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) 1993,94,95,96,97,98,99,2000 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
23 - structure the opcode space into opcode+flag.
24 - merge with glibc's regex.[ch].
25 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
26 need to modify the compiled regexp so that re_match can be reentrant.
27 - get rid of on_failure_jump_smart by doing the optimization in re_comp
28 rather than at run-time, so that re_match can be reentrant.
31 /* AIX requires this to be the first thing in the file. */
32 #if defined _AIX && !defined REGEX_MALLOC
40 #if defined STDC_HEADERS && !defined emacs
43 /* We need this for `regex.h', and perhaps for the Emacs include files. */
44 # include <sys/types.h>
47 /* Whether to use ISO C Amendment 1 wide char functions.
48 Those should not be used for Emacs since it uses its own. */
50 #define WIDE_CHAR_SUPPORT 1
52 #define WIDE_CHAR_SUPPORT \
53 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
56 /* For platform which support the ISO C amendement 1 functionality we
57 support user defined character classes. */
59 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
65 /* We have to keep the namespace clean. */
66 # define regfree(preg) __regfree (preg)
67 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
68 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
69 # define regerror(errcode, preg, errbuf, errbuf_size) \
70 __regerror(errcode, preg, errbuf, errbuf_size)
71 # define re_set_registers(bu, re, nu, st, en) \
72 __re_set_registers (bu, re, nu, st, en)
73 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
74 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
75 # define re_match(bufp, string, size, pos, regs) \
76 __re_match (bufp, string, size, pos, regs)
77 # define re_search(bufp, string, size, startpos, range, regs) \
78 __re_search (bufp, string, size, startpos, range, regs)
79 # define re_compile_pattern(pattern, length, bufp) \
80 __re_compile_pattern (pattern, length, bufp)
81 # define re_set_syntax(syntax) __re_set_syntax (syntax)
82 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
83 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
84 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
86 /* Make sure we call libc's function even if the user overrides them. */
87 # define btowc __btowc
88 # define iswctype __iswctype
89 # define wctype __wctype
91 # define WEAK_ALIAS(a,b) weak_alias (a, b)
93 /* We are also using some library internals. */
94 # include <locale/localeinfo.h>
95 # include <locale/elem-hash.h>
96 # include <langinfo.h>
98 # define WEAK_ALIAS(a,b)
101 /* This is for other GNU distributions with internationalized messages. */
102 #if HAVE_LIBINTL_H || defined _LIBC
103 # include <libintl.h>
105 # define gettext(msgid) (msgid)
109 /* This define is so xgettext can find the internationalizable
111 # define gettext_noop(String) String
114 /* The `emacs' switch turns on certain matching commands
115 that make sense only in Emacs. */
121 /* Make syntax table lookup grant data in gl_state. */
122 # define SYNTAX_ENTRY_VIA_PROPERTY
125 # include "character.h"
126 # include "category.h"
131 # define malloc xmalloc
135 # define realloc xrealloc
141 /* Converts the pointer to the char to BEG-based offset from the start. */
142 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
143 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
145 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
146 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
147 # define RE_STRING_CHAR(p, s) \
148 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
149 # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \
150 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
152 /* Set C a (possibly converted to multibyte) character before P. P
153 points into a string which is the virtual concatenation of STR1
154 (which ends at END1) or STR2 (which ends at END2). */
155 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
159 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
160 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
161 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
162 c = STRING_CHAR (dtemp, (p) - dtemp); \
166 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
167 MAKE_CHAR_MULTIBYTE (c); \
171 /* Set C a (possibly converted to multibyte) character at P, and set
172 LEN to the byte length of that character. */
173 # define GET_CHAR_AFTER(c, p, len) \
176 c = STRING_CHAR_AND_LENGTH (p, 0, len); \
181 MAKE_CHAR_MULTIBYTE (c); \
185 #else /* not emacs */
187 /* If we are not linking with Emacs proper,
188 we can't use the relocating allocator
189 even if config.h says that we can. */
192 # if defined STDC_HEADERS || defined _LIBC
199 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
200 If nothing else has been done, use the method below. */
201 # ifdef INHIBIT_STRING_HEADER
202 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
203 # if !defined bzero && !defined bcopy
204 # undef INHIBIT_STRING_HEADER
209 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
210 This is used in most programs--a few other programs avoid this
211 by defining INHIBIT_STRING_HEADER. */
212 # ifndef INHIBIT_STRING_HEADER
213 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
217 # define bzero(s, n) (memset (s, '\0', n), (s))
219 # define bzero(s, n) __bzero (s, n)
223 # include <strings.h>
225 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
228 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
233 /* Define the syntax stuff for \<, \>, etc. */
235 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
236 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1 };
238 # ifdef SWITCH_ENUM_BUG
239 # define SWITCH_ENUM_CAST(x) ((int)(x))
241 # define SWITCH_ENUM_CAST(x) (x)
244 /* Dummy macros for non-Emacs environments. */
245 # define BASE_LEADING_CODE_P(c) (0)
246 # define CHAR_CHARSET(c) 0
247 # define CHARSET_LEADING_CODE_BASE(c) 0
248 # define MAX_MULTIBYTE_LENGTH 1
249 # define RE_MULTIBYTE_P(x) 0
250 # define RE_TARGET_MULTIBYTE_P(x) 0
251 # define WORD_BOUNDARY_P(c1, c2) (0)
252 # define CHAR_HEAD_P(p) (1)
253 # define SINGLE_BYTE_CHAR_P(c) (1)
254 # define SAME_CHARSET_P(c1, c2) (1)
255 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
256 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
257 # define STRING_CHAR(p, s) (*(p))
258 # define RE_STRING_CHAR STRING_CHAR
259 # define CHAR_STRING(c, s) (*(s) = (c), 1)
260 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
261 # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH
262 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
263 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
264 # define GET_CHAR_AFTER(c, p, len) \
266 # define MAKE_CHAR(charset, c1, c2) (c1)
267 # define BYTE8_TO_CHAR(c) (c)
268 # define CHAR_BYTE8_P(c) (0)
269 # define MAKE_CHAR_MULTIBYTE(c) (c)
270 # define MAKE_CHAR_UNIBYTE(c) (c)
271 # define CHAR_LEADING_CODE(c) (c)
273 #endif /* not emacs */
276 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
277 # define RE_TRANSLATE_P(TBL) (TBL)
280 /* Get the interface, including the syntax bits. */
283 /* isalpha etc. are used for the character classes. */
288 /* 1 if C is an ASCII character. */
289 # define IS_REAL_ASCII(c) ((c) < 0200)
291 /* 1 if C is a unibyte character. */
292 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
294 /* The Emacs definitions should not be directly affected by locales. */
296 /* In Emacs, these are only used for single-byte characters. */
297 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
298 # define ISCNTRL(c) ((c) < ' ')
299 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
300 || ((c) >= 'a' && (c) <= 'f') \
301 || ((c) >= 'A' && (c) <= 'F'))
303 /* This is only used for single-byte characters. */
304 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
306 /* The rest must handle multibyte characters. */
308 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
309 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
312 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
313 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
316 # define ISALNUM(c) (IS_REAL_ASCII (c) \
317 ? (((c) >= 'a' && (c) <= 'z') \
318 || ((c) >= 'A' && (c) <= 'Z') \
319 || ((c) >= '0' && (c) <= '9')) \
320 : SYNTAX (c) == Sword)
322 # define ISALPHA(c) (IS_REAL_ASCII (c) \
323 ? (((c) >= 'a' && (c) <= 'z') \
324 || ((c) >= 'A' && (c) <= 'Z')) \
325 : SYNTAX (c) == Sword)
327 # define ISLOWER(c) (LOWERCASEP (c))
329 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
330 ? ((c) > ' ' && (c) < 0177 \
331 && !(((c) >= 'a' && (c) <= 'z') \
332 || ((c) >= 'A' && (c) <= 'Z') \
333 || ((c) >= '0' && (c) <= '9'))) \
334 : SYNTAX (c) != Sword)
336 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
338 # define ISUPPER(c) (UPPERCASEP (c))
340 # define ISWORD(c) (SYNTAX (c) == Sword)
342 #else /* not emacs */
344 /* Jim Meyering writes:
346 "... Some ctype macros are valid only for character codes that
347 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
348 using /bin/cc or gcc but without giving an ansi option). So, all
349 ctype uses should be through macros like ISPRINT... If
350 STDC_HEADERS is defined, then autoconf has verified that the ctype
351 macros don't need to be guarded with references to isascii. ...
352 Defining isascii to 1 should let any compiler worth its salt
353 eliminate the && through constant folding."
354 Solaris defines some of these symbols so we must undefine them first. */
357 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
358 # define ISASCII(c) 1
360 # define ISASCII(c) isascii(c)
363 /* 1 if C is an ASCII character. */
364 # define IS_REAL_ASCII(c) ((c) < 0200)
366 /* This distinction is not meaningful, except in Emacs. */
367 # define ISUNIBYTE(c) 1
370 # define ISBLANK(c) (ISASCII (c) && isblank (c))
372 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
375 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
377 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
381 # define ISPRINT(c) (ISASCII (c) && isprint (c))
382 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
383 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
384 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
385 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
386 # define ISLOWER(c) (ISASCII (c) && islower (c))
387 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
388 # define ISSPACE(c) (ISASCII (c) && isspace (c))
389 # define ISUPPER(c) (ISASCII (c) && isupper (c))
390 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
392 # define ISWORD(c) ISALPHA(c)
395 # define TOLOWER(c) _tolower(c)
397 # define TOLOWER(c) tolower(c)
400 /* How many characters in the character set. */
401 # define CHAR_SET_SIZE 256
405 extern char *re_syntax_table
;
407 # else /* not SYNTAX_TABLE */
409 static char re_syntax_table
[CHAR_SET_SIZE
];
420 bzero (re_syntax_table
, sizeof re_syntax_table
);
422 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
424 re_syntax_table
[c
] = Sword
;
426 re_syntax_table
['_'] = Sword
;
431 # endif /* not SYNTAX_TABLE */
433 # define SYNTAX(c) re_syntax_table[(c)]
435 #endif /* not emacs */
438 # define NULL (void *)0
441 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
442 since ours (we hope) works properly with all combinations of
443 machines, compilers, `char' and `unsigned char' argument types.
444 (Per Bothner suggested the basic approach.) */
445 #undef SIGN_EXTEND_CHAR
447 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
448 #else /* not __STDC__ */
449 /* As in Harbison and Steele. */
450 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
453 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
454 use `alloca' instead of `malloc'. This is because using malloc in
455 re_search* or re_match* could cause memory leaks when C-g is used in
456 Emacs; also, malloc is slower and causes storage fragmentation. On
457 the other hand, malloc is more portable, and easier to debug.
459 Because we sometimes use alloca, some routines have to be macros,
460 not functions -- `alloca'-allocated space disappears at the end of the
461 function it is called in. */
465 # define REGEX_ALLOCATE malloc
466 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
467 # define REGEX_FREE free
469 #else /* not REGEX_MALLOC */
471 /* Emacs already defines alloca, sometimes. */
474 /* Make alloca work the best possible way. */
476 # define alloca __builtin_alloca
477 # else /* not __GNUC__ */
478 # ifdef HAVE_ALLOCA_H
480 # endif /* HAVE_ALLOCA_H */
481 # endif /* not __GNUC__ */
483 # endif /* not alloca */
485 # define REGEX_ALLOCATE alloca
487 /* Assumes a `char *destination' variable. */
488 # define REGEX_REALLOCATE(source, osize, nsize) \
489 (destination = (char *) alloca (nsize), \
490 memcpy (destination, source, osize))
492 /* No need to do anything to free, after alloca. */
493 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
495 #endif /* not REGEX_MALLOC */
497 /* Define how to allocate the failure stack. */
499 #if defined REL_ALLOC && defined REGEX_MALLOC
501 # define REGEX_ALLOCATE_STACK(size) \
502 r_alloc (&failure_stack_ptr, (size))
503 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
504 r_re_alloc (&failure_stack_ptr, (nsize))
505 # define REGEX_FREE_STACK(ptr) \
506 r_alloc_free (&failure_stack_ptr)
508 #else /* not using relocating allocator */
512 # define REGEX_ALLOCATE_STACK malloc
513 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
514 # define REGEX_FREE_STACK free
516 # else /* not REGEX_MALLOC */
518 # define REGEX_ALLOCATE_STACK alloca
520 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
521 REGEX_REALLOCATE (source, osize, nsize)
522 /* No need to explicitly free anything. */
523 # define REGEX_FREE_STACK(arg) ((void)0)
525 # endif /* not REGEX_MALLOC */
526 #endif /* not using relocating allocator */
529 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
530 `string1' or just past its end. This works if PTR is NULL, which is
532 #define FIRST_STRING_P(ptr) \
533 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
535 /* (Re)Allocate N items of type T using malloc, or fail. */
536 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
537 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
538 #define RETALLOC_IF(addr, n, t) \
539 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
540 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
542 #define BYTEWIDTH 8 /* In bits. */
544 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
548 #define MAX(a, b) ((a) > (b) ? (a) : (b))
549 #define MIN(a, b) ((a) < (b) ? (a) : (b))
551 /* Type of source-pattern and string chars. */
552 typedef const unsigned char re_char
;
554 typedef char boolean
;
558 static int re_match_2_internal
_RE_ARGS ((struct re_pattern_buffer
*bufp
,
559 re_char
*string1
, int size1
,
560 re_char
*string2
, int size2
,
562 struct re_registers
*regs
,
565 /* These are the command codes that appear in compiled regular
566 expressions. Some opcodes are followed by argument bytes. A
567 command code can specify any interpretation whatsoever for its
568 arguments. Zero bytes may appear in the compiled regular expression. */
574 /* Succeed right away--no more backtracking. */
577 /* Followed by one byte giving n, then by n literal bytes. */
580 /* Matches any (more or less) character. */
583 /* Matches any one char belonging to specified set. First
584 following byte is number of bitmap bytes. Then come bytes
585 for a bitmap saying which chars are in. Bits in each byte
586 are ordered low-bit-first. A character is in the set if its
587 bit is 1. A character too large to have a bit in the map is
588 automatically not in the set.
590 If the length byte has the 0x80 bit set, then that stuff
591 is followed by a range table:
592 2 bytes of flags for character sets (low 8 bits, high 8 bits)
593 See RANGE_TABLE_WORK_BITS below.
594 2 bytes, the number of pairs that follow (upto 32767)
595 pairs, each 2 multibyte characters,
596 each multibyte character represented as 3 bytes. */
599 /* Same parameters as charset, but match any character that is
600 not one of those specified. */
603 /* Start remembering the text that is matched, for storing in a
604 register. Followed by one byte with the register number, in
605 the range 0 to one less than the pattern buffer's re_nsub
609 /* Stop remembering the text that is matched and store it in a
610 memory register. Followed by one byte with the register
611 number, in the range 0 to one less than `re_nsub' in the
615 /* Match a duplicate of something remembered. Followed by one
616 byte containing the register number. */
619 /* Fail unless at beginning of line. */
622 /* Fail unless at end of line. */
625 /* Succeeds if at beginning of buffer (if emacs) or at beginning
626 of string to be matched (if not). */
629 /* Analogously, for end of buffer/string. */
632 /* Followed by two byte relative address to which to jump. */
635 /* Followed by two-byte relative address of place to resume at
636 in case of failure. */
639 /* Like on_failure_jump, but pushes a placeholder instead of the
640 current string position when executed. */
641 on_failure_keep_string_jump
,
643 /* Just like `on_failure_jump', except that it checks that we
644 don't get stuck in an infinite loop (matching an empty string
646 on_failure_jump_loop
,
648 /* Just like `on_failure_jump_loop', except that it checks for
649 a different kind of loop (the kind that shows up with non-greedy
650 operators). This operation has to be immediately preceded
652 on_failure_jump_nastyloop
,
654 /* A smart `on_failure_jump' used for greedy * and + operators.
655 It analyses the loop before which it is put and if the
656 loop does not require backtracking, it changes itself to
657 `on_failure_keep_string_jump' and short-circuits the loop,
658 else it just defaults to changing itself into `on_failure_jump'.
659 It assumes that it is pointing to just past a `jump'. */
660 on_failure_jump_smart
,
662 /* Followed by two-byte relative address and two-byte number n.
663 After matching N times, jump to the address upon failure.
664 Does not work if N starts at 0: use on_failure_jump_loop
668 /* Followed by two-byte relative address, and two-byte number n.
669 Jump to the address N times, then fail. */
672 /* Set the following two-byte relative address to the
673 subsequent two-byte number. The address *includes* the two
677 wordbeg
, /* Succeeds if at word beginning. */
678 wordend
, /* Succeeds if at word end. */
680 wordbound
, /* Succeeds if at a word boundary. */
681 notwordbound
, /* Succeeds if not at a word boundary. */
683 /* Matches any character whose syntax is specified. Followed by
684 a byte which contains a syntax code, e.g., Sword. */
687 /* Matches any character whose syntax is not that specified. */
691 ,before_dot
, /* Succeeds if before point. */
692 at_dot
, /* Succeeds if at point. */
693 after_dot
, /* Succeeds if after point. */
695 /* Matches any character whose category-set contains the specified
696 category. The operator is followed by a byte which contains a
697 category code (mnemonic ASCII character). */
700 /* Matches any character whose category-set does not contain the
701 specified category. The operator is followed by a byte which
702 contains the category code (mnemonic ASCII character). */
707 /* Common operations on the compiled pattern. */
709 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
711 #define STORE_NUMBER(destination, number) \
713 (destination)[0] = (number) & 0377; \
714 (destination)[1] = (number) >> 8; \
717 /* Same as STORE_NUMBER, except increment DESTINATION to
718 the byte after where the number is stored. Therefore, DESTINATION
719 must be an lvalue. */
721 #define STORE_NUMBER_AND_INCR(destination, number) \
723 STORE_NUMBER (destination, number); \
724 (destination) += 2; \
727 /* Put into DESTINATION a number stored in two contiguous bytes starting
730 #define EXTRACT_NUMBER(destination, source) \
732 (destination) = *(source) & 0377; \
733 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
737 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
739 extract_number (dest
, source
)
743 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
744 *dest
= *source
& 0377;
748 # ifndef EXTRACT_MACROS /* To debug the macros. */
749 # undef EXTRACT_NUMBER
750 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
751 # endif /* not EXTRACT_MACROS */
755 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
756 SOURCE must be an lvalue. */
758 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
760 EXTRACT_NUMBER (destination, source); \
765 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
768 extract_number_and_incr (destination
, source
)
772 extract_number (destination
, *source
);
776 # ifndef EXTRACT_MACROS
777 # undef EXTRACT_NUMBER_AND_INCR
778 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
779 extract_number_and_incr (&dest, &src)
780 # endif /* not EXTRACT_MACROS */
784 /* Store a multibyte character in three contiguous bytes starting
785 DESTINATION, and increment DESTINATION to the byte after where the
786 character is stored. Therefore, DESTINATION must be an lvalue. */
788 #define STORE_CHARACTER_AND_INCR(destination, character) \
790 (destination)[0] = (character) & 0377; \
791 (destination)[1] = ((character) >> 8) & 0377; \
792 (destination)[2] = (character) >> 16; \
793 (destination) += 3; \
796 /* Put into DESTINATION a character stored in three contiguous bytes
797 starting at SOURCE. */
799 #define EXTRACT_CHARACTER(destination, source) \
801 (destination) = ((source)[0] \
802 | ((source)[1] << 8) \
803 | ((source)[2] << 16)); \
807 /* Macros for charset. */
809 /* Size of bitmap of charset P in bytes. P is a start of charset,
810 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
811 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
813 /* Nonzero if charset P has range table. */
814 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
816 /* Return the address of range table of charset P. But not the start
817 of table itself, but the before where the number of ranges is
818 stored. `2 +' means to skip re_opcode_t and size of bitmap,
819 and the 2 bytes of flags at the start of the range table. */
820 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
822 /* Extract the bit flags that start a range table. */
823 #define CHARSET_RANGE_TABLE_BITS(p) \
824 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
825 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
827 /* Test if C is listed in the bitmap of charset P. */
828 #define CHARSET_LOOKUP_BITMAP(p, c) \
829 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
830 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
832 /* Return the address of end of RANGE_TABLE. COUNT is number of
833 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
834 is start of range and end of range. `* 3' is size of each start
836 #define CHARSET_RANGE_TABLE_END(range_table, count) \
837 ((range_table) + (count) * 2 * 3)
839 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
840 COUNT is number of ranges in RANGE_TABLE. */
841 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
844 re_wchar_t range_start, range_end; \
846 re_char *range_table_end \
847 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
849 for (p = (range_table); p < range_table_end; p += 2 * 3) \
851 EXTRACT_CHARACTER (range_start, p); \
852 EXTRACT_CHARACTER (range_end, p + 3); \
854 if (range_start <= (c) && (c) <= range_end) \
863 /* Test if C is in range table of CHARSET. The flag NOT is negated if
864 C is listed in it. */
865 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
868 /* Number of ranges in range table. */ \
870 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
872 EXTRACT_NUMBER_AND_INCR (count, range_table); \
873 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
877 /* If DEBUG is defined, Regex prints many voluminous messages about what
878 it is doing (if the variable `debug' is nonzero). If linked with the
879 main program in `iregex.c', you can enter patterns and strings
880 interactively. And if linked with the main program in `main.c' and
881 the other test files, you can run the already-written tests. */
885 /* We use standard I/O for debugging. */
888 /* It is useful to test things that ``must'' be true when debugging. */
891 static int debug
= -100000;
893 # define DEBUG_STATEMENT(e) e
894 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
895 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
896 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
897 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
898 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
899 if (debug > 0) print_partial_compiled_pattern (s, e)
900 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
901 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
904 /* Print the fastmap in human-readable form. */
907 print_fastmap (fastmap
)
910 unsigned was_a_range
= 0;
913 while (i
< (1 << BYTEWIDTH
))
919 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
935 /* Print a compiled pattern string in human-readable form, starting at
936 the START pointer into it and ending just before the pointer END. */
939 print_partial_compiled_pattern (start
, end
)
949 fprintf (stderr
, "(null)\n");
953 /* Loop over pattern commands. */
956 fprintf (stderr
, "%d:\t", p
- start
);
958 switch ((re_opcode_t
) *p
++)
961 fprintf (stderr
, "/no_op");
965 fprintf (stderr
, "/succeed");
970 fprintf (stderr
, "/exactn/%d", mcnt
);
973 fprintf (stderr
, "/%c", *p
++);
979 fprintf (stderr
, "/start_memory/%d", *p
++);
983 fprintf (stderr
, "/stop_memory/%d", *p
++);
987 fprintf (stderr
, "/duplicate/%d", *p
++);
991 fprintf (stderr
, "/anychar");
997 register int c
, last
= -100;
998 register int in_range
= 0;
999 int length
= CHARSET_BITMAP_SIZE (p
- 1);
1000 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
1002 fprintf (stderr
, "/charset [%s",
1003 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
1005 assert (p
+ *p
< pend
);
1007 for (c
= 0; c
< 256; c
++)
1009 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
1011 /* Are we starting a range? */
1012 if (last
+ 1 == c
&& ! in_range
)
1014 fprintf (stderr
, "-");
1017 /* Have we broken a range? */
1018 else if (last
+ 1 != c
&& in_range
)
1020 fprintf (stderr
, "%c", last
);
1025 fprintf (stderr
, "%c", c
);
1031 fprintf (stderr
, "%c", last
);
1033 fprintf (stderr
, "]");
1037 if (has_range_table
)
1040 fprintf (stderr
, "has-range-table");
1042 /* ??? Should print the range table; for now, just skip it. */
1043 p
+= 2; /* skip range table bits */
1044 EXTRACT_NUMBER_AND_INCR (count
, p
);
1045 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1051 fprintf (stderr
, "/begline");
1055 fprintf (stderr
, "/endline");
1058 case on_failure_jump
:
1059 extract_number_and_incr (&mcnt
, &p
);
1060 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1063 case on_failure_keep_string_jump
:
1064 extract_number_and_incr (&mcnt
, &p
);
1065 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1068 case on_failure_jump_nastyloop
:
1069 extract_number_and_incr (&mcnt
, &p
);
1070 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1073 case on_failure_jump_loop
:
1074 extract_number_and_incr (&mcnt
, &p
);
1075 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1078 case on_failure_jump_smart
:
1079 extract_number_and_incr (&mcnt
, &p
);
1080 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1084 extract_number_and_incr (&mcnt
, &p
);
1085 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1089 extract_number_and_incr (&mcnt
, &p
);
1090 extract_number_and_incr (&mcnt2
, &p
);
1091 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1095 extract_number_and_incr (&mcnt
, &p
);
1096 extract_number_and_incr (&mcnt2
, &p
);
1097 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1101 extract_number_and_incr (&mcnt
, &p
);
1102 extract_number_and_incr (&mcnt2
, &p
);
1103 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1107 fprintf (stderr
, "/wordbound");
1111 fprintf (stderr
, "/notwordbound");
1115 fprintf (stderr
, "/wordbeg");
1119 fprintf (stderr
, "/wordend");
1122 fprintf (stderr
, "/syntaxspec");
1124 fprintf (stderr
, "/%d", mcnt
);
1128 fprintf (stderr
, "/notsyntaxspec");
1130 fprintf (stderr
, "/%d", mcnt
);
1135 fprintf (stderr
, "/before_dot");
1139 fprintf (stderr
, "/at_dot");
1143 fprintf (stderr
, "/after_dot");
1147 fprintf (stderr
, "/categoryspec");
1149 fprintf (stderr
, "/%d", mcnt
);
1152 case notcategoryspec
:
1153 fprintf (stderr
, "/notcategoryspec");
1155 fprintf (stderr
, "/%d", mcnt
);
1160 fprintf (stderr
, "/begbuf");
1164 fprintf (stderr
, "/endbuf");
1168 fprintf (stderr
, "?%d", *(p
-1));
1171 fprintf (stderr
, "\n");
1174 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1179 print_compiled_pattern (bufp
)
1180 struct re_pattern_buffer
*bufp
;
1182 re_char
*buffer
= bufp
->buffer
;
1184 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1185 printf ("%ld bytes used/%ld bytes allocated.\n",
1186 bufp
->used
, bufp
->allocated
);
1188 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1190 printf ("fastmap: ");
1191 print_fastmap (bufp
->fastmap
);
1194 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1195 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1196 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1197 printf ("no_sub: %d\t", bufp
->no_sub
);
1198 printf ("not_bol: %d\t", bufp
->not_bol
);
1199 printf ("not_eol: %d\t", bufp
->not_eol
);
1200 printf ("syntax: %lx\n", bufp
->syntax
);
1202 /* Perhaps we should print the translate table? */
1207 print_double_string (where
, string1
, size1
, string2
, size2
)
1220 if (FIRST_STRING_P (where
))
1222 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1223 putchar (string1
[this_char
]);
1228 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1229 putchar (string2
[this_char
]);
1233 #else /* not DEBUG */
1238 # define DEBUG_STATEMENT(e)
1239 # define DEBUG_PRINT1(x)
1240 # define DEBUG_PRINT2(x1, x2)
1241 # define DEBUG_PRINT3(x1, x2, x3)
1242 # define DEBUG_PRINT4(x1, x2, x3, x4)
1243 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1244 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1246 #endif /* not DEBUG */
1248 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1249 also be assigned to arbitrarily: each pattern buffer stores its own
1250 syntax, so it can be changed between regex compilations. */
1251 /* This has no initializer because initialized variables in Emacs
1252 become read-only after dumping. */
1253 reg_syntax_t re_syntax_options
;
1256 /* Specify the precise syntax of regexps for compilation. This provides
1257 for compatibility for various utilities which historically have
1258 different, incompatible syntaxes.
1260 The argument SYNTAX is a bit mask comprised of the various bits
1261 defined in regex.h. We return the old syntax. */
1264 re_set_syntax (syntax
)
1265 reg_syntax_t syntax
;
1267 reg_syntax_t ret
= re_syntax_options
;
1269 re_syntax_options
= syntax
;
1272 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1274 /* This table gives an error message for each of the error codes listed
1275 in regex.h. Obviously the order here has to be same as there.
1276 POSIX doesn't require that we do anything for REG_NOERROR,
1277 but why not be nice? */
1279 static const char *re_error_msgid
[] =
1281 gettext_noop ("Success"), /* REG_NOERROR */
1282 gettext_noop ("No match"), /* REG_NOMATCH */
1283 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1284 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1285 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1286 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1287 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1288 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1289 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1290 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1291 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1292 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1293 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1294 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1295 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1296 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1297 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1300 /* Avoiding alloca during matching, to placate r_alloc. */
1302 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1303 searching and matching functions should not call alloca. On some
1304 systems, alloca is implemented in terms of malloc, and if we're
1305 using the relocating allocator routines, then malloc could cause a
1306 relocation, which might (if the strings being searched are in the
1307 ralloc heap) shift the data out from underneath the regexp
1310 Here's another reason to avoid allocation: Emacs
1311 processes input from X in a signal handler; processing X input may
1312 call malloc; if input arrives while a matching routine is calling
1313 malloc, then we're scrod. But Emacs can't just block input while
1314 calling matching routines; then we don't notice interrupts when
1315 they come in. So, Emacs blocks input around all regexp calls
1316 except the matching calls, which it leaves unprotected, in the
1317 faith that they will not malloc. */
1319 /* Normally, this is fine. */
1320 #define MATCH_MAY_ALLOCATE
1322 /* When using GNU C, we are not REALLY using the C alloca, no matter
1323 what config.h may say. So don't take precautions for it. */
1328 /* The match routines may not allocate if (1) they would do it with malloc
1329 and (2) it's not safe for them to use malloc.
1330 Note that if REL_ALLOC is defined, matching would not use malloc for the
1331 failure stack, but we would still use it for the register vectors;
1332 so REL_ALLOC should not affect this. */
1333 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1334 # undef MATCH_MAY_ALLOCATE
1338 /* Failure stack declarations and macros; both re_compile_fastmap and
1339 re_match_2 use a failure stack. These have to be macros because of
1340 REGEX_ALLOCATE_STACK. */
1343 /* Approximate number of failure points for which to initially allocate space
1344 when matching. If this number is exceeded, we allocate more
1345 space, so it is not a hard limit. */
1346 #ifndef INIT_FAILURE_ALLOC
1347 # define INIT_FAILURE_ALLOC 20
1350 /* Roughly the maximum number of failure points on the stack. Would be
1351 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1352 This is a variable only so users of regex can assign to it; we never
1353 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1354 before using it, so it should probably be a byte-count instead. */
1355 # if defined MATCH_MAY_ALLOCATE
1356 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1357 whose default stack limit is 2mb. In order for a larger
1358 value to work reliably, you have to try to make it accord
1359 with the process stack limit. */
1360 size_t re_max_failures
= 40000;
1362 size_t re_max_failures
= 4000;
1365 union fail_stack_elt
1368 /* This should be the biggest `int' that's no bigger than a pointer. */
1372 typedef union fail_stack_elt fail_stack_elt_t
;
1376 fail_stack_elt_t
*stack
;
1378 size_t avail
; /* Offset of next open position. */
1379 size_t frame
; /* Offset of the cur constructed frame. */
1382 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1383 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1386 /* Define macros to initialize and free the failure stack.
1387 Do `return -2' if the alloc fails. */
1389 #ifdef MATCH_MAY_ALLOCATE
1390 # define INIT_FAIL_STACK() \
1392 fail_stack.stack = (fail_stack_elt_t *) \
1393 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1394 * sizeof (fail_stack_elt_t)); \
1396 if (fail_stack.stack == NULL) \
1399 fail_stack.size = INIT_FAILURE_ALLOC; \
1400 fail_stack.avail = 0; \
1401 fail_stack.frame = 0; \
1404 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1406 # define INIT_FAIL_STACK() \
1408 fail_stack.avail = 0; \
1409 fail_stack.frame = 0; \
1412 # define RESET_FAIL_STACK() ((void)0)
1416 /* Double the size of FAIL_STACK, up to a limit
1417 which allows approximately `re_max_failures' items.
1419 Return 1 if succeeds, and 0 if either ran out of memory
1420 allocating space for it or it was already too large.
1422 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1424 /* Factor to increase the failure stack size by
1425 when we increase it.
1426 This used to be 2, but 2 was too wasteful
1427 because the old discarded stacks added up to as much space
1428 were as ultimate, maximum-size stack. */
1429 #define FAIL_STACK_GROWTH_FACTOR 4
1431 #define GROW_FAIL_STACK(fail_stack) \
1432 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1433 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1435 : ((fail_stack).stack \
1436 = (fail_stack_elt_t *) \
1437 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1438 (fail_stack).size * sizeof (fail_stack_elt_t), \
1439 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1440 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1441 * FAIL_STACK_GROWTH_FACTOR))), \
1443 (fail_stack).stack == NULL \
1445 : ((fail_stack).size \
1446 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1447 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1448 * FAIL_STACK_GROWTH_FACTOR)) \
1449 / sizeof (fail_stack_elt_t)), \
1453 /* Push a pointer value onto the failure stack.
1454 Assumes the variable `fail_stack'. Probably should only
1455 be called from within `PUSH_FAILURE_POINT'. */
1456 #define PUSH_FAILURE_POINTER(item) \
1457 fail_stack.stack[fail_stack.avail++].pointer = (item)
1459 /* This pushes an integer-valued item onto the failure stack.
1460 Assumes the variable `fail_stack'. Probably should only
1461 be called from within `PUSH_FAILURE_POINT'. */
1462 #define PUSH_FAILURE_INT(item) \
1463 fail_stack.stack[fail_stack.avail++].integer = (item)
1465 /* Push a fail_stack_elt_t value onto the failure stack.
1466 Assumes the variable `fail_stack'. Probably should only
1467 be called from within `PUSH_FAILURE_POINT'. */
1468 #define PUSH_FAILURE_ELT(item) \
1469 fail_stack.stack[fail_stack.avail++] = (item)
1471 /* These three POP... operations complement the three PUSH... operations.
1472 All assume that `fail_stack' is nonempty. */
1473 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1474 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1475 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1477 /* Individual items aside from the registers. */
1478 #define NUM_NONREG_ITEMS 3
1480 /* Used to examine the stack (to detect infinite loops). */
1481 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1482 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1483 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1484 #define TOP_FAILURE_HANDLE() fail_stack.frame
1487 #define ENSURE_FAIL_STACK(space) \
1488 while (REMAINING_AVAIL_SLOTS <= space) { \
1489 if (!GROW_FAIL_STACK (fail_stack)) \
1491 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1492 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1495 /* Push register NUM onto the stack. */
1496 #define PUSH_FAILURE_REG(num) \
1498 char *destination; \
1499 ENSURE_FAIL_STACK(3); \
1500 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1501 num, regstart[num], regend[num]); \
1502 PUSH_FAILURE_POINTER (regstart[num]); \
1503 PUSH_FAILURE_POINTER (regend[num]); \
1504 PUSH_FAILURE_INT (num); \
1507 /* Change the counter's value to VAL, but make sure that it will
1508 be reset when backtracking. */
1509 #define PUSH_NUMBER(ptr,val) \
1511 char *destination; \
1513 ENSURE_FAIL_STACK(3); \
1514 EXTRACT_NUMBER (c, ptr); \
1515 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1516 PUSH_FAILURE_INT (c); \
1517 PUSH_FAILURE_POINTER (ptr); \
1518 PUSH_FAILURE_INT (-1); \
1519 STORE_NUMBER (ptr, val); \
1522 /* Pop a saved register off the stack. */
1523 #define POP_FAILURE_REG_OR_COUNT() \
1525 int reg = POP_FAILURE_INT (); \
1528 /* It's a counter. */ \
1529 /* Here, we discard `const', making re_match non-reentrant. */ \
1530 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1531 reg = POP_FAILURE_INT (); \
1532 STORE_NUMBER (ptr, reg); \
1533 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1537 regend[reg] = POP_FAILURE_POINTER (); \
1538 regstart[reg] = POP_FAILURE_POINTER (); \
1539 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1540 reg, regstart[reg], regend[reg]); \
1544 /* Check that we are not stuck in an infinite loop. */
1545 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1547 int failure = TOP_FAILURE_HANDLE (); \
1548 /* Check for infinite matching loops */ \
1549 while (failure > 0 \
1550 && (FAILURE_STR (failure) == string_place \
1551 || FAILURE_STR (failure) == NULL)) \
1553 assert (FAILURE_PAT (failure) >= bufp->buffer \
1554 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1555 if (FAILURE_PAT (failure) == pat_cur) \
1560 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1561 failure = NEXT_FAILURE_HANDLE(failure); \
1563 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1566 /* Push the information about the state we will need
1567 if we ever fail back to it.
1569 Requires variables fail_stack, regstart, regend and
1570 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1573 Does `return FAILURE_CODE' if runs out of memory. */
1575 #define PUSH_FAILURE_POINT(pattern, string_place) \
1577 char *destination; \
1578 /* Must be int, so when we don't save any registers, the arithmetic \
1579 of 0 + -1 isn't done as unsigned. */ \
1581 DEBUG_STATEMENT (nfailure_points_pushed++); \
1582 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1583 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1584 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1586 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1588 DEBUG_PRINT1 ("\n"); \
1590 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1591 PUSH_FAILURE_INT (fail_stack.frame); \
1593 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1594 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1595 DEBUG_PRINT1 ("'\n"); \
1596 PUSH_FAILURE_POINTER (string_place); \
1598 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1599 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1600 PUSH_FAILURE_POINTER (pattern); \
1602 /* Close the frame by moving the frame pointer past it. */ \
1603 fail_stack.frame = fail_stack.avail; \
1606 /* Estimate the size of data pushed by a typical failure stack entry.
1607 An estimate is all we need, because all we use this for
1608 is to choose a limit for how big to make the failure stack. */
1609 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1610 #define TYPICAL_FAILURE_SIZE 20
1612 /* How many items can still be added to the stack without overflowing it. */
1613 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1616 /* Pops what PUSH_FAIL_STACK pushes.
1618 We restore into the parameters, all of which should be lvalues:
1619 STR -- the saved data position.
1620 PAT -- the saved pattern position.
1621 REGSTART, REGEND -- arrays of string positions.
1623 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1624 `pend', `string1', `size1', `string2', and `size2'. */
1626 #define POP_FAILURE_POINT(str, pat) \
1628 assert (!FAIL_STACK_EMPTY ()); \
1630 /* Remove failure points and point to how many regs pushed. */ \
1631 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1632 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1633 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1635 /* Pop the saved registers. */ \
1636 while (fail_stack.frame < fail_stack.avail) \
1637 POP_FAILURE_REG_OR_COUNT (); \
1639 pat = POP_FAILURE_POINTER (); \
1640 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1641 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1643 /* If the saved string location is NULL, it came from an \
1644 on_failure_keep_string_jump opcode, and we want to throw away the \
1645 saved NULL, thus retaining our current position in the string. */ \
1646 str = POP_FAILURE_POINTER (); \
1647 DEBUG_PRINT2 (" Popping string %p: `", str); \
1648 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1649 DEBUG_PRINT1 ("'\n"); \
1651 fail_stack.frame = POP_FAILURE_INT (); \
1652 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1654 assert (fail_stack.avail >= 0); \
1655 assert (fail_stack.frame <= fail_stack.avail); \
1657 DEBUG_STATEMENT (nfailure_points_popped++); \
1658 } while (0) /* POP_FAILURE_POINT */
1662 /* Registers are set to a sentinel when they haven't yet matched. */
1663 #define REG_UNSET(e) ((e) == NULL)
1665 /* Subroutine declarations and macros for regex_compile. */
1667 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1668 reg_syntax_t syntax
,
1669 struct re_pattern_buffer
*bufp
));
1670 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1671 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1672 int arg1
, int arg2
));
1673 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1674 int arg
, unsigned char *end
));
1675 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1676 int arg1
, int arg2
, unsigned char *end
));
1677 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1679 reg_syntax_t syntax
));
1680 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1682 reg_syntax_t syntax
));
1683 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1684 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1685 char *fastmap
, const int multibyte
));
1687 /* Fetch the next character in the uncompiled pattern, with no
1689 #define PATFETCH(c) \
1692 if (p == pend) return REG_EEND; \
1693 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1698 /* If `translate' is non-null, return translate[D], else just D. We
1699 cast the subscript to translate because some data is declared as
1700 `char *', to avoid warnings when a string constant is passed. But
1701 when we use a character as a subscript we must make it unsigned. */
1703 # define TRANSLATE(d) \
1704 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1708 /* Macros for outputting the compiled pattern into `buffer'. */
1710 /* If the buffer isn't allocated when it comes in, use this. */
1711 #define INIT_BUF_SIZE 32
1713 /* Make sure we have at least N more bytes of space in buffer. */
1714 #define GET_BUFFER_SPACE(n) \
1715 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1718 /* Make sure we have one more byte of buffer space and then add C to it. */
1719 #define BUF_PUSH(c) \
1721 GET_BUFFER_SPACE (1); \
1722 *b++ = (unsigned char) (c); \
1726 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1727 #define BUF_PUSH_2(c1, c2) \
1729 GET_BUFFER_SPACE (2); \
1730 *b++ = (unsigned char) (c1); \
1731 *b++ = (unsigned char) (c2); \
1735 /* As with BUF_PUSH_2, except for three bytes. */
1736 #define BUF_PUSH_3(c1, c2, c3) \
1738 GET_BUFFER_SPACE (3); \
1739 *b++ = (unsigned char) (c1); \
1740 *b++ = (unsigned char) (c2); \
1741 *b++ = (unsigned char) (c3); \
1745 /* Store a jump with opcode OP at LOC to location TO. We store a
1746 relative address offset by the three bytes the jump itself occupies. */
1747 #define STORE_JUMP(op, loc, to) \
1748 store_op1 (op, loc, (to) - (loc) - 3)
1750 /* Likewise, for a two-argument jump. */
1751 #define STORE_JUMP2(op, loc, to, arg) \
1752 store_op2 (op, loc, (to) - (loc) - 3, arg)
1754 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1755 #define INSERT_JUMP(op, loc, to) \
1756 insert_op1 (op, loc, (to) - (loc) - 3, b)
1758 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1759 #define INSERT_JUMP2(op, loc, to, arg) \
1760 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1763 /* This is not an arbitrary limit: the arguments which represent offsets
1764 into the pattern are two bytes long. So if 2^16 bytes turns out to
1765 be too small, many things would have to change. */
1766 /* Any other compiler which, like MSC, has allocation limit below 2^16
1767 bytes will have to use approach similar to what was done below for
1768 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1769 reallocating to 0 bytes. Such thing is not going to work too well.
1770 You have been warned!! */
1771 #if defined _MSC_VER && !defined WIN32
1772 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1773 # define MAX_BUF_SIZE 65500L
1775 # define MAX_BUF_SIZE (1L << 16)
1778 /* Extend the buffer by twice its current size via realloc and
1779 reset the pointers that pointed into the old block to point to the
1780 correct places in the new one. If extending the buffer results in it
1781 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1782 #if __BOUNDED_POINTERS__
1783 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1784 # define MOVE_BUFFER_POINTER(P) \
1785 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1786 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1789 SET_HIGH_BOUND (b); \
1790 SET_HIGH_BOUND (begalt); \
1791 if (fixup_alt_jump) \
1792 SET_HIGH_BOUND (fixup_alt_jump); \
1794 SET_HIGH_BOUND (laststart); \
1795 if (pending_exact) \
1796 SET_HIGH_BOUND (pending_exact); \
1799 # define MOVE_BUFFER_POINTER(P) (P) += incr
1800 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1802 #define EXTEND_BUFFER() \
1804 re_char *old_buffer = bufp->buffer; \
1805 if (bufp->allocated == MAX_BUF_SIZE) \
1807 bufp->allocated <<= 1; \
1808 if (bufp->allocated > MAX_BUF_SIZE) \
1809 bufp->allocated = MAX_BUF_SIZE; \
1810 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1811 if (bufp->buffer == NULL) \
1812 return REG_ESPACE; \
1813 /* If the buffer moved, move all the pointers into it. */ \
1814 if (old_buffer != bufp->buffer) \
1816 int incr = bufp->buffer - old_buffer; \
1817 MOVE_BUFFER_POINTER (b); \
1818 MOVE_BUFFER_POINTER (begalt); \
1819 if (fixup_alt_jump) \
1820 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1822 MOVE_BUFFER_POINTER (laststart); \
1823 if (pending_exact) \
1824 MOVE_BUFFER_POINTER (pending_exact); \
1826 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1830 /* Since we have one byte reserved for the register number argument to
1831 {start,stop}_memory, the maximum number of groups we can report
1832 things about is what fits in that byte. */
1833 #define MAX_REGNUM 255
1835 /* But patterns can have more than `MAX_REGNUM' registers. We just
1836 ignore the excess. */
1837 typedef int regnum_t
;
1840 /* Macros for the compile stack. */
1842 /* Since offsets can go either forwards or backwards, this type needs to
1843 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1844 /* int may be not enough when sizeof(int) == 2. */
1845 typedef long pattern_offset_t
;
1849 pattern_offset_t begalt_offset
;
1850 pattern_offset_t fixup_alt_jump
;
1851 pattern_offset_t laststart_offset
;
1853 } compile_stack_elt_t
;
1858 compile_stack_elt_t
*stack
;
1860 unsigned avail
; /* Offset of next open position. */
1861 } compile_stack_type
;
1864 #define INIT_COMPILE_STACK_SIZE 32
1866 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1867 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1869 /* The next available element. */
1870 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1872 /* Explicit quit checking is only used on NTemacs. */
1873 #if defined WINDOWSNT && defined emacs && defined QUIT
1874 extern int immediate_quit
;
1875 # define IMMEDIATE_QUIT_CHECK \
1877 if (immediate_quit) QUIT; \
1880 # define IMMEDIATE_QUIT_CHECK ((void)0)
1883 /* Structure to manage work area for range table. */
1884 struct range_table_work_area
1886 int *table
; /* actual work area. */
1887 int allocated
; /* allocated size for work area in bytes. */
1888 int used
; /* actually used size in words. */
1889 int bits
; /* flag to record character classes */
1892 /* Make sure that WORK_AREA can hold more N multibyte characters.
1893 This is used only in set_image_of_range and set_image_of_range_1.
1894 It expects WORK_AREA to be a pointer.
1895 If it can't get the space, it returns from the surrounding function. */
1897 #define EXTEND_RANGE_TABLE(work_area, n) \
1899 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1901 extend_range_table_work_area (&work_area); \
1902 if ((work_area).table == 0) \
1903 return (REG_ESPACE); \
1907 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1908 (work_area).bits |= (bit)
1910 /* Bits used to implement the multibyte-part of the various character classes
1911 such as [:alnum:] in a charset's range table. */
1912 #define BIT_WORD 0x1
1913 #define BIT_LOWER 0x2
1914 #define BIT_PUNCT 0x4
1915 #define BIT_SPACE 0x8
1916 #define BIT_UPPER 0x10
1917 #define BIT_MULTIBYTE 0x20
1919 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1920 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1922 EXTEND_RANGE_TABLE ((work_area), 2); \
1923 (work_area).table[(work_area).used++] = (range_start); \
1924 (work_area).table[(work_area).used++] = (range_end); \
1927 /* Free allocated memory for WORK_AREA. */
1928 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1930 if ((work_area).table) \
1931 free ((work_area).table); \
1934 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1935 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1936 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1937 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1940 /* Set the bit for character C in a list. */
1941 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1946 /* Store characters in the rage range C0 to C1 in WORK_AREA while
1947 translating them and paying attention to the continuity of
1948 translated characters.
1950 Implementation note: It is better to implement this fairly big
1951 macro by a function, but it's not that easy because macros called
1952 in this macro assume various local variables already declared. */
1954 #define SETUP_MULTIBYTE_RANGE(work_area, c0, c1) \
1956 re_wchar_t c, t, t_last; \
1960 t_last = multibyte ? TRANSLATE (c) : TRANSLATE (MAKE_CHAR_MULTIBYTE (c)); \
1961 for (c++, n = 1; c <= (c1); c++, n++) \
1963 t = multibyte ? TRANSLATE (c) : TRANSLATE (MAKE_CHAR_MULTIBYTE (c)); \
1964 if (t_last + n == t) \
1966 SET_RANGE_TABLE_WORK_AREA ((work_area), t_last, t_last + n - 1); \
1971 SET_RANGE_TABLE_WORK_AREA ((work_area), t_last, t_last + n - 1); \
1976 /* Get the next unsigned number in the uncompiled pattern. */
1977 #define GET_UNSIGNED_NUMBER(num) \
1978 do { if (p != pend) \
1982 FREE_STACK_RETURN (REG_BADBR); \
1983 while ('0' <= c && c <= '9') \
1989 num = num * 10 + c - '0'; \
1990 if (num / 10 != prev) \
1991 FREE_STACK_RETURN (REG_BADBR); \
1997 FREE_STACK_RETURN (REG_BADBR); \
2001 #if WIDE_CHAR_SUPPORT
2002 /* The GNU C library provides support for user-defined character classes
2003 and the functions from ISO C amendement 1. */
2004 # ifdef CHARCLASS_NAME_MAX
2005 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2007 /* This shouldn't happen but some implementation might still have this
2008 problem. Use a reasonable default value. */
2009 # define CHAR_CLASS_MAX_LENGTH 256
2011 typedef wctype_t re_wctype_t
;
2012 typedef wchar_t re_wchar_t
;
2013 # define re_wctype wctype
2014 # define re_iswctype iswctype
2015 # define re_wctype_to_bit(cc) 0
2017 # define CHAR_CLASS_MAX_LENGTH 9 /* Namely, `multibyte'. */
2020 /* Character classes. */
2021 typedef enum { RECC_ERROR
= 0,
2022 RECC_ALNUM
, RECC_ALPHA
, RECC_WORD
,
2023 RECC_GRAPH
, RECC_PRINT
,
2024 RECC_LOWER
, RECC_UPPER
,
2025 RECC_PUNCT
, RECC_CNTRL
,
2026 RECC_DIGIT
, RECC_XDIGIT
,
2027 RECC_BLANK
, RECC_SPACE
,
2028 RECC_MULTIBYTE
, RECC_NONASCII
,
2029 RECC_ASCII
, RECC_UNIBYTE
2032 typedef int re_wchar_t
;
2034 /* Map a string to the char class it names (if any). */
2039 const char *string
= str
;
2040 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2041 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2042 else if (STREQ (string
, "word")) return RECC_WORD
;
2043 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2044 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2045 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2046 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2047 else if (STREQ (string
, "print")) return RECC_PRINT
;
2048 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2049 else if (STREQ (string
, "space")) return RECC_SPACE
;
2050 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2051 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2052 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2053 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2054 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2055 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2056 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2060 /* True iff CH is in the char class CC. */
2062 re_iswctype (ch
, cc
)
2068 case RECC_ALNUM
: return ISALNUM (ch
);
2069 case RECC_ALPHA
: return ISALPHA (ch
);
2070 case RECC_BLANK
: return ISBLANK (ch
);
2071 case RECC_CNTRL
: return ISCNTRL (ch
);
2072 case RECC_DIGIT
: return ISDIGIT (ch
);
2073 case RECC_GRAPH
: return ISGRAPH (ch
);
2074 case RECC_LOWER
: return ISLOWER (ch
);
2075 case RECC_PRINT
: return ISPRINT (ch
);
2076 case RECC_PUNCT
: return ISPUNCT (ch
);
2077 case RECC_SPACE
: return ISSPACE (ch
);
2078 case RECC_UPPER
: return ISUPPER (ch
);
2079 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2080 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2081 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2082 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2083 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2084 case RECC_WORD
: return ISWORD (ch
);
2085 case RECC_ERROR
: return false;
2091 /* Return a bit-pattern to use in the range-table bits to match multibyte
2092 chars of class CC. */
2094 re_wctype_to_bit (cc
)
2099 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2100 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2101 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2102 case RECC_LOWER
: return BIT_LOWER
;
2103 case RECC_UPPER
: return BIT_UPPER
;
2104 case RECC_PUNCT
: return BIT_PUNCT
;
2105 case RECC_SPACE
: return BIT_SPACE
;
2106 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2107 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2114 /* Filling in the work area of a range. */
2116 /* Actually extend the space in WORK_AREA. */
2119 extend_range_table_work_area (work_area
)
2120 struct range_table_work_area
*work_area
;
2122 work_area
->allocated
+= 16 * sizeof (int);
2123 if (work_area
->table
)
2125 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2128 = (int *) malloc (work_area
->allocated
);
2134 /* Carefully find the ranges of codes that are equivalent
2135 under case conversion to the range start..end when passed through
2136 TRANSLATE. Handle the case where non-letters can come in between
2137 two upper-case letters (which happens in Latin-1).
2138 Also handle the case of groups of more than 2 case-equivalent chars.
2140 The basic method is to look at consecutive characters and see
2141 if they can form a run that can be handled as one.
2143 Returns -1 if successful, REG_ESPACE if ran out of space. */
2146 set_image_of_range_1 (work_area
, start
, end
, translate
)
2147 RE_TRANSLATE_TYPE translate
;
2148 struct range_table_work_area
*work_area
;
2149 re_wchar_t start
, end
;
2151 /* `one_case' indicates a character, or a run of characters,
2152 each of which is an isolate (no case-equivalents).
2153 This includes all ASCII non-letters.
2155 `two_case' indicates a character, or a run of characters,
2156 each of which has two case-equivalent forms.
2157 This includes all ASCII letters.
2159 `strange' indicates a character that has more than one
2162 enum case_type
{one_case
, two_case
, strange
};
2164 /* Describe the run that is in progress,
2165 which the next character can try to extend.
2166 If run_type is strange, that means there really is no run.
2167 If run_type is one_case, then run_start...run_end is the run.
2168 If run_type is two_case, then the run is run_start...run_end,
2169 and the case-equivalents end at run_eqv_end. */
2171 enum case_type run_type
= strange
;
2172 int run_start
, run_end
, run_eqv_end
;
2174 Lisp_Object eqv_table
;
2176 if (!RE_TRANSLATE_P (translate
))
2178 EXTEND_RANGE_TABLE (work_area
, 2);
2179 work_area
->table
[work_area
->used
++] = (start
);
2180 work_area
->table
[work_area
->used
++] = (end
);
2184 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2186 for (; start
<= end
; start
++)
2188 enum case_type this_type
;
2189 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2190 int minchar
, maxchar
;
2192 /* Classify this character */
2194 this_type
= one_case
;
2195 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2196 this_type
= two_case
;
2198 this_type
= strange
;
2201 minchar
= start
, maxchar
= eqv
;
2203 minchar
= eqv
, maxchar
= start
;
2205 /* Can this character extend the run in progress? */
2206 if (this_type
== strange
|| this_type
!= run_type
2207 || !(minchar
== run_end
+ 1
2208 && (run_type
== two_case
2209 ? maxchar
== run_eqv_end
+ 1 : 1)))
2212 Record each of its equivalent ranges. */
2213 if (run_type
== one_case
)
2215 EXTEND_RANGE_TABLE (work_area
, 2);
2216 work_area
->table
[work_area
->used
++] = run_start
;
2217 work_area
->table
[work_area
->used
++] = run_end
;
2219 else if (run_type
== two_case
)
2221 EXTEND_RANGE_TABLE (work_area
, 4);
2222 work_area
->table
[work_area
->used
++] = run_start
;
2223 work_area
->table
[work_area
->used
++] = run_end
;
2224 work_area
->table
[work_area
->used
++]
2225 = RE_TRANSLATE (eqv_table
, run_start
);
2226 work_area
->table
[work_area
->used
++]
2227 = RE_TRANSLATE (eqv_table
, run_end
);
2232 if (this_type
== strange
)
2234 /* For a strange character, add each of its equivalents, one
2235 by one. Don't start a range. */
2238 EXTEND_RANGE_TABLE (work_area
, 2);
2239 work_area
->table
[work_area
->used
++] = eqv
;
2240 work_area
->table
[work_area
->used
++] = eqv
;
2241 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2243 while (eqv
!= start
);
2246 /* Add this char to the run, or start a new run. */
2247 else if (run_type
== strange
)
2249 /* Initialize a new range. */
2250 run_type
= this_type
;
2253 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2257 /* Extend a running range. */
2259 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2263 /* If a run is still in progress at the end, finish it now
2264 by recording its equivalent ranges. */
2265 if (run_type
== one_case
)
2267 EXTEND_RANGE_TABLE (work_area
, 2);
2268 work_area
->table
[work_area
->used
++] = run_start
;
2269 work_area
->table
[work_area
->used
++] = run_end
;
2271 else if (run_type
== two_case
)
2273 EXTEND_RANGE_TABLE (work_area
, 4);
2274 work_area
->table
[work_area
->used
++] = run_start
;
2275 work_area
->table
[work_area
->used
++] = run_end
;
2276 work_area
->table
[work_area
->used
++]
2277 = RE_TRANSLATE (eqv_table
, run_start
);
2278 work_area
->table
[work_area
->used
++]
2279 = RE_TRANSLATE (eqv_table
, run_end
);
2287 /* Record the the image of the range start..end when passed through
2288 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2289 and is not even necessarily contiguous.
2290 Normally we approximate it with the smallest contiguous range that contains
2291 all the chars we need. However, for Latin-1 we go to extra effort
2294 This function is not called for ASCII ranges.
2296 Returns -1 if successful, REG_ESPACE if ran out of space. */
2299 set_image_of_range (work_area
, start
, end
, translate
)
2300 RE_TRANSLATE_TYPE translate
;
2301 struct range_table_work_area
*work_area
;
2302 re_wchar_t start
, end
;
2304 re_wchar_t cmin
, cmax
;
2307 /* For Latin-1 ranges, use set_image_of_range_1
2308 to get proper handling of ranges that include letters and nonletters.
2309 For a range that includes the whole of Latin-1, this is not necessary.
2310 For other character sets, we don't bother to get this right. */
2311 if (RE_TRANSLATE_P (translate
) && start
< 04400
2312 && !(start
< 04200 && end
>= 04377))
2319 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2329 EXTEND_RANGE_TABLE (work_area
, 2);
2330 work_area
->table
[work_area
->used
++] = (start
);
2331 work_area
->table
[work_area
->used
++] = (end
);
2333 cmin
= -1, cmax
= -1;
2335 if (RE_TRANSLATE_P (translate
))
2339 for (ch
= start
; ch
<= end
; ch
++)
2341 re_wchar_t c
= TRANSLATE (ch
);
2342 if (! (start
<= c
&& c
<= end
))
2348 cmin
= MIN (cmin
, c
);
2349 cmax
= MAX (cmax
, c
);
2356 EXTEND_RANGE_TABLE (work_area
, 2);
2357 work_area
->table
[work_area
->used
++] = (cmin
);
2358 work_area
->table
[work_area
->used
++] = (cmax
);
2366 #ifndef MATCH_MAY_ALLOCATE
2368 /* If we cannot allocate large objects within re_match_2_internal,
2369 we make the fail stack and register vectors global.
2370 The fail stack, we grow to the maximum size when a regexp
2372 The register vectors, we adjust in size each time we
2373 compile a regexp, according to the number of registers it needs. */
2375 static fail_stack_type fail_stack
;
2377 /* Size with which the following vectors are currently allocated.
2378 That is so we can make them bigger as needed,
2379 but never make them smaller. */
2380 static int regs_allocated_size
;
2382 static re_char
** regstart
, ** regend
;
2383 static re_char
**best_regstart
, **best_regend
;
2385 /* Make the register vectors big enough for NUM_REGS registers,
2386 but don't make them smaller. */
2389 regex_grow_registers (num_regs
)
2392 if (num_regs
> regs_allocated_size
)
2394 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2395 RETALLOC_IF (regend
, num_regs
, re_char
*);
2396 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2397 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2399 regs_allocated_size
= num_regs
;
2403 #endif /* not MATCH_MAY_ALLOCATE */
2405 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2409 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2410 Returns one of error codes defined in `regex.h', or zero for success.
2412 Assumes the `allocated' (and perhaps `buffer') and `translate'
2413 fields are set in BUFP on entry.
2415 If it succeeds, results are put in BUFP (if it returns an error, the
2416 contents of BUFP are undefined):
2417 `buffer' is the compiled pattern;
2418 `syntax' is set to SYNTAX;
2419 `used' is set to the length of the compiled pattern;
2420 `fastmap_accurate' is zero;
2421 `re_nsub' is the number of subexpressions in PATTERN;
2422 `not_bol' and `not_eol' are zero;
2424 The `fastmap' field is neither examined nor set. */
2426 /* Insert the `jump' from the end of last alternative to "here".
2427 The space for the jump has already been allocated. */
2428 #define FIXUP_ALT_JUMP() \
2430 if (fixup_alt_jump) \
2431 STORE_JUMP (jump, fixup_alt_jump, b); \
2435 /* Return, freeing storage we allocated. */
2436 #define FREE_STACK_RETURN(value) \
2438 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2439 free (compile_stack.stack); \
2443 static reg_errcode_t
2444 regex_compile (pattern
, size
, syntax
, bufp
)
2447 reg_syntax_t syntax
;
2448 struct re_pattern_buffer
*bufp
;
2450 /* We fetch characters from PATTERN here. */
2451 register re_wchar_t c
, c1
;
2453 /* A random temporary spot in PATTERN. */
2456 /* Points to the end of the buffer, where we should append. */
2457 register unsigned char *b
;
2459 /* Keeps track of unclosed groups. */
2460 compile_stack_type compile_stack
;
2462 /* Points to the current (ending) position in the pattern. */
2464 /* `const' makes AIX compiler fail. */
2465 unsigned char *p
= pattern
;
2467 re_char
*p
= pattern
;
2469 re_char
*pend
= pattern
+ size
;
2471 /* How to translate the characters in the pattern. */
2472 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2474 /* Address of the count-byte of the most recently inserted `exactn'
2475 command. This makes it possible to tell if a new exact-match
2476 character can be added to that command or if the character requires
2477 a new `exactn' command. */
2478 unsigned char *pending_exact
= 0;
2480 /* Address of start of the most recently finished expression.
2481 This tells, e.g., postfix * where to find the start of its
2482 operand. Reset at the beginning of groups and alternatives. */
2483 unsigned char *laststart
= 0;
2485 /* Address of beginning of regexp, or inside of last group. */
2486 unsigned char *begalt
;
2488 /* Place in the uncompiled pattern (i.e., the {) to
2489 which to go back if the interval is invalid. */
2490 re_char
*beg_interval
;
2492 /* Address of the place where a forward jump should go to the end of
2493 the containing expression. Each alternative of an `or' -- except the
2494 last -- ends with a forward jump of this sort. */
2495 unsigned char *fixup_alt_jump
= 0;
2497 /* Counts open-groups as they are encountered. Remembered for the
2498 matching close-group on the compile stack, so the same register
2499 number is put in the stop_memory as the start_memory. */
2500 regnum_t regnum
= 0;
2502 /* Work area for range table of charset. */
2503 struct range_table_work_area range_table_work
;
2505 /* If the object matched can contain multibyte characters. */
2506 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2508 /* If a target of matching can contain multibyte characters. */
2509 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
2513 DEBUG_PRINT1 ("\nCompiling pattern: ");
2516 unsigned debug_count
;
2518 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2519 putchar (pattern
[debug_count
]);
2524 /* Initialize the compile stack. */
2525 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2526 if (compile_stack
.stack
== NULL
)
2529 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2530 compile_stack
.avail
= 0;
2532 range_table_work
.table
= 0;
2533 range_table_work
.allocated
= 0;
2535 /* Initialize the pattern buffer. */
2536 bufp
->syntax
= syntax
;
2537 bufp
->fastmap_accurate
= 0;
2538 bufp
->not_bol
= bufp
->not_eol
= 0;
2540 /* Set `used' to zero, so that if we return an error, the pattern
2541 printer (for debugging) will think there's no pattern. We reset it
2545 /* Always count groups, whether or not bufp->no_sub is set. */
2548 #if !defined emacs && !defined SYNTAX_TABLE
2549 /* Initialize the syntax table. */
2550 init_syntax_once ();
2553 if (bufp
->allocated
== 0)
2556 { /* If zero allocated, but buffer is non-null, try to realloc
2557 enough space. This loses if buffer's address is bogus, but
2558 that is the user's responsibility. */
2559 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2562 { /* Caller did not allocate a buffer. Do it for them. */
2563 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2565 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2567 bufp
->allocated
= INIT_BUF_SIZE
;
2570 begalt
= b
= bufp
->buffer
;
2572 /* Loop through the uncompiled pattern until we're at the end. */
2581 if ( /* If at start of pattern, it's an operator. */
2583 /* If context independent, it's an operator. */
2584 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2585 /* Otherwise, depends on what's come before. */
2586 || at_begline_loc_p (pattern
, p
, syntax
))
2587 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2596 if ( /* If at end of pattern, it's an operator. */
2598 /* If context independent, it's an operator. */
2599 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2600 /* Otherwise, depends on what's next. */
2601 || at_endline_loc_p (p
, pend
, syntax
))
2602 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2611 if ((syntax
& RE_BK_PLUS_QM
)
2612 || (syntax
& RE_LIMITED_OPS
))
2616 /* If there is no previous pattern... */
2619 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2620 FREE_STACK_RETURN (REG_BADRPT
);
2621 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2626 /* 1 means zero (many) matches is allowed. */
2627 boolean zero_times_ok
= 0, many_times_ok
= 0;
2630 /* If there is a sequence of repetition chars, collapse it
2631 down to just one (the right one). We can't combine
2632 interval operators with these because of, e.g., `a{2}*',
2633 which should only match an even number of `a's. */
2637 if ((syntax
& RE_FRUGAL
)
2638 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2642 zero_times_ok
|= c
!= '+';
2643 many_times_ok
|= c
!= '?';
2649 || (!(syntax
& RE_BK_PLUS_QM
)
2650 && (*p
== '+' || *p
== '?')))
2652 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2655 FREE_STACK_RETURN (REG_EESCAPE
);
2656 if (p
[1] == '+' || p
[1] == '?')
2657 PATFETCH (c
); /* Gobble up the backslash. */
2663 /* If we get here, we found another repeat character. */
2667 /* Star, etc. applied to an empty pattern is equivalent
2668 to an empty pattern. */
2669 if (!laststart
|| laststart
== b
)
2672 /* Now we know whether or not zero matches is allowed
2673 and also whether or not two or more matches is allowed. */
2678 boolean simple
= skip_one_char (laststart
) == b
;
2679 unsigned int startoffset
= 0;
2681 /* Check if the loop can match the empty string. */
2682 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2683 ? on_failure_jump
: on_failure_jump_loop
;
2684 assert (skip_one_char (laststart
) <= b
);
2686 if (!zero_times_ok
&& simple
)
2687 { /* Since simple * loops can be made faster by using
2688 on_failure_keep_string_jump, we turn simple P+
2689 into PP* if P is simple. */
2690 unsigned char *p1
, *p2
;
2691 startoffset
= b
- laststart
;
2692 GET_BUFFER_SPACE (startoffset
);
2693 p1
= b
; p2
= laststart
;
2699 GET_BUFFER_SPACE (6);
2702 STORE_JUMP (ofj
, b
, b
+ 6);
2704 /* Simple * loops can use on_failure_keep_string_jump
2705 depending on what follows. But since we don't know
2706 that yet, we leave the decision up to
2707 on_failure_jump_smart. */
2708 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2709 laststart
+ startoffset
, b
+ 6);
2711 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2716 /* A simple ? pattern. */
2717 assert (zero_times_ok
);
2718 GET_BUFFER_SPACE (3);
2719 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2723 else /* not greedy */
2724 { /* I wish the greedy and non-greedy cases could be merged. */
2726 GET_BUFFER_SPACE (7); /* We might use less. */
2729 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2731 /* The non-greedy multiple match looks like
2732 a repeat..until: we only need a conditional jump
2733 at the end of the loop. */
2734 if (emptyp
) BUF_PUSH (no_op
);
2735 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2736 : on_failure_jump
, b
, laststart
);
2740 /* The repeat...until naturally matches one or more.
2741 To also match zero times, we need to first jump to
2742 the end of the loop (its conditional jump). */
2743 INSERT_JUMP (jump
, laststart
, b
);
2749 /* non-greedy a?? */
2750 INSERT_JUMP (jump
, laststart
, b
+ 3);
2752 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2769 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2771 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2773 /* Ensure that we have enough space to push a charset: the
2774 opcode, the length count, and the bitset; 34 bytes in all. */
2775 GET_BUFFER_SPACE (34);
2779 /* We test `*p == '^' twice, instead of using an if
2780 statement, so we only need one BUF_PUSH. */
2781 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2785 /* Remember the first position in the bracket expression. */
2788 /* Push the number of bytes in the bitmap. */
2789 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2791 /* Clear the whole map. */
2792 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2794 /* charset_not matches newline according to a syntax bit. */
2795 if ((re_opcode_t
) b
[-2] == charset_not
2796 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2797 SET_LIST_BIT ('\n');
2799 /* Read in characters and ranges, setting map bits. */
2802 boolean escaped_char
= false;
2803 const unsigned char *p2
= p
;
2805 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2807 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2808 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2809 So the translation is done later in a loop. Example:
2810 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2813 /* \ might escape characters inside [...] and [^...]. */
2814 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2816 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2819 escaped_char
= true;
2823 /* Could be the end of the bracket expression. If it's
2824 not (i.e., when the bracket expression is `[]' so
2825 far), the ']' character bit gets set way below. */
2826 if (c
== ']' && p2
!= p1
)
2830 /* See if we're at the beginning of a possible character
2833 if (!escaped_char
&&
2834 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2836 /* Leave room for the null. */
2837 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2838 const unsigned char *class_beg
;
2844 /* If pattern is `[[:'. */
2845 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2850 if ((c
== ':' && *p
== ']') || p
== pend
)
2852 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2855 /* This is in any case an invalid class name. */
2860 /* If isn't a word bracketed by `[:' and `:]':
2861 undo the ending character, the letters, and
2862 leave the leading `:' and `[' (but set bits for
2864 if (c
== ':' && *p
== ']')
2870 cc
= re_wctype (str
);
2873 FREE_STACK_RETURN (REG_ECTYPE
);
2875 /* Throw away the ] at the end of the character
2879 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2881 /* Most character classes in a multibyte match
2882 just set a flag. Exceptions are is_blank,
2883 is_digit, is_cntrl, and is_xdigit, since
2884 they can only match ASCII characters. We
2885 don't need to handle them for multibyte.
2886 They are distinguished by a negative wctype. */
2888 for (ch
= 0; ch
< 128; ++ch
)
2889 if (re_iswctype (btowc (ch
), cc
))
2895 if (target_multibyte
)
2897 SET_RANGE_TABLE_WORK_AREA_BIT
2898 (range_table_work
, re_wctype_to_bit (cc
));
2902 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2905 MAKE_CHAR_MULTIBYTE (c
);
2906 if (re_iswctype (btowc (c
), cc
))
2909 MAKE_CHAR_UNIBYTE (c
);
2915 /* Repeat the loop. */
2920 /* Go back to right after the "[:". */
2924 /* Because the `:' may starts the range, we
2925 can't simply set bit and repeat the loop.
2926 Instead, just set it to C and handle below. */
2931 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2934 /* Discard the `-'. */
2937 /* Fetch the character which ends the range. */
2941 if (syntax
& RE_NO_EMPTY_RANGES
)
2942 FREE_STACK_RETURN (REG_ERANGE
);
2943 /* Else, repeat the loop. */
2947 /* Range from C to C. */
2952 c1
= TRANSLATE (c1
);
2953 /* Set the range into bitmap */
2954 for (; c
<= c1
; c
++)
2955 SET_LIST_BIT (TRANSLATE (c
));
2956 #else /* not emacs */
2957 if (target_multibyte
)
2961 re_wchar_t c0
= MAX (c
, 128);
2963 SETUP_MULTIBYTE_RANGE (range_table_work
, c0
, c1
);
2966 for (; c
<= c1
; c
++)
2967 SET_LIST_BIT (TRANSLATE (c
));
2973 for (; c
<= c1
; c
++)
2977 MAKE_CHAR_MULTIBYTE (c0
);
2978 c0
= TRANSLATE (c0
);
2979 MAKE_CHAR_UNIBYTE (c0
);
2983 #endif /* not emacs */
2986 /* Discard any (non)matching list bytes that are all 0 at the
2987 end of the map. Decrease the map-length byte too. */
2988 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2992 /* Build real range table from work area. */
2993 if (RANGE_TABLE_WORK_USED (range_table_work
)
2994 || RANGE_TABLE_WORK_BITS (range_table_work
))
2997 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
2999 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3000 bytes for flags, two for COUNT, and three bytes for
3002 GET_BUFFER_SPACE (4 + used
* 3);
3004 /* Indicate the existence of range table. */
3005 laststart
[1] |= 0x80;
3007 /* Store the character class flag bits into the range table.
3008 If not in emacs, these flag bits are always 0. */
3009 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3010 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3012 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3013 for (i
= 0; i
< used
; i
++)
3014 STORE_CHARACTER_AND_INCR
3015 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3022 if (syntax
& RE_NO_BK_PARENS
)
3029 if (syntax
& RE_NO_BK_PARENS
)
3036 if (syntax
& RE_NEWLINE_ALT
)
3043 if (syntax
& RE_NO_BK_VBAR
)
3050 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3051 goto handle_interval
;
3057 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3059 /* Do not translate the character after the \, so that we can
3060 distinguish, e.g., \B from \b, even if we normally would
3061 translate, e.g., B to b. */
3067 if (syntax
& RE_NO_BK_PARENS
)
3068 goto normal_backslash
;
3075 /* Look for a special (?...) construct */
3076 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3078 PATFETCH (c
); /* Gobble up the '?'. */
3082 case ':': shy
= 1; break;
3084 /* Only (?:...) is supported right now. */
3085 FREE_STACK_RETURN (REG_BADPAT
);
3096 if (COMPILE_STACK_FULL
)
3098 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3099 compile_stack_elt_t
);
3100 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3102 compile_stack
.size
<<= 1;
3105 /* These are the values to restore when we hit end of this
3106 group. They are all relative offsets, so that if the
3107 whole pattern moves because of realloc, they will still
3109 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3110 COMPILE_STACK_TOP
.fixup_alt_jump
3111 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3112 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3113 COMPILE_STACK_TOP
.regnum
= shy
? -regnum
: regnum
;
3116 start_memory for groups beyond the last one we can
3117 represent in the compiled pattern. */
3118 if (regnum
<= MAX_REGNUM
&& !shy
)
3119 BUF_PUSH_2 (start_memory
, regnum
);
3121 compile_stack
.avail
++;
3126 /* If we've reached MAX_REGNUM groups, then this open
3127 won't actually generate any code, so we'll have to
3128 clear pending_exact explicitly. */
3134 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3136 if (COMPILE_STACK_EMPTY
)
3138 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3139 goto normal_backslash
;
3141 FREE_STACK_RETURN (REG_ERPAREN
);
3147 /* See similar code for backslashed left paren above. */
3148 if (COMPILE_STACK_EMPTY
)
3150 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3153 FREE_STACK_RETURN (REG_ERPAREN
);
3156 /* Since we just checked for an empty stack above, this
3157 ``can't happen''. */
3158 assert (compile_stack
.avail
!= 0);
3160 /* We don't just want to restore into `regnum', because
3161 later groups should continue to be numbered higher,
3162 as in `(ab)c(de)' -- the second group is #2. */
3163 regnum_t this_group_regnum
;
3165 compile_stack
.avail
--;
3166 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3168 = COMPILE_STACK_TOP
.fixup_alt_jump
3169 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3171 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3172 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
3173 /* If we've reached MAX_REGNUM groups, then this open
3174 won't actually generate any code, so we'll have to
3175 clear pending_exact explicitly. */
3178 /* We're at the end of the group, so now we know how many
3179 groups were inside this one. */
3180 if (this_group_regnum
<= MAX_REGNUM
&& this_group_regnum
> 0)
3181 BUF_PUSH_2 (stop_memory
, this_group_regnum
);
3186 case '|': /* `\|'. */
3187 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3188 goto normal_backslash
;
3190 if (syntax
& RE_LIMITED_OPS
)
3193 /* Insert before the previous alternative a jump which
3194 jumps to this alternative if the former fails. */
3195 GET_BUFFER_SPACE (3);
3196 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3200 /* The alternative before this one has a jump after it
3201 which gets executed if it gets matched. Adjust that
3202 jump so it will jump to this alternative's analogous
3203 jump (put in below, which in turn will jump to the next
3204 (if any) alternative's such jump, etc.). The last such
3205 jump jumps to the correct final destination. A picture:
3211 If we are at `b', then fixup_alt_jump right now points to a
3212 three-byte space after `a'. We'll put in the jump, set
3213 fixup_alt_jump to right after `b', and leave behind three
3214 bytes which we'll fill in when we get to after `c'. */
3218 /* Mark and leave space for a jump after this alternative,
3219 to be filled in later either by next alternative or
3220 when know we're at the end of a series of alternatives. */
3222 GET_BUFFER_SPACE (3);
3231 /* If \{ is a literal. */
3232 if (!(syntax
& RE_INTERVALS
)
3233 /* If we're at `\{' and it's not the open-interval
3235 || (syntax
& RE_NO_BK_BRACES
))
3236 goto normal_backslash
;
3240 /* If got here, then the syntax allows intervals. */
3242 /* At least (most) this many matches must be made. */
3243 int lower_bound
= 0, upper_bound
= -1;
3248 FREE_STACK_RETURN (REG_EBRACE
);
3250 GET_UNSIGNED_NUMBER (lower_bound
);
3253 GET_UNSIGNED_NUMBER (upper_bound
);
3255 /* Interval such as `{1}' => match exactly once. */
3256 upper_bound
= lower_bound
;
3258 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3259 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3260 FREE_STACK_RETURN (REG_BADBR
);
3262 if (!(syntax
& RE_NO_BK_BRACES
))
3265 FREE_STACK_RETURN (REG_BADBR
);
3271 FREE_STACK_RETURN (REG_BADBR
);
3273 /* We just parsed a valid interval. */
3275 /* If it's invalid to have no preceding re. */
3278 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3279 FREE_STACK_RETURN (REG_BADRPT
);
3280 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3283 goto unfetch_interval
;
3286 if (upper_bound
== 0)
3287 /* If the upper bound is zero, just drop the sub pattern
3290 else if (lower_bound
== 1 && upper_bound
== 1)
3291 /* Just match it once: nothing to do here. */
3294 /* Otherwise, we have a nontrivial interval. When
3295 we're all done, the pattern will look like:
3296 set_number_at <jump count> <upper bound>
3297 set_number_at <succeed_n count> <lower bound>
3298 succeed_n <after jump addr> <succeed_n count>
3300 jump_n <succeed_n addr> <jump count>
3301 (The upper bound and `jump_n' are omitted if
3302 `upper_bound' is 1, though.) */
3304 { /* If the upper bound is > 1, we need to insert
3305 more at the end of the loop. */
3306 unsigned int nbytes
= (upper_bound
< 0 ? 3
3307 : upper_bound
> 1 ? 5 : 0);
3308 unsigned int startoffset
= 0;
3310 GET_BUFFER_SPACE (20); /* We might use less. */
3312 if (lower_bound
== 0)
3314 /* A succeed_n that starts with 0 is really a
3315 a simple on_failure_jump_loop. */
3316 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3322 /* Initialize lower bound of the `succeed_n', even
3323 though it will be set during matching by its
3324 attendant `set_number_at' (inserted next),
3325 because `re_compile_fastmap' needs to know.
3326 Jump to the `jump_n' we might insert below. */
3327 INSERT_JUMP2 (succeed_n
, laststart
,
3332 /* Code to initialize the lower bound. Insert
3333 before the `succeed_n'. The `5' is the last two
3334 bytes of this `set_number_at', plus 3 bytes of
3335 the following `succeed_n'. */
3336 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3341 if (upper_bound
< 0)
3343 /* A negative upper bound stands for infinity,
3344 in which case it degenerates to a plain jump. */
3345 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3348 else if (upper_bound
> 1)
3349 { /* More than one repetition is allowed, so
3350 append a backward jump to the `succeed_n'
3351 that starts this interval.
3353 When we've reached this during matching,
3354 we'll have matched the interval once, so
3355 jump back only `upper_bound - 1' times. */
3356 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3360 /* The location we want to set is the second
3361 parameter of the `jump_n'; that is `b-2' as
3362 an absolute address. `laststart' will be
3363 the `set_number_at' we're about to insert;
3364 `laststart+3' the number to set, the source
3365 for the relative address. But we are
3366 inserting into the middle of the pattern --
3367 so everything is getting moved up by 5.
3368 Conclusion: (b - 2) - (laststart + 3) + 5,
3369 i.e., b - laststart.
3371 We insert this at the beginning of the loop
3372 so that if we fail during matching, we'll
3373 reinitialize the bounds. */
3374 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3375 upper_bound
- 1, b
);
3380 beg_interval
= NULL
;
3385 /* If an invalid interval, match the characters as literals. */
3386 assert (beg_interval
);
3388 beg_interval
= NULL
;
3390 /* normal_char and normal_backslash need `c'. */
3393 if (!(syntax
& RE_NO_BK_BRACES
))
3395 assert (p
> pattern
&& p
[-1] == '\\');
3396 goto normal_backslash
;
3402 /* There is no way to specify the before_dot and after_dot
3403 operators. rms says this is ok. --karl */
3411 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3417 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3423 BUF_PUSH_2 (categoryspec
, c
);
3429 BUF_PUSH_2 (notcategoryspec
, c
);
3435 if (syntax
& RE_NO_GNU_OPS
)
3438 BUF_PUSH_2 (syntaxspec
, Sword
);
3443 if (syntax
& RE_NO_GNU_OPS
)
3446 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3451 if (syntax
& RE_NO_GNU_OPS
)
3457 if (syntax
& RE_NO_GNU_OPS
)
3463 if (syntax
& RE_NO_GNU_OPS
)
3465 BUF_PUSH (wordbound
);
3469 if (syntax
& RE_NO_GNU_OPS
)
3471 BUF_PUSH (notwordbound
);
3475 if (syntax
& RE_NO_GNU_OPS
)
3481 if (syntax
& RE_NO_GNU_OPS
)
3486 case '1': case '2': case '3': case '4': case '5':
3487 case '6': case '7': case '8': case '9':
3491 if (syntax
& RE_NO_BK_REFS
)
3492 goto normal_backslash
;
3496 /* Can't back reference to a subexpression before its end. */
3497 if (reg
> regnum
|| group_in_compile_stack (compile_stack
, reg
))
3498 FREE_STACK_RETURN (REG_ESUBREG
);
3501 BUF_PUSH_2 (duplicate
, reg
);
3508 if (syntax
& RE_BK_PLUS_QM
)
3511 goto normal_backslash
;
3515 /* You might think it would be useful for \ to mean
3516 not to translate; but if we don't translate it
3517 it will never match anything. */
3524 /* Expects the character in `c'. */
3526 /* If no exactn currently being built. */
3529 /* If last exactn not at current position. */
3530 || pending_exact
+ *pending_exact
+ 1 != b
3532 /* We have only one byte following the exactn for the count. */
3533 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3535 /* If followed by a repetition operator. */
3536 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3537 || ((syntax
& RE_BK_PLUS_QM
)
3538 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3539 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3540 || ((syntax
& RE_INTERVALS
)
3541 && ((syntax
& RE_NO_BK_BRACES
)
3542 ? p
!= pend
&& *p
== '{'
3543 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3545 /* Start building a new exactn. */
3549 BUF_PUSH_2 (exactn
, 0);
3550 pending_exact
= b
- 1;
3553 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3558 MAKE_CHAR_MULTIBYTE (c
);
3560 if (target_multibyte
)
3562 len
= CHAR_STRING (c
, b
);
3567 MAKE_CHAR_UNIBYTE (c
);
3571 (*pending_exact
) += len
;
3576 } /* while p != pend */
3579 /* Through the pattern now. */
3583 if (!COMPILE_STACK_EMPTY
)
3584 FREE_STACK_RETURN (REG_EPAREN
);
3586 /* If we don't want backtracking, force success
3587 the first time we reach the end of the compiled pattern. */
3588 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3591 free (compile_stack
.stack
);
3593 /* We have succeeded; set the length of the buffer. */
3594 bufp
->used
= b
- bufp
->buffer
;
3597 /* Now the buffer is adjusted for the multibyteness of a target. */
3598 bufp
->multibyte
= bufp
->target_multibyte
;
3604 re_compile_fastmap (bufp
);
3605 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3606 print_compiled_pattern (bufp
);
3611 #ifndef MATCH_MAY_ALLOCATE
3612 /* Initialize the failure stack to the largest possible stack. This
3613 isn't necessary unless we're trying to avoid calling alloca in
3614 the search and match routines. */
3616 int num_regs
= bufp
->re_nsub
+ 1;
3618 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3620 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3622 if (! fail_stack
.stack
)
3624 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3625 * sizeof (fail_stack_elt_t
));
3628 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3630 * sizeof (fail_stack_elt_t
)));
3633 regex_grow_registers (num_regs
);
3635 #endif /* not MATCH_MAY_ALLOCATE */
3638 } /* regex_compile */
3640 /* Subroutines for `regex_compile'. */
3642 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3645 store_op1 (op
, loc
, arg
)
3650 *loc
= (unsigned char) op
;
3651 STORE_NUMBER (loc
+ 1, arg
);
3655 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3658 store_op2 (op
, loc
, arg1
, arg2
)
3663 *loc
= (unsigned char) op
;
3664 STORE_NUMBER (loc
+ 1, arg1
);
3665 STORE_NUMBER (loc
+ 3, arg2
);
3669 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3670 for OP followed by two-byte integer parameter ARG. */
3673 insert_op1 (op
, loc
, arg
, end
)
3679 register unsigned char *pfrom
= end
;
3680 register unsigned char *pto
= end
+ 3;
3682 while (pfrom
!= loc
)
3685 store_op1 (op
, loc
, arg
);
3689 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3692 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3698 register unsigned char *pfrom
= end
;
3699 register unsigned char *pto
= end
+ 5;
3701 while (pfrom
!= loc
)
3704 store_op2 (op
, loc
, arg1
, arg2
);
3708 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3709 after an alternative or a begin-subexpression. We assume there is at
3710 least one character before the ^. */
3713 at_begline_loc_p (pattern
, p
, syntax
)
3714 re_char
*pattern
, *p
;
3715 reg_syntax_t syntax
;
3717 re_char
*prev
= p
- 2;
3718 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3721 /* After a subexpression? */
3722 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3723 /* After an alternative? */
3724 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3725 /* After a shy subexpression? */
3726 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3727 && prev
[-1] == '?' && prev
[-2] == '('
3728 && (syntax
& RE_NO_BK_PARENS
3729 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3733 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3734 at least one character after the $, i.e., `P < PEND'. */
3737 at_endline_loc_p (p
, pend
, syntax
)
3739 reg_syntax_t syntax
;
3742 boolean next_backslash
= *next
== '\\';
3743 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3746 /* Before a subexpression? */
3747 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3748 : next_backslash
&& next_next
&& *next_next
== ')')
3749 /* Before an alternative? */
3750 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3751 : next_backslash
&& next_next
&& *next_next
== '|');
3755 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3756 false if it's not. */
3759 group_in_compile_stack (compile_stack
, regnum
)
3760 compile_stack_type compile_stack
;
3765 for (this_element
= compile_stack
.avail
- 1;
3768 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3775 If fastmap is non-NULL, go through the pattern and fill fastmap
3776 with all the possible leading chars. If fastmap is NULL, don't
3777 bother filling it up (obviously) and only return whether the
3778 pattern could potentially match the empty string.
3780 Return 1 if p..pend might match the empty string.
3781 Return 0 if p..pend matches at least one char.
3782 Return -1 if fastmap was not updated accurately. */
3785 analyse_first (p
, pend
, fastmap
, multibyte
)
3788 const int multibyte
;
3793 /* If all elements for base leading-codes in fastmap is set, this
3794 flag is set true. */
3795 boolean match_any_multibyte_characters
= false;
3799 /* The loop below works as follows:
3800 - It has a working-list kept in the PATTERN_STACK and which basically
3801 starts by only containing a pointer to the first operation.
3802 - If the opcode we're looking at is a match against some set of
3803 chars, then we add those chars to the fastmap and go on to the
3804 next work element from the worklist (done via `break').
3805 - If the opcode is a control operator on the other hand, we either
3806 ignore it (if it's meaningless at this point, such as `start_memory')
3807 or execute it (if it's a jump). If the jump has several destinations
3808 (i.e. `on_failure_jump'), then we push the other destination onto the
3810 We guarantee termination by ignoring backward jumps (more or less),
3811 so that `p' is monotonically increasing. More to the point, we
3812 never set `p' (or push) anything `<= p1'. */
3816 /* `p1' is used as a marker of how far back a `on_failure_jump'
3817 can go without being ignored. It is normally equal to `p'
3818 (which prevents any backward `on_failure_jump') except right
3819 after a plain `jump', to allow patterns such as:
3822 10: on_failure_jump 3
3823 as used for the *? operator. */
3826 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3833 /* If the first character has to match a backreference, that means
3834 that the group was empty (since it already matched). Since this
3835 is the only case that interests us here, we can assume that the
3836 backreference must match the empty string. */
3841 /* Following are the cases which match a character. These end
3846 /* If multibyte is nonzero, the first byte of each
3847 character is an ASCII or a leading code. Otherwise,
3848 each byte is a character. Thus, this works in both
3855 /* We could put all the chars except for \n (and maybe \0)
3856 but we don't bother since it is generally not worth it. */
3857 if (!fastmap
) break;
3862 if (!fastmap
) break;
3864 /* Chars beyond end of bitmap are possible matches. */
3865 /* In a multibyte case, the bitmap is used only for ASCII
3867 int limit
= multibyte
? 128 : (1 << BYTEWIDTH
);
3869 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3876 if (!fastmap
) break;
3877 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3878 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3880 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3883 if ((not && multibyte
)
3884 /* Any leading code can possibly start a character
3885 which doesn't match the specified set of characters. */
3886 || (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3887 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3888 /* If we can match a character class, we can match
3889 any multibyte characters. */
3891 if (match_any_multibyte_characters
== false)
3893 for (j
= 0x80; j
< (1 << BYTEWIDTH
); j
++)
3895 match_any_multibyte_characters
= true;
3899 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3900 && match_any_multibyte_characters
== false)
3902 /* Set fastmap[I] to 1 where I is a leading code of each
3903 multibyte characer in the range table. */
3905 unsigned char lc1
, lc2
;
3907 /* Make P points the range table. `+ 2' is to skip flag
3908 bits for a character class. */
3909 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3911 /* Extract the number of ranges in range table into COUNT. */
3912 EXTRACT_NUMBER_AND_INCR (count
, p
);
3913 for (; count
> 0; count
--, p
+= 2 * 3) /* XXX */
3915 /* Extract the start and end of each range. */
3916 EXTRACT_CHARACTER (c
, p
);
3917 lc1
= CHAR_LEADING_CODE (c
);
3919 EXTRACT_CHARACTER (c
, p
);
3920 lc2
= CHAR_LEADING_CODE (c
);
3921 for (j
= lc1
; j
<= lc2
; j
++)
3929 if (!fastmap
) break;
3931 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
3933 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3934 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
3938 /* This match depends on text properties. These end with
3939 aborting optimizations. */
3943 case notcategoryspec
:
3944 if (!fastmap
) break;
3945 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
3947 for (j
= (multibyte
? 127 : (1 << BYTEWIDTH
)); j
>= 0; j
--)
3948 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
3953 /* Any character set can possibly contain a character
3954 whose category is K (or not). */
3955 if (match_any_multibyte_characters
== false)
3957 for (j
= 0x80; j
< (1 << BYTEWIDTH
); j
++)
3959 match_any_multibyte_characters
= true;
3964 /* All cases after this match the empty string. These end with
3984 EXTRACT_NUMBER_AND_INCR (j
, p
);
3986 /* Backward jumps can only go back to code that we've already
3987 visited. `re_compile' should make sure this is true. */
3990 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
3992 case on_failure_jump
:
3993 case on_failure_keep_string_jump
:
3994 case on_failure_jump_loop
:
3995 case on_failure_jump_nastyloop
:
3996 case on_failure_jump_smart
:
4002 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4003 to jump back to "just after here". */
4006 case on_failure_jump
:
4007 case on_failure_keep_string_jump
:
4008 case on_failure_jump_nastyloop
:
4009 case on_failure_jump_loop
:
4010 case on_failure_jump_smart
:
4011 EXTRACT_NUMBER_AND_INCR (j
, p
);
4013 ; /* Backward jump to be ignored. */
4015 { /* We have to look down both arms.
4016 We first go down the "straight" path so as to minimize
4017 stack usage when going through alternatives. */
4018 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4026 /* This code simply does not properly handle forward jump_n. */
4027 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4029 /* jump_n can either jump or fall through. The (backward) jump
4030 case has already been handled, so we only need to look at the
4031 fallthrough case. */
4035 /* If N == 0, it should be an on_failure_jump_loop instead. */
4036 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4038 /* We only care about one iteration of the loop, so we don't
4039 need to consider the case where this behaves like an
4056 abort (); /* We have listed all the cases. */
4059 /* Getting here means we have found the possible starting
4060 characters for one path of the pattern -- and that the empty
4061 string does not match. We need not follow this path further. */
4065 /* We reached the end without matching anything. */
4068 } /* analyse_first */
4070 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4071 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4072 characters can start a string that matches the pattern. This fastmap
4073 is used by re_search to skip quickly over impossible starting points.
4075 Character codes above (1 << BYTEWIDTH) are not represented in the
4076 fastmap, but the leading codes are represented. Thus, the fastmap
4077 indicates which character sets could start a match.
4079 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4080 area as BUFP->fastmap.
4082 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4085 Returns 0 if we succeed, -2 if an internal error. */
4088 re_compile_fastmap (bufp
)
4089 struct re_pattern_buffer
*bufp
;
4091 char *fastmap
= bufp
->fastmap
;
4094 assert (fastmap
&& bufp
->buffer
);
4096 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4097 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4099 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4100 fastmap
, RE_MULTIBYTE_P (bufp
));
4101 bufp
->can_be_null
= (analysis
!= 0);
4103 } /* re_compile_fastmap */
4105 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4106 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4107 this memory for recording register information. STARTS and ENDS
4108 must be allocated using the malloc library routine, and must each
4109 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4111 If NUM_REGS == 0, then subsequent matches should allocate their own
4114 Unless this function is called, the first search or match using
4115 PATTERN_BUFFER will allocate its own register data, without
4116 freeing the old data. */
4119 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4120 struct re_pattern_buffer
*bufp
;
4121 struct re_registers
*regs
;
4123 regoff_t
*starts
, *ends
;
4127 bufp
->regs_allocated
= REGS_REALLOCATE
;
4128 regs
->num_regs
= num_regs
;
4129 regs
->start
= starts
;
4134 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4136 regs
->start
= regs
->end
= (regoff_t
*) 0;
4139 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4141 /* Searching routines. */
4143 /* Like re_search_2, below, but only one string is specified, and
4144 doesn't let you say where to stop matching. */
4147 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4148 struct re_pattern_buffer
*bufp
;
4150 int size
, startpos
, range
;
4151 struct re_registers
*regs
;
4153 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4156 WEAK_ALIAS (__re_search
, re_search
)
4158 /* Head address of virtual concatenation of string. */
4159 #define HEAD_ADDR_VSTRING(P) \
4160 (((P) >= size1 ? string2 : string1))
4162 /* End address of virtual concatenation of string. */
4163 #define STOP_ADDR_VSTRING(P) \
4164 (((P) >= size1 ? string2 + size2 : string1 + size1))
4166 /* Address of POS in the concatenation of virtual string. */
4167 #define POS_ADDR_VSTRING(POS) \
4168 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4170 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4171 virtual concatenation of STRING1 and STRING2, starting first at index
4172 STARTPOS, then at STARTPOS + 1, and so on.
4174 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4176 RANGE is how far to scan while trying to match. RANGE = 0 means try
4177 only at STARTPOS; in general, the last start tried is STARTPOS +
4180 In REGS, return the indices of the virtual concatenation of STRING1
4181 and STRING2 that matched the entire BUFP->buffer and its contained
4184 Do not consider matching one past the index STOP in the virtual
4185 concatenation of STRING1 and STRING2.
4187 We return either the position in the strings at which the match was
4188 found, -1 if no match, or -2 if error (such as failure
4192 re_search_2 (bufp
, str1
, size1
, str2
, size2
, startpos
, range
, regs
, stop
)
4193 struct re_pattern_buffer
*bufp
;
4194 const char *str1
, *str2
;
4198 struct re_registers
*regs
;
4202 re_char
*string1
= (re_char
*) str1
;
4203 re_char
*string2
= (re_char
*) str2
;
4204 register char *fastmap
= bufp
->fastmap
;
4205 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4206 int total_size
= size1
+ size2
;
4207 int endpos
= startpos
+ range
;
4208 boolean anchored_start
;
4209 /* Nonzero if BUFP is setup for multibyte characters. We are sure
4210 that it is the same as RE_TARGET_MULTIBYTE_P (bufp). */
4211 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4213 /* Check for out-of-range STARTPOS. */
4214 if (startpos
< 0 || startpos
> total_size
)
4217 /* Fix up RANGE if it might eventually take us outside
4218 the virtual concatenation of STRING1 and STRING2.
4219 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4221 range
= 0 - startpos
;
4222 else if (endpos
> total_size
)
4223 range
= total_size
- startpos
;
4225 /* If the search isn't to be a backwards one, don't waste time in a
4226 search for a pattern anchored at beginning of buffer. */
4227 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4236 /* In a forward search for something that starts with \=.
4237 don't keep searching past point. */
4238 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4240 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4246 /* Update the fastmap now if not correct already. */
4247 if (fastmap
&& !bufp
->fastmap_accurate
)
4248 re_compile_fastmap (bufp
);
4250 /* See whether the pattern is anchored. */
4251 anchored_start
= (bufp
->buffer
[0] == begline
);
4254 gl_state
.object
= re_match_object
;
4256 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4258 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4262 /* Loop through the string, looking for a place to start matching. */
4265 /* If the pattern is anchored,
4266 skip quickly past places we cannot match.
4267 We don't bother to treat startpos == 0 specially
4268 because that case doesn't repeat. */
4269 if (anchored_start
&& startpos
> 0)
4271 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4272 : string2
[startpos
- size1
- 1])
4277 /* If a fastmap is supplied, skip quickly over characters that
4278 cannot be the start of a match. If the pattern can match the
4279 null string, however, we don't need to skip characters; we want
4280 the first null string. */
4281 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4283 register re_char
*d
;
4284 register re_wchar_t buf_ch
;
4286 d
= POS_ADDR_VSTRING (startpos
);
4288 if (range
> 0) /* Searching forwards. */
4290 register int lim
= 0;
4293 if (startpos
< size1
&& startpos
+ range
>= size1
)
4294 lim
= range
- (size1
- startpos
);
4296 /* Written out as an if-else to avoid testing `translate'
4298 if (RE_TRANSLATE_P (translate
))
4305 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4307 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4308 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4311 range
-= buf_charlen
;
4318 MAKE_CHAR_MULTIBYTE (buf_ch
);
4319 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4320 MAKE_CHAR_UNIBYTE (buf_ch
);
4321 if (fastmap
[buf_ch
])
4334 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4336 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4338 range
-= buf_charlen
;
4342 while (range
> lim
&& !fastmap
[*d
])
4348 startpos
+= irange
- range
;
4350 else /* Searching backwards. */
4352 int room
= (startpos
>= size1
4353 ? size2
+ size1
- startpos
4354 : size1
- startpos
);
4357 buf_ch
= STRING_CHAR (d
, room
);
4358 buf_ch
= TRANSLATE (buf_ch
);
4359 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4364 if (! fastmap
[TRANSLATE (*d
)])
4370 /* If can't match the null string, and that's all we have left, fail. */
4371 if (range
>= 0 && startpos
== total_size
&& fastmap
4372 && !bufp
->can_be_null
)
4375 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4376 startpos
, regs
, stop
);
4377 #ifndef REGEX_MALLOC
4394 /* Update STARTPOS to the next character boundary. */
4397 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4398 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4399 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
4417 /* Update STARTPOS to the previous character boundary. */
4420 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4422 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4424 /* Find the head of multibyte form. */
4425 PREV_CHAR_BOUNDARY (p
, phead
);
4426 range
+= p0
- 1 - p
;
4430 startpos
-= p0
- 1 - p
;
4436 WEAK_ALIAS (__re_search_2
, re_search_2
)
4438 /* Declarations and macros for re_match_2. */
4440 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4442 RE_TRANSLATE_TYPE translate
,
4443 const int multibyte
));
4445 /* This converts PTR, a pointer into one of the search strings `string1'
4446 and `string2' into an offset from the beginning of that string. */
4447 #define POINTER_TO_OFFSET(ptr) \
4448 (FIRST_STRING_P (ptr) \
4449 ? ((regoff_t) ((ptr) - string1)) \
4450 : ((regoff_t) ((ptr) - string2 + size1)))
4452 /* Call before fetching a character with *d. This switches over to
4453 string2 if necessary.
4454 Check re_match_2_internal for a discussion of why end_match_2 might
4455 not be within string2 (but be equal to end_match_1 instead). */
4456 #define PREFETCH() \
4459 /* End of string2 => fail. */ \
4460 if (dend == end_match_2) \
4462 /* End of string1 => advance to string2. */ \
4464 dend = end_match_2; \
4467 /* Call before fetching a char with *d if you already checked other limits.
4468 This is meant for use in lookahead operations like wordend, etc..
4469 where we might need to look at parts of the string that might be
4470 outside of the LIMITs (i.e past `stop'). */
4471 #define PREFETCH_NOLIMIT() \
4475 dend = end_match_2; \
4478 /* Test if at very beginning or at very end of the virtual concatenation
4479 of `string1' and `string2'. If only one string, it's `string2'. */
4480 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4481 #define AT_STRINGS_END(d) ((d) == end2)
4484 /* Test if D points to a character which is word-constituent. We have
4485 two special cases to check for: if past the end of string1, look at
4486 the first character in string2; and if before the beginning of
4487 string2, look at the last character in string1. */
4488 #define WORDCHAR_P(d) \
4489 (SYNTAX ((d) == end1 ? *string2 \
4490 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4493 /* Disabled due to a compiler bug -- see comment at case wordbound */
4495 /* The comment at case wordbound is following one, but we don't use
4496 AT_WORD_BOUNDARY anymore to support multibyte form.
4498 The DEC Alpha C compiler 3.x generates incorrect code for the
4499 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4500 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4501 macro and introducing temporary variables works around the bug. */
4504 /* Test if the character before D and the one at D differ with respect
4505 to being word-constituent. */
4506 #define AT_WORD_BOUNDARY(d) \
4507 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4508 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4511 /* Free everything we malloc. */
4512 #ifdef MATCH_MAY_ALLOCATE
4513 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4514 # define FREE_VARIABLES() \
4516 REGEX_FREE_STACK (fail_stack.stack); \
4517 FREE_VAR (regstart); \
4518 FREE_VAR (regend); \
4519 FREE_VAR (best_regstart); \
4520 FREE_VAR (best_regend); \
4523 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4524 #endif /* not MATCH_MAY_ALLOCATE */
4527 /* Optimization routines. */
4529 /* If the operation is a match against one or more chars,
4530 return a pointer to the next operation, else return NULL. */
4535 switch (SWITCH_ENUM_CAST (*p
++))
4546 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4549 p
= CHARSET_RANGE_TABLE (p
- 1);
4550 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4551 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4554 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4561 case notcategoryspec
:
4573 /* Jump over non-matching operations. */
4574 static unsigned char *
4575 skip_noops (p
, pend
)
4576 unsigned char *p
, *pend
;
4581 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4590 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4601 /* Non-zero if "p1 matches something" implies "p2 fails". */
4603 mutually_exclusive_p (bufp
, p1
, p2
)
4604 struct re_pattern_buffer
*bufp
;
4605 unsigned char *p1
, *p2
;
4608 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4609 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4611 assert (p1
>= bufp
->buffer
&& p1
< pend
4612 && p2
>= bufp
->buffer
&& p2
<= pend
);
4614 /* Skip over open/close-group commands.
4615 If what follows this loop is a ...+ construct,
4616 look at what begins its body, since we will have to
4617 match at least one of that. */
4618 p2
= skip_noops (p2
, pend
);
4619 /* The same skip can be done for p1, except that this function
4620 is only used in the case where p1 is a simple match operator. */
4621 /* p1 = skip_noops (p1, pend); */
4623 assert (p1
>= bufp
->buffer
&& p1
< pend
4624 && p2
>= bufp
->buffer
&& p2
<= pend
);
4626 op2
= p2
== pend
? succeed
: *p2
;
4628 switch (SWITCH_ENUM_CAST (op2
))
4632 /* If we're at the end of the pattern, we can change. */
4633 if (skip_one_char (p1
))
4635 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4643 register re_wchar_t c
4644 = (re_opcode_t
) *p2
== endline
? '\n'
4645 : RE_STRING_CHAR (p2
+ 2, pend
- p2
- 2);
4647 if ((re_opcode_t
) *p1
== exactn
)
4649 if (c
!= RE_STRING_CHAR (p1
+ 2, pend
- p1
- 2))
4651 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4656 else if ((re_opcode_t
) *p1
== charset
4657 || (re_opcode_t
) *p1
== charset_not
)
4659 int not = (re_opcode_t
) *p1
== charset_not
;
4661 /* Test if C is listed in charset (or charset_not)
4663 if (! multibyte
|| IS_REAL_ASCII (c
))
4665 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4666 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4669 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4670 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4672 /* `not' is equal to 1 if c would match, which means
4673 that we can't change to pop_failure_jump. */
4676 DEBUG_PRINT1 (" No match => fast loop.\n");
4680 else if ((re_opcode_t
) *p1
== anychar
4683 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4691 if ((re_opcode_t
) *p1
== exactn
)
4692 /* Reuse the code above. */
4693 return mutually_exclusive_p (bufp
, p2
, p1
);
4695 /* It is hard to list up all the character in charset
4696 P2 if it includes multibyte character. Give up in
4698 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4700 /* Now, we are sure that P2 has no range table.
4701 So, for the size of bitmap in P2, `p2[1]' is
4702 enough. But P1 may have range table, so the
4703 size of bitmap table of P1 is extracted by
4704 using macro `CHARSET_BITMAP_SIZE'.
4706 In a multibyte case, we know that all the character
4707 listed in P2 is ASCII. In a unibyte case, P1 has only a
4708 bitmap table. So, in both cases, it is enough to test
4709 only the bitmap table of P1. */
4711 if ((re_opcode_t
) *p1
== charset
)
4714 /* We win if the charset inside the loop
4715 has no overlap with the one after the loop. */
4718 && idx
< CHARSET_BITMAP_SIZE (p1
));
4720 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4724 || idx
== CHARSET_BITMAP_SIZE (p1
))
4726 DEBUG_PRINT1 (" No match => fast loop.\n");
4730 else if ((re_opcode_t
) *p1
== charset_not
)
4733 /* We win if the charset_not inside the loop lists
4734 every character listed in the charset after. */
4735 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4736 if (! (p2
[2 + idx
] == 0
4737 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4738 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4743 DEBUG_PRINT1 (" No match => fast loop.\n");
4752 switch (SWITCH_ENUM_CAST (*p1
))
4756 /* Reuse the code above. */
4757 return mutually_exclusive_p (bufp
, p2
, p1
);
4759 /* When we have two charset_not, it's very unlikely that
4760 they don't overlap. The union of the two sets of excluded
4761 chars should cover all possible chars, which, as a matter of
4762 fact, is virtually impossible in multibyte buffers. */
4769 return ((re_opcode_t
) *p1
== syntaxspec
4770 && p1
[1] == (op2
== wordend
? Sword
: p2
[1]));
4774 return ((re_opcode_t
) *p1
== notsyntaxspec
4775 && p1
[1] == (op2
== wordend
? Sword
: p2
[1]));
4778 return (((re_opcode_t
) *p1
== notsyntaxspec
4779 || (re_opcode_t
) *p1
== syntaxspec
)
4784 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4785 case notcategoryspec
:
4786 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4798 /* Matching routines. */
4800 #ifndef emacs /* Emacs never uses this. */
4801 /* re_match is like re_match_2 except it takes only a single string. */
4804 re_match (bufp
, string
, size
, pos
, regs
)
4805 struct re_pattern_buffer
*bufp
;
4808 struct re_registers
*regs
;
4810 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
4812 # if defined C_ALLOCA && !defined REGEX_MALLOC
4817 WEAK_ALIAS (__re_match
, re_match
)
4818 #endif /* not emacs */
4821 /* In Emacs, this is the string or buffer in which we
4822 are matching. It is used for looking up syntax properties. */
4823 Lisp_Object re_match_object
;
4826 /* re_match_2 matches the compiled pattern in BUFP against the
4827 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4828 and SIZE2, respectively). We start matching at POS, and stop
4831 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4832 store offsets for the substring each group matched in REGS. See the
4833 documentation for exactly how many groups we fill.
4835 We return -1 if no match, -2 if an internal error (such as the
4836 failure stack overflowing). Otherwise, we return the length of the
4837 matched substring. */
4840 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4841 struct re_pattern_buffer
*bufp
;
4842 const char *string1
, *string2
;
4845 struct re_registers
*regs
;
4852 gl_state
.object
= re_match_object
;
4853 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4854 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4857 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4858 (re_char
*) string2
, size2
,
4860 #if defined C_ALLOCA && !defined REGEX_MALLOC
4865 WEAK_ALIAS (__re_match_2
, re_match_2
)
4868 #define TRANSLATE_VIA_MULTIBYTE(c) \
4871 (c) = TRANSLATE (c); \
4874 MAKE_CHAR_MULTIBYTE (c); \
4875 (c) = TRANSLATE (c); \
4876 MAKE_CHAR_UNIBYTE (c); \
4881 #define TRANSLATE_VIA_MULTIBYTE(c) ((c) = TRANSLATE (c))
4885 /* This is a separate function so that we can force an alloca cleanup
4888 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4889 struct re_pattern_buffer
*bufp
;
4890 re_char
*string1
, *string2
;
4893 struct re_registers
*regs
;
4896 /* General temporaries. */
4901 /* Just past the end of the corresponding string. */
4902 re_char
*end1
, *end2
;
4904 /* Pointers into string1 and string2, just past the last characters in
4905 each to consider matching. */
4906 re_char
*end_match_1
, *end_match_2
;
4908 /* Where we are in the data, and the end of the current string. */
4911 /* Used sometimes to remember where we were before starting matching
4912 an operator so that we can go back in case of failure. This "atomic"
4913 behavior of matching opcodes is indispensable to the correctness
4914 of the on_failure_keep_string_jump optimization. */
4917 /* Where we are in the pattern, and the end of the pattern. */
4918 re_char
*p
= bufp
->buffer
;
4919 re_char
*pend
= p
+ bufp
->used
;
4921 /* We use this to map every character in the string. */
4922 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4924 /* Nonzero if BUFP is setup for multibyte characters. We are sure
4925 that it is the same as RE_TARGET_MULTIBYTE_P (bufp). */
4926 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4928 /* Failure point stack. Each place that can handle a failure further
4929 down the line pushes a failure point on this stack. It consists of
4930 regstart, and regend for all registers corresponding to
4931 the subexpressions we're currently inside, plus the number of such
4932 registers, and, finally, two char *'s. The first char * is where
4933 to resume scanning the pattern; the second one is where to resume
4934 scanning the strings. */
4935 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4936 fail_stack_type fail_stack
;
4939 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4942 #if defined REL_ALLOC && defined REGEX_MALLOC
4943 /* This holds the pointer to the failure stack, when
4944 it is allocated relocatably. */
4945 fail_stack_elt_t
*failure_stack_ptr
;
4948 /* We fill all the registers internally, independent of what we
4949 return, for use in backreferences. The number here includes
4950 an element for register zero. */
4951 size_t num_regs
= bufp
->re_nsub
+ 1;
4953 /* Information on the contents of registers. These are pointers into
4954 the input strings; they record just what was matched (on this
4955 attempt) by a subexpression part of the pattern, that is, the
4956 regnum-th regstart pointer points to where in the pattern we began
4957 matching and the regnum-th regend points to right after where we
4958 stopped matching the regnum-th subexpression. (The zeroth register
4959 keeps track of what the whole pattern matches.) */
4960 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4961 re_char
**regstart
, **regend
;
4964 /* The following record the register info as found in the above
4965 variables when we find a match better than any we've seen before.
4966 This happens as we backtrack through the failure points, which in
4967 turn happens only if we have not yet matched the entire string. */
4968 unsigned best_regs_set
= false;
4969 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4970 re_char
**best_regstart
, **best_regend
;
4973 /* Logically, this is `best_regend[0]'. But we don't want to have to
4974 allocate space for that if we're not allocating space for anything
4975 else (see below). Also, we never need info about register 0 for
4976 any of the other register vectors, and it seems rather a kludge to
4977 treat `best_regend' differently than the rest. So we keep track of
4978 the end of the best match so far in a separate variable. We
4979 initialize this to NULL so that when we backtrack the first time
4980 and need to test it, it's not garbage. */
4981 re_char
*match_end
= NULL
;
4984 /* Counts the total number of registers pushed. */
4985 unsigned num_regs_pushed
= 0;
4988 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4992 #ifdef MATCH_MAY_ALLOCATE
4993 /* Do not bother to initialize all the register variables if there are
4994 no groups in the pattern, as it takes a fair amount of time. If
4995 there are groups, we include space for register 0 (the whole
4996 pattern), even though we never use it, since it simplifies the
4997 array indexing. We should fix this. */
5000 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5001 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5002 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5003 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5005 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5013 /* We must initialize all our variables to NULL, so that
5014 `FREE_VARIABLES' doesn't try to free them. */
5015 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5017 #endif /* MATCH_MAY_ALLOCATE */
5019 /* The starting position is bogus. */
5020 if (pos
< 0 || pos
> size1
+ size2
)
5026 /* Initialize subexpression text positions to -1 to mark ones that no
5027 start_memory/stop_memory has been seen for. Also initialize the
5028 register information struct. */
5029 for (reg
= 1; reg
< num_regs
; reg
++)
5030 regstart
[reg
] = regend
[reg
] = NULL
;
5032 /* We move `string1' into `string2' if the latter's empty -- but not if
5033 `string1' is null. */
5034 if (size2
== 0 && string1
!= NULL
)
5041 end1
= string1
+ size1
;
5042 end2
= string2
+ size2
;
5044 /* `p' scans through the pattern as `d' scans through the data.
5045 `dend' is the end of the input string that `d' points within. `d'
5046 is advanced into the following input string whenever necessary, but
5047 this happens before fetching; therefore, at the beginning of the
5048 loop, `d' can be pointing at the end of a string, but it cannot
5052 /* Only match within string2. */
5053 d
= string2
+ pos
- size1
;
5054 dend
= end_match_2
= string2
+ stop
- size1
;
5055 end_match_1
= end1
; /* Just to give it a value. */
5061 /* Only match within string1. */
5062 end_match_1
= string1
+ stop
;
5064 When we reach end_match_1, PREFETCH normally switches to string2.
5065 But in the present case, this means that just doing a PREFETCH
5066 makes us jump from `stop' to `gap' within the string.
5067 What we really want here is for the search to stop as
5068 soon as we hit end_match_1. That's why we set end_match_2
5069 to end_match_1 (since PREFETCH fails as soon as we hit
5071 end_match_2
= end_match_1
;
5074 { /* It's important to use this code when stop == size so that
5075 moving `d' from end1 to string2 will not prevent the d == dend
5076 check from catching the end of string. */
5078 end_match_2
= string2
+ stop
- size1
;
5084 DEBUG_PRINT1 ("The compiled pattern is: ");
5085 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5086 DEBUG_PRINT1 ("The string to match is: `");
5087 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5088 DEBUG_PRINT1 ("'\n");
5090 /* This loops over pattern commands. It exits by returning from the
5091 function if the match is complete, or it drops through if the match
5092 fails at this starting point in the input data. */
5095 DEBUG_PRINT2 ("\n%p: ", p
);
5098 { /* End of pattern means we might have succeeded. */
5099 DEBUG_PRINT1 ("end of pattern ... ");
5101 /* If we haven't matched the entire string, and we want the
5102 longest match, try backtracking. */
5103 if (d
!= end_match_2
)
5105 /* 1 if this match ends in the same string (string1 or string2)
5106 as the best previous match. */
5107 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5108 == FIRST_STRING_P (d
));
5109 /* 1 if this match is the best seen so far. */
5110 boolean best_match_p
;
5112 /* AIX compiler got confused when this was combined
5113 with the previous declaration. */
5115 best_match_p
= d
> match_end
;
5117 best_match_p
= !FIRST_STRING_P (d
);
5119 DEBUG_PRINT1 ("backtracking.\n");
5121 if (!FAIL_STACK_EMPTY ())
5122 { /* More failure points to try. */
5124 /* If exceeds best match so far, save it. */
5125 if (!best_regs_set
|| best_match_p
)
5127 best_regs_set
= true;
5130 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5132 for (reg
= 1; reg
< num_regs
; reg
++)
5134 best_regstart
[reg
] = regstart
[reg
];
5135 best_regend
[reg
] = regend
[reg
];
5141 /* If no failure points, don't restore garbage. And if
5142 last match is real best match, don't restore second
5144 else if (best_regs_set
&& !best_match_p
)
5147 /* Restore best match. It may happen that `dend ==
5148 end_match_1' while the restored d is in string2.
5149 For example, the pattern `x.*y.*z' against the
5150 strings `x-' and `y-z-', if the two strings are
5151 not consecutive in memory. */
5152 DEBUG_PRINT1 ("Restoring best registers.\n");
5155 dend
= ((d
>= string1
&& d
<= end1
)
5156 ? end_match_1
: end_match_2
);
5158 for (reg
= 1; reg
< num_regs
; reg
++)
5160 regstart
[reg
] = best_regstart
[reg
];
5161 regend
[reg
] = best_regend
[reg
];
5164 } /* d != end_match_2 */
5167 DEBUG_PRINT1 ("Accepting match.\n");
5169 /* If caller wants register contents data back, do it. */
5170 if (regs
&& !bufp
->no_sub
)
5172 /* Have the register data arrays been allocated? */
5173 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5174 { /* No. So allocate them with malloc. We need one
5175 extra element beyond `num_regs' for the `-1' marker
5177 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5178 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5179 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5180 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5185 bufp
->regs_allocated
= REGS_REALLOCATE
;
5187 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5188 { /* Yes. If we need more elements than were already
5189 allocated, reallocate them. If we need fewer, just
5191 if (regs
->num_regs
< num_regs
+ 1)
5193 regs
->num_regs
= num_regs
+ 1;
5194 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5195 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5196 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5205 /* These braces fend off a "empty body in an else-statement"
5206 warning under GCC when assert expands to nothing. */
5207 assert (bufp
->regs_allocated
== REGS_FIXED
);
5210 /* Convert the pointer data in `regstart' and `regend' to
5211 indices. Register zero has to be set differently,
5212 since we haven't kept track of any info for it. */
5213 if (regs
->num_regs
> 0)
5215 regs
->start
[0] = pos
;
5216 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5219 /* Go through the first `min (num_regs, regs->num_regs)'
5220 registers, since that is all we initialized. */
5221 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5223 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5224 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5228 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5230 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5234 /* If the regs structure we return has more elements than
5235 were in the pattern, set the extra elements to -1. If
5236 we (re)allocated the registers, this is the case,
5237 because we always allocate enough to have at least one
5239 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5240 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5241 } /* regs && !bufp->no_sub */
5243 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5244 nfailure_points_pushed
, nfailure_points_popped
,
5245 nfailure_points_pushed
- nfailure_points_popped
);
5246 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5248 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5250 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5256 /* Otherwise match next pattern command. */
5257 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5259 /* Ignore these. Used to ignore the n of succeed_n's which
5260 currently have n == 0. */
5262 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5266 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5269 /* Match the next n pattern characters exactly. The following
5270 byte in the pattern defines n, and the n bytes after that
5271 are the characters to match. */
5274 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5276 /* Remember the start point to rollback upon failure. */
5280 /* This is written out as an if-else so we don't waste time
5281 testing `translate' inside the loop. */
5282 if (RE_TRANSLATE_P (translate
))
5286 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5306 /* The cost of testing `translate' is comparatively small. */
5310 int pat_charlen
, buf_charlen
;
5311 unsigned int pat_ch
, buf_ch
;
5314 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
5315 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5317 if (TRANSLATE (buf_ch
) != pat_ch
)
5325 mcnt
-= pat_charlen
;
5331 unsigned int buf_ch
;
5335 TRANSLATE_VIA_MULTIBYTE (buf_ch
);
5347 /* Match any character except possibly a newline or a null. */
5353 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5356 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5357 buf_ch
= TRANSLATE (buf_ch
);
5359 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5361 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5362 && buf_ch
== '\000'))
5365 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5374 register unsigned int c
;
5375 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5378 /* Start of actual range_table, or end of bitmap if there is no
5380 re_char
*range_table
;
5382 /* Nonzero if there is a range table. */
5383 int range_table_exists
;
5385 /* Number of ranges of range table. This is not included
5386 in the initial byte-length of the command. */
5389 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5391 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5393 if (range_table_exists
)
5395 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5396 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5400 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5401 TRANSLATE_VIA_MULTIBYTE (c
); /* The character to match. */
5403 if (! multibyte
|| IS_REAL_ASCII (c
))
5404 { /* Lookup bitmap. */
5405 /* Cast to `unsigned' instead of `unsigned char' in
5406 case the bit list is a full 32 bytes long. */
5407 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5408 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5412 else if (range_table_exists
)
5414 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5416 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5417 | (class_bits
& BIT_MULTIBYTE
)
5418 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5419 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5420 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5421 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5424 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5428 if (range_table_exists
)
5429 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5431 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5433 if (!not) goto fail
;
5440 /* The beginning of a group is represented by start_memory.
5441 The argument is the register number. The text
5442 matched within the group is recorded (in the internal
5443 registers data structure) under the register number. */
5445 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5447 /* In case we need to undo this operation (via backtracking). */
5448 PUSH_FAILURE_REG ((unsigned int)*p
);
5451 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5452 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5454 /* Move past the register number and inner group count. */
5459 /* The stop_memory opcode represents the end of a group. Its
5460 argument is the same as start_memory's: the register number. */
5462 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5464 assert (!REG_UNSET (regstart
[*p
]));
5465 /* Strictly speaking, there should be code such as:
5467 assert (REG_UNSET (regend[*p]));
5468 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5470 But the only info to be pushed is regend[*p] and it is known to
5471 be UNSET, so there really isn't anything to push.
5472 Not pushing anything, on the other hand deprives us from the
5473 guarantee that regend[*p] is UNSET since undoing this operation
5474 will not reset its value properly. This is not important since
5475 the value will only be read on the next start_memory or at
5476 the very end and both events can only happen if this stop_memory
5480 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5482 /* Move past the register number and the inner group count. */
5487 /* \<digit> has been turned into a `duplicate' command which is
5488 followed by the numeric value of <digit> as the register number. */
5491 register re_char
*d2
, *dend2
;
5492 int regno
= *p
++; /* Get which register to match against. */
5493 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5495 /* Can't back reference a group which we've never matched. */
5496 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5499 /* Where in input to try to start matching. */
5500 d2
= regstart
[regno
];
5502 /* Remember the start point to rollback upon failure. */
5505 /* Where to stop matching; if both the place to start and
5506 the place to stop matching are in the same string, then
5507 set to the place to stop, otherwise, for now have to use
5508 the end of the first string. */
5510 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5511 == FIRST_STRING_P (regend
[regno
]))
5512 ? regend
[regno
] : end_match_1
);
5515 /* If necessary, advance to next segment in register
5519 if (dend2
== end_match_2
) break;
5520 if (dend2
== regend
[regno
]) break;
5522 /* End of string1 => advance to string2. */
5524 dend2
= regend
[regno
];
5526 /* At end of register contents => success */
5527 if (d2
== dend2
) break;
5529 /* If necessary, advance to next segment in data. */
5532 /* How many characters left in this segment to match. */
5535 /* Want how many consecutive characters we can match in
5536 one shot, so, if necessary, adjust the count. */
5537 if (mcnt
> dend2
- d2
)
5540 /* Compare that many; failure if mismatch, else move
5542 if (RE_TRANSLATE_P (translate
)
5543 ? bcmp_translate (d
, d2
, mcnt
, translate
, multibyte
)
5544 : memcmp (d
, d2
, mcnt
))
5549 d
+= mcnt
, d2
+= mcnt
;
5555 /* begline matches the empty string at the beginning of the string
5556 (unless `not_bol' is set in `bufp'), and after newlines. */
5558 DEBUG_PRINT1 ("EXECUTING begline.\n");
5560 if (AT_STRINGS_BEG (d
))
5562 if (!bufp
->not_bol
) break;
5567 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5571 /* In all other cases, we fail. */
5575 /* endline is the dual of begline. */
5577 DEBUG_PRINT1 ("EXECUTING endline.\n");
5579 if (AT_STRINGS_END (d
))
5581 if (!bufp
->not_eol
) break;
5585 PREFETCH_NOLIMIT ();
5592 /* Match at the very beginning of the data. */
5594 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5595 if (AT_STRINGS_BEG (d
))
5600 /* Match at the very end of the data. */
5602 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5603 if (AT_STRINGS_END (d
))
5608 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5609 pushes NULL as the value for the string on the stack. Then
5610 `POP_FAILURE_POINT' will keep the current value for the
5611 string, instead of restoring it. To see why, consider
5612 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5613 then the . fails against the \n. But the next thing we want
5614 to do is match the \n against the \n; if we restored the
5615 string value, we would be back at the foo.
5617 Because this is used only in specific cases, we don't need to
5618 check all the things that `on_failure_jump' does, to make
5619 sure the right things get saved on the stack. Hence we don't
5620 share its code. The only reason to push anything on the
5621 stack at all is that otherwise we would have to change
5622 `anychar's code to do something besides goto fail in this
5623 case; that seems worse than this. */
5624 case on_failure_keep_string_jump
:
5625 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5626 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5629 PUSH_FAILURE_POINT (p
- 3, NULL
);
5632 /* A nasty loop is introduced by the non-greedy *? and +?.
5633 With such loops, the stack only ever contains one failure point
5634 at a time, so that a plain on_failure_jump_loop kind of
5635 cycle detection cannot work. Worse yet, such a detection
5636 can not only fail to detect a cycle, but it can also wrongly
5637 detect a cycle (between different instantiations of the same
5639 So the method used for those nasty loops is a little different:
5640 We use a special cycle-detection-stack-frame which is pushed
5641 when the on_failure_jump_nastyloop failure-point is *popped*.
5642 This special frame thus marks the beginning of one iteration
5643 through the loop and we can hence easily check right here
5644 whether something matched between the beginning and the end of
5646 case on_failure_jump_nastyloop
:
5647 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5648 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5651 assert ((re_opcode_t
)p
[-4] == no_op
);
5654 CHECK_INFINITE_LOOP (p
- 4, d
);
5656 /* If there's a cycle, just continue without pushing
5657 this failure point. The failure point is the "try again"
5658 option, which shouldn't be tried.
5659 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5660 PUSH_FAILURE_POINT (p
- 3, d
);
5664 /* Simple loop detecting on_failure_jump: just check on the
5665 failure stack if the same spot was already hit earlier. */
5666 case on_failure_jump_loop
:
5668 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5669 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5673 CHECK_INFINITE_LOOP (p
- 3, d
);
5675 /* If there's a cycle, get out of the loop, as if the matching
5676 had failed. We used to just `goto fail' here, but that was
5677 aborting the search a bit too early: we want to keep the
5678 empty-loop-match and keep matching after the loop.
5679 We want (x?)*y\1z to match both xxyz and xxyxz. */
5682 PUSH_FAILURE_POINT (p
- 3, d
);
5687 /* Uses of on_failure_jump:
5689 Each alternative starts with an on_failure_jump that points
5690 to the beginning of the next alternative. Each alternative
5691 except the last ends with a jump that in effect jumps past
5692 the rest of the alternatives. (They really jump to the
5693 ending jump of the following alternative, because tensioning
5694 these jumps is a hassle.)
5696 Repeats start with an on_failure_jump that points past both
5697 the repetition text and either the following jump or
5698 pop_failure_jump back to this on_failure_jump. */
5699 case on_failure_jump
:
5700 IMMEDIATE_QUIT_CHECK
;
5701 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5702 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5705 PUSH_FAILURE_POINT (p
-3, d
);
5708 /* This operation is used for greedy *.
5709 Compare the beginning of the repeat with what in the
5710 pattern follows its end. If we can establish that there
5711 is nothing that they would both match, i.e., that we
5712 would have to backtrack because of (as in, e.g., `a*a')
5713 then we can use a non-backtracking loop based on
5714 on_failure_keep_string_jump instead of on_failure_jump. */
5715 case on_failure_jump_smart
:
5716 IMMEDIATE_QUIT_CHECK
;
5717 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5718 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5721 re_char
*p1
= p
; /* Next operation. */
5722 /* Here, we discard `const', making re_match non-reentrant. */
5723 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5724 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5726 p
-= 3; /* Reset so that we will re-execute the
5727 instruction once it's been changed. */
5729 EXTRACT_NUMBER (mcnt
, p2
- 2);
5731 /* Ensure this is a indeed the trivial kind of loop
5732 we are expecting. */
5733 assert (skip_one_char (p1
) == p2
- 3);
5734 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5735 DEBUG_STATEMENT (debug
+= 2);
5736 if (mutually_exclusive_p (bufp
, p1
, p2
))
5738 /* Use a fast `on_failure_keep_string_jump' loop. */
5739 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5740 *p3
= (unsigned char) on_failure_keep_string_jump
;
5741 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5745 /* Default to a safe `on_failure_jump' loop. */
5746 DEBUG_PRINT1 (" smart default => slow loop.\n");
5747 *p3
= (unsigned char) on_failure_jump
;
5749 DEBUG_STATEMENT (debug
-= 2);
5753 /* Unconditionally jump (without popping any failure points). */
5756 IMMEDIATE_QUIT_CHECK
;
5757 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5758 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5759 p
+= mcnt
; /* Do the jump. */
5760 DEBUG_PRINT2 ("(to %p).\n", p
);
5764 /* Have to succeed matching what follows at least n times.
5765 After that, handle like `on_failure_jump'. */
5767 /* Signedness doesn't matter since we only compare MCNT to 0. */
5768 EXTRACT_NUMBER (mcnt
, p
+ 2);
5769 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5771 /* Originally, mcnt is how many times we HAVE to succeed. */
5774 /* Here, we discard `const', making re_match non-reentrant. */
5775 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5778 PUSH_NUMBER (p2
, mcnt
);
5781 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5786 /* Signedness doesn't matter since we only compare MCNT to 0. */
5787 EXTRACT_NUMBER (mcnt
, p
+ 2);
5788 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5790 /* Originally, this is how many times we CAN jump. */
5793 /* Here, we discard `const', making re_match non-reentrant. */
5794 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5796 PUSH_NUMBER (p2
, mcnt
);
5797 goto unconditional_jump
;
5799 /* If don't have to jump any more, skip over the rest of command. */
5806 unsigned char *p2
; /* Location of the counter. */
5807 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5809 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5810 /* Here, we discard `const', making re_match non-reentrant. */
5811 p2
= (unsigned char*) p
+ mcnt
;
5812 /* Signedness doesn't matter since we only copy MCNT's bits . */
5813 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5814 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
5815 PUSH_NUMBER (p2
, mcnt
);
5821 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5822 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5824 /* We SUCCEED (or FAIL) in one of the following cases: */
5826 /* Case 1: D is at the beginning or the end of string. */
5827 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5831 /* C1 is the character before D, S1 is the syntax of C1, C2
5832 is the character at D, and S2 is the syntax of C2. */
5837 int offset
= PTR_TO_OFFSET (d
- 1);
5838 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5839 UPDATE_SYNTAX_TABLE (charpos
);
5841 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5844 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5846 PREFETCH_NOLIMIT ();
5847 GET_CHAR_AFTER (c2
, d
, dummy
);
5850 if (/* Case 2: Only one of S1 and S2 is Sword. */
5851 ((s1
== Sword
) != (s2
== Sword
))
5852 /* Case 3: Both of S1 and S2 are Sword, and macro
5853 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5854 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5863 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5865 /* We FAIL in one of the following cases: */
5867 /* Case 1: D is at the end of string. */
5868 if (AT_STRINGS_END (d
))
5872 /* C1 is the character before D, S1 is the syntax of C1, C2
5873 is the character at D, and S2 is the syntax of C2. */
5878 int offset
= PTR_TO_OFFSET (d
);
5879 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5880 UPDATE_SYNTAX_TABLE (charpos
);
5883 GET_CHAR_AFTER (c2
, d
, dummy
);
5886 /* Case 2: S2 is not Sword. */
5890 /* Case 3: D is not at the beginning of string ... */
5891 if (!AT_STRINGS_BEG (d
))
5893 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5895 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5899 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5901 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5908 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5910 /* We FAIL in one of the following cases: */
5912 /* Case 1: D is at the beginning of string. */
5913 if (AT_STRINGS_BEG (d
))
5917 /* C1 is the character before D, S1 is the syntax of C1, C2
5918 is the character at D, and S2 is the syntax of C2. */
5923 int offset
= PTR_TO_OFFSET (d
) - 1;
5924 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5925 UPDATE_SYNTAX_TABLE (charpos
);
5927 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5930 /* Case 2: S1 is not Sword. */
5934 /* Case 3: D is not at the end of string ... */
5935 if (!AT_STRINGS_END (d
))
5937 PREFETCH_NOLIMIT ();
5938 GET_CHAR_AFTER (c2
, d
, dummy
);
5940 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
5944 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5946 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5954 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
5956 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
5960 int offset
= PTR_TO_OFFSET (d
);
5961 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5962 UPDATE_SYNTAX_TABLE (pos1
);
5969 GET_CHAR_AFTER (c
, d
, len
);
5970 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
5978 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5979 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
5984 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5985 if (PTR_BYTE_POS (d
) != PT_BYTE
)
5990 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5991 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
5996 case notcategoryspec
:
5997 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
5999 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
6005 GET_CHAR_AFTER (c
, d
, len
);
6006 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6017 continue; /* Successfully executed one pattern command; keep going. */
6020 /* We goto here if a matching operation fails. */
6022 IMMEDIATE_QUIT_CHECK
;
6023 if (!FAIL_STACK_EMPTY ())
6026 /* A restart point is known. Restore to that state. */
6027 DEBUG_PRINT1 ("\nFAIL:\n");
6028 POP_FAILURE_POINT (str
, pat
);
6029 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
6031 case on_failure_keep_string_jump
:
6032 assert (str
== NULL
);
6033 goto continue_failure_jump
;
6035 case on_failure_jump_nastyloop
:
6036 assert ((re_opcode_t
)pat
[-2] == no_op
);
6037 PUSH_FAILURE_POINT (pat
- 2, str
);
6040 case on_failure_jump_loop
:
6041 case on_failure_jump
:
6044 continue_failure_jump
:
6045 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6050 /* A special frame used for nastyloops. */
6057 assert (p
>= bufp
->buffer
&& p
<= pend
);
6059 if (d
>= string1
&& d
<= end1
)
6063 break; /* Matching at this starting point really fails. */
6067 goto restore_best_regs
;
6071 return -1; /* Failure to match. */
6074 /* Subroutine definitions for re_match_2. */
6076 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6077 bytes; nonzero otherwise. */
6080 bcmp_translate (s1
, s2
, len
, translate
, multibyte
)
6083 RE_TRANSLATE_TYPE translate
;
6084 const int multibyte
;
6086 register re_char
*p1
= s1
, *p2
= s2
;
6087 re_char
*p1_end
= s1
+ len
;
6088 re_char
*p2_end
= s2
+ len
;
6090 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6091 different lengths, but relying on a single `len' would break this. -sm */
6092 while (p1
< p1_end
&& p2
< p2_end
)
6094 int p1_charlen
, p2_charlen
;
6095 re_wchar_t p1_ch
, p2_ch
;
6097 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6098 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6100 if (RE_TRANSLATE (translate
, p1_ch
)
6101 != RE_TRANSLATE (translate
, p2_ch
))
6104 p1
+= p1_charlen
, p2
+= p2_charlen
;
6107 if (p1
!= p1_end
|| p2
!= p2_end
)
6113 /* Entry points for GNU code. */
6115 /* re_compile_pattern is the GNU regular expression compiler: it
6116 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6117 Returns 0 if the pattern was valid, otherwise an error string.
6119 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6120 are set in BUFP on entry.
6122 We call regex_compile to do the actual compilation. */
6125 re_compile_pattern (pattern
, length
, bufp
)
6126 const char *pattern
;
6128 struct re_pattern_buffer
*bufp
;
6132 /* GNU code is written to assume at least RE_NREGS registers will be set
6133 (and at least one extra will be -1). */
6134 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6136 /* And GNU code determines whether or not to get register information
6137 by passing null for the REGS argument to re_match, etc., not by
6141 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6145 return gettext (re_error_msgid
[(int) ret
]);
6147 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6149 /* Entry points compatible with 4.2 BSD regex library. We don't define
6150 them unless specifically requested. */
6152 #if defined _REGEX_RE_COMP || defined _LIBC
6154 /* BSD has one and only one pattern buffer. */
6155 static struct re_pattern_buffer re_comp_buf
;
6159 /* Make these definitions weak in libc, so POSIX programs can redefine
6160 these names if they don't use our functions, and still use
6161 regcomp/regexec below without link errors. */
6171 if (!re_comp_buf
.buffer
)
6172 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6173 return (char *) gettext ("No previous regular expression");
6177 if (!re_comp_buf
.buffer
)
6179 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6180 if (re_comp_buf
.buffer
== NULL
)
6181 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6182 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6183 re_comp_buf
.allocated
= 200;
6185 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6186 if (re_comp_buf
.fastmap
== NULL
)
6187 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6188 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6191 /* Since `re_exec' always passes NULL for the `regs' argument, we
6192 don't need to initialize the pattern buffer fields which affect it. */
6194 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6199 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6200 return (char *) gettext (re_error_msgid
[(int) ret
]);
6211 const int len
= strlen (s
);
6213 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6215 #endif /* _REGEX_RE_COMP */
6217 /* POSIX.2 functions. Don't define these for Emacs. */
6221 /* regcomp takes a regular expression as a string and compiles it.
6223 PREG is a regex_t *. We do not expect any fields to be initialized,
6224 since POSIX says we shouldn't. Thus, we set
6226 `buffer' to the compiled pattern;
6227 `used' to the length of the compiled pattern;
6228 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6229 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6230 RE_SYNTAX_POSIX_BASIC;
6231 `fastmap' to an allocated space for the fastmap;
6232 `fastmap_accurate' to zero;
6233 `re_nsub' to the number of subexpressions in PATTERN.
6235 PATTERN is the address of the pattern string.
6237 CFLAGS is a series of bits which affect compilation.
6239 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6240 use POSIX basic syntax.
6242 If REG_NEWLINE is set, then . and [^...] don't match newline.
6243 Also, regexec will try a match beginning after every newline.
6245 If REG_ICASE is set, then we considers upper- and lowercase
6246 versions of letters to be equivalent when matching.
6248 If REG_NOSUB is set, then when PREG is passed to regexec, that
6249 routine will report only success or failure, and nothing about the
6252 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6253 the return codes and their meanings.) */
6256 regcomp (preg
, pattern
, cflags
)
6257 regex_t
*__restrict preg
;
6258 const char *__restrict pattern
;
6263 = (cflags
& REG_EXTENDED
) ?
6264 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6266 /* regex_compile will allocate the space for the compiled pattern. */
6268 preg
->allocated
= 0;
6271 /* Try to allocate space for the fastmap. */
6272 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6274 if (cflags
& REG_ICASE
)
6279 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6280 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6281 if (preg
->translate
== NULL
)
6282 return (int) REG_ESPACE
;
6284 /* Map uppercase characters to corresponding lowercase ones. */
6285 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6286 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6289 preg
->translate
= NULL
;
6291 /* If REG_NEWLINE is set, newlines are treated differently. */
6292 if (cflags
& REG_NEWLINE
)
6293 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6294 syntax
&= ~RE_DOT_NEWLINE
;
6295 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6298 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6300 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6302 /* POSIX says a null character in the pattern terminates it, so we
6303 can use strlen here in compiling the pattern. */
6304 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6306 /* POSIX doesn't distinguish between an unmatched open-group and an
6307 unmatched close-group: both are REG_EPAREN. */
6308 if (ret
== REG_ERPAREN
)
6311 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6312 { /* Compute the fastmap now, since regexec cannot modify the pattern
6314 re_compile_fastmap (preg
);
6315 if (preg
->can_be_null
)
6316 { /* The fastmap can't be used anyway. */
6317 free (preg
->fastmap
);
6318 preg
->fastmap
= NULL
;
6323 WEAK_ALIAS (__regcomp
, regcomp
)
6326 /* regexec searches for a given pattern, specified by PREG, in the
6329 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6330 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6331 least NMATCH elements, and we set them to the offsets of the
6332 corresponding matched substrings.
6334 EFLAGS specifies `execution flags' which affect matching: if
6335 REG_NOTBOL is set, then ^ does not match at the beginning of the
6336 string; if REG_NOTEOL is set, then $ does not match at the end.
6338 We return 0 if we find a match and REG_NOMATCH if not. */
6341 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6342 const regex_t
*__restrict preg
;
6343 const char *__restrict string
;
6345 regmatch_t pmatch
[__restrict_arr
];
6349 struct re_registers regs
;
6350 regex_t private_preg
;
6351 int len
= strlen (string
);
6352 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6354 private_preg
= *preg
;
6356 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6357 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6359 /* The user has told us exactly how many registers to return
6360 information about, via `nmatch'. We have to pass that on to the
6361 matching routines. */
6362 private_preg
.regs_allocated
= REGS_FIXED
;
6366 regs
.num_regs
= nmatch
;
6367 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6368 if (regs
.start
== NULL
)
6369 return (int) REG_NOMATCH
;
6370 regs
.end
= regs
.start
+ nmatch
;
6373 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6374 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6375 was a little bit longer but still only matching the real part.
6376 This works because the `endline' will check for a '\n' and will find a
6377 '\0', correctly deciding that this is not the end of a line.
6378 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6379 a convenient '\0' there. For all we know, the string could be preceded
6380 by '\n' which would throw things off. */
6382 /* Perform the searching operation. */
6383 ret
= re_search (&private_preg
, string
, len
,
6384 /* start: */ 0, /* range: */ len
,
6385 want_reg_info
? ®s
: (struct re_registers
*) 0);
6387 /* Copy the register information to the POSIX structure. */
6394 for (r
= 0; r
< nmatch
; r
++)
6396 pmatch
[r
].rm_so
= regs
.start
[r
];
6397 pmatch
[r
].rm_eo
= regs
.end
[r
];
6401 /* If we needed the temporary register info, free the space now. */
6405 /* We want zero return to mean success, unlike `re_search'. */
6406 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6408 WEAK_ALIAS (__regexec
, regexec
)
6411 /* Returns a message corresponding to an error code, ERRCODE, returned
6412 from either regcomp or regexec. We don't use PREG here. */
6415 regerror (errcode
, preg
, errbuf
, errbuf_size
)
6417 const regex_t
*preg
;
6425 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6426 /* Only error codes returned by the rest of the code should be passed
6427 to this routine. If we are given anything else, or if other regex
6428 code generates an invalid error code, then the program has a bug.
6429 Dump core so we can fix it. */
6432 msg
= gettext (re_error_msgid
[errcode
]);
6434 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6436 if (errbuf_size
!= 0)
6438 if (msg_size
> errbuf_size
)
6440 strncpy (errbuf
, msg
, errbuf_size
- 1);
6441 errbuf
[errbuf_size
- 1] = 0;
6444 strcpy (errbuf
, msg
);
6449 WEAK_ALIAS (__regerror
, regerror
)
6452 /* Free dynamically allocated space used by PREG. */
6458 if (preg
->buffer
!= NULL
)
6459 free (preg
->buffer
);
6460 preg
->buffer
= NULL
;
6462 preg
->allocated
= 0;
6465 if (preg
->fastmap
!= NULL
)
6466 free (preg
->fastmap
);
6467 preg
->fastmap
= NULL
;
6468 preg
->fastmap_accurate
= 0;
6470 if (preg
->translate
!= NULL
)
6471 free (preg
->translate
);
6472 preg
->translate
= NULL
;
6474 WEAK_ALIAS (__regfree
, regfree
)
6476 #endif /* not emacs */