1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the
3 internationalization features.)
5 Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
6 2002, 2003, 2004, 2005, 2006, 2007 Free Software Foundation, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 This program 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
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
24 - structure the opcode space into opcode+flag.
25 - merge with glibc's regex.[ch].
26 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
27 need to modify the compiled regexp so that re_match can be reentrant.
28 - get rid of on_failure_jump_smart by doing the optimization in re_comp
29 rather than at run-time, so that re_match can be reentrant.
32 /* AIX requires this to be the first thing in the file. */
33 #if defined _AIX && !defined REGEX_MALLOC
41 #if defined STDC_HEADERS && !defined emacs
44 /* We need this for `regex.h', and perhaps for the Emacs include files. */
45 # include <sys/types.h>
48 /* Whether to use ISO C Amendment 1 wide char functions.
49 Those should not be used for Emacs since it uses its own. */
51 #define WIDE_CHAR_SUPPORT 1
53 #define WIDE_CHAR_SUPPORT \
54 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
57 /* For platform which support the ISO C amendement 1 functionality we
58 support user defined character classes. */
60 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
66 /* We have to keep the namespace clean. */
67 # define regfree(preg) __regfree (preg)
68 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
69 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
70 # define regerror(err_code, preg, errbuf, errbuf_size) \
71 __regerror(err_code, preg, errbuf, errbuf_size)
72 # define re_set_registers(bu, re, nu, st, en) \
73 __re_set_registers (bu, re, nu, st, en)
74 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
75 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
76 # define re_match(bufp, string, size, pos, regs) \
77 __re_match (bufp, string, size, pos, regs)
78 # define re_search(bufp, string, size, startpos, range, regs) \
79 __re_search (bufp, string, size, startpos, range, regs)
80 # define re_compile_pattern(pattern, length, bufp) \
81 __re_compile_pattern (pattern, length, bufp)
82 # define re_set_syntax(syntax) __re_set_syntax (syntax)
83 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
84 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
85 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
87 /* Make sure we call libc's function even if the user overrides them. */
88 # define btowc __btowc
89 # define iswctype __iswctype
90 # define wctype __wctype
92 # define WEAK_ALIAS(a,b) weak_alias (a, b)
94 /* We are also using some library internals. */
95 # include <locale/localeinfo.h>
96 # include <locale/elem-hash.h>
97 # include <langinfo.h>
99 # define WEAK_ALIAS(a,b)
102 /* This is for other GNU distributions with internationalized messages. */
103 #if HAVE_LIBINTL_H || defined _LIBC
104 # include <libintl.h>
106 # define gettext(msgid) (msgid)
110 /* This define is so xgettext can find the internationalizable
112 # define gettext_noop(String) String
115 /* The `emacs' switch turns on certain matching commands
116 that make sense only in Emacs. */
122 /* Make syntax table lookup grant data in gl_state. */
123 # define SYNTAX_ENTRY_VIA_PROPERTY
126 # include "charset.h"
127 # include "category.h"
132 # define malloc xmalloc
136 # define realloc xrealloc
142 /* Converts the pointer to the char to BEG-based offset from the start. */
143 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
144 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
146 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
147 # define RE_STRING_CHAR(p, s) \
148 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
149 # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \
150 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
152 /* Set C a (possibly multibyte) character before P. P points into a
153 string which is the virtual concatenation of STR1 (which ends at
154 END1) or STR2 (which ends at END2). */
155 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
159 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
160 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
161 re_char *d0 = dtemp; \
162 PREV_CHAR_BOUNDARY (d0, dlimit); \
163 c = STRING_CHAR (d0, dtemp - d0); \
166 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
170 #else /* not emacs */
172 /* If we are not linking with Emacs proper,
173 we can't use the relocating allocator
174 even if config.h says that we can. */
177 # if defined STDC_HEADERS || defined _LIBC
184 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
191 val
= (void *) malloc (size
);
194 write (2, "virtual memory exhausted\n", 25);
201 xrealloc (block
, size
)
206 /* We must call malloc explicitly when BLOCK is 0, since some
207 reallocs don't do this. */
209 val
= (void *) malloc (size
);
211 val
= (void *) realloc (block
, size
);
214 write (2, "virtual memory exhausted\n", 25);
223 # define malloc xmalloc
227 # define realloc xrealloc
229 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
230 If nothing else has been done, use the method below. */
231 # ifdef INHIBIT_STRING_HEADER
232 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
233 # if !defined bzero && !defined bcopy
234 # undef INHIBIT_STRING_HEADER
239 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
240 This is used in most programs--a few other programs avoid this
241 by defining INHIBIT_STRING_HEADER. */
242 # ifndef INHIBIT_STRING_HEADER
243 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
247 # define bzero(s, n) (memset (s, '\0', n), (s))
249 # define bzero(s, n) __bzero (s, n)
253 # include <strings.h>
255 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
258 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
263 /* Define the syntax stuff for \<, \>, etc. */
265 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
266 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
268 # ifdef SWITCH_ENUM_BUG
269 # define SWITCH_ENUM_CAST(x) ((int)(x))
271 # define SWITCH_ENUM_CAST(x) (x)
274 /* Dummy macros for non-Emacs environments. */
275 # define BASE_LEADING_CODE_P(c) (0)
276 # define CHAR_CHARSET(c) 0
277 # define CHARSET_LEADING_CODE_BASE(c) 0
278 # define MAX_MULTIBYTE_LENGTH 1
279 # define RE_MULTIBYTE_P(x) 0
280 # define WORD_BOUNDARY_P(c1, c2) (0)
281 # define CHAR_HEAD_P(p) (1)
282 # define SINGLE_BYTE_CHAR_P(c) (1)
283 # define SAME_CHARSET_P(c1, c2) (1)
284 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
285 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
286 # define STRING_CHAR(p, s) (*(p))
287 # define RE_STRING_CHAR STRING_CHAR
288 # define CHAR_STRING(c, s) (*(s) = (c), 1)
289 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
290 # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH
291 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
292 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
293 # define MAKE_CHAR(charset, c1, c2) (c1)
294 #endif /* not emacs */
297 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
298 # define RE_TRANSLATE_P(TBL) (TBL)
301 /* Get the interface, including the syntax bits. */
304 /* isalpha etc. are used for the character classes. */
309 /* 1 if C is an ASCII character. */
310 # define IS_REAL_ASCII(c) ((c) < 0200)
312 /* 1 if C is a unibyte character. */
313 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
315 /* The Emacs definitions should not be directly affected by locales. */
317 /* In Emacs, these are only used for single-byte characters. */
318 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
319 # define ISCNTRL(c) ((c) < ' ')
320 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
321 || ((c) >= 'a' && (c) <= 'f') \
322 || ((c) >= 'A' && (c) <= 'F'))
324 /* This is only used for single-byte characters. */
325 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
327 /* The rest must handle multibyte characters. */
329 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
330 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
333 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
334 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
337 # define ISALNUM(c) (IS_REAL_ASCII (c) \
338 ? (((c) >= 'a' && (c) <= 'z') \
339 || ((c) >= 'A' && (c) <= 'Z') \
340 || ((c) >= '0' && (c) <= '9')) \
341 : SYNTAX (c) == Sword)
343 # define ISALPHA(c) (IS_REAL_ASCII (c) \
344 ? (((c) >= 'a' && (c) <= 'z') \
345 || ((c) >= 'A' && (c) <= 'Z')) \
346 : SYNTAX (c) == Sword)
348 # define ISLOWER(c) (LOWERCASEP (c))
350 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
351 ? ((c) > ' ' && (c) < 0177 \
352 && !(((c) >= 'a' && (c) <= 'z') \
353 || ((c) >= 'A' && (c) <= 'Z') \
354 || ((c) >= '0' && (c) <= '9'))) \
355 : SYNTAX (c) != Sword)
357 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
359 # define ISUPPER(c) (UPPERCASEP (c))
361 # define ISWORD(c) (SYNTAX (c) == Sword)
363 #else /* not emacs */
365 /* Jim Meyering writes:
367 "... Some ctype macros are valid only for character codes that
368 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
369 using /bin/cc or gcc but without giving an ansi option). So, all
370 ctype uses should be through macros like ISPRINT... If
371 STDC_HEADERS is defined, then autoconf has verified that the ctype
372 macros don't need to be guarded with references to isascii. ...
373 Defining isascii to 1 should let any compiler worth its salt
374 eliminate the && through constant folding."
375 Solaris defines some of these symbols so we must undefine them first. */
378 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
379 # define ISASCII(c) 1
381 # define ISASCII(c) isascii(c)
384 /* 1 if C is an ASCII character. */
385 # define IS_REAL_ASCII(c) ((c) < 0200)
387 /* This distinction is not meaningful, except in Emacs. */
388 # define ISUNIBYTE(c) 1
391 # define ISBLANK(c) (ISASCII (c) && isblank (c))
393 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
396 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
398 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
402 # define ISPRINT(c) (ISASCII (c) && isprint (c))
403 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
404 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
405 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
406 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
407 # define ISLOWER(c) (ISASCII (c) && islower (c))
408 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
409 # define ISSPACE(c) (ISASCII (c) && isspace (c))
410 # define ISUPPER(c) (ISASCII (c) && isupper (c))
411 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
413 # define ISWORD(c) ISALPHA(c)
416 # define TOLOWER(c) _tolower(c)
418 # define TOLOWER(c) tolower(c)
421 /* How many characters in the character set. */
422 # define CHAR_SET_SIZE 256
426 extern char *re_syntax_table
;
428 # else /* not SYNTAX_TABLE */
430 static char re_syntax_table
[CHAR_SET_SIZE
];
441 bzero (re_syntax_table
, sizeof re_syntax_table
);
443 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
445 re_syntax_table
[c
] = Sword
;
447 re_syntax_table
['_'] = Ssymbol
;
452 # endif /* not SYNTAX_TABLE */
454 # define SYNTAX(c) re_syntax_table[(c)]
456 #endif /* not emacs */
459 # define NULL (void *)0
462 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
463 since ours (we hope) works properly with all combinations of
464 machines, compilers, `char' and `unsigned char' argument types.
465 (Per Bothner suggested the basic approach.) */
466 #undef SIGN_EXTEND_CHAR
468 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
469 #else /* not __STDC__ */
470 /* As in Harbison and Steele. */
471 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
474 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
475 use `alloca' instead of `malloc'. This is because using malloc in
476 re_search* or re_match* could cause memory leaks when C-g is used in
477 Emacs; also, malloc is slower and causes storage fragmentation. On
478 the other hand, malloc is more portable, and easier to debug.
480 Because we sometimes use alloca, some routines have to be macros,
481 not functions -- `alloca'-allocated space disappears at the end of the
482 function it is called in. */
486 # define REGEX_ALLOCATE malloc
487 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
488 # define REGEX_FREE free
490 #else /* not REGEX_MALLOC */
492 /* Emacs already defines alloca, sometimes. */
495 /* Make alloca work the best possible way. */
497 # define alloca __builtin_alloca
498 # else /* not __GNUC__ */
501 # endif /* HAVE_ALLOCA_H */
502 # endif /* not __GNUC__ */
504 # endif /* not alloca */
506 # define REGEX_ALLOCATE alloca
508 /* Assumes a `char *destination' variable. */
509 # define REGEX_REALLOCATE(source, osize, nsize) \
510 (destination = (char *) alloca (nsize), \
511 memcpy (destination, source, osize))
513 /* No need to do anything to free, after alloca. */
514 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
516 #endif /* not REGEX_MALLOC */
518 /* Define how to allocate the failure stack. */
520 #if defined REL_ALLOC && defined REGEX_MALLOC
522 # define REGEX_ALLOCATE_STACK(size) \
523 r_alloc (&failure_stack_ptr, (size))
524 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
525 r_re_alloc (&failure_stack_ptr, (nsize))
526 # define REGEX_FREE_STACK(ptr) \
527 r_alloc_free (&failure_stack_ptr)
529 #else /* not using relocating allocator */
533 # define REGEX_ALLOCATE_STACK malloc
534 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
535 # define REGEX_FREE_STACK free
537 # else /* not REGEX_MALLOC */
539 # define REGEX_ALLOCATE_STACK alloca
541 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
542 REGEX_REALLOCATE (source, osize, nsize)
543 /* No need to explicitly free anything. */
544 # define REGEX_FREE_STACK(arg) ((void)0)
546 # endif /* not REGEX_MALLOC */
547 #endif /* not using relocating allocator */
550 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
551 `string1' or just past its end. This works if PTR is NULL, which is
553 #define FIRST_STRING_P(ptr) \
554 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
556 /* (Re)Allocate N items of type T using malloc, or fail. */
557 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
558 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
559 #define RETALLOC_IF(addr, n, t) \
560 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
561 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
563 #define BYTEWIDTH 8 /* In bits. */
565 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
569 #define MAX(a, b) ((a) > (b) ? (a) : (b))
570 #define MIN(a, b) ((a) < (b) ? (a) : (b))
572 /* Type of source-pattern and string chars. */
573 typedef const unsigned char re_char
;
575 typedef char boolean
;
579 static int re_match_2_internal
_RE_ARGS ((struct re_pattern_buffer
*bufp
,
580 re_char
*string1
, int size1
,
581 re_char
*string2
, int size2
,
583 struct re_registers
*regs
,
586 /* These are the command codes that appear in compiled regular
587 expressions. Some opcodes are followed by argument bytes. A
588 command code can specify any interpretation whatsoever for its
589 arguments. Zero bytes may appear in the compiled regular expression. */
595 /* Succeed right away--no more backtracking. */
598 /* Followed by one byte giving n, then by n literal bytes. */
601 /* Matches any (more or less) character. */
604 /* Matches any one char belonging to specified set. First
605 following byte is number of bitmap bytes. Then come bytes
606 for a bitmap saying which chars are in. Bits in each byte
607 are ordered low-bit-first. A character is in the set if its
608 bit is 1. A character too large to have a bit in the map is
609 automatically not in the set.
611 If the length byte has the 0x80 bit set, then that stuff
612 is followed by a range table:
613 2 bytes of flags for character sets (low 8 bits, high 8 bits)
614 See RANGE_TABLE_WORK_BITS below.
615 2 bytes, the number of pairs that follow (upto 32767)
616 pairs, each 2 multibyte characters,
617 each multibyte character represented as 3 bytes. */
620 /* Same parameters as charset, but match any character that is
621 not one of those specified. */
624 /* Start remembering the text that is matched, for storing in a
625 register. Followed by one byte with the register number, in
626 the range 0 to one less than the pattern buffer's re_nsub
630 /* Stop remembering the text that is matched and store it in a
631 memory register. Followed by one byte with the register
632 number, in the range 0 to one less than `re_nsub' in the
636 /* Match a duplicate of something remembered. Followed by one
637 byte containing the register number. */
640 /* Fail unless at beginning of line. */
643 /* Fail unless at end of line. */
646 /* Succeeds if at beginning of buffer (if emacs) or at beginning
647 of string to be matched (if not). */
650 /* Analogously, for end of buffer/string. */
653 /* Followed by two byte relative address to which to jump. */
656 /* Followed by two-byte relative address of place to resume at
657 in case of failure. */
660 /* Like on_failure_jump, but pushes a placeholder instead of the
661 current string position when executed. */
662 on_failure_keep_string_jump
,
664 /* Just like `on_failure_jump', except that it checks that we
665 don't get stuck in an infinite loop (matching an empty string
667 on_failure_jump_loop
,
669 /* Just like `on_failure_jump_loop', except that it checks for
670 a different kind of loop (the kind that shows up with non-greedy
671 operators). This operation has to be immediately preceded
673 on_failure_jump_nastyloop
,
675 /* A smart `on_failure_jump' used for greedy * and + operators.
676 It analyses the loop before which it is put and if the
677 loop does not require backtracking, it changes itself to
678 `on_failure_keep_string_jump' and short-circuits the loop,
679 else it just defaults to changing itself into `on_failure_jump'.
680 It assumes that it is pointing to just past a `jump'. */
681 on_failure_jump_smart
,
683 /* Followed by two-byte relative address and two-byte number n.
684 After matching N times, jump to the address upon failure.
685 Does not work if N starts at 0: use on_failure_jump_loop
689 /* Followed by two-byte relative address, and two-byte number n.
690 Jump to the address N times, then fail. */
693 /* Set the following two-byte relative address to the
694 subsequent two-byte number. The address *includes* the two
698 wordbeg
, /* Succeeds if at word beginning. */
699 wordend
, /* Succeeds if at word end. */
701 wordbound
, /* Succeeds if at a word boundary. */
702 notwordbound
, /* Succeeds if not at a word boundary. */
704 symbeg
, /* Succeeds if at symbol beginning. */
705 symend
, /* Succeeds if at symbol end. */
707 /* Matches any character whose syntax is specified. Followed by
708 a byte which contains a syntax code, e.g., Sword. */
711 /* Matches any character whose syntax is not that specified. */
715 ,before_dot
, /* Succeeds if before point. */
716 at_dot
, /* Succeeds if at point. */
717 after_dot
, /* Succeeds if after point. */
719 /* Matches any character whose category-set contains the specified
720 category. The operator is followed by a byte which contains a
721 category code (mnemonic ASCII character). */
724 /* Matches any character whose category-set does not contain the
725 specified category. The operator is followed by a byte which
726 contains the category code (mnemonic ASCII character). */
731 /* Common operations on the compiled pattern. */
733 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
735 #define STORE_NUMBER(destination, number) \
737 (destination)[0] = (number) & 0377; \
738 (destination)[1] = (number) >> 8; \
741 /* Same as STORE_NUMBER, except increment DESTINATION to
742 the byte after where the number is stored. Therefore, DESTINATION
743 must be an lvalue. */
745 #define STORE_NUMBER_AND_INCR(destination, number) \
747 STORE_NUMBER (destination, number); \
748 (destination) += 2; \
751 /* Put into DESTINATION a number stored in two contiguous bytes starting
754 #define EXTRACT_NUMBER(destination, source) \
756 (destination) = *(source) & 0377; \
757 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
761 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
763 extract_number (dest
, source
)
767 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
768 *dest
= *source
& 0377;
772 # ifndef EXTRACT_MACROS /* To debug the macros. */
773 # undef EXTRACT_NUMBER
774 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
775 # endif /* not EXTRACT_MACROS */
779 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
780 SOURCE must be an lvalue. */
782 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
784 EXTRACT_NUMBER (destination, source); \
789 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
792 extract_number_and_incr (destination
, source
)
796 extract_number (destination
, *source
);
800 # ifndef EXTRACT_MACROS
801 # undef EXTRACT_NUMBER_AND_INCR
802 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
803 extract_number_and_incr (&dest, &src)
804 # endif /* not EXTRACT_MACROS */
808 /* Store a multibyte character in three contiguous bytes starting
809 DESTINATION, and increment DESTINATION to the byte after where the
810 character is stored. Therefore, DESTINATION must be an lvalue. */
812 #define STORE_CHARACTER_AND_INCR(destination, character) \
814 (destination)[0] = (character) & 0377; \
815 (destination)[1] = ((character) >> 8) & 0377; \
816 (destination)[2] = (character) >> 16; \
817 (destination) += 3; \
820 /* Put into DESTINATION a character stored in three contiguous bytes
821 starting at SOURCE. */
823 #define EXTRACT_CHARACTER(destination, source) \
825 (destination) = ((source)[0] \
826 | ((source)[1] << 8) \
827 | ((source)[2] << 16)); \
831 /* Macros for charset. */
833 /* Size of bitmap of charset P in bytes. P is a start of charset,
834 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
835 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
837 /* Nonzero if charset P has range table. */
838 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
840 /* Return the address of range table of charset P. But not the start
841 of table itself, but the before where the number of ranges is
842 stored. `2 +' means to skip re_opcode_t and size of bitmap,
843 and the 2 bytes of flags at the start of the range table. */
844 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
846 /* Extract the bit flags that start a range table. */
847 #define CHARSET_RANGE_TABLE_BITS(p) \
848 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
849 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
851 /* Test if C is listed in the bitmap of charset P. */
852 #define CHARSET_LOOKUP_BITMAP(p, c) \
853 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
854 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
856 /* Return the address of end of RANGE_TABLE. COUNT is number of
857 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
858 is start of range and end of range. `* 3' is size of each start
860 #define CHARSET_RANGE_TABLE_END(range_table, count) \
861 ((range_table) + (count) * 2 * 3)
863 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
864 COUNT is number of ranges in RANGE_TABLE. */
865 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
868 re_wchar_t range_start, range_end; \
870 re_char *range_table_end \
871 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
873 for (p = (range_table); p < range_table_end; p += 2 * 3) \
875 EXTRACT_CHARACTER (range_start, p); \
876 EXTRACT_CHARACTER (range_end, p + 3); \
878 if (range_start <= (c) && (c) <= range_end) \
887 /* Test if C is in range table of CHARSET. The flag NOT is negated if
888 C is listed in it. */
889 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
892 /* Number of ranges in range table. */ \
894 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
896 EXTRACT_NUMBER_AND_INCR (count, range_table); \
897 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
901 /* If DEBUG is defined, Regex prints many voluminous messages about what
902 it is doing (if the variable `debug' is nonzero). If linked with the
903 main program in `iregex.c', you can enter patterns and strings
904 interactively. And if linked with the main program in `main.c' and
905 the other test files, you can run the already-written tests. */
909 /* We use standard I/O for debugging. */
912 /* It is useful to test things that ``must'' be true when debugging. */
915 static int debug
= -100000;
917 # define DEBUG_STATEMENT(e) e
918 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
919 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
920 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
921 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
922 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
923 if (debug > 0) print_partial_compiled_pattern (s, e)
924 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
925 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
928 /* Print the fastmap in human-readable form. */
931 print_fastmap (fastmap
)
934 unsigned was_a_range
= 0;
937 while (i
< (1 << BYTEWIDTH
))
943 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
959 /* Print a compiled pattern string in human-readable form, starting at
960 the START pointer into it and ending just before the pointer END. */
963 print_partial_compiled_pattern (start
, end
)
973 fprintf (stderr
, "(null)\n");
977 /* Loop over pattern commands. */
980 fprintf (stderr
, "%d:\t", p
- start
);
982 switch ((re_opcode_t
) *p
++)
985 fprintf (stderr
, "/no_op");
989 fprintf (stderr
, "/succeed");
994 fprintf (stderr
, "/exactn/%d", mcnt
);
997 fprintf (stderr
, "/%c", *p
++);
1003 fprintf (stderr
, "/start_memory/%d", *p
++);
1007 fprintf (stderr
, "/stop_memory/%d", *p
++);
1011 fprintf (stderr
, "/duplicate/%d", *p
++);
1015 fprintf (stderr
, "/anychar");
1021 register int c
, last
= -100;
1022 register int in_range
= 0;
1023 int length
= CHARSET_BITMAP_SIZE (p
- 1);
1024 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
1026 fprintf (stderr
, "/charset [%s",
1027 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
1030 fprintf (stderr
, " !extends past end of pattern! ");
1032 for (c
= 0; c
< 256; c
++)
1034 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
1036 /* Are we starting a range? */
1037 if (last
+ 1 == c
&& ! in_range
)
1039 fprintf (stderr
, "-");
1042 /* Have we broken a range? */
1043 else if (last
+ 1 != c
&& in_range
)
1045 fprintf (stderr
, "%c", last
);
1050 fprintf (stderr
, "%c", c
);
1056 fprintf (stderr
, "%c", last
);
1058 fprintf (stderr
, "]");
1062 if (has_range_table
)
1065 fprintf (stderr
, "has-range-table");
1067 /* ??? Should print the range table; for now, just skip it. */
1068 p
+= 2; /* skip range table bits */
1069 EXTRACT_NUMBER_AND_INCR (count
, p
);
1070 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1076 fprintf (stderr
, "/begline");
1080 fprintf (stderr
, "/endline");
1083 case on_failure_jump
:
1084 extract_number_and_incr (&mcnt
, &p
);
1085 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1088 case on_failure_keep_string_jump
:
1089 extract_number_and_incr (&mcnt
, &p
);
1090 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1093 case on_failure_jump_nastyloop
:
1094 extract_number_and_incr (&mcnt
, &p
);
1095 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1098 case on_failure_jump_loop
:
1099 extract_number_and_incr (&mcnt
, &p
);
1100 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1103 case on_failure_jump_smart
:
1104 extract_number_and_incr (&mcnt
, &p
);
1105 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1109 extract_number_and_incr (&mcnt
, &p
);
1110 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1114 extract_number_and_incr (&mcnt
, &p
);
1115 extract_number_and_incr (&mcnt2
, &p
);
1116 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1120 extract_number_and_incr (&mcnt
, &p
);
1121 extract_number_and_incr (&mcnt2
, &p
);
1122 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1126 extract_number_and_incr (&mcnt
, &p
);
1127 extract_number_and_incr (&mcnt2
, &p
);
1128 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1132 fprintf (stderr
, "/wordbound");
1136 fprintf (stderr
, "/notwordbound");
1140 fprintf (stderr
, "/wordbeg");
1144 fprintf (stderr
, "/wordend");
1148 fprintf (stderr
, "/symbeg");
1152 fprintf (stderr
, "/symend");
1156 fprintf (stderr
, "/syntaxspec");
1158 fprintf (stderr
, "/%d", mcnt
);
1162 fprintf (stderr
, "/notsyntaxspec");
1164 fprintf (stderr
, "/%d", mcnt
);
1169 fprintf (stderr
, "/before_dot");
1173 fprintf (stderr
, "/at_dot");
1177 fprintf (stderr
, "/after_dot");
1181 fprintf (stderr
, "/categoryspec");
1183 fprintf (stderr
, "/%d", mcnt
);
1186 case notcategoryspec
:
1187 fprintf (stderr
, "/notcategoryspec");
1189 fprintf (stderr
, "/%d", mcnt
);
1194 fprintf (stderr
, "/begbuf");
1198 fprintf (stderr
, "/endbuf");
1202 fprintf (stderr
, "?%d", *(p
-1));
1205 fprintf (stderr
, "\n");
1208 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1213 print_compiled_pattern (bufp
)
1214 struct re_pattern_buffer
*bufp
;
1216 re_char
*buffer
= bufp
->buffer
;
1218 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1219 printf ("%ld bytes used/%ld bytes allocated.\n",
1220 bufp
->used
, bufp
->allocated
);
1222 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1224 printf ("fastmap: ");
1225 print_fastmap (bufp
->fastmap
);
1228 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1229 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1230 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1231 printf ("no_sub: %d\t", bufp
->no_sub
);
1232 printf ("not_bol: %d\t", bufp
->not_bol
);
1233 printf ("not_eol: %d\t", bufp
->not_eol
);
1234 printf ("syntax: %lx\n", bufp
->syntax
);
1236 /* Perhaps we should print the translate table? */
1241 print_double_string (where
, string1
, size1
, string2
, size2
)
1254 if (FIRST_STRING_P (where
))
1256 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1257 putchar (string1
[this_char
]);
1262 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1263 putchar (string2
[this_char
]);
1267 #else /* not DEBUG */
1272 # define DEBUG_STATEMENT(e)
1273 # define DEBUG_PRINT1(x)
1274 # define DEBUG_PRINT2(x1, x2)
1275 # define DEBUG_PRINT3(x1, x2, x3)
1276 # define DEBUG_PRINT4(x1, x2, x3, x4)
1277 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1278 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1280 #endif /* not DEBUG */
1282 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1283 also be assigned to arbitrarily: each pattern buffer stores its own
1284 syntax, so it can be changed between regex compilations. */
1285 /* This has no initializer because initialized variables in Emacs
1286 become read-only after dumping. */
1287 reg_syntax_t re_syntax_options
;
1290 /* Specify the precise syntax of regexps for compilation. This provides
1291 for compatibility for various utilities which historically have
1292 different, incompatible syntaxes.
1294 The argument SYNTAX is a bit mask comprised of the various bits
1295 defined in regex.h. We return the old syntax. */
1298 re_set_syntax (syntax
)
1299 reg_syntax_t syntax
;
1301 reg_syntax_t ret
= re_syntax_options
;
1303 re_syntax_options
= syntax
;
1306 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1308 /* Regexp to use to replace spaces, or NULL meaning don't. */
1309 static re_char
*whitespace_regexp
;
1312 re_set_whitespace_regexp (regexp
)
1315 whitespace_regexp
= (re_char
*) regexp
;
1317 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1319 /* This table gives an error message for each of the error codes listed
1320 in regex.h. Obviously the order here has to be same as there.
1321 POSIX doesn't require that we do anything for REG_NOERROR,
1322 but why not be nice? */
1324 static const char *re_error_msgid
[] =
1326 gettext_noop ("Success"), /* REG_NOERROR */
1327 gettext_noop ("No match"), /* REG_NOMATCH */
1328 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1329 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1330 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1331 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1332 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1333 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1334 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1335 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1336 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1337 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1338 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1339 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1340 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1341 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1342 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1343 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1346 /* Avoiding alloca during matching, to placate r_alloc. */
1348 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1349 searching and matching functions should not call alloca. On some
1350 systems, alloca is implemented in terms of malloc, and if we're
1351 using the relocating allocator routines, then malloc could cause a
1352 relocation, which might (if the strings being searched are in the
1353 ralloc heap) shift the data out from underneath the regexp
1356 Here's another reason to avoid allocation: Emacs
1357 processes input from X in a signal handler; processing X input may
1358 call malloc; if input arrives while a matching routine is calling
1359 malloc, then we're scrod. But Emacs can't just block input while
1360 calling matching routines; then we don't notice interrupts when
1361 they come in. So, Emacs blocks input around all regexp calls
1362 except the matching calls, which it leaves unprotected, in the
1363 faith that they will not malloc. */
1365 /* Normally, this is fine. */
1366 #define MATCH_MAY_ALLOCATE
1368 /* When using GNU C, we are not REALLY using the C alloca, no matter
1369 what config.h may say. So don't take precautions for it. */
1374 /* The match routines may not allocate if (1) they would do it with malloc
1375 and (2) it's not safe for them to use malloc.
1376 Note that if REL_ALLOC is defined, matching would not use malloc for the
1377 failure stack, but we would still use it for the register vectors;
1378 so REL_ALLOC should not affect this. */
1379 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1380 # undef MATCH_MAY_ALLOCATE
1384 /* Failure stack declarations and macros; both re_compile_fastmap and
1385 re_match_2 use a failure stack. These have to be macros because of
1386 REGEX_ALLOCATE_STACK. */
1389 /* Approximate number of failure points for which to initially allocate space
1390 when matching. If this number is exceeded, we allocate more
1391 space, so it is not a hard limit. */
1392 #ifndef INIT_FAILURE_ALLOC
1393 # define INIT_FAILURE_ALLOC 20
1396 /* Roughly the maximum number of failure points on the stack. Would be
1397 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1398 This is a variable only so users of regex can assign to it; we never
1399 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1400 before using it, so it should probably be a byte-count instead. */
1401 # if defined MATCH_MAY_ALLOCATE
1402 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1403 whose default stack limit is 2mb. In order for a larger
1404 value to work reliably, you have to try to make it accord
1405 with the process stack limit. */
1406 size_t re_max_failures
= 40000;
1408 size_t re_max_failures
= 4000;
1411 union fail_stack_elt
1414 /* This should be the biggest `int' that's no bigger than a pointer. */
1418 typedef union fail_stack_elt fail_stack_elt_t
;
1422 fail_stack_elt_t
*stack
;
1424 size_t avail
; /* Offset of next open position. */
1425 size_t frame
; /* Offset of the cur constructed frame. */
1428 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1429 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1432 /* Define macros to initialize and free the failure stack.
1433 Do `return -2' if the alloc fails. */
1435 #ifdef MATCH_MAY_ALLOCATE
1436 # define INIT_FAIL_STACK() \
1438 fail_stack.stack = (fail_stack_elt_t *) \
1439 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1440 * sizeof (fail_stack_elt_t)); \
1442 if (fail_stack.stack == NULL) \
1445 fail_stack.size = INIT_FAILURE_ALLOC; \
1446 fail_stack.avail = 0; \
1447 fail_stack.frame = 0; \
1450 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1452 # define INIT_FAIL_STACK() \
1454 fail_stack.avail = 0; \
1455 fail_stack.frame = 0; \
1458 # define RESET_FAIL_STACK() ((void)0)
1462 /* Double the size of FAIL_STACK, up to a limit
1463 which allows approximately `re_max_failures' items.
1465 Return 1 if succeeds, and 0 if either ran out of memory
1466 allocating space for it or it was already too large.
1468 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1470 /* Factor to increase the failure stack size by
1471 when we increase it.
1472 This used to be 2, but 2 was too wasteful
1473 because the old discarded stacks added up to as much space
1474 were as ultimate, maximum-size stack. */
1475 #define FAIL_STACK_GROWTH_FACTOR 4
1477 #define GROW_FAIL_STACK(fail_stack) \
1478 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1479 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1481 : ((fail_stack).stack \
1482 = (fail_stack_elt_t *) \
1483 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1484 (fail_stack).size * sizeof (fail_stack_elt_t), \
1485 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1486 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1487 * FAIL_STACK_GROWTH_FACTOR))), \
1489 (fail_stack).stack == NULL \
1491 : ((fail_stack).size \
1492 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1493 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1494 * FAIL_STACK_GROWTH_FACTOR)) \
1495 / sizeof (fail_stack_elt_t)), \
1499 /* Push a pointer value onto the failure stack.
1500 Assumes the variable `fail_stack'. Probably should only
1501 be called from within `PUSH_FAILURE_POINT'. */
1502 #define PUSH_FAILURE_POINTER(item) \
1503 fail_stack.stack[fail_stack.avail++].pointer = (item)
1505 /* This pushes an integer-valued item onto the failure stack.
1506 Assumes the variable `fail_stack'. Probably should only
1507 be called from within `PUSH_FAILURE_POINT'. */
1508 #define PUSH_FAILURE_INT(item) \
1509 fail_stack.stack[fail_stack.avail++].integer = (item)
1511 /* Push a fail_stack_elt_t value onto the failure stack.
1512 Assumes the variable `fail_stack'. Probably should only
1513 be called from within `PUSH_FAILURE_POINT'. */
1514 #define PUSH_FAILURE_ELT(item) \
1515 fail_stack.stack[fail_stack.avail++] = (item)
1517 /* These three POP... operations complement the three PUSH... operations.
1518 All assume that `fail_stack' is nonempty. */
1519 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1520 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1521 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1523 /* Individual items aside from the registers. */
1524 #define NUM_NONREG_ITEMS 3
1526 /* Used to examine the stack (to detect infinite loops). */
1527 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1528 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1529 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1530 #define TOP_FAILURE_HANDLE() fail_stack.frame
1533 #define ENSURE_FAIL_STACK(space) \
1534 while (REMAINING_AVAIL_SLOTS <= space) { \
1535 if (!GROW_FAIL_STACK (fail_stack)) \
1537 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1538 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1541 /* Push register NUM onto the stack. */
1542 #define PUSH_FAILURE_REG(num) \
1544 char *destination; \
1545 ENSURE_FAIL_STACK(3); \
1546 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1547 num, regstart[num], regend[num]); \
1548 PUSH_FAILURE_POINTER (regstart[num]); \
1549 PUSH_FAILURE_POINTER (regend[num]); \
1550 PUSH_FAILURE_INT (num); \
1553 /* Change the counter's value to VAL, but make sure that it will
1554 be reset when backtracking. */
1555 #define PUSH_NUMBER(ptr,val) \
1557 char *destination; \
1559 ENSURE_FAIL_STACK(3); \
1560 EXTRACT_NUMBER (c, ptr); \
1561 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1562 PUSH_FAILURE_INT (c); \
1563 PUSH_FAILURE_POINTER (ptr); \
1564 PUSH_FAILURE_INT (-1); \
1565 STORE_NUMBER (ptr, val); \
1568 /* Pop a saved register off the stack. */
1569 #define POP_FAILURE_REG_OR_COUNT() \
1571 int reg = POP_FAILURE_INT (); \
1574 /* It's a counter. */ \
1575 /* Here, we discard `const', making re_match non-reentrant. */ \
1576 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1577 reg = POP_FAILURE_INT (); \
1578 STORE_NUMBER (ptr, reg); \
1579 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1583 regend[reg] = POP_FAILURE_POINTER (); \
1584 regstart[reg] = POP_FAILURE_POINTER (); \
1585 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1586 reg, regstart[reg], regend[reg]); \
1590 /* Check that we are not stuck in an infinite loop. */
1591 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1593 int failure = TOP_FAILURE_HANDLE (); \
1594 /* Check for infinite matching loops */ \
1595 while (failure > 0 \
1596 && (FAILURE_STR (failure) == string_place \
1597 || FAILURE_STR (failure) == NULL)) \
1599 assert (FAILURE_PAT (failure) >= bufp->buffer \
1600 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1601 if (FAILURE_PAT (failure) == pat_cur) \
1606 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1607 failure = NEXT_FAILURE_HANDLE(failure); \
1609 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1612 /* Push the information about the state we will need
1613 if we ever fail back to it.
1615 Requires variables fail_stack, regstart, regend and
1616 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1619 Does `return FAILURE_CODE' if runs out of memory. */
1621 #define PUSH_FAILURE_POINT(pattern, string_place) \
1623 char *destination; \
1624 /* Must be int, so when we don't save any registers, the arithmetic \
1625 of 0 + -1 isn't done as unsigned. */ \
1627 DEBUG_STATEMENT (nfailure_points_pushed++); \
1628 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1629 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1630 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1632 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1634 DEBUG_PRINT1 ("\n"); \
1636 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1637 PUSH_FAILURE_INT (fail_stack.frame); \
1639 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1640 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1641 DEBUG_PRINT1 ("'\n"); \
1642 PUSH_FAILURE_POINTER (string_place); \
1644 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1645 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1646 PUSH_FAILURE_POINTER (pattern); \
1648 /* Close the frame by moving the frame pointer past it. */ \
1649 fail_stack.frame = fail_stack.avail; \
1652 /* Estimate the size of data pushed by a typical failure stack entry.
1653 An estimate is all we need, because all we use this for
1654 is to choose a limit for how big to make the failure stack. */
1655 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1656 #define TYPICAL_FAILURE_SIZE 20
1658 /* How many items can still be added to the stack without overflowing it. */
1659 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1662 /* Pops what PUSH_FAIL_STACK pushes.
1664 We restore into the parameters, all of which should be lvalues:
1665 STR -- the saved data position.
1666 PAT -- the saved pattern position.
1667 REGSTART, REGEND -- arrays of string positions.
1669 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1670 `pend', `string1', `size1', `string2', and `size2'. */
1672 #define POP_FAILURE_POINT(str, pat) \
1674 assert (!FAIL_STACK_EMPTY ()); \
1676 /* Remove failure points and point to how many regs pushed. */ \
1677 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1678 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1679 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1681 /* Pop the saved registers. */ \
1682 while (fail_stack.frame < fail_stack.avail) \
1683 POP_FAILURE_REG_OR_COUNT (); \
1685 pat = POP_FAILURE_POINTER (); \
1686 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1687 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1689 /* If the saved string location is NULL, it came from an \
1690 on_failure_keep_string_jump opcode, and we want to throw away the \
1691 saved NULL, thus retaining our current position in the string. */ \
1692 str = POP_FAILURE_POINTER (); \
1693 DEBUG_PRINT2 (" Popping string %p: `", str); \
1694 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1695 DEBUG_PRINT1 ("'\n"); \
1697 fail_stack.frame = POP_FAILURE_INT (); \
1698 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1700 assert (fail_stack.avail >= 0); \
1701 assert (fail_stack.frame <= fail_stack.avail); \
1703 DEBUG_STATEMENT (nfailure_points_popped++); \
1704 } while (0) /* POP_FAILURE_POINT */
1708 /* Registers are set to a sentinel when they haven't yet matched. */
1709 #define REG_UNSET(e) ((e) == NULL)
1711 /* Subroutine declarations and macros for regex_compile. */
1713 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1714 reg_syntax_t syntax
,
1715 struct re_pattern_buffer
*bufp
));
1716 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1717 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1718 int arg1
, int arg2
));
1719 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1720 int arg
, unsigned char *end
));
1721 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1722 int arg1
, int arg2
, unsigned char *end
));
1723 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1725 reg_syntax_t syntax
));
1726 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1728 reg_syntax_t syntax
));
1729 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1730 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1731 char *fastmap
, const int multibyte
));
1733 /* Fetch the next character in the uncompiled pattern, with no
1735 #define PATFETCH(c) \
1738 if (p == pend) return REG_EEND; \
1739 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1744 /* If `translate' is non-null, return translate[D], else just D. We
1745 cast the subscript to translate because some data is declared as
1746 `char *', to avoid warnings when a string constant is passed. But
1747 when we use a character as a subscript we must make it unsigned. */
1749 # define TRANSLATE(d) \
1750 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1754 /* Macros for outputting the compiled pattern into `buffer'. */
1756 /* If the buffer isn't allocated when it comes in, use this. */
1757 #define INIT_BUF_SIZE 32
1759 /* Make sure we have at least N more bytes of space in buffer. */
1760 #define GET_BUFFER_SPACE(n) \
1761 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1764 /* Make sure we have one more byte of buffer space and then add C to it. */
1765 #define BUF_PUSH(c) \
1767 GET_BUFFER_SPACE (1); \
1768 *b++ = (unsigned char) (c); \
1772 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1773 #define BUF_PUSH_2(c1, c2) \
1775 GET_BUFFER_SPACE (2); \
1776 *b++ = (unsigned char) (c1); \
1777 *b++ = (unsigned char) (c2); \
1781 /* As with BUF_PUSH_2, except for three bytes. */
1782 #define BUF_PUSH_3(c1, c2, c3) \
1784 GET_BUFFER_SPACE (3); \
1785 *b++ = (unsigned char) (c1); \
1786 *b++ = (unsigned char) (c2); \
1787 *b++ = (unsigned char) (c3); \
1791 /* Store a jump with opcode OP at LOC to location TO. We store a
1792 relative address offset by the three bytes the jump itself occupies. */
1793 #define STORE_JUMP(op, loc, to) \
1794 store_op1 (op, loc, (to) - (loc) - 3)
1796 /* Likewise, for a two-argument jump. */
1797 #define STORE_JUMP2(op, loc, to, arg) \
1798 store_op2 (op, loc, (to) - (loc) - 3, arg)
1800 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1801 #define INSERT_JUMP(op, loc, to) \
1802 insert_op1 (op, loc, (to) - (loc) - 3, b)
1804 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1805 #define INSERT_JUMP2(op, loc, to, arg) \
1806 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1809 /* This is not an arbitrary limit: the arguments which represent offsets
1810 into the pattern are two bytes long. So if 2^15 bytes turns out to
1811 be too small, many things would have to change. */
1812 # define MAX_BUF_SIZE (1L << 15)
1814 #if 0 /* This is when we thought it could be 2^16 bytes. */
1815 /* Any other compiler which, like MSC, has allocation limit below 2^16
1816 bytes will have to use approach similar to what was done below for
1817 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1818 reallocating to 0 bytes. Such thing is not going to work too well.
1819 You have been warned!! */
1820 #if defined _MSC_VER && !defined WIN32
1821 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1822 # define MAX_BUF_SIZE 65500L
1824 # define MAX_BUF_SIZE (1L << 16)
1828 /* Extend the buffer by twice its current size via realloc and
1829 reset the pointers that pointed into the old block to point to the
1830 correct places in the new one. If extending the buffer results in it
1831 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1832 #if __BOUNDED_POINTERS__
1833 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1834 # define MOVE_BUFFER_POINTER(P) \
1835 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1836 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1839 SET_HIGH_BOUND (b); \
1840 SET_HIGH_BOUND (begalt); \
1841 if (fixup_alt_jump) \
1842 SET_HIGH_BOUND (fixup_alt_jump); \
1844 SET_HIGH_BOUND (laststart); \
1845 if (pending_exact) \
1846 SET_HIGH_BOUND (pending_exact); \
1849 # define MOVE_BUFFER_POINTER(P) (P) += incr
1850 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1852 #define EXTEND_BUFFER() \
1854 re_char *old_buffer = bufp->buffer; \
1855 if (bufp->allocated == MAX_BUF_SIZE) \
1857 bufp->allocated <<= 1; \
1858 if (bufp->allocated > MAX_BUF_SIZE) \
1859 bufp->allocated = MAX_BUF_SIZE; \
1860 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1861 if (bufp->buffer == NULL) \
1862 return REG_ESPACE; \
1863 /* If the buffer moved, move all the pointers into it. */ \
1864 if (old_buffer != bufp->buffer) \
1866 int incr = bufp->buffer - old_buffer; \
1867 MOVE_BUFFER_POINTER (b); \
1868 MOVE_BUFFER_POINTER (begalt); \
1869 if (fixup_alt_jump) \
1870 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1872 MOVE_BUFFER_POINTER (laststart); \
1873 if (pending_exact) \
1874 MOVE_BUFFER_POINTER (pending_exact); \
1876 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1880 /* Since we have one byte reserved for the register number argument to
1881 {start,stop}_memory, the maximum number of groups we can report
1882 things about is what fits in that byte. */
1883 #define MAX_REGNUM 255
1885 /* But patterns can have more than `MAX_REGNUM' registers. We just
1886 ignore the excess. */
1887 typedef int regnum_t
;
1890 /* Macros for the compile stack. */
1892 /* Since offsets can go either forwards or backwards, this type needs to
1893 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1894 /* int may be not enough when sizeof(int) == 2. */
1895 typedef long pattern_offset_t
;
1899 pattern_offset_t begalt_offset
;
1900 pattern_offset_t fixup_alt_jump
;
1901 pattern_offset_t laststart_offset
;
1903 } compile_stack_elt_t
;
1908 compile_stack_elt_t
*stack
;
1910 unsigned avail
; /* Offset of next open position. */
1911 } compile_stack_type
;
1914 #define INIT_COMPILE_STACK_SIZE 32
1916 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1917 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1919 /* The next available element. */
1920 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1922 /* Explicit quit checking is only used on NTemacs and whenever we
1923 use polling to process input events. */
1924 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1925 extern int immediate_quit
;
1926 # define IMMEDIATE_QUIT_CHECK \
1928 if (immediate_quit) QUIT; \
1931 # define IMMEDIATE_QUIT_CHECK ((void)0)
1934 /* Structure to manage work area for range table. */
1935 struct range_table_work_area
1937 int *table
; /* actual work area. */
1938 int allocated
; /* allocated size for work area in bytes. */
1939 int used
; /* actually used size in words. */
1940 int bits
; /* flag to record character classes */
1943 /* Make sure that WORK_AREA can hold more N multibyte characters.
1944 This is used only in set_image_of_range and set_image_of_range_1.
1945 It expects WORK_AREA to be a pointer.
1946 If it can't get the space, it returns from the surrounding function. */
1948 #define EXTEND_RANGE_TABLE(work_area, n) \
1950 if (((work_area)->used + (n)) * sizeof (int) > (work_area)->allocated) \
1952 extend_range_table_work_area (work_area); \
1953 if ((work_area)->table == 0) \
1954 return (REG_ESPACE); \
1958 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1959 (work_area).bits |= (bit)
1961 /* Bits used to implement the multibyte-part of the various character classes
1962 such as [:alnum:] in a charset's range table. */
1963 #define BIT_WORD 0x1
1964 #define BIT_LOWER 0x2
1965 #define BIT_PUNCT 0x4
1966 #define BIT_SPACE 0x8
1967 #define BIT_UPPER 0x10
1968 #define BIT_MULTIBYTE 0x20
1970 /* Set a range START..END to WORK_AREA.
1971 The range is passed through TRANSLATE, so START and END
1972 should be untranslated. */
1973 #define SET_RANGE_TABLE_WORK_AREA(work_area, start, end) \
1976 tem = set_image_of_range (&work_area, start, end, translate); \
1978 FREE_STACK_RETURN (tem); \
1981 /* Free allocated memory for WORK_AREA. */
1982 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1984 if ((work_area).table) \
1985 free ((work_area).table); \
1988 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1989 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1990 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1991 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1994 /* Set the bit for character C in a list. */
1995 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1998 /* Get the next unsigned number in the uncompiled pattern. */
1999 #define GET_UNSIGNED_NUMBER(num) \
2002 FREE_STACK_RETURN (REG_EBRACE); \
2006 while ('0' <= c && c <= '9') \
2012 num = num * 10 + c - '0'; \
2013 if (num / 10 != prev) \
2014 FREE_STACK_RETURN (REG_BADBR); \
2016 FREE_STACK_RETURN (REG_EBRACE); \
2022 #if ! WIDE_CHAR_SUPPORT
2024 /* Map a string to the char class it names (if any). */
2029 const char *string
= str
;
2030 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2031 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2032 else if (STREQ (string
, "word")) return RECC_WORD
;
2033 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2034 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2035 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2036 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2037 else if (STREQ (string
, "print")) return RECC_PRINT
;
2038 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2039 else if (STREQ (string
, "space")) return RECC_SPACE
;
2040 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2041 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2042 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2043 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2044 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2045 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2046 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2050 /* True iff CH is in the char class CC. */
2052 re_iswctype (ch
, cc
)
2058 case RECC_ALNUM
: return ISALNUM (ch
);
2059 case RECC_ALPHA
: return ISALPHA (ch
);
2060 case RECC_BLANK
: return ISBLANK (ch
);
2061 case RECC_CNTRL
: return ISCNTRL (ch
);
2062 case RECC_DIGIT
: return ISDIGIT (ch
);
2063 case RECC_GRAPH
: return ISGRAPH (ch
);
2064 case RECC_LOWER
: return ISLOWER (ch
);
2065 case RECC_PRINT
: return ISPRINT (ch
);
2066 case RECC_PUNCT
: return ISPUNCT (ch
);
2067 case RECC_SPACE
: return ISSPACE (ch
);
2068 case RECC_UPPER
: return ISUPPER (ch
);
2069 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2070 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2071 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2072 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2073 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2074 case RECC_WORD
: return ISWORD (ch
);
2075 case RECC_ERROR
: return false;
2081 /* Return a bit-pattern to use in the range-table bits to match multibyte
2082 chars of class CC. */
2084 re_wctype_to_bit (cc
)
2089 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2090 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2091 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2092 case RECC_LOWER
: return BIT_LOWER
;
2093 case RECC_UPPER
: return BIT_UPPER
;
2094 case RECC_PUNCT
: return BIT_PUNCT
;
2095 case RECC_SPACE
: return BIT_SPACE
;
2096 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2097 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2104 /* Filling in the work area of a range. */
2106 /* Actually extend the space in WORK_AREA. */
2109 extend_range_table_work_area (work_area
)
2110 struct range_table_work_area
*work_area
;
2112 work_area
->allocated
+= 16 * sizeof (int);
2113 if (work_area
->table
)
2115 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2118 = (int *) malloc (work_area
->allocated
);
2123 /* Carefully find the ranges of codes that are equivalent
2124 under case conversion to the range start..end when passed through
2125 TRANSLATE. Handle the case where non-letters can come in between
2126 two upper-case letters (which happens in Latin-1).
2127 Also handle the case of groups of more than 2 case-equivalent chars.
2129 The basic method is to look at consecutive characters and see
2130 if they can form a run that can be handled as one.
2132 Returns -1 if successful, REG_ESPACE if ran out of space. */
2135 set_image_of_range_1 (work_area
, start
, end
, translate
)
2136 RE_TRANSLATE_TYPE translate
;
2137 struct range_table_work_area
*work_area
;
2138 re_wchar_t start
, end
;
2140 /* `one_case' indicates a character, or a run of characters,
2141 each of which is an isolate (no case-equivalents).
2142 This includes all ASCII non-letters.
2144 `two_case' indicates a character, or a run of characters,
2145 each of which has two case-equivalent forms.
2146 This includes all ASCII letters.
2148 `strange' indicates a character that has more than one
2151 enum case_type
{one_case
, two_case
, strange
};
2153 /* Describe the run that is in progress,
2154 which the next character can try to extend.
2155 If run_type is strange, that means there really is no run.
2156 If run_type is one_case, then run_start...run_end is the run.
2157 If run_type is two_case, then the run is run_start...run_end,
2158 and the case-equivalents end at run_eqv_end. */
2160 enum case_type run_type
= strange
;
2161 int run_start
, run_end
, run_eqv_end
;
2163 Lisp_Object eqv_table
;
2165 if (!RE_TRANSLATE_P (translate
))
2167 EXTEND_RANGE_TABLE (work_area
, 2);
2168 work_area
->table
[work_area
->used
++] = (start
);
2169 work_area
->table
[work_area
->used
++] = (end
);
2173 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2175 for (; start
<= end
; start
++)
2177 enum case_type this_type
;
2178 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2179 int minchar
, maxchar
;
2181 /* Classify this character */
2183 this_type
= one_case
;
2184 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2185 this_type
= two_case
;
2187 this_type
= strange
;
2190 minchar
= start
, maxchar
= eqv
;
2192 minchar
= eqv
, maxchar
= start
;
2194 /* Can this character extend the run in progress? */
2195 if (this_type
== strange
|| this_type
!= run_type
2196 || !(minchar
== run_end
+ 1
2197 && (run_type
== two_case
2198 ? maxchar
== run_eqv_end
+ 1 : 1)))
2201 Record each of its equivalent ranges. */
2202 if (run_type
== one_case
)
2204 EXTEND_RANGE_TABLE (work_area
, 2);
2205 work_area
->table
[work_area
->used
++] = run_start
;
2206 work_area
->table
[work_area
->used
++] = run_end
;
2208 else if (run_type
== two_case
)
2210 EXTEND_RANGE_TABLE (work_area
, 4);
2211 work_area
->table
[work_area
->used
++] = run_start
;
2212 work_area
->table
[work_area
->used
++] = run_end
;
2213 work_area
->table
[work_area
->used
++]
2214 = RE_TRANSLATE (eqv_table
, run_start
);
2215 work_area
->table
[work_area
->used
++]
2216 = RE_TRANSLATE (eqv_table
, run_end
);
2221 if (this_type
== strange
)
2223 /* For a strange character, add each of its equivalents, one
2224 by one. Don't start a range. */
2227 EXTEND_RANGE_TABLE (work_area
, 2);
2228 work_area
->table
[work_area
->used
++] = eqv
;
2229 work_area
->table
[work_area
->used
++] = eqv
;
2230 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2232 while (eqv
!= start
);
2235 /* Add this char to the run, or start a new run. */
2236 else if (run_type
== strange
)
2238 /* Initialize a new range. */
2239 run_type
= this_type
;
2242 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2246 /* Extend a running range. */
2248 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2252 /* If a run is still in progress at the end, finish it now
2253 by recording its equivalent ranges. */
2254 if (run_type
== one_case
)
2256 EXTEND_RANGE_TABLE (work_area
, 2);
2257 work_area
->table
[work_area
->used
++] = run_start
;
2258 work_area
->table
[work_area
->used
++] = run_end
;
2260 else if (run_type
== two_case
)
2262 EXTEND_RANGE_TABLE (work_area
, 4);
2263 work_area
->table
[work_area
->used
++] = run_start
;
2264 work_area
->table
[work_area
->used
++] = run_end
;
2265 work_area
->table
[work_area
->used
++]
2266 = RE_TRANSLATE (eqv_table
, run_start
);
2267 work_area
->table
[work_area
->used
++]
2268 = RE_TRANSLATE (eqv_table
, run_end
);
2276 /* Record the the image of the range start..end when passed through
2277 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2278 and is not even necessarily contiguous.
2279 Normally we approximate it with the smallest contiguous range that contains
2280 all the chars we need. However, for Latin-1 we go to extra effort
2283 This function is not called for ASCII ranges.
2285 Returns -1 if successful, REG_ESPACE if ran out of space. */
2288 set_image_of_range (work_area
, start
, end
, translate
)
2289 RE_TRANSLATE_TYPE translate
;
2290 struct range_table_work_area
*work_area
;
2291 re_wchar_t start
, end
;
2293 re_wchar_t cmin
, cmax
;
2296 /* For Latin-1 ranges, use set_image_of_range_1
2297 to get proper handling of ranges that include letters and nonletters.
2298 For a range that includes the whole of Latin-1, this is not necessary.
2299 For other character sets, we don't bother to get this right. */
2300 if (RE_TRANSLATE_P (translate
) && start
< 04400
2301 && !(start
< 04200 && end
>= 04377))
2308 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2318 EXTEND_RANGE_TABLE (work_area
, 2);
2319 work_area
->table
[work_area
->used
++] = (start
);
2320 work_area
->table
[work_area
->used
++] = (end
);
2322 cmin
= -1, cmax
= -1;
2324 if (RE_TRANSLATE_P (translate
))
2328 for (ch
= start
; ch
<= end
; ch
++)
2330 re_wchar_t c
= TRANSLATE (ch
);
2331 if (! (start
<= c
&& c
<= end
))
2337 cmin
= MIN (cmin
, c
);
2338 cmax
= MAX (cmax
, c
);
2345 EXTEND_RANGE_TABLE (work_area
, 2);
2346 work_area
->table
[work_area
->used
++] = (cmin
);
2347 work_area
->table
[work_area
->used
++] = (cmax
);
2354 #ifndef MATCH_MAY_ALLOCATE
2356 /* If we cannot allocate large objects within re_match_2_internal,
2357 we make the fail stack and register vectors global.
2358 The fail stack, we grow to the maximum size when a regexp
2360 The register vectors, we adjust in size each time we
2361 compile a regexp, according to the number of registers it needs. */
2363 static fail_stack_type fail_stack
;
2365 /* Size with which the following vectors are currently allocated.
2366 That is so we can make them bigger as needed,
2367 but never make them smaller. */
2368 static int regs_allocated_size
;
2370 static re_char
** regstart
, ** regend
;
2371 static re_char
**best_regstart
, **best_regend
;
2373 /* Make the register vectors big enough for NUM_REGS registers,
2374 but don't make them smaller. */
2377 regex_grow_registers (num_regs
)
2380 if (num_regs
> regs_allocated_size
)
2382 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2383 RETALLOC_IF (regend
, num_regs
, re_char
*);
2384 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2385 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2387 regs_allocated_size
= num_regs
;
2391 #endif /* not MATCH_MAY_ALLOCATE */
2393 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2397 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2398 Returns one of error codes defined in `regex.h', or zero for success.
2400 Assumes the `allocated' (and perhaps `buffer') and `translate'
2401 fields are set in BUFP on entry.
2403 If it succeeds, results are put in BUFP (if it returns an error, the
2404 contents of BUFP are undefined):
2405 `buffer' is the compiled pattern;
2406 `syntax' is set to SYNTAX;
2407 `used' is set to the length of the compiled pattern;
2408 `fastmap_accurate' is zero;
2409 `re_nsub' is the number of subexpressions in PATTERN;
2410 `not_bol' and `not_eol' are zero;
2412 The `fastmap' field is neither examined nor set. */
2414 /* Insert the `jump' from the end of last alternative to "here".
2415 The space for the jump has already been allocated. */
2416 #define FIXUP_ALT_JUMP() \
2418 if (fixup_alt_jump) \
2419 STORE_JUMP (jump, fixup_alt_jump, b); \
2423 /* Return, freeing storage we allocated. */
2424 #define FREE_STACK_RETURN(value) \
2426 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2427 free (compile_stack.stack); \
2431 static reg_errcode_t
2432 regex_compile (pattern
, size
, syntax
, bufp
)
2435 reg_syntax_t syntax
;
2436 struct re_pattern_buffer
*bufp
;
2438 /* We fetch characters from PATTERN here. */
2439 register re_wchar_t c
, c1
;
2441 /* A random temporary spot in PATTERN. */
2444 /* Points to the end of the buffer, where we should append. */
2445 register unsigned char *b
;
2447 /* Keeps track of unclosed groups. */
2448 compile_stack_type compile_stack
;
2450 /* Points to the current (ending) position in the pattern. */
2452 /* `const' makes AIX compiler fail. */
2453 unsigned char *p
= pattern
;
2455 re_char
*p
= pattern
;
2457 re_char
*pend
= pattern
+ size
;
2459 /* How to translate the characters in the pattern. */
2460 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2462 /* Address of the count-byte of the most recently inserted `exactn'
2463 command. This makes it possible to tell if a new exact-match
2464 character can be added to that command or if the character requires
2465 a new `exactn' command. */
2466 unsigned char *pending_exact
= 0;
2468 /* Address of start of the most recently finished expression.
2469 This tells, e.g., postfix * where to find the start of its
2470 operand. Reset at the beginning of groups and alternatives. */
2471 unsigned char *laststart
= 0;
2473 /* Address of beginning of regexp, or inside of last group. */
2474 unsigned char *begalt
;
2476 /* Place in the uncompiled pattern (i.e., the {) to
2477 which to go back if the interval is invalid. */
2478 re_char
*beg_interval
;
2480 /* Address of the place where a forward jump should go to the end of
2481 the containing expression. Each alternative of an `or' -- except the
2482 last -- ends with a forward jump of this sort. */
2483 unsigned char *fixup_alt_jump
= 0;
2485 /* Work area for range table of charset. */
2486 struct range_table_work_area range_table_work
;
2488 /* If the object matched can contain multibyte characters. */
2489 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2491 /* Nonzero if we have pushed down into a subpattern. */
2492 int in_subpattern
= 0;
2494 /* These hold the values of p, pattern, and pend from the main
2495 pattern when we have pushed into a subpattern. */
2497 re_char
*main_pattern
;
2502 DEBUG_PRINT1 ("\nCompiling pattern: ");
2505 unsigned debug_count
;
2507 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2508 putchar (pattern
[debug_count
]);
2513 /* Initialize the compile stack. */
2514 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2515 if (compile_stack
.stack
== NULL
)
2518 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2519 compile_stack
.avail
= 0;
2521 range_table_work
.table
= 0;
2522 range_table_work
.allocated
= 0;
2524 /* Initialize the pattern buffer. */
2525 bufp
->syntax
= syntax
;
2526 bufp
->fastmap_accurate
= 0;
2527 bufp
->not_bol
= bufp
->not_eol
= 0;
2528 bufp
->used_syntax
= 0;
2530 /* Set `used' to zero, so that if we return an error, the pattern
2531 printer (for debugging) will think there's no pattern. We reset it
2535 /* Always count groups, whether or not bufp->no_sub is set. */
2538 #if !defined emacs && !defined SYNTAX_TABLE
2539 /* Initialize the syntax table. */
2540 init_syntax_once ();
2543 if (bufp
->allocated
== 0)
2546 { /* If zero allocated, but buffer is non-null, try to realloc
2547 enough space. This loses if buffer's address is bogus, but
2548 that is the user's responsibility. */
2549 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2552 { /* Caller did not allocate a buffer. Do it for them. */
2553 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2555 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2557 bufp
->allocated
= INIT_BUF_SIZE
;
2560 begalt
= b
= bufp
->buffer
;
2562 /* Loop through the uncompiled pattern until we're at the end. */
2567 /* If this is the end of an included regexp,
2568 pop back to the main regexp and try again. */
2572 pattern
= main_pattern
;
2577 /* If this is the end of the main regexp, we are done. */
2589 /* If there's no special whitespace regexp, treat
2590 spaces normally. And don't try to do this recursively. */
2591 if (!whitespace_regexp
|| in_subpattern
)
2594 /* Peek past following spaces. */
2601 /* If the spaces are followed by a repetition op,
2602 treat them normally. */
2604 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2605 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2608 /* Replace the spaces with the whitespace regexp. */
2612 main_pattern
= pattern
;
2613 p
= pattern
= whitespace_regexp
;
2614 pend
= p
+ strlen (p
);
2620 if ( /* If at start of pattern, it's an operator. */
2622 /* If context independent, it's an operator. */
2623 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2624 /* Otherwise, depends on what's come before. */
2625 || at_begline_loc_p (pattern
, p
, syntax
))
2626 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2635 if ( /* If at end of pattern, it's an operator. */
2637 /* If context independent, it's an operator. */
2638 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2639 /* Otherwise, depends on what's next. */
2640 || at_endline_loc_p (p
, pend
, syntax
))
2641 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2650 if ((syntax
& RE_BK_PLUS_QM
)
2651 || (syntax
& RE_LIMITED_OPS
))
2655 /* If there is no previous pattern... */
2658 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2659 FREE_STACK_RETURN (REG_BADRPT
);
2660 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2665 /* 1 means zero (many) matches is allowed. */
2666 boolean zero_times_ok
= 0, many_times_ok
= 0;
2669 /* If there is a sequence of repetition chars, collapse it
2670 down to just one (the right one). We can't combine
2671 interval operators with these because of, e.g., `a{2}*',
2672 which should only match an even number of `a's. */
2676 if ((syntax
& RE_FRUGAL
)
2677 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2681 zero_times_ok
|= c
!= '+';
2682 many_times_ok
|= c
!= '?';
2688 || (!(syntax
& RE_BK_PLUS_QM
)
2689 && (*p
== '+' || *p
== '?')))
2691 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2694 FREE_STACK_RETURN (REG_EESCAPE
);
2695 if (p
[1] == '+' || p
[1] == '?')
2696 PATFETCH (c
); /* Gobble up the backslash. */
2702 /* If we get here, we found another repeat character. */
2706 /* Star, etc. applied to an empty pattern is equivalent
2707 to an empty pattern. */
2708 if (!laststart
|| laststart
== b
)
2711 /* Now we know whether or not zero matches is allowed
2712 and also whether or not two or more matches is allowed. */
2717 boolean simple
= skip_one_char (laststart
) == b
;
2718 unsigned int startoffset
= 0;
2720 /* Check if the loop can match the empty string. */
2721 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2722 ? on_failure_jump
: on_failure_jump_loop
;
2723 assert (skip_one_char (laststart
) <= b
);
2725 if (!zero_times_ok
&& simple
)
2726 { /* Since simple * loops can be made faster by using
2727 on_failure_keep_string_jump, we turn simple P+
2728 into PP* if P is simple. */
2729 unsigned char *p1
, *p2
;
2730 startoffset
= b
- laststart
;
2731 GET_BUFFER_SPACE (startoffset
);
2732 p1
= b
; p2
= laststart
;
2738 GET_BUFFER_SPACE (6);
2741 STORE_JUMP (ofj
, b
, b
+ 6);
2743 /* Simple * loops can use on_failure_keep_string_jump
2744 depending on what follows. But since we don't know
2745 that yet, we leave the decision up to
2746 on_failure_jump_smart. */
2747 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2748 laststart
+ startoffset
, b
+ 6);
2750 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2755 /* A simple ? pattern. */
2756 assert (zero_times_ok
);
2757 GET_BUFFER_SPACE (3);
2758 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2762 else /* not greedy */
2763 { /* I wish the greedy and non-greedy cases could be merged. */
2765 GET_BUFFER_SPACE (7); /* We might use less. */
2768 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2770 /* The non-greedy multiple match looks like
2771 a repeat..until: we only need a conditional jump
2772 at the end of the loop. */
2773 if (emptyp
) BUF_PUSH (no_op
);
2774 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2775 : on_failure_jump
, b
, laststart
);
2779 /* The repeat...until naturally matches one or more.
2780 To also match zero times, we need to first jump to
2781 the end of the loop (its conditional jump). */
2782 INSERT_JUMP (jump
, laststart
, b
);
2788 /* non-greedy a?? */
2789 INSERT_JUMP (jump
, laststart
, b
+ 3);
2791 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2808 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2810 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2812 /* Ensure that we have enough space to push a charset: the
2813 opcode, the length count, and the bitset; 34 bytes in all. */
2814 GET_BUFFER_SPACE (34);
2818 /* We test `*p == '^' twice, instead of using an if
2819 statement, so we only need one BUF_PUSH. */
2820 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2824 /* Remember the first position in the bracket expression. */
2827 /* Push the number of bytes in the bitmap. */
2828 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2830 /* Clear the whole map. */
2831 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2833 /* charset_not matches newline according to a syntax bit. */
2834 if ((re_opcode_t
) b
[-2] == charset_not
2835 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2836 SET_LIST_BIT ('\n');
2838 /* Read in characters and ranges, setting map bits. */
2841 boolean escaped_char
= false;
2842 const unsigned char *p2
= p
;
2844 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2846 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2847 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2848 So the translation is done later in a loop. Example:
2849 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2852 /* \ might escape characters inside [...] and [^...]. */
2853 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2855 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2858 escaped_char
= true;
2862 /* Could be the end of the bracket expression. If it's
2863 not (i.e., when the bracket expression is `[]' so
2864 far), the ']' character bit gets set way below. */
2865 if (c
== ']' && p2
!= p1
)
2869 /* What should we do for the character which is
2870 greater than 0x7F, but not BASE_LEADING_CODE_P?
2873 /* See if we're at the beginning of a possible character
2876 if (!escaped_char
&&
2877 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2879 /* Leave room for the null. */
2880 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2881 const unsigned char *class_beg
;
2887 /* If pattern is `[[:'. */
2888 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2893 if ((c
== ':' && *p
== ']') || p
== pend
)
2895 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2898 /* This is in any case an invalid class name. */
2903 /* If isn't a word bracketed by `[:' and `:]':
2904 undo the ending character, the letters, and
2905 leave the leading `:' and `[' (but set bits for
2907 if (c
== ':' && *p
== ']')
2912 cc
= re_wctype (str
);
2915 FREE_STACK_RETURN (REG_ECTYPE
);
2917 /* Throw away the ] at the end of the character
2921 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2923 /* Most character classes in a multibyte match
2924 just set a flag. Exceptions are is_blank,
2925 is_digit, is_cntrl, and is_xdigit, since
2926 they can only match ASCII characters. We
2927 don't need to handle them for multibyte.
2928 They are distinguished by a negative wctype. */
2931 SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work
,
2932 re_wctype_to_bit (cc
));
2934 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2936 int translated
= TRANSLATE (ch
);
2937 if (translated
< (1 << BYTEWIDTH
)
2938 && re_iswctype (btowc (ch
), cc
))
2939 SET_LIST_BIT (translated
);
2942 /* In most cases the matching rule for char classes
2943 only uses the syntax table for multibyte chars,
2944 so that the content of the syntax-table it is not
2945 hardcoded in the range_table. SPACE and WORD are
2946 the two exceptions. */
2947 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2948 bufp
->used_syntax
= 1;
2950 /* Repeat the loop. */
2955 /* Go back to right after the "[:". */
2959 /* Because the `:' may starts the range, we
2960 can't simply set bit and repeat the loop.
2961 Instead, just set it to C and handle below. */
2966 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2969 /* Discard the `-'. */
2972 /* Fetch the character which ends the range. */
2975 if (SINGLE_BYTE_CHAR_P (c
))
2977 if (! SINGLE_BYTE_CHAR_P (c1
))
2979 /* Handle a range starting with a
2980 character of less than 256, and ending
2981 with a character of not less than 256.
2982 Split that into two ranges, the low one
2983 ending at 0377, and the high one
2984 starting at the smallest character in
2985 the charset of C1 and ending at C1. */
2986 int charset
= CHAR_CHARSET (c1
);
2987 re_wchar_t c2
= MAKE_CHAR (charset
, 0, 0);
2989 SET_RANGE_TABLE_WORK_AREA (range_table_work
,
2994 else if (!SAME_CHARSET_P (c
, c1
))
2995 FREE_STACK_RETURN (REG_ERANGEX
);
2998 /* Range from C to C. */
3001 /* Set the range ... */
3002 if (SINGLE_BYTE_CHAR_P (c
))
3003 /* ... into bitmap. */
3005 re_wchar_t this_char
;
3006 re_wchar_t range_start
= c
, range_end
= c1
;
3008 /* If the start is after the end, the range is empty. */
3009 if (range_start
> range_end
)
3011 if (syntax
& RE_NO_EMPTY_RANGES
)
3012 FREE_STACK_RETURN (REG_ERANGE
);
3013 /* Else, repeat the loop. */
3017 for (this_char
= range_start
; this_char
<= range_end
;
3020 int translated
= TRANSLATE (this_char
);
3021 if (translated
< (1 << BYTEWIDTH
))
3022 SET_LIST_BIT (translated
);
3024 SET_RANGE_TABLE_WORK_AREA
3025 (range_table_work
, translated
, translated
);
3030 /* ... into range table. */
3031 SET_RANGE_TABLE_WORK_AREA (range_table_work
, c
, c1
);
3034 /* Discard any (non)matching list bytes that are all 0 at the
3035 end of the map. Decrease the map-length byte too. */
3036 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3040 /* Build real range table from work area. */
3041 if (RANGE_TABLE_WORK_USED (range_table_work
)
3042 || RANGE_TABLE_WORK_BITS (range_table_work
))
3045 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3047 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3048 bytes for flags, two for COUNT, and three bytes for
3050 GET_BUFFER_SPACE (4 + used
* 3);
3052 /* Indicate the existence of range table. */
3053 laststart
[1] |= 0x80;
3055 /* Store the character class flag bits into the range table.
3056 If not in emacs, these flag bits are always 0. */
3057 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3058 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3060 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3061 for (i
= 0; i
< used
; i
++)
3062 STORE_CHARACTER_AND_INCR
3063 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3070 if (syntax
& RE_NO_BK_PARENS
)
3077 if (syntax
& RE_NO_BK_PARENS
)
3084 if (syntax
& RE_NEWLINE_ALT
)
3091 if (syntax
& RE_NO_BK_VBAR
)
3098 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3099 goto handle_interval
;
3105 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3107 /* Do not translate the character after the \, so that we can
3108 distinguish, e.g., \B from \b, even if we normally would
3109 translate, e.g., B to b. */
3115 if (syntax
& RE_NO_BK_PARENS
)
3116 goto normal_backslash
;
3121 regnum_t regnum
= 0;
3124 /* Look for a special (?...) construct */
3125 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3127 PATFETCH (c
); /* Gobble up the '?'. */
3133 case ':': shy
= 1; break;
3135 /* An explicitly specified regnum must start
3138 FREE_STACK_RETURN (REG_BADPAT
);
3139 case '1': case '2': case '3': case '4':
3140 case '5': case '6': case '7': case '8': case '9':
3141 regnum
= 10*regnum
+ (c
- '0'); break;
3143 /* Only (?:...) is supported right now. */
3144 FREE_STACK_RETURN (REG_BADPAT
);
3151 regnum
= ++bufp
->re_nsub
;
3153 { /* It's actually not shy, but explicitly numbered. */
3155 if (regnum
> bufp
->re_nsub
)
3156 bufp
->re_nsub
= regnum
;
3157 else if (regnum
> bufp
->re_nsub
3158 /* Ideally, we'd want to check that the specified
3159 group can't have matched (i.e. all subgroups
3160 using the same regnum are in other branches of
3161 OR patterns), but we don't currently keep track
3162 of enough info to do that easily. */
3163 || group_in_compile_stack (compile_stack
, regnum
))
3164 FREE_STACK_RETURN (REG_BADPAT
);
3167 /* It's really shy. */
3168 regnum
= - bufp
->re_nsub
;
3170 if (COMPILE_STACK_FULL
)
3172 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3173 compile_stack_elt_t
);
3174 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3176 compile_stack
.size
<<= 1;
3179 /* These are the values to restore when we hit end of this
3180 group. They are all relative offsets, so that if the
3181 whole pattern moves because of realloc, they will still
3183 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3184 COMPILE_STACK_TOP
.fixup_alt_jump
3185 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3186 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3187 COMPILE_STACK_TOP
.regnum
= regnum
;
3189 /* Do not push a start_memory for groups beyond the last one
3190 we can represent in the compiled pattern. */
3191 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3192 BUF_PUSH_2 (start_memory
, regnum
);
3194 compile_stack
.avail
++;
3199 /* If we've reached MAX_REGNUM groups, then this open
3200 won't actually generate any code, so we'll have to
3201 clear pending_exact explicitly. */
3207 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3209 if (COMPILE_STACK_EMPTY
)
3211 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3212 goto normal_backslash
;
3214 FREE_STACK_RETURN (REG_ERPAREN
);
3220 /* See similar code for backslashed left paren above. */
3221 if (COMPILE_STACK_EMPTY
)
3223 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3226 FREE_STACK_RETURN (REG_ERPAREN
);
3229 /* Since we just checked for an empty stack above, this
3230 ``can't happen''. */
3231 assert (compile_stack
.avail
!= 0);
3233 /* We don't just want to restore into `regnum', because
3234 later groups should continue to be numbered higher,
3235 as in `(ab)c(de)' -- the second group is #2. */
3238 compile_stack
.avail
--;
3239 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3241 = COMPILE_STACK_TOP
.fixup_alt_jump
3242 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3244 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3245 regnum
= COMPILE_STACK_TOP
.regnum
;
3246 /* If we've reached MAX_REGNUM groups, then this open
3247 won't actually generate any code, so we'll have to
3248 clear pending_exact explicitly. */
3251 /* We're at the end of the group, so now we know how many
3252 groups were inside this one. */
3253 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3254 BUF_PUSH_2 (stop_memory
, regnum
);
3259 case '|': /* `\|'. */
3260 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3261 goto normal_backslash
;
3263 if (syntax
& RE_LIMITED_OPS
)
3266 /* Insert before the previous alternative a jump which
3267 jumps to this alternative if the former fails. */
3268 GET_BUFFER_SPACE (3);
3269 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3273 /* The alternative before this one has a jump after it
3274 which gets executed if it gets matched. Adjust that
3275 jump so it will jump to this alternative's analogous
3276 jump (put in below, which in turn will jump to the next
3277 (if any) alternative's such jump, etc.). The last such
3278 jump jumps to the correct final destination. A picture:
3284 If we are at `b', then fixup_alt_jump right now points to a
3285 three-byte space after `a'. We'll put in the jump, set
3286 fixup_alt_jump to right after `b', and leave behind three
3287 bytes which we'll fill in when we get to after `c'. */
3291 /* Mark and leave space for a jump after this alternative,
3292 to be filled in later either by next alternative or
3293 when know we're at the end of a series of alternatives. */
3295 GET_BUFFER_SPACE (3);
3304 /* If \{ is a literal. */
3305 if (!(syntax
& RE_INTERVALS
)
3306 /* If we're at `\{' and it's not the open-interval
3308 || (syntax
& RE_NO_BK_BRACES
))
3309 goto normal_backslash
;
3313 /* If got here, then the syntax allows intervals. */
3315 /* At least (most) this many matches must be made. */
3316 int lower_bound
= 0, upper_bound
= -1;
3320 GET_UNSIGNED_NUMBER (lower_bound
);
3323 GET_UNSIGNED_NUMBER (upper_bound
);
3325 /* Interval such as `{1}' => match exactly once. */
3326 upper_bound
= lower_bound
;
3328 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3329 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3330 FREE_STACK_RETURN (REG_BADBR
);
3332 if (!(syntax
& RE_NO_BK_BRACES
))
3335 FREE_STACK_RETURN (REG_BADBR
);
3337 FREE_STACK_RETURN (REG_EESCAPE
);
3342 FREE_STACK_RETURN (REG_BADBR
);
3344 /* We just parsed a valid interval. */
3346 /* If it's invalid to have no preceding re. */
3349 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3350 FREE_STACK_RETURN (REG_BADRPT
);
3351 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3354 goto unfetch_interval
;
3357 if (upper_bound
== 0)
3358 /* If the upper bound is zero, just drop the sub pattern
3361 else if (lower_bound
== 1 && upper_bound
== 1)
3362 /* Just match it once: nothing to do here. */
3365 /* Otherwise, we have a nontrivial interval. When
3366 we're all done, the pattern will look like:
3367 set_number_at <jump count> <upper bound>
3368 set_number_at <succeed_n count> <lower bound>
3369 succeed_n <after jump addr> <succeed_n count>
3371 jump_n <succeed_n addr> <jump count>
3372 (The upper bound and `jump_n' are omitted if
3373 `upper_bound' is 1, though.) */
3375 { /* If the upper bound is > 1, we need to insert
3376 more at the end of the loop. */
3377 unsigned int nbytes
= (upper_bound
< 0 ? 3
3378 : upper_bound
> 1 ? 5 : 0);
3379 unsigned int startoffset
= 0;
3381 GET_BUFFER_SPACE (20); /* We might use less. */
3383 if (lower_bound
== 0)
3385 /* A succeed_n that starts with 0 is really a
3386 a simple on_failure_jump_loop. */
3387 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3393 /* Initialize lower bound of the `succeed_n', even
3394 though it will be set during matching by its
3395 attendant `set_number_at' (inserted next),
3396 because `re_compile_fastmap' needs to know.
3397 Jump to the `jump_n' we might insert below. */
3398 INSERT_JUMP2 (succeed_n
, laststart
,
3403 /* Code to initialize the lower bound. Insert
3404 before the `succeed_n'. The `5' is the last two
3405 bytes of this `set_number_at', plus 3 bytes of
3406 the following `succeed_n'. */
3407 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3412 if (upper_bound
< 0)
3414 /* A negative upper bound stands for infinity,
3415 in which case it degenerates to a plain jump. */
3416 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3419 else if (upper_bound
> 1)
3420 { /* More than one repetition is allowed, so
3421 append a backward jump to the `succeed_n'
3422 that starts this interval.
3424 When we've reached this during matching,
3425 we'll have matched the interval once, so
3426 jump back only `upper_bound - 1' times. */
3427 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3431 /* The location we want to set is the second
3432 parameter of the `jump_n'; that is `b-2' as
3433 an absolute address. `laststart' will be
3434 the `set_number_at' we're about to insert;
3435 `laststart+3' the number to set, the source
3436 for the relative address. But we are
3437 inserting into the middle of the pattern --
3438 so everything is getting moved up by 5.
3439 Conclusion: (b - 2) - (laststart + 3) + 5,
3440 i.e., b - laststart.
3442 We insert this at the beginning of the loop
3443 so that if we fail during matching, we'll
3444 reinitialize the bounds. */
3445 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3446 upper_bound
- 1, b
);
3451 beg_interval
= NULL
;
3456 /* If an invalid interval, match the characters as literals. */
3457 assert (beg_interval
);
3459 beg_interval
= NULL
;
3461 /* normal_char and normal_backslash need `c'. */
3464 if (!(syntax
& RE_NO_BK_BRACES
))
3466 assert (p
> pattern
&& p
[-1] == '\\');
3467 goto normal_backslash
;
3473 /* There is no way to specify the before_dot and after_dot
3474 operators. rms says this is ok. --karl */
3482 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3488 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3494 BUF_PUSH_2 (categoryspec
, c
);
3500 BUF_PUSH_2 (notcategoryspec
, c
);
3506 if (syntax
& RE_NO_GNU_OPS
)
3509 BUF_PUSH_2 (syntaxspec
, Sword
);
3514 if (syntax
& RE_NO_GNU_OPS
)
3517 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3522 if (syntax
& RE_NO_GNU_OPS
)
3528 if (syntax
& RE_NO_GNU_OPS
)
3534 if (syntax
& RE_NO_GNU_OPS
)
3543 FREE_STACK_RETURN (REG_BADPAT
);
3547 if (syntax
& RE_NO_GNU_OPS
)
3549 BUF_PUSH (wordbound
);
3553 if (syntax
& RE_NO_GNU_OPS
)
3555 BUF_PUSH (notwordbound
);
3559 if (syntax
& RE_NO_GNU_OPS
)
3565 if (syntax
& RE_NO_GNU_OPS
)
3570 case '1': case '2': case '3': case '4': case '5':
3571 case '6': case '7': case '8': case '9':
3575 if (syntax
& RE_NO_BK_REFS
)
3576 goto normal_backslash
;
3580 if (reg
> bufp
->re_nsub
|| reg
< 1
3581 /* Can't back reference to a subexp before its end. */
3582 || group_in_compile_stack (compile_stack
, reg
))
3583 FREE_STACK_RETURN (REG_ESUBREG
);
3586 BUF_PUSH_2 (duplicate
, reg
);
3593 if (syntax
& RE_BK_PLUS_QM
)
3596 goto normal_backslash
;
3600 /* You might think it would be useful for \ to mean
3601 not to translate; but if we don't translate it
3602 it will never match anything. */
3609 /* Expects the character in `c'. */
3611 /* If no exactn currently being built. */
3614 /* If last exactn not at current position. */
3615 || pending_exact
+ *pending_exact
+ 1 != b
3617 /* We have only one byte following the exactn for the count. */
3618 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3620 /* If followed by a repetition operator. */
3621 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3622 || ((syntax
& RE_BK_PLUS_QM
)
3623 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3624 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3625 || ((syntax
& RE_INTERVALS
)
3626 && ((syntax
& RE_NO_BK_BRACES
)
3627 ? p
!= pend
&& *p
== '{'
3628 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3630 /* Start building a new exactn. */
3634 BUF_PUSH_2 (exactn
, 0);
3635 pending_exact
= b
- 1;
3638 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3644 len
= CHAR_STRING (c
, b
);
3648 (*pending_exact
) += len
;
3653 } /* while p != pend */
3656 /* Through the pattern now. */
3660 if (!COMPILE_STACK_EMPTY
)
3661 FREE_STACK_RETURN (REG_EPAREN
);
3663 /* If we don't want backtracking, force success
3664 the first time we reach the end of the compiled pattern. */
3665 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3668 /* We have succeeded; set the length of the buffer. */
3669 bufp
->used
= b
- bufp
->buffer
;
3674 re_compile_fastmap (bufp
);
3675 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3676 print_compiled_pattern (bufp
);
3681 #ifndef MATCH_MAY_ALLOCATE
3682 /* Initialize the failure stack to the largest possible stack. This
3683 isn't necessary unless we're trying to avoid calling alloca in
3684 the search and match routines. */
3686 int num_regs
= bufp
->re_nsub
+ 1;
3688 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3690 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3692 if (! fail_stack
.stack
)
3694 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3695 * sizeof (fail_stack_elt_t
));
3698 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3700 * sizeof (fail_stack_elt_t
)));
3703 regex_grow_registers (num_regs
);
3705 #endif /* not MATCH_MAY_ALLOCATE */
3707 FREE_STACK_RETURN (REG_NOERROR
);
3708 } /* regex_compile */
3710 /* Subroutines for `regex_compile'. */
3712 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3715 store_op1 (op
, loc
, arg
)
3720 *loc
= (unsigned char) op
;
3721 STORE_NUMBER (loc
+ 1, arg
);
3725 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3728 store_op2 (op
, loc
, arg1
, arg2
)
3733 *loc
= (unsigned char) op
;
3734 STORE_NUMBER (loc
+ 1, arg1
);
3735 STORE_NUMBER (loc
+ 3, arg2
);
3739 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3740 for OP followed by two-byte integer parameter ARG. */
3743 insert_op1 (op
, loc
, arg
, end
)
3749 register unsigned char *pfrom
= end
;
3750 register unsigned char *pto
= end
+ 3;
3752 while (pfrom
!= loc
)
3755 store_op1 (op
, loc
, arg
);
3759 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3762 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3768 register unsigned char *pfrom
= end
;
3769 register unsigned char *pto
= end
+ 5;
3771 while (pfrom
!= loc
)
3774 store_op2 (op
, loc
, arg1
, arg2
);
3778 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3779 after an alternative or a begin-subexpression. We assume there is at
3780 least one character before the ^. */
3783 at_begline_loc_p (pattern
, p
, syntax
)
3784 re_char
*pattern
, *p
;
3785 reg_syntax_t syntax
;
3787 re_char
*prev
= p
- 2;
3788 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3791 /* After a subexpression? */
3792 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3793 /* After an alternative? */
3794 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3795 /* After a shy subexpression? */
3796 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3797 && prev
[-1] == '?' && prev
[-2] == '('
3798 && (syntax
& RE_NO_BK_PARENS
3799 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3803 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3804 at least one character after the $, i.e., `P < PEND'. */
3807 at_endline_loc_p (p
, pend
, syntax
)
3809 reg_syntax_t syntax
;
3812 boolean next_backslash
= *next
== '\\';
3813 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3816 /* Before a subexpression? */
3817 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3818 : next_backslash
&& next_next
&& *next_next
== ')')
3819 /* Before an alternative? */
3820 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3821 : next_backslash
&& next_next
&& *next_next
== '|');
3825 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3826 false if it's not. */
3829 group_in_compile_stack (compile_stack
, regnum
)
3830 compile_stack_type compile_stack
;
3835 for (this_element
= compile_stack
.avail
- 1;
3838 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3845 If fastmap is non-NULL, go through the pattern and fill fastmap
3846 with all the possible leading chars. If fastmap is NULL, don't
3847 bother filling it up (obviously) and only return whether the
3848 pattern could potentially match the empty string.
3850 Return 1 if p..pend might match the empty string.
3851 Return 0 if p..pend matches at least one char.
3852 Return -1 if fastmap was not updated accurately. */
3855 analyse_first (p
, pend
, fastmap
, multibyte
)
3858 const int multibyte
;
3863 /* If all elements for base leading-codes in fastmap is set, this
3864 flag is set true. */
3865 boolean match_any_multibyte_characters
= false;
3869 /* The loop below works as follows:
3870 - It has a working-list kept in the PATTERN_STACK and which basically
3871 starts by only containing a pointer to the first operation.
3872 - If the opcode we're looking at is a match against some set of
3873 chars, then we add those chars to the fastmap and go on to the
3874 next work element from the worklist (done via `break').
3875 - If the opcode is a control operator on the other hand, we either
3876 ignore it (if it's meaningless at this point, such as `start_memory')
3877 or execute it (if it's a jump). If the jump has several destinations
3878 (i.e. `on_failure_jump'), then we push the other destination onto the
3880 We guarantee termination by ignoring backward jumps (more or less),
3881 so that `p' is monotonically increasing. More to the point, we
3882 never set `p' (or push) anything `<= p1'. */
3886 /* `p1' is used as a marker of how far back a `on_failure_jump'
3887 can go without being ignored. It is normally equal to `p'
3888 (which prevents any backward `on_failure_jump') except right
3889 after a plain `jump', to allow patterns such as:
3892 10: on_failure_jump 3
3893 as used for the *? operator. */
3896 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3903 /* If the first character has to match a backreference, that means
3904 that the group was empty (since it already matched). Since this
3905 is the only case that interests us here, we can assume that the
3906 backreference must match the empty string. */
3911 /* Following are the cases which match a character. These end
3917 int c
= RE_STRING_CHAR (p
+ 1, pend
- p
);
3918 /* When fast-scanning, the fastmap can be indexed either with
3919 a char (smaller than 256) or with the first byte of
3920 a char's byte sequence. So we have to conservatively add
3921 both to the table. */
3922 if (SINGLE_BYTE_CHAR_P (c
))
3930 /* We could put all the chars except for \n (and maybe \0)
3931 but we don't bother since it is generally not worth it. */
3932 if (!fastmap
) break;
3937 /* Chars beyond end of bitmap are possible matches.
3938 All the single-byte codes can occur in multibyte buffers.
3939 So any that are not listed in the charset
3940 are possible matches, even in multibyte buffers. */
3941 if (!fastmap
) break;
3942 /* We don't need to mark LEADING_CODE_8_BIT_CONTROL specially
3943 because it will automatically be set when needed by virtue of
3944 being larger than the highest char of its charset (0xbf) but
3945 smaller than (1<<BYTEWIDTH). */
3946 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3947 j
< (1 << BYTEWIDTH
); j
++)
3951 if (!fastmap
) break;
3952 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3953 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3955 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3959 if (j
>= 0x80 && j
< 0xa0)
3960 fastmap
[LEADING_CODE_8_BIT_CONTROL
] = 1;
3964 if ((not && multibyte
)
3965 /* Any character set can possibly contain a character
3966 which doesn't match the specified set of characters. */
3967 || (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3968 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3969 /* If we can match a character class, we can match
3970 any character set. */
3972 set_fastmap_for_multibyte_characters
:
3973 if (match_any_multibyte_characters
== false)
3975 for (j
= 0x80; j
< 0xA0; j
++) /* XXX */
3976 if (BASE_LEADING_CODE_P (j
))
3978 match_any_multibyte_characters
= true;
3982 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3983 && match_any_multibyte_characters
== false)
3985 /* Set fastmap[I] 1 where I is a base leading code of each
3986 multibyte character in the range table. */
3989 /* Make P points the range table. `+ 2' is to skip flag
3990 bits for a character class. */
3991 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3993 /* Extract the number of ranges in range table into COUNT. */
3994 EXTRACT_NUMBER_AND_INCR (count
, p
);
3995 for (; count
> 0; count
--, p
+= 2 * 3) /* XXX */
3997 /* Extract the start of each range. */
3998 EXTRACT_CHARACTER (c
, p
);
3999 j
= CHAR_CHARSET (c
);
4000 fastmap
[CHARSET_LEADING_CODE_BASE (j
)] = 1;
4007 if (!fastmap
) break;
4009 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4011 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4012 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4016 /* This match depends on text properties. These end with
4017 aborting optimizations. */
4021 case notcategoryspec
:
4022 if (!fastmap
) break;
4023 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4025 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4026 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4030 /* Any character set can possibly contain a character
4031 whose category is K (or not). */
4032 goto set_fastmap_for_multibyte_characters
;
4035 /* All cases after this match the empty string. These end with
4057 EXTRACT_NUMBER_AND_INCR (j
, p
);
4059 /* Backward jumps can only go back to code that we've already
4060 visited. `re_compile' should make sure this is true. */
4063 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4065 case on_failure_jump
:
4066 case on_failure_keep_string_jump
:
4067 case on_failure_jump_loop
:
4068 case on_failure_jump_nastyloop
:
4069 case on_failure_jump_smart
:
4075 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4076 to jump back to "just after here". */
4079 case on_failure_jump
:
4080 case on_failure_keep_string_jump
:
4081 case on_failure_jump_nastyloop
:
4082 case on_failure_jump_loop
:
4083 case on_failure_jump_smart
:
4084 EXTRACT_NUMBER_AND_INCR (j
, p
);
4086 ; /* Backward jump to be ignored. */
4088 { /* We have to look down both arms.
4089 We first go down the "straight" path so as to minimize
4090 stack usage when going through alternatives. */
4091 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4099 /* This code simply does not properly handle forward jump_n. */
4100 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4102 /* jump_n can either jump or fall through. The (backward) jump
4103 case has already been handled, so we only need to look at the
4104 fallthrough case. */
4108 /* If N == 0, it should be an on_failure_jump_loop instead. */
4109 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4111 /* We only care about one iteration of the loop, so we don't
4112 need to consider the case where this behaves like an
4129 abort (); /* We have listed all the cases. */
4132 /* Getting here means we have found the possible starting
4133 characters for one path of the pattern -- and that the empty
4134 string does not match. We need not follow this path further. */
4138 /* We reached the end without matching anything. */
4141 } /* analyse_first */
4143 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4144 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4145 characters can start a string that matches the pattern. This fastmap
4146 is used by re_search to skip quickly over impossible starting points.
4148 Character codes above (1 << BYTEWIDTH) are not represented in the
4149 fastmap, but the leading codes are represented. Thus, the fastmap
4150 indicates which character sets could start a match.
4152 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4153 area as BUFP->fastmap.
4155 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4158 Returns 0 if we succeed, -2 if an internal error. */
4161 re_compile_fastmap (bufp
)
4162 struct re_pattern_buffer
*bufp
;
4164 char *fastmap
= bufp
->fastmap
;
4167 assert (fastmap
&& bufp
->buffer
);
4169 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4170 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4172 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4173 fastmap
, RE_MULTIBYTE_P (bufp
));
4174 bufp
->can_be_null
= (analysis
!= 0);
4176 } /* re_compile_fastmap */
4178 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4179 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4180 this memory for recording register information. STARTS and ENDS
4181 must be allocated using the malloc library routine, and must each
4182 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4184 If NUM_REGS == 0, then subsequent matches should allocate their own
4187 Unless this function is called, the first search or match using
4188 PATTERN_BUFFER will allocate its own register data, without
4189 freeing the old data. */
4192 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4193 struct re_pattern_buffer
*bufp
;
4194 struct re_registers
*regs
;
4196 regoff_t
*starts
, *ends
;
4200 bufp
->regs_allocated
= REGS_REALLOCATE
;
4201 regs
->num_regs
= num_regs
;
4202 regs
->start
= starts
;
4207 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4209 regs
->start
= regs
->end
= (regoff_t
*) 0;
4212 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4214 /* Searching routines. */
4216 /* Like re_search_2, below, but only one string is specified, and
4217 doesn't let you say where to stop matching. */
4220 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4221 struct re_pattern_buffer
*bufp
;
4223 int size
, startpos
, range
;
4224 struct re_registers
*regs
;
4226 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4229 WEAK_ALIAS (__re_search
, re_search
)
4231 /* Head address of virtual concatenation of string. */
4232 #define HEAD_ADDR_VSTRING(P) \
4233 (((P) >= size1 ? string2 : string1))
4235 /* End address of virtual concatenation of string. */
4236 #define STOP_ADDR_VSTRING(P) \
4237 (((P) >= size1 ? string2 + size2 : string1 + size1))
4239 /* Address of POS in the concatenation of virtual string. */
4240 #define POS_ADDR_VSTRING(POS) \
4241 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4243 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4244 virtual concatenation of STRING1 and STRING2, starting first at index
4245 STARTPOS, then at STARTPOS + 1, and so on.
4247 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4249 RANGE is how far to scan while trying to match. RANGE = 0 means try
4250 only at STARTPOS; in general, the last start tried is STARTPOS +
4253 In REGS, return the indices of the virtual concatenation of STRING1
4254 and STRING2 that matched the entire BUFP->buffer and its contained
4257 Do not consider matching one past the index STOP in the virtual
4258 concatenation of STRING1 and STRING2.
4260 We return either the position in the strings at which the match was
4261 found, -1 if no match, or -2 if error (such as failure
4265 re_search_2 (bufp
, str1
, size1
, str2
, size2
, startpos
, range
, regs
, stop
)
4266 struct re_pattern_buffer
*bufp
;
4267 const char *str1
, *str2
;
4271 struct re_registers
*regs
;
4275 re_char
*string1
= (re_char
*) str1
;
4276 re_char
*string2
= (re_char
*) str2
;
4277 register char *fastmap
= bufp
->fastmap
;
4278 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4279 int total_size
= size1
+ size2
;
4280 int endpos
= startpos
+ range
;
4281 boolean anchored_start
;
4283 /* Nonzero if we have to concern multibyte character. */
4284 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4286 /* Check for out-of-range STARTPOS. */
4287 if (startpos
< 0 || startpos
> total_size
)
4290 /* Fix up RANGE if it might eventually take us outside
4291 the virtual concatenation of STRING1 and STRING2.
4292 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4294 range
= 0 - startpos
;
4295 else if (endpos
> total_size
)
4296 range
= total_size
- startpos
;
4298 /* If the search isn't to be a backwards one, don't waste time in a
4299 search for a pattern anchored at beginning of buffer. */
4300 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4309 /* In a forward search for something that starts with \=.
4310 don't keep searching past point. */
4311 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4313 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4319 /* Update the fastmap now if not correct already. */
4320 if (fastmap
&& !bufp
->fastmap_accurate
)
4321 re_compile_fastmap (bufp
);
4323 /* See whether the pattern is anchored. */
4324 anchored_start
= (bufp
->buffer
[0] == begline
);
4327 gl_state
.object
= re_match_object
;
4329 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4331 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4335 /* Loop through the string, looking for a place to start matching. */
4338 /* If the pattern is anchored,
4339 skip quickly past places we cannot match.
4340 We don't bother to treat startpos == 0 specially
4341 because that case doesn't repeat. */
4342 if (anchored_start
&& startpos
> 0)
4344 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4345 : string2
[startpos
- size1
- 1])
4350 /* If a fastmap is supplied, skip quickly over characters that
4351 cannot be the start of a match. If the pattern can match the
4352 null string, however, we don't need to skip characters; we want
4353 the first null string. */
4354 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4356 register re_char
*d
;
4357 register re_wchar_t buf_ch
;
4359 d
= POS_ADDR_VSTRING (startpos
);
4361 if (range
> 0) /* Searching forwards. */
4363 register int lim
= 0;
4366 if (startpos
< size1
&& startpos
+ range
>= size1
)
4367 lim
= range
- (size1
- startpos
);
4369 /* Written out as an if-else to avoid testing `translate'
4371 if (RE_TRANSLATE_P (translate
))
4378 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4381 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4386 range
-= buf_charlen
;
4391 /* Convert *d to integer to shut up GCC's
4392 whining about comparison that is always
4397 && !fastmap
[RE_TRANSLATE (translate
, di
)])
4407 re_char
*d_start
= d
;
4408 while (range
> lim
&& !fastmap
[*d
])
4414 if (multibyte
&& range
> lim
)
4416 /* Check that we are at the beginning of a char. */
4418 AT_CHAR_BOUNDARY_P (at_boundary
, d
, d_start
);
4422 { /* We have matched an internal byte of a char
4423 rather than the leading byte, so it's a false
4424 positive: we should keep scanning. */
4433 startpos
+= irange
- range
;
4435 else /* Searching backwards. */
4437 int room
= (startpos
>= size1
4438 ? size2
+ size1
- startpos
4439 : size1
- startpos
);
4440 buf_ch
= RE_STRING_CHAR (d
, room
);
4441 buf_ch
= TRANSLATE (buf_ch
);
4443 if (! (buf_ch
>= 0400
4444 || fastmap
[buf_ch
]))
4449 /* If can't match the null string, and that's all we have left, fail. */
4450 if (range
>= 0 && startpos
== total_size
&& fastmap
4451 && !bufp
->can_be_null
)
4454 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4455 startpos
, regs
, stop
);
4456 #ifndef REGEX_MALLOC
4473 /* Update STARTPOS to the next character boundary. */
4476 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4477 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4478 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
4496 /* Update STARTPOS to the previous character boundary. */
4499 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4501 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4503 /* Find the head of multibyte form. */
4504 PREV_CHAR_BOUNDARY (p
, phead
);
4505 range
+= p0
- 1 - p
;
4509 startpos
-= p0
- 1 - p
;
4515 WEAK_ALIAS (__re_search_2
, re_search_2
)
4517 /* Declarations and macros for re_match_2. */
4519 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4521 RE_TRANSLATE_TYPE translate
,
4522 const int multibyte
));
4524 /* This converts PTR, a pointer into one of the search strings `string1'
4525 and `string2' into an offset from the beginning of that string. */
4526 #define POINTER_TO_OFFSET(ptr) \
4527 (FIRST_STRING_P (ptr) \
4528 ? ((regoff_t) ((ptr) - string1)) \
4529 : ((regoff_t) ((ptr) - string2 + size1)))
4531 /* Call before fetching a character with *d. This switches over to
4532 string2 if necessary.
4533 Check re_match_2_internal for a discussion of why end_match_2 might
4534 not be within string2 (but be equal to end_match_1 instead). */
4535 #define PREFETCH() \
4538 /* End of string2 => fail. */ \
4539 if (dend == end_match_2) \
4541 /* End of string1 => advance to string2. */ \
4543 dend = end_match_2; \
4546 /* Call before fetching a char with *d if you already checked other limits.
4547 This is meant for use in lookahead operations like wordend, etc..
4548 where we might need to look at parts of the string that might be
4549 outside of the LIMITs (i.e past `stop'). */
4550 #define PREFETCH_NOLIMIT() \
4554 dend = end_match_2; \
4557 /* Test if at very beginning or at very end of the virtual concatenation
4558 of `string1' and `string2'. If only one string, it's `string2'. */
4559 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4560 #define AT_STRINGS_END(d) ((d) == end2)
4563 /* Test if D points to a character which is word-constituent. We have
4564 two special cases to check for: if past the end of string1, look at
4565 the first character in string2; and if before the beginning of
4566 string2, look at the last character in string1. */
4567 #define WORDCHAR_P(d) \
4568 (SYNTAX ((d) == end1 ? *string2 \
4569 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4572 /* Disabled due to a compiler bug -- see comment at case wordbound */
4574 /* The comment at case wordbound is following one, but we don't use
4575 AT_WORD_BOUNDARY anymore to support multibyte form.
4577 The DEC Alpha C compiler 3.x generates incorrect code for the
4578 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4579 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4580 macro and introducing temporary variables works around the bug. */
4583 /* Test if the character before D and the one at D differ with respect
4584 to being word-constituent. */
4585 #define AT_WORD_BOUNDARY(d) \
4586 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4587 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4590 /* Free everything we malloc. */
4591 #ifdef MATCH_MAY_ALLOCATE
4592 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4593 # define FREE_VARIABLES() \
4595 REGEX_FREE_STACK (fail_stack.stack); \
4596 FREE_VAR (regstart); \
4597 FREE_VAR (regend); \
4598 FREE_VAR (best_regstart); \
4599 FREE_VAR (best_regend); \
4602 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4603 #endif /* not MATCH_MAY_ALLOCATE */
4606 /* Optimization routines. */
4608 /* If the operation is a match against one or more chars,
4609 return a pointer to the next operation, else return NULL. */
4614 switch (SWITCH_ENUM_CAST (*p
++))
4625 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4628 p
= CHARSET_RANGE_TABLE (p
- 1);
4629 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4630 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4633 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4640 case notcategoryspec
:
4652 /* Jump over non-matching operations. */
4654 skip_noops (p
, pend
)
4660 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4669 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4680 /* Non-zero if "p1 matches something" implies "p2 fails". */
4682 mutually_exclusive_p (bufp
, p1
, p2
)
4683 struct re_pattern_buffer
*bufp
;
4687 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4688 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4690 assert (p1
>= bufp
->buffer
&& p1
< pend
4691 && p2
>= bufp
->buffer
&& p2
<= pend
);
4693 /* Skip over open/close-group commands.
4694 If what follows this loop is a ...+ construct,
4695 look at what begins its body, since we will have to
4696 match at least one of that. */
4697 p2
= skip_noops (p2
, pend
);
4698 /* The same skip can be done for p1, except that this function
4699 is only used in the case where p1 is a simple match operator. */
4700 /* p1 = skip_noops (p1, pend); */
4702 assert (p1
>= bufp
->buffer
&& p1
< pend
4703 && p2
>= bufp
->buffer
&& p2
<= pend
);
4705 op2
= p2
== pend
? succeed
: *p2
;
4707 switch (SWITCH_ENUM_CAST (op2
))
4711 /* If we're at the end of the pattern, we can change. */
4712 if (skip_one_char (p1
))
4714 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4722 register re_wchar_t c
4723 = (re_opcode_t
) *p2
== endline
? '\n'
4724 : RE_STRING_CHAR (p2
+ 2, pend
- p2
- 2);
4726 if ((re_opcode_t
) *p1
== exactn
)
4728 if (c
!= RE_STRING_CHAR (p1
+ 2, pend
- p1
- 2))
4730 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4735 else if ((re_opcode_t
) *p1
== charset
4736 || (re_opcode_t
) *p1
== charset_not
)
4738 int not = (re_opcode_t
) *p1
== charset_not
;
4740 /* Test if C is listed in charset (or charset_not)
4742 if (SINGLE_BYTE_CHAR_P (c
))
4744 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4745 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4748 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4749 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4751 /* `not' is equal to 1 if c would match, which means
4752 that we can't change to pop_failure_jump. */
4755 DEBUG_PRINT1 (" No match => fast loop.\n");
4759 else if ((re_opcode_t
) *p1
== anychar
4762 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4770 if ((re_opcode_t
) *p1
== exactn
)
4771 /* Reuse the code above. */
4772 return mutually_exclusive_p (bufp
, p2
, p1
);
4774 /* It is hard to list up all the character in charset
4775 P2 if it includes multibyte character. Give up in
4777 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4779 /* Now, we are sure that P2 has no range table.
4780 So, for the size of bitmap in P2, `p2[1]' is
4781 enough. But P1 may have range table, so the
4782 size of bitmap table of P1 is extracted by
4783 using macro `CHARSET_BITMAP_SIZE'.
4785 Since we know that all the character listed in
4786 P2 is ASCII, it is enough to test only bitmap
4789 if ((re_opcode_t
) *p1
== charset
)
4792 /* We win if the charset inside the loop
4793 has no overlap with the one after the loop. */
4796 && idx
< CHARSET_BITMAP_SIZE (p1
));
4798 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4802 || idx
== CHARSET_BITMAP_SIZE (p1
))
4804 DEBUG_PRINT1 (" No match => fast loop.\n");
4808 else if ((re_opcode_t
) *p1
== charset_not
)
4811 /* We win if the charset_not inside the loop lists
4812 every character listed in the charset after. */
4813 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4814 if (! (p2
[2 + idx
] == 0
4815 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4816 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4821 DEBUG_PRINT1 (" No match => fast loop.\n");
4830 switch (SWITCH_ENUM_CAST (*p1
))
4834 /* Reuse the code above. */
4835 return mutually_exclusive_p (bufp
, p2
, p1
);
4837 /* When we have two charset_not, it's very unlikely that
4838 they don't overlap. The union of the two sets of excluded
4839 chars should cover all possible chars, which, as a matter of
4840 fact, is virtually impossible in multibyte buffers. */
4846 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4848 return ((re_opcode_t
) *p1
== syntaxspec
4849 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4851 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4854 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4856 return ((re_opcode_t
) *p1
== notsyntaxspec
4857 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4859 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4862 return (((re_opcode_t
) *p1
== notsyntaxspec
4863 || (re_opcode_t
) *p1
== syntaxspec
)
4868 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4869 case notcategoryspec
:
4870 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4882 /* Matching routines. */
4884 #ifndef emacs /* Emacs never uses this. */
4885 /* re_match is like re_match_2 except it takes only a single string. */
4888 re_match (bufp
, string
, size
, pos
, regs
)
4889 struct re_pattern_buffer
*bufp
;
4892 struct re_registers
*regs
;
4894 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
4896 # if defined C_ALLOCA && !defined REGEX_MALLOC
4901 WEAK_ALIAS (__re_match
, re_match
)
4902 #endif /* not emacs */
4905 /* In Emacs, this is the string or buffer in which we
4906 are matching. It is used for looking up syntax properties. */
4907 Lisp_Object re_match_object
;
4910 /* re_match_2 matches the compiled pattern in BUFP against the
4911 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4912 and SIZE2, respectively). We start matching at POS, and stop
4915 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4916 store offsets for the substring each group matched in REGS. See the
4917 documentation for exactly how many groups we fill.
4919 We return -1 if no match, -2 if an internal error (such as the
4920 failure stack overflowing). Otherwise, we return the length of the
4921 matched substring. */
4924 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4925 struct re_pattern_buffer
*bufp
;
4926 const char *string1
, *string2
;
4929 struct re_registers
*regs
;
4936 gl_state
.object
= re_match_object
;
4937 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4938 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4941 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4942 (re_char
*) string2
, size2
,
4944 #if defined C_ALLOCA && !defined REGEX_MALLOC
4949 WEAK_ALIAS (__re_match_2
, re_match_2
)
4951 /* This is a separate function so that we can force an alloca cleanup
4954 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4955 struct re_pattern_buffer
*bufp
;
4956 re_char
*string1
, *string2
;
4959 struct re_registers
*regs
;
4962 /* General temporaries. */
4967 /* Just past the end of the corresponding string. */
4968 re_char
*end1
, *end2
;
4970 /* Pointers into string1 and string2, just past the last characters in
4971 each to consider matching. */
4972 re_char
*end_match_1
, *end_match_2
;
4974 /* Where we are in the data, and the end of the current string. */
4977 /* Used sometimes to remember where we were before starting matching
4978 an operator so that we can go back in case of failure. This "atomic"
4979 behavior of matching opcodes is indispensable to the correctness
4980 of the on_failure_keep_string_jump optimization. */
4983 /* Where we are in the pattern, and the end of the pattern. */
4984 re_char
*p
= bufp
->buffer
;
4985 re_char
*pend
= p
+ bufp
->used
;
4987 /* We use this to map every character in the string. */
4988 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4990 /* Nonzero if we have to concern multibyte character. */
4991 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4993 /* Failure point stack. Each place that can handle a failure further
4994 down the line pushes a failure point on this stack. It consists of
4995 regstart, and regend for all registers corresponding to
4996 the subexpressions we're currently inside, plus the number of such
4997 registers, and, finally, two char *'s. The first char * is where
4998 to resume scanning the pattern; the second one is where to resume
4999 scanning the strings. */
5000 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5001 fail_stack_type fail_stack
;
5004 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5007 #if defined REL_ALLOC && defined REGEX_MALLOC
5008 /* This holds the pointer to the failure stack, when
5009 it is allocated relocatably. */
5010 fail_stack_elt_t
*failure_stack_ptr
;
5013 /* We fill all the registers internally, independent of what we
5014 return, for use in backreferences. The number here includes
5015 an element for register zero. */
5016 size_t num_regs
= bufp
->re_nsub
+ 1;
5018 /* Information on the contents of registers. These are pointers into
5019 the input strings; they record just what was matched (on this
5020 attempt) by a subexpression part of the pattern, that is, the
5021 regnum-th regstart pointer points to where in the pattern we began
5022 matching and the regnum-th regend points to right after where we
5023 stopped matching the regnum-th subexpression. (The zeroth register
5024 keeps track of what the whole pattern matches.) */
5025 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5026 re_char
**regstart
, **regend
;
5029 /* The following record the register info as found in the above
5030 variables when we find a match better than any we've seen before.
5031 This happens as we backtrack through the failure points, which in
5032 turn happens only if we have not yet matched the entire string. */
5033 unsigned best_regs_set
= false;
5034 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5035 re_char
**best_regstart
, **best_regend
;
5038 /* Logically, this is `best_regend[0]'. But we don't want to have to
5039 allocate space for that if we're not allocating space for anything
5040 else (see below). Also, we never need info about register 0 for
5041 any of the other register vectors, and it seems rather a kludge to
5042 treat `best_regend' differently than the rest. So we keep track of
5043 the end of the best match so far in a separate variable. We
5044 initialize this to NULL so that when we backtrack the first time
5045 and need to test it, it's not garbage. */
5046 re_char
*match_end
= NULL
;
5049 /* Counts the total number of registers pushed. */
5050 unsigned num_regs_pushed
= 0;
5053 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5057 #ifdef MATCH_MAY_ALLOCATE
5058 /* Do not bother to initialize all the register variables if there are
5059 no groups in the pattern, as it takes a fair amount of time. If
5060 there are groups, we include space for register 0 (the whole
5061 pattern), even though we never use it, since it simplifies the
5062 array indexing. We should fix this. */
5065 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5066 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5067 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5068 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5070 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5078 /* We must initialize all our variables to NULL, so that
5079 `FREE_VARIABLES' doesn't try to free them. */
5080 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5082 #endif /* MATCH_MAY_ALLOCATE */
5084 /* The starting position is bogus. */
5085 if (pos
< 0 || pos
> size1
+ size2
)
5091 /* Initialize subexpression text positions to -1 to mark ones that no
5092 start_memory/stop_memory has been seen for. Also initialize the
5093 register information struct. */
5094 for (reg
= 1; reg
< num_regs
; reg
++)
5095 regstart
[reg
] = regend
[reg
] = NULL
;
5097 /* We move `string1' into `string2' if the latter's empty -- but not if
5098 `string1' is null. */
5099 if (size2
== 0 && string1
!= NULL
)
5106 end1
= string1
+ size1
;
5107 end2
= string2
+ size2
;
5109 /* `p' scans through the pattern as `d' scans through the data.
5110 `dend' is the end of the input string that `d' points within. `d'
5111 is advanced into the following input string whenever necessary, but
5112 this happens before fetching; therefore, at the beginning of the
5113 loop, `d' can be pointing at the end of a string, but it cannot
5117 /* Only match within string2. */
5118 d
= string2
+ pos
- size1
;
5119 dend
= end_match_2
= string2
+ stop
- size1
;
5120 end_match_1
= end1
; /* Just to give it a value. */
5126 /* Only match within string1. */
5127 end_match_1
= string1
+ stop
;
5129 When we reach end_match_1, PREFETCH normally switches to string2.
5130 But in the present case, this means that just doing a PREFETCH
5131 makes us jump from `stop' to `gap' within the string.
5132 What we really want here is for the search to stop as
5133 soon as we hit end_match_1. That's why we set end_match_2
5134 to end_match_1 (since PREFETCH fails as soon as we hit
5136 end_match_2
= end_match_1
;
5139 { /* It's important to use this code when stop == size so that
5140 moving `d' from end1 to string2 will not prevent the d == dend
5141 check from catching the end of string. */
5143 end_match_2
= string2
+ stop
- size1
;
5149 DEBUG_PRINT1 ("The compiled pattern is: ");
5150 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5151 DEBUG_PRINT1 ("The string to match is: `");
5152 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5153 DEBUG_PRINT1 ("'\n");
5155 /* This loops over pattern commands. It exits by returning from the
5156 function if the match is complete, or it drops through if the match
5157 fails at this starting point in the input data. */
5160 DEBUG_PRINT2 ("\n%p: ", p
);
5163 { /* End of pattern means we might have succeeded. */
5164 DEBUG_PRINT1 ("end of pattern ... ");
5166 /* If we haven't matched the entire string, and we want the
5167 longest match, try backtracking. */
5168 if (d
!= end_match_2
)
5170 /* 1 if this match ends in the same string (string1 or string2)
5171 as the best previous match. */
5172 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5173 == FIRST_STRING_P (d
));
5174 /* 1 if this match is the best seen so far. */
5175 boolean best_match_p
;
5177 /* AIX compiler got confused when this was combined
5178 with the previous declaration. */
5180 best_match_p
= d
> match_end
;
5182 best_match_p
= !FIRST_STRING_P (d
);
5184 DEBUG_PRINT1 ("backtracking.\n");
5186 if (!FAIL_STACK_EMPTY ())
5187 { /* More failure points to try. */
5189 /* If exceeds best match so far, save it. */
5190 if (!best_regs_set
|| best_match_p
)
5192 best_regs_set
= true;
5195 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5197 for (reg
= 1; reg
< num_regs
; reg
++)
5199 best_regstart
[reg
] = regstart
[reg
];
5200 best_regend
[reg
] = regend
[reg
];
5206 /* If no failure points, don't restore garbage. And if
5207 last match is real best match, don't restore second
5209 else if (best_regs_set
&& !best_match_p
)
5212 /* Restore best match. It may happen that `dend ==
5213 end_match_1' while the restored d is in string2.
5214 For example, the pattern `x.*y.*z' against the
5215 strings `x-' and `y-z-', if the two strings are
5216 not consecutive in memory. */
5217 DEBUG_PRINT1 ("Restoring best registers.\n");
5220 dend
= ((d
>= string1
&& d
<= end1
)
5221 ? end_match_1
: end_match_2
);
5223 for (reg
= 1; reg
< num_regs
; reg
++)
5225 regstart
[reg
] = best_regstart
[reg
];
5226 regend
[reg
] = best_regend
[reg
];
5229 } /* d != end_match_2 */
5232 DEBUG_PRINT1 ("Accepting match.\n");
5234 /* If caller wants register contents data back, do it. */
5235 if (regs
&& !bufp
->no_sub
)
5237 /* Have the register data arrays been allocated? */
5238 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5239 { /* No. So allocate them with malloc. We need one
5240 extra element beyond `num_regs' for the `-1' marker
5242 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5243 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5244 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5245 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5250 bufp
->regs_allocated
= REGS_REALLOCATE
;
5252 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5253 { /* Yes. If we need more elements than were already
5254 allocated, reallocate them. If we need fewer, just
5256 if (regs
->num_regs
< num_regs
+ 1)
5258 regs
->num_regs
= num_regs
+ 1;
5259 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5260 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5261 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5270 /* These braces fend off a "empty body in an else-statement"
5271 warning under GCC when assert expands to nothing. */
5272 assert (bufp
->regs_allocated
== REGS_FIXED
);
5275 /* Convert the pointer data in `regstart' and `regend' to
5276 indices. Register zero has to be set differently,
5277 since we haven't kept track of any info for it. */
5278 if (regs
->num_regs
> 0)
5280 regs
->start
[0] = pos
;
5281 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5284 /* Go through the first `min (num_regs, regs->num_regs)'
5285 registers, since that is all we initialized. */
5286 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5288 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5289 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5293 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5295 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5299 /* If the regs structure we return has more elements than
5300 were in the pattern, set the extra elements to -1. If
5301 we (re)allocated the registers, this is the case,
5302 because we always allocate enough to have at least one
5304 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5305 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5306 } /* regs && !bufp->no_sub */
5308 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5309 nfailure_points_pushed
, nfailure_points_popped
,
5310 nfailure_points_pushed
- nfailure_points_popped
);
5311 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5313 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5315 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5321 /* Otherwise match next pattern command. */
5322 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5324 /* Ignore these. Used to ignore the n of succeed_n's which
5325 currently have n == 0. */
5327 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5331 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5334 /* Match the next n pattern characters exactly. The following
5335 byte in the pattern defines n, and the n bytes after that
5336 are the characters to match. */
5339 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5341 /* Remember the start point to rollback upon failure. */
5344 /* This is written out as an if-else so we don't waste time
5345 testing `translate' inside the loop. */
5346 if (RE_TRANSLATE_P (translate
))
5351 int pat_charlen
, buf_charlen
;
5352 unsigned int pat_ch
, buf_ch
;
5355 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
5356 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5358 if (RE_TRANSLATE (translate
, buf_ch
)
5367 mcnt
-= pat_charlen
;
5373 /* Avoid compiler whining about comparison being
5379 if (RE_TRANSLATE (translate
, di
) != *p
++)
5404 /* Match any character except possibly a newline or a null. */
5410 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5413 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5414 buf_ch
= TRANSLATE (buf_ch
);
5416 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5418 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5419 && buf_ch
== '\000'))
5422 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5431 register unsigned int c
;
5432 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5435 /* Start of actual range_table, or end of bitmap if there is no
5437 re_char
*range_table
;
5439 /* Nonzero if there is a range table. */
5440 int range_table_exists
;
5442 /* Number of ranges of range table. This is not included
5443 in the initial byte-length of the command. */
5446 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5448 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5450 if (range_table_exists
)
5452 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5453 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5457 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5458 c
= TRANSLATE (c
); /* The character to match. */
5460 if (SINGLE_BYTE_CHAR_P (c
))
5461 { /* Lookup bitmap. */
5462 /* Cast to `unsigned' instead of `unsigned char' in
5463 case the bit list is a full 32 bytes long. */
5464 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5465 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5469 else if (range_table_exists
)
5471 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5473 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5474 | (class_bits
& BIT_MULTIBYTE
)
5475 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5476 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5477 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5478 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5481 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5485 if (range_table_exists
)
5486 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5488 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5490 if (!not) goto fail
;
5497 /* The beginning of a group is represented by start_memory.
5498 The argument is the register number. The text
5499 matched within the group is recorded (in the internal
5500 registers data structure) under the register number. */
5502 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5504 /* In case we need to undo this operation (via backtracking). */
5505 PUSH_FAILURE_REG ((unsigned int)*p
);
5508 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5509 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5511 /* Move past the register number and inner group count. */
5516 /* The stop_memory opcode represents the end of a group. Its
5517 argument is the same as start_memory's: the register number. */
5519 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5521 assert (!REG_UNSET (regstart
[*p
]));
5522 /* Strictly speaking, there should be code such as:
5524 assert (REG_UNSET (regend[*p]));
5525 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5527 But the only info to be pushed is regend[*p] and it is known to
5528 be UNSET, so there really isn't anything to push.
5529 Not pushing anything, on the other hand deprives us from the
5530 guarantee that regend[*p] is UNSET since undoing this operation
5531 will not reset its value properly. This is not important since
5532 the value will only be read on the next start_memory or at
5533 the very end and both events can only happen if this stop_memory
5537 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5539 /* Move past the register number and the inner group count. */
5544 /* \<digit> has been turned into a `duplicate' command which is
5545 followed by the numeric value of <digit> as the register number. */
5548 register re_char
*d2
, *dend2
;
5549 int regno
= *p
++; /* Get which register to match against. */
5550 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5552 /* Can't back reference a group which we've never matched. */
5553 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5556 /* Where in input to try to start matching. */
5557 d2
= regstart
[regno
];
5559 /* Remember the start point to rollback upon failure. */
5562 /* Where to stop matching; if both the place to start and
5563 the place to stop matching are in the same string, then
5564 set to the place to stop, otherwise, for now have to use
5565 the end of the first string. */
5567 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5568 == FIRST_STRING_P (regend
[regno
]))
5569 ? regend
[regno
] : end_match_1
);
5572 /* If necessary, advance to next segment in register
5576 if (dend2
== end_match_2
) break;
5577 if (dend2
== regend
[regno
]) break;
5579 /* End of string1 => advance to string2. */
5581 dend2
= regend
[regno
];
5583 /* At end of register contents => success */
5584 if (d2
== dend2
) break;
5586 /* If necessary, advance to next segment in data. */
5589 /* How many characters left in this segment to match. */
5592 /* Want how many consecutive characters we can match in
5593 one shot, so, if necessary, adjust the count. */
5594 if (mcnt
> dend2
- d2
)
5597 /* Compare that many; failure if mismatch, else move
5599 if (RE_TRANSLATE_P (translate
)
5600 ? bcmp_translate (d
, d2
, mcnt
, translate
, multibyte
)
5601 : memcmp (d
, d2
, mcnt
))
5606 d
+= mcnt
, d2
+= mcnt
;
5612 /* begline matches the empty string at the beginning of the string
5613 (unless `not_bol' is set in `bufp'), and after newlines. */
5615 DEBUG_PRINT1 ("EXECUTING begline.\n");
5617 if (AT_STRINGS_BEG (d
))
5619 if (!bufp
->not_bol
) break;
5624 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5628 /* In all other cases, we fail. */
5632 /* endline is the dual of begline. */
5634 DEBUG_PRINT1 ("EXECUTING endline.\n");
5636 if (AT_STRINGS_END (d
))
5638 if (!bufp
->not_eol
) break;
5642 PREFETCH_NOLIMIT ();
5649 /* Match at the very beginning of the data. */
5651 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5652 if (AT_STRINGS_BEG (d
))
5657 /* Match at the very end of the data. */
5659 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5660 if (AT_STRINGS_END (d
))
5665 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5666 pushes NULL as the value for the string on the stack. Then
5667 `POP_FAILURE_POINT' will keep the current value for the
5668 string, instead of restoring it. To see why, consider
5669 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5670 then the . fails against the \n. But the next thing we want
5671 to do is match the \n against the \n; if we restored the
5672 string value, we would be back at the foo.
5674 Because this is used only in specific cases, we don't need to
5675 check all the things that `on_failure_jump' does, to make
5676 sure the right things get saved on the stack. Hence we don't
5677 share its code. The only reason to push anything on the
5678 stack at all is that otherwise we would have to change
5679 `anychar's code to do something besides goto fail in this
5680 case; that seems worse than this. */
5681 case on_failure_keep_string_jump
:
5682 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5683 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5686 PUSH_FAILURE_POINT (p
- 3, NULL
);
5689 /* A nasty loop is introduced by the non-greedy *? and +?.
5690 With such loops, the stack only ever contains one failure point
5691 at a time, so that a plain on_failure_jump_loop kind of
5692 cycle detection cannot work. Worse yet, such a detection
5693 can not only fail to detect a cycle, but it can also wrongly
5694 detect a cycle (between different instantiations of the same
5696 So the method used for those nasty loops is a little different:
5697 We use a special cycle-detection-stack-frame which is pushed
5698 when the on_failure_jump_nastyloop failure-point is *popped*.
5699 This special frame thus marks the beginning of one iteration
5700 through the loop and we can hence easily check right here
5701 whether something matched between the beginning and the end of
5703 case on_failure_jump_nastyloop
:
5704 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5705 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5708 assert ((re_opcode_t
)p
[-4] == no_op
);
5711 CHECK_INFINITE_LOOP (p
- 4, d
);
5713 /* If there's a cycle, just continue without pushing
5714 this failure point. The failure point is the "try again"
5715 option, which shouldn't be tried.
5716 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5717 PUSH_FAILURE_POINT (p
- 3, d
);
5721 /* Simple loop detecting on_failure_jump: just check on the
5722 failure stack if the same spot was already hit earlier. */
5723 case on_failure_jump_loop
:
5725 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5726 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5730 CHECK_INFINITE_LOOP (p
- 3, d
);
5732 /* If there's a cycle, get out of the loop, as if the matching
5733 had failed. We used to just `goto fail' here, but that was
5734 aborting the search a bit too early: we want to keep the
5735 empty-loop-match and keep matching after the loop.
5736 We want (x?)*y\1z to match both xxyz and xxyxz. */
5739 PUSH_FAILURE_POINT (p
- 3, d
);
5744 /* Uses of on_failure_jump:
5746 Each alternative starts with an on_failure_jump that points
5747 to the beginning of the next alternative. Each alternative
5748 except the last ends with a jump that in effect jumps past
5749 the rest of the alternatives. (They really jump to the
5750 ending jump of the following alternative, because tensioning
5751 these jumps is a hassle.)
5753 Repeats start with an on_failure_jump that points past both
5754 the repetition text and either the following jump or
5755 pop_failure_jump back to this on_failure_jump. */
5756 case on_failure_jump
:
5757 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5758 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5761 PUSH_FAILURE_POINT (p
-3, d
);
5764 /* This operation is used for greedy *.
5765 Compare the beginning of the repeat with what in the
5766 pattern follows its end. If we can establish that there
5767 is nothing that they would both match, i.e., that we
5768 would have to backtrack because of (as in, e.g., `a*a')
5769 then we can use a non-backtracking loop based on
5770 on_failure_keep_string_jump instead of on_failure_jump. */
5771 case on_failure_jump_smart
:
5772 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5773 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5776 re_char
*p1
= p
; /* Next operation. */
5777 /* Here, we discard `const', making re_match non-reentrant. */
5778 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5779 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5781 p
-= 3; /* Reset so that we will re-execute the
5782 instruction once it's been changed. */
5784 EXTRACT_NUMBER (mcnt
, p2
- 2);
5786 /* Ensure this is a indeed the trivial kind of loop
5787 we are expecting. */
5788 assert (skip_one_char (p1
) == p2
- 3);
5789 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5790 DEBUG_STATEMENT (debug
+= 2);
5791 if (mutually_exclusive_p (bufp
, p1
, p2
))
5793 /* Use a fast `on_failure_keep_string_jump' loop. */
5794 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5795 *p3
= (unsigned char) on_failure_keep_string_jump
;
5796 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5800 /* Default to a safe `on_failure_jump' loop. */
5801 DEBUG_PRINT1 (" smart default => slow loop.\n");
5802 *p3
= (unsigned char) on_failure_jump
;
5804 DEBUG_STATEMENT (debug
-= 2);
5808 /* Unconditionally jump (without popping any failure points). */
5811 IMMEDIATE_QUIT_CHECK
;
5812 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5813 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5814 p
+= mcnt
; /* Do the jump. */
5815 DEBUG_PRINT2 ("(to %p).\n", p
);
5819 /* Have to succeed matching what follows at least n times.
5820 After that, handle like `on_failure_jump'. */
5822 /* Signedness doesn't matter since we only compare MCNT to 0. */
5823 EXTRACT_NUMBER (mcnt
, p
+ 2);
5824 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5826 /* Originally, mcnt is how many times we HAVE to succeed. */
5829 /* Here, we discard `const', making re_match non-reentrant. */
5830 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5833 PUSH_NUMBER (p2
, mcnt
);
5836 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5841 /* Signedness doesn't matter since we only compare MCNT to 0. */
5842 EXTRACT_NUMBER (mcnt
, p
+ 2);
5843 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5845 /* Originally, this is how many times we CAN jump. */
5848 /* Here, we discard `const', making re_match non-reentrant. */
5849 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5851 PUSH_NUMBER (p2
, mcnt
);
5852 goto unconditional_jump
;
5854 /* If don't have to jump any more, skip over the rest of command. */
5861 unsigned char *p2
; /* Location of the counter. */
5862 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5864 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5865 /* Here, we discard `const', making re_match non-reentrant. */
5866 p2
= (unsigned char*) p
+ mcnt
;
5867 /* Signedness doesn't matter since we only copy MCNT's bits . */
5868 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5869 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
5870 PUSH_NUMBER (p2
, mcnt
);
5876 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5877 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5879 /* We SUCCEED (or FAIL) in one of the following cases: */
5881 /* Case 1: D is at the beginning or the end of string. */
5882 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5886 /* C1 is the character before D, S1 is the syntax of C1, C2
5887 is the character at D, and S2 is the syntax of C2. */
5891 int offset
= PTR_TO_OFFSET (d
- 1);
5892 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5893 UPDATE_SYNTAX_TABLE (charpos
);
5895 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5898 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5900 PREFETCH_NOLIMIT ();
5901 c2
= RE_STRING_CHAR (d
, dend
- d
);
5904 if (/* Case 2: Only one of S1 and S2 is Sword. */
5905 ((s1
== Sword
) != (s2
== Sword
))
5906 /* Case 3: Both of S1 and S2 are Sword, and macro
5907 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5908 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5917 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5919 /* We FAIL in one of the following cases: */
5921 /* Case 1: D is at the end of string. */
5922 if (AT_STRINGS_END (d
))
5926 /* C1 is the character before D, S1 is the syntax of C1, C2
5927 is the character at D, and S2 is the syntax of C2. */
5931 int offset
= PTR_TO_OFFSET (d
);
5932 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5933 UPDATE_SYNTAX_TABLE (charpos
);
5936 c2
= RE_STRING_CHAR (d
, dend
- d
);
5939 /* Case 2: S2 is not Sword. */
5943 /* Case 3: D is not at the beginning of string ... */
5944 if (!AT_STRINGS_BEG (d
))
5946 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5948 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5952 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5954 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5961 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5963 /* We FAIL in one of the following cases: */
5965 /* Case 1: D is at the beginning of string. */
5966 if (AT_STRINGS_BEG (d
))
5970 /* C1 is the character before D, S1 is the syntax of C1, C2
5971 is the character at D, and S2 is the syntax of C2. */
5975 int offset
= PTR_TO_OFFSET (d
) - 1;
5976 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5977 UPDATE_SYNTAX_TABLE (charpos
);
5979 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5982 /* Case 2: S1 is not Sword. */
5986 /* Case 3: D is not at the end of string ... */
5987 if (!AT_STRINGS_END (d
))
5989 PREFETCH_NOLIMIT ();
5990 c2
= RE_STRING_CHAR (d
, dend
- d
);
5992 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5996 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5998 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6005 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6007 /* We FAIL in one of the following cases: */
6009 /* Case 1: D is at the end of string. */
6010 if (AT_STRINGS_END (d
))
6014 /* C1 is the character before D, S1 is the syntax of C1, C2
6015 is the character at D, and S2 is the syntax of C2. */
6019 int offset
= PTR_TO_OFFSET (d
);
6020 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6021 UPDATE_SYNTAX_TABLE (charpos
);
6024 c2
= RE_STRING_CHAR (d
, dend
- d
);
6027 /* Case 2: S2 is neither Sword nor Ssymbol. */
6028 if (s2
!= Sword
&& s2
!= Ssymbol
)
6031 /* Case 3: D is not at the beginning of string ... */
6032 if (!AT_STRINGS_BEG (d
))
6034 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6036 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6040 /* ... and S1 is Sword or Ssymbol. */
6041 if (s1
== Sword
|| s1
== Ssymbol
)
6048 DEBUG_PRINT1 ("EXECUTING symend.\n");
6050 /* We FAIL in one of the following cases: */
6052 /* Case 1: D is at the beginning of string. */
6053 if (AT_STRINGS_BEG (d
))
6057 /* C1 is the character before D, S1 is the syntax of C1, C2
6058 is the character at D, and S2 is the syntax of C2. */
6062 int offset
= PTR_TO_OFFSET (d
) - 1;
6063 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6064 UPDATE_SYNTAX_TABLE (charpos
);
6066 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6069 /* Case 2: S1 is neither Ssymbol nor Sword. */
6070 if (s1
!= Sword
&& s1
!= Ssymbol
)
6073 /* Case 3: D is not at the end of string ... */
6074 if (!AT_STRINGS_END (d
))
6076 PREFETCH_NOLIMIT ();
6077 c2
= RE_STRING_CHAR (d
, dend
- d
);
6079 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6083 /* ... and S2 is Sword or Ssymbol. */
6084 if (s2
== Sword
|| s2
== Ssymbol
)
6092 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6094 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
6098 int offset
= PTR_TO_OFFSET (d
);
6099 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6100 UPDATE_SYNTAX_TABLE (pos1
);
6107 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
6109 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6117 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6118 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6123 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6124 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6129 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6130 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6135 case notcategoryspec
:
6136 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6138 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
6144 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
6146 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6157 continue; /* Successfully executed one pattern command; keep going. */
6160 /* We goto here if a matching operation fails. */
6162 IMMEDIATE_QUIT_CHECK
;
6163 if (!FAIL_STACK_EMPTY ())
6166 /* A restart point is known. Restore to that state. */
6167 DEBUG_PRINT1 ("\nFAIL:\n");
6168 POP_FAILURE_POINT (str
, pat
);
6169 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
6171 case on_failure_keep_string_jump
:
6172 assert (str
== NULL
);
6173 goto continue_failure_jump
;
6175 case on_failure_jump_nastyloop
:
6176 assert ((re_opcode_t
)pat
[-2] == no_op
);
6177 PUSH_FAILURE_POINT (pat
- 2, str
);
6180 case on_failure_jump_loop
:
6181 case on_failure_jump
:
6184 continue_failure_jump
:
6185 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6190 /* A special frame used for nastyloops. */
6197 assert (p
>= bufp
->buffer
&& p
<= pend
);
6199 if (d
>= string1
&& d
<= end1
)
6203 break; /* Matching at this starting point really fails. */
6207 goto restore_best_regs
;
6211 return -1; /* Failure to match. */
6214 /* Subroutine definitions for re_match_2. */
6216 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6217 bytes; nonzero otherwise. */
6220 bcmp_translate (s1
, s2
, len
, translate
, multibyte
)
6223 RE_TRANSLATE_TYPE translate
;
6224 const int multibyte
;
6226 register re_char
*p1
= s1
, *p2
= s2
;
6227 re_char
*p1_end
= s1
+ len
;
6228 re_char
*p2_end
= s2
+ len
;
6230 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6231 different lengths, but relying on a single `len' would break this. -sm */
6232 while (p1
< p1_end
&& p2
< p2_end
)
6234 int p1_charlen
, p2_charlen
;
6235 re_wchar_t p1_ch
, p2_ch
;
6237 p1_ch
= RE_STRING_CHAR_AND_LENGTH (p1
, p1_end
- p1
, p1_charlen
);
6238 p2_ch
= RE_STRING_CHAR_AND_LENGTH (p2
, p2_end
- p2
, p2_charlen
);
6240 if (RE_TRANSLATE (translate
, p1_ch
)
6241 != RE_TRANSLATE (translate
, p2_ch
))
6244 p1
+= p1_charlen
, p2
+= p2_charlen
;
6247 if (p1
!= p1_end
|| p2
!= p2_end
)
6253 /* Entry points for GNU code. */
6255 /* re_compile_pattern is the GNU regular expression compiler: it
6256 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6257 Returns 0 if the pattern was valid, otherwise an error string.
6259 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6260 are set in BUFP on entry.
6262 We call regex_compile to do the actual compilation. */
6265 re_compile_pattern (pattern
, length
, bufp
)
6266 const char *pattern
;
6268 struct re_pattern_buffer
*bufp
;
6273 gl_state
.current_syntax_table
= current_buffer
->syntax_table
;
6276 /* GNU code is written to assume at least RE_NREGS registers will be set
6277 (and at least one extra will be -1). */
6278 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6280 /* And GNU code determines whether or not to get register information
6281 by passing null for the REGS argument to re_match, etc., not by
6285 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6289 return gettext (re_error_msgid
[(int) ret
]);
6291 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6293 /* Entry points compatible with 4.2 BSD regex library. We don't define
6294 them unless specifically requested. */
6296 #if defined _REGEX_RE_COMP || defined _LIBC
6298 /* BSD has one and only one pattern buffer. */
6299 static struct re_pattern_buffer re_comp_buf
;
6303 /* Make these definitions weak in libc, so POSIX programs can redefine
6304 these names if they don't use our functions, and still use
6305 regcomp/regexec below without link errors. */
6315 if (!re_comp_buf
.buffer
)
6316 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6317 return (char *) gettext ("No previous regular expression");
6321 if (!re_comp_buf
.buffer
)
6323 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6324 if (re_comp_buf
.buffer
== NULL
)
6325 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6326 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6327 re_comp_buf
.allocated
= 200;
6329 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6330 if (re_comp_buf
.fastmap
== NULL
)
6331 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6332 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6335 /* Since `re_exec' always passes NULL for the `regs' argument, we
6336 don't need to initialize the pattern buffer fields which affect it. */
6338 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6343 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6344 return (char *) gettext (re_error_msgid
[(int) ret
]);
6355 const int len
= strlen (s
);
6357 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6359 #endif /* _REGEX_RE_COMP */
6361 /* POSIX.2 functions. Don't define these for Emacs. */
6365 /* regcomp takes a regular expression as a string and compiles it.
6367 PREG is a regex_t *. We do not expect any fields to be initialized,
6368 since POSIX says we shouldn't. Thus, we set
6370 `buffer' to the compiled pattern;
6371 `used' to the length of the compiled pattern;
6372 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6373 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6374 RE_SYNTAX_POSIX_BASIC;
6375 `fastmap' to an allocated space for the fastmap;
6376 `fastmap_accurate' to zero;
6377 `re_nsub' to the number of subexpressions in PATTERN.
6379 PATTERN is the address of the pattern string.
6381 CFLAGS is a series of bits which affect compilation.
6383 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6384 use POSIX basic syntax.
6386 If REG_NEWLINE is set, then . and [^...] don't match newline.
6387 Also, regexec will try a match beginning after every newline.
6389 If REG_ICASE is set, then we considers upper- and lowercase
6390 versions of letters to be equivalent when matching.
6392 If REG_NOSUB is set, then when PREG is passed to regexec, that
6393 routine will report only success or failure, and nothing about the
6396 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6397 the return codes and their meanings.) */
6400 regcomp (preg
, pattern
, cflags
)
6401 regex_t
*__restrict preg
;
6402 const char *__restrict pattern
;
6407 = (cflags
& REG_EXTENDED
) ?
6408 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6410 /* regex_compile will allocate the space for the compiled pattern. */
6412 preg
->allocated
= 0;
6415 /* Try to allocate space for the fastmap. */
6416 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6418 if (cflags
& REG_ICASE
)
6423 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6424 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6425 if (preg
->translate
== NULL
)
6426 return (int) REG_ESPACE
;
6428 /* Map uppercase characters to corresponding lowercase ones. */
6429 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6430 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6433 preg
->translate
= NULL
;
6435 /* If REG_NEWLINE is set, newlines are treated differently. */
6436 if (cflags
& REG_NEWLINE
)
6437 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6438 syntax
&= ~RE_DOT_NEWLINE
;
6439 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6442 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6444 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6446 /* POSIX says a null character in the pattern terminates it, so we
6447 can use strlen here in compiling the pattern. */
6448 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6450 /* POSIX doesn't distinguish between an unmatched open-group and an
6451 unmatched close-group: both are REG_EPAREN. */
6452 if (ret
== REG_ERPAREN
)
6455 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6456 { /* Compute the fastmap now, since regexec cannot modify the pattern
6458 re_compile_fastmap (preg
);
6459 if (preg
->can_be_null
)
6460 { /* The fastmap can't be used anyway. */
6461 free (preg
->fastmap
);
6462 preg
->fastmap
= NULL
;
6467 WEAK_ALIAS (__regcomp
, regcomp
)
6470 /* regexec searches for a given pattern, specified by PREG, in the
6473 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6474 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6475 least NMATCH elements, and we set them to the offsets of the
6476 corresponding matched substrings.
6478 EFLAGS specifies `execution flags' which affect matching: if
6479 REG_NOTBOL is set, then ^ does not match at the beginning of the
6480 string; if REG_NOTEOL is set, then $ does not match at the end.
6482 We return 0 if we find a match and REG_NOMATCH if not. */
6485 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6486 const regex_t
*__restrict preg
;
6487 const char *__restrict string
;
6489 regmatch_t pmatch
[__restrict_arr
];
6493 struct re_registers regs
;
6494 regex_t private_preg
;
6495 int len
= strlen (string
);
6496 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6498 private_preg
= *preg
;
6500 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6501 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6503 /* The user has told us exactly how many registers to return
6504 information about, via `nmatch'. We have to pass that on to the
6505 matching routines. */
6506 private_preg
.regs_allocated
= REGS_FIXED
;
6510 regs
.num_regs
= nmatch
;
6511 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6512 if (regs
.start
== NULL
)
6513 return (int) REG_NOMATCH
;
6514 regs
.end
= regs
.start
+ nmatch
;
6517 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6518 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6519 was a little bit longer but still only matching the real part.
6520 This works because the `endline' will check for a '\n' and will find a
6521 '\0', correctly deciding that this is not the end of a line.
6522 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6523 a convenient '\0' there. For all we know, the string could be preceded
6524 by '\n' which would throw things off. */
6526 /* Perform the searching operation. */
6527 ret
= re_search (&private_preg
, string
, len
,
6528 /* start: */ 0, /* range: */ len
,
6529 want_reg_info
? ®s
: (struct re_registers
*) 0);
6531 /* Copy the register information to the POSIX structure. */
6538 for (r
= 0; r
< nmatch
; r
++)
6540 pmatch
[r
].rm_so
= regs
.start
[r
];
6541 pmatch
[r
].rm_eo
= regs
.end
[r
];
6545 /* If we needed the temporary register info, free the space now. */
6549 /* We want zero return to mean success, unlike `re_search'. */
6550 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6552 WEAK_ALIAS (__regexec
, regexec
)
6555 /* Returns a message corresponding to an error code, ERR_CODE, returned
6556 from either regcomp or regexec. We don't use PREG here.
6558 ERR_CODE was previously called ERRCODE, but that name causes an
6559 error with msvc8 compiler. */
6562 regerror (err_code
, preg
, errbuf
, errbuf_size
)
6564 const regex_t
*preg
;
6572 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6573 /* Only error codes returned by the rest of the code should be passed
6574 to this routine. If we are given anything else, or if other regex
6575 code generates an invalid error code, then the program has a bug.
6576 Dump core so we can fix it. */
6579 msg
= gettext (re_error_msgid
[err_code
]);
6581 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6583 if (errbuf_size
!= 0)
6585 if (msg_size
> errbuf_size
)
6587 strncpy (errbuf
, msg
, errbuf_size
- 1);
6588 errbuf
[errbuf_size
- 1] = 0;
6591 strcpy (errbuf
, msg
);
6596 WEAK_ALIAS (__regerror
, regerror
)
6599 /* Free dynamically allocated space used by PREG. */
6605 if (preg
->buffer
!= NULL
)
6606 free (preg
->buffer
);
6607 preg
->buffer
= NULL
;
6609 preg
->allocated
= 0;
6612 if (preg
->fastmap
!= NULL
)
6613 free (preg
->fastmap
);
6614 preg
->fastmap
= NULL
;
6615 preg
->fastmap_accurate
= 0;
6617 if (preg
->translate
!= NULL
)
6618 free (preg
->translate
);
6619 preg
->translate
= NULL
;
6621 WEAK_ALIAS (__regfree
, regfree
)
6623 #endif /* not emacs */
6625 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6626 (do not change this comment) */