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, 2008, 2009, 2010
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. */
124 /* Make syntax table lookup grant data in gl_state. */
125 # define SYNTAX_ENTRY_VIA_PROPERTY
128 # include "character.h"
129 # include "category.h"
134 # define malloc xmalloc
138 # define realloc xrealloc
144 /* Converts the pointer to the char to BEG-based offset from the start. */
145 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
146 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
148 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
149 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
150 # define RE_STRING_CHAR(p, multibyte) \
151 (multibyte ? (STRING_CHAR (p)) : (*(p)))
152 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
153 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
155 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
157 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
159 /* Set C a (possibly converted to multibyte) character before P. P
160 points into a string which is the virtual concatenation of STR1
161 (which ends at END1) or STR2 (which ends at END2). */
162 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
164 if (target_multibyte) \
166 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
167 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
168 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
169 c = STRING_CHAR (dtemp); \
173 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
174 (c) = RE_CHAR_TO_MULTIBYTE (c); \
178 /* Set C a (possibly converted to multibyte) character at P, and set
179 LEN to the byte length of that character. */
180 # define GET_CHAR_AFTER(c, p, len) \
182 if (target_multibyte) \
183 (c) = STRING_CHAR_AND_LENGTH (p, len); \
188 (c) = RE_CHAR_TO_MULTIBYTE (c); \
192 #else /* not emacs */
194 /* If we are not linking with Emacs proper,
195 we can't use the relocating allocator
196 even if config.h says that we can. */
199 # if defined STDC_HEADERS || defined _LIBC
206 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
213 val
= (void *) malloc (size
);
216 write (2, "virtual memory exhausted\n", 25);
223 xrealloc (block
, size
)
228 /* We must call malloc explicitly when BLOCK is 0, since some
229 reallocs don't do this. */
231 val
= (void *) malloc (size
);
233 val
= (void *) realloc (block
, size
);
236 write (2, "virtual memory exhausted\n", 25);
245 # define malloc xmalloc
249 # define realloc xrealloc
251 /* This is the normal way of making sure we have memcpy, memcmp and memset. */
252 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
255 # include <strings.h>
257 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
260 # 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 # define SWITCH_ENUM_CAST(x) (x)
271 /* Dummy macros for non-Emacs environments. */
272 # define CHAR_CHARSET(c) 0
273 # define CHARSET_LEADING_CODE_BASE(c) 0
274 # define MAX_MULTIBYTE_LENGTH 1
275 # define RE_MULTIBYTE_P(x) 0
276 # define RE_TARGET_MULTIBYTE_P(x) 0
277 # define WORD_BOUNDARY_P(c1, c2) (0)
278 # define CHAR_HEAD_P(p) (1)
279 # define SINGLE_BYTE_CHAR_P(c) (1)
280 # define SAME_CHARSET_P(c1, c2) (1)
281 # define BYTES_BY_CHAR_HEAD(p) (1)
282 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
283 # define STRING_CHAR(p) (*(p))
284 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
285 # define CHAR_STRING(c, s) (*(s) = (c), 1)
286 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
287 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
288 # define RE_CHAR_TO_MULTIBYTE(c) (c)
289 # define RE_CHAR_TO_UNIBYTE(c) (c)
290 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
291 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
292 # define GET_CHAR_AFTER(c, p, len) \
294 # define MAKE_CHAR(charset, c1, c2) (c1)
295 # define BYTE8_TO_CHAR(c) (c)
296 # define CHAR_BYTE8_P(c) (0)
297 # define CHAR_LEADING_CODE(c) (c)
299 #endif /* not emacs */
302 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
303 # define RE_TRANSLATE_P(TBL) (TBL)
306 /* Get the interface, including the syntax bits. */
309 /* isalpha etc. are used for the character classes. */
314 /* 1 if C is an ASCII character. */
315 # define IS_REAL_ASCII(c) ((c) < 0200)
317 /* 1 if C is a unibyte character. */
318 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
320 /* The Emacs definitions should not be directly affected by locales. */
322 /* In Emacs, these are only used for single-byte characters. */
323 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
324 # define ISCNTRL(c) ((c) < ' ')
325 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
326 || ((c) >= 'a' && (c) <= 'f') \
327 || ((c) >= 'A' && (c) <= 'F'))
329 /* This is only used for single-byte characters. */
330 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
332 /* The rest must handle multibyte characters. */
334 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
335 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
338 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
339 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
342 # define ISALNUM(c) (IS_REAL_ASCII (c) \
343 ? (((c) >= 'a' && (c) <= 'z') \
344 || ((c) >= 'A' && (c) <= 'Z') \
345 || ((c) >= '0' && (c) <= '9')) \
346 : SYNTAX (c) == Sword)
348 # define ISALPHA(c) (IS_REAL_ASCII (c) \
349 ? (((c) >= 'a' && (c) <= 'z') \
350 || ((c) >= 'A' && (c) <= 'Z')) \
351 : SYNTAX (c) == Sword)
353 # define ISLOWER(c) (LOWERCASEP (c))
355 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
356 ? ((c) > ' ' && (c) < 0177 \
357 && !(((c) >= 'a' && (c) <= 'z') \
358 || ((c) >= 'A' && (c) <= 'Z') \
359 || ((c) >= '0' && (c) <= '9'))) \
360 : SYNTAX (c) != Sword)
362 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
364 # define ISUPPER(c) (UPPERCASEP (c))
366 # define ISWORD(c) (SYNTAX (c) == Sword)
368 #else /* not emacs */
370 /* Jim Meyering writes:
372 "... Some ctype macros are valid only for character codes that
373 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
374 using /bin/cc or gcc but without giving an ansi option). So, all
375 ctype uses should be through macros like ISPRINT... If
376 STDC_HEADERS is defined, then autoconf has verified that the ctype
377 macros don't need to be guarded with references to isascii. ...
378 Defining isascii to 1 should let any compiler worth its salt
379 eliminate the && through constant folding."
380 Solaris defines some of these symbols so we must undefine them first. */
383 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
384 # define ISASCII(c) 1
386 # define ISASCII(c) isascii(c)
389 /* 1 if C is an ASCII character. */
390 # define IS_REAL_ASCII(c) ((c) < 0200)
392 /* This distinction is not meaningful, except in Emacs. */
393 # define ISUNIBYTE(c) 1
396 # define ISBLANK(c) (ISASCII (c) && isblank (c))
398 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
401 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
403 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
407 # define ISPRINT(c) (ISASCII (c) && isprint (c))
408 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
409 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
410 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
411 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
412 # define ISLOWER(c) (ISASCII (c) && islower (c))
413 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
414 # define ISSPACE(c) (ISASCII (c) && isspace (c))
415 # define ISUPPER(c) (ISASCII (c) && isupper (c))
416 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
418 # define ISWORD(c) ISALPHA(c)
421 # define TOLOWER(c) _tolower(c)
423 # define TOLOWER(c) tolower(c)
426 /* How many characters in the character set. */
427 # define CHAR_SET_SIZE 256
431 extern char *re_syntax_table
;
433 # else /* not SYNTAX_TABLE */
435 static char re_syntax_table
[CHAR_SET_SIZE
];
446 memset (re_syntax_table
, 0, sizeof re_syntax_table
);
448 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
450 re_syntax_table
[c
] = Sword
;
452 re_syntax_table
['_'] = Ssymbol
;
457 # endif /* not SYNTAX_TABLE */
459 # define SYNTAX(c) re_syntax_table[(c)]
461 #endif /* not emacs */
464 # define NULL (void *)0
467 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
468 since ours (we hope) works properly with all combinations of
469 machines, compilers, `char' and `unsigned char' argument types.
470 (Per Bothner suggested the basic approach.) */
471 #undef SIGN_EXTEND_CHAR
473 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
474 #else /* not __STDC__ */
475 /* As in Harbison and Steele. */
476 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
479 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
480 use `alloca' instead of `malloc'. This is because using malloc in
481 re_search* or re_match* could cause memory leaks when C-g is used in
482 Emacs; also, malloc is slower and causes storage fragmentation. On
483 the other hand, malloc is more portable, and easier to debug.
485 Because we sometimes use alloca, some routines have to be macros,
486 not functions -- `alloca'-allocated space disappears at the end of the
487 function it is called in. */
491 # define REGEX_ALLOCATE malloc
492 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
493 # define REGEX_FREE free
495 #else /* not REGEX_MALLOC */
497 /* Emacs already defines alloca, sometimes. */
500 /* Make alloca work the best possible way. */
502 # define alloca __builtin_alloca
503 # else /* not __GNUC__ */
504 # ifdef HAVE_ALLOCA_H
506 # endif /* HAVE_ALLOCA_H */
507 # endif /* not __GNUC__ */
509 # endif /* not alloca */
511 # define REGEX_ALLOCATE alloca
513 /* Assumes a `char *destination' variable. */
514 # define REGEX_REALLOCATE(source, osize, nsize) \
515 (destination = (char *) alloca (nsize), \
516 memcpy (destination, source, osize))
518 /* No need to do anything to free, after alloca. */
519 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
521 #endif /* not REGEX_MALLOC */
523 /* Define how to allocate the failure stack. */
525 #if defined REL_ALLOC && defined REGEX_MALLOC
527 # define REGEX_ALLOCATE_STACK(size) \
528 r_alloc (&failure_stack_ptr, (size))
529 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
530 r_re_alloc (&failure_stack_ptr, (nsize))
531 # define REGEX_FREE_STACK(ptr) \
532 r_alloc_free (&failure_stack_ptr)
534 #else /* not using relocating allocator */
538 # define REGEX_ALLOCATE_STACK malloc
539 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
540 # define REGEX_FREE_STACK free
542 # else /* not REGEX_MALLOC */
544 # define REGEX_ALLOCATE_STACK alloca
546 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
547 REGEX_REALLOCATE (source, osize, nsize)
548 /* No need to explicitly free anything. */
549 # define REGEX_FREE_STACK(arg) ((void)0)
551 # endif /* not REGEX_MALLOC */
552 #endif /* not using relocating allocator */
555 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
556 `string1' or just past its end. This works if PTR is NULL, which is
558 #define FIRST_STRING_P(ptr) \
559 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
561 /* (Re)Allocate N items of type T using malloc, or fail. */
562 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
563 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
564 #define RETALLOC_IF(addr, n, t) \
565 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
566 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
568 #define BYTEWIDTH 8 /* In bits. */
570 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
574 #define MAX(a, b) ((a) > (b) ? (a) : (b))
575 #define MIN(a, b) ((a) < (b) ? (a) : (b))
577 /* Type of source-pattern and string chars. */
578 typedef const unsigned char re_char
;
580 typedef char boolean
;
584 static int re_match_2_internal
_RE_ARGS ((struct re_pattern_buffer
*bufp
,
585 re_char
*string1
, int size1
,
586 re_char
*string2
, int size2
,
588 struct re_registers
*regs
,
591 /* These are the command codes that appear in compiled regular
592 expressions. Some opcodes are followed by argument bytes. A
593 command code can specify any interpretation whatsoever for its
594 arguments. Zero bytes may appear in the compiled regular expression. */
600 /* Succeed right away--no more backtracking. */
603 /* Followed by one byte giving n, then by n literal bytes. */
606 /* Matches any (more or less) character. */
609 /* Matches any one char belonging to specified set. First
610 following byte is number of bitmap bytes. Then come bytes
611 for a bitmap saying which chars are in. Bits in each byte
612 are ordered low-bit-first. A character is in the set if its
613 bit is 1. A character too large to have a bit in the map is
614 automatically not in the set.
616 If the length byte has the 0x80 bit set, then that stuff
617 is followed by a range table:
618 2 bytes of flags for character sets (low 8 bits, high 8 bits)
619 See RANGE_TABLE_WORK_BITS below.
620 2 bytes, the number of pairs that follow (upto 32767)
621 pairs, each 2 multibyte characters,
622 each multibyte character represented as 3 bytes. */
625 /* Same parameters as charset, but match any character that is
626 not one of those specified. */
629 /* Start remembering the text that is matched, for storing in a
630 register. Followed by one byte with the register number, in
631 the range 0 to one less than the pattern buffer's re_nsub
635 /* Stop remembering the text that is matched and store it in a
636 memory register. Followed by one byte with the register
637 number, in the range 0 to one less than `re_nsub' in the
641 /* Match a duplicate of something remembered. Followed by one
642 byte containing the register number. */
645 /* Fail unless at beginning of line. */
648 /* Fail unless at end of line. */
651 /* Succeeds if at beginning of buffer (if emacs) or at beginning
652 of string to be matched (if not). */
655 /* Analogously, for end of buffer/string. */
658 /* Followed by two byte relative address to which to jump. */
661 /* Followed by two-byte relative address of place to resume at
662 in case of failure. */
665 /* Like on_failure_jump, but pushes a placeholder instead of the
666 current string position when executed. */
667 on_failure_keep_string_jump
,
669 /* Just like `on_failure_jump', except that it checks that we
670 don't get stuck in an infinite loop (matching an empty string
672 on_failure_jump_loop
,
674 /* Just like `on_failure_jump_loop', except that it checks for
675 a different kind of loop (the kind that shows up with non-greedy
676 operators). This operation has to be immediately preceded
678 on_failure_jump_nastyloop
,
680 /* A smart `on_failure_jump' used for greedy * and + operators.
681 It analyses the loop before which it is put and if the
682 loop does not require backtracking, it changes itself to
683 `on_failure_keep_string_jump' and short-circuits the loop,
684 else it just defaults to changing itself into `on_failure_jump'.
685 It assumes that it is pointing to just past a `jump'. */
686 on_failure_jump_smart
,
688 /* Followed by two-byte relative address and two-byte number n.
689 After matching N times, jump to the address upon failure.
690 Does not work if N starts at 0: use on_failure_jump_loop
694 /* Followed by two-byte relative address, and two-byte number n.
695 Jump to the address N times, then fail. */
698 /* Set the following two-byte relative address to the
699 subsequent two-byte number. The address *includes* the two
703 wordbeg
, /* Succeeds if at word beginning. */
704 wordend
, /* Succeeds if at word end. */
706 wordbound
, /* Succeeds if at a word boundary. */
707 notwordbound
, /* Succeeds if not at a word boundary. */
709 symbeg
, /* Succeeds if at symbol beginning. */
710 symend
, /* Succeeds if at symbol end. */
712 /* Matches any character whose syntax is specified. Followed by
713 a byte which contains a syntax code, e.g., Sword. */
716 /* Matches any character whose syntax is not that specified. */
720 ,before_dot
, /* Succeeds if before point. */
721 at_dot
, /* Succeeds if at point. */
722 after_dot
, /* Succeeds if after point. */
724 /* Matches any character whose category-set contains the specified
725 category. The operator is followed by a byte which contains a
726 category code (mnemonic ASCII character). */
729 /* Matches any character whose category-set does not contain the
730 specified category. The operator is followed by a byte which
731 contains the category code (mnemonic ASCII character). */
736 /* Common operations on the compiled pattern. */
738 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
740 #define STORE_NUMBER(destination, number) \
742 (destination)[0] = (number) & 0377; \
743 (destination)[1] = (number) >> 8; \
746 /* Same as STORE_NUMBER, except increment DESTINATION to
747 the byte after where the number is stored. Therefore, DESTINATION
748 must be an lvalue. */
750 #define STORE_NUMBER_AND_INCR(destination, number) \
752 STORE_NUMBER (destination, number); \
753 (destination) += 2; \
756 /* Put into DESTINATION a number stored in two contiguous bytes starting
759 #define EXTRACT_NUMBER(destination, source) \
761 (destination) = *(source) & 0377; \
762 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
766 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
768 extract_number (dest
, source
)
772 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
773 *dest
= *source
& 0377;
777 # ifndef EXTRACT_MACROS /* To debug the macros. */
778 # undef EXTRACT_NUMBER
779 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
780 # endif /* not EXTRACT_MACROS */
784 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
785 SOURCE must be an lvalue. */
787 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
789 EXTRACT_NUMBER (destination, source); \
794 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
797 extract_number_and_incr (destination
, source
)
801 extract_number (destination
, *source
);
805 # ifndef EXTRACT_MACROS
806 # undef EXTRACT_NUMBER_AND_INCR
807 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
808 extract_number_and_incr (&dest, &src)
809 # endif /* not EXTRACT_MACROS */
813 /* Store a multibyte character in three contiguous bytes starting
814 DESTINATION, and increment DESTINATION to the byte after where the
815 character is stored. Therefore, DESTINATION must be an lvalue. */
817 #define STORE_CHARACTER_AND_INCR(destination, character) \
819 (destination)[0] = (character) & 0377; \
820 (destination)[1] = ((character) >> 8) & 0377; \
821 (destination)[2] = (character) >> 16; \
822 (destination) += 3; \
825 /* Put into DESTINATION a character stored in three contiguous bytes
826 starting at SOURCE. */
828 #define EXTRACT_CHARACTER(destination, source) \
830 (destination) = ((source)[0] \
831 | ((source)[1] << 8) \
832 | ((source)[2] << 16)); \
836 /* Macros for charset. */
838 /* Size of bitmap of charset P in bytes. P is a start of charset,
839 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
840 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
842 /* Nonzero if charset P has range table. */
843 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
845 /* Return the address of range table of charset P. But not the start
846 of table itself, but the before where the number of ranges is
847 stored. `2 +' means to skip re_opcode_t and size of bitmap,
848 and the 2 bytes of flags at the start of the range table. */
849 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
851 /* Extract the bit flags that start a range table. */
852 #define CHARSET_RANGE_TABLE_BITS(p) \
853 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
854 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
856 /* Test if C is listed in the bitmap of charset P. */
857 #define CHARSET_LOOKUP_BITMAP(p, c) \
858 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
859 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
861 /* Return the address of end of RANGE_TABLE. COUNT is number of
862 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
863 is start of range and end of range. `* 3' is size of each start
865 #define CHARSET_RANGE_TABLE_END(range_table, count) \
866 ((range_table) + (count) * 2 * 3)
868 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
869 COUNT is number of ranges in RANGE_TABLE. */
870 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
873 re_wchar_t range_start, range_end; \
875 re_char *range_table_end \
876 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
878 for (p = (range_table); p < range_table_end; p += 2 * 3) \
880 EXTRACT_CHARACTER (range_start, p); \
881 EXTRACT_CHARACTER (range_end, p + 3); \
883 if (range_start <= (c) && (c) <= range_end) \
892 /* Test if C is in range table of CHARSET. The flag NOT is negated if
893 C is listed in it. */
894 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
897 /* Number of ranges in range table. */ \
899 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
901 EXTRACT_NUMBER_AND_INCR (count, range_table); \
902 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
906 /* If DEBUG is defined, Regex prints many voluminous messages about what
907 it is doing (if the variable `debug' is nonzero). If linked with the
908 main program in `iregex.c', you can enter patterns and strings
909 interactively. And if linked with the main program in `main.c' and
910 the other test files, you can run the already-written tests. */
914 /* We use standard I/O for debugging. */
917 /* It is useful to test things that ``must'' be true when debugging. */
920 static int debug
= -100000;
922 # define DEBUG_STATEMENT(e) e
923 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
924 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
925 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
926 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
927 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
928 if (debug > 0) print_partial_compiled_pattern (s, e)
929 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
930 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
933 /* Print the fastmap in human-readable form. */
936 print_fastmap (fastmap
)
939 unsigned was_a_range
= 0;
942 while (i
< (1 << BYTEWIDTH
))
948 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
964 /* Print a compiled pattern string in human-readable form, starting at
965 the START pointer into it and ending just before the pointer END. */
968 print_partial_compiled_pattern (start
, end
)
978 fprintf (stderr
, "(null)\n");
982 /* Loop over pattern commands. */
985 fprintf (stderr
, "%d:\t", p
- start
);
987 switch ((re_opcode_t
) *p
++)
990 fprintf (stderr
, "/no_op");
994 fprintf (stderr
, "/succeed");
999 fprintf (stderr
, "/exactn/%d", mcnt
);
1002 fprintf (stderr
, "/%c", *p
++);
1008 fprintf (stderr
, "/start_memory/%d", *p
++);
1012 fprintf (stderr
, "/stop_memory/%d", *p
++);
1016 fprintf (stderr
, "/duplicate/%d", *p
++);
1020 fprintf (stderr
, "/anychar");
1026 register int c
, last
= -100;
1027 register int in_range
= 0;
1028 int length
= CHARSET_BITMAP_SIZE (p
- 1);
1029 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
1031 fprintf (stderr
, "/charset [%s",
1032 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
1035 fprintf (stderr
, " !extends past end of pattern! ");
1037 for (c
= 0; c
< 256; c
++)
1039 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
1041 /* Are we starting a range? */
1042 if (last
+ 1 == c
&& ! in_range
)
1044 fprintf (stderr
, "-");
1047 /* Have we broken a range? */
1048 else if (last
+ 1 != c
&& in_range
)
1050 fprintf (stderr
, "%c", last
);
1055 fprintf (stderr
, "%c", c
);
1061 fprintf (stderr
, "%c", last
);
1063 fprintf (stderr
, "]");
1067 if (has_range_table
)
1070 fprintf (stderr
, "has-range-table");
1072 /* ??? Should print the range table; for now, just skip it. */
1073 p
+= 2; /* skip range table bits */
1074 EXTRACT_NUMBER_AND_INCR (count
, p
);
1075 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1081 fprintf (stderr
, "/begline");
1085 fprintf (stderr
, "/endline");
1088 case on_failure_jump
:
1089 extract_number_and_incr (&mcnt
, &p
);
1090 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1093 case on_failure_keep_string_jump
:
1094 extract_number_and_incr (&mcnt
, &p
);
1095 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1098 case on_failure_jump_nastyloop
:
1099 extract_number_and_incr (&mcnt
, &p
);
1100 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1103 case on_failure_jump_loop
:
1104 extract_number_and_incr (&mcnt
, &p
);
1105 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1108 case on_failure_jump_smart
:
1109 extract_number_and_incr (&mcnt
, &p
);
1110 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1114 extract_number_and_incr (&mcnt
, &p
);
1115 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1119 extract_number_and_incr (&mcnt
, &p
);
1120 extract_number_and_incr (&mcnt2
, &p
);
1121 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1125 extract_number_and_incr (&mcnt
, &p
);
1126 extract_number_and_incr (&mcnt2
, &p
);
1127 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1131 extract_number_and_incr (&mcnt
, &p
);
1132 extract_number_and_incr (&mcnt2
, &p
);
1133 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1137 fprintf (stderr
, "/wordbound");
1141 fprintf (stderr
, "/notwordbound");
1145 fprintf (stderr
, "/wordbeg");
1149 fprintf (stderr
, "/wordend");
1153 fprintf (stderr
, "/symbeg");
1157 fprintf (stderr
, "/symend");
1161 fprintf (stderr
, "/syntaxspec");
1163 fprintf (stderr
, "/%d", mcnt
);
1167 fprintf (stderr
, "/notsyntaxspec");
1169 fprintf (stderr
, "/%d", mcnt
);
1174 fprintf (stderr
, "/before_dot");
1178 fprintf (stderr
, "/at_dot");
1182 fprintf (stderr
, "/after_dot");
1186 fprintf (stderr
, "/categoryspec");
1188 fprintf (stderr
, "/%d", mcnt
);
1191 case notcategoryspec
:
1192 fprintf (stderr
, "/notcategoryspec");
1194 fprintf (stderr
, "/%d", mcnt
);
1199 fprintf (stderr
, "/begbuf");
1203 fprintf (stderr
, "/endbuf");
1207 fprintf (stderr
, "?%d", *(p
-1));
1210 fprintf (stderr
, "\n");
1213 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1218 print_compiled_pattern (bufp
)
1219 struct re_pattern_buffer
*bufp
;
1221 re_char
*buffer
= bufp
->buffer
;
1223 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1224 printf ("%ld bytes used/%ld bytes allocated.\n",
1225 bufp
->used
, bufp
->allocated
);
1227 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1229 printf ("fastmap: ");
1230 print_fastmap (bufp
->fastmap
);
1233 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1234 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1235 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1236 printf ("no_sub: %d\t", bufp
->no_sub
);
1237 printf ("not_bol: %d\t", bufp
->not_bol
);
1238 printf ("not_eol: %d\t", bufp
->not_eol
);
1239 printf ("syntax: %lx\n", bufp
->syntax
);
1241 /* Perhaps we should print the translate table? */
1246 print_double_string (where
, string1
, size1
, string2
, size2
)
1259 if (FIRST_STRING_P (where
))
1261 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1262 putchar (string1
[this_char
]);
1267 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1268 putchar (string2
[this_char
]);
1272 #else /* not DEBUG */
1277 # define DEBUG_STATEMENT(e)
1278 # define DEBUG_PRINT1(x)
1279 # define DEBUG_PRINT2(x1, x2)
1280 # define DEBUG_PRINT3(x1, x2, x3)
1281 # define DEBUG_PRINT4(x1, x2, x3, x4)
1282 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1283 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1285 #endif /* not DEBUG */
1287 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1288 also be assigned to arbitrarily: each pattern buffer stores its own
1289 syntax, so it can be changed between regex compilations. */
1290 /* This has no initializer because initialized variables in Emacs
1291 become read-only after dumping. */
1292 reg_syntax_t re_syntax_options
;
1295 /* Specify the precise syntax of regexps for compilation. This provides
1296 for compatibility for various utilities which historically have
1297 different, incompatible syntaxes.
1299 The argument SYNTAX is a bit mask comprised of the various bits
1300 defined in regex.h. We return the old syntax. */
1303 re_set_syntax (reg_syntax_t syntax
)
1305 reg_syntax_t ret
= re_syntax_options
;
1307 re_syntax_options
= syntax
;
1310 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1312 /* Regexp to use to replace spaces, or NULL meaning don't. */
1313 static re_char
*whitespace_regexp
;
1316 re_set_whitespace_regexp (const char *regexp
)
1318 whitespace_regexp
= (re_char
*) regexp
;
1320 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1322 /* This table gives an error message for each of the error codes listed
1323 in regex.h. Obviously the order here has to be same as there.
1324 POSIX doesn't require that we do anything for REG_NOERROR,
1325 but why not be nice? */
1327 static const char *re_error_msgid
[] =
1329 gettext_noop ("Success"), /* REG_NOERROR */
1330 gettext_noop ("No match"), /* REG_NOMATCH */
1331 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1332 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1333 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1334 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1335 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1336 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1337 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1338 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1339 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1340 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1341 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1342 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1343 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1344 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1345 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1346 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1349 /* Avoiding alloca during matching, to placate r_alloc. */
1351 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1352 searching and matching functions should not call alloca. On some
1353 systems, alloca is implemented in terms of malloc, and if we're
1354 using the relocating allocator routines, then malloc could cause a
1355 relocation, which might (if the strings being searched are in the
1356 ralloc heap) shift the data out from underneath the regexp
1359 Here's another reason to avoid allocation: Emacs
1360 processes input from X in a signal handler; processing X input may
1361 call malloc; if input arrives while a matching routine is calling
1362 malloc, then we're scrod. But Emacs can't just block input while
1363 calling matching routines; then we don't notice interrupts when
1364 they come in. So, Emacs blocks input around all regexp calls
1365 except the matching calls, which it leaves unprotected, in the
1366 faith that they will not malloc. */
1368 /* Normally, this is fine. */
1369 #define MATCH_MAY_ALLOCATE
1371 /* The match routines may not allocate if (1) they would do it with malloc
1372 and (2) it's not safe for them to use malloc.
1373 Note that if REL_ALLOC is defined, matching would not use malloc for the
1374 failure stack, but we would still use it for the register vectors;
1375 so REL_ALLOC should not affect this. */
1376 #if defined REGEX_MALLOC && defined emacs
1377 # undef MATCH_MAY_ALLOCATE
1381 /* Failure stack declarations and macros; both re_compile_fastmap and
1382 re_match_2 use a failure stack. These have to be macros because of
1383 REGEX_ALLOCATE_STACK. */
1386 /* Approximate number of failure points for which to initially allocate space
1387 when matching. If this number is exceeded, we allocate more
1388 space, so it is not a hard limit. */
1389 #ifndef INIT_FAILURE_ALLOC
1390 # define INIT_FAILURE_ALLOC 20
1393 /* Roughly the maximum number of failure points on the stack. Would be
1394 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1395 This is a variable only so users of regex can assign to it; we never
1396 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1397 before using it, so it should probably be a byte-count instead. */
1398 # if defined MATCH_MAY_ALLOCATE
1399 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1400 whose default stack limit is 2mb. In order for a larger
1401 value to work reliably, you have to try to make it accord
1402 with the process stack limit. */
1403 size_t re_max_failures
= 40000;
1405 size_t re_max_failures
= 4000;
1408 union fail_stack_elt
1411 /* This should be the biggest `int' that's no bigger than a pointer. */
1415 typedef union fail_stack_elt fail_stack_elt_t
;
1419 fail_stack_elt_t
*stack
;
1421 size_t avail
; /* Offset of next open position. */
1422 size_t frame
; /* Offset of the cur constructed frame. */
1425 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1426 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1429 /* Define macros to initialize and free the failure stack.
1430 Do `return -2' if the alloc fails. */
1432 #ifdef MATCH_MAY_ALLOCATE
1433 # define INIT_FAIL_STACK() \
1435 fail_stack.stack = (fail_stack_elt_t *) \
1436 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1437 * sizeof (fail_stack_elt_t)); \
1439 if (fail_stack.stack == NULL) \
1442 fail_stack.size = INIT_FAILURE_ALLOC; \
1443 fail_stack.avail = 0; \
1444 fail_stack.frame = 0; \
1447 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1449 # define INIT_FAIL_STACK() \
1451 fail_stack.avail = 0; \
1452 fail_stack.frame = 0; \
1455 # define RESET_FAIL_STACK() ((void)0)
1459 /* Double the size of FAIL_STACK, up to a limit
1460 which allows approximately `re_max_failures' items.
1462 Return 1 if succeeds, and 0 if either ran out of memory
1463 allocating space for it or it was already too large.
1465 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1467 /* Factor to increase the failure stack size by
1468 when we increase it.
1469 This used to be 2, but 2 was too wasteful
1470 because the old discarded stacks added up to as much space
1471 were as ultimate, maximum-size stack. */
1472 #define FAIL_STACK_GROWTH_FACTOR 4
1474 #define GROW_FAIL_STACK(fail_stack) \
1475 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1476 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1478 : ((fail_stack).stack \
1479 = (fail_stack_elt_t *) \
1480 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1481 (fail_stack).size * sizeof (fail_stack_elt_t), \
1482 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1483 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1484 * FAIL_STACK_GROWTH_FACTOR))), \
1486 (fail_stack).stack == NULL \
1488 : ((fail_stack).size \
1489 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1490 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1491 * FAIL_STACK_GROWTH_FACTOR)) \
1492 / sizeof (fail_stack_elt_t)), \
1496 /* Push a pointer value onto the failure stack.
1497 Assumes the variable `fail_stack'. Probably should only
1498 be called from within `PUSH_FAILURE_POINT'. */
1499 #define PUSH_FAILURE_POINTER(item) \
1500 fail_stack.stack[fail_stack.avail++].pointer = (item)
1502 /* This pushes an integer-valued item onto the failure stack.
1503 Assumes the variable `fail_stack'. Probably should only
1504 be called from within `PUSH_FAILURE_POINT'. */
1505 #define PUSH_FAILURE_INT(item) \
1506 fail_stack.stack[fail_stack.avail++].integer = (item)
1508 /* Push a fail_stack_elt_t value onto the failure stack.
1509 Assumes the variable `fail_stack'. Probably should only
1510 be called from within `PUSH_FAILURE_POINT'. */
1511 #define PUSH_FAILURE_ELT(item) \
1512 fail_stack.stack[fail_stack.avail++] = (item)
1514 /* These three POP... operations complement the three PUSH... operations.
1515 All assume that `fail_stack' is nonempty. */
1516 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1517 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1518 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1520 /* Individual items aside from the registers. */
1521 #define NUM_NONREG_ITEMS 3
1523 /* Used to examine the stack (to detect infinite loops). */
1524 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1525 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1526 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1527 #define TOP_FAILURE_HANDLE() fail_stack.frame
1530 #define ENSURE_FAIL_STACK(space) \
1531 while (REMAINING_AVAIL_SLOTS <= space) { \
1532 if (!GROW_FAIL_STACK (fail_stack)) \
1534 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1535 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1538 /* Push register NUM onto the stack. */
1539 #define PUSH_FAILURE_REG(num) \
1541 char *destination; \
1542 ENSURE_FAIL_STACK(3); \
1543 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1544 num, regstart[num], regend[num]); \
1545 PUSH_FAILURE_POINTER (regstart[num]); \
1546 PUSH_FAILURE_POINTER (regend[num]); \
1547 PUSH_FAILURE_INT (num); \
1550 /* Change the counter's value to VAL, but make sure that it will
1551 be reset when backtracking. */
1552 #define PUSH_NUMBER(ptr,val) \
1554 char *destination; \
1556 ENSURE_FAIL_STACK(3); \
1557 EXTRACT_NUMBER (c, ptr); \
1558 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1559 PUSH_FAILURE_INT (c); \
1560 PUSH_FAILURE_POINTER (ptr); \
1561 PUSH_FAILURE_INT (-1); \
1562 STORE_NUMBER (ptr, val); \
1565 /* Pop a saved register off the stack. */
1566 #define POP_FAILURE_REG_OR_COUNT() \
1568 int reg = POP_FAILURE_INT (); \
1571 /* It's a counter. */ \
1572 /* Here, we discard `const', making re_match non-reentrant. */ \
1573 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1574 reg = POP_FAILURE_INT (); \
1575 STORE_NUMBER (ptr, reg); \
1576 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1580 regend[reg] = POP_FAILURE_POINTER (); \
1581 regstart[reg] = POP_FAILURE_POINTER (); \
1582 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1583 reg, regstart[reg], regend[reg]); \
1587 /* Check that we are not stuck in an infinite loop. */
1588 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1590 int failure = TOP_FAILURE_HANDLE (); \
1591 /* Check for infinite matching loops */ \
1592 while (failure > 0 \
1593 && (FAILURE_STR (failure) == string_place \
1594 || FAILURE_STR (failure) == NULL)) \
1596 assert (FAILURE_PAT (failure) >= bufp->buffer \
1597 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1598 if (FAILURE_PAT (failure) == pat_cur) \
1603 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1604 failure = NEXT_FAILURE_HANDLE(failure); \
1606 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1609 /* Push the information about the state we will need
1610 if we ever fail back to it.
1612 Requires variables fail_stack, regstart, regend and
1613 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1616 Does `return FAILURE_CODE' if runs out of memory. */
1618 #define PUSH_FAILURE_POINT(pattern, string_place) \
1620 char *destination; \
1621 /* Must be int, so when we don't save any registers, the arithmetic \
1622 of 0 + -1 isn't done as unsigned. */ \
1624 DEBUG_STATEMENT (nfailure_points_pushed++); \
1625 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1626 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1627 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1629 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1631 DEBUG_PRINT1 ("\n"); \
1633 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1634 PUSH_FAILURE_INT (fail_stack.frame); \
1636 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1637 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1638 DEBUG_PRINT1 ("'\n"); \
1639 PUSH_FAILURE_POINTER (string_place); \
1641 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1642 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1643 PUSH_FAILURE_POINTER (pattern); \
1645 /* Close the frame by moving the frame pointer past it. */ \
1646 fail_stack.frame = fail_stack.avail; \
1649 /* Estimate the size of data pushed by a typical failure stack entry.
1650 An estimate is all we need, because all we use this for
1651 is to choose a limit for how big to make the failure stack. */
1652 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1653 #define TYPICAL_FAILURE_SIZE 20
1655 /* How many items can still be added to the stack without overflowing it. */
1656 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1659 /* Pops what PUSH_FAIL_STACK pushes.
1661 We restore into the parameters, all of which should be lvalues:
1662 STR -- the saved data position.
1663 PAT -- the saved pattern position.
1664 REGSTART, REGEND -- arrays of string positions.
1666 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1667 `pend', `string1', `size1', `string2', and `size2'. */
1669 #define POP_FAILURE_POINT(str, pat) \
1671 assert (!FAIL_STACK_EMPTY ()); \
1673 /* Remove failure points and point to how many regs pushed. */ \
1674 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1675 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1676 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1678 /* Pop the saved registers. */ \
1679 while (fail_stack.frame < fail_stack.avail) \
1680 POP_FAILURE_REG_OR_COUNT (); \
1682 pat = POP_FAILURE_POINTER (); \
1683 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1684 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1686 /* If the saved string location is NULL, it came from an \
1687 on_failure_keep_string_jump opcode, and we want to throw away the \
1688 saved NULL, thus retaining our current position in the string. */ \
1689 str = POP_FAILURE_POINTER (); \
1690 DEBUG_PRINT2 (" Popping string %p: `", str); \
1691 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1692 DEBUG_PRINT1 ("'\n"); \
1694 fail_stack.frame = POP_FAILURE_INT (); \
1695 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1697 assert (fail_stack.avail >= 0); \
1698 assert (fail_stack.frame <= fail_stack.avail); \
1700 DEBUG_STATEMENT (nfailure_points_popped++); \
1701 } while (0) /* POP_FAILURE_POINT */
1705 /* Registers are set to a sentinel when they haven't yet matched. */
1706 #define REG_UNSET(e) ((e) == NULL)
1708 /* Subroutine declarations and macros for regex_compile. */
1710 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1711 reg_syntax_t syntax
,
1712 struct re_pattern_buffer
*bufp
));
1713 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1714 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1715 int arg1
, int arg2
));
1716 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1717 int arg
, unsigned char *end
));
1718 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1719 int arg1
, int arg2
, unsigned char *end
));
1720 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1722 reg_syntax_t syntax
));
1723 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1725 reg_syntax_t syntax
));
1726 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1727 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1728 char *fastmap
, const int multibyte
));
1730 /* Fetch the next character in the uncompiled pattern, with no
1732 #define PATFETCH(c) \
1735 if (p == pend) return REG_EEND; \
1736 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1741 /* If `translate' is non-null, return translate[D], else just D. We
1742 cast the subscript to translate because some data is declared as
1743 `char *', to avoid warnings when a string constant is passed. But
1744 when we use a character as a subscript we must make it unsigned. */
1746 # define TRANSLATE(d) \
1747 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1751 /* Macros for outputting the compiled pattern into `buffer'. */
1753 /* If the buffer isn't allocated when it comes in, use this. */
1754 #define INIT_BUF_SIZE 32
1756 /* Make sure we have at least N more bytes of space in buffer. */
1757 #define GET_BUFFER_SPACE(n) \
1758 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1761 /* Make sure we have one more byte of buffer space and then add C to it. */
1762 #define BUF_PUSH(c) \
1764 GET_BUFFER_SPACE (1); \
1765 *b++ = (unsigned char) (c); \
1769 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1770 #define BUF_PUSH_2(c1, c2) \
1772 GET_BUFFER_SPACE (2); \
1773 *b++ = (unsigned char) (c1); \
1774 *b++ = (unsigned char) (c2); \
1778 /* As with BUF_PUSH_2, except for three bytes. */
1779 #define BUF_PUSH_3(c1, c2, c3) \
1781 GET_BUFFER_SPACE (3); \
1782 *b++ = (unsigned char) (c1); \
1783 *b++ = (unsigned char) (c2); \
1784 *b++ = (unsigned char) (c3); \
1788 /* Store a jump with opcode OP at LOC to location TO. We store a
1789 relative address offset by the three bytes the jump itself occupies. */
1790 #define STORE_JUMP(op, loc, to) \
1791 store_op1 (op, loc, (to) - (loc) - 3)
1793 /* Likewise, for a two-argument jump. */
1794 #define STORE_JUMP2(op, loc, to, arg) \
1795 store_op2 (op, loc, (to) - (loc) - 3, arg)
1797 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1798 #define INSERT_JUMP(op, loc, to) \
1799 insert_op1 (op, loc, (to) - (loc) - 3, b)
1801 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1802 #define INSERT_JUMP2(op, loc, to, arg) \
1803 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1806 /* This is not an arbitrary limit: the arguments which represent offsets
1807 into the pattern are two bytes long. So if 2^15 bytes turns out to
1808 be too small, many things would have to change. */
1809 # define MAX_BUF_SIZE (1L << 15)
1811 #if 0 /* This is when we thought it could be 2^16 bytes. */
1812 /* Any other compiler which, like MSC, has allocation limit below 2^16
1813 bytes will have to use approach similar to what was done below for
1814 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1815 reallocating to 0 bytes. Such thing is not going to work too well.
1816 You have been warned!! */
1817 #if defined _MSC_VER && !defined WIN32
1818 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1819 # define MAX_BUF_SIZE 65500L
1821 # define MAX_BUF_SIZE (1L << 16)
1825 /* Extend the buffer by twice its current size via realloc and
1826 reset the pointers that pointed into the old block to point to the
1827 correct places in the new one. If extending the buffer results in it
1828 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1829 #if __BOUNDED_POINTERS__
1830 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1831 # define MOVE_BUFFER_POINTER(P) \
1832 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1833 SET_HIGH_BOUND (P), \
1834 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1835 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1838 SET_HIGH_BOUND (b); \
1839 SET_HIGH_BOUND (begalt); \
1840 if (fixup_alt_jump) \
1841 SET_HIGH_BOUND (fixup_alt_jump); \
1843 SET_HIGH_BOUND (laststart); \
1844 if (pending_exact) \
1845 SET_HIGH_BOUND (pending_exact); \
1848 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1849 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1851 #define EXTEND_BUFFER() \
1853 unsigned char *old_buffer = bufp->buffer; \
1854 if (bufp->allocated == MAX_BUF_SIZE) \
1856 bufp->allocated <<= 1; \
1857 if (bufp->allocated > MAX_BUF_SIZE) \
1858 bufp->allocated = MAX_BUF_SIZE; \
1859 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1860 if (bufp->buffer == NULL) \
1861 return REG_ESPACE; \
1862 /* If the buffer moved, move all the pointers into it. */ \
1863 if (old_buffer != bufp->buffer) \
1865 unsigned char *new_buffer = bufp->buffer; \
1866 MOVE_BUFFER_POINTER (b); \
1867 MOVE_BUFFER_POINTER (begalt); \
1868 if (fixup_alt_jump) \
1869 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1871 MOVE_BUFFER_POINTER (laststart); \
1872 if (pending_exact) \
1873 MOVE_BUFFER_POINTER (pending_exact); \
1875 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1879 /* Since we have one byte reserved for the register number argument to
1880 {start,stop}_memory, the maximum number of groups we can report
1881 things about is what fits in that byte. */
1882 #define MAX_REGNUM 255
1884 /* But patterns can have more than `MAX_REGNUM' registers. We just
1885 ignore the excess. */
1886 typedef int regnum_t
;
1889 /* Macros for the compile stack. */
1891 /* Since offsets can go either forwards or backwards, this type needs to
1892 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1893 /* int may be not enough when sizeof(int) == 2. */
1894 typedef long pattern_offset_t
;
1898 pattern_offset_t begalt_offset
;
1899 pattern_offset_t fixup_alt_jump
;
1900 pattern_offset_t laststart_offset
;
1902 } compile_stack_elt_t
;
1907 compile_stack_elt_t
*stack
;
1909 unsigned avail
; /* Offset of next open position. */
1910 } compile_stack_type
;
1913 #define INIT_COMPILE_STACK_SIZE 32
1915 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1916 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1918 /* The next available element. */
1919 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1921 /* Explicit quit checking is only used on NTemacs and whenever we
1922 use polling to process input events. */
1923 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1924 extern int immediate_quit
;
1925 # define IMMEDIATE_QUIT_CHECK \
1927 if (immediate_quit) QUIT; \
1930 # define IMMEDIATE_QUIT_CHECK ((void)0)
1933 /* Structure to manage work area for range table. */
1934 struct range_table_work_area
1936 int *table
; /* actual work area. */
1937 int allocated
; /* allocated size for work area in bytes. */
1938 int used
; /* actually used size in words. */
1939 int bits
; /* flag to record character classes */
1942 /* Make sure that WORK_AREA can hold more N multibyte characters.
1943 This is used only in set_image_of_range and set_image_of_range_1.
1944 It expects WORK_AREA to be a pointer.
1945 If it can't get the space, it returns from the surrounding function. */
1947 #define EXTEND_RANGE_TABLE(work_area, n) \
1949 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1951 extend_range_table_work_area (&work_area); \
1952 if ((work_area).table == 0) \
1953 return (REG_ESPACE); \
1957 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1958 (work_area).bits |= (bit)
1960 /* Bits used to implement the multibyte-part of the various character classes
1961 such as [:alnum:] in a charset's range table. */
1962 #define BIT_WORD 0x1
1963 #define BIT_LOWER 0x2
1964 #define BIT_PUNCT 0x4
1965 #define BIT_SPACE 0x8
1966 #define BIT_UPPER 0x10
1967 #define BIT_MULTIBYTE 0x20
1969 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1970 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1972 EXTEND_RANGE_TABLE ((work_area), 2); \
1973 (work_area).table[(work_area).used++] = (range_start); \
1974 (work_area).table[(work_area).used++] = (range_end); \
1977 /* Free allocated memory for WORK_AREA. */
1978 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1980 if ((work_area).table) \
1981 free ((work_area).table); \
1984 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1985 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1986 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1987 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1990 /* Set the bit for character C in a list. */
1991 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1996 /* Store characters in the range FROM to TO in the bitmap at B (for
1997 ASCII and unibyte characters) and WORK_AREA (for multibyte
1998 characters) while translating them and paying attention to the
1999 continuity of translated characters.
2001 Implementation note: It is better to implement these fairly big
2002 macros by a function, but it's not that easy because macros called
2003 in this macro assume various local variables already declared. */
2005 /* Both FROM and TO are ASCII characters. */
2007 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2011 for (C0 = (FROM); C0 <= (TO); C0++) \
2013 C1 = TRANSLATE (C0); \
2014 if (! ASCII_CHAR_P (C1)) \
2016 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2017 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2020 SET_LIST_BIT (C1); \
2025 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2027 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2029 int C0, C1, C2, I; \
2030 int USED = RANGE_TABLE_WORK_USED (work_area); \
2032 for (C0 = (FROM); C0 <= (TO); C0++) \
2034 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2035 if (CHAR_BYTE8_P (C1)) \
2036 SET_LIST_BIT (C0); \
2039 C2 = TRANSLATE (C1); \
2041 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2043 SET_LIST_BIT (C1); \
2044 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2046 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2047 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2049 if (C2 >= from - 1 && C2 <= to + 1) \
2051 if (C2 == from - 1) \
2052 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2053 else if (C2 == to + 1) \
2054 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2059 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2065 /* Both FROM and TO are multibyte characters. */
2067 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2069 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2071 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2072 for (C0 = (FROM); C0 <= (TO); C0++) \
2074 C1 = TRANSLATE (C0); \
2075 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2076 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2077 SET_LIST_BIT (C2); \
2078 if (C1 >= (FROM) && C1 <= (TO)) \
2080 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2082 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2083 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2085 if (C1 >= from - 1 && C1 <= to + 1) \
2087 if (C1 == from - 1) \
2088 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2089 else if (C1 == to + 1) \
2090 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2095 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2101 /* Get the next unsigned number in the uncompiled pattern. */
2102 #define GET_UNSIGNED_NUMBER(num) \
2105 FREE_STACK_RETURN (REG_EBRACE); \
2109 while ('0' <= c && c <= '9') \
2115 num = num * 10 + c - '0'; \
2116 if (num / 10 != prev) \
2117 FREE_STACK_RETURN (REG_BADBR); \
2119 FREE_STACK_RETURN (REG_EBRACE); \
2125 #if ! WIDE_CHAR_SUPPORT
2127 /* Map a string to the char class it names (if any). */
2129 re_wctype (const re_char
*str
)
2131 const char *string
= str
;
2132 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2133 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2134 else if (STREQ (string
, "word")) return RECC_WORD
;
2135 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2136 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2137 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2138 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2139 else if (STREQ (string
, "print")) return RECC_PRINT
;
2140 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2141 else if (STREQ (string
, "space")) return RECC_SPACE
;
2142 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2143 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2144 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2145 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2146 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2147 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2148 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2152 /* True if CH is in the char class CC. */
2154 re_iswctype (int ch
, re_wctype_t cc
)
2158 case RECC_ALNUM
: return ISALNUM (ch
);
2159 case RECC_ALPHA
: return ISALPHA (ch
);
2160 case RECC_BLANK
: return ISBLANK (ch
);
2161 case RECC_CNTRL
: return ISCNTRL (ch
);
2162 case RECC_DIGIT
: return ISDIGIT (ch
);
2163 case RECC_GRAPH
: return ISGRAPH (ch
);
2164 case RECC_LOWER
: return ISLOWER (ch
);
2165 case RECC_PRINT
: return ISPRINT (ch
);
2166 case RECC_PUNCT
: return ISPUNCT (ch
);
2167 case RECC_SPACE
: return ISSPACE (ch
);
2168 case RECC_UPPER
: return ISUPPER (ch
);
2169 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2170 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2171 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2172 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2173 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2174 case RECC_WORD
: return ISWORD (ch
);
2175 case RECC_ERROR
: return false;
2181 /* Return a bit-pattern to use in the range-table bits to match multibyte
2182 chars of class CC. */
2184 re_wctype_to_bit (re_wctype_t cc
)
2188 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2189 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2190 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2191 case RECC_LOWER
: return BIT_LOWER
;
2192 case RECC_UPPER
: return BIT_UPPER
;
2193 case RECC_PUNCT
: return BIT_PUNCT
;
2194 case RECC_SPACE
: return BIT_SPACE
;
2195 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2196 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2203 /* Filling in the work area of a range. */
2205 /* Actually extend the space in WORK_AREA. */
2208 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2210 work_area
->allocated
+= 16 * sizeof (int);
2211 if (work_area
->table
)
2213 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2216 = (int *) malloc (work_area
->allocated
);
2222 /* Carefully find the ranges of codes that are equivalent
2223 under case conversion to the range start..end when passed through
2224 TRANSLATE. Handle the case where non-letters can come in between
2225 two upper-case letters (which happens in Latin-1).
2226 Also handle the case of groups of more than 2 case-equivalent chars.
2228 The basic method is to look at consecutive characters and see
2229 if they can form a run that can be handled as one.
2231 Returns -1 if successful, REG_ESPACE if ran out of space. */
2234 set_image_of_range_1 (work_area
, start
, end
, translate
)
2235 RE_TRANSLATE_TYPE translate
;
2236 struct range_table_work_area
*work_area
;
2237 re_wchar_t start
, end
;
2239 /* `one_case' indicates a character, or a run of characters,
2240 each of which is an isolate (no case-equivalents).
2241 This includes all ASCII non-letters.
2243 `two_case' indicates a character, or a run of characters,
2244 each of which has two case-equivalent forms.
2245 This includes all ASCII letters.
2247 `strange' indicates a character that has more than one
2250 enum case_type
{one_case
, two_case
, strange
};
2252 /* Describe the run that is in progress,
2253 which the next character can try to extend.
2254 If run_type is strange, that means there really is no run.
2255 If run_type is one_case, then run_start...run_end is the run.
2256 If run_type is two_case, then the run is run_start...run_end,
2257 and the case-equivalents end at run_eqv_end. */
2259 enum case_type run_type
= strange
;
2260 int run_start
, run_end
, run_eqv_end
;
2262 Lisp_Object eqv_table
;
2264 if (!RE_TRANSLATE_P (translate
))
2266 EXTEND_RANGE_TABLE (work_area
, 2);
2267 work_area
->table
[work_area
->used
++] = (start
);
2268 work_area
->table
[work_area
->used
++] = (end
);
2272 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2274 for (; start
<= end
; start
++)
2276 enum case_type this_type
;
2277 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2278 int minchar
, maxchar
;
2280 /* Classify this character */
2282 this_type
= one_case
;
2283 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2284 this_type
= two_case
;
2286 this_type
= strange
;
2289 minchar
= start
, maxchar
= eqv
;
2291 minchar
= eqv
, maxchar
= start
;
2293 /* Can this character extend the run in progress? */
2294 if (this_type
== strange
|| this_type
!= run_type
2295 || !(minchar
== run_end
+ 1
2296 && (run_type
== two_case
2297 ? maxchar
== run_eqv_end
+ 1 : 1)))
2300 Record each of its equivalent ranges. */
2301 if (run_type
== one_case
)
2303 EXTEND_RANGE_TABLE (work_area
, 2);
2304 work_area
->table
[work_area
->used
++] = run_start
;
2305 work_area
->table
[work_area
->used
++] = run_end
;
2307 else if (run_type
== two_case
)
2309 EXTEND_RANGE_TABLE (work_area
, 4);
2310 work_area
->table
[work_area
->used
++] = run_start
;
2311 work_area
->table
[work_area
->used
++] = run_end
;
2312 work_area
->table
[work_area
->used
++]
2313 = RE_TRANSLATE (eqv_table
, run_start
);
2314 work_area
->table
[work_area
->used
++]
2315 = RE_TRANSLATE (eqv_table
, run_end
);
2320 if (this_type
== strange
)
2322 /* For a strange character, add each of its equivalents, one
2323 by one. Don't start a range. */
2326 EXTEND_RANGE_TABLE (work_area
, 2);
2327 work_area
->table
[work_area
->used
++] = eqv
;
2328 work_area
->table
[work_area
->used
++] = eqv
;
2329 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2331 while (eqv
!= start
);
2334 /* Add this char to the run, or start a new run. */
2335 else if (run_type
== strange
)
2337 /* Initialize a new range. */
2338 run_type
= this_type
;
2341 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2345 /* Extend a running range. */
2347 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2351 /* If a run is still in progress at the end, finish it now
2352 by recording its equivalent ranges. */
2353 if (run_type
== one_case
)
2355 EXTEND_RANGE_TABLE (work_area
, 2);
2356 work_area
->table
[work_area
->used
++] = run_start
;
2357 work_area
->table
[work_area
->used
++] = run_end
;
2359 else if (run_type
== two_case
)
2361 EXTEND_RANGE_TABLE (work_area
, 4);
2362 work_area
->table
[work_area
->used
++] = run_start
;
2363 work_area
->table
[work_area
->used
++] = run_end
;
2364 work_area
->table
[work_area
->used
++]
2365 = RE_TRANSLATE (eqv_table
, run_start
);
2366 work_area
->table
[work_area
->used
++]
2367 = RE_TRANSLATE (eqv_table
, run_end
);
2375 /* Record the image of the range start..end when passed through
2376 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2377 and is not even necessarily contiguous.
2378 Normally we approximate it with the smallest contiguous range that contains
2379 all the chars we need. However, for Latin-1 we go to extra effort
2382 This function is not called for ASCII ranges.
2384 Returns -1 if successful, REG_ESPACE if ran out of space. */
2387 set_image_of_range (work_area
, start
, end
, translate
)
2388 RE_TRANSLATE_TYPE translate
;
2389 struct range_table_work_area
*work_area
;
2390 re_wchar_t start
, end
;
2392 re_wchar_t cmin
, cmax
;
2395 /* For Latin-1 ranges, use set_image_of_range_1
2396 to get proper handling of ranges that include letters and nonletters.
2397 For a range that includes the whole of Latin-1, this is not necessary.
2398 For other character sets, we don't bother to get this right. */
2399 if (RE_TRANSLATE_P (translate
) && start
< 04400
2400 && !(start
< 04200 && end
>= 04377))
2407 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2417 EXTEND_RANGE_TABLE (work_area
, 2);
2418 work_area
->table
[work_area
->used
++] = (start
);
2419 work_area
->table
[work_area
->used
++] = (end
);
2421 cmin
= -1, cmax
= -1;
2423 if (RE_TRANSLATE_P (translate
))
2427 for (ch
= start
; ch
<= end
; ch
++)
2429 re_wchar_t c
= TRANSLATE (ch
);
2430 if (! (start
<= c
&& c
<= end
))
2436 cmin
= MIN (cmin
, c
);
2437 cmax
= MAX (cmax
, c
);
2444 EXTEND_RANGE_TABLE (work_area
, 2);
2445 work_area
->table
[work_area
->used
++] = (cmin
);
2446 work_area
->table
[work_area
->used
++] = (cmax
);
2454 #ifndef MATCH_MAY_ALLOCATE
2456 /* If we cannot allocate large objects within re_match_2_internal,
2457 we make the fail stack and register vectors global.
2458 The fail stack, we grow to the maximum size when a regexp
2460 The register vectors, we adjust in size each time we
2461 compile a regexp, according to the number of registers it needs. */
2463 static fail_stack_type fail_stack
;
2465 /* Size with which the following vectors are currently allocated.
2466 That is so we can make them bigger as needed,
2467 but never make them smaller. */
2468 static int regs_allocated_size
;
2470 static re_char
** regstart
, ** regend
;
2471 static re_char
**best_regstart
, **best_regend
;
2473 /* Make the register vectors big enough for NUM_REGS registers,
2474 but don't make them smaller. */
2477 regex_grow_registers (num_regs
)
2480 if (num_regs
> regs_allocated_size
)
2482 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2483 RETALLOC_IF (regend
, num_regs
, re_char
*);
2484 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2485 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2487 regs_allocated_size
= num_regs
;
2491 #endif /* not MATCH_MAY_ALLOCATE */
2493 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2497 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2498 Returns one of error codes defined in `regex.h', or zero for success.
2500 Assumes the `allocated' (and perhaps `buffer') and `translate'
2501 fields are set in BUFP on entry.
2503 If it succeeds, results are put in BUFP (if it returns an error, the
2504 contents of BUFP are undefined):
2505 `buffer' is the compiled pattern;
2506 `syntax' is set to SYNTAX;
2507 `used' is set to the length of the compiled pattern;
2508 `fastmap_accurate' is zero;
2509 `re_nsub' is the number of subexpressions in PATTERN;
2510 `not_bol' and `not_eol' are zero;
2512 The `fastmap' field is neither examined nor set. */
2514 /* Insert the `jump' from the end of last alternative to "here".
2515 The space for the jump has already been allocated. */
2516 #define FIXUP_ALT_JUMP() \
2518 if (fixup_alt_jump) \
2519 STORE_JUMP (jump, fixup_alt_jump, b); \
2523 /* Return, freeing storage we allocated. */
2524 #define FREE_STACK_RETURN(value) \
2526 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2527 free (compile_stack.stack); \
2531 static reg_errcode_t
2532 regex_compile (const re_char
*pattern
, size_t size
, reg_syntax_t syntax
, struct re_pattern_buffer
*bufp
)
2534 /* We fetch characters from PATTERN here. */
2535 register re_wchar_t c
, c1
;
2537 /* A random temporary spot in PATTERN. */
2540 /* Points to the end of the buffer, where we should append. */
2541 register unsigned char *b
;
2543 /* Keeps track of unclosed groups. */
2544 compile_stack_type compile_stack
;
2546 /* Points to the current (ending) position in the pattern. */
2548 /* `const' makes AIX compiler fail. */
2549 unsigned char *p
= pattern
;
2551 re_char
*p
= pattern
;
2553 re_char
*pend
= pattern
+ size
;
2555 /* How to translate the characters in the pattern. */
2556 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2558 /* Address of the count-byte of the most recently inserted `exactn'
2559 command. This makes it possible to tell if a new exact-match
2560 character can be added to that command or if the character requires
2561 a new `exactn' command. */
2562 unsigned char *pending_exact
= 0;
2564 /* Address of start of the most recently finished expression.
2565 This tells, e.g., postfix * where to find the start of its
2566 operand. Reset at the beginning of groups and alternatives. */
2567 unsigned char *laststart
= 0;
2569 /* Address of beginning of regexp, or inside of last group. */
2570 unsigned char *begalt
;
2572 /* Place in the uncompiled pattern (i.e., the {) to
2573 which to go back if the interval is invalid. */
2574 re_char
*beg_interval
;
2576 /* Address of the place where a forward jump should go to the end of
2577 the containing expression. Each alternative of an `or' -- except the
2578 last -- ends with a forward jump of this sort. */
2579 unsigned char *fixup_alt_jump
= 0;
2581 /* Work area for range table of charset. */
2582 struct range_table_work_area range_table_work
;
2584 /* If the object matched can contain multibyte characters. */
2585 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2587 /* If a target of matching can contain multibyte characters. */
2588 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
2590 /* Nonzero if we have pushed down into a subpattern. */
2591 int in_subpattern
= 0;
2593 /* These hold the values of p, pattern, and pend from the main
2594 pattern when we have pushed into a subpattern. */
2596 re_char
*main_pattern
;
2601 DEBUG_PRINT1 ("\nCompiling pattern: ");
2604 unsigned debug_count
;
2606 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2607 putchar (pattern
[debug_count
]);
2612 /* Initialize the compile stack. */
2613 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2614 if (compile_stack
.stack
== NULL
)
2617 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2618 compile_stack
.avail
= 0;
2620 range_table_work
.table
= 0;
2621 range_table_work
.allocated
= 0;
2623 /* Initialize the pattern buffer. */
2624 bufp
->syntax
= syntax
;
2625 bufp
->fastmap_accurate
= 0;
2626 bufp
->not_bol
= bufp
->not_eol
= 0;
2627 bufp
->used_syntax
= 0;
2629 /* Set `used' to zero, so that if we return an error, the pattern
2630 printer (for debugging) will think there's no pattern. We reset it
2634 /* Always count groups, whether or not bufp->no_sub is set. */
2637 #if !defined emacs && !defined SYNTAX_TABLE
2638 /* Initialize the syntax table. */
2639 init_syntax_once ();
2642 if (bufp
->allocated
== 0)
2645 { /* If zero allocated, but buffer is non-null, try to realloc
2646 enough space. This loses if buffer's address is bogus, but
2647 that is the user's responsibility. */
2648 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2651 { /* Caller did not allocate a buffer. Do it for them. */
2652 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2654 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2656 bufp
->allocated
= INIT_BUF_SIZE
;
2659 begalt
= b
= bufp
->buffer
;
2661 /* Loop through the uncompiled pattern until we're at the end. */
2666 /* If this is the end of an included regexp,
2667 pop back to the main regexp and try again. */
2671 pattern
= main_pattern
;
2676 /* If this is the end of the main regexp, we are done. */
2688 /* If there's no special whitespace regexp, treat
2689 spaces normally. And don't try to do this recursively. */
2690 if (!whitespace_regexp
|| in_subpattern
)
2693 /* Peek past following spaces. */
2700 /* If the spaces are followed by a repetition op,
2701 treat them normally. */
2703 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2704 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2707 /* Replace the spaces with the whitespace regexp. */
2711 main_pattern
= pattern
;
2712 p
= pattern
= whitespace_regexp
;
2713 pend
= p
+ strlen (p
);
2719 if ( /* If at start of pattern, it's an operator. */
2721 /* If context independent, it's an operator. */
2722 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2723 /* Otherwise, depends on what's come before. */
2724 || at_begline_loc_p (pattern
, p
, syntax
))
2725 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2734 if ( /* If at end of pattern, it's an operator. */
2736 /* If context independent, it's an operator. */
2737 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2738 /* Otherwise, depends on what's next. */
2739 || at_endline_loc_p (p
, pend
, syntax
))
2740 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2749 if ((syntax
& RE_BK_PLUS_QM
)
2750 || (syntax
& RE_LIMITED_OPS
))
2754 /* If there is no previous pattern... */
2757 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2758 FREE_STACK_RETURN (REG_BADRPT
);
2759 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2764 /* 1 means zero (many) matches is allowed. */
2765 boolean zero_times_ok
= 0, many_times_ok
= 0;
2768 /* If there is a sequence of repetition chars, collapse it
2769 down to just one (the right one). We can't combine
2770 interval operators with these because of, e.g., `a{2}*',
2771 which should only match an even number of `a's. */
2775 if ((syntax
& RE_FRUGAL
)
2776 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2780 zero_times_ok
|= c
!= '+';
2781 many_times_ok
|= c
!= '?';
2787 || (!(syntax
& RE_BK_PLUS_QM
)
2788 && (*p
== '+' || *p
== '?')))
2790 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2793 FREE_STACK_RETURN (REG_EESCAPE
);
2794 if (p
[1] == '+' || p
[1] == '?')
2795 PATFETCH (c
); /* Gobble up the backslash. */
2801 /* If we get here, we found another repeat character. */
2805 /* Star, etc. applied to an empty pattern is equivalent
2806 to an empty pattern. */
2807 if (!laststart
|| laststart
== b
)
2810 /* Now we know whether or not zero matches is allowed
2811 and also whether or not two or more matches is allowed. */
2816 boolean simple
= skip_one_char (laststart
) == b
;
2817 unsigned int startoffset
= 0;
2819 /* Check if the loop can match the empty string. */
2820 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2821 ? on_failure_jump
: on_failure_jump_loop
;
2822 assert (skip_one_char (laststart
) <= b
);
2824 if (!zero_times_ok
&& simple
)
2825 { /* Since simple * loops can be made faster by using
2826 on_failure_keep_string_jump, we turn simple P+
2827 into PP* if P is simple. */
2828 unsigned char *p1
, *p2
;
2829 startoffset
= b
- laststart
;
2830 GET_BUFFER_SPACE (startoffset
);
2831 p1
= b
; p2
= laststart
;
2837 GET_BUFFER_SPACE (6);
2840 STORE_JUMP (ofj
, b
, b
+ 6);
2842 /* Simple * loops can use on_failure_keep_string_jump
2843 depending on what follows. But since we don't know
2844 that yet, we leave the decision up to
2845 on_failure_jump_smart. */
2846 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2847 laststart
+ startoffset
, b
+ 6);
2849 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2854 /* A simple ? pattern. */
2855 assert (zero_times_ok
);
2856 GET_BUFFER_SPACE (3);
2857 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2861 else /* not greedy */
2862 { /* I wish the greedy and non-greedy cases could be merged. */
2864 GET_BUFFER_SPACE (7); /* We might use less. */
2867 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2869 /* The non-greedy multiple match looks like
2870 a repeat..until: we only need a conditional jump
2871 at the end of the loop. */
2872 if (emptyp
) BUF_PUSH (no_op
);
2873 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2874 : on_failure_jump
, b
, laststart
);
2878 /* The repeat...until naturally matches one or more.
2879 To also match zero times, we need to first jump to
2880 the end of the loop (its conditional jump). */
2881 INSERT_JUMP (jump
, laststart
, b
);
2887 /* non-greedy a?? */
2888 INSERT_JUMP (jump
, laststart
, b
+ 3);
2890 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2907 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2909 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2911 /* Ensure that we have enough space to push a charset: the
2912 opcode, the length count, and the bitset; 34 bytes in all. */
2913 GET_BUFFER_SPACE (34);
2917 /* We test `*p == '^' twice, instead of using an if
2918 statement, so we only need one BUF_PUSH. */
2919 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2923 /* Remember the first position in the bracket expression. */
2926 /* Push the number of bytes in the bitmap. */
2927 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2929 /* Clear the whole map. */
2930 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2932 /* charset_not matches newline according to a syntax bit. */
2933 if ((re_opcode_t
) b
[-2] == charset_not
2934 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2935 SET_LIST_BIT ('\n');
2937 /* Read in characters and ranges, setting map bits. */
2940 boolean escaped_char
= false;
2941 const unsigned char *p2
= p
;
2944 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2946 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2947 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2948 So the translation is done later in a loop. Example:
2949 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2952 /* \ might escape characters inside [...] and [^...]. */
2953 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2955 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2958 escaped_char
= true;
2962 /* Could be the end of the bracket expression. If it's
2963 not (i.e., when the bracket expression is `[]' so
2964 far), the ']' character bit gets set way below. */
2965 if (c
== ']' && p2
!= p1
)
2969 /* See if we're at the beginning of a possible character
2972 if (!escaped_char
&&
2973 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2975 /* Leave room for the null. */
2976 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2977 const unsigned char *class_beg
;
2983 /* If pattern is `[[:'. */
2984 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2989 if ((c
== ':' && *p
== ']') || p
== pend
)
2991 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2994 /* This is in any case an invalid class name. */
2999 /* If isn't a word bracketed by `[:' and `:]':
3000 undo the ending character, the letters, and
3001 leave the leading `:' and `[' (but set bits for
3003 if (c
== ':' && *p
== ']')
3008 cc
= re_wctype (str
);
3011 FREE_STACK_RETURN (REG_ECTYPE
);
3013 /* Throw away the ] at the end of the character
3017 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3020 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
3021 if (re_iswctype (btowc (ch
), cc
))
3024 if (c
< (1 << BYTEWIDTH
))
3028 /* Most character classes in a multibyte match
3029 just set a flag. Exceptions are is_blank,
3030 is_digit, is_cntrl, and is_xdigit, since
3031 they can only match ASCII characters. We
3032 don't need to handle them for multibyte.
3033 They are distinguished by a negative wctype. */
3035 /* Setup the gl_state object to its buffer-defined
3036 value. This hardcodes the buffer-global
3037 syntax-table for ASCII chars, while the other chars
3038 will obey syntax-table properties. It's not ideal,
3039 but it's the way it's been done until now. */
3040 SETUP_BUFFER_SYNTAX_TABLE ();
3042 for (ch
= 0; ch
< 256; ++ch
)
3044 c
= RE_CHAR_TO_MULTIBYTE (ch
);
3045 if (! CHAR_BYTE8_P (c
)
3046 && re_iswctype (c
, cc
))
3052 if (ASCII_CHAR_P (c1
))
3054 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
3058 SET_RANGE_TABLE_WORK_AREA_BIT
3059 (range_table_work
, re_wctype_to_bit (cc
));
3061 /* In most cases the matching rule for char classes
3062 only uses the syntax table for multibyte chars,
3063 so that the content of the syntax-table it is not
3064 hardcoded in the range_table. SPACE and WORD are
3065 the two exceptions. */
3066 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
3067 bufp
->used_syntax
= 1;
3069 /* Repeat the loop. */
3074 /* Go back to right after the "[:". */
3078 /* Because the `:' may starts the range, we
3079 can't simply set bit and repeat the loop.
3080 Instead, just set it to C and handle below. */
3085 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
3088 /* Discard the `-'. */
3091 /* Fetch the character which ends the range. */
3094 if (CHAR_BYTE8_P (c1
)
3095 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
3096 /* Treat the range from a multibyte character to
3097 raw-byte character as empty. */
3102 /* Range from C to C. */
3107 if (syntax
& RE_NO_EMPTY_RANGES
)
3108 FREE_STACK_RETURN (REG_ERANGEX
);
3109 /* Else, repeat the loop. */
3114 /* Set the range into bitmap */
3115 for (; c
<= c1
; c
++)
3118 if (ch
< (1 << BYTEWIDTH
))
3125 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
3127 if (CHAR_BYTE8_P (c1
))
3128 c
= BYTE8_TO_CHAR (128);
3132 if (CHAR_BYTE8_P (c
))
3134 c
= CHAR_TO_BYTE8 (c
);
3135 c1
= CHAR_TO_BYTE8 (c1
);
3136 for (; c
<= c1
; c
++)
3141 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
3145 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
3152 /* Discard any (non)matching list bytes that are all 0 at the
3153 end of the map. Decrease the map-length byte too. */
3154 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3158 /* Build real range table from work area. */
3159 if (RANGE_TABLE_WORK_USED (range_table_work
)
3160 || RANGE_TABLE_WORK_BITS (range_table_work
))
3163 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3165 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3166 bytes for flags, two for COUNT, and three bytes for
3168 GET_BUFFER_SPACE (4 + used
* 3);
3170 /* Indicate the existence of range table. */
3171 laststart
[1] |= 0x80;
3173 /* Store the character class flag bits into the range table.
3174 If not in emacs, these flag bits are always 0. */
3175 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3176 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3178 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3179 for (i
= 0; i
< used
; i
++)
3180 STORE_CHARACTER_AND_INCR
3181 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3188 if (syntax
& RE_NO_BK_PARENS
)
3195 if (syntax
& RE_NO_BK_PARENS
)
3202 if (syntax
& RE_NEWLINE_ALT
)
3209 if (syntax
& RE_NO_BK_VBAR
)
3216 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3217 goto handle_interval
;
3223 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3225 /* Do not translate the character after the \, so that we can
3226 distinguish, e.g., \B from \b, even if we normally would
3227 translate, e.g., B to b. */
3233 if (syntax
& RE_NO_BK_PARENS
)
3234 goto normal_backslash
;
3239 regnum_t regnum
= 0;
3242 /* Look for a special (?...) construct */
3243 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3245 PATFETCH (c
); /* Gobble up the '?'. */
3251 case ':': shy
= 1; break;
3253 /* An explicitly specified regnum must start
3256 FREE_STACK_RETURN (REG_BADPAT
);
3257 case '1': case '2': case '3': case '4':
3258 case '5': case '6': case '7': case '8': case '9':
3259 regnum
= 10*regnum
+ (c
- '0'); break;
3261 /* Only (?:...) is supported right now. */
3262 FREE_STACK_RETURN (REG_BADPAT
);
3269 regnum
= ++bufp
->re_nsub
;
3271 { /* It's actually not shy, but explicitly numbered. */
3273 if (regnum
> bufp
->re_nsub
)
3274 bufp
->re_nsub
= regnum
;
3275 else if (regnum
> bufp
->re_nsub
3276 /* Ideally, we'd want to check that the specified
3277 group can't have matched (i.e. all subgroups
3278 using the same regnum are in other branches of
3279 OR patterns), but we don't currently keep track
3280 of enough info to do that easily. */
3281 || group_in_compile_stack (compile_stack
, regnum
))
3282 FREE_STACK_RETURN (REG_BADPAT
);
3285 /* It's really shy. */
3286 regnum
= - bufp
->re_nsub
;
3288 if (COMPILE_STACK_FULL
)
3290 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3291 compile_stack_elt_t
);
3292 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3294 compile_stack
.size
<<= 1;
3297 /* These are the values to restore when we hit end of this
3298 group. They are all relative offsets, so that if the
3299 whole pattern moves because of realloc, they will still
3301 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3302 COMPILE_STACK_TOP
.fixup_alt_jump
3303 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3304 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3305 COMPILE_STACK_TOP
.regnum
= regnum
;
3307 /* Do not push a start_memory for groups beyond the last one
3308 we can represent in the compiled pattern. */
3309 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3310 BUF_PUSH_2 (start_memory
, regnum
);
3312 compile_stack
.avail
++;
3317 /* If we've reached MAX_REGNUM groups, then this open
3318 won't actually generate any code, so we'll have to
3319 clear pending_exact explicitly. */
3325 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3327 if (COMPILE_STACK_EMPTY
)
3329 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3330 goto normal_backslash
;
3332 FREE_STACK_RETURN (REG_ERPAREN
);
3338 /* See similar code for backslashed left paren above. */
3339 if (COMPILE_STACK_EMPTY
)
3341 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3344 FREE_STACK_RETURN (REG_ERPAREN
);
3347 /* Since we just checked for an empty stack above, this
3348 ``can't happen''. */
3349 assert (compile_stack
.avail
!= 0);
3351 /* We don't just want to restore into `regnum', because
3352 later groups should continue to be numbered higher,
3353 as in `(ab)c(de)' -- the second group is #2. */
3356 compile_stack
.avail
--;
3357 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3359 = COMPILE_STACK_TOP
.fixup_alt_jump
3360 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3362 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3363 regnum
= COMPILE_STACK_TOP
.regnum
;
3364 /* If we've reached MAX_REGNUM groups, then this open
3365 won't actually generate any code, so we'll have to
3366 clear pending_exact explicitly. */
3369 /* We're at the end of the group, so now we know how many
3370 groups were inside this one. */
3371 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3372 BUF_PUSH_2 (stop_memory
, regnum
);
3377 case '|': /* `\|'. */
3378 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3379 goto normal_backslash
;
3381 if (syntax
& RE_LIMITED_OPS
)
3384 /* Insert before the previous alternative a jump which
3385 jumps to this alternative if the former fails. */
3386 GET_BUFFER_SPACE (3);
3387 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3391 /* The alternative before this one has a jump after it
3392 which gets executed if it gets matched. Adjust that
3393 jump so it will jump to this alternative's analogous
3394 jump (put in below, which in turn will jump to the next
3395 (if any) alternative's such jump, etc.). The last such
3396 jump jumps to the correct final destination. A picture:
3402 If we are at `b', then fixup_alt_jump right now points to a
3403 three-byte space after `a'. We'll put in the jump, set
3404 fixup_alt_jump to right after `b', and leave behind three
3405 bytes which we'll fill in when we get to after `c'. */
3409 /* Mark and leave space for a jump after this alternative,
3410 to be filled in later either by next alternative or
3411 when know we're at the end of a series of alternatives. */
3413 GET_BUFFER_SPACE (3);
3422 /* If \{ is a literal. */
3423 if (!(syntax
& RE_INTERVALS
)
3424 /* If we're at `\{' and it's not the open-interval
3426 || (syntax
& RE_NO_BK_BRACES
))
3427 goto normal_backslash
;
3431 /* If got here, then the syntax allows intervals. */
3433 /* At least (most) this many matches must be made. */
3434 int lower_bound
= 0, upper_bound
= -1;
3438 GET_UNSIGNED_NUMBER (lower_bound
);
3441 GET_UNSIGNED_NUMBER (upper_bound
);
3443 /* Interval such as `{1}' => match exactly once. */
3444 upper_bound
= lower_bound
;
3446 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3447 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3448 FREE_STACK_RETURN (REG_BADBR
);
3450 if (!(syntax
& RE_NO_BK_BRACES
))
3453 FREE_STACK_RETURN (REG_BADBR
);
3455 FREE_STACK_RETURN (REG_EESCAPE
);
3460 FREE_STACK_RETURN (REG_BADBR
);
3462 /* We just parsed a valid interval. */
3464 /* If it's invalid to have no preceding re. */
3467 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3468 FREE_STACK_RETURN (REG_BADRPT
);
3469 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3472 goto unfetch_interval
;
3475 if (upper_bound
== 0)
3476 /* If the upper bound is zero, just drop the sub pattern
3479 else if (lower_bound
== 1 && upper_bound
== 1)
3480 /* Just match it once: nothing to do here. */
3483 /* Otherwise, we have a nontrivial interval. When
3484 we're all done, the pattern will look like:
3485 set_number_at <jump count> <upper bound>
3486 set_number_at <succeed_n count> <lower bound>
3487 succeed_n <after jump addr> <succeed_n count>
3489 jump_n <succeed_n addr> <jump count>
3490 (The upper bound and `jump_n' are omitted if
3491 `upper_bound' is 1, though.) */
3493 { /* If the upper bound is > 1, we need to insert
3494 more at the end of the loop. */
3495 unsigned int nbytes
= (upper_bound
< 0 ? 3
3496 : upper_bound
> 1 ? 5 : 0);
3497 unsigned int startoffset
= 0;
3499 GET_BUFFER_SPACE (20); /* We might use less. */
3501 if (lower_bound
== 0)
3503 /* A succeed_n that starts with 0 is really a
3504 a simple on_failure_jump_loop. */
3505 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3511 /* Initialize lower bound of the `succeed_n', even
3512 though it will be set during matching by its
3513 attendant `set_number_at' (inserted next),
3514 because `re_compile_fastmap' needs to know.
3515 Jump to the `jump_n' we might insert below. */
3516 INSERT_JUMP2 (succeed_n
, laststart
,
3521 /* Code to initialize the lower bound. Insert
3522 before the `succeed_n'. The `5' is the last two
3523 bytes of this `set_number_at', plus 3 bytes of
3524 the following `succeed_n'. */
3525 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3530 if (upper_bound
< 0)
3532 /* A negative upper bound stands for infinity,
3533 in which case it degenerates to a plain jump. */
3534 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3537 else if (upper_bound
> 1)
3538 { /* More than one repetition is allowed, so
3539 append a backward jump to the `succeed_n'
3540 that starts this interval.
3542 When we've reached this during matching,
3543 we'll have matched the interval once, so
3544 jump back only `upper_bound - 1' times. */
3545 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3549 /* The location we want to set is the second
3550 parameter of the `jump_n'; that is `b-2' as
3551 an absolute address. `laststart' will be
3552 the `set_number_at' we're about to insert;
3553 `laststart+3' the number to set, the source
3554 for the relative address. But we are
3555 inserting into the middle of the pattern --
3556 so everything is getting moved up by 5.
3557 Conclusion: (b - 2) - (laststart + 3) + 5,
3558 i.e., b - laststart.
3560 We insert this at the beginning of the loop
3561 so that if we fail during matching, we'll
3562 reinitialize the bounds. */
3563 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3564 upper_bound
- 1, b
);
3569 beg_interval
= NULL
;
3574 /* If an invalid interval, match the characters as literals. */
3575 assert (beg_interval
);
3577 beg_interval
= NULL
;
3579 /* normal_char and normal_backslash need `c'. */
3582 if (!(syntax
& RE_NO_BK_BRACES
))
3584 assert (p
> pattern
&& p
[-1] == '\\');
3585 goto normal_backslash
;
3591 /* There is no way to specify the before_dot and after_dot
3592 operators. rms says this is ok. --karl */
3600 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3606 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3612 BUF_PUSH_2 (categoryspec
, c
);
3618 BUF_PUSH_2 (notcategoryspec
, c
);
3624 if (syntax
& RE_NO_GNU_OPS
)
3627 BUF_PUSH_2 (syntaxspec
, Sword
);
3632 if (syntax
& RE_NO_GNU_OPS
)
3635 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3640 if (syntax
& RE_NO_GNU_OPS
)
3646 if (syntax
& RE_NO_GNU_OPS
)
3652 if (syntax
& RE_NO_GNU_OPS
)
3661 FREE_STACK_RETURN (REG_BADPAT
);
3665 if (syntax
& RE_NO_GNU_OPS
)
3667 BUF_PUSH (wordbound
);
3671 if (syntax
& RE_NO_GNU_OPS
)
3673 BUF_PUSH (notwordbound
);
3677 if (syntax
& RE_NO_GNU_OPS
)
3683 if (syntax
& RE_NO_GNU_OPS
)
3688 case '1': case '2': case '3': case '4': case '5':
3689 case '6': case '7': case '8': case '9':
3693 if (syntax
& RE_NO_BK_REFS
)
3694 goto normal_backslash
;
3698 if (reg
> bufp
->re_nsub
|| reg
< 1
3699 /* Can't back reference to a subexp before its end. */
3700 || group_in_compile_stack (compile_stack
, reg
))
3701 FREE_STACK_RETURN (REG_ESUBREG
);
3704 BUF_PUSH_2 (duplicate
, reg
);
3711 if (syntax
& RE_BK_PLUS_QM
)
3714 goto normal_backslash
;
3718 /* You might think it would be useful for \ to mean
3719 not to translate; but if we don't translate it
3720 it will never match anything. */
3727 /* Expects the character in `c'. */
3729 /* If no exactn currently being built. */
3732 /* If last exactn not at current position. */
3733 || pending_exact
+ *pending_exact
+ 1 != b
3735 /* We have only one byte following the exactn for the count. */
3736 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3738 /* If followed by a repetition operator. */
3739 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3740 || ((syntax
& RE_BK_PLUS_QM
)
3741 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3742 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3743 || ((syntax
& RE_INTERVALS
)
3744 && ((syntax
& RE_NO_BK_BRACES
)
3745 ? p
!= pend
&& *p
== '{'
3746 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3748 /* Start building a new exactn. */
3752 BUF_PUSH_2 (exactn
, 0);
3753 pending_exact
= b
- 1;
3756 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3763 len
= CHAR_STRING (c
, b
);
3768 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3769 if (! CHAR_BYTE8_P (c1
))
3771 re_wchar_t c2
= TRANSLATE (c1
);
3773 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3779 (*pending_exact
) += len
;
3784 } /* while p != pend */
3787 /* Through the pattern now. */
3791 if (!COMPILE_STACK_EMPTY
)
3792 FREE_STACK_RETURN (REG_EPAREN
);
3794 /* If we don't want backtracking, force success
3795 the first time we reach the end of the compiled pattern. */
3796 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3799 /* We have succeeded; set the length of the buffer. */
3800 bufp
->used
= b
- bufp
->buffer
;
3805 re_compile_fastmap (bufp
);
3806 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3807 print_compiled_pattern (bufp
);
3812 #ifndef MATCH_MAY_ALLOCATE
3813 /* Initialize the failure stack to the largest possible stack. This
3814 isn't necessary unless we're trying to avoid calling alloca in
3815 the search and match routines. */
3817 int num_regs
= bufp
->re_nsub
+ 1;
3819 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3821 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3823 if (! fail_stack
.stack
)
3825 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3826 * sizeof (fail_stack_elt_t
));
3829 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3831 * sizeof (fail_stack_elt_t
)));
3834 regex_grow_registers (num_regs
);
3836 #endif /* not MATCH_MAY_ALLOCATE */
3838 FREE_STACK_RETURN (REG_NOERROR
);
3839 } /* regex_compile */
3841 /* Subroutines for `regex_compile'. */
3843 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3846 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3848 *loc
= (unsigned char) op
;
3849 STORE_NUMBER (loc
+ 1, arg
);
3853 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3856 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3858 *loc
= (unsigned char) op
;
3859 STORE_NUMBER (loc
+ 1, arg1
);
3860 STORE_NUMBER (loc
+ 3, arg2
);
3864 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3865 for OP followed by two-byte integer parameter ARG. */
3868 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3870 register unsigned char *pfrom
= end
;
3871 register unsigned char *pto
= end
+ 3;
3873 while (pfrom
!= loc
)
3876 store_op1 (op
, loc
, arg
);
3880 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3883 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3885 register unsigned char *pfrom
= end
;
3886 register unsigned char *pto
= end
+ 5;
3888 while (pfrom
!= loc
)
3891 store_op2 (op
, loc
, arg1
, arg2
);
3895 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3896 after an alternative or a begin-subexpression. We assume there is at
3897 least one character before the ^. */
3900 at_begline_loc_p (const re_char
*pattern
, const re_char
*p
, reg_syntax_t syntax
)
3902 re_char
*prev
= p
- 2;
3903 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3906 /* After a subexpression? */
3907 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3908 /* After an alternative? */
3909 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3910 /* After a shy subexpression? */
3911 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3912 && prev
[-1] == '?' && prev
[-2] == '('
3913 && (syntax
& RE_NO_BK_PARENS
3914 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3918 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3919 at least one character after the $, i.e., `P < PEND'. */
3922 at_endline_loc_p (const re_char
*p
, const re_char
*pend
, reg_syntax_t syntax
)
3925 boolean next_backslash
= *next
== '\\';
3926 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3929 /* Before a subexpression? */
3930 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3931 : next_backslash
&& next_next
&& *next_next
== ')')
3932 /* Before an alternative? */
3933 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3934 : next_backslash
&& next_next
&& *next_next
== '|');
3938 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3939 false if it's not. */
3942 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3946 for (this_element
= compile_stack
.avail
- 1;
3949 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3956 If fastmap is non-NULL, go through the pattern and fill fastmap
3957 with all the possible leading chars. If fastmap is NULL, don't
3958 bother filling it up (obviously) and only return whether the
3959 pattern could potentially match the empty string.
3961 Return 1 if p..pend might match the empty string.
3962 Return 0 if p..pend matches at least one char.
3963 Return -1 if fastmap was not updated accurately. */
3966 analyse_first (const re_char
*p
, const re_char
*pend
, char *fastmap
, const int multibyte
)
3971 /* If all elements for base leading-codes in fastmap is set, this
3972 flag is set true. */
3973 boolean match_any_multibyte_characters
= false;
3977 /* The loop below works as follows:
3978 - It has a working-list kept in the PATTERN_STACK and which basically
3979 starts by only containing a pointer to the first operation.
3980 - If the opcode we're looking at is a match against some set of
3981 chars, then we add those chars to the fastmap and go on to the
3982 next work element from the worklist (done via `break').
3983 - If the opcode is a control operator on the other hand, we either
3984 ignore it (if it's meaningless at this point, such as `start_memory')
3985 or execute it (if it's a jump). If the jump has several destinations
3986 (i.e. `on_failure_jump'), then we push the other destination onto the
3988 We guarantee termination by ignoring backward jumps (more or less),
3989 so that `p' is monotonically increasing. More to the point, we
3990 never set `p' (or push) anything `<= p1'. */
3994 /* `p1' is used as a marker of how far back a `on_failure_jump'
3995 can go without being ignored. It is normally equal to `p'
3996 (which prevents any backward `on_failure_jump') except right
3997 after a plain `jump', to allow patterns such as:
4000 10: on_failure_jump 3
4001 as used for the *? operator. */
4004 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4011 /* If the first character has to match a backreference, that means
4012 that the group was empty (since it already matched). Since this
4013 is the only case that interests us here, we can assume that the
4014 backreference must match the empty string. */
4019 /* Following are the cases which match a character. These end
4025 /* If multibyte is nonzero, the first byte of each
4026 character is an ASCII or a leading code. Otherwise,
4027 each byte is a character. Thus, this works in both
4032 /* For the case of matching this unibyte regex
4033 against multibyte, we must set a leading code of
4034 the corresponding multibyte character. */
4035 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
4037 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
4044 /* We could put all the chars except for \n (and maybe \0)
4045 but we don't bother since it is generally not worth it. */
4046 if (!fastmap
) break;
4051 if (!fastmap
) break;
4053 /* Chars beyond end of bitmap are possible matches. */
4054 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
4055 j
< (1 << BYTEWIDTH
); j
++)
4061 if (!fastmap
) break;
4062 not = (re_opcode_t
) *(p
- 1) == charset_not
;
4063 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
4065 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
4069 if (/* Any leading code can possibly start a character
4070 which doesn't match the specified set of characters. */
4073 /* If we can match a character class, we can match any
4074 multibyte characters. */
4075 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
4076 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
4079 if (match_any_multibyte_characters
== false)
4081 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4082 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4084 match_any_multibyte_characters
= true;
4088 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
4089 && match_any_multibyte_characters
== false)
4091 /* Set fastmap[I] to 1 where I is a leading code of each
4092 multibyte characer in the range table. */
4094 unsigned char lc1
, lc2
;
4096 /* Make P points the range table. `+ 2' is to skip flag
4097 bits for a character class. */
4098 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
4100 /* Extract the number of ranges in range table into COUNT. */
4101 EXTRACT_NUMBER_AND_INCR (count
, p
);
4102 for (; count
> 0; count
--, p
+= 3)
4104 /* Extract the start and end of each range. */
4105 EXTRACT_CHARACTER (c
, p
);
4106 lc1
= CHAR_LEADING_CODE (c
);
4108 EXTRACT_CHARACTER (c
, p
);
4109 lc2
= CHAR_LEADING_CODE (c
);
4110 for (j
= lc1
; j
<= lc2
; j
++)
4119 if (!fastmap
) break;
4121 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4123 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4124 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4128 /* This match depends on text properties. These end with
4129 aborting optimizations. */
4133 case notcategoryspec
:
4134 if (!fastmap
) break;
4135 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4137 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
4138 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4141 /* Any leading code can possibly start a character which
4142 has or doesn't has the specified category. */
4143 if (match_any_multibyte_characters
== false)
4145 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4146 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4148 match_any_multibyte_characters
= true;
4152 /* All cases after this match the empty string. These end with
4174 EXTRACT_NUMBER_AND_INCR (j
, p
);
4176 /* Backward jumps can only go back to code that we've already
4177 visited. `re_compile' should make sure this is true. */
4180 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4182 case on_failure_jump
:
4183 case on_failure_keep_string_jump
:
4184 case on_failure_jump_loop
:
4185 case on_failure_jump_nastyloop
:
4186 case on_failure_jump_smart
:
4192 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4193 to jump back to "just after here". */
4196 case on_failure_jump
:
4197 case on_failure_keep_string_jump
:
4198 case on_failure_jump_nastyloop
:
4199 case on_failure_jump_loop
:
4200 case on_failure_jump_smart
:
4201 EXTRACT_NUMBER_AND_INCR (j
, p
);
4203 ; /* Backward jump to be ignored. */
4205 { /* We have to look down both arms.
4206 We first go down the "straight" path so as to minimize
4207 stack usage when going through alternatives. */
4208 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4216 /* This code simply does not properly handle forward jump_n. */
4217 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4219 /* jump_n can either jump or fall through. The (backward) jump
4220 case has already been handled, so we only need to look at the
4221 fallthrough case. */
4225 /* If N == 0, it should be an on_failure_jump_loop instead. */
4226 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4228 /* We only care about one iteration of the loop, so we don't
4229 need to consider the case where this behaves like an
4246 abort (); /* We have listed all the cases. */
4249 /* Getting here means we have found the possible starting
4250 characters for one path of the pattern -- and that the empty
4251 string does not match. We need not follow this path further. */
4255 /* We reached the end without matching anything. */
4258 } /* analyse_first */
4260 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4261 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4262 characters can start a string that matches the pattern. This fastmap
4263 is used by re_search to skip quickly over impossible starting points.
4265 Character codes above (1 << BYTEWIDTH) are not represented in the
4266 fastmap, but the leading codes are represented. Thus, the fastmap
4267 indicates which character sets could start a match.
4269 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4270 area as BUFP->fastmap.
4272 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4275 Returns 0 if we succeed, -2 if an internal error. */
4278 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4280 char *fastmap
= bufp
->fastmap
;
4283 assert (fastmap
&& bufp
->buffer
);
4285 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4286 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4288 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4289 fastmap
, RE_MULTIBYTE_P (bufp
));
4290 bufp
->can_be_null
= (analysis
!= 0);
4292 } /* re_compile_fastmap */
4294 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4295 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4296 this memory for recording register information. STARTS and ENDS
4297 must be allocated using the malloc library routine, and must each
4298 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4300 If NUM_REGS == 0, then subsequent matches should allocate their own
4303 Unless this function is called, the first search or match using
4304 PATTERN_BUFFER will allocate its own register data, without
4305 freeing the old data. */
4308 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4312 bufp
->regs_allocated
= REGS_REALLOCATE
;
4313 regs
->num_regs
= num_regs
;
4314 regs
->start
= starts
;
4319 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4321 regs
->start
= regs
->end
= (regoff_t
*) 0;
4324 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4326 /* Searching routines. */
4328 /* Like re_search_2, below, but only one string is specified, and
4329 doesn't let you say where to stop matching. */
4332 re_search (struct re_pattern_buffer
*bufp
, const char *string
, int size
, int startpos
, int range
, struct re_registers
*regs
)
4334 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4337 WEAK_ALIAS (__re_search
, re_search
)
4339 /* Head address of virtual concatenation of string. */
4340 #define HEAD_ADDR_VSTRING(P) \
4341 (((P) >= size1 ? string2 : string1))
4343 /* End address of virtual concatenation of string. */
4344 #define STOP_ADDR_VSTRING(P) \
4345 (((P) >= size1 ? string2 + size2 : string1 + size1))
4347 /* Address of POS in the concatenation of virtual string. */
4348 #define POS_ADDR_VSTRING(POS) \
4349 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4351 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4352 virtual concatenation of STRING1 and STRING2, starting first at index
4353 STARTPOS, then at STARTPOS + 1, and so on.
4355 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4357 RANGE is how far to scan while trying to match. RANGE = 0 means try
4358 only at STARTPOS; in general, the last start tried is STARTPOS +
4361 In REGS, return the indices of the virtual concatenation of STRING1
4362 and STRING2 that matched the entire BUFP->buffer and its contained
4365 Do not consider matching one past the index STOP in the virtual
4366 concatenation of STRING1 and STRING2.
4368 We return either the position in the strings at which the match was
4369 found, -1 if no match, or -2 if error (such as failure
4373 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, int size1
, const char *str2
, int size2
, int startpos
, int range
, struct re_registers
*regs
, int stop
)
4376 re_char
*string1
= (re_char
*) str1
;
4377 re_char
*string2
= (re_char
*) str2
;
4378 register char *fastmap
= bufp
->fastmap
;
4379 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4380 int total_size
= size1
+ size2
;
4381 int endpos
= startpos
+ range
;
4382 boolean anchored_start
;
4383 /* Nonzero if we are searching multibyte string. */
4384 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4386 /* Check for out-of-range STARTPOS. */
4387 if (startpos
< 0 || startpos
> total_size
)
4390 /* Fix up RANGE if it might eventually take us outside
4391 the virtual concatenation of STRING1 and STRING2.
4392 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4394 range
= 0 - startpos
;
4395 else if (endpos
> total_size
)
4396 range
= total_size
- startpos
;
4398 /* If the search isn't to be a backwards one, don't waste time in a
4399 search for a pattern anchored at beginning of buffer. */
4400 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4409 /* In a forward search for something that starts with \=.
4410 don't keep searching past point. */
4411 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4413 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4419 /* Update the fastmap now if not correct already. */
4420 if (fastmap
&& !bufp
->fastmap_accurate
)
4421 re_compile_fastmap (bufp
);
4423 /* See whether the pattern is anchored. */
4424 anchored_start
= (bufp
->buffer
[0] == begline
);
4427 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4429 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4431 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4435 /* Loop through the string, looking for a place to start matching. */
4438 /* If the pattern is anchored,
4439 skip quickly past places we cannot match.
4440 We don't bother to treat startpos == 0 specially
4441 because that case doesn't repeat. */
4442 if (anchored_start
&& startpos
> 0)
4444 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4445 : string2
[startpos
- size1
- 1])
4450 /* If a fastmap is supplied, skip quickly over characters that
4451 cannot be the start of a match. If the pattern can match the
4452 null string, however, we don't need to skip characters; we want
4453 the first null string. */
4454 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4456 register re_char
*d
;
4457 register re_wchar_t buf_ch
;
4459 d
= POS_ADDR_VSTRING (startpos
);
4461 if (range
> 0) /* Searching forwards. */
4463 register int lim
= 0;
4466 if (startpos
< size1
&& startpos
+ range
>= size1
)
4467 lim
= range
- (size1
- startpos
);
4469 /* Written out as an if-else to avoid testing `translate'
4471 if (RE_TRANSLATE_P (translate
))
4478 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4479 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4480 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4483 range
-= buf_charlen
;
4489 register re_wchar_t ch
, translated
;
4492 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4493 translated
= RE_TRANSLATE (translate
, ch
);
4494 if (translated
!= ch
4495 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4497 if (fastmap
[buf_ch
])
4510 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4511 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4513 range
-= buf_charlen
;
4517 while (range
> lim
&& !fastmap
[*d
])
4523 startpos
+= irange
- range
;
4525 else /* Searching backwards. */
4529 buf_ch
= STRING_CHAR (d
);
4530 buf_ch
= TRANSLATE (buf_ch
);
4531 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4536 register re_wchar_t ch
, translated
;
4539 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4540 translated
= TRANSLATE (ch
);
4541 if (translated
!= ch
4542 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4544 if (! fastmap
[TRANSLATE (buf_ch
)])
4550 /* If can't match the null string, and that's all we have left, fail. */
4551 if (range
>= 0 && startpos
== total_size
&& fastmap
4552 && !bufp
->can_be_null
)
4555 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4556 startpos
, regs
, stop
);
4569 /* Update STARTPOS to the next character boundary. */
4572 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4573 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4574 int len
= BYTES_BY_CHAR_HEAD (*p
);
4592 /* Update STARTPOS to the previous character boundary. */
4595 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4597 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4599 /* Find the head of multibyte form. */
4600 PREV_CHAR_BOUNDARY (p
, phead
);
4601 range
+= p0
- 1 - p
;
4605 startpos
-= p0
- 1 - p
;
4611 WEAK_ALIAS (__re_search_2
, re_search_2
)
4613 /* Declarations and macros for re_match_2. */
4615 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4617 RE_TRANSLATE_TYPE translate
,
4618 const int multibyte
));
4620 /* This converts PTR, a pointer into one of the search strings `string1'
4621 and `string2' into an offset from the beginning of that string. */
4622 #define POINTER_TO_OFFSET(ptr) \
4623 (FIRST_STRING_P (ptr) \
4624 ? ((regoff_t) ((ptr) - string1)) \
4625 : ((regoff_t) ((ptr) - string2 + size1)))
4627 /* Call before fetching a character with *d. This switches over to
4628 string2 if necessary.
4629 Check re_match_2_internal for a discussion of why end_match_2 might
4630 not be within string2 (but be equal to end_match_1 instead). */
4631 #define PREFETCH() \
4634 /* End of string2 => fail. */ \
4635 if (dend == end_match_2) \
4637 /* End of string1 => advance to string2. */ \
4639 dend = end_match_2; \
4642 /* Call before fetching a char with *d if you already checked other limits.
4643 This is meant for use in lookahead operations like wordend, etc..
4644 where we might need to look at parts of the string that might be
4645 outside of the LIMITs (i.e past `stop'). */
4646 #define PREFETCH_NOLIMIT() \
4650 dend = end_match_2; \
4653 /* Test if at very beginning or at very end of the virtual concatenation
4654 of `string1' and `string2'. If only one string, it's `string2'. */
4655 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4656 #define AT_STRINGS_END(d) ((d) == end2)
4659 /* Test if D points to a character which is word-constituent. We have
4660 two special cases to check for: if past the end of string1, look at
4661 the first character in string2; and if before the beginning of
4662 string2, look at the last character in string1. */
4663 #define WORDCHAR_P(d) \
4664 (SYNTAX ((d) == end1 ? *string2 \
4665 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4668 /* Disabled due to a compiler bug -- see comment at case wordbound */
4670 /* The comment at case wordbound is following one, but we don't use
4671 AT_WORD_BOUNDARY anymore to support multibyte form.
4673 The DEC Alpha C compiler 3.x generates incorrect code for the
4674 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4675 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4676 macro and introducing temporary variables works around the bug. */
4679 /* Test if the character before D and the one at D differ with respect
4680 to being word-constituent. */
4681 #define AT_WORD_BOUNDARY(d) \
4682 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4683 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4686 /* Free everything we malloc. */
4687 #ifdef MATCH_MAY_ALLOCATE
4688 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4689 # define FREE_VARIABLES() \
4691 REGEX_FREE_STACK (fail_stack.stack); \
4692 FREE_VAR (regstart); \
4693 FREE_VAR (regend); \
4694 FREE_VAR (best_regstart); \
4695 FREE_VAR (best_regend); \
4698 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4699 #endif /* not MATCH_MAY_ALLOCATE */
4702 /* Optimization routines. */
4704 /* If the operation is a match against one or more chars,
4705 return a pointer to the next operation, else return NULL. */
4707 skip_one_char (const re_char
*p
)
4709 switch (SWITCH_ENUM_CAST (*p
++))
4720 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4723 p
= CHARSET_RANGE_TABLE (p
- 1);
4724 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4725 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4728 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4735 case notcategoryspec
:
4747 /* Jump over non-matching operations. */
4749 skip_noops (const re_char
*p
, const re_char
*pend
)
4754 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4763 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4774 /* Non-zero if "p1 matches something" implies "p2 fails". */
4776 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, const re_char
*p1
, const re_char
*p2
)
4779 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4780 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4782 assert (p1
>= bufp
->buffer
&& p1
< pend
4783 && p2
>= bufp
->buffer
&& p2
<= pend
);
4785 /* Skip over open/close-group commands.
4786 If what follows this loop is a ...+ construct,
4787 look at what begins its body, since we will have to
4788 match at least one of that. */
4789 p2
= skip_noops (p2
, pend
);
4790 /* The same skip can be done for p1, except that this function
4791 is only used in the case where p1 is a simple match operator. */
4792 /* p1 = skip_noops (p1, pend); */
4794 assert (p1
>= bufp
->buffer
&& p1
< pend
4795 && p2
>= bufp
->buffer
&& p2
<= pend
);
4797 op2
= p2
== pend
? succeed
: *p2
;
4799 switch (SWITCH_ENUM_CAST (op2
))
4803 /* If we're at the end of the pattern, we can change. */
4804 if (skip_one_char (p1
))
4806 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4814 register re_wchar_t c
4815 = (re_opcode_t
) *p2
== endline
? '\n'
4816 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4818 if ((re_opcode_t
) *p1
== exactn
)
4820 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4822 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4827 else if ((re_opcode_t
) *p1
== charset
4828 || (re_opcode_t
) *p1
== charset_not
)
4830 int not = (re_opcode_t
) *p1
== charset_not
;
4832 /* Test if C is listed in charset (or charset_not)
4834 if (! multibyte
|| IS_REAL_ASCII (c
))
4836 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4837 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4840 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4841 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4843 /* `not' is equal to 1 if c would match, which means
4844 that we can't change to pop_failure_jump. */
4847 DEBUG_PRINT1 (" No match => fast loop.\n");
4851 else if ((re_opcode_t
) *p1
== anychar
4854 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4862 if ((re_opcode_t
) *p1
== exactn
)
4863 /* Reuse the code above. */
4864 return mutually_exclusive_p (bufp
, p2
, p1
);
4866 /* It is hard to list up all the character in charset
4867 P2 if it includes multibyte character. Give up in
4869 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4871 /* Now, we are sure that P2 has no range table.
4872 So, for the size of bitmap in P2, `p2[1]' is
4873 enough. But P1 may have range table, so the
4874 size of bitmap table of P1 is extracted by
4875 using macro `CHARSET_BITMAP_SIZE'.
4877 In a multibyte case, we know that all the character
4878 listed in P2 is ASCII. In a unibyte case, P1 has only a
4879 bitmap table. So, in both cases, it is enough to test
4880 only the bitmap table of P1. */
4882 if ((re_opcode_t
) *p1
== charset
)
4885 /* We win if the charset inside the loop
4886 has no overlap with the one after the loop. */
4889 && idx
< CHARSET_BITMAP_SIZE (p1
));
4891 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4895 || idx
== CHARSET_BITMAP_SIZE (p1
))
4897 DEBUG_PRINT1 (" No match => fast loop.\n");
4901 else if ((re_opcode_t
) *p1
== charset_not
)
4904 /* We win if the charset_not inside the loop lists
4905 every character listed in the charset after. */
4906 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4907 if (! (p2
[2 + idx
] == 0
4908 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4909 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4914 DEBUG_PRINT1 (" No match => fast loop.\n");
4923 switch (SWITCH_ENUM_CAST (*p1
))
4927 /* Reuse the code above. */
4928 return mutually_exclusive_p (bufp
, p2
, p1
);
4930 /* When we have two charset_not, it's very unlikely that
4931 they don't overlap. The union of the two sets of excluded
4932 chars should cover all possible chars, which, as a matter of
4933 fact, is virtually impossible in multibyte buffers. */
4939 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4941 return ((re_opcode_t
) *p1
== syntaxspec
4942 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4944 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4947 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4949 return ((re_opcode_t
) *p1
== notsyntaxspec
4950 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4952 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4955 return (((re_opcode_t
) *p1
== notsyntaxspec
4956 || (re_opcode_t
) *p1
== syntaxspec
)
4961 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4962 case notcategoryspec
:
4963 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4975 /* Matching routines. */
4977 #ifndef emacs /* Emacs never uses this. */
4978 /* re_match is like re_match_2 except it takes only a single string. */
4981 re_match (bufp
, string
, size
, pos
, regs
)
4982 struct re_pattern_buffer
*bufp
;
4985 struct re_registers
*regs
;
4987 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
4991 WEAK_ALIAS (__re_match
, re_match
)
4992 #endif /* not emacs */
4995 /* In Emacs, this is the string or buffer in which we
4996 are matching. It is used for looking up syntax properties. */
4997 Lisp_Object re_match_object
;
5000 /* re_match_2 matches the compiled pattern in BUFP against the
5001 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5002 and SIZE2, respectively). We start matching at POS, and stop
5005 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5006 store offsets for the substring each group matched in REGS. See the
5007 documentation for exactly how many groups we fill.
5009 We return -1 if no match, -2 if an internal error (such as the
5010 failure stack overflowing). Otherwise, we return the length of the
5011 matched substring. */
5014 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
, int size1
, const char *string2
, int size2
, int pos
, struct re_registers
*regs
, int stop
)
5020 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
5021 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
5022 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
5025 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
5026 (re_char
*) string2
, size2
,
5030 WEAK_ALIAS (__re_match_2
, re_match_2
)
5033 /* This is a separate function so that we can force an alloca cleanup
5036 re_match_2_internal (struct re_pattern_buffer
*bufp
, const re_char
*string1
, int size1
, const re_char
*string2
, int size2
, int pos
, struct re_registers
*regs
, int stop
)
5038 /* General temporaries. */
5043 /* Just past the end of the corresponding string. */
5044 re_char
*end1
, *end2
;
5046 /* Pointers into string1 and string2, just past the last characters in
5047 each to consider matching. */
5048 re_char
*end_match_1
, *end_match_2
;
5050 /* Where we are in the data, and the end of the current string. */
5053 /* Used sometimes to remember where we were before starting matching
5054 an operator so that we can go back in case of failure. This "atomic"
5055 behavior of matching opcodes is indispensable to the correctness
5056 of the on_failure_keep_string_jump optimization. */
5059 /* Where we are in the pattern, and the end of the pattern. */
5060 re_char
*p
= bufp
->buffer
;
5061 re_char
*pend
= p
+ bufp
->used
;
5063 /* We use this to map every character in the string. */
5064 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5066 /* Nonzero if BUFP is setup from a multibyte regex. */
5067 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
5069 /* Nonzero if STRING1/STRING2 are multibyte. */
5070 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
5072 /* Failure point stack. Each place that can handle a failure further
5073 down the line pushes a failure point on this stack. It consists of
5074 regstart, and regend for all registers corresponding to
5075 the subexpressions we're currently inside, plus the number of such
5076 registers, and, finally, two char *'s. The first char * is where
5077 to resume scanning the pattern; the second one is where to resume
5078 scanning the strings. */
5079 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5080 fail_stack_type fail_stack
;
5083 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5086 #if defined REL_ALLOC && defined REGEX_MALLOC
5087 /* This holds the pointer to the failure stack, when
5088 it is allocated relocatably. */
5089 fail_stack_elt_t
*failure_stack_ptr
;
5092 /* We fill all the registers internally, independent of what we
5093 return, for use in backreferences. The number here includes
5094 an element for register zero. */
5095 size_t num_regs
= bufp
->re_nsub
+ 1;
5097 /* Information on the contents of registers. These are pointers into
5098 the input strings; they record just what was matched (on this
5099 attempt) by a subexpression part of the pattern, that is, the
5100 regnum-th regstart pointer points to where in the pattern we began
5101 matching and the regnum-th regend points to right after where we
5102 stopped matching the regnum-th subexpression. (The zeroth register
5103 keeps track of what the whole pattern matches.) */
5104 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5105 re_char
**regstart
, **regend
;
5108 /* The following record the register info as found in the above
5109 variables when we find a match better than any we've seen before.
5110 This happens as we backtrack through the failure points, which in
5111 turn happens only if we have not yet matched the entire string. */
5112 unsigned best_regs_set
= false;
5113 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5114 re_char
**best_regstart
, **best_regend
;
5117 /* Logically, this is `best_regend[0]'. But we don't want to have to
5118 allocate space for that if we're not allocating space for anything
5119 else (see below). Also, we never need info about register 0 for
5120 any of the other register vectors, and it seems rather a kludge to
5121 treat `best_regend' differently than the rest. So we keep track of
5122 the end of the best match so far in a separate variable. We
5123 initialize this to NULL so that when we backtrack the first time
5124 and need to test it, it's not garbage. */
5125 re_char
*match_end
= NULL
;
5128 /* Counts the total number of registers pushed. */
5129 unsigned num_regs_pushed
= 0;
5132 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5136 #ifdef MATCH_MAY_ALLOCATE
5137 /* Do not bother to initialize all the register variables if there are
5138 no groups in the pattern, as it takes a fair amount of time. If
5139 there are groups, we include space for register 0 (the whole
5140 pattern), even though we never use it, since it simplifies the
5141 array indexing. We should fix this. */
5144 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5145 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5146 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5147 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5149 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5157 /* We must initialize all our variables to NULL, so that
5158 `FREE_VARIABLES' doesn't try to free them. */
5159 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5161 #endif /* MATCH_MAY_ALLOCATE */
5163 /* The starting position is bogus. */
5164 if (pos
< 0 || pos
> size1
+ size2
)
5170 /* Initialize subexpression text positions to -1 to mark ones that no
5171 start_memory/stop_memory has been seen for. Also initialize the
5172 register information struct. */
5173 for (reg
= 1; reg
< num_regs
; reg
++)
5174 regstart
[reg
] = regend
[reg
] = NULL
;
5176 /* We move `string1' into `string2' if the latter's empty -- but not if
5177 `string1' is null. */
5178 if (size2
== 0 && string1
!= NULL
)
5185 end1
= string1
+ size1
;
5186 end2
= string2
+ size2
;
5188 /* `p' scans through the pattern as `d' scans through the data.
5189 `dend' is the end of the input string that `d' points within. `d'
5190 is advanced into the following input string whenever necessary, but
5191 this happens before fetching; therefore, at the beginning of the
5192 loop, `d' can be pointing at the end of a string, but it cannot
5196 /* Only match within string2. */
5197 d
= string2
+ pos
- size1
;
5198 dend
= end_match_2
= string2
+ stop
- size1
;
5199 end_match_1
= end1
; /* Just to give it a value. */
5205 /* Only match within string1. */
5206 end_match_1
= string1
+ stop
;
5208 When we reach end_match_1, PREFETCH normally switches to string2.
5209 But in the present case, this means that just doing a PREFETCH
5210 makes us jump from `stop' to `gap' within the string.
5211 What we really want here is for the search to stop as
5212 soon as we hit end_match_1. That's why we set end_match_2
5213 to end_match_1 (since PREFETCH fails as soon as we hit
5215 end_match_2
= end_match_1
;
5218 { /* It's important to use this code when stop == size so that
5219 moving `d' from end1 to string2 will not prevent the d == dend
5220 check from catching the end of string. */
5222 end_match_2
= string2
+ stop
- size1
;
5228 DEBUG_PRINT1 ("The compiled pattern is: ");
5229 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5230 DEBUG_PRINT1 ("The string to match is: `");
5231 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5232 DEBUG_PRINT1 ("'\n");
5234 /* This loops over pattern commands. It exits by returning from the
5235 function if the match is complete, or it drops through if the match
5236 fails at this starting point in the input data. */
5239 DEBUG_PRINT2 ("\n%p: ", p
);
5242 { /* End of pattern means we might have succeeded. */
5243 DEBUG_PRINT1 ("end of pattern ... ");
5245 /* If we haven't matched the entire string, and we want the
5246 longest match, try backtracking. */
5247 if (d
!= end_match_2
)
5249 /* 1 if this match ends in the same string (string1 or string2)
5250 as the best previous match. */
5251 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5252 == FIRST_STRING_P (d
));
5253 /* 1 if this match is the best seen so far. */
5254 boolean best_match_p
;
5256 /* AIX compiler got confused when this was combined
5257 with the previous declaration. */
5259 best_match_p
= d
> match_end
;
5261 best_match_p
= !FIRST_STRING_P (d
);
5263 DEBUG_PRINT1 ("backtracking.\n");
5265 if (!FAIL_STACK_EMPTY ())
5266 { /* More failure points to try. */
5268 /* If exceeds best match so far, save it. */
5269 if (!best_regs_set
|| best_match_p
)
5271 best_regs_set
= true;
5274 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5276 for (reg
= 1; reg
< num_regs
; reg
++)
5278 best_regstart
[reg
] = regstart
[reg
];
5279 best_regend
[reg
] = regend
[reg
];
5285 /* If no failure points, don't restore garbage. And if
5286 last match is real best match, don't restore second
5288 else if (best_regs_set
&& !best_match_p
)
5291 /* Restore best match. It may happen that `dend ==
5292 end_match_1' while the restored d is in string2.
5293 For example, the pattern `x.*y.*z' against the
5294 strings `x-' and `y-z-', if the two strings are
5295 not consecutive in memory. */
5296 DEBUG_PRINT1 ("Restoring best registers.\n");
5299 dend
= ((d
>= string1
&& d
<= end1
)
5300 ? end_match_1
: end_match_2
);
5302 for (reg
= 1; reg
< num_regs
; reg
++)
5304 regstart
[reg
] = best_regstart
[reg
];
5305 regend
[reg
] = best_regend
[reg
];
5308 } /* d != end_match_2 */
5311 DEBUG_PRINT1 ("Accepting match.\n");
5313 /* If caller wants register contents data back, do it. */
5314 if (regs
&& !bufp
->no_sub
)
5316 /* Have the register data arrays been allocated? */
5317 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5318 { /* No. So allocate them with malloc. We need one
5319 extra element beyond `num_regs' for the `-1' marker
5321 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5322 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5323 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5324 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5329 bufp
->regs_allocated
= REGS_REALLOCATE
;
5331 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5332 { /* Yes. If we need more elements than were already
5333 allocated, reallocate them. If we need fewer, just
5335 if (regs
->num_regs
< num_regs
+ 1)
5337 regs
->num_regs
= num_regs
+ 1;
5338 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5339 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5340 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5349 /* These braces fend off a "empty body in an else-statement"
5350 warning under GCC when assert expands to nothing. */
5351 assert (bufp
->regs_allocated
== REGS_FIXED
);
5354 /* Convert the pointer data in `regstart' and `regend' to
5355 indices. Register zero has to be set differently,
5356 since we haven't kept track of any info for it. */
5357 if (regs
->num_regs
> 0)
5359 regs
->start
[0] = pos
;
5360 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5363 /* Go through the first `min (num_regs, regs->num_regs)'
5364 registers, since that is all we initialized. */
5365 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5367 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5368 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5372 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5374 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5378 /* If the regs structure we return has more elements than
5379 were in the pattern, set the extra elements to -1. If
5380 we (re)allocated the registers, this is the case,
5381 because we always allocate enough to have at least one
5383 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5384 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5385 } /* regs && !bufp->no_sub */
5387 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5388 nfailure_points_pushed
, nfailure_points_popped
,
5389 nfailure_points_pushed
- nfailure_points_popped
);
5390 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5392 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5394 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5400 /* Otherwise match next pattern command. */
5401 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5403 /* Ignore these. Used to ignore the n of succeed_n's which
5404 currently have n == 0. */
5406 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5410 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5413 /* Match the next n pattern characters exactly. The following
5414 byte in the pattern defines n, and the n bytes after that
5415 are the characters to match. */
5418 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5420 /* Remember the start point to rollback upon failure. */
5424 /* This is written out as an if-else so we don't waste time
5425 testing `translate' inside the loop. */
5426 if (RE_TRANSLATE_P (translate
))
5430 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5450 /* The cost of testing `translate' is comparatively small. */
5451 if (target_multibyte
)
5454 int pat_charlen
, buf_charlen
;
5459 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5462 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5465 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5467 if (TRANSLATE (buf_ch
) != pat_ch
)
5475 mcnt
-= pat_charlen
;
5481 int pat_charlen
, buf_charlen
;
5487 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5488 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5495 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5496 if (! CHAR_BYTE8_P (buf_ch
))
5498 buf_ch
= TRANSLATE (buf_ch
);
5499 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5505 if (buf_ch
!= pat_ch
)
5518 /* Match any character except possibly a newline or a null. */
5524 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5527 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5529 buf_ch
= TRANSLATE (buf_ch
);
5531 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5533 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5534 && buf_ch
== '\000'))
5537 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5546 register unsigned int c
;
5547 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5550 /* Start of actual range_table, or end of bitmap if there is no
5552 re_char
*range_table
;
5554 /* Nonzero if there is a range table. */
5555 int range_table_exists
;
5557 /* Number of ranges of range table. This is not included
5558 in the initial byte-length of the command. */
5561 /* Whether matching against a unibyte character. */
5562 boolean unibyte_char
= false;
5564 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5566 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5568 if (range_table_exists
)
5570 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5571 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5575 c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5576 if (target_multibyte
)
5581 c1
= RE_CHAR_TO_UNIBYTE (c
);
5584 unibyte_char
= true;
5590 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5592 if (! CHAR_BYTE8_P (c1
))
5594 c1
= TRANSLATE (c1
);
5595 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5598 unibyte_char
= true;
5603 unibyte_char
= true;
5606 if (unibyte_char
&& c
< (1 << BYTEWIDTH
))
5607 { /* Lookup bitmap. */
5608 /* Cast to `unsigned' instead of `unsigned char' in
5609 case the bit list is a full 32 bytes long. */
5610 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5611 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5615 else if (range_table_exists
)
5617 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5619 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5620 | (class_bits
& BIT_MULTIBYTE
)
5621 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5622 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5623 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5624 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5627 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5631 if (range_table_exists
)
5632 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5634 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5636 if (!not) goto fail
;
5643 /* The beginning of a group is represented by start_memory.
5644 The argument is the register number. The text
5645 matched within the group is recorded (in the internal
5646 registers data structure) under the register number. */
5648 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5650 /* In case we need to undo this operation (via backtracking). */
5651 PUSH_FAILURE_REG ((unsigned int)*p
);
5654 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5655 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5657 /* Move past the register number and inner group count. */
5662 /* The stop_memory opcode represents the end of a group. Its
5663 argument is the same as start_memory's: the register number. */
5665 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5667 assert (!REG_UNSET (regstart
[*p
]));
5668 /* Strictly speaking, there should be code such as:
5670 assert (REG_UNSET (regend[*p]));
5671 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5673 But the only info to be pushed is regend[*p] and it is known to
5674 be UNSET, so there really isn't anything to push.
5675 Not pushing anything, on the other hand deprives us from the
5676 guarantee that regend[*p] is UNSET since undoing this operation
5677 will not reset its value properly. This is not important since
5678 the value will only be read on the next start_memory or at
5679 the very end and both events can only happen if this stop_memory
5683 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5685 /* Move past the register number and the inner group count. */
5690 /* \<digit> has been turned into a `duplicate' command which is
5691 followed by the numeric value of <digit> as the register number. */
5694 register re_char
*d2
, *dend2
;
5695 int regno
= *p
++; /* Get which register to match against. */
5696 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5698 /* Can't back reference a group which we've never matched. */
5699 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5702 /* Where in input to try to start matching. */
5703 d2
= regstart
[regno
];
5705 /* Remember the start point to rollback upon failure. */
5708 /* Where to stop matching; if both the place to start and
5709 the place to stop matching are in the same string, then
5710 set to the place to stop, otherwise, for now have to use
5711 the end of the first string. */
5713 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5714 == FIRST_STRING_P (regend
[regno
]))
5715 ? regend
[regno
] : end_match_1
);
5718 /* If necessary, advance to next segment in register
5722 if (dend2
== end_match_2
) break;
5723 if (dend2
== regend
[regno
]) break;
5725 /* End of string1 => advance to string2. */
5727 dend2
= regend
[regno
];
5729 /* At end of register contents => success */
5730 if (d2
== dend2
) break;
5732 /* If necessary, advance to next segment in data. */
5735 /* How many characters left in this segment to match. */
5738 /* Want how many consecutive characters we can match in
5739 one shot, so, if necessary, adjust the count. */
5740 if (mcnt
> dend2
- d2
)
5743 /* Compare that many; failure if mismatch, else move
5745 if (RE_TRANSLATE_P (translate
)
5746 ? bcmp_translate (d
, d2
, mcnt
, translate
, target_multibyte
)
5747 : memcmp (d
, d2
, mcnt
))
5752 d
+= mcnt
, d2
+= mcnt
;
5758 /* begline matches the empty string at the beginning of the string
5759 (unless `not_bol' is set in `bufp'), and after newlines. */
5761 DEBUG_PRINT1 ("EXECUTING begline.\n");
5763 if (AT_STRINGS_BEG (d
))
5765 if (!bufp
->not_bol
) break;
5770 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5774 /* In all other cases, we fail. */
5778 /* endline is the dual of begline. */
5780 DEBUG_PRINT1 ("EXECUTING endline.\n");
5782 if (AT_STRINGS_END (d
))
5784 if (!bufp
->not_eol
) break;
5788 PREFETCH_NOLIMIT ();
5795 /* Match at the very beginning of the data. */
5797 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5798 if (AT_STRINGS_BEG (d
))
5803 /* Match at the very end of the data. */
5805 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5806 if (AT_STRINGS_END (d
))
5811 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5812 pushes NULL as the value for the string on the stack. Then
5813 `POP_FAILURE_POINT' will keep the current value for the
5814 string, instead of restoring it. To see why, consider
5815 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5816 then the . fails against the \n. But the next thing we want
5817 to do is match the \n against the \n; if we restored the
5818 string value, we would be back at the foo.
5820 Because this is used only in specific cases, we don't need to
5821 check all the things that `on_failure_jump' does, to make
5822 sure the right things get saved on the stack. Hence we don't
5823 share its code. The only reason to push anything on the
5824 stack at all is that otherwise we would have to change
5825 `anychar's code to do something besides goto fail in this
5826 case; that seems worse than this. */
5827 case on_failure_keep_string_jump
:
5828 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5829 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5832 PUSH_FAILURE_POINT (p
- 3, NULL
);
5835 /* A nasty loop is introduced by the non-greedy *? and +?.
5836 With such loops, the stack only ever contains one failure point
5837 at a time, so that a plain on_failure_jump_loop kind of
5838 cycle detection cannot work. Worse yet, such a detection
5839 can not only fail to detect a cycle, but it can also wrongly
5840 detect a cycle (between different instantiations of the same
5842 So the method used for those nasty loops is a little different:
5843 We use a special cycle-detection-stack-frame which is pushed
5844 when the on_failure_jump_nastyloop failure-point is *popped*.
5845 This special frame thus marks the beginning of one iteration
5846 through the loop and we can hence easily check right here
5847 whether something matched between the beginning and the end of
5849 case on_failure_jump_nastyloop
:
5850 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5851 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5854 assert ((re_opcode_t
)p
[-4] == no_op
);
5857 CHECK_INFINITE_LOOP (p
- 4, d
);
5859 /* If there's a cycle, just continue without pushing
5860 this failure point. The failure point is the "try again"
5861 option, which shouldn't be tried.
5862 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5863 PUSH_FAILURE_POINT (p
- 3, d
);
5867 /* Simple loop detecting on_failure_jump: just check on the
5868 failure stack if the same spot was already hit earlier. */
5869 case on_failure_jump_loop
:
5871 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5872 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5876 CHECK_INFINITE_LOOP (p
- 3, d
);
5878 /* If there's a cycle, get out of the loop, as if the matching
5879 had failed. We used to just `goto fail' here, but that was
5880 aborting the search a bit too early: we want to keep the
5881 empty-loop-match and keep matching after the loop.
5882 We want (x?)*y\1z to match both xxyz and xxyxz. */
5885 PUSH_FAILURE_POINT (p
- 3, d
);
5890 /* Uses of on_failure_jump:
5892 Each alternative starts with an on_failure_jump that points
5893 to the beginning of the next alternative. Each alternative
5894 except the last ends with a jump that in effect jumps past
5895 the rest of the alternatives. (They really jump to the
5896 ending jump of the following alternative, because tensioning
5897 these jumps is a hassle.)
5899 Repeats start with an on_failure_jump that points past both
5900 the repetition text and either the following jump or
5901 pop_failure_jump back to this on_failure_jump. */
5902 case on_failure_jump
:
5903 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5904 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5907 PUSH_FAILURE_POINT (p
-3, d
);
5910 /* This operation is used for greedy *.
5911 Compare the beginning of the repeat with what in the
5912 pattern follows its end. If we can establish that there
5913 is nothing that they would both match, i.e., that we
5914 would have to backtrack because of (as in, e.g., `a*a')
5915 then we can use a non-backtracking loop based on
5916 on_failure_keep_string_jump instead of on_failure_jump. */
5917 case on_failure_jump_smart
:
5918 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5919 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5922 re_char
*p1
= p
; /* Next operation. */
5923 /* Here, we discard `const', making re_match non-reentrant. */
5924 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5925 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5927 p
-= 3; /* Reset so that we will re-execute the
5928 instruction once it's been changed. */
5930 EXTRACT_NUMBER (mcnt
, p2
- 2);
5932 /* Ensure this is a indeed the trivial kind of loop
5933 we are expecting. */
5934 assert (skip_one_char (p1
) == p2
- 3);
5935 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5936 DEBUG_STATEMENT (debug
+= 2);
5937 if (mutually_exclusive_p (bufp
, p1
, p2
))
5939 /* Use a fast `on_failure_keep_string_jump' loop. */
5940 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5941 *p3
= (unsigned char) on_failure_keep_string_jump
;
5942 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5946 /* Default to a safe `on_failure_jump' loop. */
5947 DEBUG_PRINT1 (" smart default => slow loop.\n");
5948 *p3
= (unsigned char) on_failure_jump
;
5950 DEBUG_STATEMENT (debug
-= 2);
5954 /* Unconditionally jump (without popping any failure points). */
5957 IMMEDIATE_QUIT_CHECK
;
5958 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5959 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5960 p
+= mcnt
; /* Do the jump. */
5961 DEBUG_PRINT2 ("(to %p).\n", p
);
5965 /* Have to succeed matching what follows at least n times.
5966 After that, handle like `on_failure_jump'. */
5968 /* Signedness doesn't matter since we only compare MCNT to 0. */
5969 EXTRACT_NUMBER (mcnt
, p
+ 2);
5970 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5972 /* Originally, mcnt is how many times we HAVE to succeed. */
5975 /* Here, we discard `const', making re_match non-reentrant. */
5976 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5979 PUSH_NUMBER (p2
, mcnt
);
5982 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5987 /* Signedness doesn't matter since we only compare MCNT to 0. */
5988 EXTRACT_NUMBER (mcnt
, p
+ 2);
5989 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5991 /* Originally, this is how many times we CAN jump. */
5994 /* Here, we discard `const', making re_match non-reentrant. */
5995 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5997 PUSH_NUMBER (p2
, mcnt
);
5998 goto unconditional_jump
;
6000 /* If don't have to jump any more, skip over the rest of command. */
6007 unsigned char *p2
; /* Location of the counter. */
6008 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
6010 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6011 /* Here, we discard `const', making re_match non-reentrant. */
6012 p2
= (unsigned char*) p
+ mcnt
;
6013 /* Signedness doesn't matter since we only copy MCNT's bits . */
6014 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6015 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
6016 PUSH_NUMBER (p2
, mcnt
);
6022 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
6023 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
6025 /* We SUCCEED (or FAIL) in one of the following cases: */
6027 /* Case 1: D is at the beginning or the end of string. */
6028 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
6032 /* C1 is the character before D, S1 is the syntax of C1, C2
6033 is the character at D, and S2 is the syntax of C2. */
6038 int offset
= PTR_TO_OFFSET (d
- 1);
6039 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6040 UPDATE_SYNTAX_TABLE (charpos
);
6042 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6045 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6047 PREFETCH_NOLIMIT ();
6048 GET_CHAR_AFTER (c2
, d
, dummy
);
6051 if (/* Case 2: Only one of S1 and S2 is Sword. */
6052 ((s1
== Sword
) != (s2
== Sword
))
6053 /* Case 3: Both of S1 and S2 are Sword, and macro
6054 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6055 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
6064 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6066 /* We FAIL in one of the following cases: */
6068 /* Case 1: D is at the end of string. */
6069 if (AT_STRINGS_END (d
))
6073 /* C1 is the character before D, S1 is the syntax of C1, C2
6074 is the character at D, and S2 is the syntax of C2. */
6079 int offset
= PTR_TO_OFFSET (d
);
6080 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6081 UPDATE_SYNTAX_TABLE (charpos
);
6084 GET_CHAR_AFTER (c2
, d
, dummy
);
6087 /* Case 2: S2 is not Sword. */
6091 /* Case 3: D is not at the beginning of string ... */
6092 if (!AT_STRINGS_BEG (d
))
6094 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6096 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6100 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6102 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6109 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6111 /* We FAIL in one of the following cases: */
6113 /* Case 1: D is at the beginning of string. */
6114 if (AT_STRINGS_BEG (d
))
6118 /* C1 is the character before D, S1 is the syntax of C1, C2
6119 is the character at D, and S2 is the syntax of C2. */
6124 int offset
= PTR_TO_OFFSET (d
) - 1;
6125 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6126 UPDATE_SYNTAX_TABLE (charpos
);
6128 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6131 /* Case 2: S1 is not Sword. */
6135 /* Case 3: D is not at the end of string ... */
6136 if (!AT_STRINGS_END (d
))
6138 PREFETCH_NOLIMIT ();
6139 GET_CHAR_AFTER (c2
, d
, dummy
);
6141 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6145 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6147 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6154 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6156 /* We FAIL in one of the following cases: */
6158 /* Case 1: D is at the end of string. */
6159 if (AT_STRINGS_END (d
))
6163 /* C1 is the character before D, S1 is the syntax of C1, C2
6164 is the character at D, and S2 is the syntax of C2. */
6168 int offset
= PTR_TO_OFFSET (d
);
6169 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6170 UPDATE_SYNTAX_TABLE (charpos
);
6173 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6176 /* Case 2: S2 is neither Sword nor Ssymbol. */
6177 if (s2
!= Sword
&& s2
!= Ssymbol
)
6180 /* Case 3: D is not at the beginning of string ... */
6181 if (!AT_STRINGS_BEG (d
))
6183 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6185 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6189 /* ... and S1 is Sword or Ssymbol. */
6190 if (s1
== Sword
|| s1
== Ssymbol
)
6197 DEBUG_PRINT1 ("EXECUTING symend.\n");
6199 /* We FAIL in one of the following cases: */
6201 /* Case 1: D is at the beginning of string. */
6202 if (AT_STRINGS_BEG (d
))
6206 /* C1 is the character before D, S1 is the syntax of C1, C2
6207 is the character at D, and S2 is the syntax of C2. */
6211 int offset
= PTR_TO_OFFSET (d
) - 1;
6212 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6213 UPDATE_SYNTAX_TABLE (charpos
);
6215 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6218 /* Case 2: S1 is neither Ssymbol nor Sword. */
6219 if (s1
!= Sword
&& s1
!= Ssymbol
)
6222 /* Case 3: D is not at the end of string ... */
6223 if (!AT_STRINGS_END (d
))
6225 PREFETCH_NOLIMIT ();
6226 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6228 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6232 /* ... and S2 is Sword or Ssymbol. */
6233 if (s2
== Sword
|| s2
== Ssymbol
)
6241 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6243 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
6247 int offset
= PTR_TO_OFFSET (d
);
6248 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6249 UPDATE_SYNTAX_TABLE (pos1
);
6256 GET_CHAR_AFTER (c
, d
, len
);
6257 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6265 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6266 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6271 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6272 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6277 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6278 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6283 case notcategoryspec
:
6284 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6286 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
6292 GET_CHAR_AFTER (c
, d
, len
);
6293 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6304 continue; /* Successfully executed one pattern command; keep going. */
6307 /* We goto here if a matching operation fails. */
6309 IMMEDIATE_QUIT_CHECK
;
6310 if (!FAIL_STACK_EMPTY ())
6313 /* A restart point is known. Restore to that state. */
6314 DEBUG_PRINT1 ("\nFAIL:\n");
6315 POP_FAILURE_POINT (str
, pat
);
6316 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
6318 case on_failure_keep_string_jump
:
6319 assert (str
== NULL
);
6320 goto continue_failure_jump
;
6322 case on_failure_jump_nastyloop
:
6323 assert ((re_opcode_t
)pat
[-2] == no_op
);
6324 PUSH_FAILURE_POINT (pat
- 2, str
);
6327 case on_failure_jump_loop
:
6328 case on_failure_jump
:
6331 continue_failure_jump
:
6332 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6337 /* A special frame used for nastyloops. */
6344 assert (p
>= bufp
->buffer
&& p
<= pend
);
6346 if (d
>= string1
&& d
<= end1
)
6350 break; /* Matching at this starting point really fails. */
6354 goto restore_best_regs
;
6358 return -1; /* Failure to match. */
6361 /* Subroutine definitions for re_match_2. */
6363 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6364 bytes; nonzero otherwise. */
6367 bcmp_translate (const re_char
*s1
, const re_char
*s2
, register int len
,
6368 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6370 register re_char
*p1
= s1
, *p2
= s2
;
6371 re_char
*p1_end
= s1
+ len
;
6372 re_char
*p2_end
= s2
+ len
;
6374 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6375 different lengths, but relying on a single `len' would break this. -sm */
6376 while (p1
< p1_end
&& p2
< p2_end
)
6378 int p1_charlen
, p2_charlen
;
6379 re_wchar_t p1_ch
, p2_ch
;
6381 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6382 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6384 if (RE_TRANSLATE (translate
, p1_ch
)
6385 != RE_TRANSLATE (translate
, p2_ch
))
6388 p1
+= p1_charlen
, p2
+= p2_charlen
;
6391 if (p1
!= p1_end
|| p2
!= p2_end
)
6397 /* Entry points for GNU code. */
6399 /* re_compile_pattern is the GNU regular expression compiler: it
6400 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6401 Returns 0 if the pattern was valid, otherwise an error string.
6403 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6404 are set in BUFP on entry.
6406 We call regex_compile to do the actual compilation. */
6409 re_compile_pattern (const char *pattern
, size_t length
, struct re_pattern_buffer
*bufp
)
6413 /* GNU code is written to assume at least RE_NREGS registers will be set
6414 (and at least one extra will be -1). */
6415 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6417 /* And GNU code determines whether or not to get register information
6418 by passing null for the REGS argument to re_match, etc., not by
6422 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6426 return gettext (re_error_msgid
[(int) ret
]);
6428 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6430 /* Entry points compatible with 4.2 BSD regex library. We don't define
6431 them unless specifically requested. */
6433 #if defined _REGEX_RE_COMP || defined _LIBC
6435 /* BSD has one and only one pattern buffer. */
6436 static struct re_pattern_buffer re_comp_buf
;
6440 /* Make these definitions weak in libc, so POSIX programs can redefine
6441 these names if they don't use our functions, and still use
6442 regcomp/regexec below without link errors. */
6452 if (!re_comp_buf
.buffer
)
6453 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6454 return (char *) gettext ("No previous regular expression");
6458 if (!re_comp_buf
.buffer
)
6460 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6461 if (re_comp_buf
.buffer
== NULL
)
6462 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6463 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6464 re_comp_buf
.allocated
= 200;
6466 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6467 if (re_comp_buf
.fastmap
== NULL
)
6468 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6469 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6472 /* Since `re_exec' always passes NULL for the `regs' argument, we
6473 don't need to initialize the pattern buffer fields which affect it. */
6475 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6480 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6481 return (char *) gettext (re_error_msgid
[(int) ret
]);
6492 const int len
= strlen (s
);
6494 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6496 #endif /* _REGEX_RE_COMP */
6498 /* POSIX.2 functions. Don't define these for Emacs. */
6502 /* regcomp takes a regular expression as a string and compiles it.
6504 PREG is a regex_t *. We do not expect any fields to be initialized,
6505 since POSIX says we shouldn't. Thus, we set
6507 `buffer' to the compiled pattern;
6508 `used' to the length of the compiled pattern;
6509 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6510 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6511 RE_SYNTAX_POSIX_BASIC;
6512 `fastmap' to an allocated space for the fastmap;
6513 `fastmap_accurate' to zero;
6514 `re_nsub' to the number of subexpressions in PATTERN.
6516 PATTERN is the address of the pattern string.
6518 CFLAGS is a series of bits which affect compilation.
6520 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6521 use POSIX basic syntax.
6523 If REG_NEWLINE is set, then . and [^...] don't match newline.
6524 Also, regexec will try a match beginning after every newline.
6526 If REG_ICASE is set, then we considers upper- and lowercase
6527 versions of letters to be equivalent when matching.
6529 If REG_NOSUB is set, then when PREG is passed to regexec, that
6530 routine will report only success or failure, and nothing about the
6533 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6534 the return codes and their meanings.) */
6537 regcomp (preg
, pattern
, cflags
)
6538 regex_t
*__restrict preg
;
6539 const char *__restrict pattern
;
6544 = (cflags
& REG_EXTENDED
) ?
6545 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6547 /* regex_compile will allocate the space for the compiled pattern. */
6549 preg
->allocated
= 0;
6552 /* Try to allocate space for the fastmap. */
6553 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6555 if (cflags
& REG_ICASE
)
6560 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6561 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6562 if (preg
->translate
== NULL
)
6563 return (int) REG_ESPACE
;
6565 /* Map uppercase characters to corresponding lowercase ones. */
6566 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6567 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6570 preg
->translate
= NULL
;
6572 /* If REG_NEWLINE is set, newlines are treated differently. */
6573 if (cflags
& REG_NEWLINE
)
6574 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6575 syntax
&= ~RE_DOT_NEWLINE
;
6576 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6579 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6581 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6583 /* POSIX says a null character in the pattern terminates it, so we
6584 can use strlen here in compiling the pattern. */
6585 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6587 /* POSIX doesn't distinguish between an unmatched open-group and an
6588 unmatched close-group: both are REG_EPAREN. */
6589 if (ret
== REG_ERPAREN
)
6592 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6593 { /* Compute the fastmap now, since regexec cannot modify the pattern
6595 re_compile_fastmap (preg
);
6596 if (preg
->can_be_null
)
6597 { /* The fastmap can't be used anyway. */
6598 free (preg
->fastmap
);
6599 preg
->fastmap
= NULL
;
6604 WEAK_ALIAS (__regcomp
, regcomp
)
6607 /* regexec searches for a given pattern, specified by PREG, in the
6610 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6611 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6612 least NMATCH elements, and we set them to the offsets of the
6613 corresponding matched substrings.
6615 EFLAGS specifies `execution flags' which affect matching: if
6616 REG_NOTBOL is set, then ^ does not match at the beginning of the
6617 string; if REG_NOTEOL is set, then $ does not match at the end.
6619 We return 0 if we find a match and REG_NOMATCH if not. */
6622 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6623 const regex_t
*__restrict preg
;
6624 const char *__restrict string
;
6626 regmatch_t pmatch
[__restrict_arr
];
6630 struct re_registers regs
;
6631 regex_t private_preg
;
6632 int len
= strlen (string
);
6633 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6635 private_preg
= *preg
;
6637 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6638 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6640 /* The user has told us exactly how many registers to return
6641 information about, via `nmatch'. We have to pass that on to the
6642 matching routines. */
6643 private_preg
.regs_allocated
= REGS_FIXED
;
6647 regs
.num_regs
= nmatch
;
6648 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6649 if (regs
.start
== NULL
)
6650 return (int) REG_NOMATCH
;
6651 regs
.end
= regs
.start
+ nmatch
;
6654 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6655 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6656 was a little bit longer but still only matching the real part.
6657 This works because the `endline' will check for a '\n' and will find a
6658 '\0', correctly deciding that this is not the end of a line.
6659 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6660 a convenient '\0' there. For all we know, the string could be preceded
6661 by '\n' which would throw things off. */
6663 /* Perform the searching operation. */
6664 ret
= re_search (&private_preg
, string
, len
,
6665 /* start: */ 0, /* range: */ len
,
6666 want_reg_info
? ®s
: (struct re_registers
*) 0);
6668 /* Copy the register information to the POSIX structure. */
6675 for (r
= 0; r
< nmatch
; r
++)
6677 pmatch
[r
].rm_so
= regs
.start
[r
];
6678 pmatch
[r
].rm_eo
= regs
.end
[r
];
6682 /* If we needed the temporary register info, free the space now. */
6686 /* We want zero return to mean success, unlike `re_search'. */
6687 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6689 WEAK_ALIAS (__regexec
, regexec
)
6692 /* Returns a message corresponding to an error code, ERR_CODE, returned
6693 from either regcomp or regexec. We don't use PREG here.
6695 ERR_CODE was previously called ERRCODE, but that name causes an
6696 error with msvc8 compiler. */
6699 regerror (err_code
, preg
, errbuf
, errbuf_size
)
6701 const regex_t
*preg
;
6709 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6710 /* Only error codes returned by the rest of the code should be passed
6711 to this routine. If we are given anything else, or if other regex
6712 code generates an invalid error code, then the program has a bug.
6713 Dump core so we can fix it. */
6716 msg
= gettext (re_error_msgid
[err_code
]);
6718 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6720 if (errbuf_size
!= 0)
6722 if (msg_size
> errbuf_size
)
6724 strncpy (errbuf
, msg
, errbuf_size
- 1);
6725 errbuf
[errbuf_size
- 1] = 0;
6728 strcpy (errbuf
, msg
);
6733 WEAK_ALIAS (__regerror
, regerror
)
6736 /* Free dynamically allocated space used by PREG. */
6742 free (preg
->buffer
);
6743 preg
->buffer
= NULL
;
6745 preg
->allocated
= 0;
6748 free (preg
->fastmap
);
6749 preg
->fastmap
= NULL
;
6750 preg
->fastmap_accurate
= 0;
6752 free (preg
->translate
);
6753 preg
->translate
= NULL
;
6755 WEAK_ALIAS (__regfree
, regfree
)
6757 #endif /* not emacs */
6759 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6760 (do not change this comment) */