1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the
3 internationalization features.)
5 Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
6 2002, 2003, 2004, 2005, 2006, 2007
7 Free Software Foundation, Inc.
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 3, or (at your option)
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; if not, write to the Free Software
21 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
25 - structure the opcode space into opcode+flag.
26 - merge with glibc's regex.[ch].
27 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
28 need to modify the compiled regexp so that re_match can be reentrant.
29 - get rid of on_failure_jump_smart by doing the optimization in re_comp
30 rather than at run-time, so that re_match can be reentrant.
33 /* AIX requires this to be the first thing in the file. */
34 #if defined _AIX && !defined REGEX_MALLOC
42 #if defined STDC_HEADERS && !defined emacs
45 /* We need this for `regex.h', and perhaps for the Emacs include files. */
46 # include <sys/types.h>
49 /* Whether to use ISO C Amendment 1 wide char functions.
50 Those should not be used for Emacs since it uses its own. */
52 #define WIDE_CHAR_SUPPORT 1
54 #define WIDE_CHAR_SUPPORT \
55 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
58 /* For platform which support the ISO C amendement 1 functionality we
59 support user defined character classes. */
61 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
67 /* We have to keep the namespace clean. */
68 # define regfree(preg) __regfree (preg)
69 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
70 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
71 # define regerror(err_code, preg, errbuf, errbuf_size) \
72 __regerror(err_code, preg, errbuf, errbuf_size)
73 # define re_set_registers(bu, re, nu, st, en) \
74 __re_set_registers (bu, re, nu, st, en)
75 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
76 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
77 # define re_match(bufp, string, size, pos, regs) \
78 __re_match (bufp, string, size, pos, regs)
79 # define re_search(bufp, string, size, startpos, range, regs) \
80 __re_search (bufp, string, size, startpos, range, regs)
81 # define re_compile_pattern(pattern, length, bufp) \
82 __re_compile_pattern (pattern, length, bufp)
83 # define re_set_syntax(syntax) __re_set_syntax (syntax)
84 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
85 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
86 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
88 /* Make sure we call libc's function even if the user overrides them. */
89 # define btowc __btowc
90 # define iswctype __iswctype
91 # define wctype __wctype
93 # define WEAK_ALIAS(a,b) weak_alias (a, b)
95 /* We are also using some library internals. */
96 # include <locale/localeinfo.h>
97 # include <locale/elem-hash.h>
98 # include <langinfo.h>
100 # define WEAK_ALIAS(a,b)
103 /* This is for other GNU distributions with internationalized messages. */
104 #if HAVE_LIBINTL_H || defined _LIBC
105 # include <libintl.h>
107 # define gettext(msgid) (msgid)
111 /* This define is so xgettext can find the internationalizable
113 # define gettext_noop(String) String
116 /* The `emacs' switch turns on certain matching commands
117 that make sense only in Emacs. */
123 /* Make syntax table lookup grant data in gl_state. */
124 # define SYNTAX_ENTRY_VIA_PROPERTY
127 # include "character.h"
128 # include "category.h"
133 # define malloc xmalloc
137 # define realloc xrealloc
143 /* Converts the pointer to the char to BEG-based offset from the start. */
144 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
145 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
147 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
148 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
149 # define RE_STRING_CHAR(p, s, multibyte) \
150 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
151 # define RE_STRING_CHAR_AND_LENGTH(p, s, len, multibyte) \
152 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
154 # define RE_CHAR_TO_MULTIBYTE(c) unibyte_to_multibyte_table[(c)]
156 # define RE_CHAR_TO_UNIBYTE(c) \
157 (ASCII_CHAR_P (c) ? (c) \
158 : CHAR_BYTE8_P (c) ? CHAR_TO_BYTE8 (c) \
159 : multibyte_char_to_unibyte_safe (c))
161 /* Set C a (possibly converted to multibyte) character before P. P
162 points into a string which is the virtual concatenation of STR1
163 (which ends at END1) or STR2 (which ends at END2). */
164 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
166 if (target_multibyte) \
168 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
169 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
170 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
171 c = STRING_CHAR (dtemp, (p) - dtemp); \
175 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
176 (c) = RE_CHAR_TO_MULTIBYTE (c); \
180 /* Set C a (possibly converted to multibyte) character at P, and set
181 LEN to the byte length of that character. */
182 # define GET_CHAR_AFTER(c, p, len) \
184 if (target_multibyte) \
185 (c) = STRING_CHAR_AND_LENGTH (p, 0, len); \
190 (c) = RE_CHAR_TO_MULTIBYTE (c); \
194 #else /* not emacs */
196 /* If we are not linking with Emacs proper,
197 we can't use the relocating allocator
198 even if config.h says that we can. */
201 # if defined STDC_HEADERS || defined _LIBC
208 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
215 val
= (void *) malloc (size
);
218 write (2, "virtual memory exhausted\n", 25);
225 xrealloc (block
, size
)
230 /* We must call malloc explicitly when BLOCK is 0, since some
231 reallocs don't do this. */
233 val
= (void *) malloc (size
);
235 val
= (void *) realloc (block
, size
);
238 write (2, "virtual memory exhausted\n", 25);
247 # define malloc xmalloc
251 # define realloc xrealloc
253 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
254 If nothing else has been done, use the method below. */
255 # ifdef INHIBIT_STRING_HEADER
256 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
257 # if !defined bzero && !defined bcopy
258 # undef INHIBIT_STRING_HEADER
263 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
264 This is used in most programs--a few other programs avoid this
265 by defining INHIBIT_STRING_HEADER. */
266 # ifndef INHIBIT_STRING_HEADER
267 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
271 # define bzero(s, n) (memset (s, '\0', n), (s))
273 # define bzero(s, n) __bzero (s, n)
277 # include <strings.h>
279 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
282 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
287 /* Define the syntax stuff for \<, \>, etc. */
289 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
290 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
292 # ifdef SWITCH_ENUM_BUG
293 # define SWITCH_ENUM_CAST(x) ((int)(x))
295 # define SWITCH_ENUM_CAST(x) (x)
298 /* Dummy macros for non-Emacs environments. */
299 # define BASE_LEADING_CODE_P(c) (0)
300 # define CHAR_CHARSET(c) 0
301 # define CHARSET_LEADING_CODE_BASE(c) 0
302 # define MAX_MULTIBYTE_LENGTH 1
303 # define RE_MULTIBYTE_P(x) 0
304 # define RE_TARGET_MULTIBYTE_P(x) 0
305 # define WORD_BOUNDARY_P(c1, c2) (0)
306 # define CHAR_HEAD_P(p) (1)
307 # define SINGLE_BYTE_CHAR_P(c) (1)
308 # define SAME_CHARSET_P(c1, c2) (1)
309 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
310 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
311 # define STRING_CHAR(p, s) (*(p))
312 # define RE_STRING_CHAR(p, s, multibyte) STRING_CHAR ((p), (s))
313 # define CHAR_STRING(c, s) (*(s) = (c), 1)
314 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
315 # define RE_STRING_CHAR_AND_LENGTH(p, s, len, multibyte) STRING_CHAR_AND_LENGTH ((p), (s), (len))
316 # define RE_CHAR_TO_MULTIBYTE(c) (c)
317 # define RE_CHAR_TO_UNIBYTE(c) (c)
318 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
319 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
320 # define GET_CHAR_AFTER(c, p, len) \
322 # define MAKE_CHAR(charset, c1, c2) (c1)
323 # define BYTE8_TO_CHAR(c) (c)
324 # define CHAR_BYTE8_P(c) (0)
325 # define CHAR_LEADING_CODE(c) (c)
327 #endif /* not emacs */
330 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
331 # define RE_TRANSLATE_P(TBL) (TBL)
334 /* Get the interface, including the syntax bits. */
337 /* isalpha etc. are used for the character classes. */
342 /* 1 if C is an ASCII character. */
343 # define IS_REAL_ASCII(c) ((c) < 0200)
345 /* 1 if C is a unibyte character. */
346 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
348 /* The Emacs definitions should not be directly affected by locales. */
350 /* In Emacs, these are only used for single-byte characters. */
351 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
352 # define ISCNTRL(c) ((c) < ' ')
353 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
354 || ((c) >= 'a' && (c) <= 'f') \
355 || ((c) >= 'A' && (c) <= 'F'))
357 /* This is only used for single-byte characters. */
358 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
360 /* The rest must handle multibyte characters. */
362 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
363 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
366 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
367 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
370 # define ISALNUM(c) (IS_REAL_ASCII (c) \
371 ? (((c) >= 'a' && (c) <= 'z') \
372 || ((c) >= 'A' && (c) <= 'Z') \
373 || ((c) >= '0' && (c) <= '9')) \
374 : SYNTAX (c) == Sword)
376 # define ISALPHA(c) (IS_REAL_ASCII (c) \
377 ? (((c) >= 'a' && (c) <= 'z') \
378 || ((c) >= 'A' && (c) <= 'Z')) \
379 : SYNTAX (c) == Sword)
381 # define ISLOWER(c) (LOWERCASEP (c))
383 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
384 ? ((c) > ' ' && (c) < 0177 \
385 && !(((c) >= 'a' && (c) <= 'z') \
386 || ((c) >= 'A' && (c) <= 'Z') \
387 || ((c) >= '0' && (c) <= '9'))) \
388 : SYNTAX (c) != Sword)
390 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
392 # define ISUPPER(c) (UPPERCASEP (c))
394 # define ISWORD(c) (SYNTAX (c) == Sword)
396 #else /* not emacs */
398 /* Jim Meyering writes:
400 "... Some ctype macros are valid only for character codes that
401 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
402 using /bin/cc or gcc but without giving an ansi option). So, all
403 ctype uses should be through macros like ISPRINT... If
404 STDC_HEADERS is defined, then autoconf has verified that the ctype
405 macros don't need to be guarded with references to isascii. ...
406 Defining isascii to 1 should let any compiler worth its salt
407 eliminate the && through constant folding."
408 Solaris defines some of these symbols so we must undefine them first. */
411 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
412 # define ISASCII(c) 1
414 # define ISASCII(c) isascii(c)
417 /* 1 if C is an ASCII character. */
418 # define IS_REAL_ASCII(c) ((c) < 0200)
420 /* This distinction is not meaningful, except in Emacs. */
421 # define ISUNIBYTE(c) 1
424 # define ISBLANK(c) (ISASCII (c) && isblank (c))
426 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
429 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
431 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
435 # define ISPRINT(c) (ISASCII (c) && isprint (c))
436 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
437 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
438 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
439 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
440 # define ISLOWER(c) (ISASCII (c) && islower (c))
441 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
442 # define ISSPACE(c) (ISASCII (c) && isspace (c))
443 # define ISUPPER(c) (ISASCII (c) && isupper (c))
444 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
446 # define ISWORD(c) ISALPHA(c)
449 # define TOLOWER(c) _tolower(c)
451 # define TOLOWER(c) tolower(c)
454 /* How many characters in the character set. */
455 # define CHAR_SET_SIZE 256
459 extern char *re_syntax_table
;
461 # else /* not SYNTAX_TABLE */
463 static char re_syntax_table
[CHAR_SET_SIZE
];
474 bzero (re_syntax_table
, sizeof re_syntax_table
);
476 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
478 re_syntax_table
[c
] = Sword
;
480 re_syntax_table
['_'] = Ssymbol
;
485 # endif /* not SYNTAX_TABLE */
487 # define SYNTAX(c) re_syntax_table[(c)]
489 #endif /* not emacs */
492 # define NULL (void *)0
495 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
496 since ours (we hope) works properly with all combinations of
497 machines, compilers, `char' and `unsigned char' argument types.
498 (Per Bothner suggested the basic approach.) */
499 #undef SIGN_EXTEND_CHAR
501 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
502 #else /* not __STDC__ */
503 /* As in Harbison and Steele. */
504 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
507 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
508 use `alloca' instead of `malloc'. This is because using malloc in
509 re_search* or re_match* could cause memory leaks when C-g is used in
510 Emacs; also, malloc is slower and causes storage fragmentation. On
511 the other hand, malloc is more portable, and easier to debug.
513 Because we sometimes use alloca, some routines have to be macros,
514 not functions -- `alloca'-allocated space disappears at the end of the
515 function it is called in. */
519 # define REGEX_ALLOCATE malloc
520 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
521 # define REGEX_FREE free
523 #else /* not REGEX_MALLOC */
525 /* Emacs already defines alloca, sometimes. */
528 /* Make alloca work the best possible way. */
530 # define alloca __builtin_alloca
531 # else /* not __GNUC__ */
532 # ifdef HAVE_ALLOCA_H
534 # endif /* HAVE_ALLOCA_H */
535 # endif /* not __GNUC__ */
537 # endif /* not alloca */
539 # define REGEX_ALLOCATE alloca
541 /* Assumes a `char *destination' variable. */
542 # define REGEX_REALLOCATE(source, osize, nsize) \
543 (destination = (char *) alloca (nsize), \
544 memcpy (destination, source, osize))
546 /* No need to do anything to free, after alloca. */
547 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
549 #endif /* not REGEX_MALLOC */
551 /* Define how to allocate the failure stack. */
553 #if defined REL_ALLOC && defined REGEX_MALLOC
555 # define REGEX_ALLOCATE_STACK(size) \
556 r_alloc (&failure_stack_ptr, (size))
557 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
558 r_re_alloc (&failure_stack_ptr, (nsize))
559 # define REGEX_FREE_STACK(ptr) \
560 r_alloc_free (&failure_stack_ptr)
562 #else /* not using relocating allocator */
566 # define REGEX_ALLOCATE_STACK malloc
567 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
568 # define REGEX_FREE_STACK free
570 # else /* not REGEX_MALLOC */
572 # define REGEX_ALLOCATE_STACK alloca
574 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
575 REGEX_REALLOCATE (source, osize, nsize)
576 /* No need to explicitly free anything. */
577 # define REGEX_FREE_STACK(arg) ((void)0)
579 # endif /* not REGEX_MALLOC */
580 #endif /* not using relocating allocator */
583 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
584 `string1' or just past its end. This works if PTR is NULL, which is
586 #define FIRST_STRING_P(ptr) \
587 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
589 /* (Re)Allocate N items of type T using malloc, or fail. */
590 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
591 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
592 #define RETALLOC_IF(addr, n, t) \
593 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
594 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
596 #define BYTEWIDTH 8 /* In bits. */
598 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
602 #define MAX(a, b) ((a) > (b) ? (a) : (b))
603 #define MIN(a, b) ((a) < (b) ? (a) : (b))
605 /* Type of source-pattern and string chars. */
606 typedef const unsigned char re_char
;
608 typedef char boolean
;
612 static int re_match_2_internal
_RE_ARGS ((struct re_pattern_buffer
*bufp
,
613 re_char
*string1
, int size1
,
614 re_char
*string2
, int size2
,
616 struct re_registers
*regs
,
619 /* These are the command codes that appear in compiled regular
620 expressions. Some opcodes are followed by argument bytes. A
621 command code can specify any interpretation whatsoever for its
622 arguments. Zero bytes may appear in the compiled regular expression. */
628 /* Succeed right away--no more backtracking. */
631 /* Followed by one byte giving n, then by n literal bytes. */
634 /* Matches any (more or less) character. */
637 /* Matches any one char belonging to specified set. First
638 following byte is number of bitmap bytes. Then come bytes
639 for a bitmap saying which chars are in. Bits in each byte
640 are ordered low-bit-first. A character is in the set if its
641 bit is 1. A character too large to have a bit in the map is
642 automatically not in the set.
644 If the length byte has the 0x80 bit set, then that stuff
645 is followed by a range table:
646 2 bytes of flags for character sets (low 8 bits, high 8 bits)
647 See RANGE_TABLE_WORK_BITS below.
648 2 bytes, the number of pairs that follow (upto 32767)
649 pairs, each 2 multibyte characters,
650 each multibyte character represented as 3 bytes. */
653 /* Same parameters as charset, but match any character that is
654 not one of those specified. */
657 /* Start remembering the text that is matched, for storing in a
658 register. Followed by one byte with the register number, in
659 the range 0 to one less than the pattern buffer's re_nsub
663 /* Stop remembering the text that is matched and store it in a
664 memory register. Followed by one byte with the register
665 number, in the range 0 to one less than `re_nsub' in the
669 /* Match a duplicate of something remembered. Followed by one
670 byte containing the register number. */
673 /* Fail unless at beginning of line. */
676 /* Fail unless at end of line. */
679 /* Succeeds if at beginning of buffer (if emacs) or at beginning
680 of string to be matched (if not). */
683 /* Analogously, for end of buffer/string. */
686 /* Followed by two byte relative address to which to jump. */
689 /* Followed by two-byte relative address of place to resume at
690 in case of failure. */
693 /* Like on_failure_jump, but pushes a placeholder instead of the
694 current string position when executed. */
695 on_failure_keep_string_jump
,
697 /* Just like `on_failure_jump', except that it checks that we
698 don't get stuck in an infinite loop (matching an empty string
700 on_failure_jump_loop
,
702 /* Just like `on_failure_jump_loop', except that it checks for
703 a different kind of loop (the kind that shows up with non-greedy
704 operators). This operation has to be immediately preceded
706 on_failure_jump_nastyloop
,
708 /* A smart `on_failure_jump' used for greedy * and + operators.
709 It analyses the loop before which it is put and if the
710 loop does not require backtracking, it changes itself to
711 `on_failure_keep_string_jump' and short-circuits the loop,
712 else it just defaults to changing itself into `on_failure_jump'.
713 It assumes that it is pointing to just past a `jump'. */
714 on_failure_jump_smart
,
716 /* Followed by two-byte relative address and two-byte number n.
717 After matching N times, jump to the address upon failure.
718 Does not work if N starts at 0: use on_failure_jump_loop
722 /* Followed by two-byte relative address, and two-byte number n.
723 Jump to the address N times, then fail. */
726 /* Set the following two-byte relative address to the
727 subsequent two-byte number. The address *includes* the two
731 wordbeg
, /* Succeeds if at word beginning. */
732 wordend
, /* Succeeds if at word end. */
734 wordbound
, /* Succeeds if at a word boundary. */
735 notwordbound
, /* Succeeds if not at a word boundary. */
737 symbeg
, /* Succeeds if at symbol beginning. */
738 symend
, /* Succeeds if at symbol end. */
740 /* Matches any character whose syntax is specified. Followed by
741 a byte which contains a syntax code, e.g., Sword. */
744 /* Matches any character whose syntax is not that specified. */
748 ,before_dot
, /* Succeeds if before point. */
749 at_dot
, /* Succeeds if at point. */
750 after_dot
, /* Succeeds if after point. */
752 /* Matches any character whose category-set contains the specified
753 category. The operator is followed by a byte which contains a
754 category code (mnemonic ASCII character). */
757 /* Matches any character whose category-set does not contain the
758 specified category. The operator is followed by a byte which
759 contains the category code (mnemonic ASCII character). */
764 /* Common operations on the compiled pattern. */
766 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
768 #define STORE_NUMBER(destination, number) \
770 (destination)[0] = (number) & 0377; \
771 (destination)[1] = (number) >> 8; \
774 /* Same as STORE_NUMBER, except increment DESTINATION to
775 the byte after where the number is stored. Therefore, DESTINATION
776 must be an lvalue. */
778 #define STORE_NUMBER_AND_INCR(destination, number) \
780 STORE_NUMBER (destination, number); \
781 (destination) += 2; \
784 /* Put into DESTINATION a number stored in two contiguous bytes starting
787 #define EXTRACT_NUMBER(destination, source) \
789 (destination) = *(source) & 0377; \
790 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
794 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
796 extract_number (dest
, source
)
800 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
801 *dest
= *source
& 0377;
805 # ifndef EXTRACT_MACROS /* To debug the macros. */
806 # undef EXTRACT_NUMBER
807 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
808 # endif /* not EXTRACT_MACROS */
812 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
813 SOURCE must be an lvalue. */
815 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
817 EXTRACT_NUMBER (destination, source); \
822 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
825 extract_number_and_incr (destination
, source
)
829 extract_number (destination
, *source
);
833 # ifndef EXTRACT_MACROS
834 # undef EXTRACT_NUMBER_AND_INCR
835 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
836 extract_number_and_incr (&dest, &src)
837 # endif /* not EXTRACT_MACROS */
841 /* Store a multibyte character in three contiguous bytes starting
842 DESTINATION, and increment DESTINATION to the byte after where the
843 character is stored. Therefore, DESTINATION must be an lvalue. */
845 #define STORE_CHARACTER_AND_INCR(destination, character) \
847 (destination)[0] = (character) & 0377; \
848 (destination)[1] = ((character) >> 8) & 0377; \
849 (destination)[2] = (character) >> 16; \
850 (destination) += 3; \
853 /* Put into DESTINATION a character stored in three contiguous bytes
854 starting at SOURCE. */
856 #define EXTRACT_CHARACTER(destination, source) \
858 (destination) = ((source)[0] \
859 | ((source)[1] << 8) \
860 | ((source)[2] << 16)); \
864 /* Macros for charset. */
866 /* Size of bitmap of charset P in bytes. P is a start of charset,
867 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
868 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
870 /* Nonzero if charset P has range table. */
871 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
873 /* Return the address of range table of charset P. But not the start
874 of table itself, but the before where the number of ranges is
875 stored. `2 +' means to skip re_opcode_t and size of bitmap,
876 and the 2 bytes of flags at the start of the range table. */
877 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
879 /* Extract the bit flags that start a range table. */
880 #define CHARSET_RANGE_TABLE_BITS(p) \
881 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
882 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
884 /* Test if C is listed in the bitmap of charset P. */
885 #define CHARSET_LOOKUP_BITMAP(p, c) \
886 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
887 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
889 /* Return the address of end of RANGE_TABLE. COUNT is number of
890 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
891 is start of range and end of range. `* 3' is size of each start
893 #define CHARSET_RANGE_TABLE_END(range_table, count) \
894 ((range_table) + (count) * 2 * 3)
896 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
897 COUNT is number of ranges in RANGE_TABLE. */
898 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
901 re_wchar_t range_start, range_end; \
903 re_char *range_table_end \
904 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
906 for (p = (range_table); p < range_table_end; p += 2 * 3) \
908 EXTRACT_CHARACTER (range_start, p); \
909 EXTRACT_CHARACTER (range_end, p + 3); \
911 if (range_start <= (c) && (c) <= range_end) \
920 /* Test if C is in range table of CHARSET. The flag NOT is negated if
921 C is listed in it. */
922 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
925 /* Number of ranges in range table. */ \
927 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
929 EXTRACT_NUMBER_AND_INCR (count, range_table); \
930 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
934 /* If DEBUG is defined, Regex prints many voluminous messages about what
935 it is doing (if the variable `debug' is nonzero). If linked with the
936 main program in `iregex.c', you can enter patterns and strings
937 interactively. And if linked with the main program in `main.c' and
938 the other test files, you can run the already-written tests. */
942 /* We use standard I/O for debugging. */
945 /* It is useful to test things that ``must'' be true when debugging. */
948 static int debug
= -100000;
950 # define DEBUG_STATEMENT(e) e
951 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
952 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
953 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
954 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
955 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
956 if (debug > 0) print_partial_compiled_pattern (s, e)
957 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
958 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
961 /* Print the fastmap in human-readable form. */
964 print_fastmap (fastmap
)
967 unsigned was_a_range
= 0;
970 while (i
< (1 << BYTEWIDTH
))
976 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
992 /* Print a compiled pattern string in human-readable form, starting at
993 the START pointer into it and ending just before the pointer END. */
996 print_partial_compiled_pattern (start
, end
)
1002 re_char
*pend
= end
;
1006 fprintf (stderr
, "(null)\n");
1010 /* Loop over pattern commands. */
1013 fprintf (stderr
, "%d:\t", p
- start
);
1015 switch ((re_opcode_t
) *p
++)
1018 fprintf (stderr
, "/no_op");
1022 fprintf (stderr
, "/succeed");
1027 fprintf (stderr
, "/exactn/%d", mcnt
);
1030 fprintf (stderr
, "/%c", *p
++);
1036 fprintf (stderr
, "/start_memory/%d", *p
++);
1040 fprintf (stderr
, "/stop_memory/%d", *p
++);
1044 fprintf (stderr
, "/duplicate/%d", *p
++);
1048 fprintf (stderr
, "/anychar");
1054 register int c
, last
= -100;
1055 register int in_range
= 0;
1056 int length
= CHARSET_BITMAP_SIZE (p
- 1);
1057 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
1059 fprintf (stderr
, "/charset [%s",
1060 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
1063 fprintf (stderr
, " !extends past end of pattern! ");
1065 for (c
= 0; c
< 256; c
++)
1067 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
1069 /* Are we starting a range? */
1070 if (last
+ 1 == c
&& ! in_range
)
1072 fprintf (stderr
, "-");
1075 /* Have we broken a range? */
1076 else if (last
+ 1 != c
&& in_range
)
1078 fprintf (stderr
, "%c", last
);
1083 fprintf (stderr
, "%c", c
);
1089 fprintf (stderr
, "%c", last
);
1091 fprintf (stderr
, "]");
1095 if (has_range_table
)
1098 fprintf (stderr
, "has-range-table");
1100 /* ??? Should print the range table; for now, just skip it. */
1101 p
+= 2; /* skip range table bits */
1102 EXTRACT_NUMBER_AND_INCR (count
, p
);
1103 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1109 fprintf (stderr
, "/begline");
1113 fprintf (stderr
, "/endline");
1116 case on_failure_jump
:
1117 extract_number_and_incr (&mcnt
, &p
);
1118 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1121 case on_failure_keep_string_jump
:
1122 extract_number_and_incr (&mcnt
, &p
);
1123 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1126 case on_failure_jump_nastyloop
:
1127 extract_number_and_incr (&mcnt
, &p
);
1128 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1131 case on_failure_jump_loop
:
1132 extract_number_and_incr (&mcnt
, &p
);
1133 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1136 case on_failure_jump_smart
:
1137 extract_number_and_incr (&mcnt
, &p
);
1138 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1142 extract_number_and_incr (&mcnt
, &p
);
1143 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1147 extract_number_and_incr (&mcnt
, &p
);
1148 extract_number_and_incr (&mcnt2
, &p
);
1149 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1153 extract_number_and_incr (&mcnt
, &p
);
1154 extract_number_and_incr (&mcnt2
, &p
);
1155 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1159 extract_number_and_incr (&mcnt
, &p
);
1160 extract_number_and_incr (&mcnt2
, &p
);
1161 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1165 fprintf (stderr
, "/wordbound");
1169 fprintf (stderr
, "/notwordbound");
1173 fprintf (stderr
, "/wordbeg");
1177 fprintf (stderr
, "/wordend");
1181 fprintf (stderr
, "/symbeg");
1185 fprintf (stderr
, "/symend");
1189 fprintf (stderr
, "/syntaxspec");
1191 fprintf (stderr
, "/%d", mcnt
);
1195 fprintf (stderr
, "/notsyntaxspec");
1197 fprintf (stderr
, "/%d", mcnt
);
1202 fprintf (stderr
, "/before_dot");
1206 fprintf (stderr
, "/at_dot");
1210 fprintf (stderr
, "/after_dot");
1214 fprintf (stderr
, "/categoryspec");
1216 fprintf (stderr
, "/%d", mcnt
);
1219 case notcategoryspec
:
1220 fprintf (stderr
, "/notcategoryspec");
1222 fprintf (stderr
, "/%d", mcnt
);
1227 fprintf (stderr
, "/begbuf");
1231 fprintf (stderr
, "/endbuf");
1235 fprintf (stderr
, "?%d", *(p
-1));
1238 fprintf (stderr
, "\n");
1241 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1246 print_compiled_pattern (bufp
)
1247 struct re_pattern_buffer
*bufp
;
1249 re_char
*buffer
= bufp
->buffer
;
1251 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1252 printf ("%ld bytes used/%ld bytes allocated.\n",
1253 bufp
->used
, bufp
->allocated
);
1255 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1257 printf ("fastmap: ");
1258 print_fastmap (bufp
->fastmap
);
1261 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1262 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1263 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1264 printf ("no_sub: %d\t", bufp
->no_sub
);
1265 printf ("not_bol: %d\t", bufp
->not_bol
);
1266 printf ("not_eol: %d\t", bufp
->not_eol
);
1267 printf ("syntax: %lx\n", bufp
->syntax
);
1269 /* Perhaps we should print the translate table? */
1274 print_double_string (where
, string1
, size1
, string2
, size2
)
1287 if (FIRST_STRING_P (where
))
1289 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1290 putchar (string1
[this_char
]);
1295 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1296 putchar (string2
[this_char
]);
1300 #else /* not DEBUG */
1305 # define DEBUG_STATEMENT(e)
1306 # define DEBUG_PRINT1(x)
1307 # define DEBUG_PRINT2(x1, x2)
1308 # define DEBUG_PRINT3(x1, x2, x3)
1309 # define DEBUG_PRINT4(x1, x2, x3, x4)
1310 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1311 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1313 #endif /* not DEBUG */
1315 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1316 also be assigned to arbitrarily: each pattern buffer stores its own
1317 syntax, so it can be changed between regex compilations. */
1318 /* This has no initializer because initialized variables in Emacs
1319 become read-only after dumping. */
1320 reg_syntax_t re_syntax_options
;
1323 /* Specify the precise syntax of regexps for compilation. This provides
1324 for compatibility for various utilities which historically have
1325 different, incompatible syntaxes.
1327 The argument SYNTAX is a bit mask comprised of the various bits
1328 defined in regex.h. We return the old syntax. */
1331 re_set_syntax (syntax
)
1332 reg_syntax_t syntax
;
1334 reg_syntax_t ret
= re_syntax_options
;
1336 re_syntax_options
= syntax
;
1339 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1341 /* Regexp to use to replace spaces, or NULL meaning don't. */
1342 static re_char
*whitespace_regexp
;
1345 re_set_whitespace_regexp (regexp
)
1348 whitespace_regexp
= (re_char
*) regexp
;
1350 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1352 /* This table gives an error message for each of the error codes listed
1353 in regex.h. Obviously the order here has to be same as there.
1354 POSIX doesn't require that we do anything for REG_NOERROR,
1355 but why not be nice? */
1357 static const char *re_error_msgid
[] =
1359 gettext_noop ("Success"), /* REG_NOERROR */
1360 gettext_noop ("No match"), /* REG_NOMATCH */
1361 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1362 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1363 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1364 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1365 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1366 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1367 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1368 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1369 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1370 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1371 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1372 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1373 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1374 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1375 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1376 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1379 /* Avoiding alloca during matching, to placate r_alloc. */
1381 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1382 searching and matching functions should not call alloca. On some
1383 systems, alloca is implemented in terms of malloc, and if we're
1384 using the relocating allocator routines, then malloc could cause a
1385 relocation, which might (if the strings being searched are in the
1386 ralloc heap) shift the data out from underneath the regexp
1389 Here's another reason to avoid allocation: Emacs
1390 processes input from X in a signal handler; processing X input may
1391 call malloc; if input arrives while a matching routine is calling
1392 malloc, then we're scrod. But Emacs can't just block input while
1393 calling matching routines; then we don't notice interrupts when
1394 they come in. So, Emacs blocks input around all regexp calls
1395 except the matching calls, which it leaves unprotected, in the
1396 faith that they will not malloc. */
1398 /* Normally, this is fine. */
1399 #define MATCH_MAY_ALLOCATE
1401 /* The match routines may not allocate if (1) they would do it with malloc
1402 and (2) it's not safe for them to use malloc.
1403 Note that if REL_ALLOC is defined, matching would not use malloc for the
1404 failure stack, but we would still use it for the register vectors;
1405 so REL_ALLOC should not affect this. */
1406 #if defined REGEX_MALLOC && defined emacs
1407 # undef MATCH_MAY_ALLOCATE
1411 /* Failure stack declarations and macros; both re_compile_fastmap and
1412 re_match_2 use a failure stack. These have to be macros because of
1413 REGEX_ALLOCATE_STACK. */
1416 /* Approximate number of failure points for which to initially allocate space
1417 when matching. If this number is exceeded, we allocate more
1418 space, so it is not a hard limit. */
1419 #ifndef INIT_FAILURE_ALLOC
1420 # define INIT_FAILURE_ALLOC 20
1423 /* Roughly the maximum number of failure points on the stack. Would be
1424 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1425 This is a variable only so users of regex can assign to it; we never
1426 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1427 before using it, so it should probably be a byte-count instead. */
1428 # if defined MATCH_MAY_ALLOCATE
1429 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1430 whose default stack limit is 2mb. In order for a larger
1431 value to work reliably, you have to try to make it accord
1432 with the process stack limit. */
1433 size_t re_max_failures
= 40000;
1435 size_t re_max_failures
= 4000;
1438 union fail_stack_elt
1441 /* This should be the biggest `int' that's no bigger than a pointer. */
1445 typedef union fail_stack_elt fail_stack_elt_t
;
1449 fail_stack_elt_t
*stack
;
1451 size_t avail
; /* Offset of next open position. */
1452 size_t frame
; /* Offset of the cur constructed frame. */
1455 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1456 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1459 /* Define macros to initialize and free the failure stack.
1460 Do `return -2' if the alloc fails. */
1462 #ifdef MATCH_MAY_ALLOCATE
1463 # define INIT_FAIL_STACK() \
1465 fail_stack.stack = (fail_stack_elt_t *) \
1466 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1467 * sizeof (fail_stack_elt_t)); \
1469 if (fail_stack.stack == NULL) \
1472 fail_stack.size = INIT_FAILURE_ALLOC; \
1473 fail_stack.avail = 0; \
1474 fail_stack.frame = 0; \
1477 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1479 # define INIT_FAIL_STACK() \
1481 fail_stack.avail = 0; \
1482 fail_stack.frame = 0; \
1485 # define RESET_FAIL_STACK() ((void)0)
1489 /* Double the size of FAIL_STACK, up to a limit
1490 which allows approximately `re_max_failures' items.
1492 Return 1 if succeeds, and 0 if either ran out of memory
1493 allocating space for it or it was already too large.
1495 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1497 /* Factor to increase the failure stack size by
1498 when we increase it.
1499 This used to be 2, but 2 was too wasteful
1500 because the old discarded stacks added up to as much space
1501 were as ultimate, maximum-size stack. */
1502 #define FAIL_STACK_GROWTH_FACTOR 4
1504 #define GROW_FAIL_STACK(fail_stack) \
1505 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1506 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1508 : ((fail_stack).stack \
1509 = (fail_stack_elt_t *) \
1510 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1511 (fail_stack).size * sizeof (fail_stack_elt_t), \
1512 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1513 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1514 * FAIL_STACK_GROWTH_FACTOR))), \
1516 (fail_stack).stack == NULL \
1518 : ((fail_stack).size \
1519 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1520 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1521 * FAIL_STACK_GROWTH_FACTOR)) \
1522 / sizeof (fail_stack_elt_t)), \
1526 /* Push a pointer value onto the failure stack.
1527 Assumes the variable `fail_stack'. Probably should only
1528 be called from within `PUSH_FAILURE_POINT'. */
1529 #define PUSH_FAILURE_POINTER(item) \
1530 fail_stack.stack[fail_stack.avail++].pointer = (item)
1532 /* This pushes an integer-valued item onto the failure stack.
1533 Assumes the variable `fail_stack'. Probably should only
1534 be called from within `PUSH_FAILURE_POINT'. */
1535 #define PUSH_FAILURE_INT(item) \
1536 fail_stack.stack[fail_stack.avail++].integer = (item)
1538 /* Push a fail_stack_elt_t value onto the failure stack.
1539 Assumes the variable `fail_stack'. Probably should only
1540 be called from within `PUSH_FAILURE_POINT'. */
1541 #define PUSH_FAILURE_ELT(item) \
1542 fail_stack.stack[fail_stack.avail++] = (item)
1544 /* These three POP... operations complement the three PUSH... operations.
1545 All assume that `fail_stack' is nonempty. */
1546 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1547 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1548 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1550 /* Individual items aside from the registers. */
1551 #define NUM_NONREG_ITEMS 3
1553 /* Used to examine the stack (to detect infinite loops). */
1554 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1555 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1556 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1557 #define TOP_FAILURE_HANDLE() fail_stack.frame
1560 #define ENSURE_FAIL_STACK(space) \
1561 while (REMAINING_AVAIL_SLOTS <= space) { \
1562 if (!GROW_FAIL_STACK (fail_stack)) \
1564 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1565 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1568 /* Push register NUM onto the stack. */
1569 #define PUSH_FAILURE_REG(num) \
1571 char *destination; \
1572 ENSURE_FAIL_STACK(3); \
1573 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1574 num, regstart[num], regend[num]); \
1575 PUSH_FAILURE_POINTER (regstart[num]); \
1576 PUSH_FAILURE_POINTER (regend[num]); \
1577 PUSH_FAILURE_INT (num); \
1580 /* Change the counter's value to VAL, but make sure that it will
1581 be reset when backtracking. */
1582 #define PUSH_NUMBER(ptr,val) \
1584 char *destination; \
1586 ENSURE_FAIL_STACK(3); \
1587 EXTRACT_NUMBER (c, ptr); \
1588 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1589 PUSH_FAILURE_INT (c); \
1590 PUSH_FAILURE_POINTER (ptr); \
1591 PUSH_FAILURE_INT (-1); \
1592 STORE_NUMBER (ptr, val); \
1595 /* Pop a saved register off the stack. */
1596 #define POP_FAILURE_REG_OR_COUNT() \
1598 int reg = POP_FAILURE_INT (); \
1601 /* It's a counter. */ \
1602 /* Here, we discard `const', making re_match non-reentrant. */ \
1603 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1604 reg = POP_FAILURE_INT (); \
1605 STORE_NUMBER (ptr, reg); \
1606 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1610 regend[reg] = POP_FAILURE_POINTER (); \
1611 regstart[reg] = POP_FAILURE_POINTER (); \
1612 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1613 reg, regstart[reg], regend[reg]); \
1617 /* Check that we are not stuck in an infinite loop. */
1618 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1620 int failure = TOP_FAILURE_HANDLE (); \
1621 /* Check for infinite matching loops */ \
1622 while (failure > 0 \
1623 && (FAILURE_STR (failure) == string_place \
1624 || FAILURE_STR (failure) == NULL)) \
1626 assert (FAILURE_PAT (failure) >= bufp->buffer \
1627 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1628 if (FAILURE_PAT (failure) == pat_cur) \
1633 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1634 failure = NEXT_FAILURE_HANDLE(failure); \
1636 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1639 /* Push the information about the state we will need
1640 if we ever fail back to it.
1642 Requires variables fail_stack, regstart, regend and
1643 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1646 Does `return FAILURE_CODE' if runs out of memory. */
1648 #define PUSH_FAILURE_POINT(pattern, string_place) \
1650 char *destination; \
1651 /* Must be int, so when we don't save any registers, the arithmetic \
1652 of 0 + -1 isn't done as unsigned. */ \
1654 DEBUG_STATEMENT (nfailure_points_pushed++); \
1655 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1656 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1657 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1659 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1661 DEBUG_PRINT1 ("\n"); \
1663 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1664 PUSH_FAILURE_INT (fail_stack.frame); \
1666 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1667 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1668 DEBUG_PRINT1 ("'\n"); \
1669 PUSH_FAILURE_POINTER (string_place); \
1671 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1672 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1673 PUSH_FAILURE_POINTER (pattern); \
1675 /* Close the frame by moving the frame pointer past it. */ \
1676 fail_stack.frame = fail_stack.avail; \
1679 /* Estimate the size of data pushed by a typical failure stack entry.
1680 An estimate is all we need, because all we use this for
1681 is to choose a limit for how big to make the failure stack. */
1682 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1683 #define TYPICAL_FAILURE_SIZE 20
1685 /* How many items can still be added to the stack without overflowing it. */
1686 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1689 /* Pops what PUSH_FAIL_STACK pushes.
1691 We restore into the parameters, all of which should be lvalues:
1692 STR -- the saved data position.
1693 PAT -- the saved pattern position.
1694 REGSTART, REGEND -- arrays of string positions.
1696 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1697 `pend', `string1', `size1', `string2', and `size2'. */
1699 #define POP_FAILURE_POINT(str, pat) \
1701 assert (!FAIL_STACK_EMPTY ()); \
1703 /* Remove failure points and point to how many regs pushed. */ \
1704 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1705 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1706 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1708 /* Pop the saved registers. */ \
1709 while (fail_stack.frame < fail_stack.avail) \
1710 POP_FAILURE_REG_OR_COUNT (); \
1712 pat = POP_FAILURE_POINTER (); \
1713 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1714 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1716 /* If the saved string location is NULL, it came from an \
1717 on_failure_keep_string_jump opcode, and we want to throw away the \
1718 saved NULL, thus retaining our current position in the string. */ \
1719 str = POP_FAILURE_POINTER (); \
1720 DEBUG_PRINT2 (" Popping string %p: `", str); \
1721 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1722 DEBUG_PRINT1 ("'\n"); \
1724 fail_stack.frame = POP_FAILURE_INT (); \
1725 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1727 assert (fail_stack.avail >= 0); \
1728 assert (fail_stack.frame <= fail_stack.avail); \
1730 DEBUG_STATEMENT (nfailure_points_popped++); \
1731 } while (0) /* POP_FAILURE_POINT */
1735 /* Registers are set to a sentinel when they haven't yet matched. */
1736 #define REG_UNSET(e) ((e) == NULL)
1738 /* Subroutine declarations and macros for regex_compile. */
1740 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1741 reg_syntax_t syntax
,
1742 struct re_pattern_buffer
*bufp
));
1743 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1744 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1745 int arg1
, int arg2
));
1746 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1747 int arg
, unsigned char *end
));
1748 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1749 int arg1
, int arg2
, unsigned char *end
));
1750 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1752 reg_syntax_t syntax
));
1753 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1755 reg_syntax_t syntax
));
1756 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1757 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1758 char *fastmap
, const int multibyte
));
1760 /* Fetch the next character in the uncompiled pattern, with no
1762 #define PATFETCH(c) \
1765 if (p == pend) return REG_EEND; \
1766 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len, multibyte); \
1771 /* If `translate' is non-null, return translate[D], else just D. We
1772 cast the subscript to translate because some data is declared as
1773 `char *', to avoid warnings when a string constant is passed. But
1774 when we use a character as a subscript we must make it unsigned. */
1776 # define TRANSLATE(d) \
1777 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1781 /* Macros for outputting the compiled pattern into `buffer'. */
1783 /* If the buffer isn't allocated when it comes in, use this. */
1784 #define INIT_BUF_SIZE 32
1786 /* Make sure we have at least N more bytes of space in buffer. */
1787 #define GET_BUFFER_SPACE(n) \
1788 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1791 /* Make sure we have one more byte of buffer space and then add C to it. */
1792 #define BUF_PUSH(c) \
1794 GET_BUFFER_SPACE (1); \
1795 *b++ = (unsigned char) (c); \
1799 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1800 #define BUF_PUSH_2(c1, c2) \
1802 GET_BUFFER_SPACE (2); \
1803 *b++ = (unsigned char) (c1); \
1804 *b++ = (unsigned char) (c2); \
1808 /* As with BUF_PUSH_2, except for three bytes. */
1809 #define BUF_PUSH_3(c1, c2, c3) \
1811 GET_BUFFER_SPACE (3); \
1812 *b++ = (unsigned char) (c1); \
1813 *b++ = (unsigned char) (c2); \
1814 *b++ = (unsigned char) (c3); \
1818 /* Store a jump with opcode OP at LOC to location TO. We store a
1819 relative address offset by the three bytes the jump itself occupies. */
1820 #define STORE_JUMP(op, loc, to) \
1821 store_op1 (op, loc, (to) - (loc) - 3)
1823 /* Likewise, for a two-argument jump. */
1824 #define STORE_JUMP2(op, loc, to, arg) \
1825 store_op2 (op, loc, (to) - (loc) - 3, arg)
1827 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1828 #define INSERT_JUMP(op, loc, to) \
1829 insert_op1 (op, loc, (to) - (loc) - 3, b)
1831 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1832 #define INSERT_JUMP2(op, loc, to, arg) \
1833 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1836 /* This is not an arbitrary limit: the arguments which represent offsets
1837 into the pattern are two bytes long. So if 2^15 bytes turns out to
1838 be too small, many things would have to change. */
1839 # define MAX_BUF_SIZE (1L << 15)
1841 #if 0 /* This is when we thought it could be 2^16 bytes. */
1842 /* Any other compiler which, like MSC, has allocation limit below 2^16
1843 bytes will have to use approach similar to what was done below for
1844 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1845 reallocating to 0 bytes. Such thing is not going to work too well.
1846 You have been warned!! */
1847 #if defined _MSC_VER && !defined WIN32
1848 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1849 # define MAX_BUF_SIZE 65500L
1851 # define MAX_BUF_SIZE (1L << 16)
1855 /* Extend the buffer by twice its current size via realloc and
1856 reset the pointers that pointed into the old block to point to the
1857 correct places in the new one. If extending the buffer results in it
1858 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1859 #if __BOUNDED_POINTERS__
1860 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1861 # define MOVE_BUFFER_POINTER(P) \
1862 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1863 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1866 SET_HIGH_BOUND (b); \
1867 SET_HIGH_BOUND (begalt); \
1868 if (fixup_alt_jump) \
1869 SET_HIGH_BOUND (fixup_alt_jump); \
1871 SET_HIGH_BOUND (laststart); \
1872 if (pending_exact) \
1873 SET_HIGH_BOUND (pending_exact); \
1876 # define MOVE_BUFFER_POINTER(P) (P) += incr
1877 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1879 #define EXTEND_BUFFER() \
1881 re_char *old_buffer = bufp->buffer; \
1882 if (bufp->allocated == MAX_BUF_SIZE) \
1884 bufp->allocated <<= 1; \
1885 if (bufp->allocated > MAX_BUF_SIZE) \
1886 bufp->allocated = MAX_BUF_SIZE; \
1887 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1888 if (bufp->buffer == NULL) \
1889 return REG_ESPACE; \
1890 /* If the buffer moved, move all the pointers into it. */ \
1891 if (old_buffer != bufp->buffer) \
1893 int incr = bufp->buffer - old_buffer; \
1894 MOVE_BUFFER_POINTER (b); \
1895 MOVE_BUFFER_POINTER (begalt); \
1896 if (fixup_alt_jump) \
1897 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1899 MOVE_BUFFER_POINTER (laststart); \
1900 if (pending_exact) \
1901 MOVE_BUFFER_POINTER (pending_exact); \
1903 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1907 /* Since we have one byte reserved for the register number argument to
1908 {start,stop}_memory, the maximum number of groups we can report
1909 things about is what fits in that byte. */
1910 #define MAX_REGNUM 255
1912 /* But patterns can have more than `MAX_REGNUM' registers. We just
1913 ignore the excess. */
1914 typedef int regnum_t
;
1917 /* Macros for the compile stack. */
1919 /* Since offsets can go either forwards or backwards, this type needs to
1920 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1921 /* int may be not enough when sizeof(int) == 2. */
1922 typedef long pattern_offset_t
;
1926 pattern_offset_t begalt_offset
;
1927 pattern_offset_t fixup_alt_jump
;
1928 pattern_offset_t laststart_offset
;
1930 } compile_stack_elt_t
;
1935 compile_stack_elt_t
*stack
;
1937 unsigned avail
; /* Offset of next open position. */
1938 } compile_stack_type
;
1941 #define INIT_COMPILE_STACK_SIZE 32
1943 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1944 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1946 /* The next available element. */
1947 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1949 /* Explicit quit checking is only used on NTemacs and whenever we
1950 use polling to process input events. */
1951 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1952 extern int immediate_quit
;
1953 # define IMMEDIATE_QUIT_CHECK \
1955 if (immediate_quit) QUIT; \
1958 # define IMMEDIATE_QUIT_CHECK ((void)0)
1961 /* Structure to manage work area for range table. */
1962 struct range_table_work_area
1964 int *table
; /* actual work area. */
1965 int allocated
; /* allocated size for work area in bytes. */
1966 int used
; /* actually used size in words. */
1967 int bits
; /* flag to record character classes */
1970 /* Make sure that WORK_AREA can hold more N multibyte characters.
1971 This is used only in set_image_of_range and set_image_of_range_1.
1972 It expects WORK_AREA to be a pointer.
1973 If it can't get the space, it returns from the surrounding function. */
1975 #define EXTEND_RANGE_TABLE(work_area, n) \
1977 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1979 extend_range_table_work_area (&work_area); \
1980 if ((work_area).table == 0) \
1981 return (REG_ESPACE); \
1985 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1986 (work_area).bits |= (bit)
1988 /* Bits used to implement the multibyte-part of the various character classes
1989 such as [:alnum:] in a charset's range table. */
1990 #define BIT_WORD 0x1
1991 #define BIT_LOWER 0x2
1992 #define BIT_PUNCT 0x4
1993 #define BIT_SPACE 0x8
1994 #define BIT_UPPER 0x10
1995 #define BIT_MULTIBYTE 0x20
1997 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1998 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
2000 EXTEND_RANGE_TABLE ((work_area), 2); \
2001 (work_area).table[(work_area).used++] = (range_start); \
2002 (work_area).table[(work_area).used++] = (range_end); \
2005 /* Free allocated memory for WORK_AREA. */
2006 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2008 if ((work_area).table) \
2009 free ((work_area).table); \
2012 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2013 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2014 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2015 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2018 /* Set the bit for character C in a list. */
2019 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2024 /* Store characters in the range FROM to TO in the bitmap at B (for
2025 ASCII and unibyte characters) and WORK_AREA (for multibyte
2026 characters) while translating them and paying attention to the
2027 continuity of translated characters.
2029 Implementation note: It is better to implement these fairly big
2030 macros by a function, but it's not that easy because macros called
2031 in this macro assume various local variables already declared. */
2033 /* Both FROM and TO are ASCII characters. */
2035 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2039 for (C0 = (FROM); C0 <= (TO); C0++) \
2041 C1 = TRANSLATE (C0); \
2042 if (! ASCII_CHAR_P (C1)) \
2044 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2045 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2048 SET_LIST_BIT (C1); \
2053 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2055 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2057 int C0, C1, C2, I; \
2058 int USED = RANGE_TABLE_WORK_USED (work_area); \
2060 for (C0 = (FROM); C0 <= (TO); C0++) \
2062 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2063 if (CHAR_BYTE8_P (C1)) \
2064 SET_LIST_BIT (C0); \
2067 C2 = TRANSLATE (C1); \
2069 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2071 SET_LIST_BIT (C1); \
2072 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2074 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2075 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2077 if (C2 >= from - 1 && C2 <= to + 1) \
2079 if (C2 == from - 1) \
2080 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2081 else if (C2 == to + 1) \
2082 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2087 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2093 /* Both FROM and TO are mulitbyte characters. */
2095 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2097 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2099 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2100 for (C0 = (FROM); C0 <= (TO); C0++) \
2102 C1 = TRANSLATE (C0); \
2103 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2104 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2105 SET_LIST_BIT (C2); \
2106 if (C1 >= (FROM) && C1 <= (TO)) \
2108 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2110 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2111 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2113 if (C1 >= from - 1 && C1 <= to + 1) \
2115 if (C1 == from - 1) \
2116 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2117 else if (C1 == to + 1) \
2118 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2123 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2129 /* Get the next unsigned number in the uncompiled pattern. */
2130 #define GET_UNSIGNED_NUMBER(num) \
2133 FREE_STACK_RETURN (REG_EBRACE); \
2137 while ('0' <= c && c <= '9') \
2143 num = num * 10 + c - '0'; \
2144 if (num / 10 != prev) \
2145 FREE_STACK_RETURN (REG_BADBR); \
2147 FREE_STACK_RETURN (REG_EBRACE); \
2153 #if ! WIDE_CHAR_SUPPORT
2155 /* Map a string to the char class it names (if any). */
2160 const char *string
= str
;
2161 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2162 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2163 else if (STREQ (string
, "word")) return RECC_WORD
;
2164 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2165 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2166 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2167 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2168 else if (STREQ (string
, "print")) return RECC_PRINT
;
2169 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2170 else if (STREQ (string
, "space")) return RECC_SPACE
;
2171 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2172 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2173 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2174 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2175 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2176 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2177 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2181 /* True if CH is in the char class CC. */
2183 re_iswctype (ch
, cc
)
2189 case RECC_ALNUM
: return ISALNUM (ch
);
2190 case RECC_ALPHA
: return ISALPHA (ch
);
2191 case RECC_BLANK
: return ISBLANK (ch
);
2192 case RECC_CNTRL
: return ISCNTRL (ch
);
2193 case RECC_DIGIT
: return ISDIGIT (ch
);
2194 case RECC_GRAPH
: return ISGRAPH (ch
);
2195 case RECC_LOWER
: return ISLOWER (ch
);
2196 case RECC_PRINT
: return ISPRINT (ch
);
2197 case RECC_PUNCT
: return ISPUNCT (ch
);
2198 case RECC_SPACE
: return ISSPACE (ch
);
2199 case RECC_UPPER
: return ISUPPER (ch
);
2200 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2201 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2202 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2203 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2204 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2205 case RECC_WORD
: return ISWORD (ch
);
2206 case RECC_ERROR
: return false;
2212 /* Return a bit-pattern to use in the range-table bits to match multibyte
2213 chars of class CC. */
2215 re_wctype_to_bit (cc
)
2220 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2221 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2222 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2223 case RECC_LOWER
: return BIT_LOWER
;
2224 case RECC_UPPER
: return BIT_UPPER
;
2225 case RECC_PUNCT
: return BIT_PUNCT
;
2226 case RECC_SPACE
: return BIT_SPACE
;
2227 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2228 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2235 /* Filling in the work area of a range. */
2237 /* Actually extend the space in WORK_AREA. */
2240 extend_range_table_work_area (work_area
)
2241 struct range_table_work_area
*work_area
;
2243 work_area
->allocated
+= 16 * sizeof (int);
2244 if (work_area
->table
)
2246 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2249 = (int *) malloc (work_area
->allocated
);
2255 /* Carefully find the ranges of codes that are equivalent
2256 under case conversion to the range start..end when passed through
2257 TRANSLATE. Handle the case where non-letters can come in between
2258 two upper-case letters (which happens in Latin-1).
2259 Also handle the case of groups of more than 2 case-equivalent chars.
2261 The basic method is to look at consecutive characters and see
2262 if they can form a run that can be handled as one.
2264 Returns -1 if successful, REG_ESPACE if ran out of space. */
2267 set_image_of_range_1 (work_area
, start
, end
, translate
)
2268 RE_TRANSLATE_TYPE translate
;
2269 struct range_table_work_area
*work_area
;
2270 re_wchar_t start
, end
;
2272 /* `one_case' indicates a character, or a run of characters,
2273 each of which is an isolate (no case-equivalents).
2274 This includes all ASCII non-letters.
2276 `two_case' indicates a character, or a run of characters,
2277 each of which has two case-equivalent forms.
2278 This includes all ASCII letters.
2280 `strange' indicates a character that has more than one
2283 enum case_type
{one_case
, two_case
, strange
};
2285 /* Describe the run that is in progress,
2286 which the next character can try to extend.
2287 If run_type is strange, that means there really is no run.
2288 If run_type is one_case, then run_start...run_end is the run.
2289 If run_type is two_case, then the run is run_start...run_end,
2290 and the case-equivalents end at run_eqv_end. */
2292 enum case_type run_type
= strange
;
2293 int run_start
, run_end
, run_eqv_end
;
2295 Lisp_Object eqv_table
;
2297 if (!RE_TRANSLATE_P (translate
))
2299 EXTEND_RANGE_TABLE (work_area
, 2);
2300 work_area
->table
[work_area
->used
++] = (start
);
2301 work_area
->table
[work_area
->used
++] = (end
);
2305 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2307 for (; start
<= end
; start
++)
2309 enum case_type this_type
;
2310 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2311 int minchar
, maxchar
;
2313 /* Classify this character */
2315 this_type
= one_case
;
2316 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2317 this_type
= two_case
;
2319 this_type
= strange
;
2322 minchar
= start
, maxchar
= eqv
;
2324 minchar
= eqv
, maxchar
= start
;
2326 /* Can this character extend the run in progress? */
2327 if (this_type
== strange
|| this_type
!= run_type
2328 || !(minchar
== run_end
+ 1
2329 && (run_type
== two_case
2330 ? maxchar
== run_eqv_end
+ 1 : 1)))
2333 Record each of its equivalent ranges. */
2334 if (run_type
== one_case
)
2336 EXTEND_RANGE_TABLE (work_area
, 2);
2337 work_area
->table
[work_area
->used
++] = run_start
;
2338 work_area
->table
[work_area
->used
++] = run_end
;
2340 else if (run_type
== two_case
)
2342 EXTEND_RANGE_TABLE (work_area
, 4);
2343 work_area
->table
[work_area
->used
++] = run_start
;
2344 work_area
->table
[work_area
->used
++] = run_end
;
2345 work_area
->table
[work_area
->used
++]
2346 = RE_TRANSLATE (eqv_table
, run_start
);
2347 work_area
->table
[work_area
->used
++]
2348 = RE_TRANSLATE (eqv_table
, run_end
);
2353 if (this_type
== strange
)
2355 /* For a strange character, add each of its equivalents, one
2356 by one. Don't start a range. */
2359 EXTEND_RANGE_TABLE (work_area
, 2);
2360 work_area
->table
[work_area
->used
++] = eqv
;
2361 work_area
->table
[work_area
->used
++] = eqv
;
2362 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2364 while (eqv
!= start
);
2367 /* Add this char to the run, or start a new run. */
2368 else if (run_type
== strange
)
2370 /* Initialize a new range. */
2371 run_type
= this_type
;
2374 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2378 /* Extend a running range. */
2380 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2384 /* If a run is still in progress at the end, finish it now
2385 by recording its equivalent ranges. */
2386 if (run_type
== one_case
)
2388 EXTEND_RANGE_TABLE (work_area
, 2);
2389 work_area
->table
[work_area
->used
++] = run_start
;
2390 work_area
->table
[work_area
->used
++] = run_end
;
2392 else if (run_type
== two_case
)
2394 EXTEND_RANGE_TABLE (work_area
, 4);
2395 work_area
->table
[work_area
->used
++] = run_start
;
2396 work_area
->table
[work_area
->used
++] = run_end
;
2397 work_area
->table
[work_area
->used
++]
2398 = RE_TRANSLATE (eqv_table
, run_start
);
2399 work_area
->table
[work_area
->used
++]
2400 = RE_TRANSLATE (eqv_table
, run_end
);
2408 /* Record the the image of the range start..end when passed through
2409 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2410 and is not even necessarily contiguous.
2411 Normally we approximate it with the smallest contiguous range that contains
2412 all the chars we need. However, for Latin-1 we go to extra effort
2415 This function is not called for ASCII ranges.
2417 Returns -1 if successful, REG_ESPACE if ran out of space. */
2420 set_image_of_range (work_area
, start
, end
, translate
)
2421 RE_TRANSLATE_TYPE translate
;
2422 struct range_table_work_area
*work_area
;
2423 re_wchar_t start
, end
;
2425 re_wchar_t cmin
, cmax
;
2428 /* For Latin-1 ranges, use set_image_of_range_1
2429 to get proper handling of ranges that include letters and nonletters.
2430 For a range that includes the whole of Latin-1, this is not necessary.
2431 For other character sets, we don't bother to get this right. */
2432 if (RE_TRANSLATE_P (translate
) && start
< 04400
2433 && !(start
< 04200 && end
>= 04377))
2440 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2450 EXTEND_RANGE_TABLE (work_area
, 2);
2451 work_area
->table
[work_area
->used
++] = (start
);
2452 work_area
->table
[work_area
->used
++] = (end
);
2454 cmin
= -1, cmax
= -1;
2456 if (RE_TRANSLATE_P (translate
))
2460 for (ch
= start
; ch
<= end
; ch
++)
2462 re_wchar_t c
= TRANSLATE (ch
);
2463 if (! (start
<= c
&& c
<= end
))
2469 cmin
= MIN (cmin
, c
);
2470 cmax
= MAX (cmax
, c
);
2477 EXTEND_RANGE_TABLE (work_area
, 2);
2478 work_area
->table
[work_area
->used
++] = (cmin
);
2479 work_area
->table
[work_area
->used
++] = (cmax
);
2487 #ifndef MATCH_MAY_ALLOCATE
2489 /* If we cannot allocate large objects within re_match_2_internal,
2490 we make the fail stack and register vectors global.
2491 The fail stack, we grow to the maximum size when a regexp
2493 The register vectors, we adjust in size each time we
2494 compile a regexp, according to the number of registers it needs. */
2496 static fail_stack_type fail_stack
;
2498 /* Size with which the following vectors are currently allocated.
2499 That is so we can make them bigger as needed,
2500 but never make them smaller. */
2501 static int regs_allocated_size
;
2503 static re_char
** regstart
, ** regend
;
2504 static re_char
**best_regstart
, **best_regend
;
2506 /* Make the register vectors big enough for NUM_REGS registers,
2507 but don't make them smaller. */
2510 regex_grow_registers (num_regs
)
2513 if (num_regs
> regs_allocated_size
)
2515 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2516 RETALLOC_IF (regend
, num_regs
, re_char
*);
2517 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2518 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2520 regs_allocated_size
= num_regs
;
2524 #endif /* not MATCH_MAY_ALLOCATE */
2526 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2530 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2531 Returns one of error codes defined in `regex.h', or zero for success.
2533 Assumes the `allocated' (and perhaps `buffer') and `translate'
2534 fields are set in BUFP on entry.
2536 If it succeeds, results are put in BUFP (if it returns an error, the
2537 contents of BUFP are undefined):
2538 `buffer' is the compiled pattern;
2539 `syntax' is set to SYNTAX;
2540 `used' is set to the length of the compiled pattern;
2541 `fastmap_accurate' is zero;
2542 `re_nsub' is the number of subexpressions in PATTERN;
2543 `not_bol' and `not_eol' are zero;
2545 The `fastmap' field is neither examined nor set. */
2547 /* Insert the `jump' from the end of last alternative to "here".
2548 The space for the jump has already been allocated. */
2549 #define FIXUP_ALT_JUMP() \
2551 if (fixup_alt_jump) \
2552 STORE_JUMP (jump, fixup_alt_jump, b); \
2556 /* Return, freeing storage we allocated. */
2557 #define FREE_STACK_RETURN(value) \
2559 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2560 free (compile_stack.stack); \
2564 static reg_errcode_t
2565 regex_compile (pattern
, size
, syntax
, bufp
)
2568 reg_syntax_t syntax
;
2569 struct re_pattern_buffer
*bufp
;
2571 /* We fetch characters from PATTERN here. */
2572 register re_wchar_t c
, c1
;
2574 /* A random temporary spot in PATTERN. */
2577 /* Points to the end of the buffer, where we should append. */
2578 register unsigned char *b
;
2580 /* Keeps track of unclosed groups. */
2581 compile_stack_type compile_stack
;
2583 /* Points to the current (ending) position in the pattern. */
2585 /* `const' makes AIX compiler fail. */
2586 unsigned char *p
= pattern
;
2588 re_char
*p
= pattern
;
2590 re_char
*pend
= pattern
+ size
;
2592 /* How to translate the characters in the pattern. */
2593 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2595 /* Address of the count-byte of the most recently inserted `exactn'
2596 command. This makes it possible to tell if a new exact-match
2597 character can be added to that command or if the character requires
2598 a new `exactn' command. */
2599 unsigned char *pending_exact
= 0;
2601 /* Address of start of the most recently finished expression.
2602 This tells, e.g., postfix * where to find the start of its
2603 operand. Reset at the beginning of groups and alternatives. */
2604 unsigned char *laststart
= 0;
2606 /* Address of beginning of regexp, or inside of last group. */
2607 unsigned char *begalt
;
2609 /* Place in the uncompiled pattern (i.e., the {) to
2610 which to go back if the interval is invalid. */
2611 re_char
*beg_interval
;
2613 /* Address of the place where a forward jump should go to the end of
2614 the containing expression. Each alternative of an `or' -- except the
2615 last -- ends with a forward jump of this sort. */
2616 unsigned char *fixup_alt_jump
= 0;
2618 /* Work area for range table of charset. */
2619 struct range_table_work_area range_table_work
;
2621 /* If the object matched can contain multibyte characters. */
2622 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2624 /* If a target of matching can contain multibyte characters. */
2625 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
2627 /* Nonzero if we have pushed down into a subpattern. */
2628 int in_subpattern
= 0;
2630 /* These hold the values of p, pattern, and pend from the main
2631 pattern when we have pushed into a subpattern. */
2633 re_char
*main_pattern
;
2638 DEBUG_PRINT1 ("\nCompiling pattern: ");
2641 unsigned debug_count
;
2643 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2644 putchar (pattern
[debug_count
]);
2649 /* Initialize the compile stack. */
2650 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2651 if (compile_stack
.stack
== NULL
)
2654 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2655 compile_stack
.avail
= 0;
2657 range_table_work
.table
= 0;
2658 range_table_work
.allocated
= 0;
2660 /* Initialize the pattern buffer. */
2661 bufp
->syntax
= syntax
;
2662 bufp
->fastmap_accurate
= 0;
2663 bufp
->not_bol
= bufp
->not_eol
= 0;
2664 bufp
->used_syntax
= 0;
2666 /* Set `used' to zero, so that if we return an error, the pattern
2667 printer (for debugging) will think there's no pattern. We reset it
2671 /* Always count groups, whether or not bufp->no_sub is set. */
2674 #if !defined emacs && !defined SYNTAX_TABLE
2675 /* Initialize the syntax table. */
2676 init_syntax_once ();
2679 if (bufp
->allocated
== 0)
2682 { /* If zero allocated, but buffer is non-null, try to realloc
2683 enough space. This loses if buffer's address is bogus, but
2684 that is the user's responsibility. */
2685 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2688 { /* Caller did not allocate a buffer. Do it for them. */
2689 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2691 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2693 bufp
->allocated
= INIT_BUF_SIZE
;
2696 begalt
= b
= bufp
->buffer
;
2698 /* Loop through the uncompiled pattern until we're at the end. */
2703 /* If this is the end of an included regexp,
2704 pop back to the main regexp and try again. */
2708 pattern
= main_pattern
;
2713 /* If this is the end of the main regexp, we are done. */
2725 /* If there's no special whitespace regexp, treat
2726 spaces normally. And don't try to do this recursively. */
2727 if (!whitespace_regexp
|| in_subpattern
)
2730 /* Peek past following spaces. */
2737 /* If the spaces are followed by a repetition op,
2738 treat them normally. */
2740 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2741 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2744 /* Replace the spaces with the whitespace regexp. */
2748 main_pattern
= pattern
;
2749 p
= pattern
= whitespace_regexp
;
2750 pend
= p
+ strlen (p
);
2756 if ( /* If at start of pattern, it's an operator. */
2758 /* If context independent, it's an operator. */
2759 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2760 /* Otherwise, depends on what's come before. */
2761 || at_begline_loc_p (pattern
, p
, syntax
))
2762 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2771 if ( /* If at end of pattern, it's an operator. */
2773 /* If context independent, it's an operator. */
2774 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2775 /* Otherwise, depends on what's next. */
2776 || at_endline_loc_p (p
, pend
, syntax
))
2777 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2786 if ((syntax
& RE_BK_PLUS_QM
)
2787 || (syntax
& RE_LIMITED_OPS
))
2791 /* If there is no previous pattern... */
2794 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2795 FREE_STACK_RETURN (REG_BADRPT
);
2796 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2801 /* 1 means zero (many) matches is allowed. */
2802 boolean zero_times_ok
= 0, many_times_ok
= 0;
2805 /* If there is a sequence of repetition chars, collapse it
2806 down to just one (the right one). We can't combine
2807 interval operators with these because of, e.g., `a{2}*',
2808 which should only match an even number of `a's. */
2812 if ((syntax
& RE_FRUGAL
)
2813 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2817 zero_times_ok
|= c
!= '+';
2818 many_times_ok
|= c
!= '?';
2824 || (!(syntax
& RE_BK_PLUS_QM
)
2825 && (*p
== '+' || *p
== '?')))
2827 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2830 FREE_STACK_RETURN (REG_EESCAPE
);
2831 if (p
[1] == '+' || p
[1] == '?')
2832 PATFETCH (c
); /* Gobble up the backslash. */
2838 /* If we get here, we found another repeat character. */
2842 /* Star, etc. applied to an empty pattern is equivalent
2843 to an empty pattern. */
2844 if (!laststart
|| laststart
== b
)
2847 /* Now we know whether or not zero matches is allowed
2848 and also whether or not two or more matches is allowed. */
2853 boolean simple
= skip_one_char (laststart
) == b
;
2854 unsigned int startoffset
= 0;
2856 /* Check if the loop can match the empty string. */
2857 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2858 ? on_failure_jump
: on_failure_jump_loop
;
2859 assert (skip_one_char (laststart
) <= b
);
2861 if (!zero_times_ok
&& simple
)
2862 { /* Since simple * loops can be made faster by using
2863 on_failure_keep_string_jump, we turn simple P+
2864 into PP* if P is simple. */
2865 unsigned char *p1
, *p2
;
2866 startoffset
= b
- laststart
;
2867 GET_BUFFER_SPACE (startoffset
);
2868 p1
= b
; p2
= laststart
;
2874 GET_BUFFER_SPACE (6);
2877 STORE_JUMP (ofj
, b
, b
+ 6);
2879 /* Simple * loops can use on_failure_keep_string_jump
2880 depending on what follows. But since we don't know
2881 that yet, we leave the decision up to
2882 on_failure_jump_smart. */
2883 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2884 laststart
+ startoffset
, b
+ 6);
2886 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2891 /* A simple ? pattern. */
2892 assert (zero_times_ok
);
2893 GET_BUFFER_SPACE (3);
2894 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2898 else /* not greedy */
2899 { /* I wish the greedy and non-greedy cases could be merged. */
2901 GET_BUFFER_SPACE (7); /* We might use less. */
2904 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2906 /* The non-greedy multiple match looks like
2907 a repeat..until: we only need a conditional jump
2908 at the end of the loop. */
2909 if (emptyp
) BUF_PUSH (no_op
);
2910 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2911 : on_failure_jump
, b
, laststart
);
2915 /* The repeat...until naturally matches one or more.
2916 To also match zero times, we need to first jump to
2917 the end of the loop (its conditional jump). */
2918 INSERT_JUMP (jump
, laststart
, b
);
2924 /* non-greedy a?? */
2925 INSERT_JUMP (jump
, laststart
, b
+ 3);
2927 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2944 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2946 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2948 /* Ensure that we have enough space to push a charset: the
2949 opcode, the length count, and the bitset; 34 bytes in all. */
2950 GET_BUFFER_SPACE (34);
2954 /* We test `*p == '^' twice, instead of using an if
2955 statement, so we only need one BUF_PUSH. */
2956 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2960 /* Remember the first position in the bracket expression. */
2963 /* Push the number of bytes in the bitmap. */
2964 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2966 /* Clear the whole map. */
2967 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2969 /* charset_not matches newline according to a syntax bit. */
2970 if ((re_opcode_t
) b
[-2] == charset_not
2971 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2972 SET_LIST_BIT ('\n');
2974 /* Read in characters and ranges, setting map bits. */
2977 boolean escaped_char
= false;
2978 const unsigned char *p2
= p
;
2981 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2983 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2984 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2985 So the translation is done later in a loop. Example:
2986 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2989 /* \ might escape characters inside [...] and [^...]. */
2990 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2992 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2995 escaped_char
= true;
2999 /* Could be the end of the bracket expression. If it's
3000 not (i.e., when the bracket expression is `[]' so
3001 far), the ']' character bit gets set way below. */
3002 if (c
== ']' && p2
!= p1
)
3006 /* See if we're at the beginning of a possible character
3009 if (!escaped_char
&&
3010 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
3012 /* Leave room for the null. */
3013 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
3014 const unsigned char *class_beg
;
3020 /* If pattern is `[[:'. */
3021 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3026 if ((c
== ':' && *p
== ']') || p
== pend
)
3028 if (c1
< CHAR_CLASS_MAX_LENGTH
)
3031 /* This is in any case an invalid class name. */
3036 /* If isn't a word bracketed by `[:' and `:]':
3037 undo the ending character, the letters, and
3038 leave the leading `:' and `[' (but set bits for
3040 if (c
== ':' && *p
== ']')
3045 cc
= re_wctype (str
);
3048 FREE_STACK_RETURN (REG_ECTYPE
);
3050 /* Throw away the ] at the end of the character
3054 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3057 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
3058 if (re_iswctype (btowc (ch
), cc
))
3061 if (c
< (1 << BYTEWIDTH
))
3065 /* Most character classes in a multibyte match
3066 just set a flag. Exceptions are is_blank,
3067 is_digit, is_cntrl, and is_xdigit, since
3068 they can only match ASCII characters. We
3069 don't need to handle them for multibyte.
3070 They are distinguished by a negative wctype. */
3072 for (ch
= 0; ch
< 256; ++ch
)
3074 c
= RE_CHAR_TO_MULTIBYTE (ch
);
3075 if (! CHAR_BYTE8_P (c
)
3076 && re_iswctype (c
, cc
))
3082 if (ASCII_CHAR_P (c1
))
3084 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
3088 SET_RANGE_TABLE_WORK_AREA_BIT
3089 (range_table_work
, re_wctype_to_bit (cc
));
3091 /* In most cases the matching rule for char classes
3092 only uses the syntax table for multibyte chars,
3093 so that the content of the syntax-table it is not
3094 hardcoded in the range_table. SPACE and WORD are
3095 the two exceptions. */
3096 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
3097 bufp
->used_syntax
= 1;
3099 /* Repeat the loop. */
3104 /* Go back to right after the "[:". */
3108 /* Because the `:' may starts the range, we
3109 can't simply set bit and repeat the loop.
3110 Instead, just set it to C and handle below. */
3115 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
3118 /* Discard the `-'. */
3121 /* Fetch the character which ends the range. */
3124 if (CHAR_BYTE8_P (c1
)
3125 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
3126 /* Treat the range from a multibyte character to
3127 raw-byte character as empty. */
3132 /* Range from C to C. */
3137 if (syntax
& RE_NO_EMPTY_RANGES
)
3138 FREE_STACK_RETURN (REG_ERANGEX
);
3139 /* Else, repeat the loop. */
3144 /* Set the range into bitmap */
3145 for (; c
<= c1
; c
++)
3148 if (ch
< (1 << BYTEWIDTH
))
3155 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
3157 if (CHAR_BYTE8_P (c1
))
3158 c
= BYTE8_TO_CHAR (128);
3162 if (CHAR_BYTE8_P (c
))
3164 c
= CHAR_TO_BYTE8 (c
);
3165 c1
= CHAR_TO_BYTE8 (c1
);
3166 for (; c
<= c1
; c
++)
3171 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
3175 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
3182 /* Discard any (non)matching list bytes that are all 0 at the
3183 end of the map. Decrease the map-length byte too. */
3184 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3188 /* Build real range table from work area. */
3189 if (RANGE_TABLE_WORK_USED (range_table_work
)
3190 || RANGE_TABLE_WORK_BITS (range_table_work
))
3193 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3195 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3196 bytes for flags, two for COUNT, and three bytes for
3198 GET_BUFFER_SPACE (4 + used
* 3);
3200 /* Indicate the existence of range table. */
3201 laststart
[1] |= 0x80;
3203 /* Store the character class flag bits into the range table.
3204 If not in emacs, these flag bits are always 0. */
3205 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3206 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3208 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3209 for (i
= 0; i
< used
; i
++)
3210 STORE_CHARACTER_AND_INCR
3211 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3218 if (syntax
& RE_NO_BK_PARENS
)
3225 if (syntax
& RE_NO_BK_PARENS
)
3232 if (syntax
& RE_NEWLINE_ALT
)
3239 if (syntax
& RE_NO_BK_VBAR
)
3246 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3247 goto handle_interval
;
3253 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3255 /* Do not translate the character after the \, so that we can
3256 distinguish, e.g., \B from \b, even if we normally would
3257 translate, e.g., B to b. */
3263 if (syntax
& RE_NO_BK_PARENS
)
3264 goto normal_backslash
;
3269 regnum_t regnum
= 0;
3272 /* Look for a special (?...) construct */
3273 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3275 PATFETCH (c
); /* Gobble up the '?'. */
3281 case ':': shy
= 1; break;
3283 /* An explicitly specified regnum must start
3286 FREE_STACK_RETURN (REG_BADPAT
);
3287 case '1': case '2': case '3': case '4':
3288 case '5': case '6': case '7': case '8': case '9':
3289 regnum
= 10*regnum
+ (c
- '0'); break;
3291 /* Only (?:...) is supported right now. */
3292 FREE_STACK_RETURN (REG_BADPAT
);
3299 regnum
= ++bufp
->re_nsub
;
3301 { /* It's actually not shy, but explicitly numbered. */
3303 if (regnum
> bufp
->re_nsub
)
3304 bufp
->re_nsub
= regnum
;
3305 else if (regnum
> bufp
->re_nsub
3306 /* Ideally, we'd want to check that the specified
3307 group can't have matched (i.e. all subgroups
3308 using the same regnum are in other branches of
3309 OR patterns), but we don't currently keep track
3310 of enough info to do that easily. */
3311 || group_in_compile_stack (compile_stack
, regnum
))
3312 FREE_STACK_RETURN (REG_BADPAT
);
3315 /* It's really shy. */
3316 regnum
= - bufp
->re_nsub
;
3318 if (COMPILE_STACK_FULL
)
3320 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3321 compile_stack_elt_t
);
3322 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3324 compile_stack
.size
<<= 1;
3327 /* These are the values to restore when we hit end of this
3328 group. They are all relative offsets, so that if the
3329 whole pattern moves because of realloc, they will still
3331 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3332 COMPILE_STACK_TOP
.fixup_alt_jump
3333 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3334 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3335 COMPILE_STACK_TOP
.regnum
= regnum
;
3337 /* Do not push a start_memory for groups beyond the last one
3338 we can represent in the compiled pattern. */
3339 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3340 BUF_PUSH_2 (start_memory
, regnum
);
3342 compile_stack
.avail
++;
3347 /* If we've reached MAX_REGNUM groups, then this open
3348 won't actually generate any code, so we'll have to
3349 clear pending_exact explicitly. */
3355 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3357 if (COMPILE_STACK_EMPTY
)
3359 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3360 goto normal_backslash
;
3362 FREE_STACK_RETURN (REG_ERPAREN
);
3368 /* See similar code for backslashed left paren above. */
3369 if (COMPILE_STACK_EMPTY
)
3371 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3374 FREE_STACK_RETURN (REG_ERPAREN
);
3377 /* Since we just checked for an empty stack above, this
3378 ``can't happen''. */
3379 assert (compile_stack
.avail
!= 0);
3381 /* We don't just want to restore into `regnum', because
3382 later groups should continue to be numbered higher,
3383 as in `(ab)c(de)' -- the second group is #2. */
3386 compile_stack
.avail
--;
3387 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3389 = COMPILE_STACK_TOP
.fixup_alt_jump
3390 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3392 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3393 regnum
= COMPILE_STACK_TOP
.regnum
;
3394 /* If we've reached MAX_REGNUM groups, then this open
3395 won't actually generate any code, so we'll have to
3396 clear pending_exact explicitly. */
3399 /* We're at the end of the group, so now we know how many
3400 groups were inside this one. */
3401 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3402 BUF_PUSH_2 (stop_memory
, regnum
);
3407 case '|': /* `\|'. */
3408 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3409 goto normal_backslash
;
3411 if (syntax
& RE_LIMITED_OPS
)
3414 /* Insert before the previous alternative a jump which
3415 jumps to this alternative if the former fails. */
3416 GET_BUFFER_SPACE (3);
3417 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3421 /* The alternative before this one has a jump after it
3422 which gets executed if it gets matched. Adjust that
3423 jump so it will jump to this alternative's analogous
3424 jump (put in below, which in turn will jump to the next
3425 (if any) alternative's such jump, etc.). The last such
3426 jump jumps to the correct final destination. A picture:
3432 If we are at `b', then fixup_alt_jump right now points to a
3433 three-byte space after `a'. We'll put in the jump, set
3434 fixup_alt_jump to right after `b', and leave behind three
3435 bytes which we'll fill in when we get to after `c'. */
3439 /* Mark and leave space for a jump after this alternative,
3440 to be filled in later either by next alternative or
3441 when know we're at the end of a series of alternatives. */
3443 GET_BUFFER_SPACE (3);
3452 /* If \{ is a literal. */
3453 if (!(syntax
& RE_INTERVALS
)
3454 /* If we're at `\{' and it's not the open-interval
3456 || (syntax
& RE_NO_BK_BRACES
))
3457 goto normal_backslash
;
3461 /* If got here, then the syntax allows intervals. */
3463 /* At least (most) this many matches must be made. */
3464 int lower_bound
= 0, upper_bound
= -1;
3468 GET_UNSIGNED_NUMBER (lower_bound
);
3471 GET_UNSIGNED_NUMBER (upper_bound
);
3473 /* Interval such as `{1}' => match exactly once. */
3474 upper_bound
= lower_bound
;
3476 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3477 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3478 FREE_STACK_RETURN (REG_BADBR
);
3480 if (!(syntax
& RE_NO_BK_BRACES
))
3483 FREE_STACK_RETURN (REG_BADBR
);
3485 FREE_STACK_RETURN (REG_EESCAPE
);
3490 FREE_STACK_RETURN (REG_BADBR
);
3492 /* We just parsed a valid interval. */
3494 /* If it's invalid to have no preceding re. */
3497 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3498 FREE_STACK_RETURN (REG_BADRPT
);
3499 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3502 goto unfetch_interval
;
3505 if (upper_bound
== 0)
3506 /* If the upper bound is zero, just drop the sub pattern
3509 else if (lower_bound
== 1 && upper_bound
== 1)
3510 /* Just match it once: nothing to do here. */
3513 /* Otherwise, we have a nontrivial interval. When
3514 we're all done, the pattern will look like:
3515 set_number_at <jump count> <upper bound>
3516 set_number_at <succeed_n count> <lower bound>
3517 succeed_n <after jump addr> <succeed_n count>
3519 jump_n <succeed_n addr> <jump count>
3520 (The upper bound and `jump_n' are omitted if
3521 `upper_bound' is 1, though.) */
3523 { /* If the upper bound is > 1, we need to insert
3524 more at the end of the loop. */
3525 unsigned int nbytes
= (upper_bound
< 0 ? 3
3526 : upper_bound
> 1 ? 5 : 0);
3527 unsigned int startoffset
= 0;
3529 GET_BUFFER_SPACE (20); /* We might use less. */
3531 if (lower_bound
== 0)
3533 /* A succeed_n that starts with 0 is really a
3534 a simple on_failure_jump_loop. */
3535 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3541 /* Initialize lower bound of the `succeed_n', even
3542 though it will be set during matching by its
3543 attendant `set_number_at' (inserted next),
3544 because `re_compile_fastmap' needs to know.
3545 Jump to the `jump_n' we might insert below. */
3546 INSERT_JUMP2 (succeed_n
, laststart
,
3551 /* Code to initialize the lower bound. Insert
3552 before the `succeed_n'. The `5' is the last two
3553 bytes of this `set_number_at', plus 3 bytes of
3554 the following `succeed_n'. */
3555 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3560 if (upper_bound
< 0)
3562 /* A negative upper bound stands for infinity,
3563 in which case it degenerates to a plain jump. */
3564 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3567 else if (upper_bound
> 1)
3568 { /* More than one repetition is allowed, so
3569 append a backward jump to the `succeed_n'
3570 that starts this interval.
3572 When we've reached this during matching,
3573 we'll have matched the interval once, so
3574 jump back only `upper_bound - 1' times. */
3575 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3579 /* The location we want to set is the second
3580 parameter of the `jump_n'; that is `b-2' as
3581 an absolute address. `laststart' will be
3582 the `set_number_at' we're about to insert;
3583 `laststart+3' the number to set, the source
3584 for the relative address. But we are
3585 inserting into the middle of the pattern --
3586 so everything is getting moved up by 5.
3587 Conclusion: (b - 2) - (laststart + 3) + 5,
3588 i.e., b - laststart.
3590 We insert this at the beginning of the loop
3591 so that if we fail during matching, we'll
3592 reinitialize the bounds. */
3593 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3594 upper_bound
- 1, b
);
3599 beg_interval
= NULL
;
3604 /* If an invalid interval, match the characters as literals. */
3605 assert (beg_interval
);
3607 beg_interval
= NULL
;
3609 /* normal_char and normal_backslash need `c'. */
3612 if (!(syntax
& RE_NO_BK_BRACES
))
3614 assert (p
> pattern
&& p
[-1] == '\\');
3615 goto normal_backslash
;
3621 /* There is no way to specify the before_dot and after_dot
3622 operators. rms says this is ok. --karl */
3630 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3636 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3642 BUF_PUSH_2 (categoryspec
, c
);
3648 BUF_PUSH_2 (notcategoryspec
, c
);
3654 if (syntax
& RE_NO_GNU_OPS
)
3657 BUF_PUSH_2 (syntaxspec
, Sword
);
3662 if (syntax
& RE_NO_GNU_OPS
)
3665 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3670 if (syntax
& RE_NO_GNU_OPS
)
3676 if (syntax
& RE_NO_GNU_OPS
)
3682 if (syntax
& RE_NO_GNU_OPS
)
3691 FREE_STACK_RETURN (REG_BADPAT
);
3695 if (syntax
& RE_NO_GNU_OPS
)
3697 BUF_PUSH (wordbound
);
3701 if (syntax
& RE_NO_GNU_OPS
)
3703 BUF_PUSH (notwordbound
);
3707 if (syntax
& RE_NO_GNU_OPS
)
3713 if (syntax
& RE_NO_GNU_OPS
)
3718 case '1': case '2': case '3': case '4': case '5':
3719 case '6': case '7': case '8': case '9':
3723 if (syntax
& RE_NO_BK_REFS
)
3724 goto normal_backslash
;
3728 if (reg
> bufp
->re_nsub
|| reg
< 1
3729 /* Can't back reference to a subexp before its end. */
3730 || group_in_compile_stack (compile_stack
, reg
))
3731 FREE_STACK_RETURN (REG_ESUBREG
);
3734 BUF_PUSH_2 (duplicate
, reg
);
3741 if (syntax
& RE_BK_PLUS_QM
)
3744 goto normal_backslash
;
3748 /* You might think it would be useful for \ to mean
3749 not to translate; but if we don't translate it
3750 it will never match anything. */
3757 /* Expects the character in `c'. */
3759 /* If no exactn currently being built. */
3762 /* If last exactn not at current position. */
3763 || pending_exact
+ *pending_exact
+ 1 != b
3765 /* We have only one byte following the exactn for the count. */
3766 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3768 /* If followed by a repetition operator. */
3769 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3770 || ((syntax
& RE_BK_PLUS_QM
)
3771 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3772 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3773 || ((syntax
& RE_INTERVALS
)
3774 && ((syntax
& RE_NO_BK_BRACES
)
3775 ? p
!= pend
&& *p
== '{'
3776 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3778 /* Start building a new exactn. */
3782 BUF_PUSH_2 (exactn
, 0);
3783 pending_exact
= b
- 1;
3786 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3793 len
= CHAR_STRING (c
, b
);
3798 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3799 if (! CHAR_BYTE8_P (c1
))
3801 re_wchar_t c2
= TRANSLATE (c1
);
3803 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3809 (*pending_exact
) += len
;
3814 } /* while p != pend */
3817 /* Through the pattern now. */
3821 if (!COMPILE_STACK_EMPTY
)
3822 FREE_STACK_RETURN (REG_EPAREN
);
3824 /* If we don't want backtracking, force success
3825 the first time we reach the end of the compiled pattern. */
3826 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3829 /* We have succeeded; set the length of the buffer. */
3830 bufp
->used
= b
- bufp
->buffer
;
3835 re_compile_fastmap (bufp
);
3836 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3837 print_compiled_pattern (bufp
);
3842 #ifndef MATCH_MAY_ALLOCATE
3843 /* Initialize the failure stack to the largest possible stack. This
3844 isn't necessary unless we're trying to avoid calling alloca in
3845 the search and match routines. */
3847 int num_regs
= bufp
->re_nsub
+ 1;
3849 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3851 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3853 if (! fail_stack
.stack
)
3855 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3856 * sizeof (fail_stack_elt_t
));
3859 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3861 * sizeof (fail_stack_elt_t
)));
3864 regex_grow_registers (num_regs
);
3866 #endif /* not MATCH_MAY_ALLOCATE */
3868 FREE_STACK_RETURN (REG_NOERROR
);
3869 } /* regex_compile */
3871 /* Subroutines for `regex_compile'. */
3873 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3876 store_op1 (op
, loc
, arg
)
3881 *loc
= (unsigned char) op
;
3882 STORE_NUMBER (loc
+ 1, arg
);
3886 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3889 store_op2 (op
, loc
, arg1
, arg2
)
3894 *loc
= (unsigned char) op
;
3895 STORE_NUMBER (loc
+ 1, arg1
);
3896 STORE_NUMBER (loc
+ 3, arg2
);
3900 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3901 for OP followed by two-byte integer parameter ARG. */
3904 insert_op1 (op
, loc
, arg
, end
)
3910 register unsigned char *pfrom
= end
;
3911 register unsigned char *pto
= end
+ 3;
3913 while (pfrom
!= loc
)
3916 store_op1 (op
, loc
, arg
);
3920 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3923 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3929 register unsigned char *pfrom
= end
;
3930 register unsigned char *pto
= end
+ 5;
3932 while (pfrom
!= loc
)
3935 store_op2 (op
, loc
, arg1
, arg2
);
3939 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3940 after an alternative or a begin-subexpression. We assume there is at
3941 least one character before the ^. */
3944 at_begline_loc_p (pattern
, p
, syntax
)
3945 re_char
*pattern
, *p
;
3946 reg_syntax_t syntax
;
3948 re_char
*prev
= p
- 2;
3949 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3952 /* After a subexpression? */
3953 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3954 /* After an alternative? */
3955 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3956 /* After a shy subexpression? */
3957 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3958 && prev
[-1] == '?' && prev
[-2] == '('
3959 && (syntax
& RE_NO_BK_PARENS
3960 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3964 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3965 at least one character after the $, i.e., `P < PEND'. */
3968 at_endline_loc_p (p
, pend
, syntax
)
3970 reg_syntax_t syntax
;
3973 boolean next_backslash
= *next
== '\\';
3974 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3977 /* Before a subexpression? */
3978 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3979 : next_backslash
&& next_next
&& *next_next
== ')')
3980 /* Before an alternative? */
3981 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3982 : next_backslash
&& next_next
&& *next_next
== '|');
3986 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3987 false if it's not. */
3990 group_in_compile_stack (compile_stack
, regnum
)
3991 compile_stack_type compile_stack
;
3996 for (this_element
= compile_stack
.avail
- 1;
3999 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
4006 If fastmap is non-NULL, go through the pattern and fill fastmap
4007 with all the possible leading chars. If fastmap is NULL, don't
4008 bother filling it up (obviously) and only return whether the
4009 pattern could potentially match the empty string.
4011 Return 1 if p..pend might match the empty string.
4012 Return 0 if p..pend matches at least one char.
4013 Return -1 if fastmap was not updated accurately. */
4016 analyse_first (p
, pend
, fastmap
, multibyte
)
4019 const int multibyte
;
4024 /* If all elements for base leading-codes in fastmap is set, this
4025 flag is set true. */
4026 boolean match_any_multibyte_characters
= false;
4030 /* The loop below works as follows:
4031 - It has a working-list kept in the PATTERN_STACK and which basically
4032 starts by only containing a pointer to the first operation.
4033 - If the opcode we're looking at is a match against some set of
4034 chars, then we add those chars to the fastmap and go on to the
4035 next work element from the worklist (done via `break').
4036 - If the opcode is a control operator on the other hand, we either
4037 ignore it (if it's meaningless at this point, such as `start_memory')
4038 or execute it (if it's a jump). If the jump has several destinations
4039 (i.e. `on_failure_jump'), then we push the other destination onto the
4041 We guarantee termination by ignoring backward jumps (more or less),
4042 so that `p' is monotonically increasing. More to the point, we
4043 never set `p' (or push) anything `<= p1'. */
4047 /* `p1' is used as a marker of how far back a `on_failure_jump'
4048 can go without being ignored. It is normally equal to `p'
4049 (which prevents any backward `on_failure_jump') except right
4050 after a plain `jump', to allow patterns such as:
4053 10: on_failure_jump 3
4054 as used for the *? operator. */
4057 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4064 /* If the first character has to match a backreference, that means
4065 that the group was empty (since it already matched). Since this
4066 is the only case that interests us here, we can assume that the
4067 backreference must match the empty string. */
4072 /* Following are the cases which match a character. These end
4078 /* If multibyte is nonzero, the first byte of each
4079 character is an ASCII or a leading code. Otherwise,
4080 each byte is a character. Thus, this works in both
4085 /* For the case of matching this unibyte regex
4086 against multibyte, we must set a leading code of
4087 the corresponding multibyte character. */
4088 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
4090 if (! CHAR_BYTE8_P (c
))
4091 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
4098 /* We could put all the chars except for \n (and maybe \0)
4099 but we don't bother since it is generally not worth it. */
4100 if (!fastmap
) break;
4105 if (!fastmap
) break;
4107 /* Chars beyond end of bitmap are possible matches. */
4108 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
4109 j
< (1 << BYTEWIDTH
); j
++)
4115 if (!fastmap
) break;
4116 not = (re_opcode_t
) *(p
- 1) == charset_not
;
4117 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
4119 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
4123 if (/* Any leading code can possibly start a character
4124 which doesn't match the specified set of characters. */
4127 /* If we can match a character class, we can match any
4128 multibyte characters. */
4129 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
4130 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
4133 if (match_any_multibyte_characters
== false)
4135 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4136 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4138 match_any_multibyte_characters
= true;
4142 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
4143 && match_any_multibyte_characters
== false)
4145 /* Set fastmap[I] to 1 where I is a leading code of each
4146 multibyte characer in the range table. */
4148 unsigned char lc1
, lc2
;
4150 /* Make P points the range table. `+ 2' is to skip flag
4151 bits for a character class. */
4152 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
4154 /* Extract the number of ranges in range table into COUNT. */
4155 EXTRACT_NUMBER_AND_INCR (count
, p
);
4156 for (; count
> 0; count
--, p
+= 3)
4158 /* Extract the start and end of each range. */
4159 EXTRACT_CHARACTER (c
, p
);
4160 lc1
= CHAR_LEADING_CODE (c
);
4162 EXTRACT_CHARACTER (c
, p
);
4163 lc2
= CHAR_LEADING_CODE (c
);
4164 for (j
= lc1
; j
<= lc2
; j
++)
4173 if (!fastmap
) break;
4175 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4177 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4178 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4182 /* This match depends on text properties. These end with
4183 aborting optimizations. */
4187 case notcategoryspec
:
4188 if (!fastmap
) break;
4189 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4191 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
4192 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4195 /* Any leading code can possibly start a character which
4196 has or doesn't has the specified category. */
4197 if (match_any_multibyte_characters
== false)
4199 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4200 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4202 match_any_multibyte_characters
= true;
4206 /* All cases after this match the empty string. These end with
4228 EXTRACT_NUMBER_AND_INCR (j
, p
);
4230 /* Backward jumps can only go back to code that we've already
4231 visited. `re_compile' should make sure this is true. */
4234 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4236 case on_failure_jump
:
4237 case on_failure_keep_string_jump
:
4238 case on_failure_jump_loop
:
4239 case on_failure_jump_nastyloop
:
4240 case on_failure_jump_smart
:
4246 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4247 to jump back to "just after here". */
4250 case on_failure_jump
:
4251 case on_failure_keep_string_jump
:
4252 case on_failure_jump_nastyloop
:
4253 case on_failure_jump_loop
:
4254 case on_failure_jump_smart
:
4255 EXTRACT_NUMBER_AND_INCR (j
, p
);
4257 ; /* Backward jump to be ignored. */
4259 { /* We have to look down both arms.
4260 We first go down the "straight" path so as to minimize
4261 stack usage when going through alternatives. */
4262 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4270 /* This code simply does not properly handle forward jump_n. */
4271 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4273 /* jump_n can either jump or fall through. The (backward) jump
4274 case has already been handled, so we only need to look at the
4275 fallthrough case. */
4279 /* If N == 0, it should be an on_failure_jump_loop instead. */
4280 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4282 /* We only care about one iteration of the loop, so we don't
4283 need to consider the case where this behaves like an
4300 abort (); /* We have listed all the cases. */
4303 /* Getting here means we have found the possible starting
4304 characters for one path of the pattern -- and that the empty
4305 string does not match. We need not follow this path further. */
4309 /* We reached the end without matching anything. */
4312 } /* analyse_first */
4314 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4315 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4316 characters can start a string that matches the pattern. This fastmap
4317 is used by re_search to skip quickly over impossible starting points.
4319 Character codes above (1 << BYTEWIDTH) are not represented in the
4320 fastmap, but the leading codes are represented. Thus, the fastmap
4321 indicates which character sets could start a match.
4323 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4324 area as BUFP->fastmap.
4326 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4329 Returns 0 if we succeed, -2 if an internal error. */
4332 re_compile_fastmap (bufp
)
4333 struct re_pattern_buffer
*bufp
;
4335 char *fastmap
= bufp
->fastmap
;
4338 assert (fastmap
&& bufp
->buffer
);
4340 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4341 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4343 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4344 fastmap
, RE_MULTIBYTE_P (bufp
));
4345 bufp
->can_be_null
= (analysis
!= 0);
4347 } /* re_compile_fastmap */
4349 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4350 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4351 this memory for recording register information. STARTS and ENDS
4352 must be allocated using the malloc library routine, and must each
4353 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4355 If NUM_REGS == 0, then subsequent matches should allocate their own
4358 Unless this function is called, the first search or match using
4359 PATTERN_BUFFER will allocate its own register data, without
4360 freeing the old data. */
4363 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4364 struct re_pattern_buffer
*bufp
;
4365 struct re_registers
*regs
;
4367 regoff_t
*starts
, *ends
;
4371 bufp
->regs_allocated
= REGS_REALLOCATE
;
4372 regs
->num_regs
= num_regs
;
4373 regs
->start
= starts
;
4378 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4380 regs
->start
= regs
->end
= (regoff_t
*) 0;
4383 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4385 /* Searching routines. */
4387 /* Like re_search_2, below, but only one string is specified, and
4388 doesn't let you say where to stop matching. */
4391 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4392 struct re_pattern_buffer
*bufp
;
4394 int size
, startpos
, range
;
4395 struct re_registers
*regs
;
4397 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4400 WEAK_ALIAS (__re_search
, re_search
)
4402 /* Head address of virtual concatenation of string. */
4403 #define HEAD_ADDR_VSTRING(P) \
4404 (((P) >= size1 ? string2 : string1))
4406 /* End address of virtual concatenation of string. */
4407 #define STOP_ADDR_VSTRING(P) \
4408 (((P) >= size1 ? string2 + size2 : string1 + size1))
4410 /* Address of POS in the concatenation of virtual string. */
4411 #define POS_ADDR_VSTRING(POS) \
4412 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4414 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4415 virtual concatenation of STRING1 and STRING2, starting first at index
4416 STARTPOS, then at STARTPOS + 1, and so on.
4418 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4420 RANGE is how far to scan while trying to match. RANGE = 0 means try
4421 only at STARTPOS; in general, the last start tried is STARTPOS +
4424 In REGS, return the indices of the virtual concatenation of STRING1
4425 and STRING2 that matched the entire BUFP->buffer and its contained
4428 Do not consider matching one past the index STOP in the virtual
4429 concatenation of STRING1 and STRING2.
4431 We return either the position in the strings at which the match was
4432 found, -1 if no match, or -2 if error (such as failure
4436 re_search_2 (bufp
, str1
, size1
, str2
, size2
, startpos
, range
, regs
, stop
)
4437 struct re_pattern_buffer
*bufp
;
4438 const char *str1
, *str2
;
4442 struct re_registers
*regs
;
4446 re_char
*string1
= (re_char
*) str1
;
4447 re_char
*string2
= (re_char
*) str2
;
4448 register char *fastmap
= bufp
->fastmap
;
4449 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4450 int total_size
= size1
+ size2
;
4451 int endpos
= startpos
+ range
;
4452 boolean anchored_start
;
4453 /* Nonzero if we are searching multibyte string. */
4454 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4456 /* Check for out-of-range STARTPOS. */
4457 if (startpos
< 0 || startpos
> total_size
)
4460 /* Fix up RANGE if it might eventually take us outside
4461 the virtual concatenation of STRING1 and STRING2.
4462 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4464 range
= 0 - startpos
;
4465 else if (endpos
> total_size
)
4466 range
= total_size
- startpos
;
4468 /* If the search isn't to be a backwards one, don't waste time in a
4469 search for a pattern anchored at beginning of buffer. */
4470 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4479 /* In a forward search for something that starts with \=.
4480 don't keep searching past point. */
4481 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4483 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4489 /* Update the fastmap now if not correct already. */
4490 if (fastmap
&& !bufp
->fastmap_accurate
)
4491 re_compile_fastmap (bufp
);
4493 /* See whether the pattern is anchored. */
4494 anchored_start
= (bufp
->buffer
[0] == begline
);
4497 gl_state
.object
= re_match_object
;
4499 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4501 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4505 /* Loop through the string, looking for a place to start matching. */
4508 /* If the pattern is anchored,
4509 skip quickly past places we cannot match.
4510 We don't bother to treat startpos == 0 specially
4511 because that case doesn't repeat. */
4512 if (anchored_start
&& startpos
> 0)
4514 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4515 : string2
[startpos
- size1
- 1])
4520 /* If a fastmap is supplied, skip quickly over characters that
4521 cannot be the start of a match. If the pattern can match the
4522 null string, however, we don't need to skip characters; we want
4523 the first null string. */
4524 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4526 register re_char
*d
;
4527 register re_wchar_t buf_ch
;
4529 d
= POS_ADDR_VSTRING (startpos
);
4531 if (range
> 0) /* Searching forwards. */
4533 register int lim
= 0;
4536 if (startpos
< size1
&& startpos
+ range
>= size1
)
4537 lim
= range
- (size1
- startpos
);
4539 /* Written out as an if-else to avoid testing `translate'
4541 if (RE_TRANSLATE_P (translate
))
4548 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4550 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4551 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4554 range
-= buf_charlen
;
4560 register re_wchar_t ch
, translated
;
4563 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4564 translated
= RE_TRANSLATE (translate
, ch
);
4565 if (translated
!= ch
4566 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4568 if (fastmap
[buf_ch
])
4581 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4583 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4585 range
-= buf_charlen
;
4589 while (range
> lim
&& !fastmap
[*d
])
4595 startpos
+= irange
- range
;
4597 else /* Searching backwards. */
4599 int room
= (startpos
>= size1
4600 ? size2
+ size1
- startpos
4601 : size1
- startpos
);
4604 buf_ch
= STRING_CHAR (d
, room
);
4605 buf_ch
= TRANSLATE (buf_ch
);
4606 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4611 register re_wchar_t ch
, translated
;
4614 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4615 translated
= TRANSLATE (ch
);
4616 if (translated
!= ch
4617 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4619 if (! fastmap
[TRANSLATE (buf_ch
)])
4625 /* If can't match the null string, and that's all we have left, fail. */
4626 if (range
>= 0 && startpos
== total_size
&& fastmap
4627 && !bufp
->can_be_null
)
4630 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4631 startpos
, regs
, stop
);
4644 /* Update STARTPOS to the next character boundary. */
4647 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4648 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4649 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
4667 /* Update STARTPOS to the previous character boundary. */
4670 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4672 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4674 /* Find the head of multibyte form. */
4675 PREV_CHAR_BOUNDARY (p
, phead
);
4676 range
+= p0
- 1 - p
;
4680 startpos
-= p0
- 1 - p
;
4686 WEAK_ALIAS (__re_search_2
, re_search_2
)
4688 /* Declarations and macros for re_match_2. */
4690 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4692 RE_TRANSLATE_TYPE translate
,
4693 const int multibyte
));
4695 /* This converts PTR, a pointer into one of the search strings `string1'
4696 and `string2' into an offset from the beginning of that string. */
4697 #define POINTER_TO_OFFSET(ptr) \
4698 (FIRST_STRING_P (ptr) \
4699 ? ((regoff_t) ((ptr) - string1)) \
4700 : ((regoff_t) ((ptr) - string2 + size1)))
4702 /* Call before fetching a character with *d. This switches over to
4703 string2 if necessary.
4704 Check re_match_2_internal for a discussion of why end_match_2 might
4705 not be within string2 (but be equal to end_match_1 instead). */
4706 #define PREFETCH() \
4709 /* End of string2 => fail. */ \
4710 if (dend == end_match_2) \
4712 /* End of string1 => advance to string2. */ \
4714 dend = end_match_2; \
4717 /* Call before fetching a char with *d if you already checked other limits.
4718 This is meant for use in lookahead operations like wordend, etc..
4719 where we might need to look at parts of the string that might be
4720 outside of the LIMITs (i.e past `stop'). */
4721 #define PREFETCH_NOLIMIT() \
4725 dend = end_match_2; \
4728 /* Test if at very beginning or at very end of the virtual concatenation
4729 of `string1' and `string2'. If only one string, it's `string2'. */
4730 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4731 #define AT_STRINGS_END(d) ((d) == end2)
4734 /* Test if D points to a character which is word-constituent. We have
4735 two special cases to check for: if past the end of string1, look at
4736 the first character in string2; and if before the beginning of
4737 string2, look at the last character in string1. */
4738 #define WORDCHAR_P(d) \
4739 (SYNTAX ((d) == end1 ? *string2 \
4740 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4743 /* Disabled due to a compiler bug -- see comment at case wordbound */
4745 /* The comment at case wordbound is following one, but we don't use
4746 AT_WORD_BOUNDARY anymore to support multibyte form.
4748 The DEC Alpha C compiler 3.x generates incorrect code for the
4749 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4750 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4751 macro and introducing temporary variables works around the bug. */
4754 /* Test if the character before D and the one at D differ with respect
4755 to being word-constituent. */
4756 #define AT_WORD_BOUNDARY(d) \
4757 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4758 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4761 /* Free everything we malloc. */
4762 #ifdef MATCH_MAY_ALLOCATE
4763 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4764 # define FREE_VARIABLES() \
4766 REGEX_FREE_STACK (fail_stack.stack); \
4767 FREE_VAR (regstart); \
4768 FREE_VAR (regend); \
4769 FREE_VAR (best_regstart); \
4770 FREE_VAR (best_regend); \
4773 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4774 #endif /* not MATCH_MAY_ALLOCATE */
4777 /* Optimization routines. */
4779 /* If the operation is a match against one or more chars,
4780 return a pointer to the next operation, else return NULL. */
4785 switch (SWITCH_ENUM_CAST (*p
++))
4796 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4799 p
= CHARSET_RANGE_TABLE (p
- 1);
4800 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4801 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4804 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4811 case notcategoryspec
:
4823 /* Jump over non-matching operations. */
4825 skip_noops (p
, pend
)
4831 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4840 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4851 /* Non-zero if "p1 matches something" implies "p2 fails". */
4853 mutually_exclusive_p (bufp
, p1
, p2
)
4854 struct re_pattern_buffer
*bufp
;
4858 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4859 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4861 assert (p1
>= bufp
->buffer
&& p1
< pend
4862 && p2
>= bufp
->buffer
&& p2
<= pend
);
4864 /* Skip over open/close-group commands.
4865 If what follows this loop is a ...+ construct,
4866 look at what begins its body, since we will have to
4867 match at least one of that. */
4868 p2
= skip_noops (p2
, pend
);
4869 /* The same skip can be done for p1, except that this function
4870 is only used in the case where p1 is a simple match operator. */
4871 /* p1 = skip_noops (p1, pend); */
4873 assert (p1
>= bufp
->buffer
&& p1
< pend
4874 && p2
>= bufp
->buffer
&& p2
<= pend
);
4876 op2
= p2
== pend
? succeed
: *p2
;
4878 switch (SWITCH_ENUM_CAST (op2
))
4882 /* If we're at the end of the pattern, we can change. */
4883 if (skip_one_char (p1
))
4885 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4893 register re_wchar_t c
4894 = (re_opcode_t
) *p2
== endline
? '\n'
4895 : RE_STRING_CHAR (p2
+ 2, pend
- p2
- 2, multibyte
);
4897 if ((re_opcode_t
) *p1
== exactn
)
4899 if (c
!= RE_STRING_CHAR (p1
+ 2, pend
- p1
- 2, multibyte
))
4901 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4906 else if ((re_opcode_t
) *p1
== charset
4907 || (re_opcode_t
) *p1
== charset_not
)
4909 int not = (re_opcode_t
) *p1
== charset_not
;
4911 /* Test if C is listed in charset (or charset_not)
4913 if (! multibyte
|| IS_REAL_ASCII (c
))
4915 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4916 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4919 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4920 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4922 /* `not' is equal to 1 if c would match, which means
4923 that we can't change to pop_failure_jump. */
4926 DEBUG_PRINT1 (" No match => fast loop.\n");
4930 else if ((re_opcode_t
) *p1
== anychar
4933 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4941 if ((re_opcode_t
) *p1
== exactn
)
4942 /* Reuse the code above. */
4943 return mutually_exclusive_p (bufp
, p2
, p1
);
4945 /* It is hard to list up all the character in charset
4946 P2 if it includes multibyte character. Give up in
4948 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4950 /* Now, we are sure that P2 has no range table.
4951 So, for the size of bitmap in P2, `p2[1]' is
4952 enough. But P1 may have range table, so the
4953 size of bitmap table of P1 is extracted by
4954 using macro `CHARSET_BITMAP_SIZE'.
4956 In a multibyte case, we know that all the character
4957 listed in P2 is ASCII. In a unibyte case, P1 has only a
4958 bitmap table. So, in both cases, it is enough to test
4959 only the bitmap table of P1. */
4961 if ((re_opcode_t
) *p1
== charset
)
4964 /* We win if the charset inside the loop
4965 has no overlap with the one after the loop. */
4968 && idx
< CHARSET_BITMAP_SIZE (p1
));
4970 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4974 || idx
== CHARSET_BITMAP_SIZE (p1
))
4976 DEBUG_PRINT1 (" No match => fast loop.\n");
4980 else if ((re_opcode_t
) *p1
== charset_not
)
4983 /* We win if the charset_not inside the loop lists
4984 every character listed in the charset after. */
4985 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4986 if (! (p2
[2 + idx
] == 0
4987 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4988 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4993 DEBUG_PRINT1 (" No match => fast loop.\n");
5002 switch (SWITCH_ENUM_CAST (*p1
))
5006 /* Reuse the code above. */
5007 return mutually_exclusive_p (bufp
, p2
, p1
);
5009 /* When we have two charset_not, it's very unlikely that
5010 they don't overlap. The union of the two sets of excluded
5011 chars should cover all possible chars, which, as a matter of
5012 fact, is virtually impossible in multibyte buffers. */
5018 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
5020 return ((re_opcode_t
) *p1
== syntaxspec
5021 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
5023 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
5026 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
5028 return ((re_opcode_t
) *p1
== notsyntaxspec
5029 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
5031 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
5034 return (((re_opcode_t
) *p1
== notsyntaxspec
5035 || (re_opcode_t
) *p1
== syntaxspec
)
5040 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
5041 case notcategoryspec
:
5042 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
5054 /* Matching routines. */
5056 #ifndef emacs /* Emacs never uses this. */
5057 /* re_match is like re_match_2 except it takes only a single string. */
5060 re_match (bufp
, string
, size
, pos
, regs
)
5061 struct re_pattern_buffer
*bufp
;
5064 struct re_registers
*regs
;
5066 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
5070 WEAK_ALIAS (__re_match
, re_match
)
5071 #endif /* not emacs */
5074 /* In Emacs, this is the string or buffer in which we
5075 are matching. It is used for looking up syntax properties. */
5076 Lisp_Object re_match_object
;
5079 /* re_match_2 matches the compiled pattern in BUFP against the
5080 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5081 and SIZE2, respectively). We start matching at POS, and stop
5084 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5085 store offsets for the substring each group matched in REGS. See the
5086 documentation for exactly how many groups we fill.
5088 We return -1 if no match, -2 if an internal error (such as the
5089 failure stack overflowing). Otherwise, we return the length of the
5090 matched substring. */
5093 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
5094 struct re_pattern_buffer
*bufp
;
5095 const char *string1
, *string2
;
5098 struct re_registers
*regs
;
5105 gl_state
.object
= re_match_object
;
5106 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
5107 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
5110 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
5111 (re_char
*) string2
, size2
,
5115 WEAK_ALIAS (__re_match_2
, re_match_2
)
5118 /* This is a separate function so that we can force an alloca cleanup
5121 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
5122 struct re_pattern_buffer
*bufp
;
5123 re_char
*string1
, *string2
;
5126 struct re_registers
*regs
;
5129 /* General temporaries. */
5134 /* Just past the end of the corresponding string. */
5135 re_char
*end1
, *end2
;
5137 /* Pointers into string1 and string2, just past the last characters in
5138 each to consider matching. */
5139 re_char
*end_match_1
, *end_match_2
;
5141 /* Where we are in the data, and the end of the current string. */
5144 /* Used sometimes to remember where we were before starting matching
5145 an operator so that we can go back in case of failure. This "atomic"
5146 behavior of matching opcodes is indispensable to the correctness
5147 of the on_failure_keep_string_jump optimization. */
5150 /* Where we are in the pattern, and the end of the pattern. */
5151 re_char
*p
= bufp
->buffer
;
5152 re_char
*pend
= p
+ bufp
->used
;
5154 /* We use this to map every character in the string. */
5155 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5157 /* Nonzero if BUFP is setup from a multibyte regex. */
5158 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
5160 /* Nonzero if STRING1/STRING2 are multibyte. */
5161 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
5163 /* Failure point stack. Each place that can handle a failure further
5164 down the line pushes a failure point on this stack. It consists of
5165 regstart, and regend for all registers corresponding to
5166 the subexpressions we're currently inside, plus the number of such
5167 registers, and, finally, two char *'s. The first char * is where
5168 to resume scanning the pattern; the second one is where to resume
5169 scanning the strings. */
5170 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5171 fail_stack_type fail_stack
;
5174 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5177 #if defined REL_ALLOC && defined REGEX_MALLOC
5178 /* This holds the pointer to the failure stack, when
5179 it is allocated relocatably. */
5180 fail_stack_elt_t
*failure_stack_ptr
;
5183 /* We fill all the registers internally, independent of what we
5184 return, for use in backreferences. The number here includes
5185 an element for register zero. */
5186 size_t num_regs
= bufp
->re_nsub
+ 1;
5188 /* Information on the contents of registers. These are pointers into
5189 the input strings; they record just what was matched (on this
5190 attempt) by a subexpression part of the pattern, that is, the
5191 regnum-th regstart pointer points to where in the pattern we began
5192 matching and the regnum-th regend points to right after where we
5193 stopped matching the regnum-th subexpression. (The zeroth register
5194 keeps track of what the whole pattern matches.) */
5195 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5196 re_char
**regstart
, **regend
;
5199 /* The following record the register info as found in the above
5200 variables when we find a match better than any we've seen before.
5201 This happens as we backtrack through the failure points, which in
5202 turn happens only if we have not yet matched the entire string. */
5203 unsigned best_regs_set
= false;
5204 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5205 re_char
**best_regstart
, **best_regend
;
5208 /* Logically, this is `best_regend[0]'. But we don't want to have to
5209 allocate space for that if we're not allocating space for anything
5210 else (see below). Also, we never need info about register 0 for
5211 any of the other register vectors, and it seems rather a kludge to
5212 treat `best_regend' differently than the rest. So we keep track of
5213 the end of the best match so far in a separate variable. We
5214 initialize this to NULL so that when we backtrack the first time
5215 and need to test it, it's not garbage. */
5216 re_char
*match_end
= NULL
;
5219 /* Counts the total number of registers pushed. */
5220 unsigned num_regs_pushed
= 0;
5223 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5227 #ifdef MATCH_MAY_ALLOCATE
5228 /* Do not bother to initialize all the register variables if there are
5229 no groups in the pattern, as it takes a fair amount of time. If
5230 there are groups, we include space for register 0 (the whole
5231 pattern), even though we never use it, since it simplifies the
5232 array indexing. We should fix this. */
5235 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5236 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5237 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5238 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5240 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5248 /* We must initialize all our variables to NULL, so that
5249 `FREE_VARIABLES' doesn't try to free them. */
5250 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5252 #endif /* MATCH_MAY_ALLOCATE */
5254 /* The starting position is bogus. */
5255 if (pos
< 0 || pos
> size1
+ size2
)
5261 /* Initialize subexpression text positions to -1 to mark ones that no
5262 start_memory/stop_memory has been seen for. Also initialize the
5263 register information struct. */
5264 for (reg
= 1; reg
< num_regs
; reg
++)
5265 regstart
[reg
] = regend
[reg
] = NULL
;
5267 /* We move `string1' into `string2' if the latter's empty -- but not if
5268 `string1' is null. */
5269 if (size2
== 0 && string1
!= NULL
)
5276 end1
= string1
+ size1
;
5277 end2
= string2
+ size2
;
5279 /* `p' scans through the pattern as `d' scans through the data.
5280 `dend' is the end of the input string that `d' points within. `d'
5281 is advanced into the following input string whenever necessary, but
5282 this happens before fetching; therefore, at the beginning of the
5283 loop, `d' can be pointing at the end of a string, but it cannot
5287 /* Only match within string2. */
5288 d
= string2
+ pos
- size1
;
5289 dend
= end_match_2
= string2
+ stop
- size1
;
5290 end_match_1
= end1
; /* Just to give it a value. */
5296 /* Only match within string1. */
5297 end_match_1
= string1
+ stop
;
5299 When we reach end_match_1, PREFETCH normally switches to string2.
5300 But in the present case, this means that just doing a PREFETCH
5301 makes us jump from `stop' to `gap' within the string.
5302 What we really want here is for the search to stop as
5303 soon as we hit end_match_1. That's why we set end_match_2
5304 to end_match_1 (since PREFETCH fails as soon as we hit
5306 end_match_2
= end_match_1
;
5309 { /* It's important to use this code when stop == size so that
5310 moving `d' from end1 to string2 will not prevent the d == dend
5311 check from catching the end of string. */
5313 end_match_2
= string2
+ stop
- size1
;
5319 DEBUG_PRINT1 ("The compiled pattern is: ");
5320 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5321 DEBUG_PRINT1 ("The string to match is: `");
5322 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5323 DEBUG_PRINT1 ("'\n");
5325 /* This loops over pattern commands. It exits by returning from the
5326 function if the match is complete, or it drops through if the match
5327 fails at this starting point in the input data. */
5330 DEBUG_PRINT2 ("\n%p: ", p
);
5333 { /* End of pattern means we might have succeeded. */
5334 DEBUG_PRINT1 ("end of pattern ... ");
5336 /* If we haven't matched the entire string, and we want the
5337 longest match, try backtracking. */
5338 if (d
!= end_match_2
)
5340 /* 1 if this match ends in the same string (string1 or string2)
5341 as the best previous match. */
5342 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5343 == FIRST_STRING_P (d
));
5344 /* 1 if this match is the best seen so far. */
5345 boolean best_match_p
;
5347 /* AIX compiler got confused when this was combined
5348 with the previous declaration. */
5350 best_match_p
= d
> match_end
;
5352 best_match_p
= !FIRST_STRING_P (d
);
5354 DEBUG_PRINT1 ("backtracking.\n");
5356 if (!FAIL_STACK_EMPTY ())
5357 { /* More failure points to try. */
5359 /* If exceeds best match so far, save it. */
5360 if (!best_regs_set
|| best_match_p
)
5362 best_regs_set
= true;
5365 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5367 for (reg
= 1; reg
< num_regs
; reg
++)
5369 best_regstart
[reg
] = regstart
[reg
];
5370 best_regend
[reg
] = regend
[reg
];
5376 /* If no failure points, don't restore garbage. And if
5377 last match is real best match, don't restore second
5379 else if (best_regs_set
&& !best_match_p
)
5382 /* Restore best match. It may happen that `dend ==
5383 end_match_1' while the restored d is in string2.
5384 For example, the pattern `x.*y.*z' against the
5385 strings `x-' and `y-z-', if the two strings are
5386 not consecutive in memory. */
5387 DEBUG_PRINT1 ("Restoring best registers.\n");
5390 dend
= ((d
>= string1
&& d
<= end1
)
5391 ? end_match_1
: end_match_2
);
5393 for (reg
= 1; reg
< num_regs
; reg
++)
5395 regstart
[reg
] = best_regstart
[reg
];
5396 regend
[reg
] = best_regend
[reg
];
5399 } /* d != end_match_2 */
5402 DEBUG_PRINT1 ("Accepting match.\n");
5404 /* If caller wants register contents data back, do it. */
5405 if (regs
&& !bufp
->no_sub
)
5407 /* Have the register data arrays been allocated? */
5408 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5409 { /* No. So allocate them with malloc. We need one
5410 extra element beyond `num_regs' for the `-1' marker
5412 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5413 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5414 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5415 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5420 bufp
->regs_allocated
= REGS_REALLOCATE
;
5422 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5423 { /* Yes. If we need more elements than were already
5424 allocated, reallocate them. If we need fewer, just
5426 if (regs
->num_regs
< num_regs
+ 1)
5428 regs
->num_regs
= num_regs
+ 1;
5429 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5430 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5431 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5440 /* These braces fend off a "empty body in an else-statement"
5441 warning under GCC when assert expands to nothing. */
5442 assert (bufp
->regs_allocated
== REGS_FIXED
);
5445 /* Convert the pointer data in `regstart' and `regend' to
5446 indices. Register zero has to be set differently,
5447 since we haven't kept track of any info for it. */
5448 if (regs
->num_regs
> 0)
5450 regs
->start
[0] = pos
;
5451 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5454 /* Go through the first `min (num_regs, regs->num_regs)'
5455 registers, since that is all we initialized. */
5456 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5458 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5459 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5463 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5465 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5469 /* If the regs structure we return has more elements than
5470 were in the pattern, set the extra elements to -1. If
5471 we (re)allocated the registers, this is the case,
5472 because we always allocate enough to have at least one
5474 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5475 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5476 } /* regs && !bufp->no_sub */
5478 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5479 nfailure_points_pushed
, nfailure_points_popped
,
5480 nfailure_points_pushed
- nfailure_points_popped
);
5481 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5483 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5485 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5491 /* Otherwise match next pattern command. */
5492 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5494 /* Ignore these. Used to ignore the n of succeed_n's which
5495 currently have n == 0. */
5497 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5501 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5504 /* Match the next n pattern characters exactly. The following
5505 byte in the pattern defines n, and the n bytes after that
5506 are the characters to match. */
5509 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5511 /* Remember the start point to rollback upon failure. */
5515 /* This is written out as an if-else so we don't waste time
5516 testing `translate' inside the loop. */
5517 if (RE_TRANSLATE_P (translate
))
5521 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5541 /* The cost of testing `translate' is comparatively small. */
5542 if (target_multibyte
)
5545 int pat_charlen
, buf_charlen
;
5550 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
5553 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5556 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5558 if (TRANSLATE (buf_ch
) != pat_ch
)
5566 mcnt
-= pat_charlen
;
5572 int pat_charlen
, buf_charlen
;
5578 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
5579 if (CHAR_BYTE8_P (pat_ch
))
5580 pat_ch
= CHAR_TO_BYTE8 (pat_ch
);
5582 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5589 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5590 if (! CHAR_BYTE8_P (buf_ch
))
5592 buf_ch
= TRANSLATE (buf_ch
);
5593 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5597 if (buf_ch
!= pat_ch
)
5610 /* Match any character except possibly a newline or a null. */
5616 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5619 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
,
5621 buf_ch
= TRANSLATE (buf_ch
);
5623 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5625 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5626 && buf_ch
== '\000'))
5629 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5638 register unsigned int c
;
5639 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5642 /* Start of actual range_table, or end of bitmap if there is no
5644 re_char
*range_table
;
5646 /* Nonzero if there is a range table. */
5647 int range_table_exists
;
5649 /* Number of ranges of range table. This is not included
5650 in the initial byte-length of the command. */
5653 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5655 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5657 if (range_table_exists
)
5659 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5660 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5664 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
, target_multibyte
);
5665 if (target_multibyte
)
5670 c1
= RE_CHAR_TO_UNIBYTE (c
);
5676 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5678 if (! CHAR_BYTE8_P (c1
))
5680 c1
= TRANSLATE (c1
);
5681 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5687 if (c
< (1 << BYTEWIDTH
))
5688 { /* Lookup bitmap. */
5689 /* Cast to `unsigned' instead of `unsigned char' in
5690 case the bit list is a full 32 bytes long. */
5691 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5692 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5696 else if (range_table_exists
)
5698 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5700 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5701 | (class_bits
& BIT_MULTIBYTE
)
5702 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5703 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5704 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5705 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5708 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5712 if (range_table_exists
)
5713 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5715 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5717 if (!not) goto fail
;
5724 /* The beginning of a group is represented by start_memory.
5725 The argument is the register number. The text
5726 matched within the group is recorded (in the internal
5727 registers data structure) under the register number. */
5729 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5731 /* In case we need to undo this operation (via backtracking). */
5732 PUSH_FAILURE_REG ((unsigned int)*p
);
5735 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5736 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5738 /* Move past the register number and inner group count. */
5743 /* The stop_memory opcode represents the end of a group. Its
5744 argument is the same as start_memory's: the register number. */
5746 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5748 assert (!REG_UNSET (regstart
[*p
]));
5749 /* Strictly speaking, there should be code such as:
5751 assert (REG_UNSET (regend[*p]));
5752 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5754 But the only info to be pushed is regend[*p] and it is known to
5755 be UNSET, so there really isn't anything to push.
5756 Not pushing anything, on the other hand deprives us from the
5757 guarantee that regend[*p] is UNSET since undoing this operation
5758 will not reset its value properly. This is not important since
5759 the value will only be read on the next start_memory or at
5760 the very end and both events can only happen if this stop_memory
5764 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5766 /* Move past the register number and the inner group count. */
5771 /* \<digit> has been turned into a `duplicate' command which is
5772 followed by the numeric value of <digit> as the register number. */
5775 register re_char
*d2
, *dend2
;
5776 int regno
= *p
++; /* Get which register to match against. */
5777 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5779 /* Can't back reference a group which we've never matched. */
5780 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5783 /* Where in input to try to start matching. */
5784 d2
= regstart
[regno
];
5786 /* Remember the start point to rollback upon failure. */
5789 /* Where to stop matching; if both the place to start and
5790 the place to stop matching are in the same string, then
5791 set to the place to stop, otherwise, for now have to use
5792 the end of the first string. */
5794 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5795 == FIRST_STRING_P (regend
[regno
]))
5796 ? regend
[regno
] : end_match_1
);
5799 /* If necessary, advance to next segment in register
5803 if (dend2
== end_match_2
) break;
5804 if (dend2
== regend
[regno
]) break;
5806 /* End of string1 => advance to string2. */
5808 dend2
= regend
[regno
];
5810 /* At end of register contents => success */
5811 if (d2
== dend2
) break;
5813 /* If necessary, advance to next segment in data. */
5816 /* How many characters left in this segment to match. */
5819 /* Want how many consecutive characters we can match in
5820 one shot, so, if necessary, adjust the count. */
5821 if (mcnt
> dend2
- d2
)
5824 /* Compare that many; failure if mismatch, else move
5826 if (RE_TRANSLATE_P (translate
)
5827 ? bcmp_translate (d
, d2
, mcnt
, translate
, target_multibyte
)
5828 : memcmp (d
, d2
, mcnt
))
5833 d
+= mcnt
, d2
+= mcnt
;
5839 /* begline matches the empty string at the beginning of the string
5840 (unless `not_bol' is set in `bufp'), and after newlines. */
5842 DEBUG_PRINT1 ("EXECUTING begline.\n");
5844 if (AT_STRINGS_BEG (d
))
5846 if (!bufp
->not_bol
) break;
5851 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5855 /* In all other cases, we fail. */
5859 /* endline is the dual of begline. */
5861 DEBUG_PRINT1 ("EXECUTING endline.\n");
5863 if (AT_STRINGS_END (d
))
5865 if (!bufp
->not_eol
) break;
5869 PREFETCH_NOLIMIT ();
5876 /* Match at the very beginning of the data. */
5878 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5879 if (AT_STRINGS_BEG (d
))
5884 /* Match at the very end of the data. */
5886 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5887 if (AT_STRINGS_END (d
))
5892 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5893 pushes NULL as the value for the string on the stack. Then
5894 `POP_FAILURE_POINT' will keep the current value for the
5895 string, instead of restoring it. To see why, consider
5896 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5897 then the . fails against the \n. But the next thing we want
5898 to do is match the \n against the \n; if we restored the
5899 string value, we would be back at the foo.
5901 Because this is used only in specific cases, we don't need to
5902 check all the things that `on_failure_jump' does, to make
5903 sure the right things get saved on the stack. Hence we don't
5904 share its code. The only reason to push anything on the
5905 stack at all is that otherwise we would have to change
5906 `anychar's code to do something besides goto fail in this
5907 case; that seems worse than this. */
5908 case on_failure_keep_string_jump
:
5909 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5910 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5913 PUSH_FAILURE_POINT (p
- 3, NULL
);
5916 /* A nasty loop is introduced by the non-greedy *? and +?.
5917 With such loops, the stack only ever contains one failure point
5918 at a time, so that a plain on_failure_jump_loop kind of
5919 cycle detection cannot work. Worse yet, such a detection
5920 can not only fail to detect a cycle, but it can also wrongly
5921 detect a cycle (between different instantiations of the same
5923 So the method used for those nasty loops is a little different:
5924 We use a special cycle-detection-stack-frame which is pushed
5925 when the on_failure_jump_nastyloop failure-point is *popped*.
5926 This special frame thus marks the beginning of one iteration
5927 through the loop and we can hence easily check right here
5928 whether something matched between the beginning and the end of
5930 case on_failure_jump_nastyloop
:
5931 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5932 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5935 assert ((re_opcode_t
)p
[-4] == no_op
);
5938 CHECK_INFINITE_LOOP (p
- 4, d
);
5940 /* If there's a cycle, just continue without pushing
5941 this failure point. The failure point is the "try again"
5942 option, which shouldn't be tried.
5943 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5944 PUSH_FAILURE_POINT (p
- 3, d
);
5948 /* Simple loop detecting on_failure_jump: just check on the
5949 failure stack if the same spot was already hit earlier. */
5950 case on_failure_jump_loop
:
5952 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5953 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5957 CHECK_INFINITE_LOOP (p
- 3, d
);
5959 /* If there's a cycle, get out of the loop, as if the matching
5960 had failed. We used to just `goto fail' here, but that was
5961 aborting the search a bit too early: we want to keep the
5962 empty-loop-match and keep matching after the loop.
5963 We want (x?)*y\1z to match both xxyz and xxyxz. */
5966 PUSH_FAILURE_POINT (p
- 3, d
);
5971 /* Uses of on_failure_jump:
5973 Each alternative starts with an on_failure_jump that points
5974 to the beginning of the next alternative. Each alternative
5975 except the last ends with a jump that in effect jumps past
5976 the rest of the alternatives. (They really jump to the
5977 ending jump of the following alternative, because tensioning
5978 these jumps is a hassle.)
5980 Repeats start with an on_failure_jump that points past both
5981 the repetition text and either the following jump or
5982 pop_failure_jump back to this on_failure_jump. */
5983 case on_failure_jump
:
5984 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5985 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5988 PUSH_FAILURE_POINT (p
-3, d
);
5991 /* This operation is used for greedy *.
5992 Compare the beginning of the repeat with what in the
5993 pattern follows its end. If we can establish that there
5994 is nothing that they would both match, i.e., that we
5995 would have to backtrack because of (as in, e.g., `a*a')
5996 then we can use a non-backtracking loop based on
5997 on_failure_keep_string_jump instead of on_failure_jump. */
5998 case on_failure_jump_smart
:
5999 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6000 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
6003 re_char
*p1
= p
; /* Next operation. */
6004 /* Here, we discard `const', making re_match non-reentrant. */
6005 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
6006 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
6008 p
-= 3; /* Reset so that we will re-execute the
6009 instruction once it's been changed. */
6011 EXTRACT_NUMBER (mcnt
, p2
- 2);
6013 /* Ensure this is a indeed the trivial kind of loop
6014 we are expecting. */
6015 assert (skip_one_char (p1
) == p2
- 3);
6016 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
6017 DEBUG_STATEMENT (debug
+= 2);
6018 if (mutually_exclusive_p (bufp
, p1
, p2
))
6020 /* Use a fast `on_failure_keep_string_jump' loop. */
6021 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
6022 *p3
= (unsigned char) on_failure_keep_string_jump
;
6023 STORE_NUMBER (p2
- 2, mcnt
+ 3);
6027 /* Default to a safe `on_failure_jump' loop. */
6028 DEBUG_PRINT1 (" smart default => slow loop.\n");
6029 *p3
= (unsigned char) on_failure_jump
;
6031 DEBUG_STATEMENT (debug
-= 2);
6035 /* Unconditionally jump (without popping any failure points). */
6038 IMMEDIATE_QUIT_CHECK
;
6039 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
6040 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
6041 p
+= mcnt
; /* Do the jump. */
6042 DEBUG_PRINT2 ("(to %p).\n", p
);
6046 /* Have to succeed matching what follows at least n times.
6047 After that, handle like `on_failure_jump'. */
6049 /* Signedness doesn't matter since we only compare MCNT to 0. */
6050 EXTRACT_NUMBER (mcnt
, p
+ 2);
6051 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
6053 /* Originally, mcnt is how many times we HAVE to succeed. */
6056 /* Here, we discard `const', making re_match non-reentrant. */
6057 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
6060 PUSH_NUMBER (p2
, mcnt
);
6063 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
6068 /* Signedness doesn't matter since we only compare MCNT to 0. */
6069 EXTRACT_NUMBER (mcnt
, p
+ 2);
6070 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
6072 /* Originally, this is how many times we CAN jump. */
6075 /* Here, we discard `const', making re_match non-reentrant. */
6076 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
6078 PUSH_NUMBER (p2
, mcnt
);
6079 goto unconditional_jump
;
6081 /* If don't have to jump any more, skip over the rest of command. */
6088 unsigned char *p2
; /* Location of the counter. */
6089 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
6091 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6092 /* Here, we discard `const', making re_match non-reentrant. */
6093 p2
= (unsigned char*) p
+ mcnt
;
6094 /* Signedness doesn't matter since we only copy MCNT's bits . */
6095 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6096 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
6097 PUSH_NUMBER (p2
, mcnt
);
6103 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
6104 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
6106 /* We SUCCEED (or FAIL) in one of the following cases: */
6108 /* Case 1: D is at the beginning or the end of string. */
6109 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
6113 /* C1 is the character before D, S1 is the syntax of C1, C2
6114 is the character at D, and S2 is the syntax of C2. */
6119 int offset
= PTR_TO_OFFSET (d
- 1);
6120 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6121 UPDATE_SYNTAX_TABLE (charpos
);
6123 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6126 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6128 PREFETCH_NOLIMIT ();
6129 GET_CHAR_AFTER (c2
, d
, dummy
);
6132 if (/* Case 2: Only one of S1 and S2 is Sword. */
6133 ((s1
== Sword
) != (s2
== Sword
))
6134 /* Case 3: Both of S1 and S2 are Sword, and macro
6135 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6136 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
6145 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6147 /* We FAIL in one of the following cases: */
6149 /* Case 1: D is at the end of string. */
6150 if (AT_STRINGS_END (d
))
6154 /* C1 is the character before D, S1 is the syntax of C1, C2
6155 is the character at D, and S2 is the syntax of C2. */
6160 int offset
= PTR_TO_OFFSET (d
);
6161 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6162 UPDATE_SYNTAX_TABLE (charpos
);
6165 GET_CHAR_AFTER (c2
, d
, dummy
);
6168 /* Case 2: S2 is not Sword. */
6172 /* Case 3: D is not at the beginning of string ... */
6173 if (!AT_STRINGS_BEG (d
))
6175 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6177 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6181 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6183 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6190 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6192 /* We FAIL in one of the following cases: */
6194 /* Case 1: D is at the beginning of string. */
6195 if (AT_STRINGS_BEG (d
))
6199 /* C1 is the character before D, S1 is the syntax of C1, C2
6200 is the character at D, and S2 is the syntax of C2. */
6205 int offset
= PTR_TO_OFFSET (d
) - 1;
6206 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6207 UPDATE_SYNTAX_TABLE (charpos
);
6209 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6212 /* Case 2: S1 is not Sword. */
6216 /* Case 3: D is not at the end of string ... */
6217 if (!AT_STRINGS_END (d
))
6219 PREFETCH_NOLIMIT ();
6220 GET_CHAR_AFTER (c2
, d
, dummy
);
6222 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6226 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6228 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6235 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6237 /* We FAIL in one of the following cases: */
6239 /* Case 1: D is at the end of string. */
6240 if (AT_STRINGS_END (d
))
6244 /* C1 is the character before D, S1 is the syntax of C1, C2
6245 is the character at D, and S2 is the syntax of C2. */
6249 int offset
= PTR_TO_OFFSET (d
);
6250 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6251 UPDATE_SYNTAX_TABLE (charpos
);
6254 c2
= RE_STRING_CHAR (d
, dend
- d
, target_multibyte
);
6257 /* Case 2: S2 is neither Sword nor Ssymbol. */
6258 if (s2
!= Sword
&& s2
!= Ssymbol
)
6261 /* Case 3: D is not at the beginning of string ... */
6262 if (!AT_STRINGS_BEG (d
))
6264 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6266 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6270 /* ... and S1 is Sword or Ssymbol. */
6271 if (s1
== Sword
|| s1
== Ssymbol
)
6278 DEBUG_PRINT1 ("EXECUTING symend.\n");
6280 /* We FAIL in one of the following cases: */
6282 /* Case 1: D is at the beginning of string. */
6283 if (AT_STRINGS_BEG (d
))
6287 /* C1 is the character before D, S1 is the syntax of C1, C2
6288 is the character at D, and S2 is the syntax of C2. */
6292 int offset
= PTR_TO_OFFSET (d
) - 1;
6293 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6294 UPDATE_SYNTAX_TABLE (charpos
);
6296 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6299 /* Case 2: S1 is neither Ssymbol nor Sword. */
6300 if (s1
!= Sword
&& s1
!= Ssymbol
)
6303 /* Case 3: D is not at the end of string ... */
6304 if (!AT_STRINGS_END (d
))
6306 PREFETCH_NOLIMIT ();
6307 c2
= RE_STRING_CHAR (d
, dend
- d
, target_multibyte
);
6309 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6313 /* ... and S2 is Sword or Ssymbol. */
6314 if (s2
== Sword
|| s2
== Ssymbol
)
6322 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6324 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
6328 int offset
= PTR_TO_OFFSET (d
);
6329 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6330 UPDATE_SYNTAX_TABLE (pos1
);
6337 GET_CHAR_AFTER (c
, d
, len
);
6338 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6346 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6347 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6352 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6353 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6358 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6359 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6364 case notcategoryspec
:
6365 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6367 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
6373 GET_CHAR_AFTER (c
, d
, len
);
6374 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6385 continue; /* Successfully executed one pattern command; keep going. */
6388 /* We goto here if a matching operation fails. */
6390 IMMEDIATE_QUIT_CHECK
;
6391 if (!FAIL_STACK_EMPTY ())
6394 /* A restart point is known. Restore to that state. */
6395 DEBUG_PRINT1 ("\nFAIL:\n");
6396 POP_FAILURE_POINT (str
, pat
);
6397 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
6399 case on_failure_keep_string_jump
:
6400 assert (str
== NULL
);
6401 goto continue_failure_jump
;
6403 case on_failure_jump_nastyloop
:
6404 assert ((re_opcode_t
)pat
[-2] == no_op
);
6405 PUSH_FAILURE_POINT (pat
- 2, str
);
6408 case on_failure_jump_loop
:
6409 case on_failure_jump
:
6412 continue_failure_jump
:
6413 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6418 /* A special frame used for nastyloops. */
6425 assert (p
>= bufp
->buffer
&& p
<= pend
);
6427 if (d
>= string1
&& d
<= end1
)
6431 break; /* Matching at this starting point really fails. */
6435 goto restore_best_regs
;
6439 return -1; /* Failure to match. */
6442 /* Subroutine definitions for re_match_2. */
6444 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6445 bytes; nonzero otherwise. */
6448 bcmp_translate (s1
, s2
, len
, translate
, target_multibyte
)
6451 RE_TRANSLATE_TYPE translate
;
6452 const int target_multibyte
;
6454 register re_char
*p1
= s1
, *p2
= s2
;
6455 re_char
*p1_end
= s1
+ len
;
6456 re_char
*p2_end
= s2
+ len
;
6458 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6459 different lengths, but relying on a single `len' would break this. -sm */
6460 while (p1
< p1_end
&& p2
< p2_end
)
6462 int p1_charlen
, p2_charlen
;
6463 re_wchar_t p1_ch
, p2_ch
;
6465 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6466 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6468 if (RE_TRANSLATE (translate
, p1_ch
)
6469 != RE_TRANSLATE (translate
, p2_ch
))
6472 p1
+= p1_charlen
, p2
+= p2_charlen
;
6475 if (p1
!= p1_end
|| p2
!= p2_end
)
6481 /* Entry points for GNU code. */
6483 /* re_compile_pattern is the GNU regular expression compiler: it
6484 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6485 Returns 0 if the pattern was valid, otherwise an error string.
6487 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6488 are set in BUFP on entry.
6490 We call regex_compile to do the actual compilation. */
6493 re_compile_pattern (pattern
, length
, bufp
)
6494 const char *pattern
;
6496 struct re_pattern_buffer
*bufp
;
6501 gl_state
.current_syntax_table
= current_buffer
->syntax_table
;
6504 /* GNU code is written to assume at least RE_NREGS registers will be set
6505 (and at least one extra will be -1). */
6506 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6508 /* And GNU code determines whether or not to get register information
6509 by passing null for the REGS argument to re_match, etc., not by
6513 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6517 return gettext (re_error_msgid
[(int) ret
]);
6519 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6521 /* Entry points compatible with 4.2 BSD regex library. We don't define
6522 them unless specifically requested. */
6524 #if defined _REGEX_RE_COMP || defined _LIBC
6526 /* BSD has one and only one pattern buffer. */
6527 static struct re_pattern_buffer re_comp_buf
;
6531 /* Make these definitions weak in libc, so POSIX programs can redefine
6532 these names if they don't use our functions, and still use
6533 regcomp/regexec below without link errors. */
6543 if (!re_comp_buf
.buffer
)
6544 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6545 return (char *) gettext ("No previous regular expression");
6549 if (!re_comp_buf
.buffer
)
6551 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6552 if (re_comp_buf
.buffer
== NULL
)
6553 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6554 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6555 re_comp_buf
.allocated
= 200;
6557 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6558 if (re_comp_buf
.fastmap
== NULL
)
6559 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6560 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6563 /* Since `re_exec' always passes NULL for the `regs' argument, we
6564 don't need to initialize the pattern buffer fields which affect it. */
6566 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6571 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6572 return (char *) gettext (re_error_msgid
[(int) ret
]);
6583 const int len
= strlen (s
);
6585 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6587 #endif /* _REGEX_RE_COMP */
6589 /* POSIX.2 functions. Don't define these for Emacs. */
6593 /* regcomp takes a regular expression as a string and compiles it.
6595 PREG is a regex_t *. We do not expect any fields to be initialized,
6596 since POSIX says we shouldn't. Thus, we set
6598 `buffer' to the compiled pattern;
6599 `used' to the length of the compiled pattern;
6600 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6601 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6602 RE_SYNTAX_POSIX_BASIC;
6603 `fastmap' to an allocated space for the fastmap;
6604 `fastmap_accurate' to zero;
6605 `re_nsub' to the number of subexpressions in PATTERN.
6607 PATTERN is the address of the pattern string.
6609 CFLAGS is a series of bits which affect compilation.
6611 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6612 use POSIX basic syntax.
6614 If REG_NEWLINE is set, then . and [^...] don't match newline.
6615 Also, regexec will try a match beginning after every newline.
6617 If REG_ICASE is set, then we considers upper- and lowercase
6618 versions of letters to be equivalent when matching.
6620 If REG_NOSUB is set, then when PREG is passed to regexec, that
6621 routine will report only success or failure, and nothing about the
6624 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6625 the return codes and their meanings.) */
6628 regcomp (preg
, pattern
, cflags
)
6629 regex_t
*__restrict preg
;
6630 const char *__restrict pattern
;
6635 = (cflags
& REG_EXTENDED
) ?
6636 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6638 /* regex_compile will allocate the space for the compiled pattern. */
6640 preg
->allocated
= 0;
6643 /* Try to allocate space for the fastmap. */
6644 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6646 if (cflags
& REG_ICASE
)
6651 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6652 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6653 if (preg
->translate
== NULL
)
6654 return (int) REG_ESPACE
;
6656 /* Map uppercase characters to corresponding lowercase ones. */
6657 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6658 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6661 preg
->translate
= NULL
;
6663 /* If REG_NEWLINE is set, newlines are treated differently. */
6664 if (cflags
& REG_NEWLINE
)
6665 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6666 syntax
&= ~RE_DOT_NEWLINE
;
6667 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6670 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6672 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6674 /* POSIX says a null character in the pattern terminates it, so we
6675 can use strlen here in compiling the pattern. */
6676 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6678 /* POSIX doesn't distinguish between an unmatched open-group and an
6679 unmatched close-group: both are REG_EPAREN. */
6680 if (ret
== REG_ERPAREN
)
6683 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6684 { /* Compute the fastmap now, since regexec cannot modify the pattern
6686 re_compile_fastmap (preg
);
6687 if (preg
->can_be_null
)
6688 { /* The fastmap can't be used anyway. */
6689 free (preg
->fastmap
);
6690 preg
->fastmap
= NULL
;
6695 WEAK_ALIAS (__regcomp
, regcomp
)
6698 /* regexec searches for a given pattern, specified by PREG, in the
6701 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6702 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6703 least NMATCH elements, and we set them to the offsets of the
6704 corresponding matched substrings.
6706 EFLAGS specifies `execution flags' which affect matching: if
6707 REG_NOTBOL is set, then ^ does not match at the beginning of the
6708 string; if REG_NOTEOL is set, then $ does not match at the end.
6710 We return 0 if we find a match and REG_NOMATCH if not. */
6713 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6714 const regex_t
*__restrict preg
;
6715 const char *__restrict string
;
6717 regmatch_t pmatch
[__restrict_arr
];
6721 struct re_registers regs
;
6722 regex_t private_preg
;
6723 int len
= strlen (string
);
6724 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6726 private_preg
= *preg
;
6728 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6729 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6731 /* The user has told us exactly how many registers to return
6732 information about, via `nmatch'. We have to pass that on to the
6733 matching routines. */
6734 private_preg
.regs_allocated
= REGS_FIXED
;
6738 regs
.num_regs
= nmatch
;
6739 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6740 if (regs
.start
== NULL
)
6741 return (int) REG_NOMATCH
;
6742 regs
.end
= regs
.start
+ nmatch
;
6745 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6746 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6747 was a little bit longer but still only matching the real part.
6748 This works because the `endline' will check for a '\n' and will find a
6749 '\0', correctly deciding that this is not the end of a line.
6750 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6751 a convenient '\0' there. For all we know, the string could be preceded
6752 by '\n' which would throw things off. */
6754 /* Perform the searching operation. */
6755 ret
= re_search (&private_preg
, string
, len
,
6756 /* start: */ 0, /* range: */ len
,
6757 want_reg_info
? ®s
: (struct re_registers
*) 0);
6759 /* Copy the register information to the POSIX structure. */
6766 for (r
= 0; r
< nmatch
; r
++)
6768 pmatch
[r
].rm_so
= regs
.start
[r
];
6769 pmatch
[r
].rm_eo
= regs
.end
[r
];
6773 /* If we needed the temporary register info, free the space now. */
6777 /* We want zero return to mean success, unlike `re_search'. */
6778 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6780 WEAK_ALIAS (__regexec
, regexec
)
6783 /* Returns a message corresponding to an error code, ERR_CODE, returned
6784 from either regcomp or regexec. We don't use PREG here.
6786 ERR_CODE was previously called ERRCODE, but that name causes an
6787 error with msvc8 compiler. */
6790 regerror (err_code
, preg
, errbuf
, errbuf_size
)
6792 const regex_t
*preg
;
6800 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6801 /* Only error codes returned by the rest of the code should be passed
6802 to this routine. If we are given anything else, or if other regex
6803 code generates an invalid error code, then the program has a bug.
6804 Dump core so we can fix it. */
6807 msg
= gettext (re_error_msgid
[err_code
]);
6809 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6811 if (errbuf_size
!= 0)
6813 if (msg_size
> errbuf_size
)
6815 strncpy (errbuf
, msg
, errbuf_size
- 1);
6816 errbuf
[errbuf_size
- 1] = 0;
6819 strcpy (errbuf
, msg
);
6824 WEAK_ALIAS (__regerror
, regerror
)
6827 /* Free dynamically allocated space used by PREG. */
6833 if (preg
->buffer
!= NULL
)
6834 free (preg
->buffer
);
6835 preg
->buffer
= NULL
;
6837 preg
->allocated
= 0;
6840 if (preg
->fastmap
!= NULL
)
6841 free (preg
->fastmap
);
6842 preg
->fastmap
= NULL
;
6843 preg
->fastmap_accurate
= 0;
6845 if (preg
->translate
!= NULL
)
6846 free (preg
->translate
);
6847 preg
->translate
= NULL
;
6849 WEAK_ALIAS (__regfree
, regfree
)
6851 #endif /* not emacs */
6853 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6854 (do not change this comment) */