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
7 Free Software Foundation, Inc.
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 3, or (at your option)
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; if not, write to the Free Software
21 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
25 - structure the opcode space into opcode+flag.
26 - merge with glibc's regex.[ch].
27 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
28 need to modify the compiled regexp so that re_match can be reentrant.
29 - get rid of on_failure_jump_smart by doing the optimization in re_comp
30 rather than at run-time, so that re_match can be reentrant.
33 /* AIX requires this to be the first thing in the file. */
34 #if defined _AIX && !defined REGEX_MALLOC
42 #if defined STDC_HEADERS && !defined emacs
45 /* We need this for `regex.h', and perhaps for the Emacs include files. */
46 # include <sys/types.h>
49 /* Whether to use ISO C Amendment 1 wide char functions.
50 Those should not be used for Emacs since it uses its own. */
52 #define WIDE_CHAR_SUPPORT 1
54 #define WIDE_CHAR_SUPPORT \
55 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
58 /* For platform which support the ISO C amendement 1 functionality we
59 support user defined character classes. */
61 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
67 /* We have to keep the namespace clean. */
68 # define regfree(preg) __regfree (preg)
69 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
70 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
71 # define regerror(err_code, preg, errbuf, errbuf_size) \
72 __regerror(err_code, preg, errbuf, errbuf_size)
73 # define re_set_registers(bu, re, nu, st, en) \
74 __re_set_registers (bu, re, nu, st, en)
75 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
76 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
77 # define re_match(bufp, string, size, pos, regs) \
78 __re_match (bufp, string, size, pos, regs)
79 # define re_search(bufp, string, size, startpos, range, regs) \
80 __re_search (bufp, string, size, startpos, range, regs)
81 # define re_compile_pattern(pattern, length, bufp) \
82 __re_compile_pattern (pattern, length, bufp)
83 # define re_set_syntax(syntax) __re_set_syntax (syntax)
84 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
85 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
86 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
88 /* Make sure we call libc's function even if the user overrides them. */
89 # define btowc __btowc
90 # define iswctype __iswctype
91 # define wctype __wctype
93 # define WEAK_ALIAS(a,b) weak_alias (a, b)
95 /* We are also using some library internals. */
96 # include <locale/localeinfo.h>
97 # include <locale/elem-hash.h>
98 # include <langinfo.h>
100 # define WEAK_ALIAS(a,b)
103 /* This is for other GNU distributions with internationalized messages. */
104 #if HAVE_LIBINTL_H || defined _LIBC
105 # include <libintl.h>
107 # define gettext(msgid) (msgid)
111 /* This define is so xgettext can find the internationalizable
113 # define gettext_noop(String) String
116 /* The `emacs' switch turns on certain matching commands
117 that make sense only in Emacs. */
123 /* Make syntax table lookup grant data in gl_state. */
124 # define SYNTAX_ENTRY_VIA_PROPERTY
127 # include "charset.h"
128 # include "category.h"
133 # define malloc xmalloc
137 # define realloc xrealloc
143 /* Converts the pointer to the char to BEG-based offset from the start. */
144 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
145 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
147 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
148 # define RE_STRING_CHAR(p, s) \
149 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
150 # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \
151 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
153 /* Set C a (possibly multibyte) character before P. P points into a
154 string which is the virtual concatenation of STR1 (which ends at
155 END1) or STR2 (which ends at END2). */
156 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
160 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
161 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
162 re_char *d0 = dtemp; \
163 PREV_CHAR_BOUNDARY (d0, dlimit); \
164 c = STRING_CHAR (d0, dtemp - d0); \
167 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
171 #else /* not emacs */
173 /* If we are not linking with Emacs proper,
174 we can't use the relocating allocator
175 even if config.h says that we can. */
178 # if defined STDC_HEADERS || defined _LIBC
185 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
192 val
= (void *) malloc (size
);
195 write (2, "virtual memory exhausted\n", 25);
202 xrealloc (block
, size
)
207 /* We must call malloc explicitly when BLOCK is 0, since some
208 reallocs don't do this. */
210 val
= (void *) malloc (size
);
212 val
= (void *) realloc (block
, size
);
215 write (2, "virtual memory exhausted\n", 25);
224 # define malloc xmalloc
228 # define realloc xrealloc
230 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
231 If nothing else has been done, use the method below. */
232 # ifdef INHIBIT_STRING_HEADER
233 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
234 # if !defined bzero && !defined bcopy
235 # undef INHIBIT_STRING_HEADER
240 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
241 This is used in most programs--a few other programs avoid this
242 by defining INHIBIT_STRING_HEADER. */
243 # ifndef INHIBIT_STRING_HEADER
244 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
248 # define bzero(s, n) (memset (s, '\0', n), (s))
250 # define bzero(s, n) __bzero (s, n)
254 # include <strings.h>
256 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
259 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
264 /* Define the syntax stuff for \<, \>, etc. */
266 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
267 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
269 # ifdef SWITCH_ENUM_BUG
270 # define SWITCH_ENUM_CAST(x) ((int)(x))
272 # define SWITCH_ENUM_CAST(x) (x)
275 /* Dummy macros for non-Emacs environments. */
276 # define BASE_LEADING_CODE_P(c) (0)
277 # define CHAR_CHARSET(c) 0
278 # define CHARSET_LEADING_CODE_BASE(c) 0
279 # define MAX_MULTIBYTE_LENGTH 1
280 # define RE_MULTIBYTE_P(x) 0
281 # define WORD_BOUNDARY_P(c1, c2) (0)
282 # define CHAR_HEAD_P(p) (1)
283 # define SINGLE_BYTE_CHAR_P(c) (1)
284 # define SAME_CHARSET_P(c1, c2) (1)
285 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
286 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
287 # define STRING_CHAR(p, s) (*(p))
288 # define RE_STRING_CHAR STRING_CHAR
289 # define CHAR_STRING(c, s) (*(s) = (c), 1)
290 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
291 # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH
292 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
293 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
294 # define MAKE_CHAR(charset, c1, c2) (c1)
295 #endif /* not emacs */
298 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
299 # define RE_TRANSLATE_P(TBL) (TBL)
302 /* Get the interface, including the syntax bits. */
305 /* isalpha etc. are used for the character classes. */
310 /* 1 if C is an ASCII character. */
311 # define IS_REAL_ASCII(c) ((c) < 0200)
313 /* 1 if C is a unibyte character. */
314 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
316 /* The Emacs definitions should not be directly affected by locales. */
318 /* In Emacs, these are only used for single-byte characters. */
319 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
320 # define ISCNTRL(c) ((c) < ' ')
321 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
322 || ((c) >= 'a' && (c) <= 'f') \
323 || ((c) >= 'A' && (c) <= 'F'))
325 /* This is only used for single-byte characters. */
326 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
328 /* The rest must handle multibyte characters. */
330 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
331 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
334 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
335 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
338 # define ISALNUM(c) (IS_REAL_ASCII (c) \
339 ? (((c) >= 'a' && (c) <= 'z') \
340 || ((c) >= 'A' && (c) <= 'Z') \
341 || ((c) >= '0' && (c) <= '9')) \
342 : SYNTAX (c) == Sword)
344 # define ISALPHA(c) (IS_REAL_ASCII (c) \
345 ? (((c) >= 'a' && (c) <= 'z') \
346 || ((c) >= 'A' && (c) <= 'Z')) \
347 : SYNTAX (c) == Sword)
349 # define ISLOWER(c) (LOWERCASEP (c))
351 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
352 ? ((c) > ' ' && (c) < 0177 \
353 && !(((c) >= 'a' && (c) <= 'z') \
354 || ((c) >= 'A' && (c) <= 'Z') \
355 || ((c) >= '0' && (c) <= '9'))) \
356 : SYNTAX (c) != Sword)
358 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
360 # define ISUPPER(c) (UPPERCASEP (c))
362 # define ISWORD(c) (SYNTAX (c) == Sword)
364 #else /* not emacs */
366 /* Jim Meyering writes:
368 "... Some ctype macros are valid only for character codes that
369 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
370 using /bin/cc or gcc but without giving an ansi option). So, all
371 ctype uses should be through macros like ISPRINT... If
372 STDC_HEADERS is defined, then autoconf has verified that the ctype
373 macros don't need to be guarded with references to isascii. ...
374 Defining isascii to 1 should let any compiler worth its salt
375 eliminate the && through constant folding."
376 Solaris defines some of these symbols so we must undefine them first. */
379 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
380 # define ISASCII(c) 1
382 # define ISASCII(c) isascii(c)
385 /* 1 if C is an ASCII character. */
386 # define IS_REAL_ASCII(c) ((c) < 0200)
388 /* This distinction is not meaningful, except in Emacs. */
389 # define ISUNIBYTE(c) 1
392 # define ISBLANK(c) (ISASCII (c) && isblank (c))
394 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
397 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
399 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
403 # define ISPRINT(c) (ISASCII (c) && isprint (c))
404 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
405 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
406 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
407 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
408 # define ISLOWER(c) (ISASCII (c) && islower (c))
409 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
410 # define ISSPACE(c) (ISASCII (c) && isspace (c))
411 # define ISUPPER(c) (ISASCII (c) && isupper (c))
412 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
414 # define ISWORD(c) ISALPHA(c)
417 # define TOLOWER(c) _tolower(c)
419 # define TOLOWER(c) tolower(c)
422 /* How many characters in the character set. */
423 # define CHAR_SET_SIZE 256
427 extern char *re_syntax_table
;
429 # else /* not SYNTAX_TABLE */
431 static char re_syntax_table
[CHAR_SET_SIZE
];
442 bzero (re_syntax_table
, sizeof re_syntax_table
);
444 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
446 re_syntax_table
[c
] = Sword
;
448 re_syntax_table
['_'] = Ssymbol
;
453 # endif /* not SYNTAX_TABLE */
455 # define SYNTAX(c) re_syntax_table[(c)]
457 #endif /* not emacs */
460 # define NULL (void *)0
463 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
464 since ours (we hope) works properly with all combinations of
465 machines, compilers, `char' and `unsigned char' argument types.
466 (Per Bothner suggested the basic approach.) */
467 #undef SIGN_EXTEND_CHAR
469 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
470 #else /* not __STDC__ */
471 /* As in Harbison and Steele. */
472 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
475 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
476 use `alloca' instead of `malloc'. This is because using malloc in
477 re_search* or re_match* could cause memory leaks when C-g is used in
478 Emacs; also, malloc is slower and causes storage fragmentation. On
479 the other hand, malloc is more portable, and easier to debug.
481 Because we sometimes use alloca, some routines have to be macros,
482 not functions -- `alloca'-allocated space disappears at the end of the
483 function it is called in. */
487 # define REGEX_ALLOCATE malloc
488 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
489 # define REGEX_FREE free
491 #else /* not REGEX_MALLOC */
493 /* Emacs already defines alloca, sometimes. */
496 /* Make alloca work the best possible way. */
498 # define alloca __builtin_alloca
499 # else /* not __GNUC__ */
502 # endif /* HAVE_ALLOCA_H */
503 # endif /* not __GNUC__ */
505 # endif /* not alloca */
507 # define REGEX_ALLOCATE alloca
509 /* Assumes a `char *destination' variable. */
510 # define REGEX_REALLOCATE(source, osize, nsize) \
511 (destination = (char *) alloca (nsize), \
512 memcpy (destination, source, osize))
514 /* No need to do anything to free, after alloca. */
515 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
517 #endif /* not REGEX_MALLOC */
519 /* Define how to allocate the failure stack. */
521 #if defined REL_ALLOC && defined REGEX_MALLOC
523 # define REGEX_ALLOCATE_STACK(size) \
524 r_alloc (&failure_stack_ptr, (size))
525 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
526 r_re_alloc (&failure_stack_ptr, (nsize))
527 # define REGEX_FREE_STACK(ptr) \
528 r_alloc_free (&failure_stack_ptr)
530 #else /* not using relocating allocator */
534 # define REGEX_ALLOCATE_STACK malloc
535 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
536 # define REGEX_FREE_STACK free
538 # else /* not REGEX_MALLOC */
540 # define REGEX_ALLOCATE_STACK alloca
542 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
543 REGEX_REALLOCATE (source, osize, nsize)
544 /* No need to explicitly free anything. */
545 # define REGEX_FREE_STACK(arg) ((void)0)
547 # endif /* not REGEX_MALLOC */
548 #endif /* not using relocating allocator */
551 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
552 `string1' or just past its end. This works if PTR is NULL, which is
554 #define FIRST_STRING_P(ptr) \
555 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
557 /* (Re)Allocate N items of type T using malloc, or fail. */
558 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
559 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
560 #define RETALLOC_IF(addr, n, t) \
561 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
562 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
564 #define BYTEWIDTH 8 /* In bits. */
566 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
570 #define MAX(a, b) ((a) > (b) ? (a) : (b))
571 #define MIN(a, b) ((a) < (b) ? (a) : (b))
573 /* Type of source-pattern and string chars. */
574 typedef const unsigned char re_char
;
576 typedef char boolean
;
580 static int re_match_2_internal
_RE_ARGS ((struct re_pattern_buffer
*bufp
,
581 re_char
*string1
, int size1
,
582 re_char
*string2
, int size2
,
584 struct re_registers
*regs
,
587 /* These are the command codes that appear in compiled regular
588 expressions. Some opcodes are followed by argument bytes. A
589 command code can specify any interpretation whatsoever for its
590 arguments. Zero bytes may appear in the compiled regular expression. */
596 /* Succeed right away--no more backtracking. */
599 /* Followed by one byte giving n, then by n literal bytes. */
602 /* Matches any (more or less) character. */
605 /* Matches any one char belonging to specified set. First
606 following byte is number of bitmap bytes. Then come bytes
607 for a bitmap saying which chars are in. Bits in each byte
608 are ordered low-bit-first. A character is in the set if its
609 bit is 1. A character too large to have a bit in the map is
610 automatically not in the set.
612 If the length byte has the 0x80 bit set, then that stuff
613 is followed by a range table:
614 2 bytes of flags for character sets (low 8 bits, high 8 bits)
615 See RANGE_TABLE_WORK_BITS below.
616 2 bytes, the number of pairs that follow (upto 32767)
617 pairs, each 2 multibyte characters,
618 each multibyte character represented as 3 bytes. */
621 /* Same parameters as charset, but match any character that is
622 not one of those specified. */
625 /* Start remembering the text that is matched, for storing in a
626 register. Followed by one byte with the register number, in
627 the range 0 to one less than the pattern buffer's re_nsub
631 /* Stop remembering the text that is matched and store it in a
632 memory register. Followed by one byte with the register
633 number, in the range 0 to one less than `re_nsub' in the
637 /* Match a duplicate of something remembered. Followed by one
638 byte containing the register number. */
641 /* Fail unless at beginning of line. */
644 /* Fail unless at end of line. */
647 /* Succeeds if at beginning of buffer (if emacs) or at beginning
648 of string to be matched (if not). */
651 /* Analogously, for end of buffer/string. */
654 /* Followed by two byte relative address to which to jump. */
657 /* Followed by two-byte relative address of place to resume at
658 in case of failure. */
661 /* Like on_failure_jump, but pushes a placeholder instead of the
662 current string position when executed. */
663 on_failure_keep_string_jump
,
665 /* Just like `on_failure_jump', except that it checks that we
666 don't get stuck in an infinite loop (matching an empty string
668 on_failure_jump_loop
,
670 /* Just like `on_failure_jump_loop', except that it checks for
671 a different kind of loop (the kind that shows up with non-greedy
672 operators). This operation has to be immediately preceded
674 on_failure_jump_nastyloop
,
676 /* A smart `on_failure_jump' used for greedy * and + operators.
677 It analyses the loop before which it is put and if the
678 loop does not require backtracking, it changes itself to
679 `on_failure_keep_string_jump' and short-circuits the loop,
680 else it just defaults to changing itself into `on_failure_jump'.
681 It assumes that it is pointing to just past a `jump'. */
682 on_failure_jump_smart
,
684 /* Followed by two-byte relative address and two-byte number n.
685 After matching N times, jump to the address upon failure.
686 Does not work if N starts at 0: use on_failure_jump_loop
690 /* Followed by two-byte relative address, and two-byte number n.
691 Jump to the address N times, then fail. */
694 /* Set the following two-byte relative address to the
695 subsequent two-byte number. The address *includes* the two
699 wordbeg
, /* Succeeds if at word beginning. */
700 wordend
, /* Succeeds if at word end. */
702 wordbound
, /* Succeeds if at a word boundary. */
703 notwordbound
, /* Succeeds if not at a word boundary. */
705 symbeg
, /* Succeeds if at symbol beginning. */
706 symend
, /* Succeeds if at symbol end. */
708 /* Matches any character whose syntax is specified. Followed by
709 a byte which contains a syntax code, e.g., Sword. */
712 /* Matches any character whose syntax is not that specified. */
716 ,before_dot
, /* Succeeds if before point. */
717 at_dot
, /* Succeeds if at point. */
718 after_dot
, /* Succeeds if after point. */
720 /* Matches any character whose category-set contains the specified
721 category. The operator is followed by a byte which contains a
722 category code (mnemonic ASCII character). */
725 /* Matches any character whose category-set does not contain the
726 specified category. The operator is followed by a byte which
727 contains the category code (mnemonic ASCII character). */
732 /* Common operations on the compiled pattern. */
734 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
736 #define STORE_NUMBER(destination, number) \
738 (destination)[0] = (number) & 0377; \
739 (destination)[1] = (number) >> 8; \
742 /* Same as STORE_NUMBER, except increment DESTINATION to
743 the byte after where the number is stored. Therefore, DESTINATION
744 must be an lvalue. */
746 #define STORE_NUMBER_AND_INCR(destination, number) \
748 STORE_NUMBER (destination, number); \
749 (destination) += 2; \
752 /* Put into DESTINATION a number stored in two contiguous bytes starting
755 #define EXTRACT_NUMBER(destination, source) \
757 (destination) = *(source) & 0377; \
758 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
762 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
764 extract_number (dest
, source
)
768 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
769 *dest
= *source
& 0377;
773 # ifndef EXTRACT_MACROS /* To debug the macros. */
774 # undef EXTRACT_NUMBER
775 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
776 # endif /* not EXTRACT_MACROS */
780 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
781 SOURCE must be an lvalue. */
783 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
785 EXTRACT_NUMBER (destination, source); \
790 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
793 extract_number_and_incr (destination
, source
)
797 extract_number (destination
, *source
);
801 # ifndef EXTRACT_MACROS
802 # undef EXTRACT_NUMBER_AND_INCR
803 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
804 extract_number_and_incr (&dest, &src)
805 # endif /* not EXTRACT_MACROS */
809 /* Store a multibyte character in three contiguous bytes starting
810 DESTINATION, and increment DESTINATION to the byte after where the
811 character is stored. Therefore, DESTINATION must be an lvalue. */
813 #define STORE_CHARACTER_AND_INCR(destination, character) \
815 (destination)[0] = (character) & 0377; \
816 (destination)[1] = ((character) >> 8) & 0377; \
817 (destination)[2] = (character) >> 16; \
818 (destination) += 3; \
821 /* Put into DESTINATION a character stored in three contiguous bytes
822 starting at SOURCE. */
824 #define EXTRACT_CHARACTER(destination, source) \
826 (destination) = ((source)[0] \
827 | ((source)[1] << 8) \
828 | ((source)[2] << 16)); \
832 /* Macros for charset. */
834 /* Size of bitmap of charset P in bytes. P is a start of charset,
835 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
836 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
838 /* Nonzero if charset P has range table. */
839 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
841 /* Return the address of range table of charset P. But not the start
842 of table itself, but the before where the number of ranges is
843 stored. `2 +' means to skip re_opcode_t and size of bitmap,
844 and the 2 bytes of flags at the start of the range table. */
845 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
847 /* Extract the bit flags that start a range table. */
848 #define CHARSET_RANGE_TABLE_BITS(p) \
849 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
850 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
852 /* Test if C is listed in the bitmap of charset P. */
853 #define CHARSET_LOOKUP_BITMAP(p, c) \
854 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
855 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
857 /* Return the address of end of RANGE_TABLE. COUNT is number of
858 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
859 is start of range and end of range. `* 3' is size of each start
861 #define CHARSET_RANGE_TABLE_END(range_table, count) \
862 ((range_table) + (count) * 2 * 3)
864 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
865 COUNT is number of ranges in RANGE_TABLE. */
866 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
869 re_wchar_t range_start, range_end; \
871 re_char *range_table_end \
872 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
874 for (p = (range_table); p < range_table_end; p += 2 * 3) \
876 EXTRACT_CHARACTER (range_start, p); \
877 EXTRACT_CHARACTER (range_end, p + 3); \
879 if (range_start <= (c) && (c) <= range_end) \
888 /* Test if C is in range table of CHARSET. The flag NOT is negated if
889 C is listed in it. */
890 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
893 /* Number of ranges in range table. */ \
895 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
897 EXTRACT_NUMBER_AND_INCR (count, range_table); \
898 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
902 /* If DEBUG is defined, Regex prints many voluminous messages about what
903 it is doing (if the variable `debug' is nonzero). If linked with the
904 main program in `iregex.c', you can enter patterns and strings
905 interactively. And if linked with the main program in `main.c' and
906 the other test files, you can run the already-written tests. */
910 /* We use standard I/O for debugging. */
913 /* It is useful to test things that ``must'' be true when debugging. */
916 static int debug
= -100000;
918 # define DEBUG_STATEMENT(e) e
919 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
920 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
921 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
922 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
923 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
924 if (debug > 0) print_partial_compiled_pattern (s, e)
925 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
926 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
929 /* Print the fastmap in human-readable form. */
932 print_fastmap (fastmap
)
935 unsigned was_a_range
= 0;
938 while (i
< (1 << BYTEWIDTH
))
944 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
960 /* Print a compiled pattern string in human-readable form, starting at
961 the START pointer into it and ending just before the pointer END. */
964 print_partial_compiled_pattern (start
, end
)
974 fprintf (stderr
, "(null)\n");
978 /* Loop over pattern commands. */
981 fprintf (stderr
, "%d:\t", p
- start
);
983 switch ((re_opcode_t
) *p
++)
986 fprintf (stderr
, "/no_op");
990 fprintf (stderr
, "/succeed");
995 fprintf (stderr
, "/exactn/%d", mcnt
);
998 fprintf (stderr
, "/%c", *p
++);
1004 fprintf (stderr
, "/start_memory/%d", *p
++);
1008 fprintf (stderr
, "/stop_memory/%d", *p
++);
1012 fprintf (stderr
, "/duplicate/%d", *p
++);
1016 fprintf (stderr
, "/anychar");
1022 register int c
, last
= -100;
1023 register int in_range
= 0;
1024 int length
= CHARSET_BITMAP_SIZE (p
- 1);
1025 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
1027 fprintf (stderr
, "/charset [%s",
1028 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
1031 fprintf (stderr
, " !extends past end of pattern! ");
1033 for (c
= 0; c
< 256; c
++)
1035 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
1037 /* Are we starting a range? */
1038 if (last
+ 1 == c
&& ! in_range
)
1040 fprintf (stderr
, "-");
1043 /* Have we broken a range? */
1044 else if (last
+ 1 != c
&& in_range
)
1046 fprintf (stderr
, "%c", last
);
1051 fprintf (stderr
, "%c", c
);
1057 fprintf (stderr
, "%c", last
);
1059 fprintf (stderr
, "]");
1063 if (has_range_table
)
1066 fprintf (stderr
, "has-range-table");
1068 /* ??? Should print the range table; for now, just skip it. */
1069 p
+= 2; /* skip range table bits */
1070 EXTRACT_NUMBER_AND_INCR (count
, p
);
1071 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1077 fprintf (stderr
, "/begline");
1081 fprintf (stderr
, "/endline");
1084 case on_failure_jump
:
1085 extract_number_and_incr (&mcnt
, &p
);
1086 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1089 case on_failure_keep_string_jump
:
1090 extract_number_and_incr (&mcnt
, &p
);
1091 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1094 case on_failure_jump_nastyloop
:
1095 extract_number_and_incr (&mcnt
, &p
);
1096 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1099 case on_failure_jump_loop
:
1100 extract_number_and_incr (&mcnt
, &p
);
1101 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1104 case on_failure_jump_smart
:
1105 extract_number_and_incr (&mcnt
, &p
);
1106 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1110 extract_number_and_incr (&mcnt
, &p
);
1111 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1115 extract_number_and_incr (&mcnt
, &p
);
1116 extract_number_and_incr (&mcnt2
, &p
);
1117 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1121 extract_number_and_incr (&mcnt
, &p
);
1122 extract_number_and_incr (&mcnt2
, &p
);
1123 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1127 extract_number_and_incr (&mcnt
, &p
);
1128 extract_number_and_incr (&mcnt2
, &p
);
1129 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1133 fprintf (stderr
, "/wordbound");
1137 fprintf (stderr
, "/notwordbound");
1141 fprintf (stderr
, "/wordbeg");
1145 fprintf (stderr
, "/wordend");
1149 fprintf (stderr
, "/symbeg");
1153 fprintf (stderr
, "/symend");
1157 fprintf (stderr
, "/syntaxspec");
1159 fprintf (stderr
, "/%d", mcnt
);
1163 fprintf (stderr
, "/notsyntaxspec");
1165 fprintf (stderr
, "/%d", mcnt
);
1170 fprintf (stderr
, "/before_dot");
1174 fprintf (stderr
, "/at_dot");
1178 fprintf (stderr
, "/after_dot");
1182 fprintf (stderr
, "/categoryspec");
1184 fprintf (stderr
, "/%d", mcnt
);
1187 case notcategoryspec
:
1188 fprintf (stderr
, "/notcategoryspec");
1190 fprintf (stderr
, "/%d", mcnt
);
1195 fprintf (stderr
, "/begbuf");
1199 fprintf (stderr
, "/endbuf");
1203 fprintf (stderr
, "?%d", *(p
-1));
1206 fprintf (stderr
, "\n");
1209 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1214 print_compiled_pattern (bufp
)
1215 struct re_pattern_buffer
*bufp
;
1217 re_char
*buffer
= bufp
->buffer
;
1219 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1220 printf ("%ld bytes used/%ld bytes allocated.\n",
1221 bufp
->used
, bufp
->allocated
);
1223 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1225 printf ("fastmap: ");
1226 print_fastmap (bufp
->fastmap
);
1229 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1230 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1231 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1232 printf ("no_sub: %d\t", bufp
->no_sub
);
1233 printf ("not_bol: %d\t", bufp
->not_bol
);
1234 printf ("not_eol: %d\t", bufp
->not_eol
);
1235 printf ("syntax: %lx\n", bufp
->syntax
);
1237 /* Perhaps we should print the translate table? */
1242 print_double_string (where
, string1
, size1
, string2
, size2
)
1255 if (FIRST_STRING_P (where
))
1257 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1258 putchar (string1
[this_char
]);
1263 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1264 putchar (string2
[this_char
]);
1268 #else /* not DEBUG */
1273 # define DEBUG_STATEMENT(e)
1274 # define DEBUG_PRINT1(x)
1275 # define DEBUG_PRINT2(x1, x2)
1276 # define DEBUG_PRINT3(x1, x2, x3)
1277 # define DEBUG_PRINT4(x1, x2, x3, x4)
1278 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1279 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1281 #endif /* not DEBUG */
1283 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1284 also be assigned to arbitrarily: each pattern buffer stores its own
1285 syntax, so it can be changed between regex compilations. */
1286 /* This has no initializer because initialized variables in Emacs
1287 become read-only after dumping. */
1288 reg_syntax_t re_syntax_options
;
1291 /* Specify the precise syntax of regexps for compilation. This provides
1292 for compatibility for various utilities which historically have
1293 different, incompatible syntaxes.
1295 The argument SYNTAX is a bit mask comprised of the various bits
1296 defined in regex.h. We return the old syntax. */
1299 re_set_syntax (syntax
)
1300 reg_syntax_t syntax
;
1302 reg_syntax_t ret
= re_syntax_options
;
1304 re_syntax_options
= syntax
;
1307 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1309 /* Regexp to use to replace spaces, or NULL meaning don't. */
1310 static re_char
*whitespace_regexp
;
1313 re_set_whitespace_regexp (regexp
)
1316 whitespace_regexp
= (re_char
*) regexp
;
1318 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1320 /* This table gives an error message for each of the error codes listed
1321 in regex.h. Obviously the order here has to be same as there.
1322 POSIX doesn't require that we do anything for REG_NOERROR,
1323 but why not be nice? */
1325 static const char *re_error_msgid
[] =
1327 gettext_noop ("Success"), /* REG_NOERROR */
1328 gettext_noop ("No match"), /* REG_NOMATCH */
1329 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1330 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1331 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1332 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1333 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1334 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1335 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1336 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1337 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1338 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1339 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1340 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1341 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1342 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1343 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1344 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1347 /* Avoiding alloca during matching, to placate r_alloc. */
1349 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1350 searching and matching functions should not call alloca. On some
1351 systems, alloca is implemented in terms of malloc, and if we're
1352 using the relocating allocator routines, then malloc could cause a
1353 relocation, which might (if the strings being searched are in the
1354 ralloc heap) shift the data out from underneath the regexp
1357 Here's another reason to avoid allocation: Emacs
1358 processes input from X in a signal handler; processing X input may
1359 call malloc; if input arrives while a matching routine is calling
1360 malloc, then we're scrod. But Emacs can't just block input while
1361 calling matching routines; then we don't notice interrupts when
1362 they come in. So, Emacs blocks input around all regexp calls
1363 except the matching calls, which it leaves unprotected, in the
1364 faith that they will not malloc. */
1366 /* Normally, this is fine. */
1367 #define MATCH_MAY_ALLOCATE
1369 /* When using GNU C, we are not REALLY using the C alloca, no matter
1370 what config.h may say. So don't take precautions for it. */
1375 /* The match routines may not allocate if (1) they would do it with malloc
1376 and (2) it's not safe for them to use malloc.
1377 Note that if REL_ALLOC is defined, matching would not use malloc for the
1378 failure stack, but we would still use it for the register vectors;
1379 so REL_ALLOC should not affect this. */
1380 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1381 # undef MATCH_MAY_ALLOCATE
1385 /* Failure stack declarations and macros; both re_compile_fastmap and
1386 re_match_2 use a failure stack. These have to be macros because of
1387 REGEX_ALLOCATE_STACK. */
1390 /* Approximate number of failure points for which to initially allocate space
1391 when matching. If this number is exceeded, we allocate more
1392 space, so it is not a hard limit. */
1393 #ifndef INIT_FAILURE_ALLOC
1394 # define INIT_FAILURE_ALLOC 20
1397 /* Roughly the maximum number of failure points on the stack. Would be
1398 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1399 This is a variable only so users of regex can assign to it; we never
1400 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1401 before using it, so it should probably be a byte-count instead. */
1402 # if defined MATCH_MAY_ALLOCATE
1403 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1404 whose default stack limit is 2mb. In order for a larger
1405 value to work reliably, you have to try to make it accord
1406 with the process stack limit. */
1407 size_t re_max_failures
= 40000;
1409 size_t re_max_failures
= 4000;
1412 union fail_stack_elt
1415 /* This should be the biggest `int' that's no bigger than a pointer. */
1419 typedef union fail_stack_elt fail_stack_elt_t
;
1423 fail_stack_elt_t
*stack
;
1425 size_t avail
; /* Offset of next open position. */
1426 size_t frame
; /* Offset of the cur constructed frame. */
1429 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1430 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1433 /* Define macros to initialize and free the failure stack.
1434 Do `return -2' if the alloc fails. */
1436 #ifdef MATCH_MAY_ALLOCATE
1437 # define INIT_FAIL_STACK() \
1439 fail_stack.stack = (fail_stack_elt_t *) \
1440 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1441 * sizeof (fail_stack_elt_t)); \
1443 if (fail_stack.stack == NULL) \
1446 fail_stack.size = INIT_FAILURE_ALLOC; \
1447 fail_stack.avail = 0; \
1448 fail_stack.frame = 0; \
1451 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1453 # define INIT_FAIL_STACK() \
1455 fail_stack.avail = 0; \
1456 fail_stack.frame = 0; \
1459 # define RESET_FAIL_STACK() ((void)0)
1463 /* Double the size of FAIL_STACK, up to a limit
1464 which allows approximately `re_max_failures' items.
1466 Return 1 if succeeds, and 0 if either ran out of memory
1467 allocating space for it or it was already too large.
1469 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1471 /* Factor to increase the failure stack size by
1472 when we increase it.
1473 This used to be 2, but 2 was too wasteful
1474 because the old discarded stacks added up to as much space
1475 were as ultimate, maximum-size stack. */
1476 #define FAIL_STACK_GROWTH_FACTOR 4
1478 #define GROW_FAIL_STACK(fail_stack) \
1479 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1480 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1482 : ((fail_stack).stack \
1483 = (fail_stack_elt_t *) \
1484 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1485 (fail_stack).size * sizeof (fail_stack_elt_t), \
1486 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1487 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1488 * FAIL_STACK_GROWTH_FACTOR))), \
1490 (fail_stack).stack == NULL \
1492 : ((fail_stack).size \
1493 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1494 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1495 * FAIL_STACK_GROWTH_FACTOR)) \
1496 / sizeof (fail_stack_elt_t)), \
1500 /* Push a pointer value onto the failure stack.
1501 Assumes the variable `fail_stack'. Probably should only
1502 be called from within `PUSH_FAILURE_POINT'. */
1503 #define PUSH_FAILURE_POINTER(item) \
1504 fail_stack.stack[fail_stack.avail++].pointer = (item)
1506 /* This pushes an integer-valued item onto the failure stack.
1507 Assumes the variable `fail_stack'. Probably should only
1508 be called from within `PUSH_FAILURE_POINT'. */
1509 #define PUSH_FAILURE_INT(item) \
1510 fail_stack.stack[fail_stack.avail++].integer = (item)
1512 /* Push a fail_stack_elt_t value onto the failure stack.
1513 Assumes the variable `fail_stack'. Probably should only
1514 be called from within `PUSH_FAILURE_POINT'. */
1515 #define PUSH_FAILURE_ELT(item) \
1516 fail_stack.stack[fail_stack.avail++] = (item)
1518 /* These three POP... operations complement the three PUSH... operations.
1519 All assume that `fail_stack' is nonempty. */
1520 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1521 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1522 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1524 /* Individual items aside from the registers. */
1525 #define NUM_NONREG_ITEMS 3
1527 /* Used to examine the stack (to detect infinite loops). */
1528 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1529 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1530 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1531 #define TOP_FAILURE_HANDLE() fail_stack.frame
1534 #define ENSURE_FAIL_STACK(space) \
1535 while (REMAINING_AVAIL_SLOTS <= space) { \
1536 if (!GROW_FAIL_STACK (fail_stack)) \
1538 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1539 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1542 /* Push register NUM onto the stack. */
1543 #define PUSH_FAILURE_REG(num) \
1545 char *destination; \
1546 ENSURE_FAIL_STACK(3); \
1547 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1548 num, regstart[num], regend[num]); \
1549 PUSH_FAILURE_POINTER (regstart[num]); \
1550 PUSH_FAILURE_POINTER (regend[num]); \
1551 PUSH_FAILURE_INT (num); \
1554 /* Change the counter's value to VAL, but make sure that it will
1555 be reset when backtracking. */
1556 #define PUSH_NUMBER(ptr,val) \
1558 char *destination; \
1560 ENSURE_FAIL_STACK(3); \
1561 EXTRACT_NUMBER (c, ptr); \
1562 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1563 PUSH_FAILURE_INT (c); \
1564 PUSH_FAILURE_POINTER (ptr); \
1565 PUSH_FAILURE_INT (-1); \
1566 STORE_NUMBER (ptr, val); \
1569 /* Pop a saved register off the stack. */
1570 #define POP_FAILURE_REG_OR_COUNT() \
1572 int reg = POP_FAILURE_INT (); \
1575 /* It's a counter. */ \
1576 /* Here, we discard `const', making re_match non-reentrant. */ \
1577 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1578 reg = POP_FAILURE_INT (); \
1579 STORE_NUMBER (ptr, reg); \
1580 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1584 regend[reg] = POP_FAILURE_POINTER (); \
1585 regstart[reg] = POP_FAILURE_POINTER (); \
1586 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1587 reg, regstart[reg], regend[reg]); \
1591 /* Check that we are not stuck in an infinite loop. */
1592 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1594 int failure = TOP_FAILURE_HANDLE (); \
1595 /* Check for infinite matching loops */ \
1596 while (failure > 0 \
1597 && (FAILURE_STR (failure) == string_place \
1598 || FAILURE_STR (failure) == NULL)) \
1600 assert (FAILURE_PAT (failure) >= bufp->buffer \
1601 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1602 if (FAILURE_PAT (failure) == pat_cur) \
1607 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1608 failure = NEXT_FAILURE_HANDLE(failure); \
1610 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1613 /* Push the information about the state we will need
1614 if we ever fail back to it.
1616 Requires variables fail_stack, regstart, regend and
1617 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1620 Does `return FAILURE_CODE' if runs out of memory. */
1622 #define PUSH_FAILURE_POINT(pattern, string_place) \
1624 char *destination; \
1625 /* Must be int, so when we don't save any registers, the arithmetic \
1626 of 0 + -1 isn't done as unsigned. */ \
1628 DEBUG_STATEMENT (nfailure_points_pushed++); \
1629 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1630 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1631 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1633 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1635 DEBUG_PRINT1 ("\n"); \
1637 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1638 PUSH_FAILURE_INT (fail_stack.frame); \
1640 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1641 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1642 DEBUG_PRINT1 ("'\n"); \
1643 PUSH_FAILURE_POINTER (string_place); \
1645 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1646 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1647 PUSH_FAILURE_POINTER (pattern); \
1649 /* Close the frame by moving the frame pointer past it. */ \
1650 fail_stack.frame = fail_stack.avail; \
1653 /* Estimate the size of data pushed by a typical failure stack entry.
1654 An estimate is all we need, because all we use this for
1655 is to choose a limit for how big to make the failure stack. */
1656 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1657 #define TYPICAL_FAILURE_SIZE 20
1659 /* How many items can still be added to the stack without overflowing it. */
1660 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1663 /* Pops what PUSH_FAIL_STACK pushes.
1665 We restore into the parameters, all of which should be lvalues:
1666 STR -- the saved data position.
1667 PAT -- the saved pattern position.
1668 REGSTART, REGEND -- arrays of string positions.
1670 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1671 `pend', `string1', `size1', `string2', and `size2'. */
1673 #define POP_FAILURE_POINT(str, pat) \
1675 assert (!FAIL_STACK_EMPTY ()); \
1677 /* Remove failure points and point to how many regs pushed. */ \
1678 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1679 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1680 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1682 /* Pop the saved registers. */ \
1683 while (fail_stack.frame < fail_stack.avail) \
1684 POP_FAILURE_REG_OR_COUNT (); \
1686 pat = POP_FAILURE_POINTER (); \
1687 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1688 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1690 /* If the saved string location is NULL, it came from an \
1691 on_failure_keep_string_jump opcode, and we want to throw away the \
1692 saved NULL, thus retaining our current position in the string. */ \
1693 str = POP_FAILURE_POINTER (); \
1694 DEBUG_PRINT2 (" Popping string %p: `", str); \
1695 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1696 DEBUG_PRINT1 ("'\n"); \
1698 fail_stack.frame = POP_FAILURE_INT (); \
1699 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1701 assert (fail_stack.avail >= 0); \
1702 assert (fail_stack.frame <= fail_stack.avail); \
1704 DEBUG_STATEMENT (nfailure_points_popped++); \
1705 } while (0) /* POP_FAILURE_POINT */
1709 /* Registers are set to a sentinel when they haven't yet matched. */
1710 #define REG_UNSET(e) ((e) == NULL)
1712 /* Subroutine declarations and macros for regex_compile. */
1714 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1715 reg_syntax_t syntax
,
1716 struct re_pattern_buffer
*bufp
));
1717 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1718 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1719 int arg1
, int arg2
));
1720 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1721 int arg
, unsigned char *end
));
1722 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1723 int arg1
, int arg2
, unsigned char *end
));
1724 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1726 reg_syntax_t syntax
));
1727 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1729 reg_syntax_t syntax
));
1730 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1731 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1732 char *fastmap
, const int multibyte
));
1734 /* Fetch the next character in the uncompiled pattern, with no
1736 #define PATFETCH(c) \
1739 if (p == pend) return REG_EEND; \
1740 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1745 /* If `translate' is non-null, return translate[D], else just D. We
1746 cast the subscript to translate because some data is declared as
1747 `char *', to avoid warnings when a string constant is passed. But
1748 when we use a character as a subscript we must make it unsigned. */
1750 # define TRANSLATE(d) \
1751 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1755 /* Macros for outputting the compiled pattern into `buffer'. */
1757 /* If the buffer isn't allocated when it comes in, use this. */
1758 #define INIT_BUF_SIZE 32
1760 /* Make sure we have at least N more bytes of space in buffer. */
1761 #define GET_BUFFER_SPACE(n) \
1762 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1765 /* Make sure we have one more byte of buffer space and then add C to it. */
1766 #define BUF_PUSH(c) \
1768 GET_BUFFER_SPACE (1); \
1769 *b++ = (unsigned char) (c); \
1773 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1774 #define BUF_PUSH_2(c1, c2) \
1776 GET_BUFFER_SPACE (2); \
1777 *b++ = (unsigned char) (c1); \
1778 *b++ = (unsigned char) (c2); \
1782 /* As with BUF_PUSH_2, except for three bytes. */
1783 #define BUF_PUSH_3(c1, c2, c3) \
1785 GET_BUFFER_SPACE (3); \
1786 *b++ = (unsigned char) (c1); \
1787 *b++ = (unsigned char) (c2); \
1788 *b++ = (unsigned char) (c3); \
1792 /* Store a jump with opcode OP at LOC to location TO. We store a
1793 relative address offset by the three bytes the jump itself occupies. */
1794 #define STORE_JUMP(op, loc, to) \
1795 store_op1 (op, loc, (to) - (loc) - 3)
1797 /* Likewise, for a two-argument jump. */
1798 #define STORE_JUMP2(op, loc, to, arg) \
1799 store_op2 (op, loc, (to) - (loc) - 3, arg)
1801 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1802 #define INSERT_JUMP(op, loc, to) \
1803 insert_op1 (op, loc, (to) - (loc) - 3, b)
1805 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1806 #define INSERT_JUMP2(op, loc, to, arg) \
1807 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1810 /* This is not an arbitrary limit: the arguments which represent offsets
1811 into the pattern are two bytes long. So if 2^15 bytes turns out to
1812 be too small, many things would have to change. */
1813 # define MAX_BUF_SIZE (1L << 15)
1815 #if 0 /* This is when we thought it could be 2^16 bytes. */
1816 /* Any other compiler which, like MSC, has allocation limit below 2^16
1817 bytes will have to use approach similar to what was done below for
1818 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1819 reallocating to 0 bytes. Such thing is not going to work too well.
1820 You have been warned!! */
1821 #if defined _MSC_VER && !defined WIN32
1822 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1823 # define MAX_BUF_SIZE 65500L
1825 # define MAX_BUF_SIZE (1L << 16)
1829 /* Extend the buffer by twice its current size via realloc and
1830 reset the pointers that pointed into the old block to point to the
1831 correct places in the new one. If extending the buffer results in it
1832 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1833 #if __BOUNDED_POINTERS__
1834 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1835 # define MOVE_BUFFER_POINTER(P) \
1836 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1837 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1840 SET_HIGH_BOUND (b); \
1841 SET_HIGH_BOUND (begalt); \
1842 if (fixup_alt_jump) \
1843 SET_HIGH_BOUND (fixup_alt_jump); \
1845 SET_HIGH_BOUND (laststart); \
1846 if (pending_exact) \
1847 SET_HIGH_BOUND (pending_exact); \
1850 # define MOVE_BUFFER_POINTER(P) (P) += incr
1851 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1853 #define EXTEND_BUFFER() \
1855 re_char *old_buffer = bufp->buffer; \
1856 if (bufp->allocated == MAX_BUF_SIZE) \
1858 bufp->allocated <<= 1; \
1859 if (bufp->allocated > MAX_BUF_SIZE) \
1860 bufp->allocated = MAX_BUF_SIZE; \
1861 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1862 if (bufp->buffer == NULL) \
1863 return REG_ESPACE; \
1864 /* If the buffer moved, move all the pointers into it. */ \
1865 if (old_buffer != bufp->buffer) \
1867 int incr = bufp->buffer - old_buffer; \
1868 MOVE_BUFFER_POINTER (b); \
1869 MOVE_BUFFER_POINTER (begalt); \
1870 if (fixup_alt_jump) \
1871 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1873 MOVE_BUFFER_POINTER (laststart); \
1874 if (pending_exact) \
1875 MOVE_BUFFER_POINTER (pending_exact); \
1877 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1881 /* Since we have one byte reserved for the register number argument to
1882 {start,stop}_memory, the maximum number of groups we can report
1883 things about is what fits in that byte. */
1884 #define MAX_REGNUM 255
1886 /* But patterns can have more than `MAX_REGNUM' registers. We just
1887 ignore the excess. */
1888 typedef int regnum_t
;
1891 /* Macros for the compile stack. */
1893 /* Since offsets can go either forwards or backwards, this type needs to
1894 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1895 /* int may be not enough when sizeof(int) == 2. */
1896 typedef long pattern_offset_t
;
1900 pattern_offset_t begalt_offset
;
1901 pattern_offset_t fixup_alt_jump
;
1902 pattern_offset_t laststart_offset
;
1904 } compile_stack_elt_t
;
1909 compile_stack_elt_t
*stack
;
1911 unsigned avail
; /* Offset of next open position. */
1912 } compile_stack_type
;
1915 #define INIT_COMPILE_STACK_SIZE 32
1917 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1918 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1920 /* The next available element. */
1921 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1923 /* Explicit quit checking is only used on NTemacs and whenever we
1924 use polling to process input events. */
1925 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1926 extern int immediate_quit
;
1927 # define IMMEDIATE_QUIT_CHECK \
1929 if (immediate_quit) QUIT; \
1932 # define IMMEDIATE_QUIT_CHECK ((void)0)
1935 /* Structure to manage work area for range table. */
1936 struct range_table_work_area
1938 int *table
; /* actual work area. */
1939 int allocated
; /* allocated size for work area in bytes. */
1940 int used
; /* actually used size in words. */
1941 int bits
; /* flag to record character classes */
1944 /* Make sure that WORK_AREA can hold more N multibyte characters.
1945 This is used only in set_image_of_range and set_image_of_range_1.
1946 It expects WORK_AREA to be a pointer.
1947 If it can't get the space, it returns from the surrounding function. */
1949 #define EXTEND_RANGE_TABLE(work_area, n) \
1951 if (((work_area)->used + (n)) * sizeof (int) > (work_area)->allocated) \
1953 extend_range_table_work_area (work_area); \
1954 if ((work_area)->table == 0) \
1955 return (REG_ESPACE); \
1959 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1960 (work_area).bits |= (bit)
1962 /* Bits used to implement the multibyte-part of the various character classes
1963 such as [:alnum:] in a charset's range table. */
1964 #define BIT_WORD 0x1
1965 #define BIT_LOWER 0x2
1966 #define BIT_PUNCT 0x4
1967 #define BIT_SPACE 0x8
1968 #define BIT_UPPER 0x10
1969 #define BIT_MULTIBYTE 0x20
1971 /* Set a range START..END to WORK_AREA.
1972 The range is passed through TRANSLATE, so START and END
1973 should be untranslated. */
1974 #define SET_RANGE_TABLE_WORK_AREA(work_area, start, end) \
1977 tem = set_image_of_range (&work_area, start, end, translate); \
1979 FREE_STACK_RETURN (tem); \
1982 /* Free allocated memory for WORK_AREA. */
1983 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1985 if ((work_area).table) \
1986 free ((work_area).table); \
1989 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1990 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1991 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1992 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1995 /* Set the bit for character C in a list. */
1996 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1999 /* Get the next unsigned number in the uncompiled pattern. */
2000 #define GET_UNSIGNED_NUMBER(num) \
2003 FREE_STACK_RETURN (REG_EBRACE); \
2007 while ('0' <= c && c <= '9') \
2013 num = num * 10 + c - '0'; \
2014 if (num / 10 != prev) \
2015 FREE_STACK_RETURN (REG_BADBR); \
2017 FREE_STACK_RETURN (REG_EBRACE); \
2023 #if ! WIDE_CHAR_SUPPORT
2025 /* Map a string to the char class it names (if any). */
2030 const char *string
= str
;
2031 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2032 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2033 else if (STREQ (string
, "word")) return RECC_WORD
;
2034 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2035 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2036 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2037 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2038 else if (STREQ (string
, "print")) return RECC_PRINT
;
2039 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2040 else if (STREQ (string
, "space")) return RECC_SPACE
;
2041 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2042 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2043 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2044 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2045 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2046 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2047 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2051 /* True if CH is in the char class CC. */
2053 re_iswctype (ch
, cc
)
2059 case RECC_ALNUM
: return ISALNUM (ch
);
2060 case RECC_ALPHA
: return ISALPHA (ch
);
2061 case RECC_BLANK
: return ISBLANK (ch
);
2062 case RECC_CNTRL
: return ISCNTRL (ch
);
2063 case RECC_DIGIT
: return ISDIGIT (ch
);
2064 case RECC_GRAPH
: return ISGRAPH (ch
);
2065 case RECC_LOWER
: return ISLOWER (ch
);
2066 case RECC_PRINT
: return ISPRINT (ch
);
2067 case RECC_PUNCT
: return ISPUNCT (ch
);
2068 case RECC_SPACE
: return ISSPACE (ch
);
2069 case RECC_UPPER
: return ISUPPER (ch
);
2070 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2071 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2072 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2073 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2074 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2075 case RECC_WORD
: return ISWORD (ch
);
2076 case RECC_ERROR
: return false;
2082 /* Return a bit-pattern to use in the range-table bits to match multibyte
2083 chars of class CC. */
2085 re_wctype_to_bit (cc
)
2090 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2091 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2092 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2093 case RECC_LOWER
: return BIT_LOWER
;
2094 case RECC_UPPER
: return BIT_UPPER
;
2095 case RECC_PUNCT
: return BIT_PUNCT
;
2096 case RECC_SPACE
: return BIT_SPACE
;
2097 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2098 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2105 /* Filling in the work area of a range. */
2107 /* Actually extend the space in WORK_AREA. */
2110 extend_range_table_work_area (work_area
)
2111 struct range_table_work_area
*work_area
;
2113 work_area
->allocated
+= 16 * sizeof (int);
2114 if (work_area
->table
)
2116 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2119 = (int *) malloc (work_area
->allocated
);
2124 /* Carefully find the ranges of codes that are equivalent
2125 under case conversion to the range start..end when passed through
2126 TRANSLATE. Handle the case where non-letters can come in between
2127 two upper-case letters (which happens in Latin-1).
2128 Also handle the case of groups of more than 2 case-equivalent chars.
2130 The basic method is to look at consecutive characters and see
2131 if they can form a run that can be handled as one.
2133 Returns -1 if successful, REG_ESPACE if ran out of space. */
2136 set_image_of_range_1 (work_area
, start
, end
, translate
)
2137 RE_TRANSLATE_TYPE translate
;
2138 struct range_table_work_area
*work_area
;
2139 re_wchar_t start
, end
;
2141 /* `one_case' indicates a character, or a run of characters,
2142 each of which is an isolate (no case-equivalents).
2143 This includes all ASCII non-letters.
2145 `two_case' indicates a character, or a run of characters,
2146 each of which has two case-equivalent forms.
2147 This includes all ASCII letters.
2149 `strange' indicates a character that has more than one
2152 enum case_type
{one_case
, two_case
, strange
};
2154 /* Describe the run that is in progress,
2155 which the next character can try to extend.
2156 If run_type is strange, that means there really is no run.
2157 If run_type is one_case, then run_start...run_end is the run.
2158 If run_type is two_case, then the run is run_start...run_end,
2159 and the case-equivalents end at run_eqv_end. */
2161 enum case_type run_type
= strange
;
2162 int run_start
, run_end
, run_eqv_end
;
2164 Lisp_Object eqv_table
;
2166 if (!RE_TRANSLATE_P (translate
))
2168 EXTEND_RANGE_TABLE (work_area
, 2);
2169 work_area
->table
[work_area
->used
++] = (start
);
2170 work_area
->table
[work_area
->used
++] = (end
);
2174 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2176 for (; start
<= end
; start
++)
2178 enum case_type this_type
;
2179 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2180 int minchar
, maxchar
;
2182 /* Classify this character */
2184 this_type
= one_case
;
2185 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2186 this_type
= two_case
;
2188 this_type
= strange
;
2191 minchar
= start
, maxchar
= eqv
;
2193 minchar
= eqv
, maxchar
= start
;
2195 /* Can this character extend the run in progress? */
2196 if (this_type
== strange
|| this_type
!= run_type
2197 || !(minchar
== run_end
+ 1
2198 && (run_type
== two_case
2199 ? maxchar
== run_eqv_end
+ 1 : 1)))
2202 Record each of its equivalent ranges. */
2203 if (run_type
== one_case
)
2205 EXTEND_RANGE_TABLE (work_area
, 2);
2206 work_area
->table
[work_area
->used
++] = run_start
;
2207 work_area
->table
[work_area
->used
++] = run_end
;
2209 else if (run_type
== two_case
)
2211 EXTEND_RANGE_TABLE (work_area
, 4);
2212 work_area
->table
[work_area
->used
++] = run_start
;
2213 work_area
->table
[work_area
->used
++] = run_end
;
2214 work_area
->table
[work_area
->used
++]
2215 = RE_TRANSLATE (eqv_table
, run_start
);
2216 work_area
->table
[work_area
->used
++]
2217 = RE_TRANSLATE (eqv_table
, run_end
);
2222 if (this_type
== strange
)
2224 /* For a strange character, add each of its equivalents, one
2225 by one. Don't start a range. */
2228 EXTEND_RANGE_TABLE (work_area
, 2);
2229 work_area
->table
[work_area
->used
++] = eqv
;
2230 work_area
->table
[work_area
->used
++] = eqv
;
2231 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2233 while (eqv
!= start
);
2236 /* Add this char to the run, or start a new run. */
2237 else if (run_type
== strange
)
2239 /* Initialize a new range. */
2240 run_type
= this_type
;
2243 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2247 /* Extend a running range. */
2249 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2253 /* If a run is still in progress at the end, finish it now
2254 by recording its equivalent ranges. */
2255 if (run_type
== one_case
)
2257 EXTEND_RANGE_TABLE (work_area
, 2);
2258 work_area
->table
[work_area
->used
++] = run_start
;
2259 work_area
->table
[work_area
->used
++] = run_end
;
2261 else if (run_type
== two_case
)
2263 EXTEND_RANGE_TABLE (work_area
, 4);
2264 work_area
->table
[work_area
->used
++] = run_start
;
2265 work_area
->table
[work_area
->used
++] = run_end
;
2266 work_area
->table
[work_area
->used
++]
2267 = RE_TRANSLATE (eqv_table
, run_start
);
2268 work_area
->table
[work_area
->used
++]
2269 = RE_TRANSLATE (eqv_table
, run_end
);
2277 /* Record the the image of the range start..end when passed through
2278 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2279 and is not even necessarily contiguous.
2280 Normally we approximate it with the smallest contiguous range that contains
2281 all the chars we need. However, for Latin-1 we go to extra effort
2284 This function is not called for ASCII ranges.
2286 Returns -1 if successful, REG_ESPACE if ran out of space. */
2289 set_image_of_range (work_area
, start
, end
, translate
)
2290 RE_TRANSLATE_TYPE translate
;
2291 struct range_table_work_area
*work_area
;
2292 re_wchar_t start
, end
;
2294 re_wchar_t cmin
, cmax
;
2297 /* For Latin-1 ranges, use set_image_of_range_1
2298 to get proper handling of ranges that include letters and nonletters.
2299 For a range that includes the whole of Latin-1, this is not necessary.
2300 For other character sets, we don't bother to get this right. */
2301 if (RE_TRANSLATE_P (translate
) && start
< 04400
2302 && !(start
< 04200 && end
>= 04377))
2309 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2319 EXTEND_RANGE_TABLE (work_area
, 2);
2320 work_area
->table
[work_area
->used
++] = (start
);
2321 work_area
->table
[work_area
->used
++] = (end
);
2323 cmin
= -1, cmax
= -1;
2325 if (RE_TRANSLATE_P (translate
))
2329 for (ch
= start
; ch
<= end
; ch
++)
2331 re_wchar_t c
= TRANSLATE (ch
);
2332 if (! (start
<= c
&& c
<= end
))
2338 cmin
= MIN (cmin
, c
);
2339 cmax
= MAX (cmax
, c
);
2346 EXTEND_RANGE_TABLE (work_area
, 2);
2347 work_area
->table
[work_area
->used
++] = (cmin
);
2348 work_area
->table
[work_area
->used
++] = (cmax
);
2355 #ifndef MATCH_MAY_ALLOCATE
2357 /* If we cannot allocate large objects within re_match_2_internal,
2358 we make the fail stack and register vectors global.
2359 The fail stack, we grow to the maximum size when a regexp
2361 The register vectors, we adjust in size each time we
2362 compile a regexp, according to the number of registers it needs. */
2364 static fail_stack_type fail_stack
;
2366 /* Size with which the following vectors are currently allocated.
2367 That is so we can make them bigger as needed,
2368 but never make them smaller. */
2369 static int regs_allocated_size
;
2371 static re_char
** regstart
, ** regend
;
2372 static re_char
**best_regstart
, **best_regend
;
2374 /* Make the register vectors big enough for NUM_REGS registers,
2375 but don't make them smaller. */
2378 regex_grow_registers (num_regs
)
2381 if (num_regs
> regs_allocated_size
)
2383 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2384 RETALLOC_IF (regend
, num_regs
, re_char
*);
2385 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2386 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2388 regs_allocated_size
= num_regs
;
2392 #endif /* not MATCH_MAY_ALLOCATE */
2394 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2398 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2399 Returns one of error codes defined in `regex.h', or zero for success.
2401 Assumes the `allocated' (and perhaps `buffer') and `translate'
2402 fields are set in BUFP on entry.
2404 If it succeeds, results are put in BUFP (if it returns an error, the
2405 contents of BUFP are undefined):
2406 `buffer' is the compiled pattern;
2407 `syntax' is set to SYNTAX;
2408 `used' is set to the length of the compiled pattern;
2409 `fastmap_accurate' is zero;
2410 `re_nsub' is the number of subexpressions in PATTERN;
2411 `not_bol' and `not_eol' are zero;
2413 The `fastmap' field is neither examined nor set. */
2415 /* Insert the `jump' from the end of last alternative to "here".
2416 The space for the jump has already been allocated. */
2417 #define FIXUP_ALT_JUMP() \
2419 if (fixup_alt_jump) \
2420 STORE_JUMP (jump, fixup_alt_jump, b); \
2424 /* Return, freeing storage we allocated. */
2425 #define FREE_STACK_RETURN(value) \
2427 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2428 free (compile_stack.stack); \
2432 static reg_errcode_t
2433 regex_compile (pattern
, size
, syntax
, bufp
)
2436 reg_syntax_t syntax
;
2437 struct re_pattern_buffer
*bufp
;
2439 /* We fetch characters from PATTERN here. */
2440 register re_wchar_t c
, c1
;
2442 /* A random temporary spot in PATTERN. */
2445 /* Points to the end of the buffer, where we should append. */
2446 register unsigned char *b
;
2448 /* Keeps track of unclosed groups. */
2449 compile_stack_type compile_stack
;
2451 /* Points to the current (ending) position in the pattern. */
2453 /* `const' makes AIX compiler fail. */
2454 unsigned char *p
= pattern
;
2456 re_char
*p
= pattern
;
2458 re_char
*pend
= pattern
+ size
;
2460 /* How to translate the characters in the pattern. */
2461 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2463 /* Address of the count-byte of the most recently inserted `exactn'
2464 command. This makes it possible to tell if a new exact-match
2465 character can be added to that command or if the character requires
2466 a new `exactn' command. */
2467 unsigned char *pending_exact
= 0;
2469 /* Address of start of the most recently finished expression.
2470 This tells, e.g., postfix * where to find the start of its
2471 operand. Reset at the beginning of groups and alternatives. */
2472 unsigned char *laststart
= 0;
2474 /* Address of beginning of regexp, or inside of last group. */
2475 unsigned char *begalt
;
2477 /* Place in the uncompiled pattern (i.e., the {) to
2478 which to go back if the interval is invalid. */
2479 re_char
*beg_interval
;
2481 /* Address of the place where a forward jump should go to the end of
2482 the containing expression. Each alternative of an `or' -- except the
2483 last -- ends with a forward jump of this sort. */
2484 unsigned char *fixup_alt_jump
= 0;
2486 /* Work area for range table of charset. */
2487 struct range_table_work_area range_table_work
;
2489 /* If the object matched can contain multibyte characters. */
2490 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2492 /* Nonzero if we have pushed down into a subpattern. */
2493 int in_subpattern
= 0;
2495 /* These hold the values of p, pattern, and pend from the main
2496 pattern when we have pushed into a subpattern. */
2498 re_char
*main_pattern
;
2503 DEBUG_PRINT1 ("\nCompiling pattern: ");
2506 unsigned debug_count
;
2508 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2509 putchar (pattern
[debug_count
]);
2514 /* Initialize the compile stack. */
2515 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2516 if (compile_stack
.stack
== NULL
)
2519 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2520 compile_stack
.avail
= 0;
2522 range_table_work
.table
= 0;
2523 range_table_work
.allocated
= 0;
2525 /* Initialize the pattern buffer. */
2526 bufp
->syntax
= syntax
;
2527 bufp
->fastmap_accurate
= 0;
2528 bufp
->not_bol
= bufp
->not_eol
= 0;
2529 bufp
->used_syntax
= 0;
2531 /* Set `used' to zero, so that if we return an error, the pattern
2532 printer (for debugging) will think there's no pattern. We reset it
2536 /* Always count groups, whether or not bufp->no_sub is set. */
2539 #if !defined emacs && !defined SYNTAX_TABLE
2540 /* Initialize the syntax table. */
2541 init_syntax_once ();
2544 if (bufp
->allocated
== 0)
2547 { /* If zero allocated, but buffer is non-null, try to realloc
2548 enough space. This loses if buffer's address is bogus, but
2549 that is the user's responsibility. */
2550 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2553 { /* Caller did not allocate a buffer. Do it for them. */
2554 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2556 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2558 bufp
->allocated
= INIT_BUF_SIZE
;
2561 begalt
= b
= bufp
->buffer
;
2563 /* Loop through the uncompiled pattern until we're at the end. */
2568 /* If this is the end of an included regexp,
2569 pop back to the main regexp and try again. */
2573 pattern
= main_pattern
;
2578 /* If this is the end of the main regexp, we are done. */
2590 /* If there's no special whitespace regexp, treat
2591 spaces normally. And don't try to do this recursively. */
2592 if (!whitespace_regexp
|| in_subpattern
)
2595 /* Peek past following spaces. */
2602 /* If the spaces are followed by a repetition op,
2603 treat them normally. */
2605 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2606 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2609 /* Replace the spaces with the whitespace regexp. */
2613 main_pattern
= pattern
;
2614 p
= pattern
= whitespace_regexp
;
2615 pend
= p
+ strlen (p
);
2621 if ( /* If at start of pattern, it's an operator. */
2623 /* If context independent, it's an operator. */
2624 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2625 /* Otherwise, depends on what's come before. */
2626 || at_begline_loc_p (pattern
, p
, syntax
))
2627 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2636 if ( /* If at end of pattern, it's an operator. */
2638 /* If context independent, it's an operator. */
2639 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2640 /* Otherwise, depends on what's next. */
2641 || at_endline_loc_p (p
, pend
, syntax
))
2642 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2651 if ((syntax
& RE_BK_PLUS_QM
)
2652 || (syntax
& RE_LIMITED_OPS
))
2656 /* If there is no previous pattern... */
2659 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2660 FREE_STACK_RETURN (REG_BADRPT
);
2661 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2666 /* 1 means zero (many) matches is allowed. */
2667 boolean zero_times_ok
= 0, many_times_ok
= 0;
2670 /* If there is a sequence of repetition chars, collapse it
2671 down to just one (the right one). We can't combine
2672 interval operators with these because of, e.g., `a{2}*',
2673 which should only match an even number of `a's. */
2677 if ((syntax
& RE_FRUGAL
)
2678 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2682 zero_times_ok
|= c
!= '+';
2683 many_times_ok
|= c
!= '?';
2689 || (!(syntax
& RE_BK_PLUS_QM
)
2690 && (*p
== '+' || *p
== '?')))
2692 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2695 FREE_STACK_RETURN (REG_EESCAPE
);
2696 if (p
[1] == '+' || p
[1] == '?')
2697 PATFETCH (c
); /* Gobble up the backslash. */
2703 /* If we get here, we found another repeat character. */
2707 /* Star, etc. applied to an empty pattern is equivalent
2708 to an empty pattern. */
2709 if (!laststart
|| laststart
== b
)
2712 /* Now we know whether or not zero matches is allowed
2713 and also whether or not two or more matches is allowed. */
2718 boolean simple
= skip_one_char (laststart
) == b
;
2719 unsigned int startoffset
= 0;
2721 /* Check if the loop can match the empty string. */
2722 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2723 ? on_failure_jump
: on_failure_jump_loop
;
2724 assert (skip_one_char (laststart
) <= b
);
2726 if (!zero_times_ok
&& simple
)
2727 { /* Since simple * loops can be made faster by using
2728 on_failure_keep_string_jump, we turn simple P+
2729 into PP* if P is simple. */
2730 unsigned char *p1
, *p2
;
2731 startoffset
= b
- laststart
;
2732 GET_BUFFER_SPACE (startoffset
);
2733 p1
= b
; p2
= laststart
;
2739 GET_BUFFER_SPACE (6);
2742 STORE_JUMP (ofj
, b
, b
+ 6);
2744 /* Simple * loops can use on_failure_keep_string_jump
2745 depending on what follows. But since we don't know
2746 that yet, we leave the decision up to
2747 on_failure_jump_smart. */
2748 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2749 laststart
+ startoffset
, b
+ 6);
2751 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2756 /* A simple ? pattern. */
2757 assert (zero_times_ok
);
2758 GET_BUFFER_SPACE (3);
2759 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2763 else /* not greedy */
2764 { /* I wish the greedy and non-greedy cases could be merged. */
2766 GET_BUFFER_SPACE (7); /* We might use less. */
2769 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2771 /* The non-greedy multiple match looks like
2772 a repeat..until: we only need a conditional jump
2773 at the end of the loop. */
2774 if (emptyp
) BUF_PUSH (no_op
);
2775 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2776 : on_failure_jump
, b
, laststart
);
2780 /* The repeat...until naturally matches one or more.
2781 To also match zero times, we need to first jump to
2782 the end of the loop (its conditional jump). */
2783 INSERT_JUMP (jump
, laststart
, b
);
2789 /* non-greedy a?? */
2790 INSERT_JUMP (jump
, laststart
, b
+ 3);
2792 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2809 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2811 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2813 /* Ensure that we have enough space to push a charset: the
2814 opcode, the length count, and the bitset; 34 bytes in all. */
2815 GET_BUFFER_SPACE (34);
2819 /* We test `*p == '^' twice, instead of using an if
2820 statement, so we only need one BUF_PUSH. */
2821 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2825 /* Remember the first position in the bracket expression. */
2828 /* Push the number of bytes in the bitmap. */
2829 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2831 /* Clear the whole map. */
2832 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2834 /* charset_not matches newline according to a syntax bit. */
2835 if ((re_opcode_t
) b
[-2] == charset_not
2836 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2837 SET_LIST_BIT ('\n');
2839 /* Read in characters and ranges, setting map bits. */
2842 boolean escaped_char
= false;
2843 const unsigned char *p2
= p
;
2845 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2847 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2848 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2849 So the translation is done later in a loop. Example:
2850 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2853 /* \ might escape characters inside [...] and [^...]. */
2854 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2856 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2859 escaped_char
= true;
2863 /* Could be the end of the bracket expression. If it's
2864 not (i.e., when the bracket expression is `[]' so
2865 far), the ']' character bit gets set way below. */
2866 if (c
== ']' && p2
!= p1
)
2870 /* What should we do for the character which is
2871 greater than 0x7F, but not BASE_LEADING_CODE_P?
2874 /* See if we're at the beginning of a possible character
2877 if (!escaped_char
&&
2878 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2880 /* Leave room for the null. */
2881 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2882 const unsigned char *class_beg
;
2888 /* If pattern is `[[:'. */
2889 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2894 if ((c
== ':' && *p
== ']') || p
== pend
)
2896 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2899 /* This is in any case an invalid class name. */
2904 /* If isn't a word bracketed by `[:' and `:]':
2905 undo the ending character, the letters, and
2906 leave the leading `:' and `[' (but set bits for
2908 if (c
== ':' && *p
== ']')
2913 cc
= re_wctype (str
);
2916 FREE_STACK_RETURN (REG_ECTYPE
);
2918 /* Throw away the ] at the end of the character
2922 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2924 /* Most character classes in a multibyte match
2925 just set a flag. Exceptions are is_blank,
2926 is_digit, is_cntrl, and is_xdigit, since
2927 they can only match ASCII characters. We
2928 don't need to handle them for multibyte.
2929 They are distinguished by a negative wctype. */
2932 SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work
,
2933 re_wctype_to_bit (cc
));
2935 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2937 int translated
= TRANSLATE (ch
);
2938 if (translated
< (1 << BYTEWIDTH
)
2939 && re_iswctype (btowc (ch
), cc
))
2940 SET_LIST_BIT (translated
);
2943 /* In most cases the matching rule for char classes
2944 only uses the syntax table for multibyte chars,
2945 so that the content of the syntax-table it is not
2946 hardcoded in the range_table. SPACE and WORD are
2947 the two exceptions. */
2948 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2949 bufp
->used_syntax
= 1;
2951 /* Repeat the loop. */
2956 /* Go back to right after the "[:". */
2960 /* Because the `:' may starts the range, we
2961 can't simply set bit and repeat the loop.
2962 Instead, just set it to C and handle below. */
2967 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2970 /* Discard the `-'. */
2973 /* Fetch the character which ends the range. */
2976 if (SINGLE_BYTE_CHAR_P (c
))
2978 if (! SINGLE_BYTE_CHAR_P (c1
))
2980 /* Handle a range starting with a
2981 character of less than 256, and ending
2982 with a character of not less than 256.
2983 Split that into two ranges, the low one
2984 ending at 0377, and the high one
2985 starting at the smallest character in
2986 the charset of C1 and ending at C1. */
2987 int charset
= CHAR_CHARSET (c1
);
2988 re_wchar_t c2
= MAKE_CHAR (charset
, 0, 0);
2990 SET_RANGE_TABLE_WORK_AREA (range_table_work
,
2995 else if (!SAME_CHARSET_P (c
, c1
))
2996 FREE_STACK_RETURN (REG_ERANGEX
);
2999 /* Range from C to C. */
3002 /* Set the range ... */
3003 if (SINGLE_BYTE_CHAR_P (c
))
3004 /* ... into bitmap. */
3006 re_wchar_t this_char
;
3007 re_wchar_t range_start
= c
, range_end
= c1
;
3009 /* If the start is after the end, the range is empty. */
3010 if (range_start
> range_end
)
3012 if (syntax
& RE_NO_EMPTY_RANGES
)
3013 FREE_STACK_RETURN (REG_ERANGE
);
3014 /* Else, repeat the loop. */
3018 for (this_char
= range_start
; this_char
<= range_end
;
3021 int translated
= TRANSLATE (this_char
);
3022 if (translated
< (1 << BYTEWIDTH
))
3023 SET_LIST_BIT (translated
);
3025 SET_RANGE_TABLE_WORK_AREA
3026 (range_table_work
, translated
, translated
);
3031 /* ... into range table. */
3032 SET_RANGE_TABLE_WORK_AREA (range_table_work
, c
, c1
);
3035 /* Discard any (non)matching list bytes that are all 0 at the
3036 end of the map. Decrease the map-length byte too. */
3037 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3041 /* Build real range table from work area. */
3042 if (RANGE_TABLE_WORK_USED (range_table_work
)
3043 || RANGE_TABLE_WORK_BITS (range_table_work
))
3046 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3048 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3049 bytes for flags, two for COUNT, and three bytes for
3051 GET_BUFFER_SPACE (4 + used
* 3);
3053 /* Indicate the existence of range table. */
3054 laststart
[1] |= 0x80;
3056 /* Store the character class flag bits into the range table.
3057 If not in emacs, these flag bits are always 0. */
3058 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3059 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3061 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3062 for (i
= 0; i
< used
; i
++)
3063 STORE_CHARACTER_AND_INCR
3064 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3071 if (syntax
& RE_NO_BK_PARENS
)
3078 if (syntax
& RE_NO_BK_PARENS
)
3085 if (syntax
& RE_NEWLINE_ALT
)
3092 if (syntax
& RE_NO_BK_VBAR
)
3099 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3100 goto handle_interval
;
3106 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3108 /* Do not translate the character after the \, so that we can
3109 distinguish, e.g., \B from \b, even if we normally would
3110 translate, e.g., B to b. */
3116 if (syntax
& RE_NO_BK_PARENS
)
3117 goto normal_backslash
;
3122 regnum_t regnum
= 0;
3125 /* Look for a special (?...) construct */
3126 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3128 PATFETCH (c
); /* Gobble up the '?'. */
3134 case ':': shy
= 1; break;
3136 /* An explicitly specified regnum must start
3139 FREE_STACK_RETURN (REG_BADPAT
);
3140 case '1': case '2': case '3': case '4':
3141 case '5': case '6': case '7': case '8': case '9':
3142 regnum
= 10*regnum
+ (c
- '0'); break;
3144 /* Only (?:...) is supported right now. */
3145 FREE_STACK_RETURN (REG_BADPAT
);
3152 regnum
= ++bufp
->re_nsub
;
3154 { /* It's actually not shy, but explicitly numbered. */
3156 if (regnum
> bufp
->re_nsub
)
3157 bufp
->re_nsub
= regnum
;
3158 else if (regnum
> bufp
->re_nsub
3159 /* Ideally, we'd want to check that the specified
3160 group can't have matched (i.e. all subgroups
3161 using the same regnum are in other branches of
3162 OR patterns), but we don't currently keep track
3163 of enough info to do that easily. */
3164 || group_in_compile_stack (compile_stack
, regnum
))
3165 FREE_STACK_RETURN (REG_BADPAT
);
3168 /* It's really shy. */
3169 regnum
= - bufp
->re_nsub
;
3171 if (COMPILE_STACK_FULL
)
3173 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3174 compile_stack_elt_t
);
3175 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3177 compile_stack
.size
<<= 1;
3180 /* These are the values to restore when we hit end of this
3181 group. They are all relative offsets, so that if the
3182 whole pattern moves because of realloc, they will still
3184 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3185 COMPILE_STACK_TOP
.fixup_alt_jump
3186 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3187 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3188 COMPILE_STACK_TOP
.regnum
= regnum
;
3190 /* Do not push a start_memory for groups beyond the last one
3191 we can represent in the compiled pattern. */
3192 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3193 BUF_PUSH_2 (start_memory
, regnum
);
3195 compile_stack
.avail
++;
3200 /* If we've reached MAX_REGNUM groups, then this open
3201 won't actually generate any code, so we'll have to
3202 clear pending_exact explicitly. */
3208 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3210 if (COMPILE_STACK_EMPTY
)
3212 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3213 goto normal_backslash
;
3215 FREE_STACK_RETURN (REG_ERPAREN
);
3221 /* See similar code for backslashed left paren above. */
3222 if (COMPILE_STACK_EMPTY
)
3224 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3227 FREE_STACK_RETURN (REG_ERPAREN
);
3230 /* Since we just checked for an empty stack above, this
3231 ``can't happen''. */
3232 assert (compile_stack
.avail
!= 0);
3234 /* We don't just want to restore into `regnum', because
3235 later groups should continue to be numbered higher,
3236 as in `(ab)c(de)' -- the second group is #2. */
3239 compile_stack
.avail
--;
3240 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3242 = COMPILE_STACK_TOP
.fixup_alt_jump
3243 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3245 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3246 regnum
= COMPILE_STACK_TOP
.regnum
;
3247 /* If we've reached MAX_REGNUM groups, then this open
3248 won't actually generate any code, so we'll have to
3249 clear pending_exact explicitly. */
3252 /* We're at the end of the group, so now we know how many
3253 groups were inside this one. */
3254 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3255 BUF_PUSH_2 (stop_memory
, regnum
);
3260 case '|': /* `\|'. */
3261 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3262 goto normal_backslash
;
3264 if (syntax
& RE_LIMITED_OPS
)
3267 /* Insert before the previous alternative a jump which
3268 jumps to this alternative if the former fails. */
3269 GET_BUFFER_SPACE (3);
3270 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3274 /* The alternative before this one has a jump after it
3275 which gets executed if it gets matched. Adjust that
3276 jump so it will jump to this alternative's analogous
3277 jump (put in below, which in turn will jump to the next
3278 (if any) alternative's such jump, etc.). The last such
3279 jump jumps to the correct final destination. A picture:
3285 If we are at `b', then fixup_alt_jump right now points to a
3286 three-byte space after `a'. We'll put in the jump, set
3287 fixup_alt_jump to right after `b', and leave behind three
3288 bytes which we'll fill in when we get to after `c'. */
3292 /* Mark and leave space for a jump after this alternative,
3293 to be filled in later either by next alternative or
3294 when know we're at the end of a series of alternatives. */
3296 GET_BUFFER_SPACE (3);
3305 /* If \{ is a literal. */
3306 if (!(syntax
& RE_INTERVALS
)
3307 /* If we're at `\{' and it's not the open-interval
3309 || (syntax
& RE_NO_BK_BRACES
))
3310 goto normal_backslash
;
3314 /* If got here, then the syntax allows intervals. */
3316 /* At least (most) this many matches must be made. */
3317 int lower_bound
= 0, upper_bound
= -1;
3321 GET_UNSIGNED_NUMBER (lower_bound
);
3324 GET_UNSIGNED_NUMBER (upper_bound
);
3326 /* Interval such as `{1}' => match exactly once. */
3327 upper_bound
= lower_bound
;
3329 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3330 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3331 FREE_STACK_RETURN (REG_BADBR
);
3333 if (!(syntax
& RE_NO_BK_BRACES
))
3336 FREE_STACK_RETURN (REG_BADBR
);
3338 FREE_STACK_RETURN (REG_EESCAPE
);
3343 FREE_STACK_RETURN (REG_BADBR
);
3345 /* We just parsed a valid interval. */
3347 /* If it's invalid to have no preceding re. */
3350 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3351 FREE_STACK_RETURN (REG_BADRPT
);
3352 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3355 goto unfetch_interval
;
3358 if (upper_bound
== 0)
3359 /* If the upper bound is zero, just drop the sub pattern
3362 else if (lower_bound
== 1 && upper_bound
== 1)
3363 /* Just match it once: nothing to do here. */
3366 /* Otherwise, we have a nontrivial interval. When
3367 we're all done, the pattern will look like:
3368 set_number_at <jump count> <upper bound>
3369 set_number_at <succeed_n count> <lower bound>
3370 succeed_n <after jump addr> <succeed_n count>
3372 jump_n <succeed_n addr> <jump count>
3373 (The upper bound and `jump_n' are omitted if
3374 `upper_bound' is 1, though.) */
3376 { /* If the upper bound is > 1, we need to insert
3377 more at the end of the loop. */
3378 unsigned int nbytes
= (upper_bound
< 0 ? 3
3379 : upper_bound
> 1 ? 5 : 0);
3380 unsigned int startoffset
= 0;
3382 GET_BUFFER_SPACE (20); /* We might use less. */
3384 if (lower_bound
== 0)
3386 /* A succeed_n that starts with 0 is really a
3387 a simple on_failure_jump_loop. */
3388 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3394 /* Initialize lower bound of the `succeed_n', even
3395 though it will be set during matching by its
3396 attendant `set_number_at' (inserted next),
3397 because `re_compile_fastmap' needs to know.
3398 Jump to the `jump_n' we might insert below. */
3399 INSERT_JUMP2 (succeed_n
, laststart
,
3404 /* Code to initialize the lower bound. Insert
3405 before the `succeed_n'. The `5' is the last two
3406 bytes of this `set_number_at', plus 3 bytes of
3407 the following `succeed_n'. */
3408 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3413 if (upper_bound
< 0)
3415 /* A negative upper bound stands for infinity,
3416 in which case it degenerates to a plain jump. */
3417 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3420 else if (upper_bound
> 1)
3421 { /* More than one repetition is allowed, so
3422 append a backward jump to the `succeed_n'
3423 that starts this interval.
3425 When we've reached this during matching,
3426 we'll have matched the interval once, so
3427 jump back only `upper_bound - 1' times. */
3428 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3432 /* The location we want to set is the second
3433 parameter of the `jump_n'; that is `b-2' as
3434 an absolute address. `laststart' will be
3435 the `set_number_at' we're about to insert;
3436 `laststart+3' the number to set, the source
3437 for the relative address. But we are
3438 inserting into the middle of the pattern --
3439 so everything is getting moved up by 5.
3440 Conclusion: (b - 2) - (laststart + 3) + 5,
3441 i.e., b - laststart.
3443 We insert this at the beginning of the loop
3444 so that if we fail during matching, we'll
3445 reinitialize the bounds. */
3446 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3447 upper_bound
- 1, b
);
3452 beg_interval
= NULL
;
3457 /* If an invalid interval, match the characters as literals. */
3458 assert (beg_interval
);
3460 beg_interval
= NULL
;
3462 /* normal_char and normal_backslash need `c'. */
3465 if (!(syntax
& RE_NO_BK_BRACES
))
3467 assert (p
> pattern
&& p
[-1] == '\\');
3468 goto normal_backslash
;
3474 /* There is no way to specify the before_dot and after_dot
3475 operators. rms says this is ok. --karl */
3483 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3489 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3495 BUF_PUSH_2 (categoryspec
, c
);
3501 BUF_PUSH_2 (notcategoryspec
, c
);
3507 if (syntax
& RE_NO_GNU_OPS
)
3510 BUF_PUSH_2 (syntaxspec
, Sword
);
3515 if (syntax
& RE_NO_GNU_OPS
)
3518 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3523 if (syntax
& RE_NO_GNU_OPS
)
3529 if (syntax
& RE_NO_GNU_OPS
)
3535 if (syntax
& RE_NO_GNU_OPS
)
3544 FREE_STACK_RETURN (REG_BADPAT
);
3548 if (syntax
& RE_NO_GNU_OPS
)
3550 BUF_PUSH (wordbound
);
3554 if (syntax
& RE_NO_GNU_OPS
)
3556 BUF_PUSH (notwordbound
);
3560 if (syntax
& RE_NO_GNU_OPS
)
3566 if (syntax
& RE_NO_GNU_OPS
)
3571 case '1': case '2': case '3': case '4': case '5':
3572 case '6': case '7': case '8': case '9':
3576 if (syntax
& RE_NO_BK_REFS
)
3577 goto normal_backslash
;
3581 if (reg
> bufp
->re_nsub
|| reg
< 1
3582 /* Can't back reference to a subexp before its end. */
3583 || group_in_compile_stack (compile_stack
, reg
))
3584 FREE_STACK_RETURN (REG_ESUBREG
);
3587 BUF_PUSH_2 (duplicate
, reg
);
3594 if (syntax
& RE_BK_PLUS_QM
)
3597 goto normal_backslash
;
3601 /* You might think it would be useful for \ to mean
3602 not to translate; but if we don't translate it
3603 it will never match anything. */
3610 /* Expects the character in `c'. */
3612 /* If no exactn currently being built. */
3615 /* If last exactn not at current position. */
3616 || pending_exact
+ *pending_exact
+ 1 != b
3618 /* We have only one byte following the exactn for the count. */
3619 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3621 /* If followed by a repetition operator. */
3622 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3623 || ((syntax
& RE_BK_PLUS_QM
)
3624 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3625 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3626 || ((syntax
& RE_INTERVALS
)
3627 && ((syntax
& RE_NO_BK_BRACES
)
3628 ? p
!= pend
&& *p
== '{'
3629 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3631 /* Start building a new exactn. */
3635 BUF_PUSH_2 (exactn
, 0);
3636 pending_exact
= b
- 1;
3639 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3645 len
= CHAR_STRING (c
, b
);
3649 (*pending_exact
) += len
;
3654 } /* while p != pend */
3657 /* Through the pattern now. */
3661 if (!COMPILE_STACK_EMPTY
)
3662 FREE_STACK_RETURN (REG_EPAREN
);
3664 /* If we don't want backtracking, force success
3665 the first time we reach the end of the compiled pattern. */
3666 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3669 /* We have succeeded; set the length of the buffer. */
3670 bufp
->used
= b
- bufp
->buffer
;
3675 re_compile_fastmap (bufp
);
3676 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3677 print_compiled_pattern (bufp
);
3682 #ifndef MATCH_MAY_ALLOCATE
3683 /* Initialize the failure stack to the largest possible stack. This
3684 isn't necessary unless we're trying to avoid calling alloca in
3685 the search and match routines. */
3687 int num_regs
= bufp
->re_nsub
+ 1;
3689 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3691 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3693 if (! fail_stack
.stack
)
3695 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3696 * sizeof (fail_stack_elt_t
));
3699 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3701 * sizeof (fail_stack_elt_t
)));
3704 regex_grow_registers (num_regs
);
3706 #endif /* not MATCH_MAY_ALLOCATE */
3708 FREE_STACK_RETURN (REG_NOERROR
);
3709 } /* regex_compile */
3711 /* Subroutines for `regex_compile'. */
3713 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3716 store_op1 (op
, loc
, arg
)
3721 *loc
= (unsigned char) op
;
3722 STORE_NUMBER (loc
+ 1, arg
);
3726 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3729 store_op2 (op
, loc
, arg1
, arg2
)
3734 *loc
= (unsigned char) op
;
3735 STORE_NUMBER (loc
+ 1, arg1
);
3736 STORE_NUMBER (loc
+ 3, arg2
);
3740 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3741 for OP followed by two-byte integer parameter ARG. */
3744 insert_op1 (op
, loc
, arg
, end
)
3750 register unsigned char *pfrom
= end
;
3751 register unsigned char *pto
= end
+ 3;
3753 while (pfrom
!= loc
)
3756 store_op1 (op
, loc
, arg
);
3760 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3763 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3769 register unsigned char *pfrom
= end
;
3770 register unsigned char *pto
= end
+ 5;
3772 while (pfrom
!= loc
)
3775 store_op2 (op
, loc
, arg1
, arg2
);
3779 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3780 after an alternative or a begin-subexpression. We assume there is at
3781 least one character before the ^. */
3784 at_begline_loc_p (pattern
, p
, syntax
)
3785 re_char
*pattern
, *p
;
3786 reg_syntax_t syntax
;
3788 re_char
*prev
= p
- 2;
3789 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3792 /* After a subexpression? */
3793 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3794 /* After an alternative? */
3795 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3796 /* After a shy subexpression? */
3797 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3798 && prev
[-1] == '?' && prev
[-2] == '('
3799 && (syntax
& RE_NO_BK_PARENS
3800 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3804 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3805 at least one character after the $, i.e., `P < PEND'. */
3808 at_endline_loc_p (p
, pend
, syntax
)
3810 reg_syntax_t syntax
;
3813 boolean next_backslash
= *next
== '\\';
3814 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3817 /* Before a subexpression? */
3818 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3819 : next_backslash
&& next_next
&& *next_next
== ')')
3820 /* Before an alternative? */
3821 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3822 : next_backslash
&& next_next
&& *next_next
== '|');
3826 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3827 false if it's not. */
3830 group_in_compile_stack (compile_stack
, regnum
)
3831 compile_stack_type compile_stack
;
3836 for (this_element
= compile_stack
.avail
- 1;
3839 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3846 If fastmap is non-NULL, go through the pattern and fill fastmap
3847 with all the possible leading chars. If fastmap is NULL, don't
3848 bother filling it up (obviously) and only return whether the
3849 pattern could potentially match the empty string.
3851 Return 1 if p..pend might match the empty string.
3852 Return 0 if p..pend matches at least one char.
3853 Return -1 if fastmap was not updated accurately. */
3856 analyse_first (p
, pend
, fastmap
, multibyte
)
3859 const int multibyte
;
3864 /* If all elements for base leading-codes in fastmap is set, this
3865 flag is set true. */
3866 boolean match_any_multibyte_characters
= false;
3870 /* The loop below works as follows:
3871 - It has a working-list kept in the PATTERN_STACK and which basically
3872 starts by only containing a pointer to the first operation.
3873 - If the opcode we're looking at is a match against some set of
3874 chars, then we add those chars to the fastmap and go on to the
3875 next work element from the worklist (done via `break').
3876 - If the opcode is a control operator on the other hand, we either
3877 ignore it (if it's meaningless at this point, such as `start_memory')
3878 or execute it (if it's a jump). If the jump has several destinations
3879 (i.e. `on_failure_jump'), then we push the other destination onto the
3881 We guarantee termination by ignoring backward jumps (more or less),
3882 so that `p' is monotonically increasing. More to the point, we
3883 never set `p' (or push) anything `<= p1'. */
3887 /* `p1' is used as a marker of how far back a `on_failure_jump'
3888 can go without being ignored. It is normally equal to `p'
3889 (which prevents any backward `on_failure_jump') except right
3890 after a plain `jump', to allow patterns such as:
3893 10: on_failure_jump 3
3894 as used for the *? operator. */
3897 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3904 /* If the first character has to match a backreference, that means
3905 that the group was empty (since it already matched). Since this
3906 is the only case that interests us here, we can assume that the
3907 backreference must match the empty string. */
3912 /* Following are the cases which match a character. These end
3918 int c
= RE_STRING_CHAR (p
+ 1, pend
- p
);
3919 /* When fast-scanning, the fastmap can be indexed either with
3920 a char (smaller than 256) or with the first byte of
3921 a char's byte sequence. So we have to conservatively add
3922 both to the table. */
3923 if (SINGLE_BYTE_CHAR_P (c
))
3931 /* We could put all the chars except for \n (and maybe \0)
3932 but we don't bother since it is generally not worth it. */
3933 if (!fastmap
) break;
3938 /* Chars beyond end of bitmap are possible matches.
3939 All the single-byte codes can occur in multibyte buffers.
3940 So any that are not listed in the charset
3941 are possible matches, even in multibyte buffers. */
3942 if (!fastmap
) break;
3943 /* We don't need to mark LEADING_CODE_8_BIT_CONTROL specially
3944 because it will automatically be set when needed by virtue of
3945 being larger than the highest char of its charset (0xbf) but
3946 smaller than (1<<BYTEWIDTH). */
3947 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3948 j
< (1 << BYTEWIDTH
); j
++)
3952 if (!fastmap
) break;
3953 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3954 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3956 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3960 if (j
>= 0x80 && j
< 0xa0)
3961 fastmap
[LEADING_CODE_8_BIT_CONTROL
] = 1;
3965 if ((not && multibyte
)
3966 /* Any character set can possibly contain a character
3967 which doesn't match the specified set of characters. */
3968 || (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3969 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3970 /* If we can match a character class, we can match
3971 any character set. */
3973 set_fastmap_for_multibyte_characters
:
3974 if (match_any_multibyte_characters
== false)
3976 for (j
= 0x80; j
< 0xA0; j
++) /* XXX */
3977 if (BASE_LEADING_CODE_P (j
))
3979 match_any_multibyte_characters
= true;
3983 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3984 && match_any_multibyte_characters
== false)
3986 /* Set fastmap[I] 1 where I is a base leading code of each
3987 multibyte character in the range table. */
3990 /* Make P points the range table. `+ 2' is to skip flag
3991 bits for a character class. */
3992 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3994 /* Extract the number of ranges in range table into COUNT. */
3995 EXTRACT_NUMBER_AND_INCR (count
, p
);
3996 for (; count
> 0; count
--, p
+= 2 * 3) /* XXX */
3998 /* Extract the start of each range. */
3999 EXTRACT_CHARACTER (c
, p
);
4000 j
= CHAR_CHARSET (c
);
4001 fastmap
[CHARSET_LEADING_CODE_BASE (j
)] = 1;
4008 if (!fastmap
) break;
4010 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4012 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4013 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4017 /* This match depends on text properties. These end with
4018 aborting optimizations. */
4022 case notcategoryspec
:
4023 if (!fastmap
) break;
4024 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4026 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4027 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4031 /* Any character set can possibly contain a character
4032 whose category is K (or not). */
4033 goto set_fastmap_for_multibyte_characters
;
4036 /* All cases after this match the empty string. These end with
4058 EXTRACT_NUMBER_AND_INCR (j
, p
);
4060 /* Backward jumps can only go back to code that we've already
4061 visited. `re_compile' should make sure this is true. */
4064 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4066 case on_failure_jump
:
4067 case on_failure_keep_string_jump
:
4068 case on_failure_jump_loop
:
4069 case on_failure_jump_nastyloop
:
4070 case on_failure_jump_smart
:
4076 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4077 to jump back to "just after here". */
4080 case on_failure_jump
:
4081 case on_failure_keep_string_jump
:
4082 case on_failure_jump_nastyloop
:
4083 case on_failure_jump_loop
:
4084 case on_failure_jump_smart
:
4085 EXTRACT_NUMBER_AND_INCR (j
, p
);
4087 ; /* Backward jump to be ignored. */
4089 { /* We have to look down both arms.
4090 We first go down the "straight" path so as to minimize
4091 stack usage when going through alternatives. */
4092 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4100 /* This code simply does not properly handle forward jump_n. */
4101 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4103 /* jump_n can either jump or fall through. The (backward) jump
4104 case has already been handled, so we only need to look at the
4105 fallthrough case. */
4109 /* If N == 0, it should be an on_failure_jump_loop instead. */
4110 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4112 /* We only care about one iteration of the loop, so we don't
4113 need to consider the case where this behaves like an
4130 abort (); /* We have listed all the cases. */
4133 /* Getting here means we have found the possible starting
4134 characters for one path of the pattern -- and that the empty
4135 string does not match. We need not follow this path further. */
4139 /* We reached the end without matching anything. */
4142 } /* analyse_first */
4144 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4145 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4146 characters can start a string that matches the pattern. This fastmap
4147 is used by re_search to skip quickly over impossible starting points.
4149 Character codes above (1 << BYTEWIDTH) are not represented in the
4150 fastmap, but the leading codes are represented. Thus, the fastmap
4151 indicates which character sets could start a match.
4153 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4154 area as BUFP->fastmap.
4156 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4159 Returns 0 if we succeed, -2 if an internal error. */
4162 re_compile_fastmap (bufp
)
4163 struct re_pattern_buffer
*bufp
;
4165 char *fastmap
= bufp
->fastmap
;
4168 assert (fastmap
&& bufp
->buffer
);
4170 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4171 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4173 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4174 fastmap
, RE_MULTIBYTE_P (bufp
));
4175 bufp
->can_be_null
= (analysis
!= 0);
4177 } /* re_compile_fastmap */
4179 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4180 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4181 this memory for recording register information. STARTS and ENDS
4182 must be allocated using the malloc library routine, and must each
4183 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4185 If NUM_REGS == 0, then subsequent matches should allocate their own
4188 Unless this function is called, the first search or match using
4189 PATTERN_BUFFER will allocate its own register data, without
4190 freeing the old data. */
4193 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4194 struct re_pattern_buffer
*bufp
;
4195 struct re_registers
*regs
;
4197 regoff_t
*starts
, *ends
;
4201 bufp
->regs_allocated
= REGS_REALLOCATE
;
4202 regs
->num_regs
= num_regs
;
4203 regs
->start
= starts
;
4208 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4210 regs
->start
= regs
->end
= (regoff_t
*) 0;
4213 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4215 /* Searching routines. */
4217 /* Like re_search_2, below, but only one string is specified, and
4218 doesn't let you say where to stop matching. */
4221 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4222 struct re_pattern_buffer
*bufp
;
4224 int size
, startpos
, range
;
4225 struct re_registers
*regs
;
4227 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4230 WEAK_ALIAS (__re_search
, re_search
)
4232 /* Head address of virtual concatenation of string. */
4233 #define HEAD_ADDR_VSTRING(P) \
4234 (((P) >= size1 ? string2 : string1))
4236 /* End address of virtual concatenation of string. */
4237 #define STOP_ADDR_VSTRING(P) \
4238 (((P) >= size1 ? string2 + size2 : string1 + size1))
4240 /* Address of POS in the concatenation of virtual string. */
4241 #define POS_ADDR_VSTRING(POS) \
4242 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4244 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4245 virtual concatenation of STRING1 and STRING2, starting first at index
4246 STARTPOS, then at STARTPOS + 1, and so on.
4248 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4250 RANGE is how far to scan while trying to match. RANGE = 0 means try
4251 only at STARTPOS; in general, the last start tried is STARTPOS +
4254 In REGS, return the indices of the virtual concatenation of STRING1
4255 and STRING2 that matched the entire BUFP->buffer and its contained
4258 Do not consider matching one past the index STOP in the virtual
4259 concatenation of STRING1 and STRING2.
4261 We return either the position in the strings at which the match was
4262 found, -1 if no match, or -2 if error (such as failure
4266 re_search_2 (bufp
, str1
, size1
, str2
, size2
, startpos
, range
, regs
, stop
)
4267 struct re_pattern_buffer
*bufp
;
4268 const char *str1
, *str2
;
4272 struct re_registers
*regs
;
4276 re_char
*string1
= (re_char
*) str1
;
4277 re_char
*string2
= (re_char
*) str2
;
4278 register char *fastmap
= bufp
->fastmap
;
4279 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4280 int total_size
= size1
+ size2
;
4281 int endpos
= startpos
+ range
;
4282 boolean anchored_start
;
4284 /* Nonzero if we have to concern multibyte character. */
4285 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4287 /* Check for out-of-range STARTPOS. */
4288 if (startpos
< 0 || startpos
> total_size
)
4291 /* Fix up RANGE if it might eventually take us outside
4292 the virtual concatenation of STRING1 and STRING2.
4293 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4295 range
= 0 - startpos
;
4296 else if (endpos
> total_size
)
4297 range
= total_size
- startpos
;
4299 /* If the search isn't to be a backwards one, don't waste time in a
4300 search for a pattern anchored at beginning of buffer. */
4301 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4310 /* In a forward search for something that starts with \=.
4311 don't keep searching past point. */
4312 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4314 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4320 /* Update the fastmap now if not correct already. */
4321 if (fastmap
&& !bufp
->fastmap_accurate
)
4322 re_compile_fastmap (bufp
);
4324 /* See whether the pattern is anchored. */
4325 anchored_start
= (bufp
->buffer
[0] == begline
);
4328 gl_state
.object
= re_match_object
;
4330 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4332 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4336 /* Loop through the string, looking for a place to start matching. */
4339 /* If the pattern is anchored,
4340 skip quickly past places we cannot match.
4341 We don't bother to treat startpos == 0 specially
4342 because that case doesn't repeat. */
4343 if (anchored_start
&& startpos
> 0)
4345 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4346 : string2
[startpos
- size1
- 1])
4351 /* If a fastmap is supplied, skip quickly over characters that
4352 cannot be the start of a match. If the pattern can match the
4353 null string, however, we don't need to skip characters; we want
4354 the first null string. */
4355 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4357 register re_char
*d
;
4358 register re_wchar_t buf_ch
;
4360 d
= POS_ADDR_VSTRING (startpos
);
4362 if (range
> 0) /* Searching forwards. */
4364 register int lim
= 0;
4367 if (startpos
< size1
&& startpos
+ range
>= size1
)
4368 lim
= range
- (size1
- startpos
);
4370 /* Written out as an if-else to avoid testing `translate'
4372 if (RE_TRANSLATE_P (translate
))
4379 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4382 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4387 range
-= buf_charlen
;
4392 /* Convert *d to integer to shut up GCC's
4393 whining about comparison that is always
4398 && !fastmap
[RE_TRANSLATE (translate
, di
)])
4408 re_char
*d_start
= d
;
4409 while (range
> lim
&& !fastmap
[*d
])
4415 if (multibyte
&& range
> lim
)
4417 /* Check that we are at the beginning of a char. */
4419 AT_CHAR_BOUNDARY_P (at_boundary
, d
, d_start
);
4423 { /* We have matched an internal byte of a char
4424 rather than the leading byte, so it's a false
4425 positive: we should keep scanning. */
4434 startpos
+= irange
- range
;
4436 else /* Searching backwards. */
4438 int room
= (startpos
>= size1
4439 ? size2
+ size1
- startpos
4440 : size1
- startpos
);
4441 buf_ch
= RE_STRING_CHAR (d
, room
);
4442 buf_ch
= TRANSLATE (buf_ch
);
4444 if (! (buf_ch
>= 0400
4445 || fastmap
[buf_ch
]))
4450 /* If can't match the null string, and that's all we have left, fail. */
4451 if (range
>= 0 && startpos
== total_size
&& fastmap
4452 && !bufp
->can_be_null
)
4455 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4456 startpos
, regs
, stop
);
4457 #ifndef REGEX_MALLOC
4474 /* Update STARTPOS to the next character boundary. */
4477 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4478 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4479 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
4497 /* Update STARTPOS to the previous character boundary. */
4500 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4502 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4504 /* Find the head of multibyte form. */
4505 PREV_CHAR_BOUNDARY (p
, phead
);
4506 range
+= p0
- 1 - p
;
4510 startpos
-= p0
- 1 - p
;
4516 WEAK_ALIAS (__re_search_2
, re_search_2
)
4518 /* Declarations and macros for re_match_2. */
4520 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4522 RE_TRANSLATE_TYPE translate
,
4523 const int multibyte
));
4525 /* This converts PTR, a pointer into one of the search strings `string1'
4526 and `string2' into an offset from the beginning of that string. */
4527 #define POINTER_TO_OFFSET(ptr) \
4528 (FIRST_STRING_P (ptr) \
4529 ? ((regoff_t) ((ptr) - string1)) \
4530 : ((regoff_t) ((ptr) - string2 + size1)))
4532 /* Call before fetching a character with *d. This switches over to
4533 string2 if necessary.
4534 Check re_match_2_internal for a discussion of why end_match_2 might
4535 not be within string2 (but be equal to end_match_1 instead). */
4536 #define PREFETCH() \
4539 /* End of string2 => fail. */ \
4540 if (dend == end_match_2) \
4542 /* End of string1 => advance to string2. */ \
4544 dend = end_match_2; \
4547 /* Call before fetching a char with *d if you already checked other limits.
4548 This is meant for use in lookahead operations like wordend, etc..
4549 where we might need to look at parts of the string that might be
4550 outside of the LIMITs (i.e past `stop'). */
4551 #define PREFETCH_NOLIMIT() \
4555 dend = end_match_2; \
4558 /* Test if at very beginning or at very end of the virtual concatenation
4559 of `string1' and `string2'. If only one string, it's `string2'. */
4560 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4561 #define AT_STRINGS_END(d) ((d) == end2)
4564 /* Test if D points to a character which is word-constituent. We have
4565 two special cases to check for: if past the end of string1, look at
4566 the first character in string2; and if before the beginning of
4567 string2, look at the last character in string1. */
4568 #define WORDCHAR_P(d) \
4569 (SYNTAX ((d) == end1 ? *string2 \
4570 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4573 /* Disabled due to a compiler bug -- see comment at case wordbound */
4575 /* The comment at case wordbound is following one, but we don't use
4576 AT_WORD_BOUNDARY anymore to support multibyte form.
4578 The DEC Alpha C compiler 3.x generates incorrect code for the
4579 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4580 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4581 macro and introducing temporary variables works around the bug. */
4584 /* Test if the character before D and the one at D differ with respect
4585 to being word-constituent. */
4586 #define AT_WORD_BOUNDARY(d) \
4587 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4588 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4591 /* Free everything we malloc. */
4592 #ifdef MATCH_MAY_ALLOCATE
4593 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4594 # define FREE_VARIABLES() \
4596 REGEX_FREE_STACK (fail_stack.stack); \
4597 FREE_VAR (regstart); \
4598 FREE_VAR (regend); \
4599 FREE_VAR (best_regstart); \
4600 FREE_VAR (best_regend); \
4603 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4604 #endif /* not MATCH_MAY_ALLOCATE */
4607 /* Optimization routines. */
4609 /* If the operation is a match against one or more chars,
4610 return a pointer to the next operation, else return NULL. */
4615 switch (SWITCH_ENUM_CAST (*p
++))
4626 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4629 p
= CHARSET_RANGE_TABLE (p
- 1);
4630 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4631 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4634 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4641 case notcategoryspec
:
4653 /* Jump over non-matching operations. */
4655 skip_noops (p
, pend
)
4661 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4670 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4681 /* Non-zero if "p1 matches something" implies "p2 fails". */
4683 mutually_exclusive_p (bufp
, p1
, p2
)
4684 struct re_pattern_buffer
*bufp
;
4688 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4689 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4691 assert (p1
>= bufp
->buffer
&& p1
< pend
4692 && p2
>= bufp
->buffer
&& p2
<= pend
);
4694 /* Skip over open/close-group commands.
4695 If what follows this loop is a ...+ construct,
4696 look at what begins its body, since we will have to
4697 match at least one of that. */
4698 p2
= skip_noops (p2
, pend
);
4699 /* The same skip can be done for p1, except that this function
4700 is only used in the case where p1 is a simple match operator. */
4701 /* p1 = skip_noops (p1, pend); */
4703 assert (p1
>= bufp
->buffer
&& p1
< pend
4704 && p2
>= bufp
->buffer
&& p2
<= pend
);
4706 op2
= p2
== pend
? succeed
: *p2
;
4708 switch (SWITCH_ENUM_CAST (op2
))
4712 /* If we're at the end of the pattern, we can change. */
4713 if (skip_one_char (p1
))
4715 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4723 register re_wchar_t c
4724 = (re_opcode_t
) *p2
== endline
? '\n'
4725 : RE_STRING_CHAR (p2
+ 2, pend
- p2
- 2);
4727 if ((re_opcode_t
) *p1
== exactn
)
4729 if (c
!= RE_STRING_CHAR (p1
+ 2, pend
- p1
- 2))
4731 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4736 else if ((re_opcode_t
) *p1
== charset
4737 || (re_opcode_t
) *p1
== charset_not
)
4739 int not = (re_opcode_t
) *p1
== charset_not
;
4741 /* Test if C is listed in charset (or charset_not)
4743 if (SINGLE_BYTE_CHAR_P (c
))
4745 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4746 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4749 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4750 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4752 /* `not' is equal to 1 if c would match, which means
4753 that we can't change to pop_failure_jump. */
4756 DEBUG_PRINT1 (" No match => fast loop.\n");
4760 else if ((re_opcode_t
) *p1
== anychar
4763 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4771 if ((re_opcode_t
) *p1
== exactn
)
4772 /* Reuse the code above. */
4773 return mutually_exclusive_p (bufp
, p2
, p1
);
4775 /* It is hard to list up all the character in charset
4776 P2 if it includes multibyte character. Give up in
4778 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4780 /* Now, we are sure that P2 has no range table.
4781 So, for the size of bitmap in P2, `p2[1]' is
4782 enough. But P1 may have range table, so the
4783 size of bitmap table of P1 is extracted by
4784 using macro `CHARSET_BITMAP_SIZE'.
4786 Since we know that all the character listed in
4787 P2 is ASCII, it is enough to test only bitmap
4790 if ((re_opcode_t
) *p1
== charset
)
4793 /* We win if the charset inside the loop
4794 has no overlap with the one after the loop. */
4797 && idx
< CHARSET_BITMAP_SIZE (p1
));
4799 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4803 || idx
== CHARSET_BITMAP_SIZE (p1
))
4805 DEBUG_PRINT1 (" No match => fast loop.\n");
4809 else if ((re_opcode_t
) *p1
== charset_not
)
4812 /* We win if the charset_not inside the loop lists
4813 every character listed in the charset after. */
4814 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4815 if (! (p2
[2 + idx
] == 0
4816 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4817 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4822 DEBUG_PRINT1 (" No match => fast loop.\n");
4831 switch (SWITCH_ENUM_CAST (*p1
))
4835 /* Reuse the code above. */
4836 return mutually_exclusive_p (bufp
, p2
, p1
);
4838 /* When we have two charset_not, it's very unlikely that
4839 they don't overlap. The union of the two sets of excluded
4840 chars should cover all possible chars, which, as a matter of
4841 fact, is virtually impossible in multibyte buffers. */
4847 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4849 return ((re_opcode_t
) *p1
== syntaxspec
4850 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4852 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4855 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4857 return ((re_opcode_t
) *p1
== notsyntaxspec
4858 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4860 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4863 return (((re_opcode_t
) *p1
== notsyntaxspec
4864 || (re_opcode_t
) *p1
== syntaxspec
)
4869 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4870 case notcategoryspec
:
4871 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4883 /* Matching routines. */
4885 #ifndef emacs /* Emacs never uses this. */
4886 /* re_match is like re_match_2 except it takes only a single string. */
4889 re_match (bufp
, string
, size
, pos
, regs
)
4890 struct re_pattern_buffer
*bufp
;
4893 struct re_registers
*regs
;
4895 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
4897 # if defined C_ALLOCA && !defined REGEX_MALLOC
4902 WEAK_ALIAS (__re_match
, re_match
)
4903 #endif /* not emacs */
4906 /* In Emacs, this is the string or buffer in which we
4907 are matching. It is used for looking up syntax properties. */
4908 Lisp_Object re_match_object
;
4911 /* re_match_2 matches the compiled pattern in BUFP against the
4912 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4913 and SIZE2, respectively). We start matching at POS, and stop
4916 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4917 store offsets for the substring each group matched in REGS. See the
4918 documentation for exactly how many groups we fill.
4920 We return -1 if no match, -2 if an internal error (such as the
4921 failure stack overflowing). Otherwise, we return the length of the
4922 matched substring. */
4925 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4926 struct re_pattern_buffer
*bufp
;
4927 const char *string1
, *string2
;
4930 struct re_registers
*regs
;
4937 gl_state
.object
= re_match_object
;
4938 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4939 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4942 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4943 (re_char
*) string2
, size2
,
4945 #if defined C_ALLOCA && !defined REGEX_MALLOC
4950 WEAK_ALIAS (__re_match_2
, re_match_2
)
4952 /* This is a separate function so that we can force an alloca cleanup
4955 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4956 struct re_pattern_buffer
*bufp
;
4957 re_char
*string1
, *string2
;
4960 struct re_registers
*regs
;
4963 /* General temporaries. */
4968 /* Just past the end of the corresponding string. */
4969 re_char
*end1
, *end2
;
4971 /* Pointers into string1 and string2, just past the last characters in
4972 each to consider matching. */
4973 re_char
*end_match_1
, *end_match_2
;
4975 /* Where we are in the data, and the end of the current string. */
4978 /* Used sometimes to remember where we were before starting matching
4979 an operator so that we can go back in case of failure. This "atomic"
4980 behavior of matching opcodes is indispensable to the correctness
4981 of the on_failure_keep_string_jump optimization. */
4984 /* Where we are in the pattern, and the end of the pattern. */
4985 re_char
*p
= bufp
->buffer
;
4986 re_char
*pend
= p
+ bufp
->used
;
4988 /* We use this to map every character in the string. */
4989 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4991 /* Nonzero if we have to concern multibyte character. */
4992 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4994 /* Failure point stack. Each place that can handle a failure further
4995 down the line pushes a failure point on this stack. It consists of
4996 regstart, and regend for all registers corresponding to
4997 the subexpressions we're currently inside, plus the number of such
4998 registers, and, finally, two char *'s. The first char * is where
4999 to resume scanning the pattern; the second one is where to resume
5000 scanning the strings. */
5001 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5002 fail_stack_type fail_stack
;
5005 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5008 #if defined REL_ALLOC && defined REGEX_MALLOC
5009 /* This holds the pointer to the failure stack, when
5010 it is allocated relocatably. */
5011 fail_stack_elt_t
*failure_stack_ptr
;
5014 /* We fill all the registers internally, independent of what we
5015 return, for use in backreferences. The number here includes
5016 an element for register zero. */
5017 size_t num_regs
= bufp
->re_nsub
+ 1;
5019 /* Information on the contents of registers. These are pointers into
5020 the input strings; they record just what was matched (on this
5021 attempt) by a subexpression part of the pattern, that is, the
5022 regnum-th regstart pointer points to where in the pattern we began
5023 matching and the regnum-th regend points to right after where we
5024 stopped matching the regnum-th subexpression. (The zeroth register
5025 keeps track of what the whole pattern matches.) */
5026 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5027 re_char
**regstart
, **regend
;
5030 /* The following record the register info as found in the above
5031 variables when we find a match better than any we've seen before.
5032 This happens as we backtrack through the failure points, which in
5033 turn happens only if we have not yet matched the entire string. */
5034 unsigned best_regs_set
= false;
5035 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5036 re_char
**best_regstart
, **best_regend
;
5039 /* Logically, this is `best_regend[0]'. But we don't want to have to
5040 allocate space for that if we're not allocating space for anything
5041 else (see below). Also, we never need info about register 0 for
5042 any of the other register vectors, and it seems rather a kludge to
5043 treat `best_regend' differently than the rest. So we keep track of
5044 the end of the best match so far in a separate variable. We
5045 initialize this to NULL so that when we backtrack the first time
5046 and need to test it, it's not garbage. */
5047 re_char
*match_end
= NULL
;
5050 /* Counts the total number of registers pushed. */
5051 unsigned num_regs_pushed
= 0;
5054 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5058 #ifdef MATCH_MAY_ALLOCATE
5059 /* Do not bother to initialize all the register variables if there are
5060 no groups in the pattern, as it takes a fair amount of time. If
5061 there are groups, we include space for register 0 (the whole
5062 pattern), even though we never use it, since it simplifies the
5063 array indexing. We should fix this. */
5066 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5067 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5068 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5069 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5071 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5079 /* We must initialize all our variables to NULL, so that
5080 `FREE_VARIABLES' doesn't try to free them. */
5081 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5083 #endif /* MATCH_MAY_ALLOCATE */
5085 /* The starting position is bogus. */
5086 if (pos
< 0 || pos
> size1
+ size2
)
5092 /* Initialize subexpression text positions to -1 to mark ones that no
5093 start_memory/stop_memory has been seen for. Also initialize the
5094 register information struct. */
5095 for (reg
= 1; reg
< num_regs
; reg
++)
5096 regstart
[reg
] = regend
[reg
] = NULL
;
5098 /* We move `string1' into `string2' if the latter's empty -- but not if
5099 `string1' is null. */
5100 if (size2
== 0 && string1
!= NULL
)
5107 end1
= string1
+ size1
;
5108 end2
= string2
+ size2
;
5110 /* `p' scans through the pattern as `d' scans through the data.
5111 `dend' is the end of the input string that `d' points within. `d'
5112 is advanced into the following input string whenever necessary, but
5113 this happens before fetching; therefore, at the beginning of the
5114 loop, `d' can be pointing at the end of a string, but it cannot
5118 /* Only match within string2. */
5119 d
= string2
+ pos
- size1
;
5120 dend
= end_match_2
= string2
+ stop
- size1
;
5121 end_match_1
= end1
; /* Just to give it a value. */
5127 /* Only match within string1. */
5128 end_match_1
= string1
+ stop
;
5130 When we reach end_match_1, PREFETCH normally switches to string2.
5131 But in the present case, this means that just doing a PREFETCH
5132 makes us jump from `stop' to `gap' within the string.
5133 What we really want here is for the search to stop as
5134 soon as we hit end_match_1. That's why we set end_match_2
5135 to end_match_1 (since PREFETCH fails as soon as we hit
5137 end_match_2
= end_match_1
;
5140 { /* It's important to use this code when stop == size so that
5141 moving `d' from end1 to string2 will not prevent the d == dend
5142 check from catching the end of string. */
5144 end_match_2
= string2
+ stop
- size1
;
5150 DEBUG_PRINT1 ("The compiled pattern is: ");
5151 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5152 DEBUG_PRINT1 ("The string to match is: `");
5153 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5154 DEBUG_PRINT1 ("'\n");
5156 /* This loops over pattern commands. It exits by returning from the
5157 function if the match is complete, or it drops through if the match
5158 fails at this starting point in the input data. */
5161 DEBUG_PRINT2 ("\n%p: ", p
);
5164 { /* End of pattern means we might have succeeded. */
5165 DEBUG_PRINT1 ("end of pattern ... ");
5167 /* If we haven't matched the entire string, and we want the
5168 longest match, try backtracking. */
5169 if (d
!= end_match_2
)
5171 /* 1 if this match ends in the same string (string1 or string2)
5172 as the best previous match. */
5173 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5174 == FIRST_STRING_P (d
));
5175 /* 1 if this match is the best seen so far. */
5176 boolean best_match_p
;
5178 /* AIX compiler got confused when this was combined
5179 with the previous declaration. */
5181 best_match_p
= d
> match_end
;
5183 best_match_p
= !FIRST_STRING_P (d
);
5185 DEBUG_PRINT1 ("backtracking.\n");
5187 if (!FAIL_STACK_EMPTY ())
5188 { /* More failure points to try. */
5190 /* If exceeds best match so far, save it. */
5191 if (!best_regs_set
|| best_match_p
)
5193 best_regs_set
= true;
5196 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5198 for (reg
= 1; reg
< num_regs
; reg
++)
5200 best_regstart
[reg
] = regstart
[reg
];
5201 best_regend
[reg
] = regend
[reg
];
5207 /* If no failure points, don't restore garbage. And if
5208 last match is real best match, don't restore second
5210 else if (best_regs_set
&& !best_match_p
)
5213 /* Restore best match. It may happen that `dend ==
5214 end_match_1' while the restored d is in string2.
5215 For example, the pattern `x.*y.*z' against the
5216 strings `x-' and `y-z-', if the two strings are
5217 not consecutive in memory. */
5218 DEBUG_PRINT1 ("Restoring best registers.\n");
5221 dend
= ((d
>= string1
&& d
<= end1
)
5222 ? end_match_1
: end_match_2
);
5224 for (reg
= 1; reg
< num_regs
; reg
++)
5226 regstart
[reg
] = best_regstart
[reg
];
5227 regend
[reg
] = best_regend
[reg
];
5230 } /* d != end_match_2 */
5233 DEBUG_PRINT1 ("Accepting match.\n");
5235 /* If caller wants register contents data back, do it. */
5236 if (regs
&& !bufp
->no_sub
)
5238 /* Have the register data arrays been allocated? */
5239 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5240 { /* No. So allocate them with malloc. We need one
5241 extra element beyond `num_regs' for the `-1' marker
5243 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5244 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5245 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5246 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5251 bufp
->regs_allocated
= REGS_REALLOCATE
;
5253 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5254 { /* Yes. If we need more elements than were already
5255 allocated, reallocate them. If we need fewer, just
5257 if (regs
->num_regs
< num_regs
+ 1)
5259 regs
->num_regs
= num_regs
+ 1;
5260 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5261 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5262 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5271 /* These braces fend off a "empty body in an else-statement"
5272 warning under GCC when assert expands to nothing. */
5273 assert (bufp
->regs_allocated
== REGS_FIXED
);
5276 /* Convert the pointer data in `regstart' and `regend' to
5277 indices. Register zero has to be set differently,
5278 since we haven't kept track of any info for it. */
5279 if (regs
->num_regs
> 0)
5281 regs
->start
[0] = pos
;
5282 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5285 /* Go through the first `min (num_regs, regs->num_regs)'
5286 registers, since that is all we initialized. */
5287 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5289 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5290 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5294 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5296 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5300 /* If the regs structure we return has more elements than
5301 were in the pattern, set the extra elements to -1. If
5302 we (re)allocated the registers, this is the case,
5303 because we always allocate enough to have at least one
5305 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5306 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5307 } /* regs && !bufp->no_sub */
5309 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5310 nfailure_points_pushed
, nfailure_points_popped
,
5311 nfailure_points_pushed
- nfailure_points_popped
);
5312 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5314 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5316 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5322 /* Otherwise match next pattern command. */
5323 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5325 /* Ignore these. Used to ignore the n of succeed_n's which
5326 currently have n == 0. */
5328 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5332 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5335 /* Match the next n pattern characters exactly. The following
5336 byte in the pattern defines n, and the n bytes after that
5337 are the characters to match. */
5340 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5342 /* Remember the start point to rollback upon failure. */
5345 /* This is written out as an if-else so we don't waste time
5346 testing `translate' inside the loop. */
5347 if (RE_TRANSLATE_P (translate
))
5352 int pat_charlen
, buf_charlen
;
5353 unsigned int pat_ch
, buf_ch
;
5356 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
5357 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5359 if (RE_TRANSLATE (translate
, buf_ch
)
5368 mcnt
-= pat_charlen
;
5374 /* Avoid compiler whining about comparison being
5380 if (RE_TRANSLATE (translate
, di
) != *p
++)
5405 /* Match any character except possibly a newline or a null. */
5411 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5414 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5415 buf_ch
= TRANSLATE (buf_ch
);
5417 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5419 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5420 && buf_ch
== '\000'))
5423 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5432 register unsigned int c
;
5433 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5436 /* Start of actual range_table, or end of bitmap if there is no
5438 re_char
*range_table
;
5440 /* Nonzero if there is a range table. */
5441 int range_table_exists
;
5443 /* Number of ranges of range table. This is not included
5444 in the initial byte-length of the command. */
5447 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5449 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5451 if (range_table_exists
)
5453 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5454 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5458 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5459 c
= TRANSLATE (c
); /* The character to match. */
5461 if (SINGLE_BYTE_CHAR_P (c
))
5462 { /* Lookup bitmap. */
5463 /* Cast to `unsigned' instead of `unsigned char' in
5464 case the bit list is a full 32 bytes long. */
5465 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5466 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5470 else if (range_table_exists
)
5472 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5474 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5475 | (class_bits
& BIT_MULTIBYTE
)
5476 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5477 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5478 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5479 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5482 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5486 if (range_table_exists
)
5487 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5489 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5491 if (!not) goto fail
;
5498 /* The beginning of a group is represented by start_memory.
5499 The argument is the register number. The text
5500 matched within the group is recorded (in the internal
5501 registers data structure) under the register number. */
5503 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5505 /* In case we need to undo this operation (via backtracking). */
5506 PUSH_FAILURE_REG ((unsigned int)*p
);
5509 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5510 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5512 /* Move past the register number and inner group count. */
5517 /* The stop_memory opcode represents the end of a group. Its
5518 argument is the same as start_memory's: the register number. */
5520 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5522 assert (!REG_UNSET (regstart
[*p
]));
5523 /* Strictly speaking, there should be code such as:
5525 assert (REG_UNSET (regend[*p]));
5526 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5528 But the only info to be pushed is regend[*p] and it is known to
5529 be UNSET, so there really isn't anything to push.
5530 Not pushing anything, on the other hand deprives us from the
5531 guarantee that regend[*p] is UNSET since undoing this operation
5532 will not reset its value properly. This is not important since
5533 the value will only be read on the next start_memory or at
5534 the very end and both events can only happen if this stop_memory
5538 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5540 /* Move past the register number and the inner group count. */
5545 /* \<digit> has been turned into a `duplicate' command which is
5546 followed by the numeric value of <digit> as the register number. */
5549 register re_char
*d2
, *dend2
;
5550 int regno
= *p
++; /* Get which register to match against. */
5551 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5553 /* Can't back reference a group which we've never matched. */
5554 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5557 /* Where in input to try to start matching. */
5558 d2
= regstart
[regno
];
5560 /* Remember the start point to rollback upon failure. */
5563 /* Where to stop matching; if both the place to start and
5564 the place to stop matching are in the same string, then
5565 set to the place to stop, otherwise, for now have to use
5566 the end of the first string. */
5568 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5569 == FIRST_STRING_P (regend
[regno
]))
5570 ? regend
[regno
] : end_match_1
);
5573 /* If necessary, advance to next segment in register
5577 if (dend2
== end_match_2
) break;
5578 if (dend2
== regend
[regno
]) break;
5580 /* End of string1 => advance to string2. */
5582 dend2
= regend
[regno
];
5584 /* At end of register contents => success */
5585 if (d2
== dend2
) break;
5587 /* If necessary, advance to next segment in data. */
5590 /* How many characters left in this segment to match. */
5593 /* Want how many consecutive characters we can match in
5594 one shot, so, if necessary, adjust the count. */
5595 if (mcnt
> dend2
- d2
)
5598 /* Compare that many; failure if mismatch, else move
5600 if (RE_TRANSLATE_P (translate
)
5601 ? bcmp_translate (d
, d2
, mcnt
, translate
, multibyte
)
5602 : memcmp (d
, d2
, mcnt
))
5607 d
+= mcnt
, d2
+= mcnt
;
5613 /* begline matches the empty string at the beginning of the string
5614 (unless `not_bol' is set in `bufp'), and after newlines. */
5616 DEBUG_PRINT1 ("EXECUTING begline.\n");
5618 if (AT_STRINGS_BEG (d
))
5620 if (!bufp
->not_bol
) break;
5625 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5629 /* In all other cases, we fail. */
5633 /* endline is the dual of begline. */
5635 DEBUG_PRINT1 ("EXECUTING endline.\n");
5637 if (AT_STRINGS_END (d
))
5639 if (!bufp
->not_eol
) break;
5643 PREFETCH_NOLIMIT ();
5650 /* Match at the very beginning of the data. */
5652 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5653 if (AT_STRINGS_BEG (d
))
5658 /* Match at the very end of the data. */
5660 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5661 if (AT_STRINGS_END (d
))
5666 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5667 pushes NULL as the value for the string on the stack. Then
5668 `POP_FAILURE_POINT' will keep the current value for the
5669 string, instead of restoring it. To see why, consider
5670 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5671 then the . fails against the \n. But the next thing we want
5672 to do is match the \n against the \n; if we restored the
5673 string value, we would be back at the foo.
5675 Because this is used only in specific cases, we don't need to
5676 check all the things that `on_failure_jump' does, to make
5677 sure the right things get saved on the stack. Hence we don't
5678 share its code. The only reason to push anything on the
5679 stack at all is that otherwise we would have to change
5680 `anychar's code to do something besides goto fail in this
5681 case; that seems worse than this. */
5682 case on_failure_keep_string_jump
:
5683 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5684 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5687 PUSH_FAILURE_POINT (p
- 3, NULL
);
5690 /* A nasty loop is introduced by the non-greedy *? and +?.
5691 With such loops, the stack only ever contains one failure point
5692 at a time, so that a plain on_failure_jump_loop kind of
5693 cycle detection cannot work. Worse yet, such a detection
5694 can not only fail to detect a cycle, but it can also wrongly
5695 detect a cycle (between different instantiations of the same
5697 So the method used for those nasty loops is a little different:
5698 We use a special cycle-detection-stack-frame which is pushed
5699 when the on_failure_jump_nastyloop failure-point is *popped*.
5700 This special frame thus marks the beginning of one iteration
5701 through the loop and we can hence easily check right here
5702 whether something matched between the beginning and the end of
5704 case on_failure_jump_nastyloop
:
5705 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5706 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5709 assert ((re_opcode_t
)p
[-4] == no_op
);
5712 CHECK_INFINITE_LOOP (p
- 4, d
);
5714 /* If there's a cycle, just continue without pushing
5715 this failure point. The failure point is the "try again"
5716 option, which shouldn't be tried.
5717 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5718 PUSH_FAILURE_POINT (p
- 3, d
);
5722 /* Simple loop detecting on_failure_jump: just check on the
5723 failure stack if the same spot was already hit earlier. */
5724 case on_failure_jump_loop
:
5726 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5727 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5731 CHECK_INFINITE_LOOP (p
- 3, d
);
5733 /* If there's a cycle, get out of the loop, as if the matching
5734 had failed. We used to just `goto fail' here, but that was
5735 aborting the search a bit too early: we want to keep the
5736 empty-loop-match and keep matching after the loop.
5737 We want (x?)*y\1z to match both xxyz and xxyxz. */
5740 PUSH_FAILURE_POINT (p
- 3, d
);
5745 /* Uses of on_failure_jump:
5747 Each alternative starts with an on_failure_jump that points
5748 to the beginning of the next alternative. Each alternative
5749 except the last ends with a jump that in effect jumps past
5750 the rest of the alternatives. (They really jump to the
5751 ending jump of the following alternative, because tensioning
5752 these jumps is a hassle.)
5754 Repeats start with an on_failure_jump that points past both
5755 the repetition text and either the following jump or
5756 pop_failure_jump back to this on_failure_jump. */
5757 case on_failure_jump
:
5758 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5759 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5762 PUSH_FAILURE_POINT (p
-3, d
);
5765 /* This operation is used for greedy *.
5766 Compare the beginning of the repeat with what in the
5767 pattern follows its end. If we can establish that there
5768 is nothing that they would both match, i.e., that we
5769 would have to backtrack because of (as in, e.g., `a*a')
5770 then we can use a non-backtracking loop based on
5771 on_failure_keep_string_jump instead of on_failure_jump. */
5772 case on_failure_jump_smart
:
5773 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5774 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5777 re_char
*p1
= p
; /* Next operation. */
5778 /* Here, we discard `const', making re_match non-reentrant. */
5779 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5780 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5782 p
-= 3; /* Reset so that we will re-execute the
5783 instruction once it's been changed. */
5785 EXTRACT_NUMBER (mcnt
, p2
- 2);
5787 /* Ensure this is a indeed the trivial kind of loop
5788 we are expecting. */
5789 assert (skip_one_char (p1
) == p2
- 3);
5790 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5791 DEBUG_STATEMENT (debug
+= 2);
5792 if (mutually_exclusive_p (bufp
, p1
, p2
))
5794 /* Use a fast `on_failure_keep_string_jump' loop. */
5795 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5796 *p3
= (unsigned char) on_failure_keep_string_jump
;
5797 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5801 /* Default to a safe `on_failure_jump' loop. */
5802 DEBUG_PRINT1 (" smart default => slow loop.\n");
5803 *p3
= (unsigned char) on_failure_jump
;
5805 DEBUG_STATEMENT (debug
-= 2);
5809 /* Unconditionally jump (without popping any failure points). */
5812 IMMEDIATE_QUIT_CHECK
;
5813 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5814 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5815 p
+= mcnt
; /* Do the jump. */
5816 DEBUG_PRINT2 ("(to %p).\n", p
);
5820 /* Have to succeed matching what follows at least n times.
5821 After that, handle like `on_failure_jump'. */
5823 /* Signedness doesn't matter since we only compare MCNT to 0. */
5824 EXTRACT_NUMBER (mcnt
, p
+ 2);
5825 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5827 /* Originally, mcnt is how many times we HAVE to succeed. */
5830 /* Here, we discard `const', making re_match non-reentrant. */
5831 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5834 PUSH_NUMBER (p2
, mcnt
);
5837 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5842 /* Signedness doesn't matter since we only compare MCNT to 0. */
5843 EXTRACT_NUMBER (mcnt
, p
+ 2);
5844 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5846 /* Originally, this is how many times we CAN jump. */
5849 /* Here, we discard `const', making re_match non-reentrant. */
5850 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5852 PUSH_NUMBER (p2
, mcnt
);
5853 goto unconditional_jump
;
5855 /* If don't have to jump any more, skip over the rest of command. */
5862 unsigned char *p2
; /* Location of the counter. */
5863 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5865 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5866 /* Here, we discard `const', making re_match non-reentrant. */
5867 p2
= (unsigned char*) p
+ mcnt
;
5868 /* Signedness doesn't matter since we only copy MCNT's bits . */
5869 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5870 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
5871 PUSH_NUMBER (p2
, mcnt
);
5877 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5878 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5880 /* We SUCCEED (or FAIL) in one of the following cases: */
5882 /* Case 1: D is at the beginning or the end of string. */
5883 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5887 /* C1 is the character before D, S1 is the syntax of C1, C2
5888 is the character at D, and S2 is the syntax of C2. */
5892 int offset
= PTR_TO_OFFSET (d
- 1);
5893 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5894 UPDATE_SYNTAX_TABLE (charpos
);
5896 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5899 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5901 PREFETCH_NOLIMIT ();
5902 c2
= RE_STRING_CHAR (d
, dend
- d
);
5905 if (/* Case 2: Only one of S1 and S2 is Sword. */
5906 ((s1
== Sword
) != (s2
== Sword
))
5907 /* Case 3: Both of S1 and S2 are Sword, and macro
5908 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5909 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5918 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5920 /* We FAIL in one of the following cases: */
5922 /* Case 1: D is at the end of string. */
5923 if (AT_STRINGS_END (d
))
5927 /* C1 is the character before D, S1 is the syntax of C1, C2
5928 is the character at D, and S2 is the syntax of C2. */
5932 int offset
= PTR_TO_OFFSET (d
);
5933 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5934 UPDATE_SYNTAX_TABLE (charpos
);
5937 c2
= RE_STRING_CHAR (d
, dend
- d
);
5940 /* Case 2: S2 is not Sword. */
5944 /* Case 3: D is not at the beginning of string ... */
5945 if (!AT_STRINGS_BEG (d
))
5947 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5949 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5953 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5955 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5962 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5964 /* We FAIL in one of the following cases: */
5966 /* Case 1: D is at the beginning of string. */
5967 if (AT_STRINGS_BEG (d
))
5971 /* C1 is the character before D, S1 is the syntax of C1, C2
5972 is the character at D, and S2 is the syntax of C2. */
5976 int offset
= PTR_TO_OFFSET (d
) - 1;
5977 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5978 UPDATE_SYNTAX_TABLE (charpos
);
5980 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5983 /* Case 2: S1 is not Sword. */
5987 /* Case 3: D is not at the end of string ... */
5988 if (!AT_STRINGS_END (d
))
5990 PREFETCH_NOLIMIT ();
5991 c2
= RE_STRING_CHAR (d
, dend
- d
);
5993 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5997 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5999 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6006 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6008 /* We FAIL in one of the following cases: */
6010 /* Case 1: D is at the end of string. */
6011 if (AT_STRINGS_END (d
))
6015 /* C1 is the character before D, S1 is the syntax of C1, C2
6016 is the character at D, and S2 is the syntax of C2. */
6020 int offset
= PTR_TO_OFFSET (d
);
6021 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6022 UPDATE_SYNTAX_TABLE (charpos
);
6025 c2
= RE_STRING_CHAR (d
, dend
- d
);
6028 /* Case 2: S2 is neither Sword nor Ssymbol. */
6029 if (s2
!= Sword
&& s2
!= Ssymbol
)
6032 /* Case 3: D is not at the beginning of string ... */
6033 if (!AT_STRINGS_BEG (d
))
6035 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6037 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6041 /* ... and S1 is Sword or Ssymbol. */
6042 if (s1
== Sword
|| s1
== Ssymbol
)
6049 DEBUG_PRINT1 ("EXECUTING symend.\n");
6051 /* We FAIL in one of the following cases: */
6053 /* Case 1: D is at the beginning of string. */
6054 if (AT_STRINGS_BEG (d
))
6058 /* C1 is the character before D, S1 is the syntax of C1, C2
6059 is the character at D, and S2 is the syntax of C2. */
6063 int offset
= PTR_TO_OFFSET (d
) - 1;
6064 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6065 UPDATE_SYNTAX_TABLE (charpos
);
6067 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6070 /* Case 2: S1 is neither Ssymbol nor Sword. */
6071 if (s1
!= Sword
&& s1
!= Ssymbol
)
6074 /* Case 3: D is not at the end of string ... */
6075 if (!AT_STRINGS_END (d
))
6077 PREFETCH_NOLIMIT ();
6078 c2
= RE_STRING_CHAR (d
, dend
- d
);
6080 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6084 /* ... and S2 is Sword or Ssymbol. */
6085 if (s2
== Sword
|| s2
== Ssymbol
)
6093 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6095 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
6099 int offset
= PTR_TO_OFFSET (d
);
6100 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6101 UPDATE_SYNTAX_TABLE (pos1
);
6108 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
6110 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6118 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6119 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6124 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6125 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6130 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6131 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6136 case notcategoryspec
:
6137 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6139 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
6145 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
6147 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6158 continue; /* Successfully executed one pattern command; keep going. */
6161 /* We goto here if a matching operation fails. */
6163 IMMEDIATE_QUIT_CHECK
;
6164 if (!FAIL_STACK_EMPTY ())
6167 /* A restart point is known. Restore to that state. */
6168 DEBUG_PRINT1 ("\nFAIL:\n");
6169 POP_FAILURE_POINT (str
, pat
);
6170 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
6172 case on_failure_keep_string_jump
:
6173 assert (str
== NULL
);
6174 goto continue_failure_jump
;
6176 case on_failure_jump_nastyloop
:
6177 assert ((re_opcode_t
)pat
[-2] == no_op
);
6178 PUSH_FAILURE_POINT (pat
- 2, str
);
6181 case on_failure_jump_loop
:
6182 case on_failure_jump
:
6185 continue_failure_jump
:
6186 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6191 /* A special frame used for nastyloops. */
6198 assert (p
>= bufp
->buffer
&& p
<= pend
);
6200 if (d
>= string1
&& d
<= end1
)
6204 break; /* Matching at this starting point really fails. */
6208 goto restore_best_regs
;
6212 return -1; /* Failure to match. */
6215 /* Subroutine definitions for re_match_2. */
6217 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6218 bytes; nonzero otherwise. */
6221 bcmp_translate (s1
, s2
, len
, translate
, multibyte
)
6224 RE_TRANSLATE_TYPE translate
;
6225 const int multibyte
;
6227 register re_char
*p1
= s1
, *p2
= s2
;
6228 re_char
*p1_end
= s1
+ len
;
6229 re_char
*p2_end
= s2
+ len
;
6231 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6232 different lengths, but relying on a single `len' would break this. -sm */
6233 while (p1
< p1_end
&& p2
< p2_end
)
6235 int p1_charlen
, p2_charlen
;
6236 re_wchar_t p1_ch
, p2_ch
;
6238 p1_ch
= RE_STRING_CHAR_AND_LENGTH (p1
, p1_end
- p1
, p1_charlen
);
6239 p2_ch
= RE_STRING_CHAR_AND_LENGTH (p2
, p2_end
- p2
, p2_charlen
);
6241 if (RE_TRANSLATE (translate
, p1_ch
)
6242 != RE_TRANSLATE (translate
, p2_ch
))
6245 p1
+= p1_charlen
, p2
+= p2_charlen
;
6248 if (p1
!= p1_end
|| p2
!= p2_end
)
6254 /* Entry points for GNU code. */
6256 /* re_compile_pattern is the GNU regular expression compiler: it
6257 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6258 Returns 0 if the pattern was valid, otherwise an error string.
6260 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6261 are set in BUFP on entry.
6263 We call regex_compile to do the actual compilation. */
6266 re_compile_pattern (pattern
, length
, bufp
)
6267 const char *pattern
;
6269 struct re_pattern_buffer
*bufp
;
6274 gl_state
.current_syntax_table
= current_buffer
->syntax_table
;
6277 /* GNU code is written to assume at least RE_NREGS registers will be set
6278 (and at least one extra will be -1). */
6279 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6281 /* And GNU code determines whether or not to get register information
6282 by passing null for the REGS argument to re_match, etc., not by
6286 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6290 return gettext (re_error_msgid
[(int) ret
]);
6292 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6294 /* Entry points compatible with 4.2 BSD regex library. We don't define
6295 them unless specifically requested. */
6297 #if defined _REGEX_RE_COMP || defined _LIBC
6299 /* BSD has one and only one pattern buffer. */
6300 static struct re_pattern_buffer re_comp_buf
;
6304 /* Make these definitions weak in libc, so POSIX programs can redefine
6305 these names if they don't use our functions, and still use
6306 regcomp/regexec below without link errors. */
6316 if (!re_comp_buf
.buffer
)
6317 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6318 return (char *) gettext ("No previous regular expression");
6322 if (!re_comp_buf
.buffer
)
6324 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6325 if (re_comp_buf
.buffer
== NULL
)
6326 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6327 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6328 re_comp_buf
.allocated
= 200;
6330 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6331 if (re_comp_buf
.fastmap
== NULL
)
6332 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6333 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6336 /* Since `re_exec' always passes NULL for the `regs' argument, we
6337 don't need to initialize the pattern buffer fields which affect it. */
6339 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6344 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6345 return (char *) gettext (re_error_msgid
[(int) ret
]);
6356 const int len
= strlen (s
);
6358 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6360 #endif /* _REGEX_RE_COMP */
6362 /* POSIX.2 functions. Don't define these for Emacs. */
6366 /* regcomp takes a regular expression as a string and compiles it.
6368 PREG is a regex_t *. We do not expect any fields to be initialized,
6369 since POSIX says we shouldn't. Thus, we set
6371 `buffer' to the compiled pattern;
6372 `used' to the length of the compiled pattern;
6373 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6374 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6375 RE_SYNTAX_POSIX_BASIC;
6376 `fastmap' to an allocated space for the fastmap;
6377 `fastmap_accurate' to zero;
6378 `re_nsub' to the number of subexpressions in PATTERN.
6380 PATTERN is the address of the pattern string.
6382 CFLAGS is a series of bits which affect compilation.
6384 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6385 use POSIX basic syntax.
6387 If REG_NEWLINE is set, then . and [^...] don't match newline.
6388 Also, regexec will try a match beginning after every newline.
6390 If REG_ICASE is set, then we considers upper- and lowercase
6391 versions of letters to be equivalent when matching.
6393 If REG_NOSUB is set, then when PREG is passed to regexec, that
6394 routine will report only success or failure, and nothing about the
6397 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6398 the return codes and their meanings.) */
6401 regcomp (preg
, pattern
, cflags
)
6402 regex_t
*__restrict preg
;
6403 const char *__restrict pattern
;
6408 = (cflags
& REG_EXTENDED
) ?
6409 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6411 /* regex_compile will allocate the space for the compiled pattern. */
6413 preg
->allocated
= 0;
6416 /* Try to allocate space for the fastmap. */
6417 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6419 if (cflags
& REG_ICASE
)
6424 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6425 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6426 if (preg
->translate
== NULL
)
6427 return (int) REG_ESPACE
;
6429 /* Map uppercase characters to corresponding lowercase ones. */
6430 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6431 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6434 preg
->translate
= NULL
;
6436 /* If REG_NEWLINE is set, newlines are treated differently. */
6437 if (cflags
& REG_NEWLINE
)
6438 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6439 syntax
&= ~RE_DOT_NEWLINE
;
6440 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6443 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6445 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6447 /* POSIX says a null character in the pattern terminates it, so we
6448 can use strlen here in compiling the pattern. */
6449 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6451 /* POSIX doesn't distinguish between an unmatched open-group and an
6452 unmatched close-group: both are REG_EPAREN. */
6453 if (ret
== REG_ERPAREN
)
6456 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6457 { /* Compute the fastmap now, since regexec cannot modify the pattern
6459 re_compile_fastmap (preg
);
6460 if (preg
->can_be_null
)
6461 { /* The fastmap can't be used anyway. */
6462 free (preg
->fastmap
);
6463 preg
->fastmap
= NULL
;
6468 WEAK_ALIAS (__regcomp
, regcomp
)
6471 /* regexec searches for a given pattern, specified by PREG, in the
6474 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6475 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6476 least NMATCH elements, and we set them to the offsets of the
6477 corresponding matched substrings.
6479 EFLAGS specifies `execution flags' which affect matching: if
6480 REG_NOTBOL is set, then ^ does not match at the beginning of the
6481 string; if REG_NOTEOL is set, then $ does not match at the end.
6483 We return 0 if we find a match and REG_NOMATCH if not. */
6486 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6487 const regex_t
*__restrict preg
;
6488 const char *__restrict string
;
6490 regmatch_t pmatch
[__restrict_arr
];
6494 struct re_registers regs
;
6495 regex_t private_preg
;
6496 int len
= strlen (string
);
6497 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6499 private_preg
= *preg
;
6501 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6502 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6504 /* The user has told us exactly how many registers to return
6505 information about, via `nmatch'. We have to pass that on to the
6506 matching routines. */
6507 private_preg
.regs_allocated
= REGS_FIXED
;
6511 regs
.num_regs
= nmatch
;
6512 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6513 if (regs
.start
== NULL
)
6514 return (int) REG_NOMATCH
;
6515 regs
.end
= regs
.start
+ nmatch
;
6518 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6519 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6520 was a little bit longer but still only matching the real part.
6521 This works because the `endline' will check for a '\n' and will find a
6522 '\0', correctly deciding that this is not the end of a line.
6523 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6524 a convenient '\0' there. For all we know, the string could be preceded
6525 by '\n' which would throw things off. */
6527 /* Perform the searching operation. */
6528 ret
= re_search (&private_preg
, string
, len
,
6529 /* start: */ 0, /* range: */ len
,
6530 want_reg_info
? ®s
: (struct re_registers
*) 0);
6532 /* Copy the register information to the POSIX structure. */
6539 for (r
= 0; r
< nmatch
; r
++)
6541 pmatch
[r
].rm_so
= regs
.start
[r
];
6542 pmatch
[r
].rm_eo
= regs
.end
[r
];
6546 /* If we needed the temporary register info, free the space now. */
6550 /* We want zero return to mean success, unlike `re_search'. */
6551 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6553 WEAK_ALIAS (__regexec
, regexec
)
6556 /* Returns a message corresponding to an error code, ERR_CODE, returned
6557 from either regcomp or regexec. We don't use PREG here.
6559 ERR_CODE was previously called ERRCODE, but that name causes an
6560 error with msvc8 compiler. */
6563 regerror (err_code
, preg
, errbuf
, errbuf_size
)
6565 const regex_t
*preg
;
6573 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6574 /* Only error codes returned by the rest of the code should be passed
6575 to this routine. If we are given anything else, or if other regex
6576 code generates an invalid error code, then the program has a bug.
6577 Dump core so we can fix it. */
6580 msg
= gettext (re_error_msgid
[err_code
]);
6582 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6584 if (errbuf_size
!= 0)
6586 if (msg_size
> errbuf_size
)
6588 strncpy (errbuf
, msg
, errbuf_size
- 1);
6589 errbuf
[errbuf_size
- 1] = 0;
6592 strcpy (errbuf
, msg
);
6597 WEAK_ALIAS (__regerror
, regerror
)
6600 /* Free dynamically allocated space used by PREG. */
6606 if (preg
->buffer
!= NULL
)
6607 free (preg
->buffer
);
6608 preg
->buffer
= NULL
;
6610 preg
->allocated
= 0;
6613 if (preg
->fastmap
!= NULL
)
6614 free (preg
->fastmap
);
6615 preg
->fastmap
= NULL
;
6616 preg
->fastmap_accurate
= 0;
6618 if (preg
->translate
!= NULL
)
6619 free (preg
->translate
);
6620 preg
->translate
= NULL
;
6622 WEAK_ALIAS (__regfree
, regfree
)
6624 #endif /* not emacs */
6626 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6627 (do not change this comment) */