535 if (symb_table[2 * elem] == hash
3536 && c1 == extra[symb_table[2 * elem + 1]]
3538 &extra[symb_table[2 * elem + 1]
3542 /* Yep, this is the entry. */
3543 idx = symb_table[2 * elem + 1];
3544 idx += 1 + extra[idx];
3552 if (symb_table[2 * elem] == 0)
3553 /* This is no valid character. */
3554 FREE_STACK_RETURN (REG_ECOLLATE);
3556 /* Throw away the ] at the end of the equivalence
3560 /* Now add the multibyte character(s) we found
3563 XXX Note that this is not entirely correct.
3564 we would have to match multibyte sequences
3565 but this is not possible with the current
3566 implementation. Also, we have to match
3567 collating symbols, which expand to more than
3568 one file, as a whole and not allow the
3569 individual bytes. */
3572 range_start = extra[idx];
3575 SET_LIST_BIT (extra[idx]);
3580 had_char_class = false;
3590 had_char_class = false;
3595 had_char_class = false;
3601 /* Discard any (non)matching list bytes that are all 0 at the
3602 end of the map. Decrease the map-length byte too. */
3603 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3612 if (syntax & RE_NO_BK_PARENS)
3619 if (syntax & RE_NO_BK_PARENS)
3626 if (syntax & RE_NEWLINE_ALT)
3633 if (syntax & RE_NO_BK_VBAR)
3640 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3641 goto handle_interval;
3647 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3649 /* Do not translate the character after the \, so that we can
3650 distinguish, e.g., \B from \b, even if we normally would
3651 translate, e.g., B to b. */
3657 if (syntax & RE_NO_BK_PARENS)
3658 goto normal_backslash;
3664 if (COMPILE_STACK_FULL)
3666 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3667 compile_stack_elt_t);
3668 if (compile_stack.stack == NULL) return REG_ESPACE;
3670 compile_stack.size <<= 1;
3673 /* These are the values to restore when we hit end of this
3674 group. They are all relative offsets, so that if the
3675 whole pattern moves because of realloc, they will still
3677 COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR;
3678 COMPILE_STACK_TOP.fixup_alt_jump
3679 = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0;
3680 COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR;
3681 COMPILE_STACK_TOP.regnum = regnum;
3683 /* We will eventually replace the 0 with the number of
3684 groups inner to this one. But do not push a
3685 start_memory for groups beyond the last one we can
3686 represent in the compiled pattern. */
3687 if (regnum <= MAX_REGNUM)
3689 COMPILE_STACK_TOP.inner_group_offset = b
3690 - COMPILED_BUFFER_VAR + 2;
3691 BUF_PUSH_3 (start_memory, regnum, 0);
3694 compile_stack.avail++;
3699 /* If we've reached MAX_REGNUM groups, then this open
3700 won't actually generate any code, so we'll have to
3701 clear pending_exact explicitly. */
3707 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3709 if (COMPILE_STACK_EMPTY)
3711 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3712 goto normal_backslash;
3714 FREE_STACK_RETURN (REG_ERPAREN);
3719 { /* Push a dummy failure point at the end of the
3720 alternative for a possible future
3721 `pop_failure_jump' to pop. See comments at
3722 `push_dummy_failure' in `re_match_2'. */
3723 BUF_PUSH (push_dummy_failure);
3725 /* We allocated space for this jump when we assigned
3726 to `fixup_alt_jump', in the `handle_alt' case below. */
3727 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
3730 /* See similar code for backslashed left paren above. */
3731 if (COMPILE_STACK_EMPTY)
3733 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3736 FREE_STACK_RETURN (REG_ERPAREN);
3739 /* Since we just checked for an empty stack above, this
3740 ``can't happen''. */
3741 assert (compile_stack.avail != 0);
3743 /* We don't just want to restore into `regnum', because
3744 later groups should continue to be numbered higher,
3745 as in `(ab)c(de)' -- the second group is #2. */
3746 regnum_t this_group_regnum;
3748 compile_stack.avail--;
3749 begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset;
3751 = COMPILE_STACK_TOP.fixup_alt_jump
3752 ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1
3754 laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset;
3755 this_group_regnum = COMPILE_STACK_TOP.regnum;
3756 /* If we've reached MAX_REGNUM groups, then this open
3757 won't actually generate any code, so we'll have to
3758 clear pending_exact explicitly. */
3761 /* We're at the end of the group, so now we know how many
3762 groups were inside this one. */
3763 if (this_group_regnum <= MAX_REGNUM)
3765 UCHAR_T *inner_group_loc
3766 = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset;
3768 *inner_group_loc = regnum - this_group_regnum;
3769 BUF_PUSH_3 (stop_memory, this_group_regnum,
3770 regnum - this_group_regnum);
3776 case '|': /* `\|'. */
3777 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3778 goto normal_backslash;
3780 if (syntax & RE_LIMITED_OPS)
3783 /* Insert before the previous alternative a jump which
3784 jumps to this alternative if the former fails. */
3785 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3786 INSERT_JUMP (on_failure_jump, begalt,
3787 b + 2 + 2 * OFFSET_ADDRESS_SIZE);
3789 b += 1 + OFFSET_ADDRESS_SIZE;
3791 /* The alternative before this one has a jump after it
3792 which gets executed if it gets matched. Adjust that
3793 jump so it will jump to this alternative's analogous
3794 jump (put in below, which in turn will jump to the next
3795 (if any) alternative's such jump, etc.). The last such
3796 jump jumps to the correct final destination. A picture:
3802 If we are at `b', then fixup_alt_jump right now points to a
3803 three-byte space after `a'. We'll put in the jump, set
3804 fixup_alt_jump to right after `b', and leave behind three
3805 bytes which we'll fill in when we get to after `c'. */
3808 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3810 /* Mark and leave space for a jump after this alternative,
3811 to be filled in later either by next alternative or
3812 when know we're at the end of a series of alternatives. */
3814 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3815 b += 1 + OFFSET_ADDRESS_SIZE;
3823 /* If \{ is a literal. */
3824 if (!(syntax & RE_INTERVALS)
3825 /* If we're at `\{' and it's not the open-interval
3827 || (syntax & RE_NO_BK_BRACES))
3828 goto normal_backslash;
3832 /* If got here, then the syntax allows intervals. */
3834 /* At least (most) this many matches must be made. */
3835 int lower_bound = -1, upper_bound = -1;
3837 /* Place in the uncompiled pattern (i.e., just after
3838 the '{') to go back to if the interval is invalid. */
3839 const CHAR_T *beg_interval = p;
3842 goto invalid_interval;
3844 GET_UNSIGNED_NUMBER (lower_bound);
3848 GET_UNSIGNED_NUMBER (upper_bound);
3849 if (upper_bound < 0)
3850 upper_bound = RE_DUP_MAX;
3853 /* Interval such as `{1}' => match exactly once. */
3854 upper_bound = lower_bound;
3856 if (! (0 <= lower_bound && lower_bound <= upper_bound))
3857 goto invalid_interval;
3859 if (!(syntax & RE_NO_BK_BRACES))
3861 if (c != '\\' || p == pend)
3862 goto invalid_interval;
3867 goto invalid_interval;
3869 /* If it's invalid to have no preceding re. */
3872 if (syntax & RE_CONTEXT_INVALID_OPS
3873 && !(syntax & RE_INVALID_INTERVAL_ORD))
3874 FREE_STACK_RETURN (REG_BADRPT);
3875 else if (syntax & RE_CONTEXT_INDEP_OPS)
3878 goto unfetch_interval;
3881 /* We just parsed a valid interval. */
3883 if (RE_DUP_MAX < upper_bound)
3884 FREE_STACK_RETURN (REG_BADBR);
3886 /* If the upper bound is zero, don't want to succeed at
3887 all; jump from `laststart' to `b + 3', which will be
3888 the end of the buffer after we insert the jump. */
3889 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3890 instead of 'b + 3'. */
3891 if (upper_bound == 0)
3893 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3894 INSERT_JUMP (jump, laststart, b + 1
3895 + OFFSET_ADDRESS_SIZE);
3896 b += 1 + OFFSET_ADDRESS_SIZE;
3899 /* Otherwise, we have a nontrivial interval. When
3900 we're all done, the pattern will look like:
3901 set_number_at <jump count> <upper bound>
3902 set_number_at <succeed_n count> <lower bound>
3903 succeed_n <after jump addr> <succeed_n count>
3905 jump_n <succeed_n addr> <jump count>
3906 (The upper bound and `jump_n' are omitted if
3907 `upper_bound' is 1, though.) */
3909 { /* If the upper bound is > 1, we need to insert
3910 more at the end of the loop. */
3911 unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE +
3912 (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE);
3914 GET_BUFFER_SPACE (nbytes);
3916 /* Initialize lower bound of the `succeed_n', even
3917 though it will be set during matching by its
3918 attendant `set_number_at' (inserted next),
3919 because `re_compile_fastmap' needs to know.
3920 Jump to the `jump_n' we might insert below. */
3921 INSERT_JUMP2 (succeed_n, laststart,
3922 b + 1 + 2 * OFFSET_ADDRESS_SIZE
3923 + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE)
3925 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3927 /* Code to initialize the lower bound. Insert
3928 before the `succeed_n'. The `5' is the last two
3929 bytes of this `set_number_at', plus 3 bytes of
3930 the following `succeed_n'. */
3931 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
3932 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
3933 of the following `succeed_n'. */
3934 PREFIX(insert_op2) (set_number_at, laststart, 1
3935 + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b);
3936 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3938 if (upper_bound > 1)
3939 { /* More than one repetition is allowed, so
3940 append a backward jump to the `succeed_n'
3941 that starts this interval.
3943 When we've reached this during matching,
3944 we'll have matched the interval once, so
3945 jump back only `upper_bound - 1' times. */
3946 STORE_JUMP2 (jump_n, b, laststart
3947 + 2 * OFFSET_ADDRESS_SIZE + 1,
3949 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3951 /* The location we want to set is the second
3952 parameter of the `jump_n'; that is `b-2' as
3953 an absolute address. `laststart' will be
3954 the `set_number_at' we're about to insert;
3955 `laststart+3' the number to set, the source
3956 for the relative address. But we are
3957 inserting into the middle of the pattern --
3958 so everything is getting moved up by 5.
3959 Conclusion: (b - 2) - (laststart + 3) + 5,
3960 i.e., b - laststart.
3962 We insert this at the beginning of the loop
3963 so that if we fail during matching, we'll
3964 reinitialize the bounds. */
3965 PREFIX(insert_op2) (set_number_at, laststart,
3967 upper_bound - 1, b);
3968 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3975 if (!(syntax & RE_INVALID_INTERVAL_ORD))
3976 FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR);
3978 /* Match the characters as literals. */
3981 if (syntax & RE_NO_BK_BRACES)
3984 goto normal_backslash;
3988 /* There is no way to specify the before_dot and after_dot
3989 operators. rms says this is ok. --karl */
3997 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
4003 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
4009 if (syntax & RE_NO_GNU_OPS)
4012 BUF_PUSH (wordchar);
4017 if (syntax & RE_NO_GNU_OPS)
4020 BUF_PUSH (notwordchar);
4025 if (syntax & RE_NO_GNU_OPS)
4031 if (syntax & RE_NO_GNU_OPS)
4037 if (syntax & RE_NO_GNU_OPS)
4039 BUF_PUSH (wordbound);
4043 if (syntax & RE_NO_GNU_OPS)
4045 BUF_PUSH (notwordbound);
4049 if (syntax & RE_NO_GNU_OPS)
4055 if (syntax & RE_NO_GNU_OPS)
4060 case '1': case '2': case '3': case '4': case '5':
4061 case '6': case '7': case '8': case '9':
4062 if (syntax & RE_NO_BK_REFS)
4068 FREE_STACK_RETURN (REG_ESUBREG);
4070 /* Can't back reference to a subexpression if inside of it. */
4071 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
4075 BUF_PUSH_2 (duplicate, c1);
4081 if (syntax & RE_BK_PLUS_QM)
4084 goto normal_backslash;
4088 /* You might think it would be useful for \ to mean
4089 not to translate; but if we don't translate it
4090 it will never match anything. */
4098 /* Expects the character in `c'. */
4100 /* If no exactn currently being built. */
4103 /* If last exactn handle binary(or character) and
4104 new exactn handle character(or binary). */
4105 || is_exactn_bin != is_binary[p - 1 - pattern]
4108 /* If last exactn not at current position. */
4109 || pending_exact + *pending_exact + 1 != b
4111 /* We have only one byte following the exactn for the count. */
4112 || *pending_exact == (1 << BYTEWIDTH) - 1
4114 /* If followed by a repetition operator. */
4115 || *p == '*' || *p == '^'
4116 || ((syntax & RE_BK_PLUS_QM)
4117 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
4118 : (*p == '+' || *p == '?'))
4119 || ((syntax & RE_INTERVALS)
4120 && ((syntax & RE_NO_BK_BRACES)
4122 : (p[0] == '\\' && p[1] == '{'))))
4124 /* Start building a new exactn. */
4129 /* Is this exactn binary data or character? */
4130 is_exactn_bin = is_binary[p - 1 - pattern];
4132 BUF_PUSH_2 (exactn_bin, 0);
4134 BUF_PUSH_2 (exactn, 0);
4136 BUF_PUSH_2 (exactn, 0);
4138 pending_exact = b - 1;
4145 } /* while p != pend */
4148 /* Through the pattern now. */
4151 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
4153 if (!COMPILE_STACK_EMPTY)
4154 FREE_STACK_RETURN (REG_EPAREN);
4156 /* If we don't want backtracking, force success
4157 the first time we reach the end of the compiled pattern. */
4158 if (syntax & RE_NO_POSIX_BACKTRACKING)
4166 free (compile_stack.stack);
4168 /* We have succeeded; set the length of the buffer. */
4170 bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR;
4172 bufp->used = b - bufp->buffer;
4178 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4179 PREFIX(print_compiled_pattern) (bufp);
4183 #ifndef MATCH_MAY_ALLOCATE
4184 /* Initialize the failure stack to the largest possible stack. This
4185 isn't necessary unless we're trying to avoid calling alloca in
4186 the search and match routines. */
4188 int num_regs = bufp->re_nsub + 1;
4190 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4191 is strictly greater than re_max_failures, the largest possible stack
4192 is 2 * re_max_failures failure points. */
4193 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
4195 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
4198 if (! fail_stack.stack)
4200 = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size
4201 * sizeof (PREFIX(fail_stack_elt_t)));
4204 = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack,
4206 * sizeof (PREFIX(fail_stack_elt_t))));
4207 # else /* not emacs */
4208 if (! fail_stack.stack)
4210 = malloc (fail_stack.size * sizeof (PREFIX(fail_stack_elt_t)));
4213 = realloc (fail_stack.stack,
4214 fail_stack.size * sizeof (PREFIX(fail_stack_elt_t)));
4215 # endif /* not emacs */
4218 PREFIX(regex_grow_registers) (num_regs);
4220 #endif /* not MATCH_MAY_ALLOCATE */
4223 } /* regex_compile */
4225 /* Subroutines for `regex_compile'. */
4227 /* Store OP at LOC followed by two-byte integer parameter ARG. */
4228 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4231 PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg)
4233 *loc = (UCHAR_T) op;
4234 STORE_NUMBER (loc + 1, arg);
4238 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4239 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4242 PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc, int arg1, int arg2)
4244 *loc = (UCHAR_T) op;
4245 STORE_NUMBER (loc + 1, arg1);
4246 STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2);
4250 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
4251 for OP followed by two-byte integer parameter ARG. */
4252 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4255 PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc, int arg, UCHAR_T *end)
4257 register UCHAR_T *pfrom = end;
4258 register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE;
4260 while (pfrom != loc)
4263 PREFIX(store_op1) (op, loc, arg);
4267 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4268 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4271 PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc, int arg1, int arg2,
4274 register UCHAR_T *pfrom = end;
4275 register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE;
4277 while (pfrom != loc)
4280 PREFIX(store_op2) (op, loc, arg1, arg2);
4284 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
4285 after an alternative or a begin-subexpression. We assume there is at
4286 least one character before the ^. */
4289 PREFIX(at_begline_loc_p) (const CHAR_T *pattern, const CHAR_T *p,
4290 reg_syntax_t syntax)
4292 const CHAR_T *prev = p - 2;
4293 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
4296 /* After a subexpression? */
4297 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
4298 /* After an alternative? */
4299 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
4303 /* The dual of at_begline_loc_p. This one is for $. We assume there is
4304 at least one character after the $, i.e., `P < PEND'. */
4307 PREFIX(at_endline_loc_p) (const CHAR_T *p, const CHAR_T *pend,
4308 reg_syntax_t syntax)
4310 const CHAR_T *next = p;
4311 boolean next_backslash = *next == '\\';
4312 const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0;
4315 /* Before a subexpression? */
4316 (syntax & RE_NO_BK_PARENS ? *next == ')'
4317 : next_backslash && next_next && *next_next == ')')
4318 /* Before an alternative? */
4319 || (syntax & RE_NO_BK_VBAR ? *next == '|'
4320 : next_backslash && next_next && *next_next == '|');
4323 #else /* not INSIDE_RECURSION */
4325 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4326 false if it's not. */
4329 group_in_compile_stack (compile_stack_type compile_stack,
4334 for (this_element = compile_stack.avail - 1;
4337 if (compile_stack.stack[this_element].regnum == regnum)
4342 #endif /* not INSIDE_RECURSION */
4344 #ifdef INSIDE_RECURSION
4347 /* This insert space, which size is "num", into the pattern at "loc".
4348 "end" must point the end of the allocated buffer. */
4350 insert_space (int num, CHAR_T *loc, CHAR_T *end)
4352 register CHAR_T *pto = end;
4353 register CHAR_T *pfrom = end - num;
4355 while (pfrom >= loc)
4361 static reg_errcode_t
4362 wcs_compile_range (CHAR_T range_start_char,
4363 const CHAR_T **p_ptr, const CHAR_T *pend,
4364 RE_TRANSLATE_TYPE translate, reg_syntax_t syntax,
4365 CHAR_T *b, CHAR_T *char_set)
4367 const CHAR_T *p = *p_ptr;
4368 CHAR_T range_start, range_end;
4372 uint32_t start_val, end_val;
4378 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4381 const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE,
4382 _NL_COLLATE_COLLSEQWC);
4383 const unsigned char *extra = (const unsigned char *)
4384 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4386 if (range_start_char < -1)
4388 /* range_start is a collating symbol. */
4390 /* Retreive the index and get collation sequence value. */
4391 wextra = (int32_t*)(extra + char_set[-range_start_char]);
4392 start_val = wextra[1 + *wextra];
4395 start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char));
4397 end_val = collseq_table_lookup (collseq, TRANSLATE (p[0]));
4399 /* Report an error if the range is empty and the syntax prohibits
4401 ret = ((syntax & RE_NO_EMPTY_RANGES)
4402 && (start_val > end_val))? REG_ERANGE : REG_NOERROR;
4404 /* Insert space to the end of the char_ranges. */
4405 insert_space(2, b - char_set[5] - 2, b - 1);
4406 *(b - char_set[5] - 2) = (wchar_t)start_val;
4407 *(b - char_set[5] - 1) = (wchar_t)end_val;
4408 char_set[4]++; /* ranges_index */
4413 range_start = (range_start_char >= 0)? TRANSLATE (range_start_char):
4415 range_end = TRANSLATE (p[0]);
4416 /* Report an error if the range is empty and the syntax prohibits
4418 ret = ((syntax & RE_NO_EMPTY_RANGES)
4419 && (range_start > range_end))? REG_ERANGE : REG_NOERROR;
4421 /* Insert space to the end of the char_ranges. */
4422 insert_space(2, b - char_set[5] - 2, b - 1);
4423 *(b - char_set[5] - 2) = range_start;
4424 *(b - char_set[5] - 1) = range_end;
4425 char_set[4]++; /* ranges_index */
4427 /* Have to increment the pointer into the pattern string, so the
4428 caller isn't still at the ending character. */
4434 /* Read the ending character of a range (in a bracket expression) from the
4435 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4436 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4437 Then we set the translation of all bits between the starting and
4438 ending characters (inclusive) in the compiled pattern B.
4440 Return an error code.
4442 We use these short variable names so we can use the same macros as
4443 `regex_compile' itself. */
4445 static reg_errcode_t
4446 byte_compile_range (unsigned int range_start_char,
4447 const char **p_ptr, const char *pend,
4448 RE_TRANSLATE_TYPE translate, reg_syntax_t syntax,
4452 const char *p = *p_ptr;
4455 const unsigned char *collseq;
4456 unsigned int start_colseq;
4457 unsigned int end_colseq;
4465 /* Have to increment the pointer into the pattern string, so the
4466 caller isn't still at the ending character. */
4469 /* Report an error if the range is empty and the syntax prohibits this. */
4470 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
4473 collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
4474 _NL_COLLATE_COLLSEQMB);
4476 start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)];
4477 end_colseq = collseq[(unsigned char) TRANSLATE (p[0])];
4478 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
4480 unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)];
4482 if (start_colseq <= this_colseq && this_colseq <= end_colseq)
4484 SET_LIST_BIT (TRANSLATE (this_char));
4489 /* Here we see why `this_char' has to be larger than an `unsigned
4490 char' -- we would otherwise go into an infinite loop, since all
4491 characters <= 0xff. */
4492 range_start_char = TRANSLATE (range_start_char);
4493 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4494 and some compilers cast it to int implicitly, so following for_loop
4495 may fall to (almost) infinite loop.
4496 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4497 To avoid this, we cast p[0] to unsigned int and truncate it. */
4498 end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1));
4500 for (this_char = range_start_char; this_char <= end_char; ++this_char)
4502 SET_LIST_BIT (TRANSLATE (this_char));
4511 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4512 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4513 characters can start a string that matches the pattern. This fastmap
4514 is used by re_search to skip quickly over impossible starting points.
4516 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4517 area as BUFP->fastmap.
4519 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4522 Returns 0 if we succeed, -2 if an internal error. */
4525 /* local function for re_compile_fastmap.
4526 truncate wchar_t character to char. */
4528 static unsigned char
4529 truncate_wchar (CHAR_T c)
4531 unsigned char buf[MB_CUR_MAX];
4534 memset (&state, '\0', sizeof (state));
4535 retval = wcrtomb (buf, c, &state);
4536 return retval > 0 ? buf[0] : (unsigned char) c;
4541 PREFIX(re_compile_fastmap) (struct re_pattern_buffer *bufp)
4544 #ifdef MATCH_MAY_ALLOCATE
4545 PREFIX(fail_stack_type) fail_stack;
4547 #ifndef REGEX_MALLOC
4551 register char *fastmap = bufp->fastmap;
4554 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4555 pattern to (char*) in regex_compile. */
4556 UCHAR_T *pattern = (UCHAR_T*)bufp->buffer;
4557 register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used);
4559 UCHAR_T *pattern = bufp->buffer;
4560 register UCHAR_T *pend = pattern + bufp->used;
4562 UCHAR_T *p = pattern;
4565 /* This holds the pointer to the failure stack, when
4566 it is allocated relocatably. */
4567 fail_stack_elt_t *failure_stack_ptr;
4570 /* Assume that each path through the pattern can be null until
4571 proven otherwise. We set this false at the bottom of switch
4572 statement, to which we get only if a particular path doesn't
4573 match the empty string. */
4574 boolean path_can_be_null = true;
4576 /* We aren't doing a `succeed_n' to begin with. */
4577 boolean succeed_n_p = false;
4579 assert (fastmap != NULL && p != NULL);
4582 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4583 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4584 bufp->can_be_null = 0;
4588 if (p == pend || *p == succeed)
4590 /* We have reached the (effective) end of pattern. */
4591 if (!FAIL_STACK_EMPTY ())
4593 bufp->can_be_null |= path_can_be_null;
4595 /* Reset for next path. */
4596 path_can_be_null = true;
4598 p = fail_stack.stack[--fail_stack.avail].pointer;
4606 /* We should never be about to go beyond the end of the pattern. */
4609 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4612 /* I guess the idea here is to simply not bother with a fastmap
4613 if a backreference is used, since it's too hard to figure out
4614 the fastmap for the corresponding group. Setting
4615 `can_be_null' stops `re_search_2' from using the fastmap, so
4616 that is all we do. */
4618 bufp->can_be_null = 1;
4622 /* Following are the cases which match a character. These end
4627 fastmap[truncate_wchar(p[1])] = 1;
4641 /* It is hard to distinguish fastmap from (multi byte) characters
4642 which depends on current locale. */
4647 bufp->can_be_null = 1;
4651 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4652 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
4658 /* Chars beyond end of map must be allowed. */
4659 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
4662 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4663 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
4669 for (j = 0; j < (1 << BYTEWIDTH); j++)
4670 if (SYNTAX (j) == Sword)
4676 for (j = 0; j < (1 << BYTEWIDTH); j++)
4677 if (SYNTAX (j) != Sword)
4684 int fastmap_newline = fastmap['\n'];
4686 /* `.' matches anything ... */
4687 for (j = 0; j < (1 << BYTEWIDTH); j++)
4690 /* ... except perhaps newline. */
4691 if (!(bufp->syntax & RE_DOT_NEWLINE))
4692 fastmap['\n'] = fastmap_newline;
4694 /* Return if we have already set `can_be_null'; if we have,
4695 then the fastmap is irrelevant. Something's wrong here. */
4696 else if (bufp->can_be_null)
4699 /* Otherwise, have to check alternative paths. */
4706 for (j = 0; j < (1 << BYTEWIDTH); j++)
4707 if (SYNTAX (j) == (enum syntaxcode) k)
4714 for (j = 0; j < (1 << BYTEWIDTH); j++)
4715 if (SYNTAX (j) != (enum syntaxcode) k)
4720 /* All cases after this match the empty string. These end with
4740 case push_dummy_failure:
4745 case pop_failure_jump:
4746 case maybe_pop_jump:
4749 case dummy_failure_jump:
4750 EXTRACT_NUMBER_AND_INCR (j, p);
4755 /* Jump backward implies we just went through the body of a
4756 loop and matched nothing. Opcode jumped to should be
4757 `on_failure_jump' or `succeed_n'. Just treat it like an
4758 ordinary jump. For a * loop, it has pushed its failure
4759 point already; if so, discard that as redundant. */
4760 if ((re_opcode_t) *p != on_failure_jump
4761 && (re_opcode_t) *p != succeed_n)
4765 EXTRACT_NUMBER_AND_INCR (j, p);
4768 /* If what's on the stack is where we are now, pop it. */
4769 if (!FAIL_STACK_EMPTY ()
4770 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
4776 case on_failure_jump:
4777 case on_failure_keep_string_jump:
4778 handle_on_failure_jump:
4779 EXTRACT_NUMBER_AND_INCR (j, p);
4781 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4782 end of the pattern. We don't want to push such a point,
4783 since when we restore it above, entering the switch will
4784 increment `p' past the end of the pattern. We don't need
4785 to push such a point since we obviously won't find any more
4786 fastmap entries beyond `pend'. Such a pattern can match
4787 the null string, though. */
4790 if (!PUSH_PATTERN_OP (p + j, fail_stack))
4792 RESET_FAIL_STACK ();
4797 bufp->can_be_null = 1;
4801 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
4802 succeed_n_p = false;
4809 /* Get to the number of times to succeed. */
4810 p += OFFSET_ADDRESS_SIZE;
4812 /* Increment p past the n for when k != 0. */
4813 EXTRACT_NUMBER_AND_INCR (k, p);
4816 p -= 2 * OFFSET_ADDRESS_SIZE;
4817 succeed_n_p = true; /* Spaghetti code alert. */
4818 goto handle_on_failure_jump;
4824 p += 2 * OFFSET_ADDRESS_SIZE;
4835 abort (); /* We have listed all the cases. */
4838 /* Getting here means we have found the possible starting
4839 characters for one path of the pattern -- and that the empty
4840 string does not match. We need not follow this path further.
4841 Instead, look at the next alternative (remembered on the
4842 stack), or quit if no more. The test at the top of the loop
4843 does these things. */
4844 path_can_be_null = false;
4848 /* Set `can_be_null' for the last path (also the first path, if the
4849 pattern is empty). */
4850 bufp->can_be_null |= path_can_be_null;
4853 RESET_FAIL_STACK ();
4857 #else /* not INSIDE_RECURSION */
4860 re_compile_fastmap (struct re_pattern_buffer *bufp)
4863 if (MB_CUR_MAX != 1)
4864 return wcs_re_compile_fastmap(bufp);
4867 return byte_re_compile_fastmap(bufp);
4868 } /* re_compile_fastmap */
4870 weak_alias (__re_compile_fastmap, re_compile_fastmap)
4874 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4875 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4876 this memory for recording register information. STARTS and ENDS
4877 must be allocated using the malloc library routine, and must each
4878 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4880 If NUM_REGS == 0, then subsequent matches should allocate their own
4883 Unless this function is called, the first search or match using
4884 PATTERN_BUFFER will allocate its own register data, without
4885 freeing the old data. */
4888 re_set_registers (struct re_pattern_buffer *bufp,
4889 struct re_registers *regs,
4890 unsigned int num_regs,
4891 regoff_t *starts, regoff_t *ends)
4895 bufp->regs_allocated = REGS_REALLOCATE;
4896 regs->num_regs = num_regs;
4897 regs->start = starts;
4902 bufp->regs_allocated = REGS_UNALLOCATED;
4904 regs->start = regs->end = (regoff_t *) 0;
4908 weak_alias (__re_set_registers, re_set_registers)
4911 /* Searching routines. */
4913 /* Like re_search_2, below, but only one string is specified, and
4914 doesn't let you say where to stop matching. */
4917 re_search (struct re_pattern_buffer *bufp,
4919 int size, int startpos, int range,
4920 struct re_registers *regs)
4922 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4926 weak_alias (__re_search, re_search)
4930 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4931 virtual concatenation of STRING1 and STRING2, starting first at index
4932 STARTPOS, then at STARTPOS + 1, and so on.
4934 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4936 RANGE is how far to scan while trying to match. RANGE = 0 means try
4937 only at STARTPOS; in general, the last start tried is STARTPOS +
4940 In REGS, return the indices of the virtual concatenation of STRING1
4941 and STRING2 that matched the entire BUFP->buffer and its contained
4944 Do not consider matching one past the index STOP in the virtual
4945 concatenation of STRING1 and STRING2.
4947 We return either the position in the strings at which the match was
4948 found, -1 if no match, or -2 if error (such as failure
4952 re_search_2 (struct re_pattern_buffer *bufp,
4953 const char *string1, int size1,
4954 const char *string2, int size2,
4955 int startpos, int range,
4956 struct re_registers *regs,
4960 if (MB_CUR_MAX != 1)
4961 return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos,
4965 return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos,
4969 weak_alias (__re_search_2, re_search_2)
4972 #endif /* not INSIDE_RECURSION */
4974 #ifdef INSIDE_RECURSION
4976 #ifdef MATCH_MAY_ALLOCATE
4977 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
4979 # define FREE_VAR(var) if (var) free (var); var = NULL
4983 # define MAX_ALLOCA_SIZE 2000
4985 # define FREE_WCS_BUFFERS() \
4987 if (size1 > MAX_ALLOCA_SIZE) \
4989 free (wcs_string1); \
4990 free (mbs_offset1); \
4994 FREE_VAR (wcs_string1); \
4995 FREE_VAR (mbs_offset1); \
4997 if (size2 > MAX_ALLOCA_SIZE) \
4999 free (wcs_string2); \
5000 free (mbs_offset2); \
5004 FREE_VAR (wcs_string2); \
5005 FREE_VAR (mbs_offset2); \
5013 PREFIX(re_search_2) (struct re_pattern_buffer *bufp,
5014 const char *string1, int size1,
5015 const char *string2, int size2,
5016 int startpos, int range,
5017 struct re_registers *regs,
5021 register char *fastmap = bufp->fastmap;
5022 register RE_TRANSLATE_TYPE translate = bufp->translate;
5023 int total_size = size1 + size2;
5024 int endpos = startpos + range;
5026 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5027 wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL;
5028 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5029 int wcs_size1 = 0, wcs_size2 = 0;
5030 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5031 int *mbs_offset1 = NULL, *mbs_offset2 = NULL;
5032 /* They hold whether each wchar_t is binary data or not. */
5033 char *is_binary = NULL;
5036 /* Check for out-of-range STARTPOS. */
5037 if (startpos < 0 || startpos > total_size)
5040 /* Fix up RANGE if it might eventually take us outside
5041 the virtual concatenation of STRING1 and STRING2.
5042 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5044 range = 0 - startpos;
5045 else if (endpos > total_size)
5046 range = total_size - startpos;
5048 /* If the search isn't to be a backwards one, don't waste time in a
5049 search for a pattern that must be anchored. */
5050 if (bufp->used > 0 && range > 0
5051 && ((re_opcode_t) bufp->buffer[0] == begbuf
5052 /* `begline' is like `begbuf' if it cannot match at newlines. */
5053 || ((re_opcode_t) bufp->buffer[0] == begline
5054 && !bufp->newline_anchor)))
5063 /* In a forward search for something that starts with \=.
5064 don't keep searching past point. */
5065 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
5067 range = PT - startpos;
5073 /* Update the fastmap now if not correct already. */
5074 if (fastmap && !bufp->fastmap_accurate)
5075 if (re_compile_fastmap (bufp) == -2)
5079 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5080 fill them with converted string. */
5083 if (size1 > MAX_ALLOCA_SIZE)
5085 wcs_string1 = TALLOC (size1 + 1, CHAR_T);
5086 mbs_offset1 = TALLOC (size1 + 1, int);
5087 is_binary = TALLOC (size1 + 1, char);
5091 wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T);
5092 mbs_offset1 = REGEX_TALLOC (size1 + 1, int);
5093 is_binary = REGEX_TALLOC (size1 + 1, char);
5095 if (!wcs_string1 || !mbs_offset1 || !is_binary)
5097 if (size1 > MAX_ALLOCA_SIZE)
5105 FREE_VAR (wcs_string1);
5106 FREE_VAR (mbs_offset1);
5107 FREE_VAR (is_binary);
5111 wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1,
5112 mbs_offset1, is_binary);
5113 wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */
5114 if (size1 > MAX_ALLOCA_SIZE)
5117 FREE_VAR (is_binary);
5121 if (size2 > MAX_ALLOCA_SIZE)
5123 wcs_string2 = TALLOC (size2 + 1, CHAR_T);
5124 mbs_offset2 = TALLOC (size2 + 1, int);
5125 is_binary = TALLOC (size2 + 1, char);
5129 wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T);
5130 mbs_offset2 = REGEX_TALLOC (size2 + 1, int);
5131 is_binary = REGEX_TALLOC (size2 + 1, char);
5133 if (!wcs_string2 || !mbs_offset2 || !is_binary)
5135 FREE_WCS_BUFFERS ();
5136 if (size2 > MAX_ALLOCA_SIZE)
5139 FREE_VAR (is_binary);
5142 wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2,
5143 mbs_offset2, is_binary);
5144 wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */
5145 if (size2 > MAX_ALLOCA_SIZE)
5148 FREE_VAR (is_binary);
5153 /* Loop through the string, looking for a place to start matching. */
5156 /* If a fastmap is supplied, skip quickly over characters that
5157 cannot be the start of a match. If the pattern can match the
5158 null string, however, we don't need to skip characters; we want
5159 the first null string. */
5160 if (fastmap && startpos < total_size && !bufp->can_be_null)
5162 if (range > 0) /* Searching forwards. */
5164 register const char *d;
5165 register int lim = 0;
5168 if (startpos < size1 && startpos + range >= size1)
5169 lim = range - (size1 - startpos);
5171 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
5173 /* Written out as an if-else to avoid testing `translate'
5177 && !fastmap[(unsigned char)
5178 translate[(unsigned char) *d++]])
5181 while (range > lim && !fastmap[(unsigned char) *d++])
5184 startpos += irange - range;
5186 else /* Searching backwards. */
5188 register CHAR_T c = (size1 == 0 || startpos >= size1
5189 ? string2[startpos - size1]
5190 : string1[startpos]);
5192 if (!fastmap[(unsigned char) TRANSLATE (c)])
5197 /* If can't match the null string, and that's all we have left, fail. */
5198 if (range >= 0 && startpos == total_size && fastmap
5199 && !bufp->can_be_null)
5202 FREE_WCS_BUFFERS ();
5208 val = wcs_re_match_2_internal (bufp, string1, size1, string2,
5209 size2, startpos, regs, stop,
5210 wcs_string1, wcs_size1,
5211 wcs_string2, wcs_size2,
5212 mbs_offset1, mbs_offset2);
5214 val = byte_re_match_2_internal (bufp, string1, size1, string2,
5215 size2, startpos, regs, stop);
5218 #ifndef REGEX_MALLOC
5227 FREE_WCS_BUFFERS ();
5235 FREE_WCS_BUFFERS ();
5255 FREE_WCS_BUFFERS ();
5261 /* This converts PTR, a pointer into one of the search wchar_t strings
5262 `string1' and `string2' into an multibyte string offset from the
5263 beginning of that string. We use mbs_offset to optimize.
5264 See convert_mbs_to_wcs. */
5265 # define POINTER_TO_OFFSET(ptr) \
5266 (FIRST_STRING_P (ptr) \
5267 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5268 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5271 /* This converts PTR, a pointer into one of the search strings `string1'
5272 and `string2' into an offset from the beginning of that string. */
5273 # define POINTER_TO_OFFSET(ptr) \
5274 (FIRST_STRING_P (ptr) \
5275 ? ((regoff_t) ((ptr) - string1)) \
5276 : ((regoff_t) ((ptr) - string2 + size1)))
5279 /* Macros for dealing with the split strings in re_match_2. */
5281 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5283 /* Call before fetching a character with *d. This switches over to
5284 string2 if necessary. */
5285 #define PREFETCH() \
5288 /* End of string2 => fail. */ \
5289 if (dend == end_match_2) \
5291 /* End of string1 => advance to string2. */ \
5293 dend = end_match_2; \
5296 /* Test if at very beginning or at very end of the virtual concatenation
5297 of `string1' and `string2'. If only one string, it's `string2'. */
5298 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5299 #define AT_STRINGS_END(d) ((d) == end2)
5302 /* Test if D points to a character which is word-constituent. We have
5303 two special cases to check for: if past the end of string1, look at
5304 the first character in string2; and if before the beginning of
5305 string2, look at the last character in string1. */
5307 /* Use internationalized API instead of SYNTAX. */
5308 # define WORDCHAR_P(d) \
5309 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5310 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5311 || ((d) == end1 ? *string2 \
5312 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5314 # define WORDCHAR_P(d) \
5315 (SYNTAX ((d) == end1 ? *string2 \
5316 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5320 /* Disabled due to a compiler bug -- see comment at case wordbound */
5322 /* Test if the character before D and the one at D differ with respect
5323 to being word-constituent. */
5324 #define AT_WORD_BOUNDARY(d) \
5325 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5326 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5329 /* Free everything we malloc. */
5330 #ifdef MATCH_MAY_ALLOCATE
5332 # define FREE_VARIABLES() \
5334 REGEX_FREE_STACK (fail_stack.stack); \
5335 FREE_VAR (regstart); \
5336 FREE_VAR (regend); \
5337 FREE_VAR (old_regstart); \
5338 FREE_VAR (old_regend); \
5339 FREE_VAR (best_regstart); \
5340 FREE_VAR (best_regend); \
5341 FREE_VAR (reg_info); \
5342 FREE_VAR (reg_dummy); \
5343 FREE_VAR (reg_info_dummy); \
5344 if (!cant_free_wcs_buf) \
5346 FREE_VAR (string1); \
5347 FREE_VAR (string2); \
5348 FREE_VAR (mbs_offset1); \
5349 FREE_VAR (mbs_offset2); \
5353 # define FREE_VARIABLES() \
5355 REGEX_FREE_STACK (fail_stack.stack); \
5356 FREE_VAR (regstart); \
5357 FREE_VAR (regend); \
5358 FREE_VAR (old_regstart); \
5359 FREE_VAR (old_regend); \
5360 FREE_VAR (best_regstart); \
5361 FREE_VAR (best_regend); \
5362 FREE_VAR (reg_info); \
5363 FREE_VAR (reg_dummy); \
5364 FREE_VAR (reg_info_dummy); \
5369 # define FREE_VARIABLES() \
5371 if (!cant_free_wcs_buf) \
5373 FREE_VAR (string1); \
5374 FREE_VAR (string2); \
5375 FREE_VAR (mbs_offset1); \
5376 FREE_VAR (mbs_offset2); \
5380 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5382 #endif /* not MATCH_MAY_ALLOCATE */
5384 /* These values must meet several constraints. They must not be valid
5385 register values; since we have a limit of 255 registers (because
5386 we use only one byte in the pattern for the register number), we can
5387 use numbers larger than 255. They must differ by 1, because of
5388 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5389 be larger than the value for the highest register, so we do not try
5390 to actually save any registers when none are active. */
5391 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5392 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5394 #else /* not INSIDE_RECURSION */
5395 /* Matching routines. */
5397 #ifndef emacs /* Emacs never uses this. */
5398 /* re_match is like re_match_2 except it takes only a single string. */
5401 re_match (struct re_pattern_buffer *bufp,
5404 struct re_registers *regs)
5408 if (MB_CUR_MAX != 1)
5409 result = wcs_re_match_2_internal (bufp, NULL, 0, string, size,
5411 NULL, 0, NULL, 0, NULL, NULL);
5414 result = byte_re_match_2_internal (bufp, NULL, 0, string, size,
5416 # ifndef REGEX_MALLOC
5424 weak_alias (__re_match, re_match)
5426 #endif /* not emacs */
5428 #endif /* not INSIDE_RECURSION */
5430 #ifdef INSIDE_RECURSION
5431 static boolean PREFIX(group_match_null_string_p) (UCHAR_T **p,
5433 PREFIX(register_info_type) *reg_info);
5434 static boolean PREFIX(alt_match_null_string_p) (UCHAR_T *p,
5436 PREFIX(register_info_type) *reg_info);
5437 static boolean PREFIX(common_op_match_null_string_p) (UCHAR_T **p,
5439 PREFIX(register_info_type) *reg_info);
5440 static int PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2,
5441 int len, char *translate);
5442 #else /* not INSIDE_RECURSION */
5444 /* re_match_2 matches the compiled pattern in BUFP against the
5445 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5446 and SIZE2, respectively). We start matching at POS, and stop
5449 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5450 store offsets for the substring each group matched in REGS. See the
5451 documentation for exactly how many groups we fill.
5453 We return -1 if no match, -2 if an internal error (such as the
5454 failure stack overflowing). Otherwise, we return the length of the
5455 matched substring. */
5458 re_match_2 (struct re_pattern_buffer *bufp,
5459 const char *string1, int size1,
5460 const char *string2, int size2,
5461 int pos, struct re_registers *regs,
5466 if (MB_CUR_MAX != 1)
5467 result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2,
5469 NULL, 0, NULL, 0, NULL, NULL);
5472 result = byte_re_match_2_internal (bufp, string1, size1, string2, size2,
5475 #ifndef REGEX_MALLOC
5483 weak_alias (__re_match_2, re_match_2)
5486 #endif /* not INSIDE_RECURSION */
5488 #ifdef INSIDE_RECURSION
5492 /* This check the substring (from 0, to length) of the multibyte string,
5493 to which offset_buffer correspond. And count how many wchar_t_characters
5494 the substring occupy. We use offset_buffer to optimization.
5495 See convert_mbs_to_wcs. */
5498 count_mbs_length (int *offset_buffer, int length)
5502 /* Check whether the size is valid. */
5506 if (offset_buffer == NULL)
5509 /* If there are no multibyte character, offset_buffer[i] == i.
5510 Optmize for this case. */
5511 if (offset_buffer[length] == length)
5514 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5520 int middle = (lower + upper) / 2;
5521 if (middle == lower || middle == upper)
5523 if (offset_buffer[middle] > length)
5525 else if (offset_buffer[middle] < length)
5535 /* This is a separate function so that we can force an alloca cleanup
5539 wcs_re_match_2_internal (struct re_pattern_buffer *bufp,
5540 const char *cstring1, int csize1,
5541 const char *cstring2, int csize2,
5543 struct re_registers *regs,
5545 /* string1 == string2 == NULL means
5546 string1/2, size1/2 and mbs_offset1/2 need
5547 setting up in this function. */
5548 /* We need wchar_t * buffers corresponding to
5549 cstring1, cstring2. */
5550 wchar_t *string1, int size1,
5551 wchar_t *string2, int size2,
5552 /* Offset buffer for optimization. See
5553 convert_mbs_to_wc. */
5558 byte_re_match_2_internal (struct re_pattern_buffer *bufp,
5559 const char *string1, int size1,
5560 const char *string2, int size2,
5562 struct re_registers *regs,
5566 /* General temporaries. */
5570 /* They hold whether each wchar_t is binary data or not. */
5571 char *is_binary = NULL;
5572 /* If true, we can't free string1/2, mbs_offset1/2. */
5573 int cant_free_wcs_buf = 1;
5576 /* Just past the end of the corresponding string. */
5577 const CHAR_T *end1, *end2;
5579 /* Pointers into string1 and string2, just past the last characters in
5580 each to consider matching. */
5581 const CHAR_T *end_match_1, *end_match_2;
5583 /* Where we are in the data, and the end of the current string. */
5584 const CHAR_T *d, *dend;
5586 /* Where we are in the pattern, and the end of the pattern. */
5588 UCHAR_T *pattern, *p;
5589 register UCHAR_T *pend;
5591 UCHAR_T *p = bufp->buffer;
5592 register UCHAR_T *pend = p + bufp->used;
5595 /* Mark the opcode just after a start_memory, so we can test for an
5596 empty subpattern when we get to the stop_memory. */
5597 UCHAR_T *just_past_start_mem = 0;
5599 /* We use this to map every character in the string. */
5600 RE_TRANSLATE_TYPE translate = bufp->translate;
5602 /* Failure point stack. Each place that can handle a failure further
5603 down the line pushes a failure point on this stack. It consists of
5604 restart, regend, and reg_info for all registers corresponding to
5605 the subexpressions we're currently inside, plus the number of such
5606 registers, and, finally, two char *'s. The first char * is where
5607 to resume scanning the pattern; the second one is where to resume
5608 scanning the strings. If the latter is zero, the failure point is
5609 a ``dummy''; if a failure happens and the failure point is a dummy,
5610 it gets discarded and the next next one is tried. */
5611 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5612 PREFIX(fail_stack_type) fail_stack;
5615 static unsigned failure_id;
5616 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5620 /* This holds the pointer to the failure stack, when
5621 it is allocated relocatably. */
5622 fail_stack_elt_t *failure_stack_ptr;
5625 /* We fill all the registers internally, independent of what we
5626 return, for use in backreferences. The number here includes
5627 an element for register zero. */
5628 size_t num_regs = bufp->re_nsub + 1;
5630 /* The currently active registers. */
5631 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
5632 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
5634 /* Information on the contents of registers. These are pointers into
5635 the input strings; they record just what was matched (on this
5636 attempt) by a subexpression part of the pattern, that is, the
5637 regnum-th regstart pointer points to where in the pattern we began
5638 matching and the regnum-th regend points to right after where we
5639 stopped matching the regnum-th subexpression. (The zeroth register
5640 keeps track of what the whole pattern matches.) */
5641 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5642 const CHAR_T **regstart, **regend;
5645 /* If a group that's operated upon by a repetition operator fails to
5646 match anything, then the register for its start will need to be
5647 restored because it will have been set to wherever in the string we
5648 are when we last see its open-group operator. Similarly for a
5650 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5651 const CHAR_T **old_regstart, **old_regend;
5654 /* The is_active field of reg_info helps us keep track of which (possibly
5655 nested) subexpressions we are currently in. The matched_something
5656 field of reg_info[reg_num] helps us tell whether or not we have
5657 matched any of the pattern so far this time through the reg_num-th
5658 subexpression. These two fields get reset each time through any
5659 loop their register is in. */
5660 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5661 PREFIX(register_info_type) *reg_info;
5664 /* The following record the register info as found in the above
5665 variables when we find a match better than any we've seen before.
5666 This happens as we backtrack through the failure points, which in
5667 turn happens only if we have not yet matched the entire string. */
5668 unsigned best_regs_set = false;
5669 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5670 const CHAR_T **best_regstart, **best_regend;
5673 /* Logically, this is `best_regend[0]'. But we don't want to have to
5674 allocate space for that if we're not allocating space for anything
5675 else (see below). Also, we never need info about register 0 for
5676 any of the other register vectors, and it seems rather a kludge to
5677 treat `best_regend' differently than the rest. So we keep track of
5678 the end of the best match so far in a separate variable. We
5679 initialize this to NULL so that when we backtrack the first time
5680 and need to test it, it's not garbage. */
5681 const CHAR_T *match_end = NULL;
5683 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5684 int set_regs_matched_done = 0;
5686 /* Used when we pop values we don't care about. */
5687 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5688 const CHAR_T **reg_dummy;
5689 PREFIX(register_info_type) *reg_info_dummy;
5693 /* Counts the total number of registers pushed. */
5694 unsigned num_regs_pushed = 0;
5697 /* Definitions for state transitions. More efficiently for gcc. */
5699 # if defined HAVE_SUBTRACT_LOCAL_LABELS && defined SHARED
5704 const void *__unbounded ptr; \
5705 offset = (p == pend \
5706 ? 0 : jmptable[SWITCH_ENUM_CAST ((re_opcode_t) *p++)]); \
5707 ptr = &&end_of_pattern + offset; \
5712 &&label_##x - &&end_of_pattern
5713 # define JUMP_TABLE_TYPE const int
5718 const void *__unbounded ptr; \
5719 ptr = (p == pend ? &&end_of_pattern \
5720 : jmptable[SWITCH_ENUM_CAST ((re_opcode_t) *p++)]); \
5726 # define JUMP_TABLE_TYPE const void *const
5728 # define CASE(x) label_##x
5729 static JUMP_TABLE_TYPE jmptable[] =
5748 REF (jump_past_alt),
5749 REF (on_failure_jump),
5750 REF (on_failure_keep_string_jump),
5751 REF (pop_failure_jump),
5752 REF (maybe_pop_jump),
5753 REF (dummy_failure_jump),
5754 REF (push_dummy_failure),
5757 REF (set_number_at),
5779 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5783 #ifdef MATCH_MAY_ALLOCATE
5784 /* Do not bother to initialize all the register variables if there are
5785 no groups in the pattern, as it takes a fair amount of time. If
5786 there are groups, we include space for register 0 (the whole
5787 pattern), even though we never use it, since it simplifies the
5788 array indexing. We should fix this. */
5791 regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5792 regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5793 old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5794 old_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5795 best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5796 best_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5797 reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5798 reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *);
5799 reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5801 if (!(regstart && regend && old_regstart && old_regend && reg_info
5802 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
5810 /* We must initialize all our variables to NULL, so that
5811 `FREE_VARIABLES' doesn't try to free them. */
5812 regstart = regend = old_regstart = old_regend = best_regstart
5813 = best_regend = reg_dummy = NULL;
5814 reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL;
5816 #endif /* MATCH_MAY_ALLOCATE */
5818 /* The starting position is bogus. */
5820 if (pos < 0 || pos > csize1 + csize2)
5822 if (pos < 0 || pos > size1 + size2)
5830 /* Allocate wchar_t array for string1 and string2 and
5831 fill them with converted string. */
5832 if (string1 == NULL && string2 == NULL)
5834 /* We need seting up buffers here. */
5836 /* We must free wcs buffers in this function. */
5837 cant_free_wcs_buf = 0;
5841 string1 = REGEX_TALLOC (csize1 + 1, CHAR_T);
5842 mbs_offset1 = REGEX_TALLOC (csize1 + 1, int);
5843 is_binary = REGEX_TALLOC (csize1 + 1, char);
5844 if (!string1 || !mbs_offset1 || !is_binary)
5847 FREE_VAR (mbs_offset1);
5848 FREE_VAR (is_binary);
5854 string2 = REGEX_TALLOC (csize2 + 1, CHAR_T);
5855 mbs_offset2 = REGEX_TALLOC (csize2 + 1, int);
5856 is_binary = REGEX_TALLOC (csize2 + 1, char);
5857 if (!string2 || !mbs_offset2 || !is_binary)
5860 FREE_VAR (mbs_offset1);
5862 FREE_VAR (mbs_offset2);
5863 FREE_VAR (is_binary);
5866 size2 = convert_mbs_to_wcs(string2, cstring2, csize2,
5867 mbs_offset2, is_binary);
5868 string2[size2] = L'\0'; /* for a sentinel */
5869 FREE_VAR (is_binary);
5873 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5874 pattern to (char*) in regex_compile. */
5875 p = pattern = (CHAR_T*)bufp->buffer;
5876 pend = (CHAR_T*)(bufp->buffer + bufp->used);
5880 /* Initialize subexpression text positions to -1 to mark ones that no
5881 start_memory/stop_memory has been seen for. Also initialize the
5882 register information struct. */
5883 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5885 regstart[mcnt] = regend[mcnt]
5886 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
5888 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
5889 IS_ACTIVE (reg_info[mcnt]) = 0;
5890 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5891 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5894 /* We move `string1' into `string2' if the latter's empty -- but not if
5895 `string1' is null. */
5896 if (size2 == 0 && string1 != NULL)
5903 mbs_offset2 = mbs_offset1;
5909 end1 = string1 + size1;
5910 end2 = string2 + size2;
5912 /* Compute where to stop matching, within the two strings. */
5916 mcnt = count_mbs_length(mbs_offset1, stop);
5917 end_match_1 = string1 + mcnt;
5918 end_match_2 = string2;
5922 if (stop > csize1 + csize2)
5923 stop = csize1 + csize2;
5925 mcnt = count_mbs_length(mbs_offset2, stop-csize1);
5926 end_match_2 = string2 + mcnt;
5929 { /* count_mbs_length return error. */
5936 end_match_1 = string1 + stop;
5937 end_match_2 = string2;
5942 end_match_2 = string2 + stop - size1;
5946 /* `p' scans through the pattern as `d' scans through the data.
5947 `dend' is the end of the input string that `d' points within. `d'
5948 is advanced into the following input string whenever necessary, but
5949 this happens before fetching; therefore, at the beginning of the
5950 loop, `d' can be pointing at the end of a string, but it cannot
5953 if (size1 > 0 && pos <= csize1)
5955 mcnt = count_mbs_length(mbs_offset1, pos);
5961 mcnt = count_mbs_length(mbs_offset2, pos-csize1);
5967 { /* count_mbs_length return error. */
5972 if (size1 > 0 && pos <= size1)
5979 d = string2 + pos - size1;
5984 DEBUG_PRINT1 ("The compiled pattern is:\n");
5985 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5986 DEBUG_PRINT1 ("The string to match is: `");
5987 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5988 DEBUG_PRINT1 ("'\n");
5990 /* This loops over pattern commands. It exits by returning from the
5991 function if the match is complete, or it drops through if the match
5992 fails at this starting point in the input data. */
5996 DEBUG_PRINT2 ("\n%p: ", p);
5998 DEBUG_PRINT2 ("\n0x%x: ", p);
6010 /* End of pattern means we might have succeeded. */
6011 DEBUG_PRINT1 ("end of pattern ... ");
6013 /* If we haven't matched the entire string, and we want the
6014 longest match, try backtracking. */
6015 if (d != end_match_2)
6017 /* 1 if this match is the best seen so far. */
6018 boolean best_match_p;
6021 /* 1 if this match ends in the same string (string1 or string2)
6022 as the best previous match. */
6023 boolean same_str_p = (FIRST_STRING_P (match_end)
6024 == MATCHING_IN_FIRST_STRING);
6026 /* AIX compiler got confused when this was combined
6027 with the previous declaration. */
6029 best_match_p = d > match_end;
6031 best_match_p = !MATCHING_IN_FIRST_STRING;
6034 DEBUG_PRINT1 ("backtracking.\n");
6036 if (!FAIL_STACK_EMPTY ())
6037 { /* More failure points to try. */
6039 /* If exceeds best match so far, save it. */
6040 if (!best_regs_set || best_match_p)
6042 best_regs_set = true;
6045 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6047 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6049 best_regstart[mcnt] = regstart[mcnt];
6050 best_regend[mcnt] = regend[mcnt];
6056 /* If no failure points, don't restore garbage. And if
6057 last match is real best match, don't restore second
6059 else if (best_regs_set && !best_match_p)
6062 /* Restore best match. It may happen that `dend ==
6063 end_match_1' while the restored d is in string2.
6064 For example, the pattern `x.*y.*z' against the
6065 strings `x-' and `y-z-', if the two strings are
6066 not consecutive in memory. */
6067 DEBUG_PRINT1 ("Restoring best registers.\n");
6070 dend = ((d >= string1 && d <= end1)
6071 ? end_match_1 : end_match_2);
6073 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6075 regstart[mcnt] = best_regstart[mcnt];
6076 regend[mcnt] = best_regend[mcnt];
6079 } /* d != end_match_2 */
6082 DEBUG_PRINT1 ("Accepting match.\n");
6083 /* If caller wants register contents data back, do it. */
6084 if (regs && !bufp->no_sub)
6086 /* Have the register data arrays been allocated? */
6087 if (bufp->regs_allocated == REGS_UNALLOCATED)
6088 { /* No. So allocate them with malloc. We need one
6089 extra element beyond `num_regs' for the `-1' marker
6091 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
6092 regs->start = TALLOC (regs->num_regs, regoff_t);
6093 regs->end = TALLOC (regs->num_regs, regoff_t);
6094 if (regs->start == NULL || regs->end == NULL)
6099 bufp->regs_allocated = REGS_REALLOCATE;
6101 else if (bufp->regs_allocated == REGS_REALLOCATE)
6102 { /* Yes. If we need more elements than were already
6103 allocated, reallocate them. If we need fewer, just
6105 if (regs->num_regs < num_regs + 1)
6107 regs->num_regs = num_regs + 1;
6108 RETALLOC (regs->start, regs->num_regs, regoff_t);
6109 RETALLOC (regs->end, regs->num_regs, regoff_t);
6110 if (regs->start == NULL || regs->end == NULL)
6119 /* These braces fend off a "empty body in an else-statement"
6120 warning under GCC when assert expands to nothing. */
6121 assert (bufp->regs_allocated == REGS_FIXED);
6124 /* Convert the pointer data in `regstart' and `regend' to
6125 indices. Register zero has to be set differently,
6126 since we haven't kept track of any info for it. */
6127 if (regs->num_regs > 0)
6129 regs->start[0] = pos;
6131 if (MATCHING_IN_FIRST_STRING)
6132 regs->end[0] = (mbs_offset1 != NULL ?
6133 mbs_offset1[d-string1] : 0);
6135 regs->end[0] = csize1 + (mbs_offset2 != NULL
6136 ? mbs_offset2[d-string2] : 0);
6138 regs->end[0] = (MATCHING_IN_FIRST_STRING
6139 ? ((regoff_t) (d - string1))
6140 : ((regoff_t) (d - string2 + size1)));
6144 /* Go through the first `min (num_regs, regs->num_regs)'
6145 registers, since that is all we initialized. */
6146 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
6149 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
6150 regs->start[mcnt] = regs->end[mcnt] = -1;
6154 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
6156 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
6160 /* If the regs structure we return has more elements than
6161 were in the pattern, set the extra elements to -1. If
6162 we (re)allocated the registers, this is the case,
6163 because we always allocate enough to have at least one
6165 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
6166 regs->start[mcnt] = regs->end[mcnt] = -1;
6167 } /* regs && !bufp->no_sub */
6169 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6170 nfailure_points_pushed, nfailure_points_popped,
6171 nfailure_points_pushed - nfailure_points_popped);
6172 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
6175 if (MATCHING_IN_FIRST_STRING)
6176 mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0;
6178 mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) +
6182 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
6183 ? string1 : string2 - size1);
6186 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
6193 /* Otherwise match next pattern command. */
6194 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
6197 /* Ignore these. Used to ignore the n of succeed_n's which
6198 currently have n == 0. */
6200 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6204 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6207 /* Match the next n pattern characters exactly. The following
6208 byte in the pattern defines n, and the n bytes after that
6209 are the characters to match. */
6215 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
6217 /* This is written out as an if-else so we don't waste time
6218 testing `translate' inside the loop. */
6227 if ((UCHAR_T) translate[(unsigned char) *d++]
6233 if (*d++ != (CHAR_T) *p++)
6237 if ((UCHAR_T) translate[(unsigned char) *d++]
6249 if (*d++ != (CHAR_T) *p++) goto fail;
6253 SET_REGS_MATCHED ();
6257 /* Match any character except possibly a newline or a null. */
6259 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6263 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
6264 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
6267 SET_REGS_MATCHED ();
6268 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d);
6278 unsigned int i, char_class_length, coll_symbol_length,
6279 equiv_class_length, ranges_length, chars_length, length;
6280 CHAR_T *workp, *workp2, *charset_top;
6281 #define WORK_BUFFER_SIZE 128
6282 CHAR_T str_buf[WORK_BUFFER_SIZE];
6287 boolean negate = (re_opcode_t) *(p - 1) == charset_not;
6289 DEBUG_PRINT2 ("EXECUTING charset%s.\n", negate ? "_not" : "");
6291 c = TRANSLATE (*d); /* The character to match. */
6294 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
6296 charset_top = p - 1;
6297 char_class_length = *p++;
6298 coll_symbol_length = *p++;
6299 equiv_class_length = *p++;
6300 ranges_length = *p++;
6301 chars_length = *p++;
6302 /* p points charset[6], so the address of the next instruction
6303 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6304 where l=length of char_classes, m=length of collating_symbol,
6305 n=equivalence_class, o=length of char_range,
6306 p'=length of character. */
6308 /* Update p to indicate the next instruction. */
6309 p += char_class_length + coll_symbol_length+ equiv_class_length +
6310 2*ranges_length + chars_length;
6312 /* match with char_class? */
6313 for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE)
6316 uintptr_t alignedp = ((uintptr_t)workp
6317 + __alignof__(wctype_t) - 1)
6318 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6319 wctype = *((wctype_t*)alignedp);
6320 workp += CHAR_CLASS_SIZE;
6321 if (iswctype((wint_t)c, wctype))
6322 goto char_set_matched;
6325 /* match with collating_symbol? */
6329 const unsigned char *extra = (const unsigned char *)
6330 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
6332 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;
6336 wextra = (int32_t*)(extra + *workp++);
6337 for (i = 0; i < *wextra; ++i)
6338 if (TRANSLATE(d[i]) != wextra[1 + i])
6343 /* Update d, however d will be incremented at
6344 char_set_matched:, we decrement d here. */
6346 goto char_set_matched;
6350 else /* (nrules == 0) */
6352 /* If we can't look up collation data, we use wcscoll
6355 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;)
6357 const CHAR_T *backup_d = d, *backup_dend = dend;
6358 length = wcslen (workp);
6360 /* If wcscoll(the collating symbol, whole string) > 0,
6361 any substring of the string never match with the
6362 collating symbol. */
6363 if (wcscoll (workp, d) > 0)
6365 workp += length + 1;
6369 /* First, we compare the collating symbol with
6370 the first character of the string.
6371 If it don't match, we add the next character to
6372 the compare buffer in turn. */
6373 for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++)
6378 if (dend == end_match_2)
6384 /* add next character to the compare buffer. */
6385 str_buf[i] = TRANSLATE(*d);
6386 str_buf[i+1] = '\0';
6388 match = wcscoll (workp, str_buf);
6390 goto char_set_matched;
6393 /* (str_buf > workp) indicate (str_buf + X > workp),
6394 because for all X (str_buf + X > str_buf).
6395 So we don't need continue this loop. */
6398 /* Otherwise(str_buf < workp),
6399 (str_buf+next_character) may equals (workp).
6400 So we continue this loop. */
6405 workp += length + 1;
6408 /* match with equivalence_class? */
6412 const CHAR_T *backup_d = d, *backup_dend = dend;
6413 /* Try to match the equivalence class against
6414 those known to the collate implementation. */
6415 const int32_t *table;
6416 const int32_t *weights;
6417 const int32_t *extra;
6418 const int32_t *indirect;
6423 /* This #include defines a local function! */
6424 # include <locale/weightwc.h>
6426 table = (const int32_t *)
6427 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
6428 weights = (const wint_t *)
6429 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
6430 extra = (const wint_t *)
6431 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
6432 indirect = (const int32_t *)
6433 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
6435 /* Write 1 collating element to str_buf, and
6439 for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++)
6441 cp = (wint_t*)str_buf;
6444 if (dend == end_match_2)
6449 str_buf[i] = TRANSLATE(*(d+i));
6450 str_buf[i+1] = '\0'; /* sentinel */
6451 idx2 = findidx ((const wint_t**)&cp);
6454 /* Update d, however d will be incremented at
6455 char_set_matched:, we decrement d here. */
6456 d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1);
6459 if (dend == end_match_2)
6468 len = weights[idx2];
6470 for (workp2 = workp + equiv_class_length ; workp < workp2 ;
6473 idx = (int32_t)*workp;
6474 /* We already checked idx != 0 in regex_compile. */
6476 if (idx2 != 0 && len == weights[idx])
6479 while (cnt < len && (weights[idx + 1 + cnt]
6480 == weights[idx2 + 1 + cnt]))
6484 goto char_set_matched;
6491 else /* (nrules == 0) */
6493 /* If we can't look up collation data, we use wcscoll
6496 for (workp2 = workp + equiv_class_length ; workp < workp2 ;)
6498 const CHAR_T *backup_d = d, *backup_dend = dend;
6499 length = wcslen (workp);
6501 /* If wcscoll(the collating symbol, whole string) > 0,
6502 any substring of the string never match with the
6503 collating symbol. */
6504 if (wcscoll (workp, d) > 0)
6506 workp += length + 1;
6510 /* First, we compare the equivalence class with
6511 the first character of the string.
6512 If it don't match, we add the next character to
6513 the compare buffer in turn. */
6514 for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++)
6519 if (dend == end_match_2)
6525 /* add next character to the compare buffer. */
6526 str_buf[i] = TRANSLATE(*d);
6527 str_buf[i+1] = '\0';
6529 match = wcscoll (workp, str_buf);
6532 goto char_set_matched;
6535 /* (str_buf > workp) indicate (str_buf + X > workp),
6536 because for all X (str_buf + X > str_buf).
6537 So we don't need continue this loop. */
6540 /* Otherwise(str_buf < workp),
6541 (str_buf+next_character) may equals (workp).
6542 So we continue this loop. */
6547 workp += length + 1;
6551 /* match with char_range? */
6555 uint32_t collseqval;
6556 const char *collseq = (const char *)
6557 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
6559 collseqval = collseq_table_lookup (collseq, c);
6561 for (; workp < p - chars_length ;)
6563 uint32_t start_val, end_val;
6565 /* We already compute the collation sequence value
6566 of the characters (or collating symbols). */
6567 start_val = (uint32_t) *workp++; /* range_start */
6568 end_val = (uint32_t) *workp++; /* range_end */
6570 if (start_val <= collseqval && collseqval <= end_val)
6571 goto char_set_matched;
6577 /* We set range_start_char at str_buf[0], range_end_char
6578 at str_buf[4], and compared char at str_buf[2]. */
6583 for (; workp < p - chars_length ;)
6585 wchar_t *range_start_char, *range_end_char;
6587 /* match if (range_start_char <= c <= range_end_char). */
6589 /* If range_start(or end) < 0, we assume -range_start(end)
6590 is the offset of the collating symbol which is specified
6591 as the character of the range start(end). */
6595 range_start_char = charset_top - (*workp++);
6598 str_buf[0] = *workp++;
6599 range_start_char = str_buf;
6604 range_end_char = charset_top - (*workp++);
6607 str_buf[4] = *workp++;
6608 range_end_char = str_buf + 4;
6611 if (wcscoll (range_start_char, str_buf+2) <= 0
6612 && wcscoll (str_buf+2, range_end_char) <= 0)
6613 goto char_set_matched;
6617 /* match with char? */
6618 for (; workp < p ; workp++)
6620 goto char_set_matched;
6625 if (negate) goto fail;
6627 /* Cast to `unsigned' instead of `unsigned char' in case the
6628 bit list is a full 32 bytes long. */
6629 if (c < (unsigned) (*p * BYTEWIDTH)
6630 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6635 if (!negate) goto fail;
6636 #undef WORK_BUFFER_SIZE
6638 SET_REGS_MATCHED ();
6644 /* The beginning of a group is represented by start_memory.
6645 The arguments are the register number in the next byte, and the
6646 number of groups inner to this one in the next. The text
6647 matched within the group is recorded (in the internal
6648 registers data structure) under the register number. */
6649 CASE (start_memory):
6650 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6651 (long int) *p, (long int) p[1]);
6653 /* Find out if this group can match the empty string. */
6654 p1 = p; /* To send to group_match_null_string_p. */
6656 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
6657 REG_MATCH_NULL_STRING_P (reg_info[*p])
6658 = PREFIX(group_match_null_string_p) (&p1, pend, reg_info);
6660 /* Save the position in the string where we were the last time
6661 we were at this open-group operator in case the group is
6662 operated upon by a repetition operator, e.g., with `(a*)*b'
6663 against `ab'; then we want to ignore where we are now in
6664 the string in case this attempt to match fails. */
6665 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6666 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
6668 DEBUG_PRINT2 (" old_regstart: %d\n",
6669 POINTER_TO_OFFSET (old_regstart[*p]));
6672 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
6674 IS_ACTIVE (reg_info[*p]) = 1;
6675 MATCHED_SOMETHING (reg_info[*p]) = 0;
6677 /* Clear this whenever we change the register activity status. */
6678 set_regs_matched_done = 0;
6680 /* This is the new highest active register. */
6681 highest_active_reg = *p;
6683 /* If nothing was active before, this is the new lowest active
6685 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6686 lowest_active_reg = *p;
6688 /* Move past the register number and inner group count. */
6690 just_past_start_mem = p;
6695 /* The stop_memory opcode represents the end of a group. Its
6696 arguments are the same as start_memory's: the register
6697 number, and the number of inner groups. */
6699 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6700 (long int) *p, (long int) p[1]);
6702 /* We need to save the string position the last time we were at
6703 this close-group operator in case the group is operated
6704 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6705 against `aba'; then we want to ignore where we are now in
6706 the string in case this attempt to match fails. */
6707 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6708 ? REG_UNSET (regend[*p]) ? d : regend[*p]
6710 DEBUG_PRINT2 (" old_regend: %d\n",
6711 POINTER_TO_OFFSET (old_regend[*p]));
6714 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
6716 /* This register isn't active anymore. */
6717 IS_ACTIVE (reg_info[*p]) = 0;
6719 /* Clear this whenever we change the register activity status. */
6720 set_regs_matched_done = 0;
6722 /* If this was the only register active, nothing is active
6724 if (lowest_active_reg == highest_active_reg)
6726 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6727 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6730 { /* We must scan for the new highest active register, since
6731 it isn't necessarily one less than now: consider
6732 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6733 new highest active register is 1. */
6735 while (r > 0 && !IS_ACTIVE (reg_info[r]))
6738 /* If we end up at register zero, that means that we saved
6739 the registers as the result of an `on_failure_jump', not
6740 a `start_memory', and we jumped to past the innermost
6741 `stop_memory'. For example, in ((.)*) we save
6742 registers 1 and 2 as a result of the *, but when we pop
6743 back to the second ), we are at the stop_memory 1.
6744 Thus, nothing is active. */
6747 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6748 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6751 highest_active_reg = r;
6754 /* If just failed to match something this time around with a
6755 group that's operated on by a repetition operator, try to
6756 force exit from the ``loop'', and restore the register
6757 information for this group that we had before trying this
6759 if ((!MATCHED_SOMETHING (reg_info[*p])
6760 || just_past_start_mem == p - 1)
6763 boolean is_a_jump_n = false;
6767 switch ((re_opcode_t) *p1++)
6771 case pop_failure_jump:
6772 case maybe_pop_jump:
6774 case dummy_failure_jump:
6775 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6777 p1 += OFFSET_ADDRESS_SIZE;
6785 /* If the next operation is a jump backwards in the pattern
6786 to an on_failure_jump right before the start_memory
6787 corresponding to this stop_memory, exit from the loop
6788 by forcing a failure after pushing on the stack the
6789 on_failure_jump's jump in the pattern, and d. */
6790 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
6791 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory
6792 && p1[2+OFFSET_ADDRESS_SIZE] == *p)
6794 /* If this group ever matched anything, then restore
6795 what its registers were before trying this last
6796 failed match, e.g., with `(a*)*b' against `ab' for
6797 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6798 against `aba' for regend[3].
6800 Also restore the registers for inner groups for,
6801 e.g., `((a*)(b*))*' against `aba' (register 3 would
6802 otherwise get trashed). */
6804 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
6808 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
6810 /* Restore this and inner groups' (if any) registers. */
6811 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
6814 regstart[r] = old_regstart[r];
6816 /* xx why this test? */
6817 if (old_regend[r] >= regstart[r])
6818 regend[r] = old_regend[r];
6822 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6823 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
6829 /* Move past the register number and the inner group count. */
6834 /* \<digit> has been turned into a `duplicate' command which is
6835 followed by the numeric value of <digit> as the register number. */
6838 register const CHAR_T *d2, *dend2;
6839 int regno = *p++; /* Get which register to match against. */
6840 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
6842 /* Can't back reference a group which we've never matched. */
6843 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
6846 /* Where in input to try to start matching. */
6847 d2 = regstart[regno];
6849 /* Where to stop matching; if both the place to start and
6850 the place to stop matching are in the same string, then
6851 set to the place to stop, otherwise, for now have to use
6852 the end of the first string. */
6854 dend2 = ((FIRST_STRING_P (regstart[regno])
6855 == FIRST_STRING_P (regend[regno]))
6856 ? regend[regno] : end_match_1);
6859 /* If necessary, advance to next segment in register
6863 if (dend2 == end_match_2) break;
6864 if (dend2 == regend[regno]) break;
6866 /* End of string1 => advance to string2. */
6868 dend2 = regend[regno];
6870 /* At end of register contents => success */
6871 if (d2 == dend2) break;
6873 /* If necessary, advance to next segment in data. */
6876 /* How many characters left in this segment to match. */
6879 /* Want how many consecutive characters we can match in
6880 one shot, so, if necessary, adjust the count. */
6881 if (mcnt > dend2 - d2)
6884 /* Compare that many; failure if mismatch, else move
6887 ? PREFIX(bcmp_translate) (d, d2, mcnt, translate)
6888 : memcmp (d, d2, mcnt*sizeof(UCHAR_T)))
6890 d += mcnt, d2 += mcnt;
6892 /* Do this because we've match some characters. */
6893 SET_REGS_MATCHED ();
6899 /* begline matches the empty string at the beginning of the string
6900 (unless `not_bol' is set in `bufp'), and, if
6901 `newline_anchor' is set, after newlines. */
6903 DEBUG_PRINT1 ("EXECUTING begline.\n");
6905 if (AT_STRINGS_BEG (d))
6912 else if (d[-1] == '\n' && bufp->newline_anchor)
6916 /* In all other cases, we fail. */
6920 /* endline is the dual of begline. */
6922 DEBUG_PRINT1 ("EXECUTING endline.\n");
6924 if (AT_STRINGS_END (d))
6932 /* We have to ``prefetch'' the next character. */
6933 else if ((d == end1 ? *string2 : *d) == '\n'
6934 && bufp->newline_anchor)
6941 /* Match at the very beginning of the data. */
6943 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
6944 if (AT_STRINGS_BEG (d))
6951 /* Match at the very end of the data. */
6953 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
6954 if (AT_STRINGS_END (d))
6961 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
6962 pushes NULL as the value for the string on the stack. Then
6963 `pop_failure_point' will keep the current value for the
6964 string, instead of restoring it. To see why, consider
6965 matching `foo\nbar' against `.*\n'. The .* matches the foo;
6966 then the . fails against the \n. But the next thing we want
6967 to do is match the \n against the \n; if we restored the
6968 string value, we would be back at the foo.
6970 Because this is used only in specific cases, we don't need to
6971 check all the things that `on_failure_jump' does, to make
6972 sure the right things get saved on the stack. Hence we don't
6973 share its code. The only reason to push anything on the
6974 stack at all is that otherwise we would have to change
6975 `anychar's code to do something besides goto fail in this
6976 case; that seems worse than this. */
6977 CASE (on_failure_keep_string_jump):
6978 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
6980 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6982 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
6984 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
6987 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
6991 /* Uses of on_failure_jump:
6993 Each alternative starts with an on_failure_jump that points
6994 to the beginning of the next alternative. Each alternative
6995 except the last ends with a jump that in effect jumps past
6996 the rest of the alternatives. (They really jump to the
6997 ending jump of the following alternative, because tensioning
6998 these jumps is a hassle.)
7000 Repeats start with an on_failure_jump that points past both
7001 the repetition text and either the following jump or
7002 pop_failure_jump back to this on_failure_jump. */
7003 CASE (on_failure_jump):
7005 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
7007 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7009 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
7011 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
7014 /* If this on_failure_jump comes right before a group (i.e.,
7015 the original * applied to a group), save the information
7016 for that group and all inner ones, so that if we fail back
7017 to this point, the group's information will be correct.
7018 For example, in \(a*\)*\1, we need the preceding group,
7019 and in \(zz\(a*\)b*\)\2, we need the inner group. */
7021 /* We can't use `p' to check ahead because we push
7022 a failure point to `p + mcnt' after we do this. */
7025 /* We need to skip no_op's before we look for the
7026 start_memory in case this on_failure_jump is happening as
7027 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7029 while (p1 < pend && (re_opcode_t) *p1 == no_op)
7032 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
7034 /* We have a new highest active register now. This will
7035 get reset at the start_memory we are about to get to,
7036 but we will have saved all the registers relevant to
7037 this repetition op, as described above. */
7038 highest_active_reg = *(p1 + 1) + *(p1 + 2);
7039 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
7040 lowest_active_reg = *(p1 + 1);
7043 DEBUG_PRINT1 (":\n");
7044 PUSH_FAILURE_POINT (p + mcnt, d, -2);
7048 /* A smart repeat ends with `maybe_pop_jump'.
7049 We change it to either `pop_failure_jump' or `jump'. */
7050 CASE (maybe_pop_jump):
7051 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7052 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
7054 register UCHAR_T *p2 = p;
7056 /* Compare the beginning of the repeat with what in the
7057 pattern follows its end. If we can establish that there
7058 is nothing that they would both match, i.e., that we
7059 would have to backtrack because of (as in, e.g., `a*a')
7060 then we can change to pop_failure_jump, because we'll
7061 never have to backtrack.
7063 This is not true in the case of alternatives: in
7064 `(a|ab)*' we do need to backtrack to the `ab' alternative
7065 (e.g., if the string was `ab'). But instead of trying to
7066 detect that here, the alternative has put on a dummy
7067 failure point which is what we will end up popping. */
7069 /* Skip over open/close-group commands.
7070 If what follows this loop is a ...+ construct,
7071 look at what begins its body, since we will have to
7072 match at least one of that. */
7076 && ((re_opcode_t) *p2 == stop_memory
7077 || (re_opcode_t) *p2 == start_memory))
7079 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend
7080 && (re_opcode_t) *p2 == dummy_failure_jump)
7081 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE;
7087 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7088 to the `maybe_finalize_jump' of this case. Examine what
7091 /* If we're at the end of the pattern, we can change. */
7094 /* Consider what happens when matching ":\(.*\)"
7095 against ":/". I don't really understand this code
7097 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7100 (" End of pattern: change to `pop_failure_jump'.\n");
7103 else if ((re_opcode_t) *p2 == exactn
7105 || (re_opcode_t) *p2 == exactn_bin
7107 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
7110 = *p2 == (UCHAR_T) endline ? '\n' : p2[2];
7112 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn
7114 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin
7116 ) && p1[3+OFFSET_ADDRESS_SIZE] != c)
7118 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7121 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7123 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]);
7125 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7127 (char) p1[3+OFFSET_ADDRESS_SIZE]);
7132 else if ((re_opcode_t) p1[3] == charset
7133 || (re_opcode_t) p1[3] == charset_not)
7135 int negate = (re_opcode_t) p1[3] == charset_not;
7137 if (c < (unsigned) (p1[4] * BYTEWIDTH)
7138 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
7141 /* `negate' is equal to 1 if c would match, which means
7142 that we can't change to pop_failure_jump. */
7145 p[-3] = (unsigned char) pop_failure_jump;
7146 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7149 #endif /* not WCHAR */
7152 else if ((re_opcode_t) *p2 == charset)
7154 /* We win if the first character of the loop is not part
7156 if ((re_opcode_t) p1[3] == exactn
7157 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
7158 && (p2[2 + p1[5] / BYTEWIDTH]
7159 & (1 << (p1[5] % BYTEWIDTH)))))
7161 p[-3] = (unsigned char) pop_failure_jump;
7162 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7165 else if ((re_opcode_t) p1[3] == charset_not)
7168 /* We win if the charset_not inside the loop
7169 lists every character listed in the charset after. */
7170 for (idx = 0; idx < (int) p2[1]; idx++)
7171 if (! (p2[2 + idx] == 0
7172 || (idx < (int) p1[4]
7173 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
7178 p[-3] = (unsigned char) pop_failure_jump;
7179 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7182 else if ((re_opcode_t) p1[3] == charset)
7185 /* We win if the charset inside the loop
7186 has no overlap with the one after the loop. */
7188 idx < (int) p2[1] && idx < (int) p1[4];
7190 if ((p2[2 + idx] & p1[5 + idx]) != 0)
7193 if (idx == p2[1] || idx == p1[4])
7195 p[-3] = (unsigned char) pop_failure_jump;
7196 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7200 #endif /* not WCHAR */
7202 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */
7203 if ((re_opcode_t) p[-1] != pop_failure_jump)
7205 p[-1] = (UCHAR_T) jump;
7206 DEBUG_PRINT1 (" Match => jump.\n");
7207 goto unconditional_jump;
7209 /* Note fall through. */
7212 /* The end of a simple repeat has a pop_failure_jump back to
7213 its matching on_failure_jump, where the latter will push a
7214 failure point. The pop_failure_jump takes off failure
7215 points put on by this pop_failure_jump's matching
7216 on_failure_jump; we got through the pattern to here from the
7217 matching on_failure_jump, so didn't fail. */
7218 CASE (pop_failure_jump):
7220 /* We need to pass separate storage for the lowest and
7221 highest registers, even though we don't care about the
7222 actual values. Otherwise, we will restore only one
7223 register from the stack, since lowest will == highest in
7224 `pop_failure_point'. */
7225 active_reg_t dummy_low_reg, dummy_high_reg;
7226 UCHAR_T *pdummy = NULL;
7227 const CHAR_T *sdummy = NULL;
7229 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7230 POP_FAILURE_POINT (sdummy, pdummy,
7231 dummy_low_reg, dummy_high_reg,
7232 reg_dummy, reg_dummy, reg_info_dummy);
7234 /* Note fall through. */
7238 DEBUG_PRINT2 ("\n%p: ", p);
7240 DEBUG_PRINT2 ("\n0x%x: ", p);
7242 /* Note fall through. */
7244 /* Unconditionally jump (without popping any failure points). */
7246 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
7247 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
7248 p += mcnt; /* Do the jump. */
7250 DEBUG_PRINT2 ("(to %p).\n", p);
7252 DEBUG_PRINT2 ("(to 0x%x).\n", p);
7257 /* We need this opcode so we can detect where alternatives end
7258 in `group_match_null_string_p' et al. */
7259 CASE (jump_past_alt):
7260 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7261 goto unconditional_jump;
7264 /* Normally, the on_failure_jump pushes a failure point, which
7265 then gets popped at pop_failure_jump. We will end up at
7266 pop_failure_jump, also, and with a pattern of, say, `a+', we
7267 are skipping over the on_failure_jump, so we have to push
7268 something meaningless for pop_failure_jump to pop. */
7269 CASE (dummy_failure_jump):
7270 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7271 /* It doesn't matter what we push for the string here. What
7272 the code at `fail' tests is the value for the pattern. */
7273 PUSH_FAILURE_POINT (NULL, NULL, -2);
7274 goto unconditional_jump;
7277 /* At the end of an alternative, we need to push a dummy failure
7278 point in case we are followed by a `pop_failure_jump', because
7279 we don't want the failure point for the alternative to be
7280 popped. For example, matching `(a|ab)*' against `aab'
7281 requires that we match the `ab' alternative. */
7282 CASE (push_dummy_failure):
7283 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7284 /* See comments just above at `dummy_failure_jump' about the
7286 PUSH_FAILURE_POINT (NULL, NULL, -2);
7289 /* Have to succeed matching what follows at least n times.
7290 After that, handle like `on_failure_jump'. */
7292 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7293 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
7296 /* Originally, this is how many times we HAVE to succeed. */
7300 p += OFFSET_ADDRESS_SIZE;
7301 STORE_NUMBER_AND_INCR (p, mcnt);
7303 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE
7306 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE
7313 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7314 p + OFFSET_ADDRESS_SIZE);
7316 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7317 p + OFFSET_ADDRESS_SIZE);
7321 p[1] = (UCHAR_T) no_op;
7323 p[2] = (UCHAR_T) no_op;
7324 p[3] = (UCHAR_T) no_op;
7331 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7332 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
7334 /* Originally, this is how many times we CAN jump. */
7338 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt);
7341 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE,
7344 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE,
7347 goto unconditional_jump;
7349 /* If don't have to jump any more, skip over the rest of command. */
7351 p += 2 * OFFSET_ADDRESS_SIZE;
7354 CASE (set_number_at):
7356 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7358 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7360 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7362 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
7364 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
7366 STORE_NUMBER (p1, mcnt);
7371 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7372 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7373 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7374 macro and introducing temporary variables works around the bug. */
7377 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7378 if (AT_WORD_BOUNDARY (d))
7384 CASE (notwordbound):
7385 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7386 if (AT_WORD_BOUNDARY (d))
7392 boolean prevchar, thischar;
7394 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7395 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7400 prevchar = WORDCHAR_P (d - 1);
7401 thischar = WORDCHAR_P (d);
7402 if (prevchar != thischar)
7409 CASE (notwordbound):
7411 boolean prevchar, thischar;
7413 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7414 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7417 prevchar = WORDCHAR_P (d - 1);
7418 thischar = WORDCHAR_P (d);
7419 if (prevchar != thischar)
7426 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7427 if (!AT_STRINGS_END (d) && WORDCHAR_P (d)
7428 && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
7435 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7436 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
7437 && (AT_STRINGS_END (d) || !WORDCHAR_P (d)))
7445 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7446 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
7451 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7452 if (PTR_CHAR_POS ((unsigned char *) d) != point)
7457 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7458 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
7463 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
7468 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7472 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7474 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
7476 SET_REGS_MATCHED ();
7479 CASE (notsyntaxspec):
7480 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
7482 goto matchnotsyntax;
7485 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7489 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7491 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
7493 SET_REGS_MATCHED ();
7496 #else /* not emacs */
7498 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7500 if (!WORDCHAR_P (d))
7502 SET_REGS_MATCHED ();
7507 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7511 SET_REGS_MATCHED ();
7514 #endif /* not emacs */
7520 continue; /* Successfully executed one pattern command; keep going. */
7524 /* We goto here if a matching operation fails. */
7526 if (!FAIL_STACK_EMPTY ())
7527 { /* A restart point is known. Restore to that state. */
7528 DEBUG_PRINT1 ("\nFAIL:\n");
7529 POP_FAILURE_POINT (d, p,
7530 lowest_active_reg, highest_active_reg,
7531 regstart, regend, reg_info);
7533 /* If this failure point is a dummy, try the next one. */
7537 /* If we failed to the end of the pattern, don't examine *p. */
7541 boolean is_a_jump_n = false;
7543 /* If failed to a backwards jump that's part of a repetition
7544 loop, need to pop this failure point and use the next one. */
7545 switch ((re_opcode_t) *p)
7549 case maybe_pop_jump:
7550 case pop_failure_jump:
7553 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7556 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
7558 && (re_opcode_t) *p1 == on_failure_jump))
7566 if (d >= string1 && d <= end1)
7570 break; /* Matching at this starting point really fails. */
7574 goto restore_best_regs;
7578 return -1; /* Failure to match. */
7581 /* Subroutine definitions for re_match_2. */
7584 /* We are passed P pointing to a register number after a start_memory.
7586 Return true if the pattern up to the corresponding stop_memory can
7587 match the empty string, and false otherwise.
7589 If we find the matching stop_memory, sets P to point to one past its number.
7590 Otherwise, sets P to an undefined byte less than or equal to END.
7592 We don't handle duplicates properly (yet). */
7595 PREFIX(group_match_null_string_p) (UCHAR_T **p, UCHAR_T *end,
7596 PREFIX(register_info_type) *reg_info)
7599 /* Point to after the args to the start_memory. */
7600 UCHAR_T *p1 = *p + 2;
7604 /* Skip over opcodes that can match nothing, and return true or
7605 false, as appropriate, when we get to one that can't, or to the
7606 matching stop_memory. */
7608 switch ((re_opcode_t) *p1)
7610 /* Could be either a loop or a series of alternatives. */
7611 case on_failure_jump:
7613 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7615 /* If the next operation is not a jump backwards in the
7620 /* Go through the on_failure_jumps of the alternatives,
7621 seeing if any of the alternatives cannot match nothing.
7622 The last alternative starts with only a jump,
7623 whereas the rest start with on_failure_jump and end
7624 with a jump, e.g., here is the pattern for `a|b|c':
7626 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7627 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7630 So, we have to first go through the first (n-1)
7631 alternatives and then deal with the last one separately. */
7634 /* Deal with the first (n-1) alternatives, which start
7635 with an on_failure_jump (see above) that jumps to right
7636 past a jump_past_alt. */
7638 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] ==
7641 /* `mcnt' holds how many bytes long the alternative
7642 is, including the ending `jump_past_alt' and
7645 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt -
7646 (1 + OFFSET_ADDRESS_SIZE),
7650 /* Move to right after this alternative, including the
7654 /* Break if it's the beginning of an n-th alternative
7655 that doesn't begin with an on_failure_jump. */
7656 if ((re_opcode_t) *p1 != on_failure_jump)
7659 /* Still have to check that it's not an n-th
7660 alternative that starts with an on_failure_jump. */
7662 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7663 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] !=
7666 /* Get to the beginning of the n-th alternative. */
7667 p1 -= 1 + OFFSET_ADDRESS_SIZE;
7672 /* Deal with the last alternative: go back and get number
7673 of the `jump_past_alt' just before it. `mcnt' contains
7674 the length of the alternative. */
7675 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE);
7677 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info))
7680 p1 += mcnt; /* Get past the n-th alternative. */
7686 assert (p1[1] == **p);
7692 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7695 } /* while p1 < end */
7698 } /* group_match_null_string_p */
7701 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7702 It expects P to be the first byte of a single alternative and END one
7703 byte past the last. The alternative can contain groups. */
7706 PREFIX(alt_match_null_string_p) (UCHAR_T *p, UCHAR_T *end,
7707 PREFIX(register_info_type) *reg_info)
7714 /* Skip over opcodes that can match nothing, and break when we get
7715 to one that can't. */
7717 switch ((re_opcode_t) *p1)
7720 case on_failure_jump:
7722 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7727 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7730 } /* while p1 < end */
7733 } /* alt_match_null_string_p */
7736 /* Deals with the ops common to group_match_null_string_p and
7737 alt_match_null_string_p.
7739 Sets P to one after the op and its arguments, if any. */
7742 PREFIX(common_op_match_null_string_p) (UCHAR_T **p, UCHAR_T *end,
7743 PREFIX(register_info_type) *reg_info)
7750 switch ((re_opcode_t) *p1++)
7770 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
7771 ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info);
7773 /* Have to set this here in case we're checking a group which
7774 contains a group and a back reference to it. */
7776 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
7777 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
7783 /* If this is an optimized succeed_n for zero times, make the jump. */
7785 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7793 /* Get to the number of times to succeed. */
7794 p1 += OFFSET_ADDRESS_SIZE;
7795 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7799 p1 -= 2 * OFFSET_ADDRESS_SIZE;
7800 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7808 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
7813 p1 += 2 * OFFSET_ADDRESS_SIZE;
7816 /* All other opcodes mean we cannot match the empty string. */
7822 } /* common_op_match_null_string_p */
7825 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7826 bytes; nonzero otherwise. */
7829 PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2,
7831 RE_TRANSLATE_TYPE translate)
7833 register const UCHAR_T *p1 = (const UCHAR_T *) s1;
7834 register const UCHAR_T *p2 = (const UCHAR_T *) s2;
7838 if (((*p1<=0xff)?translate[*p1++]:*p1++)
7839 != ((*p2<=0xff)?translate[*p2++]:*p2++))
7842 if (translate[*p1++] != translate[*p2++]) return 1;
7850 #else /* not INSIDE_RECURSION */
7852 /* Entry points for GNU code. */
7854 /* re_compile_pattern is the GNU regular expression compiler: it
7855 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7856 Returns 0 if the pattern was valid, otherwise an error string.
7858 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7859 are set in BUFP on entry.
7861 We call regex_compile to do the actual compilation. */
7864 re_compile_pattern (const char *pattern,
7866 struct re_pattern_buffer *bufp)
7870 /* GNU code is written to assume at least RE_NREGS registers will be set
7871 (and at least one extra will be -1). */
7872 bufp->regs_allocated = REGS_UNALLOCATED;
7874 /* And GNU code determines whether or not to get register information
7875 by passing null for the REGS argument to re_match, etc., not by
7879 /* Match anchors at newline. */
7880 bufp->newline_anchor = 1;
7883 if (MB_CUR_MAX != 1)
7884 ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp);
7887 ret = byte_regex_compile (pattern, length, re_syntax_options, bufp);
7891 return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7894 weak_alias (__re_compile_pattern, re_compile_pattern)
7897 /* Entry points compatible with 4.2 BSD regex library. We don't define
7898 them unless specifically requested. */
7900 #if defined _REGEX_RE_COMP || defined _LIBC
7902 /* BSD has one and only one pattern buffer. */
7903 static struct re_pattern_buffer re_comp_buf;
7907 /* Make these definitions weak in libc, so POSIX programs can redefine
7908 these names if they don't use our functions, and still use
7909 regcomp/regexec below without link errors. */
7912 re_comp (const char *s)
7918 if (!re_comp_buf.buffer)
7919 return (char *) gettext ("No previous regular expression");
7923 if (!re_comp_buf.buffer)
7925 re_comp_buf.buffer = malloc (200);
7926 if (re_comp_buf.buffer == NULL)
7927 return (char *) gettext (re_error_msgid
7928 + re_error_msgid_idx[(int) REG_ESPACE]);
7929 re_comp_buf.allocated = 200;
7931 re_comp_buf.fastmap = malloc (1 << BYTEWIDTH);
7932 if (re_comp_buf.fastmap == NULL)
7933 return (char *) gettext (re_error_msgid
7934 + re_error_msgid_idx[(int) REG_ESPACE]);
7937 /* Since `re_exec' always passes NULL for the `regs' argument, we
7938 don't need to initialize the pattern buffer fields which affect it. */
7940 /* Match anchors at newlines. */
7941 re_comp_buf.newline_anchor = 1;
7944 if (MB_CUR_MAX != 1)
7945 ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7948 ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7953 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
7954 return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7962 re_exec (const char *s)
7964 const int len = strlen (s);
7966 0 <= re_search (&re_comp_buf, s, len, 0, len, 0);
7969 #endif /* _REGEX_RE_COMP */
7971 /* POSIX.2 functions. Don't define these for Emacs. */
7975 /* regcomp takes a regular expression as a string and compiles it.
7977 PREG is a regex_t *. We do not expect any fields to be initialized,
7978 since POSIX says we shouldn't. Thus, we set
7980 `buffer' to the compiled pattern;
7981 `used' to the length of the compiled pattern;
7982 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
7983 REG_EXTENDED bit in CFLAGS is set; otherwise, to
7984 RE_SYNTAX_POSIX_BASIC;
7985 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
7986 `fastmap' to an allocated space for the fastmap;
7987 `fastmap_accurate' to zero;
7988 `re_nsub' to the number of subexpressions in PATTERN.
7990 PATTERN is the address of the pattern string.
7992 CFLAGS is a series of bits which affect compilation.
7994 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
7995 use POSIX basic syntax.
7997 If REG_NEWLINE is set, then . and [^...] don't match newline.
7998 Also, regexec will try a match beginning after every newline.
8000 If REG_ICASE is set, then we considers upper- and lowercase
8001 versions of letters to be equivalent when matching.
8003 If REG_NOSUB is set, then when PREG is passed to regexec, that
8004 routine will report only success or failure, and nothing about the
8007 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
8008 the return codes and their meanings.) */
8011 regcomp (regex_t *preg, const char *pattern, int cflags)
8015 = (cflags & REG_EXTENDED) ?
8016 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
8018 /* regex_compile will allocate the space for the compiled pattern. */
8020 preg->allocated = 0;
8023 /* Try to allocate space for the fastmap. */
8024 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
8026 if (cflags & REG_ICASE)
8032 malloc (CHAR_SET_SIZE * sizeof (*(RE_TRANSLATE_TYPE)0));
8033 if (preg->translate == NULL)
8034 return (int) REG_ESPACE;
8036 /* Map uppercase characters to corresponding lowercase ones. */
8037 for (i = 0; i < CHAR_SET_SIZE; i++)
8038 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
8041 preg->translate = NULL;
8043 /* If REG_NEWLINE is set, newlines are treated differently. */
8044 if (cflags & REG_NEWLINE)
8045 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
8046 syntax &= ~RE_DOT_NEWLINE;
8047 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
8048 /* It also changes the matching behavior. */
8049 preg->newline_anchor = 1;
8052 preg->newline_anchor = 0;
8054 preg->no_sub = !!(cflags & REG_NOSUB);
8056 /* POSIX says a null character in the pattern terminates it, so we
8057 can use strlen here in compiling the pattern. */
8059 if (MB_CUR_MAX != 1)
8060 ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg);
8063 ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg);
8065 /* POSIX doesn't distinguish between an unmatched open-group and an
8066 unmatched close-group: both are REG_EPAREN. */
8067 if (ret == REG_ERPAREN) ret = REG_EPAREN;
8069 if (ret == REG_NOERROR && preg->fastmap)
8071 /* Compute the fastmap now, since regexec cannot modify the pattern
8073 if (re_compile_fastmap (preg) == -2)
8075 /* Some error occurred while computing the fastmap, just forget
8077 free (preg->fastmap);
8078 preg->fastmap = NULL;
8085 weak_alias (__regcomp, regcomp)
8089 /* regexec searches for a given pattern, specified by PREG, in the
8092 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8093 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8094 least NMATCH elements, and we set them to the offsets of the
8095 corresponding matched substrings.
8097 EFLAGS specifies `execution flags' which affect matching: if
8098 REG_NOTBOL is set, then ^ does not match at the beginning of the
8099 string; if REG_NOTEOL is set, then $ does not match at the end.
8101 We return 0 if we find a match and REG_NOMATCH if not. */
8104 regexec (const regex_t *preg, const char *string,
8105 size_t nmatch, regmatch_t pmatch[], int eflags)
8108 struct re_registers regs;
8109 regex_t private_preg;
8110 int len = strlen (string);
8111 boolean want_reg_info = !preg->no_sub && nmatch > 0;
8113 private_preg = *preg;
8115 private_preg.not_bol = !!(eflags & REG_NOTBOL);
8116 private_preg.not_eol = !!(eflags & REG_NOTEOL);
8118 /* The user has told us exactly how many registers to return
8119 information about, via `nmatch'. We have to pass that on to the
8120 matching routines. */
8121 private_preg.regs_allocated = REGS_FIXED;
8125 regs.num_regs = nmatch;
8126 regs.start = TALLOC (nmatch * 2, regoff_t);
8127 if (regs.start == NULL)
8128 return (int) REG_NOMATCH;
8129 regs.end = regs.start + nmatch;
8132 /* Perform the searching operation. */
8133 ret = re_search (&private_preg, string, len,
8134 /* start: */ 0, /* range: */ len,
8135 want_reg_info ? ®s : 0);
8137 /* Copy the register information to the POSIX structure. */
8144 for (r = 0; r < nmatch; r++)
8146 pmatch[r].rm_so = regs.start[r];
8147 pmatch[r].rm_eo = regs.end[r];
8151 /* If we needed the temporary register info, free the space now. */
8155 /* We want zero return to mean success, unlike `re_search'. */
8156 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
8159 weak_alias (__regexec, regexec)
8163 /* Returns a message corresponding to an error code, ERRCODE, returned
8164 from either regcomp or regexec. We don't use PREG here. */
8167 regerror (int errcode, const regex_t *preg, char *errbuf, size_t errbuf_size)
8173 || errcode >= (int) (sizeof (re_error_msgid_idx)
8174 / sizeof (re_error_msgid_idx[0])))
8175 /* Only error codes returned by the rest of the code should be passed
8176 to this routine. If we are given anything else, or if other regex
8177 code generates an invalid error code, then the program has a bug.
8178 Dump core so we can fix it. */
8181 msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
8183 msg_size = strlen (msg) + 1; /* Includes the null. */
8185 if (errbuf_size != 0)
8187 if (msg_size > errbuf_size)
8189 #if defined HAVE_MEMPCPY || defined _LIBC
8190 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
8192 memcpy (errbuf, msg, errbuf_size - 1);
8193 errbuf[errbuf_size - 1] = 0;
8197 memcpy (errbuf, msg, msg_size);
8203 weak_alias (__regerror, regerror)
8207 /* Free dynamically allocated space used by PREG. */
8210 regfree (regex_t *preg)
8212 if (preg->buffer != NULL)
8213 free (preg->buffer);
8214 preg->buffer = NULL;
8216 preg->allocated = 0;
8219 if (preg->fastmap != NULL)
8220 free (preg->fastmap);
8221 preg->fastmap = NULL;
8222 preg->fastmap_accurate = 0;
8224 if (preg->translate != NULL)
8225 free (preg->translate);
8226 preg->translate = NULL;
8229 weak_alias (__regfree, regfree)
8232 #endif /* not emacs */
8234 #endif /* not INSIDE_RECURSION */
8238 #undef STORE_NUMBER_AND_INCR
8239 #undef EXTRACT_NUMBER
8240 #undef EXTRACT_NUMBER_AND_INCR
8242 #undef DEBUG_PRINT_COMPILED_PATTERN
8243 #undef DEBUG_PRINT_DOUBLE_STRING
8245 #undef INIT_FAIL_STACK
8246 #undef RESET_FAIL_STACK
8247 #undef DOUBLE_FAIL_STACK
8248 #undef PUSH_PATTERN_OP
8249 #undef PUSH_FAILURE_POINTER
8250 #undef PUSH_FAILURE_INT
8251 #undef PUSH_FAILURE_ELT
8252 #undef POP_FAILURE_POINTER
8253 #undef POP_FAILURE_INT
8254 #undef POP_FAILURE_ELT
8257 #undef PUSH_FAILURE_POINT
8258 #undef POP_FAILURE_POINT
8260 #undef REG_UNSET_VALUE
8268 #undef INIT_BUF_SIZE
8269 #undef GET_BUFFER_SPACE
8277 #undef EXTEND_BUFFER
8278 #undef GET_UNSIGNED_NUMBER
8279 #undef FREE_STACK_RETURN
8281 # undef POINTER_TO_OFFSET
8282 # undef MATCHING_IN_FRST_STRING
8284 # undef AT_STRINGS_BEG
8285 # undef AT_STRINGS_END
8288 # undef FREE_VARIABLES
8289 # undef NO_HIGHEST_ACTIVE_REG
8290 # undef NO_LOWEST_ACTIVE_REG
8294 # undef COMPILED_BUFFER_VAR
8295 # undef OFFSET_ADDRESS_SIZE
8296 # undef CHAR_CLASS_SIZE
8303 # define DEFINED_ONCE