1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
5 Copyright (C) 1993-1999, 2000-2002 Free Software Foundation, Inc.
6 This file is part of the GNU C Library.
8 The GNU C Library is free software; you can redistribute it and/or
9 modify it under the terms of the GNU Lesser General Public
10 License as published by the Free Software Foundation; either
11 version 2.1 of the License, or (at your option) any later version.
13 The GNU C Library is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public
19 License along with the GNU C Library; if not, write to the Free
20 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
23 /* AIX requires this to be the first thing in the file. */
24 #if defined _AIX && !defined REGEX_MALLOC
36 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
37 # define PARAMS(args) args
39 # define PARAMS(args) ()
41 #endif /* Not PARAMS. */
43 #ifndef INSIDE_RECURSION
45 # if defined STDC_HEADERS && !defined emacs
48 /* We need this for `regex.h', and perhaps for the Emacs include files. */
49 # include <sys/types.h>
52 # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
54 /* For platform which support the ISO C amendement 1 functionality we
55 support user defined character classes. */
56 # if defined _LIBC || WIDE_CHAR_SUPPORT
57 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
63 /* We have to keep the namespace clean. */
64 # define regfree(preg) __regfree (preg)
65 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
66 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
67 # define regerror(errcode, preg, errbuf, errbuf_size) \
68 __regerror(errcode, preg, errbuf, errbuf_size)
69 # define re_set_registers(bu, re, nu, st, en) \
70 __re_set_registers (bu, re, nu, st, en)
71 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
72 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
73 # define re_match(bufp, string, size, pos, regs) \
74 __re_match (bufp, string, size, pos, regs)
75 # define re_search(bufp, string, size, startpos, range, regs) \
76 __re_search (bufp, string, size, startpos, range, regs)
77 # define re_compile_pattern(pattern, length, bufp) \
78 __re_compile_pattern (pattern, length, bufp)
79 # define re_set_syntax(syntax) __re_set_syntax (syntax)
80 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
81 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
82 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
84 # define btowc __btowc
85 # define iswctype __iswctype
86 # define mbrtowc __mbrtowc
87 # define wcslen __wcslen
88 # define wcscoll __wcscoll
89 # define wcrtomb __wcrtomb
91 /* We are also using some library internals. */
92 # include <locale/localeinfo.h>
93 # include <locale/elem-hash.h>
94 # include <langinfo.h>
95 # include <locale/coll-lookup.h>
101 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
102 /* This define is so xgettext can find the internationalizable strings. */
103 # define gettext_noop(msgid) msgid
105 /* This is for other GNU distributions with internationalized messages. */
106 # include "gettext.h"
109 /* Support for bounded pointers. */
110 # if !defined _LIBC && !defined __BOUNDED_POINTERS__
111 # define __bounded /* nothing */
112 # define __unbounded /* nothing */
113 # define __ptrvalue /* nothing */
116 /* The `emacs' switch turns on certain matching commands
117 that make sense only in Emacs. */
124 # else /* not emacs */
126 /* If we are not linking with Emacs proper,
127 we can't use the relocating allocator
128 even if config.h says that we can. */
131 # if defined STDC_HEADERS || defined _LIBC
138 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
139 If nothing else has been done, use the method below. */
140 # ifdef INHIBIT_STRING_HEADER
141 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
142 # if !defined bzero && !defined bcopy
143 # undef INHIBIT_STRING_HEADER
148 /* This is the normal way of making sure we have a bcopy and a bzero.
149 This is used in most programs--a few other programs avoid this
150 by defining INHIBIT_STRING_HEADER. */
151 # ifndef INHIBIT_STRING_HEADER
152 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
156 # define bzero(s, n) (memset (s, '\0', n), (s))
158 # define bzero(s, n) __bzero (s, n)
162 # include <strings.h>
164 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
167 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
172 /* Define the syntax stuff for \<, \>, etc. */
174 /* This must be nonzero for the wordchar and notwordchar pattern
175 commands in re_match_2. */
180 # ifdef SWITCH_ENUM_BUG
181 # define SWITCH_ENUM_CAST(x) ((int)(x))
183 # define SWITCH_ENUM_CAST(x) (x)
186 # endif /* not emacs */
188 # if defined _LIBC || HAVE_LIMITS_H
193 # define MB_LEN_MAX 1
196 /* Get the interface, including the syntax bits. */
199 /* isalpha etc. are used for the character classes. */
202 /* Jim Meyering writes:
204 "... Some ctype macros are valid only for character codes that
205 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
206 using /bin/cc or gcc but without giving an ansi option). So, all
207 ctype uses should be through macros like ISPRINT... If
208 STDC_HEADERS is defined, then autoconf has verified that the ctype
209 macros don't need to be guarded with references to isascii. ...
210 Defining isascii to 1 should let any compiler worth its salt
211 eliminate the && through constant folding."
212 Solaris defines some of these symbols so we must undefine them first. */
214 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
215 # define IN_CTYPE_DOMAIN(c) 1
217 # define IN_CTYPE_DOMAIN(c) isascii(c)
221 # define ISBLANK(c) (IN_CTYPE_DOMAIN (c) && isblank (c))
223 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
226 # define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isgraph (c))
228 # define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isprint (c) && !isspace (c))
232 # define ISPRINT(c) (IN_CTYPE_DOMAIN (c) && isprint (c))
233 # define ISDIGIT(c) (IN_CTYPE_DOMAIN (c) && isdigit (c))
234 # define ISALNUM(c) (IN_CTYPE_DOMAIN (c) && isalnum (c))
235 # define ISALPHA(c) (IN_CTYPE_DOMAIN (c) && isalpha (c))
236 # define ISCNTRL(c) (IN_CTYPE_DOMAIN (c) && iscntrl (c))
237 # define ISLOWER(c) (IN_CTYPE_DOMAIN (c) && islower (c))
238 # define ISPUNCT(c) (IN_CTYPE_DOMAIN (c) && ispunct (c))
239 # define ISSPACE(c) (IN_CTYPE_DOMAIN (c) && isspace (c))
240 # define ISUPPER(c) (IN_CTYPE_DOMAIN (c) && isupper (c))
241 # define ISXDIGIT(c) (IN_CTYPE_DOMAIN (c) && isxdigit (c))
244 # define TOLOWER(c) _tolower(c)
246 # define TOLOWER(c) tolower(c)
250 # define NULL (void *)0
253 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
254 since ours (we hope) works properly with all combinations of
255 machines, compilers, `char' and `unsigned char' argument types.
256 (Per Bothner suggested the basic approach.) */
257 # undef SIGN_EXTEND_CHAR
259 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
260 # else /* not __STDC__ */
261 /* As in Harbison and Steele. */
262 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
266 /* How many characters in the character set. */
267 # define CHAR_SET_SIZE 256
271 extern char *re_syntax_table;
273 # else /* not SYNTAX_TABLE */
275 static char re_syntax_table[CHAR_SET_SIZE];
277 static void init_syntax_once PARAMS ((void));
287 bzero (re_syntax_table, sizeof re_syntax_table);
289 for (c = 0; c < CHAR_SET_SIZE; ++c)
291 re_syntax_table[c] = Sword;
293 re_syntax_table['_'] = Sword;
298 # endif /* not SYNTAX_TABLE */
300 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
304 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
305 use `alloca' instead of `malloc'. This is because using malloc in
306 re_search* or re_match* could cause memory leaks when C-g is used in
307 Emacs; also, malloc is slower and causes storage fragmentation. On
308 the other hand, malloc is more portable, and easier to debug.
310 Because we sometimes use alloca, some routines have to be macros,
311 not functions -- `alloca'-allocated space disappears at the end of the
312 function it is called in. */
316 # define REGEX_ALLOCATE malloc
317 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
318 # define REGEX_FREE free
320 # else /* not REGEX_MALLOC */
322 /* Emacs already defines alloca, sometimes. */
325 /* Make alloca work the best possible way. */
327 # define alloca __builtin_alloca
328 # else /* not __GNUC__ */
331 # endif /* HAVE_ALLOCA_H */
332 # endif /* not __GNUC__ */
334 # endif /* not alloca */
336 # define REGEX_ALLOCATE alloca
338 /* Assumes a `char *destination' variable. */
339 # define REGEX_REALLOCATE(source, osize, nsize) \
340 (destination = (char *) alloca (nsize), \
341 memcpy (destination, source, osize))
343 /* No need to do anything to free, after alloca. */
344 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
346 # endif /* not REGEX_MALLOC */
348 /* Define how to allocate the failure stack. */
350 # if defined REL_ALLOC && defined REGEX_MALLOC
352 # define REGEX_ALLOCATE_STACK(size) \
353 r_alloc (&failure_stack_ptr, (size))
354 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
355 r_re_alloc (&failure_stack_ptr, (nsize))
356 # define REGEX_FREE_STACK(ptr) \
357 r_alloc_free (&failure_stack_ptr)
359 # else /* not using relocating allocator */
363 # define REGEX_ALLOCATE_STACK malloc
364 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
365 # define REGEX_FREE_STACK free
367 # else /* not REGEX_MALLOC */
369 # define REGEX_ALLOCATE_STACK alloca
371 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
372 REGEX_REALLOCATE (source, osize, nsize)
373 /* No need to explicitly free anything. */
374 # define REGEX_FREE_STACK(arg)
376 # endif /* not REGEX_MALLOC */
377 # endif /* not using relocating allocator */
380 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
381 `string1' or just past its end. This works if PTR is NULL, which is
383 # define FIRST_STRING_P(ptr) \
384 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
386 /* (Re)Allocate N items of type T using malloc, or fail. */
387 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
388 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
389 # define RETALLOC_IF(addr, n, t) \
390 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
391 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
393 # define BYTEWIDTH 8 /* In bits. */
395 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
399 # define MAX(a, b) ((a) > (b) ? (a) : (b))
400 # define MIN(a, b) ((a) < (b) ? (a) : (b))
402 typedef char boolean;
406 static reg_errcode_t byte_regex_compile _RE_ARGS ((const char *pattern, size_t size,
408 struct re_pattern_buffer *bufp));
410 static int byte_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
411 const char *string1, int size1,
412 const char *string2, int size2,
414 struct re_registers *regs,
416 static int byte_re_search_2 PARAMS ((struct re_pattern_buffer *bufp,
417 const char *string1, int size1,
418 const char *string2, int size2,
419 int startpos, int range,
420 struct re_registers *regs, int stop));
421 static int byte_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
424 static reg_errcode_t wcs_regex_compile _RE_ARGS ((const char *pattern, size_t size,
426 struct re_pattern_buffer *bufp));
429 static int wcs_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
430 const char *cstring1, int csize1,
431 const char *cstring2, int csize2,
433 struct re_registers *regs,
435 wchar_t *string1, int size1,
436 wchar_t *string2, int size2,
437 int *mbs_offset1, int *mbs_offset2));
438 static int wcs_re_search_2 PARAMS ((struct re_pattern_buffer *bufp,
439 const char *string1, int size1,
440 const char *string2, int size2,
441 int startpos, int range,
442 struct re_registers *regs, int stop));
443 static int wcs_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
446 /* These are the command codes that appear in compiled regular
447 expressions. Some opcodes are followed by argument bytes. A
448 command code can specify any interpretation whatsoever for its
449 arguments. Zero bytes may appear in the compiled regular expression. */
455 /* Succeed right away--no more backtracking. */
458 /* Followed by one byte giving n, then by n literal bytes. */
462 /* Same as exactn, but contains binary data. */
466 /* Matches any (more or less) character. */
469 /* Matches any one char belonging to specified set. First
470 following byte is number of bitmap bytes. Then come bytes
471 for a bitmap saying which chars are in. Bits in each byte
472 are ordered low-bit-first. A character is in the set if its
473 bit is 1. A character too large to have a bit in the map is
474 automatically not in the set. */
475 /* ifdef MBS_SUPPORT, following element is length of character
476 classes, length of collating symbols, length of equivalence
477 classes, length of character ranges, and length of characters.
478 Next, character class element, collating symbols elements,
479 equivalence class elements, range elements, and character
481 See regex_compile function. */
484 /* Same parameters as charset, but match any character that is
485 not one of those specified. */
488 /* Start remembering the text that is matched, for storing in a
489 register. Followed by one byte with the register number, in
490 the range 0 to one less than the pattern buffer's re_nsub
491 field. Then followed by one byte with the number of groups
492 inner to this one. (This last has to be part of the
493 start_memory only because we need it in the on_failure_jump
497 /* Stop remembering the text that is matched and store it in a
498 memory register. Followed by one byte with the register
499 number, in the range 0 to one less than `re_nsub' in the
500 pattern buffer, and one byte with the number of inner groups,
501 just like `start_memory'. (We need the number of inner
502 groups here because we don't have any easy way of finding the
503 corresponding start_memory when we're at a stop_memory.) */
506 /* Match a duplicate of something remembered. Followed by one
507 byte containing the register number. */
510 /* Fail unless at beginning of line. */
513 /* Fail unless at end of line. */
516 /* Succeeds if at beginning of buffer (if emacs) or at beginning
517 of string to be matched (if not). */
520 /* Analogously, for end of buffer/string. */
523 /* Followed by two byte relative address to which to jump. */
526 /* Same as jump, but marks the end of an alternative. */
529 /* Followed by two-byte relative address of place to resume at
530 in case of failure. */
531 /* ifdef MBS_SUPPORT, the size of address is 1. */
534 /* Like on_failure_jump, but pushes a placeholder instead of the
535 current string position when executed. */
536 on_failure_keep_string_jump,
538 /* Throw away latest failure point and then jump to following
539 two-byte relative address. */
540 /* ifdef MBS_SUPPORT, the size of address is 1. */
543 /* Change to pop_failure_jump if know won't have to backtrack to
544 match; otherwise change to jump. This is used to jump
545 back to the beginning of a repeat. If what follows this jump
546 clearly won't match what the repeat does, such that we can be
547 sure that there is no use backtracking out of repetitions
548 already matched, then we change it to a pop_failure_jump.
549 Followed by two-byte address. */
550 /* ifdef MBS_SUPPORT, the size of address is 1. */
553 /* Jump to following two-byte address, and push a dummy failure
554 point. This failure point will be thrown away if an attempt
555 is made to use it for a failure. A `+' construct makes this
556 before the first repeat. Also used as an intermediary kind
557 of jump when compiling an alternative. */
558 /* ifdef MBS_SUPPORT, the size of address is 1. */
561 /* Push a dummy failure point and continue. Used at the end of
565 /* Followed by two-byte relative address and two-byte number n.
566 After matching N times, jump to the address upon failure. */
567 /* ifdef MBS_SUPPORT, the size of address is 1. */
570 /* Followed by two-byte relative address, and two-byte number n.
571 Jump to the address N times, then fail. */
572 /* ifdef MBS_SUPPORT, the size of address is 1. */
575 /* Set the following two-byte relative address to the
576 subsequent two-byte number. The address *includes* the two
578 /* ifdef MBS_SUPPORT, the size of address is 1. */
581 wordchar, /* Matches any word-constituent character. */
582 notwordchar, /* Matches any char that is not a word-constituent. */
584 wordbeg, /* Succeeds if at word beginning. */
585 wordend, /* Succeeds if at word end. */
587 wordbound, /* Succeeds if at a word boundary. */
588 notwordbound /* Succeeds if not at a word boundary. */
591 ,before_dot, /* Succeeds if before point. */
592 at_dot, /* Succeeds if at point. */
593 after_dot, /* Succeeds if after point. */
595 /* Matches any character whose syntax is specified. Followed by
596 a byte which contains a syntax code, e.g., Sword. */
599 /* Matches any character whose syntax is not that specified. */
603 #endif /* not INSIDE_RECURSION */
608 # define UCHAR_T unsigned char
609 # define COMPILED_BUFFER_VAR bufp->buffer
610 # define OFFSET_ADDRESS_SIZE 2
611 # define PREFIX(name) byte_##name
612 # define ARG_PREFIX(name) name
613 # define PUT_CHAR(c) putchar (c)
616 # define CHAR_T wchar_t
617 # define UCHAR_T wchar_t
618 # define COMPILED_BUFFER_VAR wc_buffer
619 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
620 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
621 # define PREFIX(name) wcs_##name
622 # define ARG_PREFIX(name) c##name
623 /* Should we use wide stream?? */
624 # define PUT_CHAR(c) printf ("%C", c);
630 # define INSIDE_RECURSION
632 # undef INSIDE_RECURSION
635 # define INSIDE_RECURSION
637 # undef INSIDE_RECURSION
640 #include "unlocked-io.h"
642 #ifdef INSIDE_RECURSION
643 /* Common operations on the compiled pattern. */
645 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
646 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
649 # define STORE_NUMBER(destination, number) \
651 *(destination) = (UCHAR_T)(number); \
654 # define STORE_NUMBER(destination, number) \
656 (destination)[0] = (number) & 0377; \
657 (destination)[1] = (number) >> 8; \
661 /* Same as STORE_NUMBER, except increment DESTINATION to
662 the byte after where the number is stored. Therefore, DESTINATION
663 must be an lvalue. */
664 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
666 # define STORE_NUMBER_AND_INCR(destination, number) \
668 STORE_NUMBER (destination, number); \
669 (destination) += OFFSET_ADDRESS_SIZE; \
672 /* Put into DESTINATION a number stored in two contiguous bytes starting
674 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
677 # define EXTRACT_NUMBER(destination, source) \
679 (destination) = *(source); \
682 # define EXTRACT_NUMBER(destination, source) \
684 (destination) = *(source) & 0377; \
685 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
690 static void PREFIX(extract_number) _RE_ARGS ((int *dest, UCHAR_T *source));
692 PREFIX(extract_number) (dest, source)
699 int temp = SIGN_EXTEND_CHAR (*(source + 1));
700 *dest = *source & 0377;
705 # ifndef EXTRACT_MACROS /* To debug the macros. */
706 # undef EXTRACT_NUMBER
707 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
708 # endif /* not EXTRACT_MACROS */
712 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
713 SOURCE must be an lvalue. */
715 # define EXTRACT_NUMBER_AND_INCR(destination, source) \
717 EXTRACT_NUMBER (destination, source); \
718 (source) += OFFSET_ADDRESS_SIZE; \
722 static void PREFIX(extract_number_and_incr) _RE_ARGS ((int *destination,
725 PREFIX(extract_number_and_incr) (destination, source)
729 PREFIX(extract_number) (destination, *source);
730 *source += OFFSET_ADDRESS_SIZE;
733 # ifndef EXTRACT_MACROS
734 # undef EXTRACT_NUMBER_AND_INCR
735 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
736 PREFIX(extract_number_and_incr) (&dest, &src)
737 # endif /* not EXTRACT_MACROS */
743 /* If DEBUG is defined, Regex prints many voluminous messages about what
744 it is doing (if the variable `debug' is nonzero). If linked with the
745 main program in `iregex.c', you can enter patterns and strings
746 interactively. And if linked with the main program in `main.c' and
747 the other test files, you can run the already-written tests. */
751 # ifndef DEFINED_ONCE
753 /* We use standard I/O for debugging. */
756 /* It is useful to test things that ``must'' be true when debugging. */
761 # define DEBUG_STATEMENT(e) e
762 # define DEBUG_PRINT1(x) if (debug) printf (x)
763 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
764 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
765 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
766 # endif /* not DEFINED_ONCE */
768 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
769 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
770 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
771 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
774 /* Print the fastmap in human-readable form. */
776 # ifndef DEFINED_ONCE
778 print_fastmap (fastmap)
781 unsigned was_a_range = 0;
784 while (i < (1 << BYTEWIDTH))
790 while (i < (1 << BYTEWIDTH) && fastmap[i])
804 # endif /* not DEFINED_ONCE */
807 /* Print a compiled pattern string in human-readable form, starting at
808 the START pointer into it and ending just before the pointer END. */
811 PREFIX(print_partial_compiled_pattern) (start, end)
826 /* Loop over pattern commands. */
830 printf ("%td:\t", p - start);
832 printf ("%ld:\t", (long int) (p - start));
835 switch ((re_opcode_t) *p++)
843 printf ("/exactn/%d", mcnt);
855 printf ("/exactn_bin/%d", mcnt);
858 printf("/%lx", (long int) *p++);
862 # endif /* MBS_SUPPORT */
866 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++);
871 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++);
875 printf ("/duplicate/%ld", (long int) *p++);
888 printf ("/charset [%s",
889 (re_opcode_t) *(workp - 1) == charset_not ? "^" : "");
891 length = *workp++; /* the length of char_classes */
892 for (i=0 ; i<length ; i++)
893 printf("[:%lx:]", (long int) *p++);
894 length = *workp++; /* the length of collating_symbol */
895 for (i=0 ; i<length ;)
899 PUT_CHAR((i++,*p++));
903 length = *workp++; /* the length of equivalence_class */
904 for (i=0 ; i<length ;)
908 PUT_CHAR((i++,*p++));
912 length = *workp++; /* the length of char_range */
913 for (i=0 ; i<length ; i++)
915 wchar_t range_start = *p++;
916 wchar_t range_end = *p++;
917 printf("%C-%C", range_start, range_end);
919 length = *workp++; /* the length of char */
920 for (i=0 ; i<length ; i++)
924 register int c, last = -100;
925 register int in_range = 0;
927 printf ("/charset [%s",
928 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
930 assert (p + *p < pend);
932 for (c = 0; c < 256; c++)
934 && (p[1 + (c/8)] & (1 << (c % 8))))
936 /* Are we starting a range? */
937 if (last + 1 == c && ! in_range)
942 /* Have we broken a range? */
943 else if (last + 1 != c && in_range)
973 case on_failure_jump:
974 PREFIX(extract_number_and_incr) (&mcnt, &p);
976 printf ("/on_failure_jump to %td", p + mcnt - start);
978 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
982 case on_failure_keep_string_jump:
983 PREFIX(extract_number_and_incr) (&mcnt, &p);
985 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start);
987 printf ("/on_failure_keep_string_jump to %ld",
988 (long int) (p + mcnt - start));
992 case dummy_failure_jump:
993 PREFIX(extract_number_and_incr) (&mcnt, &p);
995 printf ("/dummy_failure_jump to %td", p + mcnt - start);
997 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
1001 case push_dummy_failure:
1002 printf ("/push_dummy_failure");
1005 case maybe_pop_jump:
1006 PREFIX(extract_number_and_incr) (&mcnt, &p);
1008 printf ("/maybe_pop_jump to %td", p + mcnt - start);
1010 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
1014 case pop_failure_jump:
1015 PREFIX(extract_number_and_incr) (&mcnt, &p);
1017 printf ("/pop_failure_jump to %td", p + mcnt - start);
1019 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
1024 PREFIX(extract_number_and_incr) (&mcnt, &p);
1026 printf ("/jump_past_alt to %td", p + mcnt - start);
1028 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
1033 PREFIX(extract_number_and_incr) (&mcnt, &p);
1035 printf ("/jump to %td", p + mcnt - start);
1037 printf ("/jump to %ld", (long int) (p + mcnt - start));
1042 PREFIX(extract_number_and_incr) (&mcnt, &p);
1044 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1046 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2);
1048 printf ("/succeed_n to %ld, %d times",
1049 (long int) (p1 - start), mcnt2);
1054 PREFIX(extract_number_and_incr) (&mcnt, &p);
1056 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1057 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
1061 PREFIX(extract_number_and_incr) (&mcnt, &p);
1063 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1065 printf ("/set_number_at location %td to %d", p1 - start, mcnt2);
1067 printf ("/set_number_at location %ld to %d",
1068 (long int) (p1 - start), mcnt2);
1073 printf ("/wordbound");
1077 printf ("/notwordbound");
1081 printf ("/wordbeg");
1085 printf ("/wordend");
1090 printf ("/before_dot");
1098 printf ("/after_dot");
1102 printf ("/syntaxspec");
1104 printf ("/%d", mcnt);
1108 printf ("/notsyntaxspec");
1110 printf ("/%d", mcnt);
1115 printf ("/wordchar");
1119 printf ("/notwordchar");
1131 printf ("?%ld", (long int) *(p-1));
1138 printf ("%td:\tend of pattern.\n", p - start);
1140 printf ("%ld:\tend of pattern.\n", (long int) (p - start));
1146 PREFIX(print_compiled_pattern) (bufp)
1147 struct re_pattern_buffer *bufp;
1149 UCHAR_T *buffer = (UCHAR_T*) bufp->buffer;
1151 PREFIX(print_partial_compiled_pattern) (buffer, buffer
1152 + bufp->used / sizeof(UCHAR_T));
1153 printf ("%ld bytes used/%ld bytes allocated.\n",
1154 bufp->used, bufp->allocated);
1156 if (bufp->fastmap_accurate && bufp->fastmap)
1158 printf ("fastmap: ");
1159 print_fastmap (bufp->fastmap);
1163 printf ("re_nsub: %Zd\t", bufp->re_nsub);
1165 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
1167 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1168 printf ("can_be_null: %d\t", bufp->can_be_null);
1169 printf ("newline_anchor: %d\n", bufp->newline_anchor);
1170 printf ("no_sub: %d\t", bufp->no_sub);
1171 printf ("not_bol: %d\t", bufp->not_bol);
1172 printf ("not_eol: %d\t", bufp->not_eol);
1173 printf ("syntax: %lx\n", bufp->syntax);
1174 /* Perhaps we should print the translate table? */
1179 PREFIX(print_double_string) (where, string1, size1, string2, size2)
1180 const CHAR_T *where;
1181 const CHAR_T *string1;
1182 const CHAR_T *string2;
1194 if (FIRST_STRING_P (where))
1196 for (this_char = where - string1; this_char < size1; this_char++)
1197 PUT_CHAR (string1[this_char]);
1203 for (this_char = where - string2; this_char < size2; this_char++)
1205 PUT_CHAR (string2[this_char]);
1208 fputs ("...", stdout);
1215 # ifndef DEFINED_ONCE
1224 # else /* not DEBUG */
1226 # ifndef DEFINED_ONCE
1230 # define DEBUG_STATEMENT(e)
1231 # define DEBUG_PRINT1(x)
1232 # define DEBUG_PRINT2(x1, x2)
1233 # define DEBUG_PRINT3(x1, x2, x3)
1234 # define DEBUG_PRINT4(x1, x2, x3, x4)
1235 # endif /* not DEFINED_ONCE */
1236 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1237 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1239 # endif /* not DEBUG */
1244 /* This convert a multibyte string to a wide character string.
1245 And write their correspondances to offset_buffer(see below)
1246 and write whether each wchar_t is binary data to is_binary.
1247 This assume invalid multibyte sequences as binary data.
1248 We assume offset_buffer and is_binary is already allocated
1251 static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src,
1252 size_t len, int *offset_buffer,
1255 convert_mbs_to_wcs (dest, src, len, offset_buffer, is_binary)
1257 const unsigned char* src;
1258 size_t len; /* the length of multibyte string. */
1260 /* It hold correspondances between src(char string) and
1261 dest(wchar_t string) for optimization.
1263 dest = {'X', 'Y', 'Z'}
1264 (each "xxx", "y" and "zz" represent one multibyte character
1265 corresponding to 'X', 'Y' and 'Z'.)
1266 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1272 wchar_t *pdest = dest;
1273 const unsigned char *psrc = src;
1274 size_t wc_count = 0;
1278 size_t mb_remain = len;
1279 size_t mb_count = 0;
1281 /* Initialize the conversion state. */
1282 memset (&mbs, 0, sizeof (mbstate_t));
1284 offset_buffer[0] = 0;
1285 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed,
1288 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs);
1291 /* failed to convert. maybe src contains binary data.
1292 So we consume 1 byte manualy. */
1296 is_binary[wc_count] = TRUE;
1299 is_binary[wc_count] = FALSE;
1300 /* In sjis encoding, we use yen sign as escape character in
1301 place of reverse solidus. So we convert 0x5c(yen sign in
1302 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1303 solidus in UCS2). */
1304 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5)
1305 *pdest = (wchar_t) *psrc;
1307 offset_buffer[wc_count + 1] = mb_count += consumed;
1310 /* Fill remain of the buffer with sentinel. */
1311 for (i = wc_count + 1 ; i <= len ; i++)
1312 offset_buffer[i] = mb_count + 1;
1319 #else /* not INSIDE_RECURSION */
1321 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1322 also be assigned to arbitrarily: each pattern buffer stores its own
1323 syntax, so it can be changed between regex compilations. */
1324 /* This has no initializer because initialized variables in Emacs
1325 become read-only after dumping. */
1326 reg_syntax_t re_syntax_options;
1329 /* Specify the precise syntax of regexps for compilation. This provides
1330 for compatibility for various utilities which historically have
1331 different, incompatible syntaxes.
1333 The argument SYNTAX is a bit mask comprised of the various bits
1334 defined in regex.h. We return the old syntax. */
1337 re_set_syntax (syntax)
1338 reg_syntax_t syntax;
1340 reg_syntax_t ret = re_syntax_options;
1342 re_syntax_options = syntax;
1344 if (syntax & RE_DEBUG)
1346 else if (debug) /* was on but now is not */
1352 weak_alias (__re_set_syntax, re_set_syntax)
1355 /* This table gives an error message for each of the error codes listed
1356 in regex.h. Obviously the order here has to be same as there.
1357 POSIX doesn't require that we do anything for REG_NOERROR,
1358 but why not be nice? */
1360 static const char re_error_msgid[] =
1362 # define REG_NOERROR_IDX 0
1363 gettext_noop ("Success") /* REG_NOERROR */
1365 # define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1366 gettext_noop ("No match") /* REG_NOMATCH */
1368 # define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1369 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1371 # define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1372 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1374 # define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1375 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1377 # define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1378 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1380 # define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1381 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1383 # define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1384 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1386 # define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1387 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1389 # define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1390 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1392 # define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1393 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1395 # define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1396 gettext_noop ("Invalid range end") /* REG_ERANGE */
1398 # define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1399 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1401 # define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1402 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1404 # define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1405 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1407 # define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1408 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1410 # define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1411 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1414 static const size_t re_error_msgid_idx[] =
1435 #endif /* INSIDE_RECURSION */
1437 #ifndef DEFINED_ONCE
1438 /* Avoiding alloca during matching, to placate r_alloc. */
1440 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1441 searching and matching functions should not call alloca. On some
1442 systems, alloca is implemented in terms of malloc, and if we're
1443 using the relocating allocator routines, then malloc could cause a
1444 relocation, which might (if the strings being searched are in the
1445 ralloc heap) shift the data out from underneath the regexp
1448 Here's another reason to avoid allocation: Emacs
1449 processes input from X in a signal handler; processing X input may
1450 call malloc; if input arrives while a matching routine is calling
1451 malloc, then we're scrod. But Emacs can't just block input while
1452 calling matching routines; then we don't notice interrupts when
1453 they come in. So, Emacs blocks input around all regexp calls
1454 except the matching calls, which it leaves unprotected, in the
1455 faith that they will not malloc. */
1457 /* Normally, this is fine. */
1458 # define MATCH_MAY_ALLOCATE
1460 /* When using GNU C, we are not REALLY using the C alloca, no matter
1461 what config.h may say. So don't take precautions for it. */
1466 /* The match routines may not allocate if (1) they would do it with malloc
1467 and (2) it's not safe for them to use malloc.
1468 Note that if REL_ALLOC is defined, matching would not use malloc for the
1469 failure stack, but we would still use it for the register vectors;
1470 so REL_ALLOC should not affect this. */
1471 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1472 # undef MATCH_MAY_ALLOCATE
1474 #endif /* not DEFINED_ONCE */
1476 #ifdef INSIDE_RECURSION
1477 /* Failure stack declarations and macros; both re_compile_fastmap and
1478 re_match_2 use a failure stack. These have to be macros because of
1479 REGEX_ALLOCATE_STACK. */
1482 /* Number of failure points for which to initially allocate space
1483 when matching. If this number is exceeded, we allocate more
1484 space, so it is not a hard limit. */
1485 # ifndef INIT_FAILURE_ALLOC
1486 # define INIT_FAILURE_ALLOC 5
1489 /* Roughly the maximum number of failure points on the stack. Would be
1490 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1491 This is a variable only so users of regex can assign to it; we never
1492 change it ourselves. */
1494 # ifdef INT_IS_16BIT
1496 # ifndef DEFINED_ONCE
1497 # if defined MATCH_MAY_ALLOCATE
1498 /* 4400 was enough to cause a crash on Alpha OSF/1,
1499 whose default stack limit is 2mb. */
1500 long int re_max_failures = 4000;
1502 long int re_max_failures = 2000;
1506 union PREFIX(fail_stack_elt)
1512 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1516 PREFIX(fail_stack_elt_t) *stack;
1517 unsigned long int size;
1518 unsigned long int avail; /* Offset of next open position. */
1519 } PREFIX(fail_stack_type);
1521 # else /* not INT_IS_16BIT */
1523 # ifndef DEFINED_ONCE
1524 # if defined MATCH_MAY_ALLOCATE
1525 /* 4400 was enough to cause a crash on Alpha OSF/1,
1526 whose default stack limit is 2mb. */
1527 int re_max_failures = 4000;
1529 int re_max_failures = 2000;
1533 union PREFIX(fail_stack_elt)
1539 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1543 PREFIX(fail_stack_elt_t) *stack;
1545 unsigned avail; /* Offset of next open position. */
1546 } PREFIX(fail_stack_type);
1548 # endif /* INT_IS_16BIT */
1550 # ifndef DEFINED_ONCE
1551 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1552 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1553 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1557 /* Define macros to initialize and free the failure stack.
1558 Do `return -2' if the alloc fails. */
1560 # ifdef MATCH_MAY_ALLOCATE
1561 # define INIT_FAIL_STACK() \
1563 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1564 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1566 if (fail_stack.stack == NULL) \
1569 fail_stack.size = INIT_FAILURE_ALLOC; \
1570 fail_stack.avail = 0; \
1573 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1575 # define INIT_FAIL_STACK() \
1577 fail_stack.avail = 0; \
1580 # define RESET_FAIL_STACK()
1584 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1586 Return 1 if succeeds, and 0 if either ran out of memory
1587 allocating space for it or it was already too large.
1589 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1591 # define DOUBLE_FAIL_STACK(fail_stack) \
1592 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1594 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1595 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1596 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1597 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1599 (fail_stack).stack == NULL \
1601 : ((fail_stack).size <<= 1, \
1605 /* Push pointer POINTER on FAIL_STACK.
1606 Return 1 if was able to do so and 0 if ran out of memory allocating
1608 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1609 ((FAIL_STACK_FULL () \
1610 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1612 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1615 /* Push a pointer value onto the failure stack.
1616 Assumes the variable `fail_stack'. Probably should only
1617 be called from within `PUSH_FAILURE_POINT'. */
1618 # define PUSH_FAILURE_POINTER(item) \
1619 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1621 /* This pushes an integer-valued item onto the failure stack.
1622 Assumes the variable `fail_stack'. Probably should only
1623 be called from within `PUSH_FAILURE_POINT'. */
1624 # define PUSH_FAILURE_INT(item) \
1625 fail_stack.stack[fail_stack.avail++].integer = (item)
1627 /* Push a fail_stack_elt_t value onto the failure stack.
1628 Assumes the variable `fail_stack'. Probably should only
1629 be called from within `PUSH_FAILURE_POINT'. */
1630 # define PUSH_FAILURE_ELT(item) \
1631 fail_stack.stack[fail_stack.avail++] = (item)
1633 /* These three POP... operations complement the three PUSH... operations.
1634 All assume that `fail_stack' is nonempty. */
1635 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1636 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1637 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1639 /* Used to omit pushing failure point id's when we're not debugging. */
1641 # define DEBUG_PUSH PUSH_FAILURE_INT
1642 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1644 # define DEBUG_PUSH(item)
1645 # define DEBUG_POP(item_addr)
1649 /* Push the information about the state we will need
1650 if we ever fail back to it.
1652 Requires variables fail_stack, regstart, regend, reg_info, and
1653 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1656 Does `return FAILURE_CODE' if runs out of memory. */
1658 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1660 char *destination; \
1661 /* Must be int, so when we don't save any registers, the arithmetic \
1662 of 0 + -1 isn't done as unsigned. */ \
1663 /* Can't be int, since there is not a shred of a guarantee that int \
1664 is wide enough to hold a value of something to which pointer can \
1666 active_reg_t this_reg; \
1668 DEBUG_STATEMENT (failure_id++); \
1669 DEBUG_STATEMENT (nfailure_points_pushed++); \
1670 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1671 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1672 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1674 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1675 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1677 /* Ensure we have enough space allocated for what we will push. */ \
1678 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1680 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1681 return failure_code; \
1683 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1684 (fail_stack).size); \
1685 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1688 /* Push the info, starting with the registers. */ \
1689 DEBUG_PRINT1 ("\n"); \
1692 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1695 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1696 DEBUG_STATEMENT (num_regs_pushed++); \
1698 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1699 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1701 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1702 PUSH_FAILURE_POINTER (regend[this_reg]); \
1704 DEBUG_PRINT2 (" info: %p\n ", \
1705 reg_info[this_reg].word.pointer); \
1706 DEBUG_PRINT2 (" match_null=%d", \
1707 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1708 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1709 DEBUG_PRINT2 (" matched_something=%d", \
1710 MATCHED_SOMETHING (reg_info[this_reg])); \
1711 DEBUG_PRINT2 (" ever_matched=%d", \
1712 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1713 DEBUG_PRINT1 ("\n"); \
1714 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1717 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1718 PUSH_FAILURE_INT (lowest_active_reg); \
1720 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1721 PUSH_FAILURE_INT (highest_active_reg); \
1723 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1724 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1725 PUSH_FAILURE_POINTER (pattern_place); \
1727 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1728 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1730 DEBUG_PRINT1 ("'\n"); \
1731 PUSH_FAILURE_POINTER (string_place); \
1733 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1734 DEBUG_PUSH (failure_id); \
1737 # ifndef DEFINED_ONCE
1738 /* This is the number of items that are pushed and popped on the stack
1739 for each register. */
1740 # define NUM_REG_ITEMS 3
1742 /* Individual items aside from the registers. */
1744 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1746 # define NUM_NONREG_ITEMS 4
1749 /* We push at most this many items on the stack. */
1750 /* We used to use (num_regs - 1), which is the number of registers
1751 this regexp will save; but that was changed to 5
1752 to avoid stack overflow for a regexp with lots of parens. */
1753 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1755 /* We actually push this many items. */
1756 # define NUM_FAILURE_ITEMS \
1758 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1762 /* How many items can still be added to the stack without overflowing it. */
1763 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1764 # endif /* not DEFINED_ONCE */
1767 /* Pops what PUSH_FAIL_STACK pushes.
1769 We restore into the parameters, all of which should be lvalues:
1770 STR -- the saved data position.
1771 PAT -- the saved pattern position.
1772 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1773 REGSTART, REGEND -- arrays of string positions.
1774 REG_INFO -- array of information about each subexpression.
1776 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1777 `pend', `string1', `size1', `string2', and `size2'. */
1778 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1780 DEBUG_STATEMENT (unsigned failure_id;) \
1781 active_reg_t this_reg; \
1782 const UCHAR_T *string_temp; \
1784 assert (!FAIL_STACK_EMPTY ()); \
1786 /* Remove failure points and point to how many regs pushed. */ \
1787 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1788 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1789 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1791 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1793 DEBUG_POP (&failure_id); \
1794 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1796 /* If the saved string location is NULL, it came from an \
1797 on_failure_keep_string_jump opcode, and we want to throw away the \
1798 saved NULL, thus retaining our current position in the string. */ \
1799 string_temp = POP_FAILURE_POINTER (); \
1800 if (string_temp != NULL) \
1801 str = (const CHAR_T *) string_temp; \
1803 DEBUG_PRINT2 (" Popping string %p: `", str); \
1804 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1805 DEBUG_PRINT1 ("'\n"); \
1807 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1808 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1809 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1811 /* Restore register info. */ \
1812 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1813 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1815 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1816 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1819 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1821 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1823 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1824 DEBUG_PRINT2 (" info: %p\n", \
1825 reg_info[this_reg].word.pointer); \
1827 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1828 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1830 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1831 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1835 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1837 reg_info[this_reg].word.integer = 0; \
1838 regend[this_reg] = 0; \
1839 regstart[this_reg] = 0; \
1841 highest_active_reg = high_reg; \
1844 set_regs_matched_done = 0; \
1845 DEBUG_STATEMENT (nfailure_points_popped++); \
1846 } /* POP_FAILURE_POINT */
1848 /* Structure for per-register (a.k.a. per-group) information.
1849 Other register information, such as the
1850 starting and ending positions (which are addresses), and the list of
1851 inner groups (which is a bits list) are maintained in separate
1854 We are making a (strictly speaking) nonportable assumption here: that
1855 the compiler will pack our bit fields into something that fits into
1856 the type of `word', i.e., is something that fits into one item on the
1860 /* Declarations and macros for re_match_2. */
1864 PREFIX(fail_stack_elt_t) word;
1867 /* This field is one if this group can match the empty string,
1868 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1869 # define MATCH_NULL_UNSET_VALUE 3
1870 unsigned match_null_string_p : 2;
1871 unsigned is_active : 1;
1872 unsigned matched_something : 1;
1873 unsigned ever_matched_something : 1;
1875 } PREFIX(register_info_type);
1877 # ifndef DEFINED_ONCE
1878 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1879 # define IS_ACTIVE(R) ((R).bits.is_active)
1880 # define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1881 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1884 /* Call this when have matched a real character; it sets `matched' flags
1885 for the subexpressions which we are currently inside. Also records
1886 that those subexprs have matched. */
1887 # define SET_REGS_MATCHED() \
1890 if (!set_regs_matched_done) \
1893 set_regs_matched_done = 1; \
1894 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1896 MATCHED_SOMETHING (reg_info[r]) \
1897 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1903 # endif /* not DEFINED_ONCE */
1905 /* Registers are set to a sentinel when they haven't yet matched. */
1906 static CHAR_T PREFIX(reg_unset_dummy);
1907 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1908 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1910 /* Subroutine declarations and macros for regex_compile. */
1911 static void PREFIX(store_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, int arg));
1912 static void PREFIX(store_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1913 int arg1, int arg2));
1914 static void PREFIX(insert_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1915 int arg, UCHAR_T *end));
1916 static void PREFIX(insert_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1917 int arg1, int arg2, UCHAR_T *end));
1918 static boolean PREFIX(at_begline_loc_p) _RE_ARGS ((const CHAR_T *pattern,
1920 reg_syntax_t syntax));
1921 static boolean PREFIX(at_endline_loc_p) _RE_ARGS ((const CHAR_T *p,
1923 reg_syntax_t syntax));
1925 static reg_errcode_t wcs_compile_range _RE_ARGS ((CHAR_T range_start,
1926 const CHAR_T **p_ptr,
1929 reg_syntax_t syntax,
1932 static void insert_space _RE_ARGS ((int num, CHAR_T *loc, CHAR_T *end));
1934 static reg_errcode_t byte_compile_range _RE_ARGS ((unsigned int range_start,
1938 reg_syntax_t syntax,
1942 /* Fetch the next character in the uncompiled pattern---translating it
1943 if necessary. Also cast from a signed character in the constant
1944 string passed to us by the user to an unsigned char that we can use
1945 as an array index (in, e.g., `translate'). */
1946 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1947 because it is impossible to allocate 4GB array for some encodings
1948 which have 4 byte character_set like UCS4. */
1951 # define PATFETCH(c) \
1952 do {if (p == pend) return REG_EEND; \
1953 c = (UCHAR_T) *p++; \
1954 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1957 # define PATFETCH(c) \
1958 do {if (p == pend) return REG_EEND; \
1959 c = (unsigned char) *p++; \
1960 if (translate) c = (unsigned char) translate[c]; \
1965 /* Fetch the next character in the uncompiled pattern, with no
1967 # define PATFETCH_RAW(c) \
1968 do {if (p == pend) return REG_EEND; \
1969 c = (UCHAR_T) *p++; \
1972 /* Go backwards one character in the pattern. */
1973 # define PATUNFETCH p--
1976 /* If `translate' is non-null, return translate[D], else just D. We
1977 cast the subscript to translate because some data is declared as
1978 `char *', to avoid warnings when a string constant is passed. But
1979 when we use a character as a subscript we must make it unsigned. */
1980 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1981 because it is impossible to allocate 4GB array for some encodings
1982 which have 4 byte character_set like UCS4. */
1986 # define TRANSLATE(d) \
1987 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1988 ? (char) translate[(unsigned char) (d)] : (d))
1990 # define TRANSLATE(d) \
1991 (translate ? (char) translate[(unsigned char) (d)] : (d))
1996 /* Macros for outputting the compiled pattern into `buffer'. */
1998 /* If the buffer isn't allocated when it comes in, use this. */
1999 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
2001 /* Make sure we have at least N more bytes of space in buffer. */
2003 # define GET_BUFFER_SPACE(n) \
2004 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
2005 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
2008 # define GET_BUFFER_SPACE(n) \
2009 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
2013 /* Make sure we have one more byte of buffer space and then add C to it. */
2014 # define BUF_PUSH(c) \
2016 GET_BUFFER_SPACE (1); \
2017 *b++ = (UCHAR_T) (c); \
2021 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
2022 # define BUF_PUSH_2(c1, c2) \
2024 GET_BUFFER_SPACE (2); \
2025 *b++ = (UCHAR_T) (c1); \
2026 *b++ = (UCHAR_T) (c2); \
2030 /* As with BUF_PUSH_2, except for three bytes. */
2031 # define BUF_PUSH_3(c1, c2, c3) \
2033 GET_BUFFER_SPACE (3); \
2034 *b++ = (UCHAR_T) (c1); \
2035 *b++ = (UCHAR_T) (c2); \
2036 *b++ = (UCHAR_T) (c3); \
2039 /* Store a jump with opcode OP at LOC to location TO. We store a
2040 relative address offset by the three bytes the jump itself occupies. */
2041 # define STORE_JUMP(op, loc, to) \
2042 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
2044 /* Likewise, for a two-argument jump. */
2045 # define STORE_JUMP2(op, loc, to, arg) \
2046 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
2048 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
2049 # define INSERT_JUMP(op, loc, to) \
2050 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
2052 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
2053 # define INSERT_JUMP2(op, loc, to, arg) \
2054 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
2057 /* This is not an arbitrary limit: the arguments which represent offsets
2058 into the pattern are two bytes long. So if 2^16 bytes turns out to
2059 be too small, many things would have to change. */
2060 /* Any other compiler which, like MSC, has allocation limit below 2^16
2061 bytes will have to use approach similar to what was done below for
2062 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2063 reallocating to 0 bytes. Such thing is not going to work too well.
2064 You have been warned!! */
2065 # ifndef DEFINED_ONCE
2066 # if defined _MSC_VER && !defined WIN32
2067 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2068 The REALLOC define eliminates a flurry of conversion warnings,
2069 but is not required. */
2070 # define MAX_BUF_SIZE 65500L
2071 # define REALLOC(p,s) realloc ((p), (size_t) (s))
2073 # define MAX_BUF_SIZE (1L << 16)
2074 # define REALLOC(p,s) realloc ((p), (s))
2077 /* Extend the buffer by twice its current size via realloc and
2078 reset the pointers that pointed into the old block to point to the
2079 correct places in the new one. If extending the buffer results in it
2080 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2081 # if __BOUNDED_POINTERS__
2082 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2083 # define MOVE_BUFFER_POINTER(P) \
2084 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2085 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2088 SET_HIGH_BOUND (b); \
2089 SET_HIGH_BOUND (begalt); \
2090 if (fixup_alt_jump) \
2091 SET_HIGH_BOUND (fixup_alt_jump); \
2093 SET_HIGH_BOUND (laststart); \
2094 if (pending_exact) \
2095 SET_HIGH_BOUND (pending_exact); \
2098 # define MOVE_BUFFER_POINTER(P) (P) += incr
2099 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
2101 # endif /* not DEFINED_ONCE */
2104 # define EXTEND_BUFFER() \
2106 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2108 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2110 bufp->allocated <<= 1; \
2111 if (bufp->allocated > MAX_BUF_SIZE) \
2112 bufp->allocated = MAX_BUF_SIZE; \
2113 /* How many characters the new buffer can have? */ \
2114 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2115 if (wchar_count == 0) wchar_count = 1; \
2116 /* Truncate the buffer to CHAR_T align. */ \
2117 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2118 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2119 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2120 if (COMPILED_BUFFER_VAR == NULL) \
2121 return REG_ESPACE; \
2122 /* If the buffer moved, move all the pointers into it. */ \
2123 if (old_buffer != COMPILED_BUFFER_VAR) \
2125 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2126 MOVE_BUFFER_POINTER (b); \
2127 MOVE_BUFFER_POINTER (begalt); \
2128 if (fixup_alt_jump) \
2129 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2131 MOVE_BUFFER_POINTER (laststart); \
2132 if (pending_exact) \
2133 MOVE_BUFFER_POINTER (pending_exact); \
2135 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2138 # define EXTEND_BUFFER() \
2140 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2141 if (bufp->allocated == MAX_BUF_SIZE) \
2143 bufp->allocated <<= 1; \
2144 if (bufp->allocated > MAX_BUF_SIZE) \
2145 bufp->allocated = MAX_BUF_SIZE; \
2146 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2148 if (COMPILED_BUFFER_VAR == NULL) \
2149 return REG_ESPACE; \
2150 /* If the buffer moved, move all the pointers into it. */ \
2151 if (old_buffer != COMPILED_BUFFER_VAR) \
2153 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2154 MOVE_BUFFER_POINTER (b); \
2155 MOVE_BUFFER_POINTER (begalt); \
2156 if (fixup_alt_jump) \
2157 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2159 MOVE_BUFFER_POINTER (laststart); \
2160 if (pending_exact) \
2161 MOVE_BUFFER_POINTER (pending_exact); \
2163 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2167 # ifndef DEFINED_ONCE
2168 /* Since we have one byte reserved for the register number argument to
2169 {start,stop}_memory, the maximum number of groups we can report
2170 things about is what fits in that byte. */
2171 # define MAX_REGNUM 255
2173 /* But patterns can have more than `MAX_REGNUM' registers. We just
2174 ignore the excess. */
2175 typedef unsigned regnum_t;
2178 /* Macros for the compile stack. */
2180 /* Since offsets can go either forwards or backwards, this type needs to
2181 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2182 /* int may be not enough when sizeof(int) == 2. */
2183 typedef long pattern_offset_t;
2187 pattern_offset_t begalt_offset;
2188 pattern_offset_t fixup_alt_jump;
2189 pattern_offset_t inner_group_offset;
2190 pattern_offset_t laststart_offset;
2192 } compile_stack_elt_t;
2197 compile_stack_elt_t *stack;
2199 unsigned avail; /* Offset of next open position. */
2200 } compile_stack_type;
2203 # define INIT_COMPILE_STACK_SIZE 32
2205 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2206 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2208 /* The next available element. */
2209 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2211 # endif /* not DEFINED_ONCE */
2213 /* Set the bit for character C in a list. */
2214 # ifndef DEFINED_ONCE
2215 # define SET_LIST_BIT(c) \
2216 (b[((unsigned char) (c)) / BYTEWIDTH] \
2217 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2218 # endif /* DEFINED_ONCE */
2220 /* Get the next unsigned number in the uncompiled pattern. */
2221 # define GET_UNSIGNED_NUMBER(num) \
2226 if (c < '0' || c > '9') \
2228 if (num <= RE_DUP_MAX) \
2232 num = num * 10 + c - '0'; \
2237 # ifndef DEFINED_ONCE
2238 # if defined _LIBC || WIDE_CHAR_SUPPORT
2239 /* The GNU C library provides support for user-defined character classes
2240 and the functions from ISO C amendement 1. */
2241 # ifdef CHARCLASS_NAME_MAX
2242 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2244 /* This shouldn't happen but some implementation might still have this
2245 problem. Use a reasonable default value. */
2246 # define CHAR_CLASS_MAX_LENGTH 256
2250 # define IS_CHAR_CLASS(string) __wctype (string)
2252 # define IS_CHAR_CLASS(string) wctype (string)
2255 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2257 # define IS_CHAR_CLASS(string) \
2258 (STREQ (string, "alpha") || STREQ (string, "upper") \
2259 || STREQ (string, "lower") || STREQ (string, "digit") \
2260 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2261 || STREQ (string, "space") || STREQ (string, "print") \
2262 || STREQ (string, "punct") || STREQ (string, "graph") \
2263 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2265 # endif /* DEFINED_ONCE */
2267 # ifndef MATCH_MAY_ALLOCATE
2269 /* If we cannot allocate large objects within re_match_2_internal,
2270 we make the fail stack and register vectors global.
2271 The fail stack, we grow to the maximum size when a regexp
2273 The register vectors, we adjust in size each time we
2274 compile a regexp, according to the number of registers it needs. */
2276 static PREFIX(fail_stack_type) fail_stack;
2278 /* Size with which the following vectors are currently allocated.
2279 That is so we can make them bigger as needed,
2280 but never make them smaller. */
2281 # ifdef DEFINED_ONCE
2282 static int regs_allocated_size;
2284 static const char ** regstart, ** regend;
2285 static const char ** old_regstart, ** old_regend;
2286 static const char **best_regstart, **best_regend;
2287 static const char **reg_dummy;
2288 # endif /* DEFINED_ONCE */
2290 static PREFIX(register_info_type) *PREFIX(reg_info);
2291 static PREFIX(register_info_type) *PREFIX(reg_info_dummy);
2293 /* Make the register vectors big enough for NUM_REGS registers,
2294 but don't make them smaller. */
2297 PREFIX(regex_grow_registers) (num_regs)
2300 if (num_regs > regs_allocated_size)
2302 RETALLOC_IF (regstart, num_regs, const char *);
2303 RETALLOC_IF (regend, num_regs, const char *);
2304 RETALLOC_IF (old_regstart, num_regs, const char *);
2305 RETALLOC_IF (old_regend, num_regs, const char *);
2306 RETALLOC_IF (best_regstart, num_regs, const char *);
2307 RETALLOC_IF (best_regend, num_regs, const char *);
2308 RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type));
2309 RETALLOC_IF (reg_dummy, num_regs, const char *);
2310 RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type));
2312 regs_allocated_size = num_regs;
2316 # endif /* not MATCH_MAY_ALLOCATE */
2318 # ifndef DEFINED_ONCE
2319 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2322 # endif /* not DEFINED_ONCE */
2324 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2325 Returns one of error codes defined in `regex.h', or zero for success.
2327 Assumes the `allocated' (and perhaps `buffer') and `translate'
2328 fields are set in BUFP on entry.
2330 If it succeeds, results are put in BUFP (if it returns an error, the
2331 contents of BUFP are undefined):
2332 `buffer' is the compiled pattern;
2333 `syntax' is set to SYNTAX;
2334 `used' is set to the length of the compiled pattern;
2335 `fastmap_accurate' is zero;
2336 `re_nsub' is the number of subexpressions in PATTERN;
2337 `not_bol' and `not_eol' are zero;
2339 The `fastmap' and `newline_anchor' fields are neither
2340 examined nor set. */
2342 /* Return, freeing storage we allocated. */
2344 # define FREE_STACK_RETURN(value) \
2345 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2347 # define FREE_STACK_RETURN(value) \
2348 return (free (compile_stack.stack), value)
2351 static reg_errcode_t
2352 PREFIX(regex_compile) (ARG_PREFIX(pattern), ARG_PREFIX(size), syntax, bufp)
2353 const char *ARG_PREFIX(pattern);
2354 size_t ARG_PREFIX(size);
2355 reg_syntax_t syntax;
2356 struct re_pattern_buffer *bufp;
2358 /* We fetch characters from PATTERN here. Even though PATTERN is
2359 `char *' (i.e., signed), we declare these variables as unsigned, so
2360 they can be reliably used as array indices. */
2361 register UCHAR_T c, c1;
2364 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2365 CHAR_T *pattern, *COMPILED_BUFFER_VAR;
2367 /* offset buffer for optimization. See convert_mbs_to_wc. */
2368 int *mbs_offset = NULL;
2369 /* It hold whether each wchar_t is binary data or not. */
2370 char *is_binary = NULL;
2371 /* A flag whether exactn is handling binary data or not. */
2372 char is_exactn_bin = FALSE;
2375 /* A random temporary spot in PATTERN. */
2378 /* Points to the end of the buffer, where we should append. */
2379 register UCHAR_T *b;
2381 /* Keeps track of unclosed groups. */
2382 compile_stack_type compile_stack;
2384 /* Points to the current (ending) position in the pattern. */
2389 const CHAR_T *p = pattern;
2390 const CHAR_T *pend = pattern + size;
2393 /* How to translate the characters in the pattern. */
2394 RE_TRANSLATE_TYPE translate = bufp->translate;
2396 /* Address of the count-byte of the most recently inserted `exactn'
2397 command. This makes it possible to tell if a new exact-match
2398 character can be added to that command or if the character requires
2399 a new `exactn' command. */
2400 UCHAR_T *pending_exact = 0;
2402 /* Address of start of the most recently finished expression.
2403 This tells, e.g., postfix * where to find the start of its
2404 operand. Reset at the beginning of groups and alternatives. */
2405 UCHAR_T *laststart = 0;
2407 /* Address of beginning of regexp, or inside of last group. */
2410 /* Address of the place where a forward jump should go to the end of
2411 the containing expression. Each alternative of an `or' -- except the
2412 last -- ends with a forward jump of this sort. */
2413 UCHAR_T *fixup_alt_jump = 0;
2415 /* Counts open-groups as they are encountered. Remembered for the
2416 matching close-group on the compile stack, so the same register
2417 number is put in the stop_memory as the start_memory. */
2418 regnum_t regnum = 0;
2421 /* Initialize the wchar_t PATTERN and offset_buffer. */
2422 p = pend = pattern = TALLOC(csize + 1, CHAR_T);
2423 mbs_offset = TALLOC(csize + 1, int);
2424 is_binary = TALLOC(csize + 1, char);
2425 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL)
2432 pattern[csize] = L'\0'; /* sentinel */
2433 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary);
2445 DEBUG_PRINT1 ("\nCompiling pattern: ");
2448 unsigned debug_count;
2450 for (debug_count = 0; debug_count < size; debug_count++)
2451 PUT_CHAR (pattern[debug_count]);
2456 /* Initialize the compile stack. */
2457 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2458 if (compile_stack.stack == NULL)
2468 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2469 compile_stack.avail = 0;
2471 /* Initialize the pattern buffer. */
2472 bufp->syntax = syntax;
2473 bufp->fastmap_accurate = 0;
2474 bufp->not_bol = bufp->not_eol = 0;
2476 /* Set `used' to zero, so that if we return an error, the pattern
2477 printer (for debugging) will think there's no pattern. We reset it
2481 /* Always count groups, whether or not bufp->no_sub is set. */
2484 #if !defined emacs && !defined SYNTAX_TABLE
2485 /* Initialize the syntax table. */
2486 init_syntax_once ();
2489 if (bufp->allocated == 0)
2492 { /* If zero allocated, but buffer is non-null, try to realloc
2493 enough space. This loses if buffer's address is bogus, but
2494 that is the user's responsibility. */
2496 /* Free bufp->buffer and allocate an array for wchar_t pattern
2499 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T),
2502 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T);
2506 { /* Caller did not allocate a buffer. Do it for them. */
2507 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T),
2511 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE);
2513 bufp->buffer = (char*)COMPILED_BUFFER_VAR;
2515 bufp->allocated = INIT_BUF_SIZE;
2519 COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer;
2522 begalt = b = COMPILED_BUFFER_VAR;
2524 /* Loop through the uncompiled pattern until we're at the end. */
2533 if ( /* If at start of pattern, it's an operator. */
2535 /* If context independent, it's an operator. */
2536 || syntax & RE_CONTEXT_INDEP_ANCHORS
2537 /* Otherwise, depends on what's come before. */
2538 || PREFIX(at_begline_loc_p) (pattern, p, syntax))
2548 if ( /* If at end of pattern, it's an operator. */
2550 /* If context independent, it's an operator. */
2551 || syntax & RE_CONTEXT_INDEP_ANCHORS
2552 /* Otherwise, depends on what's next. */
2553 || PREFIX(at_endline_loc_p) (p, pend, syntax))
2563 if ((syntax & RE_BK_PLUS_QM)
2564 || (syntax & RE_LIMITED_OPS))
2568 /* If there is no previous pattern... */
2571 if (syntax & RE_CONTEXT_INVALID_OPS)
2572 FREE_STACK_RETURN (REG_BADRPT);
2573 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2578 /* Are we optimizing this jump? */
2579 boolean keep_string_p = false;
2581 /* 1 means zero (many) matches is allowed. */
2582 char zero_times_ok = 0, many_times_ok = 0;
2584 /* If there is a sequence of repetition chars, collapse it
2585 down to just one (the right one). We can't combine
2586 interval operators with these because of, e.g., `a{2}*',
2587 which should only match an even number of `a's. */
2591 zero_times_ok |= c != '+';
2592 many_times_ok |= c != '?';
2600 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2603 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2605 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2608 if (!(c1 == '+' || c1 == '?'))
2623 /* If we get here, we found another repeat character. */
2626 /* Star, etc. applied to an empty pattern is equivalent
2627 to an empty pattern. */
2631 /* Now we know whether or not zero matches is allowed
2632 and also whether or not two or more matches is allowed. */
2634 { /* More than one repetition is allowed, so put in at the
2635 end a backward relative jump from `b' to before the next
2636 jump we're going to put in below (which jumps from
2637 laststart to after this jump).
2639 But if we are at the `*' in the exact sequence `.*\n',
2640 insert an unconditional jump backwards to the .,
2641 instead of the beginning of the loop. This way we only
2642 push a failure point once, instead of every time
2643 through the loop. */
2644 assert (p - 1 > pattern);
2646 /* Allocate the space for the jump. */
2647 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2649 /* We know we are not at the first character of the pattern,
2650 because laststart was nonzero. And we've already
2651 incremented `p', by the way, to be the character after
2652 the `*'. Do we have to do something analogous here
2653 for null bytes, because of RE_DOT_NOT_NULL? */
2654 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2656 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2657 && !(syntax & RE_DOT_NEWLINE))
2658 { /* We have .*\n. */
2659 STORE_JUMP (jump, b, laststart);
2660 keep_string_p = true;
2663 /* Anything else. */
2664 STORE_JUMP (maybe_pop_jump, b, laststart -
2665 (1 + OFFSET_ADDRESS_SIZE));
2667 /* We've added more stuff to the buffer. */
2668 b += 1 + OFFSET_ADDRESS_SIZE;
2671 /* On failure, jump from laststart to b + 3, which will be the
2672 end of the buffer after this jump is inserted. */
2673 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2675 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2676 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2678 laststart, b + 1 + OFFSET_ADDRESS_SIZE);
2680 b += 1 + OFFSET_ADDRESS_SIZE;
2684 /* At least one repetition is required, so insert a
2685 `dummy_failure_jump' before the initial
2686 `on_failure_jump' instruction of the loop. This
2687 effects a skip over that instruction the first time
2688 we hit that loop. */
2689 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2690 INSERT_JUMP (dummy_failure_jump, laststart, laststart +
2691 2 + 2 * OFFSET_ADDRESS_SIZE);
2692 b += 1 + OFFSET_ADDRESS_SIZE;
2706 boolean had_char_class = false;
2708 CHAR_T range_start = 0xffffffff;
2710 unsigned int range_start = 0xffffffff;
2712 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2715 /* We assume a charset(_not) structure as a wchar_t array.
2716 charset[0] = (re_opcode_t) charset(_not)
2717 charset[1] = l (= length of char_classes)
2718 charset[2] = m (= length of collating_symbols)
2719 charset[3] = n (= length of equivalence_classes)
2720 charset[4] = o (= length of char_ranges)
2721 charset[5] = p (= length of chars)
2723 charset[6] = char_class (wctype_t)
2724 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2726 charset[l+5] = char_class (wctype_t)
2728 charset[l+6] = collating_symbol (wchar_t)
2730 charset[l+m+5] = collating_symbol (wchar_t)
2731 ifdef _LIBC we use the index if
2732 _NL_COLLATE_SYMB_EXTRAMB instead of
2735 charset[l+m+6] = equivalence_classes (wchar_t)
2737 charset[l+m+n+5] = equivalence_classes (wchar_t)
2738 ifdef _LIBC we use the index in
2739 _NL_COLLATE_WEIGHT instead of
2742 charset[l+m+n+6] = range_start
2743 charset[l+m+n+7] = range_end
2745 charset[l+m+n+2o+4] = range_start
2746 charset[l+m+n+2o+5] = range_end
2747 ifdef _LIBC we use the value looked up
2748 in _NL_COLLATE_COLLSEQ instead of
2751 charset[l+m+n+2o+6] = char
2753 charset[l+m+n+2o+p+5] = char
2757 /* We need at least 6 spaces: the opcode, the length of
2758 char_classes, the length of collating_symbols, the length of
2759 equivalence_classes, the length of char_ranges, the length of
2761 GET_BUFFER_SPACE (6);
2763 /* Save b as laststart. And We use laststart as the pointer
2764 to the first element of the charset here.
2765 In other words, laststart[i] indicates charset[i]. */
2768 /* We test `*p == '^' twice, instead of using an if
2769 statement, so we only need one BUF_PUSH. */
2770 BUF_PUSH (*p == '^' ? charset_not : charset);
2774 /* Push the length of char_classes, the length of
2775 collating_symbols, the length of equivalence_classes, the
2776 length of char_ranges and the length of chars. */
2777 BUF_PUSH_3 (0, 0, 0);
2780 /* Remember the first position in the bracket expression. */
2783 /* charset_not matches newline according to a syntax bit. */
2784 if ((re_opcode_t) b[-6] == charset_not
2785 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2788 laststart[5]++; /* Update the length of characters */
2791 /* Read in characters and ranges, setting map bits. */
2794 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2798 /* \ might escape characters inside [...] and [^...]. */
2799 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2801 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2805 laststart[5]++; /* Update the length of chars */
2810 /* Could be the end of the bracket expression. If it's
2811 not (i.e., when the bracket expression is `[]' so
2812 far), the ']' character bit gets set way below. */
2813 if (c == ']' && p != p1 + 1)
2816 /* Look ahead to see if it's a range when the last thing
2817 was a character class. */
2818 if (had_char_class && c == '-' && *p != ']')
2819 FREE_STACK_RETURN (REG_ERANGE);
2821 /* Look ahead to see if it's a range when the last thing
2822 was a character: if this is a hyphen not at the
2823 beginning or the end of a list, then it's the range
2826 && !(p - 2 >= pattern && p[-2] == '[')
2827 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2831 /* Allocate the space for range_start and range_end. */
2832 GET_BUFFER_SPACE (2);
2833 /* Update the pointer to indicate end of buffer. */
2835 ret = wcs_compile_range (range_start, &p, pend, translate,
2836 syntax, b, laststart);
2837 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2838 range_start = 0xffffffff;
2840 else if (p[0] == '-' && p[1] != ']')
2841 { /* This handles ranges made up of characters only. */
2844 /* Move past the `-'. */
2846 /* Allocate the space for range_start and range_end. */
2847 GET_BUFFER_SPACE (2);
2848 /* Update the pointer to indicate end of buffer. */
2850 ret = wcs_compile_range (c, &p, pend, translate, syntax, b,
2852 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2853 range_start = 0xffffffff;
2856 /* See if we're at the beginning of a possible character
2858 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2859 { /* Leave room for the null. */
2860 char str[CHAR_CLASS_MAX_LENGTH + 1];
2865 /* If pattern is `[[:'. */
2866 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2871 if ((c == ':' && *p == ']') || p == pend)
2873 if (c1 < CHAR_CLASS_MAX_LENGTH)
2876 /* This is in any case an invalid class name. */
2881 /* If isn't a word bracketed by `[:' and `:]':
2882 undo the ending character, the letters, and leave
2883 the leading `:' and `[' (but store them as character). */
2884 if (c == ':' && *p == ']')
2889 /* Query the character class as wctype_t. */
2890 wt = IS_CHAR_CLASS (str);
2892 FREE_STACK_RETURN (REG_ECTYPE);
2894 /* Throw away the ] at the end of the character
2898 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2900 /* Allocate the space for character class. */
2901 GET_BUFFER_SPACE(CHAR_CLASS_SIZE);
2902 /* Update the pointer to indicate end of buffer. */
2903 b += CHAR_CLASS_SIZE;
2904 /* Move data which follow character classes
2905 not to violate the data. */
2906 insert_space(CHAR_CLASS_SIZE,
2907 laststart + 6 + laststart[1],
2909 alignedp = ((uintptr_t)(laststart + 6 + laststart[1])
2910 + __alignof__(wctype_t) - 1)
2911 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2912 /* Store the character class. */
2913 *((wctype_t*)alignedp) = wt;
2914 /* Update length of char_classes */
2915 laststart[1] += CHAR_CLASS_SIZE;
2917 had_char_class = true;
2926 laststart[5] += 2; /* Update the length of characters */
2928 had_char_class = false;
2931 else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '='
2934 CHAR_T str[128]; /* Should be large enough. */
2935 CHAR_T delim = *p; /* '=' or '.' */
2938 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2943 /* If pattern is `[[=' or '[[.'. */
2944 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2949 if ((c == delim && *p == ']') || p == pend)
2951 if (c1 < sizeof (str) - 1)
2954 /* This is in any case an invalid class name. */
2959 if (c == delim && *p == ']' && str[0] != '\0')
2961 unsigned int i, offset;
2962 /* If we have no collation data we use the default
2963 collation in which each character is in a class
2964 by itself. It also means that ASCII is the
2965 character set and therefore we cannot have character
2966 with more than one byte in the multibyte
2969 /* If not defined _LIBC, we push the name and
2970 `\0' for the sake of matching performance. */
2971 int datasize = c1 + 1;
2979 FREE_STACK_RETURN (REG_ECOLLATE);
2984 const int32_t *table;
2985 const int32_t *weights;
2986 const int32_t *extra;
2987 const int32_t *indirect;
2990 /* This #include defines a local function! */
2991 # include <locale/weightwc.h>
2995 /* We push the index for equivalence class. */
2998 table = (const int32_t *)
2999 _NL_CURRENT (LC_COLLATE,
3000 _NL_COLLATE_TABLEWC);
3001 weights = (const int32_t *)
3002 _NL_CURRENT (LC_COLLATE,
3003 _NL_COLLATE_WEIGHTWC);
3004 extra = (const int32_t *)
3005 _NL_CURRENT (LC_COLLATE,
3006 _NL_COLLATE_EXTRAWC);
3007 indirect = (const int32_t *)
3008 _NL_CURRENT (LC_COLLATE,
3009 _NL_COLLATE_INDIRECTWC);
3011 idx = findidx ((const wint_t**)&cp);
3012 if (idx == 0 || cp < (wint_t*) str + c1)
3013 /* This is no valid character. */
3014 FREE_STACK_RETURN (REG_ECOLLATE);
3016 str[0] = (wchar_t)idx;
3018 else /* delim == '.' */
3020 /* We push collation sequence value
3021 for collating symbol. */
3023 const int32_t *symb_table;
3024 const unsigned char *extra;
3031 /* We have to convert the name to a single-byte
3032 string. This is possible since the names
3033 consist of ASCII characters and the internal
3034 representation is UCS4. */
3035 for (i = 0; i < c1; ++i)
3036 char_str[i] = str[i];
3039 _NL_CURRENT_WORD (LC_COLLATE,
3040 _NL_COLLATE_SYMB_HASH_SIZEMB);
3041 symb_table = (const int32_t *)
3042 _NL_CURRENT (LC_COLLATE,
3043 _NL_COLLATE_SYMB_TABLEMB);
3044 extra = (const unsigned char *)
3045 _NL_CURRENT (LC_COLLATE,
3046 _NL_COLLATE_SYMB_EXTRAMB);
3048 /* Locate the character in the hashing table. */
3049 hash = elem_hash (char_str, c1);
3052 elem = hash % table_size;
3053 second = hash % (table_size - 2);
3054 while (symb_table[2 * elem] != 0)
3056 /* First compare the hashing value. */
3057 if (symb_table[2 * elem] == hash
3058 && c1 == extra[symb_table[2 * elem + 1]]
3059 && memcmp (char_str,
3060 &extra[symb_table[2 * elem + 1]
3063 /* Yep, this is the entry. */
3064 idx = symb_table[2 * elem + 1];
3065 idx += 1 + extra[idx];
3073 if (symb_table[2 * elem] != 0)
3075 /* Compute the index of the byte sequence
3077 idx += 1 + extra[idx];
3078 /* Adjust for the alignment. */
3079 idx = (idx + 3) & ~3;
3081 str[0] = (wchar_t) idx + 4;
3083 else if (symb_table[2 * elem] == 0 && c1 == 1)
3085 /* No valid character. Match it as a
3086 single byte character. */
3087 had_char_class = false;
3089 /* Update the length of characters */
3091 range_start = str[0];
3093 /* Throw away the ] at the end of the
3094 collating symbol. */
3096 /* exit from the switch block. */
3100 FREE_STACK_RETURN (REG_ECOLLATE);
3105 /* Throw away the ] at the end of the equivalence
3106 class (or collating symbol). */
3109 /* Allocate the space for the equivalence class
3110 (or collating symbol) (and '\0' if needed). */
3111 GET_BUFFER_SPACE(datasize);
3112 /* Update the pointer to indicate end of buffer. */
3116 { /* equivalence class */
3117 /* Calculate the offset of char_ranges,
3118 which is next to equivalence_classes. */
3119 offset = laststart[1] + laststart[2]
3122 insert_space(datasize, laststart + offset, b - 1);
3124 /* Write the equivalence_class and \0. */
3125 for (i = 0 ; i < datasize ; i++)
3126 laststart[offset + i] = str[i];
3128 /* Update the length of equivalence_classes. */
3129 laststart[3] += datasize;
3130 had_char_class = true;
3132 else /* delim == '.' */
3133 { /* collating symbol */
3134 /* Calculate the offset of the equivalence_classes,
3135 which is next to collating_symbols. */
3136 offset = laststart[1] + laststart[2] + 6;
3137 /* Insert space and write the collationg_symbol
3139 insert_space(datasize, laststart + offset, b-1);
3140 for (i = 0 ; i < datasize ; i++)
3141 laststart[offset + i] = str[i];
3143 /* In re_match_2_internal if range_start < -1, we
3144 assume -range_start is the offset of the
3145 collating symbol which is specified as
3146 the character of the range start. So we assign
3147 -(laststart[1] + laststart[2] + 6) to
3149 range_start = -(laststart[1] + laststart[2] + 6);
3150 /* Update the length of collating_symbol. */
3151 laststart[2] += datasize;
3152 had_char_class = false;
3162 laststart[5] += 2; /* Update the length of characters */
3163 range_start = delim;
3164 had_char_class = false;
3169 had_char_class = false;
3171 laststart[5]++; /* Update the length of characters */
3177 /* Ensure that we have enough space to push a charset: the
3178 opcode, the length count, and the bitset; 34 bytes in all. */
3179 GET_BUFFER_SPACE (34);
3183 /* We test `*p == '^' twice, instead of using an if
3184 statement, so we only need one BUF_PUSH. */
3185 BUF_PUSH (*p == '^' ? charset_not : charset);
3189 /* Remember the first position in the bracket expression. */
3192 /* Push the number of bytes in the bitmap. */
3193 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
3195 /* Clear the whole map. */
3196 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
3198 /* charset_not matches newline according to a syntax bit. */
3199 if ((re_opcode_t) b[-2] == charset_not
3200 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
3201 SET_LIST_BIT ('\n');
3203 /* Read in characters and ranges, setting map bits. */
3206 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3210 /* \ might escape characters inside [...] and [^...]. */
3211 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
3213 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3221 /* Could be the end of the bracket expression. If it's
3222 not (i.e., when the bracket expression is `[]' so
3223 far), the ']' character bit gets set way below. */
3224 if (c == ']' && p != p1 + 1)
3227 /* Look ahead to see if it's a range when the last thing
3228 was a character class. */
3229 if (had_char_class && c == '-' && *p != ']')
3230 FREE_STACK_RETURN (REG_ERANGE);
3232 /* Look ahead to see if it's a range when the last thing
3233 was a character: if this is a hyphen not at the
3234 beginning or the end of a list, then it's the range
3237 && !(p - 2 >= pattern && p[-2] == '[')
3238 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
3242 = byte_compile_range (range_start, &p, pend, translate,
3244 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3245 range_start = 0xffffffff;
3248 else if (p[0] == '-' && p[1] != ']')
3249 { /* This handles ranges made up of characters only. */
3252 /* Move past the `-'. */
3255 ret = byte_compile_range (c, &p, pend, translate, syntax, b);
3256 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3257 range_start = 0xffffffff;
3260 /* See if we're at the beginning of a possible character
3263 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3264 { /* Leave room for the null. */
3265 char str[CHAR_CLASS_MAX_LENGTH + 1];
3270 /* If pattern is `[[:'. */
3271 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3276 if ((c == ':' && *p == ']') || p == pend)
3278 if (c1 < CHAR_CLASS_MAX_LENGTH)
3281 /* This is in any case an invalid class name. */
3286 /* If isn't a word bracketed by `[:' and `:]':
3287 undo the ending character, the letters, and leave
3288 the leading `:' and `[' (but set bits for them). */
3289 if (c == ':' && *p == ']')
3291 # if defined _LIBC || WIDE_CHAR_SUPPORT
3292 boolean is_lower = STREQ (str, "lower");
3293 boolean is_upper = STREQ (str, "upper");
3297 wt = IS_CHAR_CLASS (str);
3299 FREE_STACK_RETURN (REG_ECTYPE);
3301 /* Throw away the ] at the end of the character
3305 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3307 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
3309 if (iswctype (btowc (ch), wt))
3312 if (translate && (is_upper || is_lower)
3313 && (ISUPPER (ch) || ISLOWER (ch)))
3317 had_char_class = true;
3320 boolean is_alnum = STREQ (str, "alnum");
3321 boolean is_alpha = STREQ (str, "alpha");
3322 boolean is_blank = STREQ (str, "blank");
3323 boolean is_cntrl = STREQ (str, "cntrl");
3324 boolean is_digit = STREQ (str, "digit");
3325 boolean is_graph = STREQ (str, "graph");
3326 boolean is_lower = STREQ (str, "lower");
3327 boolean is_print = STREQ (str, "print");
3328 boolean is_punct = STREQ (str, "punct");
3329 boolean is_space = STREQ (str, "space");
3330 boolean is_upper = STREQ (str, "upper");
3331 boolean is_xdigit = STREQ (str, "xdigit");
3333 if (!IS_CHAR_CLASS (str))
3334 FREE_STACK_RETURN (REG_ECTYPE);
3336 /* Throw away the ] at the end of the character
3340 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3342 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
3344 /* This was split into 3 if's to
3345 avoid an arbitrary limit in some compiler. */
3346 if ( (is_alnum && ISALNUM (ch))
3347 || (is_alpha && ISALPHA (ch))
3348 || (is_blank && ISBLANK (ch))
3349 || (is_cntrl && ISCNTRL (ch)))
3351 if ( (is_digit && ISDIGIT (ch))
3352 || (is_graph && ISGRAPH (ch))
3353 || (is_lower && ISLOWER (ch))
3354 || (is_print && ISPRINT (ch)))
3356 if ( (is_punct && ISPUNCT (ch))
3357 || (is_space && ISSPACE (ch))
3358 || (is_upper && ISUPPER (ch))
3359 || (is_xdigit && ISXDIGIT (ch)))
3361 if ( translate && (is_upper || is_lower)
3362 && (ISUPPER (ch) || ISLOWER (ch)))
3365 had_char_class = true;
3366 # endif /* libc || wctype.h */
3376 had_char_class = false;
3379 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
3381 unsigned char str[MB_LEN_MAX + 1];
3384 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3390 /* If pattern is `[[='. */
3391 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3396 if ((c == '=' && *p == ']') || p == pend)
3398 if (c1 < MB_LEN_MAX)
3401 /* This is in any case an invalid class name. */
3406 if (c == '=' && *p == ']' && str[0] != '\0')
3408 /* If we have no collation data we use the default
3409 collation in which each character is in a class
3410 by itself. It also means that ASCII is the
3411 character set and therefore we cannot have character
3412 with more than one byte in the multibyte
3419 FREE_STACK_RETURN (REG_ECOLLATE);
3421 /* Throw away the ] at the end of the equivalence
3425 /* Set the bit for the character. */
3426 SET_LIST_BIT (str[0]);
3431 /* Try to match the byte sequence in `str' against
3432 those known to the collate implementation.
3433 First find out whether the bytes in `str' are
3434 actually from exactly one character. */
3435 const int32_t *table;
3436 const unsigned char *weights;
3437 const unsigned char *extra;
3438 const int32_t *indirect;
3440 const unsigned char *cp = str;
3443 /* This #include defines a local function! */
3444 # include <locale/weight.h>
3446 table = (const int32_t *)
3447 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3448 weights = (const unsigned char *)
3449 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3450 extra = (const unsigned char *)
3451 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3452 indirect = (const int32_t *)
3453 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3455 idx = findidx (&cp);
3456 if (idx == 0 || cp < str + c1)
3457 /* This is no valid character. */
3458 FREE_STACK_RETURN (REG_ECOLLATE);
3460 /* Throw away the ] at the end of the equivalence
3464 /* Now we have to go throught the whole table
3465 and find all characters which have the same
3468 XXX Note that this is not entirely correct.
3469 we would have to match multibyte sequences
3470 but this is not possible with the current
3472 for (ch = 1; ch < 256; ++ch)
3473 /* XXX This test would have to be changed if we
3474 would allow matching multibyte sequences. */
3477 int32_t idx2 = table[ch];
3478 size_t len = weights[idx2];
3480 /* Test whether the lenghts match. */
3481 if (weights[idx] == len)
3483 /* They do. New compare the bytes of
3488 && (weights[idx + 1 + cnt]
3489 == weights[idx2 + 1 + cnt]))
3493 /* They match. Mark the character as
3500 had_char_class = true;
3510 had_char_class = false;
3513 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
3515 unsigned char str[128]; /* Should be large enough. */
3518 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3524 /* If pattern is `[[.'. */
3525 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3530 if ((c == '.' && *p == ']') || p == pend)
3532 if (c1 < sizeof (str))
3535 /* This is in any case an invalid class name. */
3540 if (c == '.' && *p == ']' && str[0] != '\0')
3542 /* If we have no collation data we use the default
3543 collation in which each character is the name
3544 for its own class which contains only the one
3545 character. It also means that ASCII is the
3546 character set and therefore we cannot have character
3547 with more than one byte in the multibyte
3554 FREE_STACK_RETURN (REG_ECOLLATE);
3556 /* Throw away the ] at the end of the equivalence
3560 /* Set the bit for the character. */
3561 SET_LIST_BIT (str[0]);
3562 range_start = ((const unsigned char *) str)[0];
3567 /* Try to match the byte sequence in `str' against
3568 those known to the collate implementation.
3569 First find out whether the bytes in `str' are
3570 actually from exactly one character. */
3572 const int32_t *symb_table;
3573 const unsigned char *extra;
3580 _NL_CURRENT_WORD (LC_COLLATE,
3581 _NL_COLLATE_SYMB_HASH_SIZEMB);
3582 symb_table = (const int32_t *)
3583 _NL_CURRENT (LC_COLLATE,
3584 _NL_COLLATE_SYMB_TABLEMB);
3585 extra = (const unsigned char *)
3586 _NL_CURRENT (LC_COLLATE,
3587 _NL_COLLATE_SYMB_EXTRAMB);
3589 /* Locate the character in the hashing table. */
3590 hash = elem_hash (str, c1);
3593 elem = hash % table_size;
3594 second = hash % (table_size - 2);
3595 while (symb_table[2 * elem] != 0)
3597 /* First compare the hashing value. */
3598 if (symb_table[2 * elem] == hash
3599 && c1 == extra[symb_table[2 * elem + 1]]
3601 &extra[symb_table[2 * elem + 1]
3605 /* Yep, this is the entry. */
3606 idx = symb_table[2 * elem + 1];
3607 idx += 1 + extra[idx];
3615 if (symb_table[2 * elem] == 0)
3616 /* This is no valid character. */
3617 FREE_STACK_RETURN (REG_ECOLLATE);
3619 /* Throw away the ] at the end of the equivalence
3623 /* Now add the multibyte character(s) we found
3626 XXX Note that this is not entirely correct.
3627 we would have to match multibyte sequences
3628 but this is not possible with the current
3629 implementation. Also, we have to match
3630 collating symbols, which expand to more than
3631 one file, as a whole and not allow the
3632 individual bytes. */
3635 range_start = extra[idx];
3638 SET_LIST_BIT (extra[idx]);
3643 had_char_class = false;
3653 had_char_class = false;
3658 had_char_class = false;
3664 /* Discard any (non)matching list bytes that are all 0 at the
3665 end of the map. Decrease the map-length byte too. */
3666 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3675 if (syntax & RE_NO_BK_PARENS)
3682 if (syntax & RE_NO_BK_PARENS)
3689 if (syntax & RE_NEWLINE_ALT)
3696 if (syntax & RE_NO_BK_VBAR)
3703 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3704 goto handle_interval;
3710 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3712 /* Do not translate the character after the \, so that we can
3713 distinguish, e.g., \B from \b, even if we normally would
3714 translate, e.g., B to b. */
3720 if (syntax & RE_NO_BK_PARENS)
3721 goto normal_backslash;
3727 if (COMPILE_STACK_FULL)
3729 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3730 compile_stack_elt_t);
3731 if (compile_stack.stack == NULL) return REG_ESPACE;
3733 compile_stack.size <<= 1;
3736 /* These are the values to restore when we hit end of this
3737 group. They are all relative offsets, so that if the
3738 whole pattern moves because of realloc, they will still
3740 COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR;
3741 COMPILE_STACK_TOP.fixup_alt_jump
3742 = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0;
3743 COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR;
3744 COMPILE_STACK_TOP.regnum = regnum;
3746 /* We will eventually replace the 0 with the number of
3747 groups inner to this one. But do not push a
3748 start_memory for groups beyond the last one we can
3749 represent in the compiled pattern. */
3750 if (regnum <= MAX_REGNUM)
3752 COMPILE_STACK_TOP.inner_group_offset = b
3753 - COMPILED_BUFFER_VAR + 2;
3754 BUF_PUSH_3 (start_memory, regnum, 0);
3757 compile_stack.avail++;
3762 /* If we've reached MAX_REGNUM groups, then this open
3763 won't actually generate any code, so we'll have to
3764 clear pending_exact explicitly. */
3770 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3772 if (COMPILE_STACK_EMPTY)
3774 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3775 goto normal_backslash;
3777 FREE_STACK_RETURN (REG_ERPAREN);
3782 { /* Push a dummy failure point at the end of the
3783 alternative for a possible future
3784 `pop_failure_jump' to pop. See comments at
3785 `push_dummy_failure' in `re_match_2'. */
3786 BUF_PUSH (push_dummy_failure);
3788 /* We allocated space for this jump when we assigned
3789 to `fixup_alt_jump', in the `handle_alt' case below. */
3790 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
3793 /* See similar code for backslashed left paren above. */
3794 if (COMPILE_STACK_EMPTY)
3796 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3799 FREE_STACK_RETURN (REG_ERPAREN);
3802 /* Since we just checked for an empty stack above, this
3803 ``can't happen''. */
3804 assert (compile_stack.avail != 0);
3806 /* We don't just want to restore into `regnum', because
3807 later groups should continue to be numbered higher,
3808 as in `(ab)c(de)' -- the second group is #2. */
3809 regnum_t this_group_regnum;
3811 compile_stack.avail--;
3812 begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset;
3814 = COMPILE_STACK_TOP.fixup_alt_jump
3815 ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1
3817 laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset;
3818 this_group_regnum = COMPILE_STACK_TOP.regnum;
3819 /* If we've reached MAX_REGNUM groups, then this open
3820 won't actually generate any code, so we'll have to
3821 clear pending_exact explicitly. */
3824 /* We're at the end of the group, so now we know how many
3825 groups were inside this one. */
3826 if (this_group_regnum <= MAX_REGNUM)
3828 UCHAR_T *inner_group_loc
3829 = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset;
3831 *inner_group_loc = regnum - this_group_regnum;
3832 BUF_PUSH_3 (stop_memory, this_group_regnum,
3833 regnum - this_group_regnum);
3839 case '|': /* `\|'. */
3840 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3841 goto normal_backslash;
3843 if (syntax & RE_LIMITED_OPS)
3846 /* Insert before the previous alternative a jump which
3847 jumps to this alternative if the former fails. */
3848 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3849 INSERT_JUMP (on_failure_jump, begalt,
3850 b + 2 + 2 * OFFSET_ADDRESS_SIZE);
3852 b += 1 + OFFSET_ADDRESS_SIZE;
3854 /* The alternative before this one has a jump after it
3855 which gets executed if it gets matched. Adjust that
3856 jump so it will jump to this alternative's analogous
3857 jump (put in below, which in turn will jump to the next
3858 (if any) alternative's such jump, etc.). The last such
3859 jump jumps to the correct final destination. A picture:
3865 If we are at `b', then fixup_alt_jump right now points to a
3866 three-byte space after `a'. We'll put in the jump, set
3867 fixup_alt_jump to right after `b', and leave behind three
3868 bytes which we'll fill in when we get to after `c'. */
3871 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3873 /* Mark and leave space for a jump after this alternative,
3874 to be filled in later either by next alternative or
3875 when know we're at the end of a series of alternatives. */
3877 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3878 b += 1 + OFFSET_ADDRESS_SIZE;
3886 /* If \{ is a literal. */
3887 if (!(syntax & RE_INTERVALS)
3888 /* If we're at `\{' and it's not the open-interval
3890 || (syntax & RE_NO_BK_BRACES))
3891 goto normal_backslash;
3895 /* If got here, then the syntax allows intervals. */
3897 /* At least (most) this many matches must be made. */
3898 int lower_bound = -1, upper_bound = -1;
3900 /* Place in the uncompiled pattern (i.e., just after
3901 the '{') to go back to if the interval is invalid. */
3902 const CHAR_T *beg_interval = p;
3905 goto invalid_interval;
3907 GET_UNSIGNED_NUMBER (lower_bound);
3911 GET_UNSIGNED_NUMBER (upper_bound);
3912 if (upper_bound < 0)
3913 upper_bound = RE_DUP_MAX;
3916 /* Interval such as `{1}' => match exactly once. */
3917 upper_bound = lower_bound;
3919 if (! (0 <= lower_bound && lower_bound <= upper_bound))
3920 goto invalid_interval;
3922 if (!(syntax & RE_NO_BK_BRACES))
3924 if (c != '\\' || p == pend)
3925 goto invalid_interval;
3930 goto invalid_interval;
3932 /* If it's invalid to have no preceding re. */
3935 if (syntax & RE_CONTEXT_INVALID_OPS
3936 && !(syntax & RE_INVALID_INTERVAL_ORD))
3937 FREE_STACK_RETURN (REG_BADRPT);
3938 else if (syntax & RE_CONTEXT_INDEP_OPS)
3941 goto unfetch_interval;
3944 /* We just parsed a valid interval. */
3946 if (RE_DUP_MAX < upper_bound)
3947 FREE_STACK_RETURN (REG_BADBR);
3949 /* If the upper bound is zero, don't want to succeed at
3950 all; jump from `laststart' to `b + 3', which will be
3951 the end of the buffer after we insert the jump. */
3952 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3953 instead of 'b + 3'. */
3954 if (upper_bound == 0)
3956 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3957 INSERT_JUMP (jump, laststart, b + 1
3958 + OFFSET_ADDRESS_SIZE);
3959 b += 1 + OFFSET_ADDRESS_SIZE;
3962 /* Otherwise, we have a nontrivial interval. When
3963 we're all done, the pattern will look like:
3964 set_number_at <jump count> <upper bound>
3965 set_number_at <succeed_n count> <lower bound>
3966 succeed_n <after jump addr> <succeed_n count>
3968 jump_n <succeed_n addr> <jump count>
3969 (The upper bound and `jump_n' are omitted if
3970 `upper_bound' is 1, though.) */
3972 { /* If the upper bound is > 1, we need to insert
3973 more at the end of the loop. */
3974 unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE +
3975 (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE);
3977 GET_BUFFER_SPACE (nbytes);
3979 /* Initialize lower bound of the `succeed_n', even
3980 though it will be set during matching by its
3981 attendant `set_number_at' (inserted next),
3982 because `re_compile_fastmap' needs to know.
3983 Jump to the `jump_n' we might insert below. */
3984 INSERT_JUMP2 (succeed_n, laststart,
3985 b + 1 + 2 * OFFSET_ADDRESS_SIZE
3986 + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE)
3988 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3990 /* Code to initialize the lower bound. Insert
3991 before the `succeed_n'. The `5' is the last two
3992 bytes of this `set_number_at', plus 3 bytes of
3993 the following `succeed_n'. */
3994 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
3995 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
3996 of the following `succeed_n'. */
3997 PREFIX(insert_op2) (set_number_at, laststart, 1
3998 + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b);
3999 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4001 if (upper_bound > 1)
4002 { /* More than one repetition is allowed, so
4003 append a backward jump to the `succeed_n'
4004 that starts this interval.
4006 When we've reached this during matching,
4007 we'll have matched the interval once, so
4008 jump back only `upper_bound - 1' times. */
4009 STORE_JUMP2 (jump_n, b, laststart
4010 + 2 * OFFSET_ADDRESS_SIZE + 1,
4012 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4014 /* The location we want to set is the second
4015 parameter of the `jump_n'; that is `b-2' as
4016 an absolute address. `laststart' will be
4017 the `set_number_at' we're about to insert;
4018 `laststart+3' the number to set, the source
4019 for the relative address. But we are
4020 inserting into the middle of the pattern --
4021 so everything is getting moved up by 5.
4022 Conclusion: (b - 2) - (laststart + 3) + 5,
4023 i.e., b - laststart.
4025 We insert this at the beginning of the loop
4026 so that if we fail during matching, we'll
4027 reinitialize the bounds. */
4028 PREFIX(insert_op2) (set_number_at, laststart,
4030 upper_bound - 1, b);
4031 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4038 if (!(syntax & RE_INVALID_INTERVAL_ORD))
4039 FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR);
4041 /* Match the characters as literals. */
4044 if (syntax & RE_NO_BK_BRACES)
4047 goto normal_backslash;
4051 /* There is no way to specify the before_dot and after_dot
4052 operators. rms says this is ok. --karl */
4060 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
4066 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
4072 if (syntax & RE_NO_GNU_OPS)
4075 BUF_PUSH (wordchar);
4080 if (syntax & RE_NO_GNU_OPS)
4083 BUF_PUSH (notwordchar);
4088 if (syntax & RE_NO_GNU_OPS)
4094 if (syntax & RE_NO_GNU_OPS)
4100 if (syntax & RE_NO_GNU_OPS)
4102 BUF_PUSH (wordbound);
4106 if (syntax & RE_NO_GNU_OPS)
4108 BUF_PUSH (notwordbound);
4112 if (syntax & RE_NO_GNU_OPS)
4118 if (syntax & RE_NO_GNU_OPS)
4123 case '1': case '2': case '3': case '4': case '5':
4124 case '6': case '7': case '8': case '9':
4125 if (syntax & RE_NO_BK_REFS)
4131 FREE_STACK_RETURN (REG_ESUBREG);
4133 /* Can't back reference to a subexpression if inside of it. */
4134 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
4138 BUF_PUSH_2 (duplicate, c1);
4144 if (syntax & RE_BK_PLUS_QM)
4147 goto normal_backslash;
4151 /* You might think it would be useful for \ to mean
4152 not to translate; but if we don't translate it
4153 it will never match anything. */
4161 /* Expects the character in `c'. */
4163 /* If no exactn currently being built. */
4166 /* If last exactn handle binary(or character) and
4167 new exactn handle character(or binary). */
4168 || is_exactn_bin != is_binary[p - 1 - pattern]
4171 /* If last exactn not at current position. */
4172 || pending_exact + *pending_exact + 1 != b
4174 /* We have only one byte following the exactn for the count. */
4175 || *pending_exact == (1 << BYTEWIDTH) - 1
4177 /* If followed by a repetition operator. */
4178 || *p == '*' || *p == '^'
4179 || ((syntax & RE_BK_PLUS_QM)
4180 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
4181 : (*p == '+' || *p == '?'))
4182 || ((syntax & RE_INTERVALS)
4183 && ((syntax & RE_NO_BK_BRACES)
4185 : (p[0] == '\\' && p[1] == '{'))))
4187 /* Start building a new exactn. */
4192 /* Is this exactn binary data or character? */
4193 is_exactn_bin = is_binary[p - 1 - pattern];
4195 BUF_PUSH_2 (exactn_bin, 0);
4197 BUF_PUSH_2 (exactn, 0);
4199 BUF_PUSH_2 (exactn, 0);
4201 pending_exact = b - 1;
4208 } /* while p != pend */
4211 /* Through the pattern now. */
4214 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
4216 if (!COMPILE_STACK_EMPTY)
4217 FREE_STACK_RETURN (REG_EPAREN);
4219 /* If we don't want backtracking, force success
4220 the first time we reach the end of the compiled pattern. */
4221 if (syntax & RE_NO_POSIX_BACKTRACKING)
4229 free (compile_stack.stack);
4231 /* We have succeeded; set the length of the buffer. */
4233 bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR;
4235 bufp->used = b - bufp->buffer;
4241 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4242 PREFIX(print_compiled_pattern) (bufp);
4246 #ifndef MATCH_MAY_ALLOCATE
4247 /* Initialize the failure stack to the largest possible stack. This
4248 isn't necessary unless we're trying to avoid calling alloca in
4249 the search and match routines. */
4251 int num_regs = bufp->re_nsub + 1;
4253 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4254 is strictly greater than re_max_failures, the largest possible stack
4255 is 2 * re_max_failures failure points. */
4256 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
4258 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
4261 if (! fail_stack.stack)
4263 = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size
4264 * sizeof (PREFIX(fail_stack_elt_t)));
4267 = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack,
4269 * sizeof (PREFIX(fail_stack_elt_t))));
4270 # else /* not emacs */
4271 if (! fail_stack.stack)
4273 = (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size
4274 * sizeof (PREFIX(fail_stack_elt_t)));
4277 = (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack,
4279 * sizeof (PREFIX(fail_stack_elt_t))));
4280 # endif /* not emacs */
4283 PREFIX(regex_grow_registers) (num_regs);
4285 #endif /* not MATCH_MAY_ALLOCATE */
4288 } /* regex_compile */
4290 /* Subroutines for `regex_compile'. */
4292 /* Store OP at LOC followed by two-byte integer parameter ARG. */
4293 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4296 PREFIX(store_op1) (op, loc, arg)
4301 *loc = (UCHAR_T) op;
4302 STORE_NUMBER (loc + 1, arg);
4306 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4307 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4310 PREFIX(store_op2) (op, loc, arg1, arg2)
4315 *loc = (UCHAR_T) op;
4316 STORE_NUMBER (loc + 1, arg1);
4317 STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2);
4321 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
4322 for OP followed by two-byte integer parameter ARG. */
4323 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4326 PREFIX(insert_op1) (op, loc, arg, end)
4332 register UCHAR_T *pfrom = end;
4333 register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE;
4335 while (pfrom != loc)
4338 PREFIX(store_op1) (op, loc, arg);
4342 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4343 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4346 PREFIX(insert_op2) (op, loc, arg1, arg2, end)
4352 register UCHAR_T *pfrom = end;
4353 register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE;
4355 while (pfrom != loc)
4358 PREFIX(store_op2) (op, loc, arg1, arg2);
4362 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
4363 after an alternative or a begin-subexpression. We assume there is at
4364 least one character before the ^. */
4367 PREFIX(at_begline_loc_p) (pattern, p, syntax)
4368 const CHAR_T *pattern, *p;
4369 reg_syntax_t syntax;
4371 const CHAR_T *prev = p - 2;
4372 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
4375 /* After a subexpression? */
4376 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
4377 /* After an alternative? */
4378 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
4382 /* The dual of at_begline_loc_p. This one is for $. We assume there is
4383 at least one character after the $, i.e., `P < PEND'. */
4386 PREFIX(at_endline_loc_p) (p, pend, syntax)
4387 const CHAR_T *p, *pend;
4388 reg_syntax_t syntax;
4390 const CHAR_T *next = p;
4391 boolean next_backslash = *next == '\\';
4392 const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0;
4395 /* Before a subexpression? */
4396 (syntax & RE_NO_BK_PARENS ? *next == ')'
4397 : next_backslash && next_next && *next_next == ')')
4398 /* Before an alternative? */
4399 || (syntax & RE_NO_BK_VBAR ? *next == '|'
4400 : next_backslash && next_next && *next_next == '|');
4403 #else /* not INSIDE_RECURSION */
4405 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4406 false if it's not. */
4409 group_in_compile_stack (compile_stack, regnum)
4410 compile_stack_type compile_stack;
4415 for (this_element = compile_stack.avail - 1;
4418 if (compile_stack.stack[this_element].regnum == regnum)
4423 #endif /* not INSIDE_RECURSION */
4425 #ifdef INSIDE_RECURSION
4428 /* This insert space, which size is "num", into the pattern at "loc".
4429 "end" must point the end of the allocated buffer. */
4431 insert_space (num, loc, end)
4436 register CHAR_T *pto = end;
4437 register CHAR_T *pfrom = end - num;
4439 while (pfrom >= loc)
4445 static reg_errcode_t
4446 wcs_compile_range (range_start_char, p_ptr, pend, translate, syntax, b,
4448 CHAR_T range_start_char;
4449 const CHAR_T **p_ptr, *pend;
4450 CHAR_T *char_set, *b;
4451 RE_TRANSLATE_TYPE translate;
4452 reg_syntax_t syntax;
4454 const CHAR_T *p = *p_ptr;
4455 CHAR_T range_start, range_end;
4459 uint32_t start_val, end_val;
4465 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4468 const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE,
4469 _NL_COLLATE_COLLSEQWC);
4470 const unsigned char *extra = (const unsigned char *)
4471 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4473 if (range_start_char < -1)
4475 /* range_start is a collating symbol. */
4477 /* Retreive the index and get collation sequence value. */
4478 wextra = (int32_t*)(extra + char_set[-range_start_char]);
4479 start_val = wextra[1 + *wextra];
4482 start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char));
4484 end_val = collseq_table_lookup (collseq, TRANSLATE (p[0]));
4486 /* Report an error if the range is empty and the syntax prohibits
4488 ret = ((syntax & RE_NO_EMPTY_RANGES)
4489 && (start_val > end_val))? REG_ERANGE : REG_NOERROR;
4491 /* Insert space to the end of the char_ranges. */
4492 insert_space(2, b - char_set[5] - 2, b - 1);
4493 *(b - char_set[5] - 2) = (wchar_t)start_val;
4494 *(b - char_set[5] - 1) = (wchar_t)end_val;
4495 char_set[4]++; /* ranges_index */
4500 range_start = (range_start_char >= 0)? TRANSLATE (range_start_char):
4502 range_end = TRANSLATE (p[0]);
4503 /* Report an error if the range is empty and the syntax prohibits
4505 ret = ((syntax & RE_NO_EMPTY_RANGES)
4506 && (range_start > range_end))? REG_ERANGE : REG_NOERROR;
4508 /* Insert space to the end of the char_ranges. */
4509 insert_space(2, b - char_set[5] - 2, b - 1);
4510 *(b - char_set[5] - 2) = range_start;
4511 *(b - char_set[5] - 1) = range_end;
4512 char_set[4]++; /* ranges_index */
4514 /* Have to increment the pointer into the pattern string, so the
4515 caller isn't still at the ending character. */
4521 /* Read the ending character of a range (in a bracket expression) from the
4522 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4523 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4524 Then we set the translation of all bits between the starting and
4525 ending characters (inclusive) in the compiled pattern B.
4527 Return an error code.
4529 We use these short variable names so we can use the same macros as
4530 `regex_compile' itself. */
4532 static reg_errcode_t
4533 byte_compile_range (range_start_char, p_ptr, pend, translate, syntax, b)
4534 unsigned int range_start_char;
4535 const char **p_ptr, *pend;
4536 RE_TRANSLATE_TYPE translate;
4537 reg_syntax_t syntax;
4541 const char *p = *p_ptr;
4544 const unsigned char *collseq;
4545 unsigned int start_colseq;
4546 unsigned int end_colseq;
4554 /* Have to increment the pointer into the pattern string, so the
4555 caller isn't still at the ending character. */
4558 /* Report an error if the range is empty and the syntax prohibits this. */
4559 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
4562 collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
4563 _NL_COLLATE_COLLSEQMB);
4565 start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)];
4566 end_colseq = collseq[(unsigned char) TRANSLATE (p[0])];
4567 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
4569 unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)];
4571 if (start_colseq <= this_colseq && this_colseq <= end_colseq)
4573 SET_LIST_BIT (TRANSLATE (this_char));
4578 /* Here we see why `this_char' has to be larger than an `unsigned
4579 char' -- we would otherwise go into an infinite loop, since all
4580 characters <= 0xff. */
4581 range_start_char = TRANSLATE (range_start_char);
4582 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4583 and some compilers cast it to int implicitly, so following for_loop
4584 may fall to (almost) infinite loop.
4585 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4586 To avoid this, we cast p[0] to unsigned int and truncate it. */
4587 end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1));
4589 for (this_char = range_start_char; this_char <= end_char; ++this_char)
4591 SET_LIST_BIT (TRANSLATE (this_char));
4600 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4601 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4602 characters can start a string that matches the pattern. This fastmap
4603 is used by re_search to skip quickly over impossible starting points.
4605 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4606 area as BUFP->fastmap.
4608 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4611 Returns 0 if we succeed, -2 if an internal error. */
4614 /* local function for re_compile_fastmap.
4615 truncate wchar_t character to char. */
4616 static unsigned char truncate_wchar (CHAR_T c);
4618 static unsigned char
4622 unsigned char buf[MB_CUR_MAX];
4625 memset (&state, '\0', sizeof (state));
4626 retval = wcrtomb (buf, c, &state);
4627 return retval > 0 ? buf[0] : (unsigned char) c;
4632 PREFIX(re_compile_fastmap) (bufp)
4633 struct re_pattern_buffer *bufp;
4636 #ifdef MATCH_MAY_ALLOCATE
4637 PREFIX(fail_stack_type) fail_stack;
4639 #ifndef REGEX_MALLOC
4643 register char *fastmap = bufp->fastmap;
4646 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4647 pattern to (char*) in regex_compile. */
4648 UCHAR_T *pattern = (UCHAR_T*)bufp->buffer;
4649 register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used);
4651 UCHAR_T *pattern = bufp->buffer;
4652 register UCHAR_T *pend = pattern + bufp->used;
4654 UCHAR_T *p = pattern;
4657 /* This holds the pointer to the failure stack, when
4658 it is allocated relocatably. */
4659 fail_stack_elt_t *failure_stack_ptr;
4662 /* Assume that each path through the pattern can be null until
4663 proven otherwise. We set this false at the bottom of switch
4664 statement, to which we get only if a particular path doesn't
4665 match the empty string. */
4666 boolean path_can_be_null = true;
4668 /* We aren't doing a `succeed_n' to begin with. */
4669 boolean succeed_n_p = false;
4671 assert (fastmap != NULL && p != NULL);
4674 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4675 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4676 bufp->can_be_null = 0;
4680 if (p == pend || *p == succeed)
4682 /* We have reached the (effective) end of pattern. */
4683 if (!FAIL_STACK_EMPTY ())
4685 bufp->can_be_null |= path_can_be_null;
4687 /* Reset for next path. */
4688 path_can_be_null = true;
4690 p = fail_stack.stack[--fail_stack.avail].pointer;
4698 /* We should never be about to go beyond the end of the pattern. */
4701 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4704 /* I guess the idea here is to simply not bother with a fastmap
4705 if a backreference is used, since it's too hard to figure out
4706 the fastmap for the corresponding group. Setting
4707 `can_be_null' stops `re_search_2' from using the fastmap, so
4708 that is all we do. */
4710 bufp->can_be_null = 1;
4714 /* Following are the cases which match a character. These end
4719 fastmap[truncate_wchar(p[1])] = 1;
4733 /* It is hard to distinguish fastmap from (multi byte) characters
4734 which depends on current locale. */
4739 bufp->can_be_null = 1;
4743 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4744 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
4750 /* Chars beyond end of map must be allowed. */
4751 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
4754 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4755 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
4761 for (j = 0; j < (1 << BYTEWIDTH); j++)
4762 if (SYNTAX (j) == Sword)
4768 for (j = 0; j < (1 << BYTEWIDTH); j++)
4769 if (SYNTAX (j) != Sword)
4776 int fastmap_newline = fastmap['\n'];
4778 /* `.' matches anything ... */
4779 for (j = 0; j < (1 << BYTEWIDTH); j++)
4782 /* ... except perhaps newline. */
4783 if (!(bufp->syntax & RE_DOT_NEWLINE))
4784 fastmap['\n'] = fastmap_newline;
4786 /* Return if we have already set `can_be_null'; if we have,
4787 then the fastmap is irrelevant. Something's wrong here. */
4788 else if (bufp->can_be_null)
4791 /* Otherwise, have to check alternative paths. */
4798 for (j = 0; j < (1 << BYTEWIDTH); j++)
4799 if (SYNTAX (j) == (enum syntaxcode) k)
4806 for (j = 0; j < (1 << BYTEWIDTH); j++)
4807 if (SYNTAX (j) != (enum syntaxcode) k)
4812 /* All cases after this match the empty string. These end with
4832 case push_dummy_failure:
4837 case pop_failure_jump:
4838 case maybe_pop_jump:
4841 case dummy_failure_jump:
4842 EXTRACT_NUMBER_AND_INCR (j, p);
4847 /* Jump backward implies we just went through the body of a
4848 loop and matched nothing. Opcode jumped to should be
4849 `on_failure_jump' or `succeed_n'. Just treat it like an
4850 ordinary jump. For a * loop, it has pushed its failure
4851 point already; if so, discard that as redundant. */
4852 if ((re_opcode_t) *p != on_failure_jump
4853 && (re_opcode_t) *p != succeed_n)
4857 EXTRACT_NUMBER_AND_INCR (j, p);
4860 /* If what's on the stack is where we are now, pop it. */
4861 if (!FAIL_STACK_EMPTY ()
4862 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
4868 case on_failure_jump:
4869 case on_failure_keep_string_jump:
4870 handle_on_failure_jump:
4871 EXTRACT_NUMBER_AND_INCR (j, p);
4873 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4874 end of the pattern. We don't want to push such a point,
4875 since when we restore it above, entering the switch will
4876 increment `p' past the end of the pattern. We don't need
4877 to push such a point since we obviously won't find any more
4878 fastmap entries beyond `pend'. Such a pattern can match
4879 the null string, though. */
4882 if (!PUSH_PATTERN_OP (p + j, fail_stack))
4884 RESET_FAIL_STACK ();
4889 bufp->can_be_null = 1;
4893 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
4894 succeed_n_p = false;
4901 /* Get to the number of times to succeed. */
4902 p += OFFSET_ADDRESS_SIZE;
4904 /* Increment p past the n for when k != 0. */
4905 EXTRACT_NUMBER_AND_INCR (k, p);
4908 p -= 2 * OFFSET_ADDRESS_SIZE;
4909 succeed_n_p = true; /* Spaghetti code alert. */
4910 goto handle_on_failure_jump;
4916 p += 2 * OFFSET_ADDRESS_SIZE;
4927 abort (); /* We have listed all the cases. */
4930 /* Getting here means we have found the possible starting
4931 characters for one path of the pattern -- and that the empty
4932 string does not match. We need not follow this path further.
4933 Instead, look at the next alternative (remembered on the
4934 stack), or quit if no more. The test at the top of the loop
4935 does these things. */
4936 path_can_be_null = false;
4940 /* Set `can_be_null' for the last path (also the first path, if the
4941 pattern is empty). */
4942 bufp->can_be_null |= path_can_be_null;
4945 RESET_FAIL_STACK ();
4949 #else /* not INSIDE_RECURSION */
4952 re_compile_fastmap (bufp)
4953 struct re_pattern_buffer *bufp;
4956 if (MB_CUR_MAX != 1)
4957 return wcs_re_compile_fastmap(bufp);
4960 return byte_re_compile_fastmap(bufp);
4961 } /* re_compile_fastmap */
4963 weak_alias (__re_compile_fastmap, re_compile_fastmap)
4967 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4968 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4969 this memory for recording register information. STARTS and ENDS
4970 must be allocated using the malloc library routine, and must each
4971 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4973 If NUM_REGS == 0, then subsequent matches should allocate their own
4976 Unless this function is called, the first search or match using
4977 PATTERN_BUFFER will allocate its own register data, without
4978 freeing the old data. */
4981 re_set_registers (bufp, regs, num_regs, starts, ends)
4982 struct re_pattern_buffer *bufp;
4983 struct re_registers *regs;
4985 regoff_t *starts, *ends;
4989 bufp->regs_allocated = REGS_REALLOCATE;
4990 regs->num_regs = num_regs;
4991 regs->start = starts;
4996 bufp->regs_allocated = REGS_UNALLOCATED;
4998 regs->start = regs->end = (regoff_t *) 0;
5002 weak_alias (__re_set_registers, re_set_registers)
5005 /* Searching routines. */
5007 /* Like re_search_2, below, but only one string is specified, and
5008 doesn't let you say where to stop matching. */
5011 re_search (bufp, string, size, startpos, range, regs)
5012 struct re_pattern_buffer *bufp;
5014 int size, startpos, range;
5015 struct re_registers *regs;
5017 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
5021 weak_alias (__re_search, re_search)
5025 /* Using the compiled pattern in BUFP->buffer, first tries to match the
5026 virtual concatenation of STRING1 and STRING2, starting first at index
5027 STARTPOS, then at STARTPOS + 1, and so on.
5029 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
5031 RANGE is how far to scan while trying to match. RANGE = 0 means try
5032 only at STARTPOS; in general, the last start tried is STARTPOS +
5035 In REGS, return the indices of the virtual concatenation of STRING1
5036 and STRING2 that matched the entire BUFP->buffer and its contained
5039 Do not consider matching one past the index STOP in the virtual
5040 concatenation of STRING1 and STRING2.
5042 We return either the position in the strings at which the match was
5043 found, -1 if no match, or -2 if error (such as failure
5047 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
5048 struct re_pattern_buffer *bufp;
5049 const char *string1, *string2;
5053 struct re_registers *regs;
5057 if (MB_CUR_MAX != 1)
5058 return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5062 return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5066 weak_alias (__re_search_2, re_search_2)
5069 #endif /* not INSIDE_RECURSION */
5071 #ifdef INSIDE_RECURSION
5073 #ifdef MATCH_MAY_ALLOCATE
5074 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5076 # define FREE_VAR(var) if (var) free (var); var = NULL
5080 # define MAX_ALLOCA_SIZE 2000
5082 # define FREE_WCS_BUFFERS() \
5084 if (size1 > MAX_ALLOCA_SIZE) \
5086 free (wcs_string1); \
5087 free (mbs_offset1); \
5091 FREE_VAR (wcs_string1); \
5092 FREE_VAR (mbs_offset1); \
5094 if (size2 > MAX_ALLOCA_SIZE) \
5096 free (wcs_string2); \
5097 free (mbs_offset2); \
5101 FREE_VAR (wcs_string2); \
5102 FREE_VAR (mbs_offset2); \
5110 PREFIX(re_search_2) (bufp, string1, size1, string2, size2, startpos, range,
5112 struct re_pattern_buffer *bufp;
5113 const char *string1, *string2;
5117 struct re_registers *regs;
5121 register char *fastmap = bufp->fastmap;
5122 register RE_TRANSLATE_TYPE translate = bufp->translate;
5123 int total_size = size1 + size2;
5124 int endpos = startpos + range;
5126 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5127 wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL;
5128 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5129 int wcs_size1 = 0, wcs_size2 = 0;
5130 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5131 int *mbs_offset1 = NULL, *mbs_offset2 = NULL;
5132 /* They hold whether each wchar_t is binary data or not. */
5133 char *is_binary = NULL;
5136 /* Check for out-of-range STARTPOS. */
5137 if (startpos < 0 || startpos > total_size)
5140 /* Fix up RANGE if it might eventually take us outside
5141 the virtual concatenation of STRING1 and STRING2.
5142 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5144 range = 0 - startpos;
5145 else if (endpos > total_size)
5146 range = total_size - startpos;
5148 /* If the search isn't to be a backwards one, don't waste time in a
5149 search for a pattern that must be anchored. */
5150 if (bufp->used > 0 && range > 0
5151 && ((re_opcode_t) bufp->buffer[0] == begbuf
5152 /* `begline' is like `begbuf' if it cannot match at newlines. */
5153 || ((re_opcode_t) bufp->buffer[0] == begline
5154 && !bufp->newline_anchor)))
5163 /* In a forward search for something that starts with \=.
5164 don't keep searching past point. */
5165 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
5167 range = PT - startpos;
5173 /* Update the fastmap now if not correct already. */
5174 if (fastmap && !bufp->fastmap_accurate)
5175 if (re_compile_fastmap (bufp) == -2)
5179 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5180 fill them with converted string. */
5183 if (size1 > MAX_ALLOCA_SIZE)
5185 wcs_string1 = TALLOC (size1 + 1, CHAR_T);
5186 mbs_offset1 = TALLOC (size1 + 1, int);
5187 is_binary = TALLOC (size1 + 1, char);
5191 wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T);
5192 mbs_offset1 = REGEX_TALLOC (size1 + 1, int);
5193 is_binary = REGEX_TALLOC (size1 + 1, char);
5195 if (!wcs_string1 || !mbs_offset1 || !is_binary)
5197 if (size1 > MAX_ALLOCA_SIZE)
5205 FREE_VAR (wcs_string1);
5206 FREE_VAR (mbs_offset1);
5207 FREE_VAR (is_binary);
5211 wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1,
5212 mbs_offset1, is_binary);
5213 wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */
5214 if (size1 > MAX_ALLOCA_SIZE)
5217 FREE_VAR (is_binary);
5221 if (size2 > MAX_ALLOCA_SIZE)
5223 wcs_string2 = TALLOC (size2 + 1, CHAR_T);
5224 mbs_offset2 = TALLOC (size2 + 1, int);
5225 is_binary = TALLOC (size2 + 1, char);
5229 wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T);
5230 mbs_offset2 = REGEX_TALLOC (size2 + 1, int);
5231 is_binary = REGEX_TALLOC (size2 + 1, char);
5233 if (!wcs_string2 || !mbs_offset2 || !is_binary)
5235 FREE_WCS_BUFFERS ();
5236 if (size2 > MAX_ALLOCA_SIZE)
5239 FREE_VAR (is_binary);
5242 wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2,
5243 mbs_offset2, is_binary);
5244 wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */
5245 if (size2 > MAX_ALLOCA_SIZE)
5248 FREE_VAR (is_binary);
5253 /* Loop through the string, looking for a place to start matching. */
5256 /* If a fastmap is supplied, skip quickly over characters that
5257 cannot be the start of a match. If the pattern can match the
5258 null string, however, we don't need to skip characters; we want
5259 the first null string. */
5260 if (fastmap && startpos < total_size && !bufp->can_be_null)
5262 if (range > 0) /* Searching forwards. */
5264 register const char *d;
5265 register int lim = 0;
5268 if (startpos < size1 && startpos + range >= size1)
5269 lim = range - (size1 - startpos);
5271 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
5273 /* Written out as an if-else to avoid testing `translate'
5277 && !fastmap[(unsigned char)
5278 translate[(unsigned char) *d++]])
5281 while (range > lim && !fastmap[(unsigned char) *d++])
5284 startpos += irange - range;
5286 else /* Searching backwards. */
5288 register CHAR_T c = (size1 == 0 || startpos >= size1
5289 ? string2[startpos - size1]
5290 : string1[startpos]);
5292 if (!fastmap[(unsigned char) TRANSLATE (c)])
5297 /* If can't match the null string, and that's all we have left, fail. */
5298 if (range >= 0 && startpos == total_size && fastmap
5299 && !bufp->can_be_null)
5302 FREE_WCS_BUFFERS ();
5308 val = wcs_re_match_2_internal (bufp, string1, size1, string2,
5309 size2, startpos, regs, stop,
5310 wcs_string1, wcs_size1,
5311 wcs_string2, wcs_size2,
5312 mbs_offset1, mbs_offset2);
5314 val = byte_re_match_2_internal (bufp, string1, size1, string2,
5315 size2, startpos, regs, stop);
5318 #ifndef REGEX_MALLOC
5327 FREE_WCS_BUFFERS ();
5335 FREE_WCS_BUFFERS ();
5355 FREE_WCS_BUFFERS ();
5361 /* This converts PTR, a pointer into one of the search wchar_t strings
5362 `string1' and `string2' into an multibyte string offset from the
5363 beginning of that string. We use mbs_offset to optimize.
5364 See convert_mbs_to_wcs. */
5365 # define POINTER_TO_OFFSET(ptr) \
5366 (FIRST_STRING_P (ptr) \
5367 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5368 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5371 /* This converts PTR, a pointer into one of the search strings `string1'
5372 and `string2' into an offset from the beginning of that string. */
5373 # define POINTER_TO_OFFSET(ptr) \
5374 (FIRST_STRING_P (ptr) \
5375 ? ((regoff_t) ((ptr) - string1)) \
5376 : ((regoff_t) ((ptr) - string2 + size1)))
5379 /* Macros for dealing with the split strings in re_match_2. */
5381 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5383 /* Call before fetching a character with *d. This switches over to
5384 string2 if necessary. */
5385 #define PREFETCH() \
5388 /* End of string2 => fail. */ \
5389 if (dend == end_match_2) \
5391 /* End of string1 => advance to string2. */ \
5393 dend = end_match_2; \
5396 /* Test if at very beginning or at very end of the virtual concatenation
5397 of `string1' and `string2'. If only one string, it's `string2'. */
5398 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5399 #define AT_STRINGS_END(d) ((d) == end2)
5402 /* Test if D points to a character which is word-constituent. We have
5403 two special cases to check for: if past the end of string1, look at
5404 the first character in string2; and if before the beginning of
5405 string2, look at the last character in string1. */
5407 /* Use internationalized API instead of SYNTAX. */
5408 # define WORDCHAR_P(d) \
5409 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5410 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5411 || ((d) == end1 ? *string2 \
5412 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5414 # define WORDCHAR_P(d) \
5415 (SYNTAX ((d) == end1 ? *string2 \
5416 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5420 /* Disabled due to a compiler bug -- see comment at case wordbound */
5422 /* Test if the character before D and the one at D differ with respect
5423 to being word-constituent. */
5424 #define AT_WORD_BOUNDARY(d) \
5425 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5426 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5429 /* Free everything we malloc. */
5430 #ifdef MATCH_MAY_ALLOCATE
5432 # define FREE_VARIABLES() \
5434 REGEX_FREE_STACK (fail_stack.stack); \
5435 FREE_VAR (regstart); \
5436 FREE_VAR (regend); \
5437 FREE_VAR (old_regstart); \
5438 FREE_VAR (old_regend); \
5439 FREE_VAR (best_regstart); \
5440 FREE_VAR (best_regend); \
5441 FREE_VAR (reg_info); \
5442 FREE_VAR (reg_dummy); \
5443 FREE_VAR (reg_info_dummy); \
5444 if (!cant_free_wcs_buf) \
5446 FREE_VAR (string1); \
5447 FREE_VAR (string2); \
5448 FREE_VAR (mbs_offset1); \
5449 FREE_VAR (mbs_offset2); \
5453 # define FREE_VARIABLES() \
5455 REGEX_FREE_STACK (fail_stack.stack); \
5456 FREE_VAR (regstart); \
5457 FREE_VAR (regend); \
5458 FREE_VAR (old_regstart); \
5459 FREE_VAR (old_regend); \
5460 FREE_VAR (best_regstart); \
5461 FREE_VAR (best_regend); \
5462 FREE_VAR (reg_info); \
5463 FREE_VAR (reg_dummy); \
5464 FREE_VAR (reg_info_dummy); \
5469 # define FREE_VARIABLES() \
5471 if (!cant_free_wcs_buf) \
5473 FREE_VAR (string1); \
5474 FREE_VAR (string2); \
5475 FREE_VAR (mbs_offset1); \
5476 FREE_VAR (mbs_offset2); \
5480 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5482 #endif /* not MATCH_MAY_ALLOCATE */
5484 /* These values must meet several constraints. They must not be valid
5485 register values; since we have a limit of 255 registers (because
5486 we use only one byte in the pattern for the register number), we can
5487 use numbers larger than 255. They must differ by 1, because of
5488 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5489 be larger than the value for the highest register, so we do not try
5490 to actually save any registers when none are active. */
5491 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5492 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5494 #else /* not INSIDE_RECURSION */
5495 /* Matching routines. */
5497 #ifndef emacs /* Emacs never uses this. */
5498 /* re_match is like re_match_2 except it takes only a single string. */
5501 re_match (bufp, string, size, pos, regs)
5502 struct re_pattern_buffer *bufp;
5505 struct re_registers *regs;
5509 if (MB_CUR_MAX != 1)
5510 result = wcs_re_match_2_internal (bufp, NULL, 0, string, size,
5512 NULL, 0, NULL, 0, NULL, NULL);
5515 result = byte_re_match_2_internal (bufp, NULL, 0, string, size,
5517 # ifndef REGEX_MALLOC
5525 weak_alias (__re_match, re_match)
5527 #endif /* not emacs */
5529 #endif /* not INSIDE_RECURSION */
5531 #ifdef INSIDE_RECURSION
5532 static boolean PREFIX(group_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5534 PREFIX(register_info_type) *reg_info));
5535 static boolean PREFIX(alt_match_null_string_p) _RE_ARGS ((UCHAR_T *p,
5537 PREFIX(register_info_type) *reg_info));
5538 static boolean PREFIX(common_op_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5540 PREFIX(register_info_type) *reg_info));
5541 static int PREFIX(bcmp_translate) _RE_ARGS ((const CHAR_T *s1, const CHAR_T *s2,
5542 int len, char *translate));
5543 #else /* not INSIDE_RECURSION */
5545 /* re_match_2 matches the compiled pattern in BUFP against the
5546 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5547 and SIZE2, respectively). We start matching at POS, and stop
5550 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5551 store offsets for the substring each group matched in REGS. See the
5552 documentation for exactly how many groups we fill.
5554 We return -1 if no match, -2 if an internal error (such as the
5555 failure stack overflowing). Otherwise, we return the length of the
5556 matched substring. */
5559 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5560 struct re_pattern_buffer *bufp;
5561 const char *string1, *string2;
5564 struct re_registers *regs;
5569 if (MB_CUR_MAX != 1)
5570 result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2,
5572 NULL, 0, NULL, 0, NULL, NULL);
5575 result = byte_re_match_2_internal (bufp, string1, size1, string2, size2,
5578 #ifndef REGEX_MALLOC
5586 weak_alias (__re_match_2, re_match_2)
5589 #endif /* not INSIDE_RECURSION */
5591 #ifdef INSIDE_RECURSION
5594 static int count_mbs_length PARAMS ((int *, int));
5596 /* This check the substring (from 0, to length) of the multibyte string,
5597 to which offset_buffer correspond. And count how many wchar_t_characters
5598 the substring occupy. We use offset_buffer to optimization.
5599 See convert_mbs_to_wcs. */
5602 count_mbs_length(offset_buffer, length)
5608 /* Check whether the size is valid. */
5612 if (offset_buffer == NULL)
5615 /* If there are no multibyte character, offset_buffer[i] == i.
5616 Optmize for this case. */
5617 if (offset_buffer[length] == length)
5620 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5626 int middle = (lower + upper) / 2;
5627 if (middle == lower || middle == upper)
5629 if (offset_buffer[middle] > length)
5631 else if (offset_buffer[middle] < length)
5641 /* This is a separate function so that we can force an alloca cleanup
5645 wcs_re_match_2_internal (bufp, cstring1, csize1, cstring2, csize2, pos,
5646 regs, stop, string1, size1, string2, size2,
5647 mbs_offset1, mbs_offset2)
5648 struct re_pattern_buffer *bufp;
5649 const char *cstring1, *cstring2;
5652 struct re_registers *regs;
5654 /* string1 == string2 == NULL means string1/2, size1/2 and
5655 mbs_offset1/2 need seting up in this function. */
5656 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5657 wchar_t *string1, *string2;
5658 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5660 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5661 int *mbs_offset1, *mbs_offset2;
5664 byte_re_match_2_internal (bufp, string1, size1,string2, size2, pos,
5666 struct re_pattern_buffer *bufp;
5667 const char *string1, *string2;
5670 struct re_registers *regs;
5674 /* General temporaries. */
5678 /* They hold whether each wchar_t is binary data or not. */
5679 char *is_binary = NULL;
5680 /* If true, we can't free string1/2, mbs_offset1/2. */
5681 int cant_free_wcs_buf = 1;
5684 /* Just past the end of the corresponding string. */
5685 const CHAR_T *end1, *end2;
5687 /* Pointers into string1 and string2, just past the last characters in
5688 each to consider matching. */
5689 const CHAR_T *end_match_1, *end_match_2;
5691 /* Where we are in the data, and the end of the current string. */
5692 const CHAR_T *d, *dend;
5694 /* Where we are in the pattern, and the end of the pattern. */
5696 UCHAR_T *pattern, *p;
5697 register UCHAR_T *pend;
5699 UCHAR_T *p = bufp->buffer;
5700 register UCHAR_T *pend = p + bufp->used;
5703 /* Mark the opcode just after a start_memory, so we can test for an
5704 empty subpattern when we get to the stop_memory. */
5705 UCHAR_T *just_past_start_mem = 0;
5707 /* We use this to map every character in the string. */
5708 RE_TRANSLATE_TYPE translate = bufp->translate;
5710 /* Failure point stack. Each place that can handle a failure further
5711 down the line pushes a failure point on this stack. It consists of
5712 restart, regend, and reg_info for all registers corresponding to
5713 the subexpressions we're currently inside, plus the number of such
5714 registers, and, finally, two char *'s. The first char * is where
5715 to resume scanning the pattern; the second one is where to resume
5716 scanning the strings. If the latter is zero, the failure point is
5717 a ``dummy''; if a failure happens and the failure point is a dummy,
5718 it gets discarded and the next next one is tried. */
5719 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5720 PREFIX(fail_stack_type) fail_stack;
5723 static unsigned failure_id;
5724 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5728 /* This holds the pointer to the failure stack, when
5729 it is allocated relocatably. */
5730 fail_stack_elt_t *failure_stack_ptr;
5733 /* We fill all the registers internally, independent of what we
5734 return, for use in backreferences. The number here includes
5735 an element for register zero. */
5736 size_t num_regs = bufp->re_nsub + 1;
5738 /* The currently active registers. */
5739 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
5740 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
5742 /* Information on the contents of registers. These are pointers into
5743 the input strings; they record just what was matched (on this
5744 attempt) by a subexpression part of the pattern, that is, the
5745 regnum-th regstart pointer points to where in the pattern we began
5746 matching and the regnum-th regend points to right after where we
5747 stopped matching the regnum-th subexpression. (The zeroth register
5748 keeps track of what the whole pattern matches.) */
5749 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5750 const CHAR_T **regstart, **regend;
5753 /* If a group that's operated upon by a repetition operator fails to
5754 match anything, then the register for its start will need to be
5755 restored because it will have been set to wherever in the string we
5756 are when we last see its open-group operator. Similarly for a
5758 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5759 const CHAR_T **old_regstart, **old_regend;
5762 /* The is_active field of reg_info helps us keep track of which (possibly
5763 nested) subexpressions we are currently in. The matched_something
5764 field of reg_info[reg_num] helps us tell whether or not we have
5765 matched any of the pattern so far this time through the reg_num-th
5766 subexpression. These two fields get reset each time through any
5767 loop their register is in. */
5768 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5769 PREFIX(register_info_type) *reg_info;
5772 /* The following record the register info as found in the above
5773 variables when we find a match better than any we've seen before.
5774 This happens as we backtrack through the failure points, which in
5775 turn happens only if we have not yet matched the entire string. */
5776 unsigned best_regs_set = false;
5777 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5778 const CHAR_T **best_regstart, **best_regend;
5781 /* Logically, this is `best_regend[0]'. But we don't want to have to
5782 allocate space for that if we're not allocating space for anything
5783 else (see below). Also, we never need info about register 0 for
5784 any of the other register vectors, and it seems rather a kludge to
5785 treat `best_regend' differently than the rest. So we keep track of
5786 the end of the best match so far in a separate variable. We
5787 initialize this to NULL so that when we backtrack the first time
5788 and need to test it, it's not garbage. */
5789 const CHAR_T *match_end = NULL;
5791 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5792 int set_regs_matched_done = 0;
5794 /* Used when we pop values we don't care about. */
5795 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5796 const CHAR_T **reg_dummy;
5797 PREFIX(register_info_type) *reg_info_dummy;
5801 /* Counts the total number of registers pushed. */
5802 unsigned num_regs_pushed = 0;
5805 /* Definitions for state transitions. More efficiently for gcc. */
5807 # if defined HAVE_SUBTRACT_LOCAL_LABELS && defined SHARED
5812 const void *__unbounded ptr; \
5813 offset = (p == pend \
5814 ? 0 : jmptable[SWITCH_ENUM_CAST ((re_opcode_t) *p++)]); \
5815 ptr = &&end_of_pattern + offset; \
5820 &&label_##x - &&end_of_pattern
5821 # define JUMP_TABLE_TYPE const int
5826 const void *__unbounded ptr; \
5827 ptr = (p == pend ? &&end_of_pattern \
5828 : jmptable[SWITCH_ENUM_CAST ((re_opcode_t) *p++)]); \
5834 # define JUMP_TABLE_TYPE const void *const
5836 # define CASE(x) label_##x
5837 static JUMP_TABLE_TYPE jmptable[] =
5856 REF (jump_past_alt),
5857 REF (on_failure_jump),
5858 REF (on_failure_keep_string_jump),
5859 REF (pop_failure_jump),
5860 REF (maybe_pop_jump),
5861 REF (dummy_failure_jump),
5862 REF (push_dummy_failure),
5865 REF (set_number_at),
5887 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5891 #ifdef MATCH_MAY_ALLOCATE
5892 /* Do not bother to initialize all the register variables if there are
5893 no groups in the pattern, as it takes a fair amount of time. If
5894 there are groups, we include space for register 0 (the whole
5895 pattern), even though we never use it, since it simplifies the
5896 array indexing. We should fix this. */
5899 regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5900 regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5901 old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5902 old_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5903 best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5904 best_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5905 reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5906 reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *);
5907 reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5909 if (!(regstart && regend && old_regstart && old_regend && reg_info
5910 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
5918 /* We must initialize all our variables to NULL, so that
5919 `FREE_VARIABLES' doesn't try to free them. */
5920 regstart = regend = old_regstart = old_regend = best_regstart
5921 = best_regend = reg_dummy = NULL;
5922 reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL;
5924 #endif /* MATCH_MAY_ALLOCATE */
5926 /* The starting position is bogus. */
5928 if (pos < 0 || pos > csize1 + csize2)
5930 if (pos < 0 || pos > size1 + size2)
5938 /* Allocate wchar_t array for string1 and string2 and
5939 fill them with converted string. */
5940 if (string1 == NULL && string2 == NULL)
5942 /* We need seting up buffers here. */
5944 /* We must free wcs buffers in this function. */
5945 cant_free_wcs_buf = 0;
5949 string1 = REGEX_TALLOC (csize1 + 1, CHAR_T);
5950 mbs_offset1 = REGEX_TALLOC (csize1 + 1, int);
5951 is_binary = REGEX_TALLOC (csize1 + 1, char);
5952 if (!string1 || !mbs_offset1 || !is_binary)
5955 FREE_VAR (mbs_offset1);
5956 FREE_VAR (is_binary);
5962 string2 = REGEX_TALLOC (csize2 + 1, CHAR_T);
5963 mbs_offset2 = REGEX_TALLOC (csize2 + 1, int);
5964 is_binary = REGEX_TALLOC (csize2 + 1, char);
5965 if (!string2 || !mbs_offset2 || !is_binary)
5968 FREE_VAR (mbs_offset1);
5970 FREE_VAR (mbs_offset2);
5971 FREE_VAR (is_binary);
5974 size2 = convert_mbs_to_wcs(string2, cstring2, csize2,
5975 mbs_offset2, is_binary);
5976 string2[size2] = L'\0'; /* for a sentinel */
5977 FREE_VAR (is_binary);
5981 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5982 pattern to (char*) in regex_compile. */
5983 p = pattern = (CHAR_T*)bufp->buffer;
5984 pend = (CHAR_T*)(bufp->buffer + bufp->used);
5988 /* Initialize subexpression text positions to -1 to mark ones that no
5989 start_memory/stop_memory has been seen for. Also initialize the
5990 register information struct. */
5991 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5993 regstart[mcnt] = regend[mcnt]
5994 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
5996 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
5997 IS_ACTIVE (reg_info[mcnt]) = 0;
5998 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5999 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
6002 /* We move `string1' into `string2' if the latter's empty -- but not if
6003 `string1' is null. */
6004 if (size2 == 0 && string1 != NULL)
6011 mbs_offset2 = mbs_offset1;
6017 end1 = string1 + size1;
6018 end2 = string2 + size2;
6020 /* Compute where to stop matching, within the two strings. */
6024 mcnt = count_mbs_length(mbs_offset1, stop);
6025 end_match_1 = string1 + mcnt;
6026 end_match_2 = string2;
6030 if (stop > csize1 + csize2)
6031 stop = csize1 + csize2;
6033 mcnt = count_mbs_length(mbs_offset2, stop-csize1);
6034 end_match_2 = string2 + mcnt;
6037 { /* count_mbs_length return error. */
6044 end_match_1 = string1 + stop;
6045 end_match_2 = string2;
6050 end_match_2 = string2 + stop - size1;
6054 /* `p' scans through the pattern as `d' scans through the data.
6055 `dend' is the end of the input string that `d' points within. `d'
6056 is advanced into the following input string whenever necessary, but
6057 this happens before fetching; therefore, at the beginning of the
6058 loop, `d' can be pointing at the end of a string, but it cannot
6061 if (size1 > 0 && pos <= csize1)
6063 mcnt = count_mbs_length(mbs_offset1, pos);
6069 mcnt = count_mbs_length(mbs_offset2, pos-csize1);
6075 { /* count_mbs_length return error. */
6080 if (size1 > 0 && pos <= size1)
6087 d = string2 + pos - size1;
6092 DEBUG_PRINT1 ("The compiled pattern is:\n");
6093 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
6094 DEBUG_PRINT1 ("The string to match is: `");
6095 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
6096 DEBUG_PRINT1 ("'\n");
6098 /* This loops over pattern commands. It exits by returning from the
6099 function if the match is complete, or it drops through if the match
6100 fails at this starting point in the input data. */
6104 DEBUG_PRINT2 ("\n%p: ", p);
6106 DEBUG_PRINT2 ("\n0x%x: ", p);
6118 /* End of pattern means we might have succeeded. */
6119 DEBUG_PRINT1 ("end of pattern ... ");
6121 /* If we haven't matched the entire string, and we want the
6122 longest match, try backtracking. */
6123 if (d != end_match_2)
6125 /* 1 if this match ends in the same string (string1 or string2)
6126 as the best previous match. */
6127 boolean same_str_p = (FIRST_STRING_P (match_end)
6128 == MATCHING_IN_FIRST_STRING);
6129 /* 1 if this match is the best seen so far. */
6130 boolean best_match_p;
6132 /* AIX compiler got confused when this was combined
6133 with the previous declaration. */
6135 best_match_p = d > match_end;
6137 best_match_p = !MATCHING_IN_FIRST_STRING;
6139 DEBUG_PRINT1 ("backtracking.\n");
6141 if (!FAIL_STACK_EMPTY ())
6142 { /* More failure points to try. */
6144 /* If exceeds best match so far, save it. */
6145 if (!best_regs_set || best_match_p)
6147 best_regs_set = true;
6150 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6152 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6154 best_regstart[mcnt] = regstart[mcnt];
6155 best_regend[mcnt] = regend[mcnt];
6161 /* If no failure points, don't restore garbage. And if
6162 last match is real best match, don't restore second
6164 else if (best_regs_set && !best_match_p)
6167 /* Restore best match. It may happen that `dend ==
6168 end_match_1' while the restored d is in string2.
6169 For example, the pattern `x.*y.*z' against the
6170 strings `x-' and `y-z-', if the two strings are
6171 not consecutive in memory. */
6172 DEBUG_PRINT1 ("Restoring best registers.\n");
6175 dend = ((d >= string1 && d <= end1)
6176 ? end_match_1 : end_match_2);
6178 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6180 regstart[mcnt] = best_regstart[mcnt];
6181 regend[mcnt] = best_regend[mcnt];
6184 } /* d != end_match_2 */
6187 DEBUG_PRINT1 ("Accepting match.\n");
6188 /* If caller wants register contents data back, do it. */
6189 if (regs && !bufp->no_sub)
6191 /* Have the register data arrays been allocated? */
6192 if (bufp->regs_allocated == REGS_UNALLOCATED)
6193 { /* No. So allocate them with malloc. We need one
6194 extra element beyond `num_regs' for the `-1' marker
6196 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
6197 regs->start = TALLOC (regs->num_regs, regoff_t);
6198 regs->end = TALLOC (regs->num_regs, regoff_t);
6199 if (regs->start == NULL || regs->end == NULL)
6204 bufp->regs_allocated = REGS_REALLOCATE;
6206 else if (bufp->regs_allocated == REGS_REALLOCATE)
6207 { /* Yes. If we need more elements than were already
6208 allocated, reallocate them. If we need fewer, just
6210 if (regs->num_regs < num_regs + 1)
6212 regs->num_regs = num_regs + 1;
6213 RETALLOC (regs->start, regs->num_regs, regoff_t);
6214 RETALLOC (regs->end, regs->num_regs, regoff_t);
6215 if (regs->start == NULL || regs->end == NULL)
6224 /* These braces fend off a "empty body in an else-statement"
6225 warning under GCC when assert expands to nothing. */
6226 assert (bufp->regs_allocated == REGS_FIXED);
6229 /* Convert the pointer data in `regstart' and `regend' to
6230 indices. Register zero has to be set differently,
6231 since we haven't kept track of any info for it. */
6232 if (regs->num_regs > 0)
6234 regs->start[0] = pos;
6236 if (MATCHING_IN_FIRST_STRING)
6237 regs->end[0] = (mbs_offset1 != NULL ?
6238 mbs_offset1[d-string1] : 0);
6240 regs->end[0] = csize1 + (mbs_offset2 != NULL
6241 ? mbs_offset2[d-string2] : 0);
6243 regs->end[0] = (MATCHING_IN_FIRST_STRING
6244 ? ((regoff_t) (d - string1))
6245 : ((regoff_t) (d - string2 + size1)));
6249 /* Go through the first `min (num_regs, regs->num_regs)'
6250 registers, since that is all we initialized. */
6251 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
6254 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
6255 regs->start[mcnt] = regs->end[mcnt] = -1;
6259 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
6261 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
6265 /* If the regs structure we return has more elements than
6266 were in the pattern, set the extra elements to -1. If
6267 we (re)allocated the registers, this is the case,
6268 because we always allocate enough to have at least one
6270 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
6271 regs->start[mcnt] = regs->end[mcnt] = -1;
6272 } /* regs && !bufp->no_sub */
6274 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6275 nfailure_points_pushed, nfailure_points_popped,
6276 nfailure_points_pushed - nfailure_points_popped);
6277 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
6280 if (MATCHING_IN_FIRST_STRING)
6281 mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0;
6283 mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) +
6287 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
6288 ? string1 : string2 - size1);
6291 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
6298 /* Otherwise match next pattern command. */
6299 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
6302 /* Ignore these. Used to ignore the n of succeed_n's which
6303 currently have n == 0. */
6305 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6309 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6312 /* Match the next n pattern characters exactly. The following
6313 byte in the pattern defines n, and the n bytes after that
6314 are the characters to match. */
6320 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
6322 /* This is written out as an if-else so we don't waste time
6323 testing `translate' inside the loop. */
6332 if ((UCHAR_T) translate[(unsigned char) *d++]
6338 if (*d++ != (CHAR_T) *p++)
6342 if ((UCHAR_T) translate[(unsigned char) *d++]
6354 if (*d++ != (CHAR_T) *p++) goto fail;
6358 SET_REGS_MATCHED ();
6362 /* Match any character except possibly a newline or a null. */
6364 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6368 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
6369 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
6372 SET_REGS_MATCHED ();
6373 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d);
6383 unsigned int i, char_class_length, coll_symbol_length,
6384 equiv_class_length, ranges_length, chars_length, length;
6385 CHAR_T *workp, *workp2, *charset_top;
6386 #define WORK_BUFFER_SIZE 128
6387 CHAR_T str_buf[WORK_BUFFER_SIZE];
6392 boolean not = (re_opcode_t) *(p - 1) == charset_not;
6394 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6396 c = TRANSLATE (*d); /* The character to match. */
6399 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
6401 charset_top = p - 1;
6402 char_class_length = *p++;
6403 coll_symbol_length = *p++;
6404 equiv_class_length = *p++;
6405 ranges_length = *p++;
6406 chars_length = *p++;
6407 /* p points charset[6], so the address of the next instruction
6408 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6409 where l=length of char_classes, m=length of collating_symbol,
6410 n=equivalence_class, o=length of char_range,
6411 p'=length of character. */
6413 /* Update p to indicate the next instruction. */
6414 p += char_class_length + coll_symbol_length+ equiv_class_length +
6415 2*ranges_length + chars_length;
6417 /* match with char_class? */
6418 for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE)
6421 uintptr_t alignedp = ((uintptr_t)workp
6422 + __alignof__(wctype_t) - 1)
6423 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6424 wctype = *((wctype_t*)alignedp);
6425 workp += CHAR_CLASS_SIZE;
6426 if (iswctype((wint_t)c, wctype))
6427 goto char_set_matched;
6430 /* match with collating_symbol? */
6434 const unsigned char *extra = (const unsigned char *)
6435 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
6437 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;
6441 wextra = (int32_t*)(extra + *workp++);
6442 for (i = 0; i < *wextra; ++i)
6443 if (TRANSLATE(d[i]) != wextra[1 + i])
6448 /* Update d, however d will be incremented at
6449 char_set_matched:, we decrement d here. */
6451 goto char_set_matched;
6455 else /* (nrules == 0) */
6457 /* If we can't look up collation data, we use wcscoll
6460 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;)
6462 const CHAR_T *backup_d = d, *backup_dend = dend;
6463 length = wcslen (workp);
6465 /* If wcscoll(the collating symbol, whole string) > 0,
6466 any substring of the string never match with the
6467 collating symbol. */
6468 if (wcscoll (workp, d) > 0)
6470 workp += length + 1;
6474 /* First, we compare the collating symbol with
6475 the first character of the string.
6476 If it don't match, we add the next character to
6477 the compare buffer in turn. */
6478 for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++)
6483 if (dend == end_match_2)
6489 /* add next character to the compare buffer. */
6490 str_buf[i] = TRANSLATE(*d);
6491 str_buf[i+1] = '\0';
6493 match = wcscoll (workp, str_buf);
6495 goto char_set_matched;
6498 /* (str_buf > workp) indicate (str_buf + X > workp),
6499 because for all X (str_buf + X > str_buf).
6500 So we don't need continue this loop. */
6503 /* Otherwise(str_buf < workp),
6504 (str_buf+next_character) may equals (workp).
6505 So we continue this loop. */
6510 workp += length + 1;
6513 /* match with equivalence_class? */
6517 const CHAR_T *backup_d = d, *backup_dend = dend;
6518 /* Try to match the equivalence class against
6519 those known to the collate implementation. */
6520 const int32_t *table;
6521 const int32_t *weights;
6522 const int32_t *extra;
6523 const int32_t *indirect;
6528 /* This #include defines a local function! */
6529 # include <locale/weightwc.h>
6531 table = (const int32_t *)
6532 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
6533 weights = (const wint_t *)
6534 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
6535 extra = (const wint_t *)
6536 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
6537 indirect = (const int32_t *)
6538 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
6540 /* Write 1 collating element to str_buf, and
6544 for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++)
6546 cp = (wint_t*)str_buf;
6549 if (dend == end_match_2)
6554 str_buf[i] = TRANSLATE(*(d+i));
6555 str_buf[i+1] = '\0'; /* sentinel */
6556 idx2 = findidx ((const wint_t**)&cp);
6559 /* Update d, however d will be incremented at
6560 char_set_matched:, we decrement d here. */
6561 d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1);
6564 if (dend == end_match_2)
6573 len = weights[idx2];
6575 for (workp2 = workp + equiv_class_length ; workp < workp2 ;
6578 idx = (int32_t)*workp;
6579 /* We already checked idx != 0 in regex_compile. */
6581 if (idx2 != 0 && len == weights[idx])
6584 while (cnt < len && (weights[idx + 1 + cnt]
6585 == weights[idx2 + 1 + cnt]))
6589 goto char_set_matched;
6596 else /* (nrules == 0) */
6598 /* If we can't look up collation data, we use wcscoll
6601 for (workp2 = workp + equiv_class_length ; workp < workp2 ;)
6603 const CHAR_T *backup_d = d, *backup_dend = dend;
6604 length = wcslen (workp);
6606 /* If wcscoll(the collating symbol, whole string) > 0,
6607 any substring of the string never match with the
6608 collating symbol. */
6609 if (wcscoll (workp, d) > 0)
6611 workp += length + 1;
6615 /* First, we compare the equivalence class with
6616 the first character of the string.
6617 If it don't match, we add the next character to
6618 the compare buffer in turn. */
6619 for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++)
6624 if (dend == end_match_2)
6630 /* add next character to the compare buffer. */
6631 str_buf[i] = TRANSLATE(*d);
6632 str_buf[i+1] = '\0';
6634 match = wcscoll (workp, str_buf);
6637 goto char_set_matched;
6640 /* (str_buf > workp) indicate (str_buf + X > workp),
6641 because for all X (str_buf + X > str_buf).
6642 So we don't need continue this loop. */
6645 /* Otherwise(str_buf < workp),
6646 (str_buf+next_character) may equals (workp).
6647 So we continue this loop. */
6652 workp += length + 1;
6656 /* match with char_range? */
6660 uint32_t collseqval;
6661 const char *collseq = (const char *)
6662 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
6664 collseqval = collseq_table_lookup (collseq, c);
6666 for (; workp < p - chars_length ;)
6668 uint32_t start_val, end_val;
6670 /* We already compute the collation sequence value
6671 of the characters (or collating symbols). */
6672 start_val = (uint32_t) *workp++; /* range_start */
6673 end_val = (uint32_t) *workp++; /* range_end */
6675 if (start_val <= collseqval && collseqval <= end_val)
6676 goto char_set_matched;
6682 /* We set range_start_char at str_buf[0], range_end_char
6683 at str_buf[4], and compared char at str_buf[2]. */
6688 for (; workp < p - chars_length ;)
6690 wchar_t *range_start_char, *range_end_char;
6692 /* match if (range_start_char <= c <= range_end_char). */
6694 /* If range_start(or end) < 0, we assume -range_start(end)
6695 is the offset of the collating symbol which is specified
6696 as the character of the range start(end). */
6700 range_start_char = charset_top - (*workp++);
6703 str_buf[0] = *workp++;
6704 range_start_char = str_buf;
6709 range_end_char = charset_top - (*workp++);
6712 str_buf[4] = *workp++;
6713 range_end_char = str_buf + 4;
6716 if (wcscoll (range_start_char, str_buf+2) <= 0
6717 && wcscoll (str_buf+2, range_end_char) <= 0)
6718 goto char_set_matched;
6722 /* match with char? */
6723 for (; workp < p ; workp++)
6725 goto char_set_matched;
6732 /* Cast to `unsigned' instead of `unsigned char' in case the
6733 bit list is a full 32 bytes long. */
6734 if (c < (unsigned) (*p * BYTEWIDTH)
6735 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6740 if (!not) goto fail;
6741 #undef WORK_BUFFER_SIZE
6743 SET_REGS_MATCHED ();
6749 /* The beginning of a group is represented by start_memory.
6750 The arguments are the register number in the next byte, and the
6751 number of groups inner to this one in the next. The text
6752 matched within the group is recorded (in the internal
6753 registers data structure) under the register number. */
6754 CASE (start_memory):
6755 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6756 (long int) *p, (long int) p[1]);
6758 /* Find out if this group can match the empty string. */
6759 p1 = p; /* To send to group_match_null_string_p. */
6761 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
6762 REG_MATCH_NULL_STRING_P (reg_info[*p])
6763 = PREFIX(group_match_null_string_p) (&p1, pend, reg_info);
6765 /* Save the position in the string where we were the last time
6766 we were at this open-group operator in case the group is
6767 operated upon by a repetition operator, e.g., with `(a*)*b'
6768 against `ab'; then we want to ignore where we are now in
6769 the string in case this attempt to match fails. */
6770 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6771 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
6773 DEBUG_PRINT2 (" old_regstart: %d\n",
6774 POINTER_TO_OFFSET (old_regstart[*p]));
6777 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
6779 IS_ACTIVE (reg_info[*p]) = 1;
6780 MATCHED_SOMETHING (reg_info[*p]) = 0;
6782 /* Clear this whenever we change the register activity status. */
6783 set_regs_matched_done = 0;
6785 /* This is the new highest active register. */
6786 highest_active_reg = *p;
6788 /* If nothing was active before, this is the new lowest active
6790 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6791 lowest_active_reg = *p;
6793 /* Move past the register number and inner group count. */
6795 just_past_start_mem = p;
6800 /* The stop_memory opcode represents the end of a group. Its
6801 arguments are the same as start_memory's: the register
6802 number, and the number of inner groups. */
6804 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6805 (long int) *p, (long int) p[1]);
6807 /* We need to save the string position the last time we were at
6808 this close-group operator in case the group is operated
6809 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6810 against `aba'; then we want to ignore where we are now in
6811 the string in case this attempt to match fails. */
6812 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6813 ? REG_UNSET (regend[*p]) ? d : regend[*p]
6815 DEBUG_PRINT2 (" old_regend: %d\n",
6816 POINTER_TO_OFFSET (old_regend[*p]));
6819 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
6821 /* This register isn't active anymore. */
6822 IS_ACTIVE (reg_info[*p]) = 0;
6824 /* Clear this whenever we change the register activity status. */
6825 set_regs_matched_done = 0;
6827 /* If this was the only register active, nothing is active
6829 if (lowest_active_reg == highest_active_reg)
6831 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6832 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6835 { /* We must scan for the new highest active register, since
6836 it isn't necessarily one less than now: consider
6837 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6838 new highest active register is 1. */
6840 while (r > 0 && !IS_ACTIVE (reg_info[r]))
6843 /* If we end up at register zero, that means that we saved
6844 the registers as the result of an `on_failure_jump', not
6845 a `start_memory', and we jumped to past the innermost
6846 `stop_memory'. For example, in ((.)*) we save
6847 registers 1 and 2 as a result of the *, but when we pop
6848 back to the second ), we are at the stop_memory 1.
6849 Thus, nothing is active. */
6852 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6853 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6856 highest_active_reg = r;
6859 /* If just failed to match something this time around with a
6860 group that's operated on by a repetition operator, try to
6861 force exit from the ``loop'', and restore the register
6862 information for this group that we had before trying this
6864 if ((!MATCHED_SOMETHING (reg_info[*p])
6865 || just_past_start_mem == p - 1)
6868 boolean is_a_jump_n = false;
6872 switch ((re_opcode_t) *p1++)
6876 case pop_failure_jump:
6877 case maybe_pop_jump:
6879 case dummy_failure_jump:
6880 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6882 p1 += OFFSET_ADDRESS_SIZE;
6890 /* If the next operation is a jump backwards in the pattern
6891 to an on_failure_jump right before the start_memory
6892 corresponding to this stop_memory, exit from the loop
6893 by forcing a failure after pushing on the stack the
6894 on_failure_jump's jump in the pattern, and d. */
6895 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
6896 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory
6897 && p1[2+OFFSET_ADDRESS_SIZE] == *p)
6899 /* If this group ever matched anything, then restore
6900 what its registers were before trying this last
6901 failed match, e.g., with `(a*)*b' against `ab' for
6902 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6903 against `aba' for regend[3].
6905 Also restore the registers for inner groups for,
6906 e.g., `((a*)(b*))*' against `aba' (register 3 would
6907 otherwise get trashed). */
6909 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
6913 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
6915 /* Restore this and inner groups' (if any) registers. */
6916 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
6919 regstart[r] = old_regstart[r];
6921 /* xx why this test? */
6922 if (old_regend[r] >= regstart[r])
6923 regend[r] = old_regend[r];
6927 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6928 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
6934 /* Move past the register number and the inner group count. */
6939 /* \<digit> has been turned into a `duplicate' command which is
6940 followed by the numeric value of <digit> as the register number. */
6943 register const CHAR_T *d2, *dend2;
6944 int regno = *p++; /* Get which register to match against. */
6945 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
6947 /* Can't back reference a group which we've never matched. */
6948 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
6951 /* Where in input to try to start matching. */
6952 d2 = regstart[regno];
6954 /* Where to stop matching; if both the place to start and
6955 the place to stop matching are in the same string, then
6956 set to the place to stop, otherwise, for now have to use
6957 the end of the first string. */
6959 dend2 = ((FIRST_STRING_P (regstart[regno])
6960 == FIRST_STRING_P (regend[regno]))
6961 ? regend[regno] : end_match_1);
6964 /* If necessary, advance to next segment in register
6968 if (dend2 == end_match_2) break;
6969 if (dend2 == regend[regno]) break;
6971 /* End of string1 => advance to string2. */
6973 dend2 = regend[regno];
6975 /* At end of register contents => success */
6976 if (d2 == dend2) break;
6978 /* If necessary, advance to next segment in data. */
6981 /* How many characters left in this segment to match. */
6984 /* Want how many consecutive characters we can match in
6985 one shot, so, if necessary, adjust the count. */
6986 if (mcnt > dend2 - d2)
6989 /* Compare that many; failure if mismatch, else move
6992 ? PREFIX(bcmp_translate) (d, d2, mcnt, translate)
6993 : memcmp (d, d2, mcnt*sizeof(UCHAR_T)))
6995 d += mcnt, d2 += mcnt;
6997 /* Do this because we've match some characters. */
6998 SET_REGS_MATCHED ();
7004 /* begline matches the empty string at the beginning of the string
7005 (unless `not_bol' is set in `bufp'), and, if
7006 `newline_anchor' is set, after newlines. */
7008 DEBUG_PRINT1 ("EXECUTING begline.\n");
7010 if (AT_STRINGS_BEG (d))
7017 else if (d[-1] == '\n' && bufp->newline_anchor)
7021 /* In all other cases, we fail. */
7025 /* endline is the dual of begline. */
7027 DEBUG_PRINT1 ("EXECUTING endline.\n");
7029 if (AT_STRINGS_END (d))
7037 /* We have to ``prefetch'' the next character. */
7038 else if ((d == end1 ? *string2 : *d) == '\n'
7039 && bufp->newline_anchor)
7046 /* Match at the very beginning of the data. */
7048 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
7049 if (AT_STRINGS_BEG (d))
7056 /* Match at the very end of the data. */
7058 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
7059 if (AT_STRINGS_END (d))
7066 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
7067 pushes NULL as the value for the string on the stack. Then
7068 `pop_failure_point' will keep the current value for the
7069 string, instead of restoring it. To see why, consider
7070 matching `foo\nbar' against `.*\n'. The .* matches the foo;
7071 then the . fails against the \n. But the next thing we want
7072 to do is match the \n against the \n; if we restored the
7073 string value, we would be back at the foo.
7075 Because this is used only in specific cases, we don't need to
7076 check all the things that `on_failure_jump' does, to make
7077 sure the right things get saved on the stack. Hence we don't
7078 share its code. The only reason to push anything on the
7079 stack at all is that otherwise we would have to change
7080 `anychar's code to do something besides goto fail in this
7081 case; that seems worse than this. */
7082 CASE (on_failure_keep_string_jump):
7083 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
7085 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7087 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
7089 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
7092 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
7096 /* Uses of on_failure_jump:
7098 Each alternative starts with an on_failure_jump that points
7099 to the beginning of the next alternative. Each alternative
7100 except the last ends with a jump that in effect jumps past
7101 the rest of the alternatives. (They really jump to the
7102 ending jump of the following alternative, because tensioning
7103 these jumps is a hassle.)
7105 Repeats start with an on_failure_jump that points past both
7106 the repetition text and either the following jump or
7107 pop_failure_jump back to this on_failure_jump. */
7108 CASE (on_failure_jump):
7110 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
7112 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7114 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
7116 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
7119 /* If this on_failure_jump comes right before a group (i.e.,
7120 the original * applied to a group), save the information
7121 for that group and all inner ones, so that if we fail back
7122 to this point, the group's information will be correct.
7123 For example, in \(a*\)*\1, we need the preceding group,
7124 and in \(zz\(a*\)b*\)\2, we need the inner group. */
7126 /* We can't use `p' to check ahead because we push
7127 a failure point to `p + mcnt' after we do this. */
7130 /* We need to skip no_op's before we look for the
7131 start_memory in case this on_failure_jump is happening as
7132 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7134 while (p1 < pend && (re_opcode_t) *p1 == no_op)
7137 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
7139 /* We have a new highest active register now. This will
7140 get reset at the start_memory we are about to get to,
7141 but we will have saved all the registers relevant to
7142 this repetition op, as described above. */
7143 highest_active_reg = *(p1 + 1) + *(p1 + 2);
7144 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
7145 lowest_active_reg = *(p1 + 1);
7148 DEBUG_PRINT1 (":\n");
7149 PUSH_FAILURE_POINT (p + mcnt, d, -2);
7153 /* A smart repeat ends with `maybe_pop_jump'.
7154 We change it to either `pop_failure_jump' or `jump'. */
7155 CASE (maybe_pop_jump):
7156 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7157 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
7159 register UCHAR_T *p2 = p;
7161 /* Compare the beginning of the repeat with what in the
7162 pattern follows its end. If we can establish that there
7163 is nothing that they would both match, i.e., that we
7164 would have to backtrack because of (as in, e.g., `a*a')
7165 then we can change to pop_failure_jump, because we'll
7166 never have to backtrack.
7168 This is not true in the case of alternatives: in
7169 `(a|ab)*' we do need to backtrack to the `ab' alternative
7170 (e.g., if the string was `ab'). But instead of trying to
7171 detect that here, the alternative has put on a dummy
7172 failure point which is what we will end up popping. */
7174 /* Skip over open/close-group commands.
7175 If what follows this loop is a ...+ construct,
7176 look at what begins its body, since we will have to
7177 match at least one of that. */
7181 && ((re_opcode_t) *p2 == stop_memory
7182 || (re_opcode_t) *p2 == start_memory))
7184 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend
7185 && (re_opcode_t) *p2 == dummy_failure_jump)
7186 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE;
7192 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7193 to the `maybe_finalize_jump' of this case. Examine what
7196 /* If we're at the end of the pattern, we can change. */
7199 /* Consider what happens when matching ":\(.*\)"
7200 against ":/". I don't really understand this code
7202 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7205 (" End of pattern: change to `pop_failure_jump'.\n");
7208 else if ((re_opcode_t) *p2 == exactn
7210 || (re_opcode_t) *p2 == exactn_bin
7212 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
7215 = *p2 == (UCHAR_T) endline ? '\n' : p2[2];
7217 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn
7219 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin
7221 ) && p1[3+OFFSET_ADDRESS_SIZE] != c)
7223 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7226 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7228 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]);
7230 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7232 (char) p1[3+OFFSET_ADDRESS_SIZE]);
7237 else if ((re_opcode_t) p1[3] == charset
7238 || (re_opcode_t) p1[3] == charset_not)
7240 int not = (re_opcode_t) p1[3] == charset_not;
7242 if (c < (unsigned) (p1[4] * BYTEWIDTH)
7243 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
7246 /* `not' is equal to 1 if c would match, which means
7247 that we can't change to pop_failure_jump. */
7250 p[-3] = (unsigned char) pop_failure_jump;
7251 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7254 #endif /* not WCHAR */
7257 else if ((re_opcode_t) *p2 == charset)
7259 /* We win if the first character of the loop is not part
7261 if ((re_opcode_t) p1[3] == exactn
7262 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
7263 && (p2[2 + p1[5] / BYTEWIDTH]
7264 & (1 << (p1[5] % BYTEWIDTH)))))
7266 p[-3] = (unsigned char) pop_failure_jump;
7267 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7270 else if ((re_opcode_t) p1[3] == charset_not)
7273 /* We win if the charset_not inside the loop
7274 lists every character listed in the charset after. */
7275 for (idx = 0; idx < (int) p2[1]; idx++)
7276 if (! (p2[2 + idx] == 0
7277 || (idx < (int) p1[4]
7278 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
7283 p[-3] = (unsigned char) pop_failure_jump;
7284 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7287 else if ((re_opcode_t) p1[3] == charset)
7290 /* We win if the charset inside the loop
7291 has no overlap with the one after the loop. */
7293 idx < (int) p2[1] && idx < (int) p1[4];
7295 if ((p2[2 + idx] & p1[5 + idx]) != 0)
7298 if (idx == p2[1] || idx == p1[4])
7300 p[-3] = (unsigned char) pop_failure_jump;
7301 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7305 #endif /* not WCHAR */
7307 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */
7308 if ((re_opcode_t) p[-1] != pop_failure_jump)
7310 p[-1] = (UCHAR_T) jump;
7311 DEBUG_PRINT1 (" Match => jump.\n");
7312 goto unconditional_jump;
7314 /* Note fall through. */
7317 /* The end of a simple repeat has a pop_failure_jump back to
7318 its matching on_failure_jump, where the latter will push a
7319 failure point. The pop_failure_jump takes off failure
7320 points put on by this pop_failure_jump's matching
7321 on_failure_jump; we got through the pattern to here from the
7322 matching on_failure_jump, so didn't fail. */
7323 CASE (pop_failure_jump):
7325 /* We need to pass separate storage for the lowest and
7326 highest registers, even though we don't care about the
7327 actual values. Otherwise, we will restore only one
7328 register from the stack, since lowest will == highest in
7329 `pop_failure_point'. */
7330 active_reg_t dummy_low_reg, dummy_high_reg;
7331 UCHAR_T *pdummy = NULL;
7332 const CHAR_T *sdummy = NULL;
7334 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7335 POP_FAILURE_POINT (sdummy, pdummy,
7336 dummy_low_reg, dummy_high_reg,
7337 reg_dummy, reg_dummy, reg_info_dummy);
7339 /* Note fall through. */
7343 DEBUG_PRINT2 ("\n%p: ", p);
7345 DEBUG_PRINT2 ("\n0x%x: ", p);
7347 /* Note fall through. */
7349 /* Unconditionally jump (without popping any failure points). */
7351 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
7352 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
7353 p += mcnt; /* Do the jump. */
7355 DEBUG_PRINT2 ("(to %p).\n", p);
7357 DEBUG_PRINT2 ("(to 0x%x).\n", p);
7362 /* We need this opcode so we can detect where alternatives end
7363 in `group_match_null_string_p' et al. */
7364 CASE (jump_past_alt):
7365 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7366 goto unconditional_jump;
7369 /* Normally, the on_failure_jump pushes a failure point, which
7370 then gets popped at pop_failure_jump. We will end up at
7371 pop_failure_jump, also, and with a pattern of, say, `a+', we
7372 are skipping over the on_failure_jump, so we have to push
7373 something meaningless for pop_failure_jump to pop. */
7374 CASE (dummy_failure_jump):
7375 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7376 /* It doesn't matter what we push for the string here. What
7377 the code at `fail' tests is the value for the pattern. */
7378 PUSH_FAILURE_POINT (NULL, NULL, -2);
7379 goto unconditional_jump;
7382 /* At the end of an alternative, we need to push a dummy failure
7383 point in case we are followed by a `pop_failure_jump', because
7384 we don't want the failure point for the alternative to be
7385 popped. For example, matching `(a|ab)*' against `aab'
7386 requires that we match the `ab' alternative. */
7387 CASE (push_dummy_failure):
7388 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7389 /* See comments just above at `dummy_failure_jump' about the
7391 PUSH_FAILURE_POINT (NULL, NULL, -2);
7394 /* Have to succeed matching what follows at least n times.
7395 After that, handle like `on_failure_jump'. */
7397 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7398 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
7401 /* Originally, this is how many times we HAVE to succeed. */
7405 p += OFFSET_ADDRESS_SIZE;
7406 STORE_NUMBER_AND_INCR (p, mcnt);
7408 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE
7411 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE
7418 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7419 p + OFFSET_ADDRESS_SIZE);
7421 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7422 p + OFFSET_ADDRESS_SIZE);
7426 p[1] = (UCHAR_T) no_op;
7428 p[2] = (UCHAR_T) no_op;
7429 p[3] = (UCHAR_T) no_op;
7436 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7437 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
7439 /* Originally, this is how many times we CAN jump. */
7443 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt);
7446 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE,
7449 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE,
7452 goto unconditional_jump;
7454 /* If don't have to jump any more, skip over the rest of command. */
7456 p += 2 * OFFSET_ADDRESS_SIZE;
7459 CASE (set_number_at):
7461 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7463 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7465 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7467 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
7469 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
7471 STORE_NUMBER (p1, mcnt);
7476 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7477 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7478 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7479 macro and introducing temporary variables works around the bug. */
7482 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7483 if (AT_WORD_BOUNDARY (d))
7489 CASE (notwordbound):
7490 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7491 if (AT_WORD_BOUNDARY (d))
7497 boolean prevchar, thischar;
7499 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7500 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7505 prevchar = WORDCHAR_P (d - 1);
7506 thischar = WORDCHAR_P (d);
7507 if (prevchar != thischar)
7514 CASE (notwordbound):
7516 boolean prevchar, thischar;
7518 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7519 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7522 prevchar = WORDCHAR_P (d - 1);
7523 thischar = WORDCHAR_P (d);
7524 if (prevchar != thischar)
7531 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7532 if (!AT_STRINGS_END (d) && WORDCHAR_P (d)
7533 && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
7540 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7541 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
7542 && (AT_STRINGS_END (d) || !WORDCHAR_P (d)))
7550 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7551 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
7556 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7557 if (PTR_CHAR_POS ((unsigned char *) d) != point)
7562 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7563 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
7568 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
7573 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7577 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7579 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
7581 SET_REGS_MATCHED ();
7584 CASE (notsyntaxspec):
7585 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
7587 goto matchnotsyntax;
7590 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7594 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7596 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
7598 SET_REGS_MATCHED ();
7601 #else /* not emacs */
7603 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7605 if (!WORDCHAR_P (d))
7607 SET_REGS_MATCHED ();
7612 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7616 SET_REGS_MATCHED ();
7619 #endif /* not emacs */
7625 continue; /* Successfully executed one pattern command; keep going. */
7629 /* We goto here if a matching operation fails. */
7631 if (!FAIL_STACK_EMPTY ())
7632 { /* A restart point is known. Restore to that state. */
7633 DEBUG_PRINT1 ("\nFAIL:\n");
7634 POP_FAILURE_POINT (d, p,
7635 lowest_active_reg, highest_active_reg,
7636 regstart, regend, reg_info);
7638 /* If this failure point is a dummy, try the next one. */
7642 /* If we failed to the end of the pattern, don't examine *p. */
7646 boolean is_a_jump_n = false;
7648 /* If failed to a backwards jump that's part of a repetition
7649 loop, need to pop this failure point and use the next one. */
7650 switch ((re_opcode_t) *p)
7654 case maybe_pop_jump:
7655 case pop_failure_jump:
7658 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7661 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
7663 && (re_opcode_t) *p1 == on_failure_jump))
7671 if (d >= string1 && d <= end1)
7675 break; /* Matching at this starting point really fails. */
7679 goto restore_best_regs;
7683 return -1; /* Failure to match. */
7686 /* Subroutine definitions for re_match_2. */
7689 /* We are passed P pointing to a register number after a start_memory.
7691 Return true if the pattern up to the corresponding stop_memory can
7692 match the empty string, and false otherwise.
7694 If we find the matching stop_memory, sets P to point to one past its number.
7695 Otherwise, sets P to an undefined byte less than or equal to END.
7697 We don't handle duplicates properly (yet). */
7700 PREFIX(group_match_null_string_p) (p, end, reg_info)
7702 PREFIX(register_info_type) *reg_info;
7705 /* Point to after the args to the start_memory. */
7706 UCHAR_T *p1 = *p + 2;
7710 /* Skip over opcodes that can match nothing, and return true or
7711 false, as appropriate, when we get to one that can't, or to the
7712 matching stop_memory. */
7714 switch ((re_opcode_t) *p1)
7716 /* Could be either a loop or a series of alternatives. */
7717 case on_failure_jump:
7719 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7721 /* If the next operation is not a jump backwards in the
7726 /* Go through the on_failure_jumps of the alternatives,
7727 seeing if any of the alternatives cannot match nothing.
7728 The last alternative starts with only a jump,
7729 whereas the rest start with on_failure_jump and end
7730 with a jump, e.g., here is the pattern for `a|b|c':
7732 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7733 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7736 So, we have to first go through the first (n-1)
7737 alternatives and then deal with the last one separately. */
7740 /* Deal with the first (n-1) alternatives, which start
7741 with an on_failure_jump (see above) that jumps to right
7742 past a jump_past_alt. */
7744 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] ==
7747 /* `mcnt' holds how many bytes long the alternative
7748 is, including the ending `jump_past_alt' and
7751 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt -
7752 (1 + OFFSET_ADDRESS_SIZE),
7756 /* Move to right after this alternative, including the
7760 /* Break if it's the beginning of an n-th alternative
7761 that doesn't begin with an on_failure_jump. */
7762 if ((re_opcode_t) *p1 != on_failure_jump)
7765 /* Still have to check that it's not an n-th
7766 alternative that starts with an on_failure_jump. */
7768 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7769 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] !=
7772 /* Get to the beginning of the n-th alternative. */
7773 p1 -= 1 + OFFSET_ADDRESS_SIZE;
7778 /* Deal with the last alternative: go back and get number
7779 of the `jump_past_alt' just before it. `mcnt' contains
7780 the length of the alternative. */
7781 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE);
7783 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info))
7786 p1 += mcnt; /* Get past the n-th alternative. */
7792 assert (p1[1] == **p);
7798 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7801 } /* while p1 < end */
7804 } /* group_match_null_string_p */
7807 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7808 It expects P to be the first byte of a single alternative and END one
7809 byte past the last. The alternative can contain groups. */
7812 PREFIX(alt_match_null_string_p) (p, end, reg_info)
7814 PREFIX(register_info_type) *reg_info;
7821 /* Skip over opcodes that can match nothing, and break when we get
7822 to one that can't. */
7824 switch ((re_opcode_t) *p1)
7827 case on_failure_jump:
7829 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7834 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7837 } /* while p1 < end */
7840 } /* alt_match_null_string_p */
7843 /* Deals with the ops common to group_match_null_string_p and
7844 alt_match_null_string_p.
7846 Sets P to one after the op and its arguments, if any. */
7849 PREFIX(common_op_match_null_string_p) (p, end, reg_info)
7851 PREFIX(register_info_type) *reg_info;
7858 switch ((re_opcode_t) *p1++)
7878 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
7879 ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info);
7881 /* Have to set this here in case we're checking a group which
7882 contains a group and a back reference to it. */
7884 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
7885 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
7891 /* If this is an optimized succeed_n for zero times, make the jump. */
7893 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7901 /* Get to the number of times to succeed. */
7902 p1 += OFFSET_ADDRESS_SIZE;
7903 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7907 p1 -= 2 * OFFSET_ADDRESS_SIZE;
7908 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7916 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
7921 p1 += 2 * OFFSET_ADDRESS_SIZE;
7924 /* All other opcodes mean we cannot match the empty string. */
7930 } /* common_op_match_null_string_p */
7933 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7934 bytes; nonzero otherwise. */
7937 PREFIX(bcmp_translate) (s1, s2, len, translate)
7938 const CHAR_T *s1, *s2;
7940 RE_TRANSLATE_TYPE translate;
7942 register const UCHAR_T *p1 = (const UCHAR_T *) s1;
7943 register const UCHAR_T *p2 = (const UCHAR_T *) s2;
7947 if (((*p1<=0xff)?translate[*p1++]:*p1++)
7948 != ((*p2<=0xff)?translate[*p2++]:*p2++))
7951 if (translate[*p1++] != translate[*p2++]) return 1;
7959 #else /* not INSIDE_RECURSION */
7961 /* Entry points for GNU code. */
7963 /* re_compile_pattern is the GNU regular expression compiler: it
7964 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7965 Returns 0 if the pattern was valid, otherwise an error string.
7967 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7968 are set in BUFP on entry.
7970 We call regex_compile to do the actual compilation. */
7973 re_compile_pattern (pattern, length, bufp)
7974 const char *pattern;
7976 struct re_pattern_buffer *bufp;
7980 /* GNU code is written to assume at least RE_NREGS registers will be set
7981 (and at least one extra will be -1). */
7982 bufp->regs_allocated = REGS_UNALLOCATED;
7984 /* And GNU code determines whether or not to get register information
7985 by passing null for the REGS argument to re_match, etc., not by
7989 /* Match anchors at newline. */
7990 bufp->newline_anchor = 1;
7993 if (MB_CUR_MAX != 1)
7994 ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp);
7997 ret = byte_regex_compile (pattern, length, re_syntax_options, bufp);
8001 return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
8004 weak_alias (__re_compile_pattern, re_compile_pattern)
8007 /* Entry points compatible with 4.2 BSD regex library. We don't define
8008 them unless specifically requested. */
8010 #if defined _REGEX_RE_COMP || defined _LIBC
8012 /* BSD has one and only one pattern buffer. */
8013 static struct re_pattern_buffer re_comp_buf;
8017 /* Make these definitions weak in libc, so POSIX programs can redefine
8018 these names if they don't use our functions, and still use
8019 regcomp/regexec below without link errors. */
8029 if (!re_comp_buf.buffer)
8030 return gettext ("No previous regular expression");
8034 if (!re_comp_buf.buffer)
8036 re_comp_buf.buffer = (unsigned char *) malloc (200);
8037 if (re_comp_buf.buffer == NULL)
8038 return (char *) gettext (re_error_msgid
8039 + re_error_msgid_idx[(int) REG_ESPACE]);
8040 re_comp_buf.allocated = 200;
8042 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
8043 if (re_comp_buf.fastmap == NULL)
8044 return (char *) gettext (re_error_msgid
8045 + re_error_msgid_idx[(int) REG_ESPACE]);
8048 /* Since `re_exec' always passes NULL for the `regs' argument, we
8049 don't need to initialize the pattern buffer fields which affect it. */
8051 /* Match anchors at newlines. */
8052 re_comp_buf.newline_anchor = 1;
8055 if (MB_CUR_MAX != 1)
8056 ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
8059 ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
8064 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
8065 return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
8076 const int len = strlen (s);
8078 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
8081 #endif /* _REGEX_RE_COMP */
8083 /* POSIX.2 functions. Don't define these for Emacs. */
8087 /* regcomp takes a regular expression as a string and compiles it.
8089 PREG is a regex_t *. We do not expect any fields to be initialized,
8090 since POSIX says we shouldn't. Thus, we set
8092 `buffer' to the compiled pattern;
8093 `used' to the length of the compiled pattern;
8094 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
8095 REG_EXTENDED bit in CFLAGS is set; otherwise, to
8096 RE_SYNTAX_POSIX_BASIC;
8097 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
8098 `fastmap' to an allocated space for the fastmap;
8099 `fastmap_accurate' to zero;
8100 `re_nsub' to the number of subexpressions in PATTERN.
8102 PATTERN is the address of the pattern string.
8104 CFLAGS is a series of bits which affect compilation.
8106 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
8107 use POSIX basic syntax.
8109 If REG_NEWLINE is set, then . and [^...] don't match newline.
8110 Also, regexec will try a match beginning after every newline.
8112 If REG_ICASE is set, then we considers upper- and lowercase
8113 versions of letters to be equivalent when matching.
8115 If REG_NOSUB is set, then when PREG is passed to regexec, that
8116 routine will report only success or failure, and nothing about the
8119 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
8120 the return codes and their meanings.) */
8123 regcomp (preg, pattern, cflags)
8125 const char *pattern;
8130 = (cflags & REG_EXTENDED) ?
8131 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
8133 /* regex_compile will allocate the space for the compiled pattern. */
8135 preg->allocated = 0;
8138 /* Try to allocate space for the fastmap. */
8139 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
8141 if (cflags & REG_ICASE)
8146 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
8147 * sizeof (*(RE_TRANSLATE_TYPE)0));
8148 if (preg->translate == NULL)
8149 return (int) REG_ESPACE;
8151 /* Map uppercase characters to corresponding lowercase ones. */
8152 for (i = 0; i < CHAR_SET_SIZE; i++)
8153 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
8156 preg->translate = NULL;
8158 /* If REG_NEWLINE is set, newlines are treated differently. */
8159 if (cflags & REG_NEWLINE)
8160 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
8161 syntax &= ~RE_DOT_NEWLINE;
8162 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
8163 /* It also changes the matching behavior. */
8164 preg->newline_anchor = 1;
8167 preg->newline_anchor = 0;
8169 preg->no_sub = !!(cflags & REG_NOSUB);
8171 /* POSIX says a null character in the pattern terminates it, so we
8172 can use strlen here in compiling the pattern. */
8174 if (MB_CUR_MAX != 1)
8175 ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg);
8178 ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg);
8180 /* POSIX doesn't distinguish between an unmatched open-group and an
8181 unmatched close-group: both are REG_EPAREN. */
8182 if (ret == REG_ERPAREN) ret = REG_EPAREN;
8184 if (ret == REG_NOERROR && preg->fastmap)
8186 /* Compute the fastmap now, since regexec cannot modify the pattern
8188 if (re_compile_fastmap (preg) == -2)
8190 /* Some error occurred while computing the fastmap, just forget
8192 free (preg->fastmap);
8193 preg->fastmap = NULL;
8200 weak_alias (__regcomp, regcomp)
8204 /* regexec searches for a given pattern, specified by PREG, in the
8207 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8208 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8209 least NMATCH elements, and we set them to the offsets of the
8210 corresponding matched substrings.
8212 EFLAGS specifies `execution flags' which affect matching: if
8213 REG_NOTBOL is set, then ^ does not match at the beginning of the
8214 string; if REG_NOTEOL is set, then $ does not match at the end.
8216 We return 0 if we find a match and REG_NOMATCH if not. */
8219 regexec (preg, string, nmatch, pmatch, eflags)
8220 const regex_t *preg;
8223 regmatch_t pmatch[];
8227 struct re_registers regs;
8228 regex_t private_preg;
8229 int len = strlen (string);
8230 boolean want_reg_info = !preg->no_sub && nmatch > 0;
8232 private_preg = *preg;
8234 private_preg.not_bol = !!(eflags & REG_NOTBOL);
8235 private_preg.not_eol = !!(eflags & REG_NOTEOL);
8237 /* The user has told us exactly how many registers to return
8238 information about, via `nmatch'. We have to pass that on to the
8239 matching routines. */
8240 private_preg.regs_allocated = REGS_FIXED;
8244 regs.num_regs = nmatch;
8245 regs.start = TALLOC (nmatch * 2, regoff_t);
8246 if (regs.start == NULL)
8247 return (int) REG_NOMATCH;
8248 regs.end = regs.start + nmatch;
8251 /* Perform the searching operation. */
8252 ret = re_search (&private_preg, string, len,
8253 /* start: */ 0, /* range: */ len,
8254 want_reg_info ? ®s : (struct re_registers *) 0);
8256 /* Copy the register information to the POSIX structure. */
8263 for (r = 0; r < nmatch; r++)
8265 pmatch[r].rm_so = regs.start[r];
8266 pmatch[r].rm_eo = regs.end[r];
8270 /* If we needed the temporary register info, free the space now. */
8274 /* We want zero return to mean success, unlike `re_search'. */
8275 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
8278 weak_alias (__regexec, regexec)
8282 /* Returns a message corresponding to an error code, ERRCODE, returned
8283 from either regcomp or regexec. We don't use PREG here. */
8286 regerror (errcode, preg, errbuf, errbuf_size)
8288 const regex_t *preg;
8296 || errcode >= (int) (sizeof (re_error_msgid_idx)
8297 / sizeof (re_error_msgid_idx[0])))
8298 /* Only error codes returned by the rest of the code should be passed
8299 to this routine. If we are given anything else, or if other regex
8300 code generates an invalid error code, then the program has a bug.
8301 Dump core so we can fix it. */
8304 msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
8306 msg_size = strlen (msg) + 1; /* Includes the null. */
8308 if (errbuf_size != 0)
8310 if (msg_size > errbuf_size)
8312 #if defined HAVE_MEMPCPY || defined _LIBC
8313 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
8315 memcpy (errbuf, msg, errbuf_size - 1);
8316 errbuf[errbuf_size - 1] = 0;
8320 memcpy (errbuf, msg, msg_size);
8326 weak_alias (__regerror, regerror)
8330 /* Free dynamically allocated space used by PREG. */
8336 if (preg->buffer != NULL)
8337 free (preg->buffer);
8338 preg->buffer = NULL;
8340 preg->allocated = 0;
8343 if (preg->fastmap != NULL)
8344 free (preg->fastmap);
8345 preg->fastmap = NULL;
8346 preg->fastmap_accurate = 0;
8348 if (preg->translate != NULL)
8349 free (preg->translate);
8350 preg->translate = NULL;
8353 weak_alias (__regfree, regfree)
8356 #endif /* not emacs */
8358 #endif /* not INSIDE_RECURSION */
8362 #undef STORE_NUMBER_AND_INCR
8363 #undef EXTRACT_NUMBER
8364 #undef EXTRACT_NUMBER_AND_INCR
8366 #undef DEBUG_PRINT_COMPILED_PATTERN
8367 #undef DEBUG_PRINT_DOUBLE_STRING
8369 #undef INIT_FAIL_STACK
8370 #undef RESET_FAIL_STACK
8371 #undef DOUBLE_FAIL_STACK
8372 #undef PUSH_PATTERN_OP
8373 #undef PUSH_FAILURE_POINTER
8374 #undef PUSH_FAILURE_INT
8375 #undef PUSH_FAILURE_ELT
8376 #undef POP_FAILURE_POINTER
8377 #undef POP_FAILURE_INT
8378 #undef POP_FAILURE_ELT
8381 #undef PUSH_FAILURE_POINT
8382 #undef POP_FAILURE_POINT
8384 #undef REG_UNSET_VALUE
8392 #undef INIT_BUF_SIZE
8393 #undef GET_BUFFER_SPACE
8401 #undef EXTEND_BUFFER
8402 #undef GET_UNSIGNED_NUMBER
8403 #undef FREE_STACK_RETURN
8405 # undef POINTER_TO_OFFSET
8406 # undef MATCHING_IN_FRST_STRING
8408 # undef AT_STRINGS_BEG
8409 # undef AT_STRINGS_END
8412 # undef FREE_VARIABLES
8413 # undef NO_HIGHEST_ACTIVE_REG
8414 # undef NO_LOWEST_ACTIVE_REG
8418 # undef COMPILED_BUFFER_VAR
8419 # undef OFFSET_ADDRESS_SIZE
8420 # undef CHAR_CLASS_SIZE
8427 # define DEFINED_ONCE