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) 1993,94,95,96,97,98,99,2000,04 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, Ssymbol
= 2 };
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
['_'] = Ssymbol
;
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 symbeg
, /* Succeeds if at symbol beginning. */
684 symend
, /* Succeeds if at symbol end. */
686 /* Matches any character whose syntax is specified. Followed by
687 a byte which contains a syntax code, e.g., Sword. */
690 /* Matches any character whose syntax is not that specified. */
694 ,before_dot
, /* Succeeds if before point. */
695 at_dot
, /* Succeeds if at point. */
696 after_dot
, /* Succeeds if after point. */
698 /* Matches any character whose category-set contains the specified
699 category. The operator is followed by a byte which contains a
700 category code (mnemonic ASCII character). */
703 /* Matches any character whose category-set does not contain the
704 specified category. The operator is followed by a byte which
705 contains the category code (mnemonic ASCII character). */
710 /* Common operations on the compiled pattern. */
712 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
714 #define STORE_NUMBER(destination, number) \
716 (destination)[0] = (number) & 0377; \
717 (destination)[1] = (number) >> 8; \
720 /* Same as STORE_NUMBER, except increment DESTINATION to
721 the byte after where the number is stored. Therefore, DESTINATION
722 must be an lvalue. */
724 #define STORE_NUMBER_AND_INCR(destination, number) \
726 STORE_NUMBER (destination, number); \
727 (destination) += 2; \
730 /* Put into DESTINATION a number stored in two contiguous bytes starting
733 #define EXTRACT_NUMBER(destination, source) \
735 (destination) = *(source) & 0377; \
736 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
740 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
742 extract_number (dest
, source
)
746 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
747 *dest
= *source
& 0377;
751 # ifndef EXTRACT_MACROS /* To debug the macros. */
752 # undef EXTRACT_NUMBER
753 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
754 # endif /* not EXTRACT_MACROS */
758 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
759 SOURCE must be an lvalue. */
761 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
763 EXTRACT_NUMBER (destination, source); \
768 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
771 extract_number_and_incr (destination
, source
)
775 extract_number (destination
, *source
);
779 # ifndef EXTRACT_MACROS
780 # undef EXTRACT_NUMBER_AND_INCR
781 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
782 extract_number_and_incr (&dest, &src)
783 # endif /* not EXTRACT_MACROS */
787 /* Store a multibyte character in three contiguous bytes starting
788 DESTINATION, and increment DESTINATION to the byte after where the
789 character is stored. Therefore, DESTINATION must be an lvalue. */
791 #define STORE_CHARACTER_AND_INCR(destination, character) \
793 (destination)[0] = (character) & 0377; \
794 (destination)[1] = ((character) >> 8) & 0377; \
795 (destination)[2] = (character) >> 16; \
796 (destination) += 3; \
799 /* Put into DESTINATION a character stored in three contiguous bytes
800 starting at SOURCE. */
802 #define EXTRACT_CHARACTER(destination, source) \
804 (destination) = ((source)[0] \
805 | ((source)[1] << 8) \
806 | ((source)[2] << 16)); \
810 /* Macros for charset. */
812 /* Size of bitmap of charset P in bytes. P is a start of charset,
813 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
814 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
816 /* Nonzero if charset P has range table. */
817 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
819 /* Return the address of range table of charset P. But not the start
820 of table itself, but the before where the number of ranges is
821 stored. `2 +' means to skip re_opcode_t and size of bitmap,
822 and the 2 bytes of flags at the start of the range table. */
823 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
825 /* Extract the bit flags that start a range table. */
826 #define CHARSET_RANGE_TABLE_BITS(p) \
827 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
828 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
830 /* Test if C is listed in the bitmap of charset P. */
831 #define CHARSET_LOOKUP_BITMAP(p, c) \
832 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
833 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
835 /* Return the address of end of RANGE_TABLE. COUNT is number of
836 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
837 is start of range and end of range. `* 3' is size of each start
839 #define CHARSET_RANGE_TABLE_END(range_table, count) \
840 ((range_table) + (count) * 2 * 3)
842 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
843 COUNT is number of ranges in RANGE_TABLE. */
844 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
847 re_wchar_t range_start, range_end; \
849 re_char *range_table_end \
850 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
852 for (p = (range_table); p < range_table_end; p += 2 * 3) \
854 EXTRACT_CHARACTER (range_start, p); \
855 EXTRACT_CHARACTER (range_end, p + 3); \
857 if (range_start <= (c) && (c) <= range_end) \
866 /* Test if C is in range table of CHARSET. The flag NOT is negated if
867 C is listed in it. */
868 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
871 /* Number of ranges in range table. */ \
873 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
875 EXTRACT_NUMBER_AND_INCR (count, range_table); \
876 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
880 /* If DEBUG is defined, Regex prints many voluminous messages about what
881 it is doing (if the variable `debug' is nonzero). If linked with the
882 main program in `iregex.c', you can enter patterns and strings
883 interactively. And if linked with the main program in `main.c' and
884 the other test files, you can run the already-written tests. */
888 /* We use standard I/O for debugging. */
891 /* It is useful to test things that ``must'' be true when debugging. */
894 static int debug
= -100000;
896 # define DEBUG_STATEMENT(e) e
897 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
898 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
899 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
900 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
901 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
902 if (debug > 0) print_partial_compiled_pattern (s, e)
903 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
904 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
907 /* Print the fastmap in human-readable form. */
910 print_fastmap (fastmap
)
913 unsigned was_a_range
= 0;
916 while (i
< (1 << BYTEWIDTH
))
922 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
938 /* Print a compiled pattern string in human-readable form, starting at
939 the START pointer into it and ending just before the pointer END. */
942 print_partial_compiled_pattern (start
, end
)
952 fprintf (stderr
, "(null)\n");
956 /* Loop over pattern commands. */
959 fprintf (stderr
, "%d:\t", p
- start
);
961 switch ((re_opcode_t
) *p
++)
964 fprintf (stderr
, "/no_op");
968 fprintf (stderr
, "/succeed");
973 fprintf (stderr
, "/exactn/%d", mcnt
);
976 fprintf (stderr
, "/%c", *p
++);
982 fprintf (stderr
, "/start_memory/%d", *p
++);
986 fprintf (stderr
, "/stop_memory/%d", *p
++);
990 fprintf (stderr
, "/duplicate/%d", *p
++);
994 fprintf (stderr
, "/anychar");
1000 register int c
, last
= -100;
1001 register int in_range
= 0;
1002 int length
= CHARSET_BITMAP_SIZE (p
- 1);
1003 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
1005 fprintf (stderr
, "/charset [%s",
1006 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
1009 fprintf (stderr
, " !extends past end of pattern! ");
1011 for (c
= 0; c
< 256; c
++)
1013 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
1015 /* Are we starting a range? */
1016 if (last
+ 1 == c
&& ! in_range
)
1018 fprintf (stderr
, "-");
1021 /* Have we broken a range? */
1022 else if (last
+ 1 != c
&& in_range
)
1024 fprintf (stderr
, "%c", last
);
1029 fprintf (stderr
, "%c", c
);
1035 fprintf (stderr
, "%c", last
);
1037 fprintf (stderr
, "]");
1041 if (has_range_table
)
1044 fprintf (stderr
, "has-range-table");
1046 /* ??? Should print the range table; for now, just skip it. */
1047 p
+= 2; /* skip range table bits */
1048 EXTRACT_NUMBER_AND_INCR (count
, p
);
1049 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1055 fprintf (stderr
, "/begline");
1059 fprintf (stderr
, "/endline");
1062 case on_failure_jump
:
1063 extract_number_and_incr (&mcnt
, &p
);
1064 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1067 case on_failure_keep_string_jump
:
1068 extract_number_and_incr (&mcnt
, &p
);
1069 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1072 case on_failure_jump_nastyloop
:
1073 extract_number_and_incr (&mcnt
, &p
);
1074 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1077 case on_failure_jump_loop
:
1078 extract_number_and_incr (&mcnt
, &p
);
1079 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1082 case on_failure_jump_smart
:
1083 extract_number_and_incr (&mcnt
, &p
);
1084 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1088 extract_number_and_incr (&mcnt
, &p
);
1089 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1093 extract_number_and_incr (&mcnt
, &p
);
1094 extract_number_and_incr (&mcnt2
, &p
);
1095 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1099 extract_number_and_incr (&mcnt
, &p
);
1100 extract_number_and_incr (&mcnt2
, &p
);
1101 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1105 extract_number_and_incr (&mcnt
, &p
);
1106 extract_number_and_incr (&mcnt2
, &p
);
1107 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1111 fprintf (stderr
, "/wordbound");
1115 fprintf (stderr
, "/notwordbound");
1119 fprintf (stderr
, "/wordbeg");
1123 fprintf (stderr
, "/wordend");
1127 fprintf (stderr
, "/symbeg");
1131 fprintf (stderr
, "/symend");
1135 fprintf (stderr
, "/syntaxspec");
1137 fprintf (stderr
, "/%d", mcnt
);
1141 fprintf (stderr
, "/notsyntaxspec");
1143 fprintf (stderr
, "/%d", mcnt
);
1148 fprintf (stderr
, "/before_dot");
1152 fprintf (stderr
, "/at_dot");
1156 fprintf (stderr
, "/after_dot");
1160 fprintf (stderr
, "/categoryspec");
1162 fprintf (stderr
, "/%d", mcnt
);
1165 case notcategoryspec
:
1166 fprintf (stderr
, "/notcategoryspec");
1168 fprintf (stderr
, "/%d", mcnt
);
1173 fprintf (stderr
, "/begbuf");
1177 fprintf (stderr
, "/endbuf");
1181 fprintf (stderr
, "?%d", *(p
-1));
1184 fprintf (stderr
, "\n");
1187 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1192 print_compiled_pattern (bufp
)
1193 struct re_pattern_buffer
*bufp
;
1195 re_char
*buffer
= bufp
->buffer
;
1197 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1198 printf ("%ld bytes used/%ld bytes allocated.\n",
1199 bufp
->used
, bufp
->allocated
);
1201 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1203 printf ("fastmap: ");
1204 print_fastmap (bufp
->fastmap
);
1207 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1208 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1209 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1210 printf ("no_sub: %d\t", bufp
->no_sub
);
1211 printf ("not_bol: %d\t", bufp
->not_bol
);
1212 printf ("not_eol: %d\t", bufp
->not_eol
);
1213 printf ("syntax: %lx\n", bufp
->syntax
);
1215 /* Perhaps we should print the translate table? */
1220 print_double_string (where
, string1
, size1
, string2
, size2
)
1233 if (FIRST_STRING_P (where
))
1235 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1236 putchar (string1
[this_char
]);
1241 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1242 putchar (string2
[this_char
]);
1246 #else /* not DEBUG */
1251 # define DEBUG_STATEMENT(e)
1252 # define DEBUG_PRINT1(x)
1253 # define DEBUG_PRINT2(x1, x2)
1254 # define DEBUG_PRINT3(x1, x2, x3)
1255 # define DEBUG_PRINT4(x1, x2, x3, x4)
1256 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1257 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1259 #endif /* not DEBUG */
1261 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1262 also be assigned to arbitrarily: each pattern buffer stores its own
1263 syntax, so it can be changed between regex compilations. */
1264 /* This has no initializer because initialized variables in Emacs
1265 become read-only after dumping. */
1266 reg_syntax_t re_syntax_options
;
1269 /* Specify the precise syntax of regexps for compilation. This provides
1270 for compatibility for various utilities which historically have
1271 different, incompatible syntaxes.
1273 The argument SYNTAX is a bit mask comprised of the various bits
1274 defined in regex.h. We return the old syntax. */
1277 re_set_syntax (syntax
)
1278 reg_syntax_t syntax
;
1280 reg_syntax_t ret
= re_syntax_options
;
1282 re_syntax_options
= syntax
;
1285 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1287 /* Regexp to use to replace spaces, or NULL meaning don't. */
1288 static re_char
*whitespace_regexp
;
1291 re_set_whitespace_regexp (regexp
)
1294 whitespace_regexp
= regexp
;
1296 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1298 /* This table gives an error message for each of the error codes listed
1299 in regex.h. Obviously the order here has to be same as there.
1300 POSIX doesn't require that we do anything for REG_NOERROR,
1301 but why not be nice? */
1303 static const char *re_error_msgid
[] =
1305 gettext_noop ("Success"), /* REG_NOERROR */
1306 gettext_noop ("No match"), /* REG_NOMATCH */
1307 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1308 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1309 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1310 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1311 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1312 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1313 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1314 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1315 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1316 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1317 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1318 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1319 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1320 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1321 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1322 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1325 /* Avoiding alloca during matching, to placate r_alloc. */
1327 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1328 searching and matching functions should not call alloca. On some
1329 systems, alloca is implemented in terms of malloc, and if we're
1330 using the relocating allocator routines, then malloc could cause a
1331 relocation, which might (if the strings being searched are in the
1332 ralloc heap) shift the data out from underneath the regexp
1335 Here's another reason to avoid allocation: Emacs
1336 processes input from X in a signal handler; processing X input may
1337 call malloc; if input arrives while a matching routine is calling
1338 malloc, then we're scrod. But Emacs can't just block input while
1339 calling matching routines; then we don't notice interrupts when
1340 they come in. So, Emacs blocks input around all regexp calls
1341 except the matching calls, which it leaves unprotected, in the
1342 faith that they will not malloc. */
1344 /* Normally, this is fine. */
1345 #define MATCH_MAY_ALLOCATE
1347 /* When using GNU C, we are not REALLY using the C alloca, no matter
1348 what config.h may say. So don't take precautions for it. */
1353 /* The match routines may not allocate if (1) they would do it with malloc
1354 and (2) it's not safe for them to use malloc.
1355 Note that if REL_ALLOC is defined, matching would not use malloc for the
1356 failure stack, but we would still use it for the register vectors;
1357 so REL_ALLOC should not affect this. */
1358 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1359 # undef MATCH_MAY_ALLOCATE
1363 /* Failure stack declarations and macros; both re_compile_fastmap and
1364 re_match_2 use a failure stack. These have to be macros because of
1365 REGEX_ALLOCATE_STACK. */
1368 /* Approximate number of failure points for which to initially allocate space
1369 when matching. If this number is exceeded, we allocate more
1370 space, so it is not a hard limit. */
1371 #ifndef INIT_FAILURE_ALLOC
1372 # define INIT_FAILURE_ALLOC 20
1375 /* Roughly the maximum number of failure points on the stack. Would be
1376 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1377 This is a variable only so users of regex can assign to it; we never
1378 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1379 before using it, so it should probably be a byte-count instead. */
1380 # if defined MATCH_MAY_ALLOCATE
1381 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1382 whose default stack limit is 2mb. In order for a larger
1383 value to work reliably, you have to try to make it accord
1384 with the process stack limit. */
1385 size_t re_max_failures
= 40000;
1387 size_t re_max_failures
= 4000;
1390 union fail_stack_elt
1393 /* This should be the biggest `int' that's no bigger than a pointer. */
1397 typedef union fail_stack_elt fail_stack_elt_t
;
1401 fail_stack_elt_t
*stack
;
1403 size_t avail
; /* Offset of next open position. */
1404 size_t frame
; /* Offset of the cur constructed frame. */
1407 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1408 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1411 /* Define macros to initialize and free the failure stack.
1412 Do `return -2' if the alloc fails. */
1414 #ifdef MATCH_MAY_ALLOCATE
1415 # define INIT_FAIL_STACK() \
1417 fail_stack.stack = (fail_stack_elt_t *) \
1418 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1419 * sizeof (fail_stack_elt_t)); \
1421 if (fail_stack.stack == NULL) \
1424 fail_stack.size = INIT_FAILURE_ALLOC; \
1425 fail_stack.avail = 0; \
1426 fail_stack.frame = 0; \
1429 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1431 # define INIT_FAIL_STACK() \
1433 fail_stack.avail = 0; \
1434 fail_stack.frame = 0; \
1437 # define RESET_FAIL_STACK() ((void)0)
1441 /* Double the size of FAIL_STACK, up to a limit
1442 which allows approximately `re_max_failures' items.
1444 Return 1 if succeeds, and 0 if either ran out of memory
1445 allocating space for it or it was already too large.
1447 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1449 /* Factor to increase the failure stack size by
1450 when we increase it.
1451 This used to be 2, but 2 was too wasteful
1452 because the old discarded stacks added up to as much space
1453 were as ultimate, maximum-size stack. */
1454 #define FAIL_STACK_GROWTH_FACTOR 4
1456 #define GROW_FAIL_STACK(fail_stack) \
1457 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1458 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1460 : ((fail_stack).stack \
1461 = (fail_stack_elt_t *) \
1462 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1463 (fail_stack).size * sizeof (fail_stack_elt_t), \
1464 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1465 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1466 * FAIL_STACK_GROWTH_FACTOR))), \
1468 (fail_stack).stack == NULL \
1470 : ((fail_stack).size \
1471 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1472 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1473 * FAIL_STACK_GROWTH_FACTOR)) \
1474 / sizeof (fail_stack_elt_t)), \
1478 /* Push a pointer value onto the failure stack.
1479 Assumes the variable `fail_stack'. Probably should only
1480 be called from within `PUSH_FAILURE_POINT'. */
1481 #define PUSH_FAILURE_POINTER(item) \
1482 fail_stack.stack[fail_stack.avail++].pointer = (item)
1484 /* This pushes an integer-valued item onto the failure stack.
1485 Assumes the variable `fail_stack'. Probably should only
1486 be called from within `PUSH_FAILURE_POINT'. */
1487 #define PUSH_FAILURE_INT(item) \
1488 fail_stack.stack[fail_stack.avail++].integer = (item)
1490 /* Push a fail_stack_elt_t value onto the failure stack.
1491 Assumes the variable `fail_stack'. Probably should only
1492 be called from within `PUSH_FAILURE_POINT'. */
1493 #define PUSH_FAILURE_ELT(item) \
1494 fail_stack.stack[fail_stack.avail++] = (item)
1496 /* These three POP... operations complement the three PUSH... operations.
1497 All assume that `fail_stack' is nonempty. */
1498 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1499 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1500 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1502 /* Individual items aside from the registers. */
1503 #define NUM_NONREG_ITEMS 3
1505 /* Used to examine the stack (to detect infinite loops). */
1506 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1507 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1508 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1509 #define TOP_FAILURE_HANDLE() fail_stack.frame
1512 #define ENSURE_FAIL_STACK(space) \
1513 while (REMAINING_AVAIL_SLOTS <= space) { \
1514 if (!GROW_FAIL_STACK (fail_stack)) \
1516 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1517 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1520 /* Push register NUM onto the stack. */
1521 #define PUSH_FAILURE_REG(num) \
1523 char *destination; \
1524 ENSURE_FAIL_STACK(3); \
1525 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1526 num, regstart[num], regend[num]); \
1527 PUSH_FAILURE_POINTER (regstart[num]); \
1528 PUSH_FAILURE_POINTER (regend[num]); \
1529 PUSH_FAILURE_INT (num); \
1532 /* Change the counter's value to VAL, but make sure that it will
1533 be reset when backtracking. */
1534 #define PUSH_NUMBER(ptr,val) \
1536 char *destination; \
1538 ENSURE_FAIL_STACK(3); \
1539 EXTRACT_NUMBER (c, ptr); \
1540 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1541 PUSH_FAILURE_INT (c); \
1542 PUSH_FAILURE_POINTER (ptr); \
1543 PUSH_FAILURE_INT (-1); \
1544 STORE_NUMBER (ptr, val); \
1547 /* Pop a saved register off the stack. */
1548 #define POP_FAILURE_REG_OR_COUNT() \
1550 int reg = POP_FAILURE_INT (); \
1553 /* It's a counter. */ \
1554 /* Here, we discard `const', making re_match non-reentrant. */ \
1555 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1556 reg = POP_FAILURE_INT (); \
1557 STORE_NUMBER (ptr, reg); \
1558 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1562 regend[reg] = POP_FAILURE_POINTER (); \
1563 regstart[reg] = POP_FAILURE_POINTER (); \
1564 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1565 reg, regstart[reg], regend[reg]); \
1569 /* Check that we are not stuck in an infinite loop. */
1570 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1572 int failure = TOP_FAILURE_HANDLE (); \
1573 /* Check for infinite matching loops */ \
1574 while (failure > 0 \
1575 && (FAILURE_STR (failure) == string_place \
1576 || FAILURE_STR (failure) == NULL)) \
1578 assert (FAILURE_PAT (failure) >= bufp->buffer \
1579 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1580 if (FAILURE_PAT (failure) == pat_cur) \
1585 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1586 failure = NEXT_FAILURE_HANDLE(failure); \
1588 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1591 /* Push the information about the state we will need
1592 if we ever fail back to it.
1594 Requires variables fail_stack, regstart, regend and
1595 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1598 Does `return FAILURE_CODE' if runs out of memory. */
1600 #define PUSH_FAILURE_POINT(pattern, string_place) \
1602 char *destination; \
1603 /* Must be int, so when we don't save any registers, the arithmetic \
1604 of 0 + -1 isn't done as unsigned. */ \
1606 DEBUG_STATEMENT (nfailure_points_pushed++); \
1607 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1608 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1609 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1611 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1613 DEBUG_PRINT1 ("\n"); \
1615 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1616 PUSH_FAILURE_INT (fail_stack.frame); \
1618 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1619 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1620 DEBUG_PRINT1 ("'\n"); \
1621 PUSH_FAILURE_POINTER (string_place); \
1623 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1624 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1625 PUSH_FAILURE_POINTER (pattern); \
1627 /* Close the frame by moving the frame pointer past it. */ \
1628 fail_stack.frame = fail_stack.avail; \
1631 /* Estimate the size of data pushed by a typical failure stack entry.
1632 An estimate is all we need, because all we use this for
1633 is to choose a limit for how big to make the failure stack. */
1634 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1635 #define TYPICAL_FAILURE_SIZE 20
1637 /* How many items can still be added to the stack without overflowing it. */
1638 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1641 /* Pops what PUSH_FAIL_STACK pushes.
1643 We restore into the parameters, all of which should be lvalues:
1644 STR -- the saved data position.
1645 PAT -- the saved pattern position.
1646 REGSTART, REGEND -- arrays of string positions.
1648 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1649 `pend', `string1', `size1', `string2', and `size2'. */
1651 #define POP_FAILURE_POINT(str, pat) \
1653 assert (!FAIL_STACK_EMPTY ()); \
1655 /* Remove failure points and point to how many regs pushed. */ \
1656 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1657 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1658 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1660 /* Pop the saved registers. */ \
1661 while (fail_stack.frame < fail_stack.avail) \
1662 POP_FAILURE_REG_OR_COUNT (); \
1664 pat = POP_FAILURE_POINTER (); \
1665 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1666 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1668 /* If the saved string location is NULL, it came from an \
1669 on_failure_keep_string_jump opcode, and we want to throw away the \
1670 saved NULL, thus retaining our current position in the string. */ \
1671 str = POP_FAILURE_POINTER (); \
1672 DEBUG_PRINT2 (" Popping string %p: `", str); \
1673 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1674 DEBUG_PRINT1 ("'\n"); \
1676 fail_stack.frame = POP_FAILURE_INT (); \
1677 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1679 assert (fail_stack.avail >= 0); \
1680 assert (fail_stack.frame <= fail_stack.avail); \
1682 DEBUG_STATEMENT (nfailure_points_popped++); \
1683 } while (0) /* POP_FAILURE_POINT */
1687 /* Registers are set to a sentinel when they haven't yet matched. */
1688 #define REG_UNSET(e) ((e) == NULL)
1690 /* Subroutine declarations and macros for regex_compile. */
1692 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1693 reg_syntax_t syntax
,
1694 struct re_pattern_buffer
*bufp
));
1695 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1696 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1697 int arg1
, int arg2
));
1698 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1699 int arg
, unsigned char *end
));
1700 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1701 int arg1
, int arg2
, unsigned char *end
));
1702 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1704 reg_syntax_t syntax
));
1705 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1707 reg_syntax_t syntax
));
1708 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1709 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1710 char *fastmap
, const int multibyte
));
1712 /* Fetch the next character in the uncompiled pattern, with no
1714 #define PATFETCH(c) \
1717 if (p == pend) return REG_EEND; \
1718 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1723 /* If `translate' is non-null, return translate[D], else just D. We
1724 cast the subscript to translate because some data is declared as
1725 `char *', to avoid warnings when a string constant is passed. But
1726 when we use a character as a subscript we must make it unsigned. */
1728 # define TRANSLATE(d) \
1729 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1733 /* Macros for outputting the compiled pattern into `buffer'. */
1735 /* If the buffer isn't allocated when it comes in, use this. */
1736 #define INIT_BUF_SIZE 32
1738 /* Make sure we have at least N more bytes of space in buffer. */
1739 #define GET_BUFFER_SPACE(n) \
1740 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1743 /* Make sure we have one more byte of buffer space and then add C to it. */
1744 #define BUF_PUSH(c) \
1746 GET_BUFFER_SPACE (1); \
1747 *b++ = (unsigned char) (c); \
1751 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1752 #define BUF_PUSH_2(c1, c2) \
1754 GET_BUFFER_SPACE (2); \
1755 *b++ = (unsigned char) (c1); \
1756 *b++ = (unsigned char) (c2); \
1760 /* As with BUF_PUSH_2, except for three bytes. */
1761 #define BUF_PUSH_3(c1, c2, c3) \
1763 GET_BUFFER_SPACE (3); \
1764 *b++ = (unsigned char) (c1); \
1765 *b++ = (unsigned char) (c2); \
1766 *b++ = (unsigned char) (c3); \
1770 /* Store a jump with opcode OP at LOC to location TO. We store a
1771 relative address offset by the three bytes the jump itself occupies. */
1772 #define STORE_JUMP(op, loc, to) \
1773 store_op1 (op, loc, (to) - (loc) - 3)
1775 /* Likewise, for a two-argument jump. */
1776 #define STORE_JUMP2(op, loc, to, arg) \
1777 store_op2 (op, loc, (to) - (loc) - 3, arg)
1779 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1780 #define INSERT_JUMP(op, loc, to) \
1781 insert_op1 (op, loc, (to) - (loc) - 3, b)
1783 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1784 #define INSERT_JUMP2(op, loc, to, arg) \
1785 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1788 /* This is not an arbitrary limit: the arguments which represent offsets
1789 into the pattern are two bytes long. So if 2^15 bytes turns out to
1790 be too small, many things would have to change. */
1791 # define MAX_BUF_SIZE (1L << 15)
1793 #if 0 /* This is when we thought it could be 2^16 bytes. */
1794 /* Any other compiler which, like MSC, has allocation limit below 2^16
1795 bytes will have to use approach similar to what was done below for
1796 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1797 reallocating to 0 bytes. Such thing is not going to work too well.
1798 You have been warned!! */
1799 #if defined _MSC_VER && !defined WIN32
1800 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1801 # define MAX_BUF_SIZE 65500L
1803 # define MAX_BUF_SIZE (1L << 16)
1807 /* Extend the buffer by twice its current size via realloc and
1808 reset the pointers that pointed into the old block to point to the
1809 correct places in the new one. If extending the buffer results in it
1810 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1811 #if __BOUNDED_POINTERS__
1812 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1813 # define MOVE_BUFFER_POINTER(P) \
1814 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1815 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1818 SET_HIGH_BOUND (b); \
1819 SET_HIGH_BOUND (begalt); \
1820 if (fixup_alt_jump) \
1821 SET_HIGH_BOUND (fixup_alt_jump); \
1823 SET_HIGH_BOUND (laststart); \
1824 if (pending_exact) \
1825 SET_HIGH_BOUND (pending_exact); \
1828 # define MOVE_BUFFER_POINTER(P) (P) += incr
1829 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1831 #define EXTEND_BUFFER() \
1833 re_char *old_buffer = bufp->buffer; \
1834 if (bufp->allocated == MAX_BUF_SIZE) \
1836 bufp->allocated <<= 1; \
1837 if (bufp->allocated > MAX_BUF_SIZE) \
1838 bufp->allocated = MAX_BUF_SIZE; \
1839 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1840 if (bufp->buffer == NULL) \
1841 return REG_ESPACE; \
1842 /* If the buffer moved, move all the pointers into it. */ \
1843 if (old_buffer != bufp->buffer) \
1845 int incr = bufp->buffer - old_buffer; \
1846 MOVE_BUFFER_POINTER (b); \
1847 MOVE_BUFFER_POINTER (begalt); \
1848 if (fixup_alt_jump) \
1849 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1851 MOVE_BUFFER_POINTER (laststart); \
1852 if (pending_exact) \
1853 MOVE_BUFFER_POINTER (pending_exact); \
1855 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1859 /* Since we have one byte reserved for the register number argument to
1860 {start,stop}_memory, the maximum number of groups we can report
1861 things about is what fits in that byte. */
1862 #define MAX_REGNUM 255
1864 /* But patterns can have more than `MAX_REGNUM' registers. We just
1865 ignore the excess. */
1866 typedef int regnum_t
;
1869 /* Macros for the compile stack. */
1871 /* Since offsets can go either forwards or backwards, this type needs to
1872 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1873 /* int may be not enough when sizeof(int) == 2. */
1874 typedef long pattern_offset_t
;
1878 pattern_offset_t begalt_offset
;
1879 pattern_offset_t fixup_alt_jump
;
1880 pattern_offset_t laststart_offset
;
1882 } compile_stack_elt_t
;
1887 compile_stack_elt_t
*stack
;
1889 unsigned avail
; /* Offset of next open position. */
1890 } compile_stack_type
;
1893 #define INIT_COMPILE_STACK_SIZE 32
1895 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1896 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1898 /* The next available element. */
1899 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1901 /* Explicit quit checking is only used on NTemacs. */
1902 #if defined WINDOWSNT && defined emacs && defined QUIT
1903 extern int immediate_quit
;
1904 # define IMMEDIATE_QUIT_CHECK \
1906 if (immediate_quit) QUIT; \
1909 # define IMMEDIATE_QUIT_CHECK ((void)0)
1912 /* Structure to manage work area for range table. */
1913 struct range_table_work_area
1915 int *table
; /* actual work area. */
1916 int allocated
; /* allocated size for work area in bytes. */
1917 int used
; /* actually used size in words. */
1918 int bits
; /* flag to record character classes */
1921 /* Make sure that WORK_AREA can hold more N multibyte characters.
1922 This is used only in set_image_of_range and set_image_of_range_1.
1923 It expects WORK_AREA to be a pointer.
1924 If it can't get the space, it returns from the surrounding function. */
1926 #define EXTEND_RANGE_TABLE(work_area, n) \
1928 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1930 extend_range_table_work_area (&work_area); \
1931 if ((work_area).table == 0) \
1932 return (REG_ESPACE); \
1936 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1937 (work_area).bits |= (bit)
1939 /* Bits used to implement the multibyte-part of the various character classes
1940 such as [:alnum:] in a charset's range table. */
1941 #define BIT_WORD 0x1
1942 #define BIT_LOWER 0x2
1943 #define BIT_PUNCT 0x4
1944 #define BIT_SPACE 0x8
1945 #define BIT_UPPER 0x10
1946 #define BIT_MULTIBYTE 0x20
1948 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1949 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1951 EXTEND_RANGE_TABLE ((work_area), 2); \
1952 (work_area).table[(work_area).used++] = (range_start); \
1953 (work_area).table[(work_area).used++] = (range_end); \
1956 /* Free allocated memory for WORK_AREA. */
1957 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1959 if ((work_area).table) \
1960 free ((work_area).table); \
1963 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1964 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1965 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1966 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1969 /* Set the bit for character C in a list. */
1970 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1975 /* Store characters in the rage range C0 to C1 in WORK_AREA while
1976 translating them and paying attention to the continuity of
1977 translated characters.
1979 Implementation note: It is better to implement this fairly big
1980 macro by a function, but it's not that easy because macros called
1981 in this macro assume various local variables already declared. */
1983 #define SETUP_MULTIBYTE_RANGE(work_area, c0, c1) \
1985 re_wchar_t c, t, t_last; \
1989 t_last = multibyte ? TRANSLATE (c) : TRANSLATE (MAKE_CHAR_MULTIBYTE (c)); \
1990 for (c++, n = 1; c <= (c1); c++, n++) \
1992 t = multibyte ? TRANSLATE (c) : TRANSLATE (MAKE_CHAR_MULTIBYTE (c)); \
1993 if (t_last + n == t) \
1995 SET_RANGE_TABLE_WORK_AREA ((work_area), t_last, t_last + n - 1); \
2000 SET_RANGE_TABLE_WORK_AREA ((work_area), t_last, t_last + n - 1); \
2005 /* Get the next unsigned number in the uncompiled pattern. */
2006 #define GET_UNSIGNED_NUMBER(num) \
2009 FREE_STACK_RETURN (REG_EBRACE); \
2013 while ('0' <= c && c <= '9') \
2019 num = num * 10 + c - '0'; \
2020 if (num / 10 != prev) \
2021 FREE_STACK_RETURN (REG_BADBR); \
2023 FREE_STACK_RETURN (REG_EBRACE); \
2029 #if ! WIDE_CHAR_SUPPORT
2031 /* Map a string to the char class it names (if any). */
2036 const char *string
= str
;
2037 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2038 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2039 else if (STREQ (string
, "word")) return RECC_WORD
;
2040 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2041 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2042 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2043 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2044 else if (STREQ (string
, "print")) return RECC_PRINT
;
2045 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2046 else if (STREQ (string
, "space")) return RECC_SPACE
;
2047 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2048 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2049 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2050 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2051 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2052 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2053 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2057 /* True iff CH is in the char class CC. */
2059 re_iswctype (ch
, cc
)
2065 case RECC_ALNUM
: return ISALNUM (ch
);
2066 case RECC_ALPHA
: return ISALPHA (ch
);
2067 case RECC_BLANK
: return ISBLANK (ch
);
2068 case RECC_CNTRL
: return ISCNTRL (ch
);
2069 case RECC_DIGIT
: return ISDIGIT (ch
);
2070 case RECC_GRAPH
: return ISGRAPH (ch
);
2071 case RECC_LOWER
: return ISLOWER (ch
);
2072 case RECC_PRINT
: return ISPRINT (ch
);
2073 case RECC_PUNCT
: return ISPUNCT (ch
);
2074 case RECC_SPACE
: return ISSPACE (ch
);
2075 case RECC_UPPER
: return ISUPPER (ch
);
2076 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2077 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2078 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2079 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2080 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2081 case RECC_WORD
: return ISWORD (ch
);
2082 case RECC_ERROR
: return false;
2088 /* Return a bit-pattern to use in the range-table bits to match multibyte
2089 chars of class CC. */
2091 re_wctype_to_bit (cc
)
2096 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2097 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2098 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2099 case RECC_LOWER
: return BIT_LOWER
;
2100 case RECC_UPPER
: return BIT_UPPER
;
2101 case RECC_PUNCT
: return BIT_PUNCT
;
2102 case RECC_SPACE
: return BIT_SPACE
;
2103 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2104 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2111 /* Filling in the work area of a range. */
2113 /* Actually extend the space in WORK_AREA. */
2116 extend_range_table_work_area (work_area
)
2117 struct range_table_work_area
*work_area
;
2119 work_area
->allocated
+= 16 * sizeof (int);
2120 if (work_area
->table
)
2122 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2125 = (int *) malloc (work_area
->allocated
);
2131 /* Carefully find the ranges of codes that are equivalent
2132 under case conversion to the range start..end when passed through
2133 TRANSLATE. Handle the case where non-letters can come in between
2134 two upper-case letters (which happens in Latin-1).
2135 Also handle the case of groups of more than 2 case-equivalent chars.
2137 The basic method is to look at consecutive characters and see
2138 if they can form a run that can be handled as one.
2140 Returns -1 if successful, REG_ESPACE if ran out of space. */
2143 set_image_of_range_1 (work_area
, start
, end
, translate
)
2144 RE_TRANSLATE_TYPE translate
;
2145 struct range_table_work_area
*work_area
;
2146 re_wchar_t start
, end
;
2148 /* `one_case' indicates a character, or a run of characters,
2149 each of which is an isolate (no case-equivalents).
2150 This includes all ASCII non-letters.
2152 `two_case' indicates a character, or a run of characters,
2153 each of which has two case-equivalent forms.
2154 This includes all ASCII letters.
2156 `strange' indicates a character that has more than one
2159 enum case_type
{one_case
, two_case
, strange
};
2161 /* Describe the run that is in progress,
2162 which the next character can try to extend.
2163 If run_type is strange, that means there really is no run.
2164 If run_type is one_case, then run_start...run_end is the run.
2165 If run_type is two_case, then the run is run_start...run_end,
2166 and the case-equivalents end at run_eqv_end. */
2168 enum case_type run_type
= strange
;
2169 int run_start
, run_end
, run_eqv_end
;
2171 Lisp_Object eqv_table
;
2173 if (!RE_TRANSLATE_P (translate
))
2175 EXTEND_RANGE_TABLE (work_area
, 2);
2176 work_area
->table
[work_area
->used
++] = (start
);
2177 work_area
->table
[work_area
->used
++] = (end
);
2181 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2183 for (; start
<= end
; start
++)
2185 enum case_type this_type
;
2186 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2187 int minchar
, maxchar
;
2189 /* Classify this character */
2191 this_type
= one_case
;
2192 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2193 this_type
= two_case
;
2195 this_type
= strange
;
2198 minchar
= start
, maxchar
= eqv
;
2200 minchar
= eqv
, maxchar
= start
;
2202 /* Can this character extend the run in progress? */
2203 if (this_type
== strange
|| this_type
!= run_type
2204 || !(minchar
== run_end
+ 1
2205 && (run_type
== two_case
2206 ? maxchar
== run_eqv_end
+ 1 : 1)))
2209 Record each of its equivalent ranges. */
2210 if (run_type
== one_case
)
2212 EXTEND_RANGE_TABLE (work_area
, 2);
2213 work_area
->table
[work_area
->used
++] = run_start
;
2214 work_area
->table
[work_area
->used
++] = run_end
;
2216 else if (run_type
== two_case
)
2218 EXTEND_RANGE_TABLE (work_area
, 4);
2219 work_area
->table
[work_area
->used
++] = run_start
;
2220 work_area
->table
[work_area
->used
++] = run_end
;
2221 work_area
->table
[work_area
->used
++]
2222 = RE_TRANSLATE (eqv_table
, run_start
);
2223 work_area
->table
[work_area
->used
++]
2224 = RE_TRANSLATE (eqv_table
, run_end
);
2229 if (this_type
== strange
)
2231 /* For a strange character, add each of its equivalents, one
2232 by one. Don't start a range. */
2235 EXTEND_RANGE_TABLE (work_area
, 2);
2236 work_area
->table
[work_area
->used
++] = eqv
;
2237 work_area
->table
[work_area
->used
++] = eqv
;
2238 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2240 while (eqv
!= start
);
2243 /* Add this char to the run, or start a new run. */
2244 else if (run_type
== strange
)
2246 /* Initialize a new range. */
2247 run_type
= this_type
;
2250 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2254 /* Extend a running range. */
2256 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2260 /* If a run is still in progress at the end, finish it now
2261 by recording its equivalent ranges. */
2262 if (run_type
== one_case
)
2264 EXTEND_RANGE_TABLE (work_area
, 2);
2265 work_area
->table
[work_area
->used
++] = run_start
;
2266 work_area
->table
[work_area
->used
++] = run_end
;
2268 else if (run_type
== two_case
)
2270 EXTEND_RANGE_TABLE (work_area
, 4);
2271 work_area
->table
[work_area
->used
++] = run_start
;
2272 work_area
->table
[work_area
->used
++] = run_end
;
2273 work_area
->table
[work_area
->used
++]
2274 = RE_TRANSLATE (eqv_table
, run_start
);
2275 work_area
->table
[work_area
->used
++]
2276 = RE_TRANSLATE (eqv_table
, run_end
);
2284 /* Record the the image of the range start..end when passed through
2285 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2286 and is not even necessarily contiguous.
2287 Normally we approximate it with the smallest contiguous range that contains
2288 all the chars we need. However, for Latin-1 we go to extra effort
2291 This function is not called for ASCII ranges.
2293 Returns -1 if successful, REG_ESPACE if ran out of space. */
2296 set_image_of_range (work_area
, start
, end
, translate
)
2297 RE_TRANSLATE_TYPE translate
;
2298 struct range_table_work_area
*work_area
;
2299 re_wchar_t start
, end
;
2301 re_wchar_t cmin
, cmax
;
2304 /* For Latin-1 ranges, use set_image_of_range_1
2305 to get proper handling of ranges that include letters and nonletters.
2306 For a range that includes the whole of Latin-1, this is not necessary.
2307 For other character sets, we don't bother to get this right. */
2308 if (RE_TRANSLATE_P (translate
) && start
< 04400
2309 && !(start
< 04200 && end
>= 04377))
2316 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2326 EXTEND_RANGE_TABLE (work_area
, 2);
2327 work_area
->table
[work_area
->used
++] = (start
);
2328 work_area
->table
[work_area
->used
++] = (end
);
2330 cmin
= -1, cmax
= -1;
2332 if (RE_TRANSLATE_P (translate
))
2336 for (ch
= start
; ch
<= end
; ch
++)
2338 re_wchar_t c
= TRANSLATE (ch
);
2339 if (! (start
<= c
&& c
<= end
))
2345 cmin
= MIN (cmin
, c
);
2346 cmax
= MAX (cmax
, c
);
2353 EXTEND_RANGE_TABLE (work_area
, 2);
2354 work_area
->table
[work_area
->used
++] = (cmin
);
2355 work_area
->table
[work_area
->used
++] = (cmax
);
2363 #ifndef MATCH_MAY_ALLOCATE
2365 /* If we cannot allocate large objects within re_match_2_internal,
2366 we make the fail stack and register vectors global.
2367 The fail stack, we grow to the maximum size when a regexp
2369 The register vectors, we adjust in size each time we
2370 compile a regexp, according to the number of registers it needs. */
2372 static fail_stack_type fail_stack
;
2374 /* Size with which the following vectors are currently allocated.
2375 That is so we can make them bigger as needed,
2376 but never make them smaller. */
2377 static int regs_allocated_size
;
2379 static re_char
** regstart
, ** regend
;
2380 static re_char
**best_regstart
, **best_regend
;
2382 /* Make the register vectors big enough for NUM_REGS registers,
2383 but don't make them smaller. */
2386 regex_grow_registers (num_regs
)
2389 if (num_regs
> regs_allocated_size
)
2391 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2392 RETALLOC_IF (regend
, num_regs
, re_char
*);
2393 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2394 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2396 regs_allocated_size
= num_regs
;
2400 #endif /* not MATCH_MAY_ALLOCATE */
2402 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2406 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2407 Returns one of error codes defined in `regex.h', or zero for success.
2409 Assumes the `allocated' (and perhaps `buffer') and `translate'
2410 fields are set in BUFP on entry.
2412 If it succeeds, results are put in BUFP (if it returns an error, the
2413 contents of BUFP are undefined):
2414 `buffer' is the compiled pattern;
2415 `syntax' is set to SYNTAX;
2416 `used' is set to the length of the compiled pattern;
2417 `fastmap_accurate' is zero;
2418 `re_nsub' is the number of subexpressions in PATTERN;
2419 `not_bol' and `not_eol' are zero;
2421 The `fastmap' field is neither examined nor set. */
2423 /* Insert the `jump' from the end of last alternative to "here".
2424 The space for the jump has already been allocated. */
2425 #define FIXUP_ALT_JUMP() \
2427 if (fixup_alt_jump) \
2428 STORE_JUMP (jump, fixup_alt_jump, b); \
2432 /* Return, freeing storage we allocated. */
2433 #define FREE_STACK_RETURN(value) \
2435 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2436 free (compile_stack.stack); \
2440 static reg_errcode_t
2441 regex_compile (pattern
, size
, syntax
, bufp
)
2444 reg_syntax_t syntax
;
2445 struct re_pattern_buffer
*bufp
;
2447 /* We fetch characters from PATTERN here. */
2448 register re_wchar_t c
, c1
;
2450 /* A random temporary spot in PATTERN. */
2453 /* Points to the end of the buffer, where we should append. */
2454 register unsigned char *b
;
2456 /* Keeps track of unclosed groups. */
2457 compile_stack_type compile_stack
;
2459 /* Points to the current (ending) position in the pattern. */
2461 /* `const' makes AIX compiler fail. */
2462 unsigned char *p
= pattern
;
2464 re_char
*p
= pattern
;
2466 re_char
*pend
= pattern
+ size
;
2468 /* How to translate the characters in the pattern. */
2469 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2471 /* Address of the count-byte of the most recently inserted `exactn'
2472 command. This makes it possible to tell if a new exact-match
2473 character can be added to that command or if the character requires
2474 a new `exactn' command. */
2475 unsigned char *pending_exact
= 0;
2477 /* Address of start of the most recently finished expression.
2478 This tells, e.g., postfix * where to find the start of its
2479 operand. Reset at the beginning of groups and alternatives. */
2480 unsigned char *laststart
= 0;
2482 /* Address of beginning of regexp, or inside of last group. */
2483 unsigned char *begalt
;
2485 /* Place in the uncompiled pattern (i.e., the {) to
2486 which to go back if the interval is invalid. */
2487 re_char
*beg_interval
;
2489 /* Address of the place where a forward jump should go to the end of
2490 the containing expression. Each alternative of an `or' -- except the
2491 last -- ends with a forward jump of this sort. */
2492 unsigned char *fixup_alt_jump
= 0;
2494 /* Counts open-groups as they are encountered. Remembered for the
2495 matching close-group on the compile stack, so the same register
2496 number is put in the stop_memory as the start_memory. */
2497 regnum_t regnum
= 0;
2499 /* Work area for range table of charset. */
2500 struct range_table_work_area range_table_work
;
2502 /* If the object matched can contain multibyte characters. */
2503 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2505 /* If a target of matching can contain multibyte characters. */
2506 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
2508 /* Nonzero if we have pushed down into a subpattern. */
2509 int in_subpattern
= 0;
2511 /* These hold the values of p, pattern, and pend from the main
2512 pattern when we have pushed into a subpattern. */
2514 re_char
*main_pattern
;
2519 DEBUG_PRINT1 ("\nCompiling pattern: ");
2522 unsigned debug_count
;
2524 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2525 putchar (pattern
[debug_count
]);
2530 /* Initialize the compile stack. */
2531 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2532 if (compile_stack
.stack
== NULL
)
2535 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2536 compile_stack
.avail
= 0;
2538 range_table_work
.table
= 0;
2539 range_table_work
.allocated
= 0;
2541 /* Initialize the pattern buffer. */
2542 bufp
->syntax
= syntax
;
2543 bufp
->fastmap_accurate
= 0;
2544 bufp
->not_bol
= bufp
->not_eol
= 0;
2546 /* Set `used' to zero, so that if we return an error, the pattern
2547 printer (for debugging) will think there's no pattern. We reset it
2551 /* Always count groups, whether or not bufp->no_sub is set. */
2554 #if !defined emacs && !defined SYNTAX_TABLE
2555 /* Initialize the syntax table. */
2556 init_syntax_once ();
2559 if (bufp
->allocated
== 0)
2562 { /* If zero allocated, but buffer is non-null, try to realloc
2563 enough space. This loses if buffer's address is bogus, but
2564 that is the user's responsibility. */
2565 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2568 { /* Caller did not allocate a buffer. Do it for them. */
2569 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2571 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2573 bufp
->allocated
= INIT_BUF_SIZE
;
2576 begalt
= b
= bufp
->buffer
;
2578 /* Loop through the uncompiled pattern until we're at the end. */
2583 /* If this is the end of an included regexp,
2584 pop back to the main regexp and try again. */
2588 pattern
= main_pattern
;
2593 /* If this is the end of the main regexp, we are done. */
2605 /* If there's no special whitespace regexp, treat
2606 spaces normally. And don't try to do this recursively. */
2607 if (!whitespace_regexp
|| in_subpattern
)
2610 /* Peek past following spaces. */
2617 /* If the spaces are followed by a repetition op,
2618 treat them normally. */
2620 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2621 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2624 /* Replace the spaces with the whitespace regexp. */
2628 main_pattern
= pattern
;
2629 p
= pattern
= whitespace_regexp
;
2630 pend
= p
+ strlen (p
);
2636 if ( /* If at start of pattern, it's an operator. */
2638 /* If context independent, it's an operator. */
2639 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2640 /* Otherwise, depends on what's come before. */
2641 || at_begline_loc_p (pattern
, p
, syntax
))
2642 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2651 if ( /* If at end of pattern, it's an operator. */
2653 /* If context independent, it's an operator. */
2654 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2655 /* Otherwise, depends on what's next. */
2656 || at_endline_loc_p (p
, pend
, syntax
))
2657 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2666 if ((syntax
& RE_BK_PLUS_QM
)
2667 || (syntax
& RE_LIMITED_OPS
))
2671 /* If there is no previous pattern... */
2674 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2675 FREE_STACK_RETURN (REG_BADRPT
);
2676 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2681 /* 1 means zero (many) matches is allowed. */
2682 boolean zero_times_ok
= 0, many_times_ok
= 0;
2685 /* If there is a sequence of repetition chars, collapse it
2686 down to just one (the right one). We can't combine
2687 interval operators with these because of, e.g., `a{2}*',
2688 which should only match an even number of `a's. */
2692 if ((syntax
& RE_FRUGAL
)
2693 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2697 zero_times_ok
|= c
!= '+';
2698 many_times_ok
|= c
!= '?';
2704 || (!(syntax
& RE_BK_PLUS_QM
)
2705 && (*p
== '+' || *p
== '?')))
2707 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2710 FREE_STACK_RETURN (REG_EESCAPE
);
2711 if (p
[1] == '+' || p
[1] == '?')
2712 PATFETCH (c
); /* Gobble up the backslash. */
2718 /* If we get here, we found another repeat character. */
2722 /* Star, etc. applied to an empty pattern is equivalent
2723 to an empty pattern. */
2724 if (!laststart
|| laststart
== b
)
2727 /* Now we know whether or not zero matches is allowed
2728 and also whether or not two or more matches is allowed. */
2733 boolean simple
= skip_one_char (laststart
) == b
;
2734 unsigned int startoffset
= 0;
2736 /* Check if the loop can match the empty string. */
2737 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2738 ? on_failure_jump
: on_failure_jump_loop
;
2739 assert (skip_one_char (laststart
) <= b
);
2741 if (!zero_times_ok
&& simple
)
2742 { /* Since simple * loops can be made faster by using
2743 on_failure_keep_string_jump, we turn simple P+
2744 into PP* if P is simple. */
2745 unsigned char *p1
, *p2
;
2746 startoffset
= b
- laststart
;
2747 GET_BUFFER_SPACE (startoffset
);
2748 p1
= b
; p2
= laststart
;
2754 GET_BUFFER_SPACE (6);
2757 STORE_JUMP (ofj
, b
, b
+ 6);
2759 /* Simple * loops can use on_failure_keep_string_jump
2760 depending on what follows. But since we don't know
2761 that yet, we leave the decision up to
2762 on_failure_jump_smart. */
2763 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2764 laststart
+ startoffset
, b
+ 6);
2766 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2771 /* A simple ? pattern. */
2772 assert (zero_times_ok
);
2773 GET_BUFFER_SPACE (3);
2774 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2778 else /* not greedy */
2779 { /* I wish the greedy and non-greedy cases could be merged. */
2781 GET_BUFFER_SPACE (7); /* We might use less. */
2784 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2786 /* The non-greedy multiple match looks like
2787 a repeat..until: we only need a conditional jump
2788 at the end of the loop. */
2789 if (emptyp
) BUF_PUSH (no_op
);
2790 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2791 : on_failure_jump
, b
, laststart
);
2795 /* The repeat...until naturally matches one or more.
2796 To also match zero times, we need to first jump to
2797 the end of the loop (its conditional jump). */
2798 INSERT_JUMP (jump
, laststart
, b
);
2804 /* non-greedy a?? */
2805 INSERT_JUMP (jump
, laststart
, b
+ 3);
2807 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2824 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2826 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2828 /* Ensure that we have enough space to push a charset: the
2829 opcode, the length count, and the bitset; 34 bytes in all. */
2830 GET_BUFFER_SPACE (34);
2834 /* We test `*p == '^' twice, instead of using an if
2835 statement, so we only need one BUF_PUSH. */
2836 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2840 /* Remember the first position in the bracket expression. */
2843 /* Push the number of bytes in the bitmap. */
2844 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2846 /* Clear the whole map. */
2847 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2849 /* charset_not matches newline according to a syntax bit. */
2850 if ((re_opcode_t
) b
[-2] == charset_not
2851 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2852 SET_LIST_BIT ('\n');
2854 /* Read in characters and ranges, setting map bits. */
2857 boolean escaped_char
= false;
2858 const unsigned char *p2
= p
;
2860 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2862 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2863 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2864 So the translation is done later in a loop. Example:
2865 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2868 /* \ might escape characters inside [...] and [^...]. */
2869 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2871 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2874 escaped_char
= true;
2878 /* Could be the end of the bracket expression. If it's
2879 not (i.e., when the bracket expression is `[]' so
2880 far), the ']' character bit gets set way below. */
2881 if (c
== ']' && p2
!= p1
)
2885 /* See if we're at the beginning of a possible character
2888 if (!escaped_char
&&
2889 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2891 /* Leave room for the null. */
2892 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2893 const unsigned char *class_beg
;
2899 /* If pattern is `[[:'. */
2900 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2905 if ((c
== ':' && *p
== ']') || p
== pend
)
2907 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2910 /* This is in any case an invalid class name. */
2915 /* If isn't a word bracketed by `[:' and `:]':
2916 undo the ending character, the letters, and
2917 leave the leading `:' and `[' (but set bits for
2919 if (c
== ':' && *p
== ']')
2925 cc
= re_wctype (str
);
2928 FREE_STACK_RETURN (REG_ECTYPE
);
2930 /* Throw away the ] at the end of the character
2934 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2936 /* Most character classes in a multibyte match
2937 just set a flag. Exceptions are is_blank,
2938 is_digit, is_cntrl, and is_xdigit, since
2939 they can only match ASCII characters. We
2940 don't need to handle them for multibyte.
2941 They are distinguished by a negative wctype. */
2943 for (ch
= 0; ch
< 128; ++ch
)
2944 if (re_iswctype (btowc (ch
), cc
))
2950 if (target_multibyte
)
2952 SET_RANGE_TABLE_WORK_AREA_BIT
2953 (range_table_work
, re_wctype_to_bit (cc
));
2957 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2960 MAKE_CHAR_MULTIBYTE (c
);
2961 if (re_iswctype (btowc (c
), cc
))
2964 MAKE_CHAR_UNIBYTE (c
);
2970 /* Repeat the loop. */
2975 /* Go back to right after the "[:". */
2979 /* Because the `:' may starts the range, we
2980 can't simply set bit and repeat the loop.
2981 Instead, just set it to C and handle below. */
2986 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2989 /* Discard the `-'. */
2992 /* Fetch the character which ends the range. */
2996 if (syntax
& RE_NO_EMPTY_RANGES
)
2997 FREE_STACK_RETURN (REG_ERANGEX
);
2998 /* Else, repeat the loop. */
3002 /* Range from C to C. */
3007 c1
= TRANSLATE (c1
);
3008 /* Set the range into bitmap */
3009 for (; c
<= c1
; c
++)
3010 SET_LIST_BIT (TRANSLATE (c
));
3011 #else /* not emacs */
3012 if (target_multibyte
)
3016 re_wchar_t c0
= MAX (c
, 128);
3018 SETUP_MULTIBYTE_RANGE (range_table_work
, c0
, c1
);
3021 for (; c
<= c1
; c
++)
3022 SET_LIST_BIT (TRANSLATE (c
));
3028 for (; c
<= c1
; c
++)
3032 MAKE_CHAR_MULTIBYTE (c0
);
3033 c0
= TRANSLATE (c0
);
3034 MAKE_CHAR_UNIBYTE (c0
);
3038 #endif /* not emacs */
3041 /* Discard any (non)matching list bytes that are all 0 at the
3042 end of the map. Decrease the map-length byte too. */
3043 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3047 /* Build real range table from work area. */
3048 if (RANGE_TABLE_WORK_USED (range_table_work
)
3049 || RANGE_TABLE_WORK_BITS (range_table_work
))
3052 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3054 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3055 bytes for flags, two for COUNT, and three bytes for
3057 GET_BUFFER_SPACE (4 + used
* 3);
3059 /* Indicate the existence of range table. */
3060 laststart
[1] |= 0x80;
3062 /* Store the character class flag bits into the range table.
3063 If not in emacs, these flag bits are always 0. */
3064 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3065 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3067 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3068 for (i
= 0; i
< used
; i
++)
3069 STORE_CHARACTER_AND_INCR
3070 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3077 if (syntax
& RE_NO_BK_PARENS
)
3084 if (syntax
& RE_NO_BK_PARENS
)
3091 if (syntax
& RE_NEWLINE_ALT
)
3098 if (syntax
& RE_NO_BK_VBAR
)
3105 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3106 goto handle_interval
;
3112 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3114 /* Do not translate the character after the \, so that we can
3115 distinguish, e.g., \B from \b, even if we normally would
3116 translate, e.g., B to b. */
3122 if (syntax
& RE_NO_BK_PARENS
)
3123 goto normal_backslash
;
3130 /* Look for a special (?...) construct */
3131 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3133 PATFETCH (c
); /* Gobble up the '?'. */
3137 case ':': shy
= 1; break;
3139 /* Only (?:...) is supported right now. */
3140 FREE_STACK_RETURN (REG_BADPAT
);
3151 if (COMPILE_STACK_FULL
)
3153 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3154 compile_stack_elt_t
);
3155 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3157 compile_stack
.size
<<= 1;
3160 /* These are the values to restore when we hit end of this
3161 group. They are all relative offsets, so that if the
3162 whole pattern moves because of realloc, they will still
3164 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3165 COMPILE_STACK_TOP
.fixup_alt_jump
3166 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3167 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3168 COMPILE_STACK_TOP
.regnum
= shy
? -regnum
: regnum
;
3171 start_memory for groups beyond the last one we can
3172 represent in the compiled pattern. */
3173 if (regnum
<= MAX_REGNUM
&& !shy
)
3174 BUF_PUSH_2 (start_memory
, regnum
);
3176 compile_stack
.avail
++;
3181 /* If we've reached MAX_REGNUM groups, then this open
3182 won't actually generate any code, so we'll have to
3183 clear pending_exact explicitly. */
3189 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3191 if (COMPILE_STACK_EMPTY
)
3193 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3194 goto normal_backslash
;
3196 FREE_STACK_RETURN (REG_ERPAREN
);
3202 /* See similar code for backslashed left paren above. */
3203 if (COMPILE_STACK_EMPTY
)
3205 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3208 FREE_STACK_RETURN (REG_ERPAREN
);
3211 /* Since we just checked for an empty stack above, this
3212 ``can't happen''. */
3213 assert (compile_stack
.avail
!= 0);
3215 /* We don't just want to restore into `regnum', because
3216 later groups should continue to be numbered higher,
3217 as in `(ab)c(de)' -- the second group is #2. */
3218 regnum_t this_group_regnum
;
3220 compile_stack
.avail
--;
3221 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3223 = COMPILE_STACK_TOP
.fixup_alt_jump
3224 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3226 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3227 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
3228 /* If we've reached MAX_REGNUM groups, then this open
3229 won't actually generate any code, so we'll have to
3230 clear pending_exact explicitly. */
3233 /* We're at the end of the group, so now we know how many
3234 groups were inside this one. */
3235 if (this_group_regnum
<= MAX_REGNUM
&& this_group_regnum
> 0)
3236 BUF_PUSH_2 (stop_memory
, this_group_regnum
);
3241 case '|': /* `\|'. */
3242 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3243 goto normal_backslash
;
3245 if (syntax
& RE_LIMITED_OPS
)
3248 /* Insert before the previous alternative a jump which
3249 jumps to this alternative if the former fails. */
3250 GET_BUFFER_SPACE (3);
3251 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3255 /* The alternative before this one has a jump after it
3256 which gets executed if it gets matched. Adjust that
3257 jump so it will jump to this alternative's analogous
3258 jump (put in below, which in turn will jump to the next
3259 (if any) alternative's such jump, etc.). The last such
3260 jump jumps to the correct final destination. A picture:
3266 If we are at `b', then fixup_alt_jump right now points to a
3267 three-byte space after `a'. We'll put in the jump, set
3268 fixup_alt_jump to right after `b', and leave behind three
3269 bytes which we'll fill in when we get to after `c'. */
3273 /* Mark and leave space for a jump after this alternative,
3274 to be filled in later either by next alternative or
3275 when know we're at the end of a series of alternatives. */
3277 GET_BUFFER_SPACE (3);
3286 /* If \{ is a literal. */
3287 if (!(syntax
& RE_INTERVALS
)
3288 /* If we're at `\{' and it's not the open-interval
3290 || (syntax
& RE_NO_BK_BRACES
))
3291 goto normal_backslash
;
3295 /* If got here, then the syntax allows intervals. */
3297 /* At least (most) this many matches must be made. */
3298 int lower_bound
= 0, upper_bound
= -1;
3302 GET_UNSIGNED_NUMBER (lower_bound
);
3305 GET_UNSIGNED_NUMBER (upper_bound
);
3307 /* Interval such as `{1}' => match exactly once. */
3308 upper_bound
= lower_bound
;
3310 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3311 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3312 FREE_STACK_RETURN (REG_BADBR
);
3314 if (!(syntax
& RE_NO_BK_BRACES
))
3317 FREE_STACK_RETURN (REG_BADBR
);
3319 FREE_STACK_RETURN (REG_EESCAPE
);
3324 FREE_STACK_RETURN (REG_BADBR
);
3326 /* We just parsed a valid interval. */
3328 /* If it's invalid to have no preceding re. */
3331 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3332 FREE_STACK_RETURN (REG_BADRPT
);
3333 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3336 goto unfetch_interval
;
3339 if (upper_bound
== 0)
3340 /* If the upper bound is zero, just drop the sub pattern
3343 else if (lower_bound
== 1 && upper_bound
== 1)
3344 /* Just match it once: nothing to do here. */
3347 /* Otherwise, we have a nontrivial interval. When
3348 we're all done, the pattern will look like:
3349 set_number_at <jump count> <upper bound>
3350 set_number_at <succeed_n count> <lower bound>
3351 succeed_n <after jump addr> <succeed_n count>
3353 jump_n <succeed_n addr> <jump count>
3354 (The upper bound and `jump_n' are omitted if
3355 `upper_bound' is 1, though.) */
3357 { /* If the upper bound is > 1, we need to insert
3358 more at the end of the loop. */
3359 unsigned int nbytes
= (upper_bound
< 0 ? 3
3360 : upper_bound
> 1 ? 5 : 0);
3361 unsigned int startoffset
= 0;
3363 GET_BUFFER_SPACE (20); /* We might use less. */
3365 if (lower_bound
== 0)
3367 /* A succeed_n that starts with 0 is really a
3368 a simple on_failure_jump_loop. */
3369 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3375 /* Initialize lower bound of the `succeed_n', even
3376 though it will be set during matching by its
3377 attendant `set_number_at' (inserted next),
3378 because `re_compile_fastmap' needs to know.
3379 Jump to the `jump_n' we might insert below. */
3380 INSERT_JUMP2 (succeed_n
, laststart
,
3385 /* Code to initialize the lower bound. Insert
3386 before the `succeed_n'. The `5' is the last two
3387 bytes of this `set_number_at', plus 3 bytes of
3388 the following `succeed_n'. */
3389 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3394 if (upper_bound
< 0)
3396 /* A negative upper bound stands for infinity,
3397 in which case it degenerates to a plain jump. */
3398 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3401 else if (upper_bound
> 1)
3402 { /* More than one repetition is allowed, so
3403 append a backward jump to the `succeed_n'
3404 that starts this interval.
3406 When we've reached this during matching,
3407 we'll have matched the interval once, so
3408 jump back only `upper_bound - 1' times. */
3409 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3413 /* The location we want to set is the second
3414 parameter of the `jump_n'; that is `b-2' as
3415 an absolute address. `laststart' will be
3416 the `set_number_at' we're about to insert;
3417 `laststart+3' the number to set, the source
3418 for the relative address. But we are
3419 inserting into the middle of the pattern --
3420 so everything is getting moved up by 5.
3421 Conclusion: (b - 2) - (laststart + 3) + 5,
3422 i.e., b - laststart.
3424 We insert this at the beginning of the loop
3425 so that if we fail during matching, we'll
3426 reinitialize the bounds. */
3427 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3428 upper_bound
- 1, b
);
3433 beg_interval
= NULL
;
3438 /* If an invalid interval, match the characters as literals. */
3439 assert (beg_interval
);
3441 beg_interval
= NULL
;
3443 /* normal_char and normal_backslash need `c'. */
3446 if (!(syntax
& RE_NO_BK_BRACES
))
3448 assert (p
> pattern
&& p
[-1] == '\\');
3449 goto normal_backslash
;
3455 /* There is no way to specify the before_dot and after_dot
3456 operators. rms says this is ok. --karl */
3464 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3470 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3476 BUF_PUSH_2 (categoryspec
, c
);
3482 BUF_PUSH_2 (notcategoryspec
, c
);
3488 if (syntax
& RE_NO_GNU_OPS
)
3491 BUF_PUSH_2 (syntaxspec
, Sword
);
3496 if (syntax
& RE_NO_GNU_OPS
)
3499 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3504 if (syntax
& RE_NO_GNU_OPS
)
3510 if (syntax
& RE_NO_GNU_OPS
)
3516 if (syntax
& RE_NO_GNU_OPS
)
3525 FREE_STACK_RETURN (REG_BADPAT
);
3529 if (syntax
& RE_NO_GNU_OPS
)
3531 BUF_PUSH (wordbound
);
3535 if (syntax
& RE_NO_GNU_OPS
)
3537 BUF_PUSH (notwordbound
);
3541 if (syntax
& RE_NO_GNU_OPS
)
3547 if (syntax
& RE_NO_GNU_OPS
)
3552 case '1': case '2': case '3': case '4': case '5':
3553 case '6': case '7': case '8': case '9':
3557 if (syntax
& RE_NO_BK_REFS
)
3558 goto normal_backslash
;
3562 /* Can't back reference to a subexpression before its end. */
3563 if (reg
> regnum
|| group_in_compile_stack (compile_stack
, reg
))
3564 FREE_STACK_RETURN (REG_ESUBREG
);
3567 BUF_PUSH_2 (duplicate
, reg
);
3574 if (syntax
& RE_BK_PLUS_QM
)
3577 goto normal_backslash
;
3581 /* You might think it would be useful for \ to mean
3582 not to translate; but if we don't translate it
3583 it will never match anything. */
3590 /* Expects the character in `c'. */
3592 /* If no exactn currently being built. */
3595 /* If last exactn not at current position. */
3596 || pending_exact
+ *pending_exact
+ 1 != b
3598 /* We have only one byte following the exactn for the count. */
3599 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3601 /* If followed by a repetition operator. */
3602 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3603 || ((syntax
& RE_BK_PLUS_QM
)
3604 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3605 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3606 || ((syntax
& RE_INTERVALS
)
3607 && ((syntax
& RE_NO_BK_BRACES
)
3608 ? p
!= pend
&& *p
== '{'
3609 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3611 /* Start building a new exactn. */
3615 BUF_PUSH_2 (exactn
, 0);
3616 pending_exact
= b
- 1;
3619 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3624 MAKE_CHAR_MULTIBYTE (c
);
3626 if (target_multibyte
)
3628 len
= CHAR_STRING (c
, b
);
3633 MAKE_CHAR_UNIBYTE (c
);
3637 (*pending_exact
) += len
;
3642 } /* while p != pend */
3645 /* Through the pattern now. */
3649 if (!COMPILE_STACK_EMPTY
)
3650 FREE_STACK_RETURN (REG_EPAREN
);
3652 /* If we don't want backtracking, force success
3653 the first time we reach the end of the compiled pattern. */
3654 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3657 /* We have succeeded; set the length of the buffer. */
3658 bufp
->used
= b
- bufp
->buffer
;
3661 /* Now the buffer is adjusted for the multibyteness of a target. */
3662 bufp
->multibyte
= bufp
->target_multibyte
;
3668 re_compile_fastmap (bufp
);
3669 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3670 print_compiled_pattern (bufp
);
3675 #ifndef MATCH_MAY_ALLOCATE
3676 /* Initialize the failure stack to the largest possible stack. This
3677 isn't necessary unless we're trying to avoid calling alloca in
3678 the search and match routines. */
3680 int num_regs
= bufp
->re_nsub
+ 1;
3682 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3684 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3686 if (! fail_stack
.stack
)
3688 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3689 * sizeof (fail_stack_elt_t
));
3692 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3694 * sizeof (fail_stack_elt_t
)));
3697 regex_grow_registers (num_regs
);
3699 #endif /* not MATCH_MAY_ALLOCATE */
3701 FREE_STACK_RETURN (REG_NOERROR
);
3702 } /* regex_compile */
3704 /* Subroutines for `regex_compile'. */
3706 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3709 store_op1 (op
, loc
, arg
)
3714 *loc
= (unsigned char) op
;
3715 STORE_NUMBER (loc
+ 1, arg
);
3719 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3722 store_op2 (op
, loc
, arg1
, arg2
)
3727 *loc
= (unsigned char) op
;
3728 STORE_NUMBER (loc
+ 1, arg1
);
3729 STORE_NUMBER (loc
+ 3, arg2
);
3733 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3734 for OP followed by two-byte integer parameter ARG. */
3737 insert_op1 (op
, loc
, arg
, end
)
3743 register unsigned char *pfrom
= end
;
3744 register unsigned char *pto
= end
+ 3;
3746 while (pfrom
!= loc
)
3749 store_op1 (op
, loc
, arg
);
3753 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3756 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3762 register unsigned char *pfrom
= end
;
3763 register unsigned char *pto
= end
+ 5;
3765 while (pfrom
!= loc
)
3768 store_op2 (op
, loc
, arg1
, arg2
);
3772 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3773 after an alternative or a begin-subexpression. We assume there is at
3774 least one character before the ^. */
3777 at_begline_loc_p (pattern
, p
, syntax
)
3778 re_char
*pattern
, *p
;
3779 reg_syntax_t syntax
;
3781 re_char
*prev
= p
- 2;
3782 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3785 /* After a subexpression? */
3786 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3787 /* After an alternative? */
3788 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3789 /* After a shy subexpression? */
3790 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3791 && prev
[-1] == '?' && prev
[-2] == '('
3792 && (syntax
& RE_NO_BK_PARENS
3793 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3797 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3798 at least one character after the $, i.e., `P < PEND'. */
3801 at_endline_loc_p (p
, pend
, syntax
)
3803 reg_syntax_t syntax
;
3806 boolean next_backslash
= *next
== '\\';
3807 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3810 /* Before a subexpression? */
3811 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3812 : next_backslash
&& next_next
&& *next_next
== ')')
3813 /* Before an alternative? */
3814 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3815 : next_backslash
&& next_next
&& *next_next
== '|');
3819 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3820 false if it's not. */
3823 group_in_compile_stack (compile_stack
, regnum
)
3824 compile_stack_type compile_stack
;
3829 for (this_element
= compile_stack
.avail
- 1;
3832 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3839 If fastmap is non-NULL, go through the pattern and fill fastmap
3840 with all the possible leading chars. If fastmap is NULL, don't
3841 bother filling it up (obviously) and only return whether the
3842 pattern could potentially match the empty string.
3844 Return 1 if p..pend might match the empty string.
3845 Return 0 if p..pend matches at least one char.
3846 Return -1 if fastmap was not updated accurately. */
3849 analyse_first (p
, pend
, fastmap
, multibyte
)
3852 const int multibyte
;
3857 /* If all elements for base leading-codes in fastmap is set, this
3858 flag is set true. */
3859 boolean match_any_multibyte_characters
= false;
3863 /* The loop below works as follows:
3864 - It has a working-list kept in the PATTERN_STACK and which basically
3865 starts by only containing a pointer to the first operation.
3866 - If the opcode we're looking at is a match against some set of
3867 chars, then we add those chars to the fastmap and go on to the
3868 next work element from the worklist (done via `break').
3869 - If the opcode is a control operator on the other hand, we either
3870 ignore it (if it's meaningless at this point, such as `start_memory')
3871 or execute it (if it's a jump). If the jump has several destinations
3872 (i.e. `on_failure_jump'), then we push the other destination onto the
3874 We guarantee termination by ignoring backward jumps (more or less),
3875 so that `p' is monotonically increasing. More to the point, we
3876 never set `p' (or push) anything `<= p1'. */
3880 /* `p1' is used as a marker of how far back a `on_failure_jump'
3881 can go without being ignored. It is normally equal to `p'
3882 (which prevents any backward `on_failure_jump') except right
3883 after a plain `jump', to allow patterns such as:
3886 10: on_failure_jump 3
3887 as used for the *? operator. */
3890 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3897 /* If the first character has to match a backreference, that means
3898 that the group was empty (since it already matched). Since this
3899 is the only case that interests us here, we can assume that the
3900 backreference must match the empty string. */
3905 /* Following are the cases which match a character. These end
3910 /* If multibyte is nonzero, the first byte of each
3911 character is an ASCII or a leading code. Otherwise,
3912 each byte is a character. Thus, this works in both
3919 /* We could put all the chars except for \n (and maybe \0)
3920 but we don't bother since it is generally not worth it. */
3921 if (!fastmap
) break;
3926 if (!fastmap
) break;
3928 /* Chars beyond end of bitmap are possible matches. */
3929 /* In a multibyte case, the bitmap is used only for ASCII
3931 int limit
= multibyte
? 128 : (1 << BYTEWIDTH
);
3933 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3940 if (!fastmap
) break;
3941 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3942 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3944 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3947 if ((not && multibyte
)
3948 /* Any leading code can possibly start a character
3949 which doesn't match the specified set of characters. */
3950 || (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3951 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3952 /* If we can match a character class, we can match
3953 any multibyte characters. */
3955 if (match_any_multibyte_characters
== false)
3957 for (j
= 0x80; j
< (1 << BYTEWIDTH
); j
++)
3959 match_any_multibyte_characters
= true;
3963 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3964 && match_any_multibyte_characters
== false)
3966 /* Set fastmap[I] to 1 where I is a leading code of each
3967 multibyte characer in the range table. */
3969 unsigned char lc1
, lc2
;
3971 /* Make P points the range table. `+ 2' is to skip flag
3972 bits for a character class. */
3973 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3975 /* Extract the number of ranges in range table into COUNT. */
3976 EXTRACT_NUMBER_AND_INCR (count
, p
);
3977 for (; count
> 0; count
--, p
+= 2 * 3) /* XXX */
3979 /* Extract the start and end of each range. */
3980 EXTRACT_CHARACTER (c
, p
);
3981 lc1
= CHAR_LEADING_CODE (c
);
3983 EXTRACT_CHARACTER (c
, p
);
3984 lc2
= CHAR_LEADING_CODE (c
);
3985 for (j
= lc1
; j
<= lc2
; j
++)
3993 if (!fastmap
) break;
3995 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
3997 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3998 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4002 /* This match depends on text properties. These end with
4003 aborting optimizations. */
4007 case notcategoryspec
:
4008 if (!fastmap
) break;
4009 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4011 for (j
= (multibyte
? 127 : (1 << BYTEWIDTH
)); j
>= 0; j
--)
4012 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4017 /* Any character set can possibly contain a character
4018 whose category is K (or not). */
4019 if (match_any_multibyte_characters
== false)
4021 for (j
= 0x80; j
< (1 << BYTEWIDTH
); j
++)
4023 match_any_multibyte_characters
= true;
4028 /* All cases after this match the empty string. These end with
4050 EXTRACT_NUMBER_AND_INCR (j
, p
);
4052 /* Backward jumps can only go back to code that we've already
4053 visited. `re_compile' should make sure this is true. */
4056 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4058 case on_failure_jump
:
4059 case on_failure_keep_string_jump
:
4060 case on_failure_jump_loop
:
4061 case on_failure_jump_nastyloop
:
4062 case on_failure_jump_smart
:
4068 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4069 to jump back to "just after here". */
4072 case on_failure_jump
:
4073 case on_failure_keep_string_jump
:
4074 case on_failure_jump_nastyloop
:
4075 case on_failure_jump_loop
:
4076 case on_failure_jump_smart
:
4077 EXTRACT_NUMBER_AND_INCR (j
, p
);
4079 ; /* Backward jump to be ignored. */
4081 { /* We have to look down both arms.
4082 We first go down the "straight" path so as to minimize
4083 stack usage when going through alternatives. */
4084 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4092 /* This code simply does not properly handle forward jump_n. */
4093 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4095 /* jump_n can either jump or fall through. The (backward) jump
4096 case has already been handled, so we only need to look at the
4097 fallthrough case. */
4101 /* If N == 0, it should be an on_failure_jump_loop instead. */
4102 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4104 /* We only care about one iteration of the loop, so we don't
4105 need to consider the case where this behaves like an
4122 abort (); /* We have listed all the cases. */
4125 /* Getting here means we have found the possible starting
4126 characters for one path of the pattern -- and that the empty
4127 string does not match. We need not follow this path further. */
4131 /* We reached the end without matching anything. */
4134 } /* analyse_first */
4136 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4137 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4138 characters can start a string that matches the pattern. This fastmap
4139 is used by re_search to skip quickly over impossible starting points.
4141 Character codes above (1 << BYTEWIDTH) are not represented in the
4142 fastmap, but the leading codes are represented. Thus, the fastmap
4143 indicates which character sets could start a match.
4145 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4146 area as BUFP->fastmap.
4148 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4151 Returns 0 if we succeed, -2 if an internal error. */
4154 re_compile_fastmap (bufp
)
4155 struct re_pattern_buffer
*bufp
;
4157 char *fastmap
= bufp
->fastmap
;
4160 assert (fastmap
&& bufp
->buffer
);
4162 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4163 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4165 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4166 fastmap
, RE_MULTIBYTE_P (bufp
));
4167 bufp
->can_be_null
= (analysis
!= 0);
4169 } /* re_compile_fastmap */
4171 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4172 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4173 this memory for recording register information. STARTS and ENDS
4174 must be allocated using the malloc library routine, and must each
4175 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4177 If NUM_REGS == 0, then subsequent matches should allocate their own
4180 Unless this function is called, the first search or match using
4181 PATTERN_BUFFER will allocate its own register data, without
4182 freeing the old data. */
4185 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4186 struct re_pattern_buffer
*bufp
;
4187 struct re_registers
*regs
;
4189 regoff_t
*starts
, *ends
;
4193 bufp
->regs_allocated
= REGS_REALLOCATE
;
4194 regs
->num_regs
= num_regs
;
4195 regs
->start
= starts
;
4200 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4202 regs
->start
= regs
->end
= (regoff_t
*) 0;
4205 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4207 /* Searching routines. */
4209 /* Like re_search_2, below, but only one string is specified, and
4210 doesn't let you say where to stop matching. */
4213 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4214 struct re_pattern_buffer
*bufp
;
4216 int size
, startpos
, range
;
4217 struct re_registers
*regs
;
4219 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4222 WEAK_ALIAS (__re_search
, re_search
)
4224 /* Head address of virtual concatenation of string. */
4225 #define HEAD_ADDR_VSTRING(P) \
4226 (((P) >= size1 ? string2 : string1))
4228 /* End address of virtual concatenation of string. */
4229 #define STOP_ADDR_VSTRING(P) \
4230 (((P) >= size1 ? string2 + size2 : string1 + size1))
4232 /* Address of POS in the concatenation of virtual string. */
4233 #define POS_ADDR_VSTRING(POS) \
4234 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4236 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4237 virtual concatenation of STRING1 and STRING2, starting first at index
4238 STARTPOS, then at STARTPOS + 1, and so on.
4240 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4242 RANGE is how far to scan while trying to match. RANGE = 0 means try
4243 only at STARTPOS; in general, the last start tried is STARTPOS +
4246 In REGS, return the indices of the virtual concatenation of STRING1
4247 and STRING2 that matched the entire BUFP->buffer and its contained
4250 Do not consider matching one past the index STOP in the virtual
4251 concatenation of STRING1 and STRING2.
4253 We return either the position in the strings at which the match was
4254 found, -1 if no match, or -2 if error (such as failure
4258 re_search_2 (bufp
, str1
, size1
, str2
, size2
, startpos
, range
, regs
, stop
)
4259 struct re_pattern_buffer
*bufp
;
4260 const char *str1
, *str2
;
4264 struct re_registers
*regs
;
4268 re_char
*string1
= (re_char
*) str1
;
4269 re_char
*string2
= (re_char
*) str2
;
4270 register char *fastmap
= bufp
->fastmap
;
4271 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4272 int total_size
= size1
+ size2
;
4273 int endpos
= startpos
+ range
;
4274 boolean anchored_start
;
4275 /* Nonzero if BUFP is setup for multibyte characters. We are sure
4276 that it is the same as RE_TARGET_MULTIBYTE_P (bufp). */
4277 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4279 /* Check for out-of-range STARTPOS. */
4280 if (startpos
< 0 || startpos
> total_size
)
4283 /* Fix up RANGE if it might eventually take us outside
4284 the virtual concatenation of STRING1 and STRING2.
4285 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4287 range
= 0 - startpos
;
4288 else if (endpos
> total_size
)
4289 range
= total_size
- startpos
;
4291 /* If the search isn't to be a backwards one, don't waste time in a
4292 search for a pattern anchored at beginning of buffer. */
4293 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4302 /* In a forward search for something that starts with \=.
4303 don't keep searching past point. */
4304 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4306 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4312 /* Update the fastmap now if not correct already. */
4313 if (fastmap
&& !bufp
->fastmap_accurate
)
4314 re_compile_fastmap (bufp
);
4316 /* See whether the pattern is anchored. */
4317 anchored_start
= (bufp
->buffer
[0] == begline
);
4320 gl_state
.object
= re_match_object
;
4322 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4324 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4328 /* Loop through the string, looking for a place to start matching. */
4331 /* If the pattern is anchored,
4332 skip quickly past places we cannot match.
4333 We don't bother to treat startpos == 0 specially
4334 because that case doesn't repeat. */
4335 if (anchored_start
&& startpos
> 0)
4337 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4338 : string2
[startpos
- size1
- 1])
4343 /* If a fastmap is supplied, skip quickly over characters that
4344 cannot be the start of a match. If the pattern can match the
4345 null string, however, we don't need to skip characters; we want
4346 the first null string. */
4347 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4349 register re_char
*d
;
4350 register re_wchar_t buf_ch
;
4352 d
= POS_ADDR_VSTRING (startpos
);
4354 if (range
> 0) /* Searching forwards. */
4356 register int lim
= 0;
4359 if (startpos
< size1
&& startpos
+ range
>= size1
)
4360 lim
= range
- (size1
- startpos
);
4362 /* Written out as an if-else to avoid testing `translate'
4364 if (RE_TRANSLATE_P (translate
))
4371 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4373 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4374 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4377 range
-= buf_charlen
;
4384 MAKE_CHAR_MULTIBYTE (buf_ch
);
4385 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4386 MAKE_CHAR_UNIBYTE (buf_ch
);
4387 if (fastmap
[buf_ch
])
4400 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4402 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4404 range
-= buf_charlen
;
4408 while (range
> lim
&& !fastmap
[*d
])
4414 startpos
+= irange
- range
;
4416 else /* Searching backwards. */
4418 int room
= (startpos
>= size1
4419 ? size2
+ size1
- startpos
4420 : size1
- startpos
);
4423 buf_ch
= STRING_CHAR (d
, room
);
4424 buf_ch
= TRANSLATE (buf_ch
);
4425 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4430 if (! fastmap
[TRANSLATE (*d
)])
4436 /* If can't match the null string, and that's all we have left, fail. */
4437 if (range
>= 0 && startpos
== total_size
&& fastmap
4438 && !bufp
->can_be_null
)
4441 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4442 startpos
, regs
, stop
);
4443 #ifndef REGEX_MALLOC
4460 /* Update STARTPOS to the next character boundary. */
4463 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4464 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4465 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
4483 /* Update STARTPOS to the previous character boundary. */
4486 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4488 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4490 /* Find the head of multibyte form. */
4491 PREV_CHAR_BOUNDARY (p
, phead
);
4492 range
+= p0
- 1 - p
;
4496 startpos
-= p0
- 1 - p
;
4502 WEAK_ALIAS (__re_search_2
, re_search_2
)
4504 /* Declarations and macros for re_match_2. */
4506 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4508 RE_TRANSLATE_TYPE translate
,
4509 const int multibyte
));
4511 /* This converts PTR, a pointer into one of the search strings `string1'
4512 and `string2' into an offset from the beginning of that string. */
4513 #define POINTER_TO_OFFSET(ptr) \
4514 (FIRST_STRING_P (ptr) \
4515 ? ((regoff_t) ((ptr) - string1)) \
4516 : ((regoff_t) ((ptr) - string2 + size1)))
4518 /* Call before fetching a character with *d. This switches over to
4519 string2 if necessary.
4520 Check re_match_2_internal for a discussion of why end_match_2 might
4521 not be within string2 (but be equal to end_match_1 instead). */
4522 #define PREFETCH() \
4525 /* End of string2 => fail. */ \
4526 if (dend == end_match_2) \
4528 /* End of string1 => advance to string2. */ \
4530 dend = end_match_2; \
4533 /* Call before fetching a char with *d if you already checked other limits.
4534 This is meant for use in lookahead operations like wordend, etc..
4535 where we might need to look at parts of the string that might be
4536 outside of the LIMITs (i.e past `stop'). */
4537 #define PREFETCH_NOLIMIT() \
4541 dend = end_match_2; \
4544 /* Test if at very beginning or at very end of the virtual concatenation
4545 of `string1' and `string2'. If only one string, it's `string2'. */
4546 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4547 #define AT_STRINGS_END(d) ((d) == end2)
4550 /* Test if D points to a character which is word-constituent. We have
4551 two special cases to check for: if past the end of string1, look at
4552 the first character in string2; and if before the beginning of
4553 string2, look at the last character in string1. */
4554 #define WORDCHAR_P(d) \
4555 (SYNTAX ((d) == end1 ? *string2 \
4556 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4559 /* Disabled due to a compiler bug -- see comment at case wordbound */
4561 /* The comment at case wordbound is following one, but we don't use
4562 AT_WORD_BOUNDARY anymore to support multibyte form.
4564 The DEC Alpha C compiler 3.x generates incorrect code for the
4565 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4566 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4567 macro and introducing temporary variables works around the bug. */
4570 /* Test if the character before D and the one at D differ with respect
4571 to being word-constituent. */
4572 #define AT_WORD_BOUNDARY(d) \
4573 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4574 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4577 /* Free everything we malloc. */
4578 #ifdef MATCH_MAY_ALLOCATE
4579 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4580 # define FREE_VARIABLES() \
4582 REGEX_FREE_STACK (fail_stack.stack); \
4583 FREE_VAR (regstart); \
4584 FREE_VAR (regend); \
4585 FREE_VAR (best_regstart); \
4586 FREE_VAR (best_regend); \
4589 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4590 #endif /* not MATCH_MAY_ALLOCATE */
4593 /* Optimization routines. */
4595 /* If the operation is a match against one or more chars,
4596 return a pointer to the next operation, else return NULL. */
4601 switch (SWITCH_ENUM_CAST (*p
++))
4612 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4615 p
= CHARSET_RANGE_TABLE (p
- 1);
4616 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4617 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4620 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4627 case notcategoryspec
:
4639 /* Jump over non-matching operations. */
4641 skip_noops (p
, pend
)
4647 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4656 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4667 /* Non-zero if "p1 matches something" implies "p2 fails". */
4669 mutually_exclusive_p (bufp
, p1
, p2
)
4670 struct re_pattern_buffer
*bufp
;
4674 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4675 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4677 assert (p1
>= bufp
->buffer
&& p1
< pend
4678 && p2
>= bufp
->buffer
&& p2
<= pend
);
4680 /* Skip over open/close-group commands.
4681 If what follows this loop is a ...+ construct,
4682 look at what begins its body, since we will have to
4683 match at least one of that. */
4684 p2
= skip_noops (p2
, pend
);
4685 /* The same skip can be done for p1, except that this function
4686 is only used in the case where p1 is a simple match operator. */
4687 /* p1 = skip_noops (p1, pend); */
4689 assert (p1
>= bufp
->buffer
&& p1
< pend
4690 && p2
>= bufp
->buffer
&& p2
<= pend
);
4692 op2
= p2
== pend
? succeed
: *p2
;
4694 switch (SWITCH_ENUM_CAST (op2
))
4698 /* If we're at the end of the pattern, we can change. */
4699 if (skip_one_char (p1
))
4701 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4709 register re_wchar_t c
4710 = (re_opcode_t
) *p2
== endline
? '\n'
4711 : RE_STRING_CHAR (p2
+ 2, pend
- p2
- 2);
4713 if ((re_opcode_t
) *p1
== exactn
)
4715 if (c
!= RE_STRING_CHAR (p1
+ 2, pend
- p1
- 2))
4717 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4722 else if ((re_opcode_t
) *p1
== charset
4723 || (re_opcode_t
) *p1
== charset_not
)
4725 int not = (re_opcode_t
) *p1
== charset_not
;
4727 /* Test if C is listed in charset (or charset_not)
4729 if (! multibyte
|| IS_REAL_ASCII (c
))
4731 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4732 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4735 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4736 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4738 /* `not' is equal to 1 if c would match, which means
4739 that we can't change to pop_failure_jump. */
4742 DEBUG_PRINT1 (" No match => fast loop.\n");
4746 else if ((re_opcode_t
) *p1
== anychar
4749 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4757 if ((re_opcode_t
) *p1
== exactn
)
4758 /* Reuse the code above. */
4759 return mutually_exclusive_p (bufp
, p2
, p1
);
4761 /* It is hard to list up all the character in charset
4762 P2 if it includes multibyte character. Give up in
4764 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4766 /* Now, we are sure that P2 has no range table.
4767 So, for the size of bitmap in P2, `p2[1]' is
4768 enough. But P1 may have range table, so the
4769 size of bitmap table of P1 is extracted by
4770 using macro `CHARSET_BITMAP_SIZE'.
4772 In a multibyte case, we know that all the character
4773 listed in P2 is ASCII. In a unibyte case, P1 has only a
4774 bitmap table. So, in both cases, it is enough to test
4775 only the bitmap table of P1. */
4777 if ((re_opcode_t
) *p1
== charset
)
4780 /* We win if the charset inside the loop
4781 has no overlap with the one after the loop. */
4784 && idx
< CHARSET_BITMAP_SIZE (p1
));
4786 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4790 || idx
== CHARSET_BITMAP_SIZE (p1
))
4792 DEBUG_PRINT1 (" No match => fast loop.\n");
4796 else if ((re_opcode_t
) *p1
== charset_not
)
4799 /* We win if the charset_not inside the loop lists
4800 every character listed in the charset after. */
4801 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4802 if (! (p2
[2 + idx
] == 0
4803 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4804 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4809 DEBUG_PRINT1 (" No match => fast loop.\n");
4818 switch (SWITCH_ENUM_CAST (*p1
))
4822 /* Reuse the code above. */
4823 return mutually_exclusive_p (bufp
, p2
, p1
);
4825 /* When we have two charset_not, it's very unlikely that
4826 they don't overlap. The union of the two sets of excluded
4827 chars should cover all possible chars, which, as a matter of
4828 fact, is virtually impossible in multibyte buffers. */
4834 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4836 return ((re_opcode_t
) *p1
== syntaxspec
4837 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4839 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4842 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4844 return ((re_opcode_t
) *p1
== notsyntaxspec
4845 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4847 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4850 return (((re_opcode_t
) *p1
== notsyntaxspec
4851 || (re_opcode_t
) *p1
== syntaxspec
)
4856 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4857 case notcategoryspec
:
4858 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4870 /* Matching routines. */
4872 #ifndef emacs /* Emacs never uses this. */
4873 /* re_match is like re_match_2 except it takes only a single string. */
4876 re_match (bufp
, string
, size
, pos
, regs
)
4877 struct re_pattern_buffer
*bufp
;
4880 struct re_registers
*regs
;
4882 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
4884 # if defined C_ALLOCA && !defined REGEX_MALLOC
4889 WEAK_ALIAS (__re_match
, re_match
)
4890 #endif /* not emacs */
4893 /* In Emacs, this is the string or buffer in which we
4894 are matching. It is used for looking up syntax properties. */
4895 Lisp_Object re_match_object
;
4898 /* re_match_2 matches the compiled pattern in BUFP against the
4899 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4900 and SIZE2, respectively). We start matching at POS, and stop
4903 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4904 store offsets for the substring each group matched in REGS. See the
4905 documentation for exactly how many groups we fill.
4907 We return -1 if no match, -2 if an internal error (such as the
4908 failure stack overflowing). Otherwise, we return the length of the
4909 matched substring. */
4912 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4913 struct re_pattern_buffer
*bufp
;
4914 const char *string1
, *string2
;
4917 struct re_registers
*regs
;
4924 gl_state
.object
= re_match_object
;
4925 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4926 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4929 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4930 (re_char
*) string2
, size2
,
4932 #if defined C_ALLOCA && !defined REGEX_MALLOC
4937 WEAK_ALIAS (__re_match_2
, re_match_2
)
4940 #define TRANSLATE_VIA_MULTIBYTE(c) \
4943 (c) = TRANSLATE (c); \
4946 MAKE_CHAR_MULTIBYTE (c); \
4947 (c) = TRANSLATE (c); \
4948 MAKE_CHAR_UNIBYTE (c); \
4953 #define TRANSLATE_VIA_MULTIBYTE(c) ((c) = TRANSLATE (c))
4957 /* This is a separate function so that we can force an alloca cleanup
4960 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4961 struct re_pattern_buffer
*bufp
;
4962 re_char
*string1
, *string2
;
4965 struct re_registers
*regs
;
4968 /* General temporaries. */
4973 /* Just past the end of the corresponding string. */
4974 re_char
*end1
, *end2
;
4976 /* Pointers into string1 and string2, just past the last characters in
4977 each to consider matching. */
4978 re_char
*end_match_1
, *end_match_2
;
4980 /* Where we are in the data, and the end of the current string. */
4983 /* Used sometimes to remember where we were before starting matching
4984 an operator so that we can go back in case of failure. This "atomic"
4985 behavior of matching opcodes is indispensable to the correctness
4986 of the on_failure_keep_string_jump optimization. */
4989 /* Where we are in the pattern, and the end of the pattern. */
4990 re_char
*p
= bufp
->buffer
;
4991 re_char
*pend
= p
+ bufp
->used
;
4993 /* We use this to map every character in the string. */
4994 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4996 /* Nonzero if BUFP is setup for multibyte characters. We are sure
4997 that it is the same as RE_TARGET_MULTIBYTE_P (bufp). */
4998 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
5000 /* Failure point stack. Each place that can handle a failure further
5001 down the line pushes a failure point on this stack. It consists of
5002 regstart, and regend for all registers corresponding to
5003 the subexpressions we're currently inside, plus the number of such
5004 registers, and, finally, two char *'s. The first char * is where
5005 to resume scanning the pattern; the second one is where to resume
5006 scanning the strings. */
5007 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5008 fail_stack_type fail_stack
;
5011 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5014 #if defined REL_ALLOC && defined REGEX_MALLOC
5015 /* This holds the pointer to the failure stack, when
5016 it is allocated relocatably. */
5017 fail_stack_elt_t
*failure_stack_ptr
;
5020 /* We fill all the registers internally, independent of what we
5021 return, for use in backreferences. The number here includes
5022 an element for register zero. */
5023 size_t num_regs
= bufp
->re_nsub
+ 1;
5025 /* Information on the contents of registers. These are pointers into
5026 the input strings; they record just what was matched (on this
5027 attempt) by a subexpression part of the pattern, that is, the
5028 regnum-th regstart pointer points to where in the pattern we began
5029 matching and the regnum-th regend points to right after where we
5030 stopped matching the regnum-th subexpression. (The zeroth register
5031 keeps track of what the whole pattern matches.) */
5032 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5033 re_char
**regstart
, **regend
;
5036 /* The following record the register info as found in the above
5037 variables when we find a match better than any we've seen before.
5038 This happens as we backtrack through the failure points, which in
5039 turn happens only if we have not yet matched the entire string. */
5040 unsigned best_regs_set
= false;
5041 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5042 re_char
**best_regstart
, **best_regend
;
5045 /* Logically, this is `best_regend[0]'. But we don't want to have to
5046 allocate space for that if we're not allocating space for anything
5047 else (see below). Also, we never need info about register 0 for
5048 any of the other register vectors, and it seems rather a kludge to
5049 treat `best_regend' differently than the rest. So we keep track of
5050 the end of the best match so far in a separate variable. We
5051 initialize this to NULL so that when we backtrack the first time
5052 and need to test it, it's not garbage. */
5053 re_char
*match_end
= NULL
;
5056 /* Counts the total number of registers pushed. */
5057 unsigned num_regs_pushed
= 0;
5060 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5064 #ifdef MATCH_MAY_ALLOCATE
5065 /* Do not bother to initialize all the register variables if there are
5066 no groups in the pattern, as it takes a fair amount of time. If
5067 there are groups, we include space for register 0 (the whole
5068 pattern), even though we never use it, since it simplifies the
5069 array indexing. We should fix this. */
5072 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5073 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5074 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5075 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5077 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5085 /* We must initialize all our variables to NULL, so that
5086 `FREE_VARIABLES' doesn't try to free them. */
5087 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5089 #endif /* MATCH_MAY_ALLOCATE */
5091 /* The starting position is bogus. */
5092 if (pos
< 0 || pos
> size1
+ size2
)
5098 /* Initialize subexpression text positions to -1 to mark ones that no
5099 start_memory/stop_memory has been seen for. Also initialize the
5100 register information struct. */
5101 for (reg
= 1; reg
< num_regs
; reg
++)
5102 regstart
[reg
] = regend
[reg
] = NULL
;
5104 /* We move `string1' into `string2' if the latter's empty -- but not if
5105 `string1' is null. */
5106 if (size2
== 0 && string1
!= NULL
)
5113 end1
= string1
+ size1
;
5114 end2
= string2
+ size2
;
5116 /* `p' scans through the pattern as `d' scans through the data.
5117 `dend' is the end of the input string that `d' points within. `d'
5118 is advanced into the following input string whenever necessary, but
5119 this happens before fetching; therefore, at the beginning of the
5120 loop, `d' can be pointing at the end of a string, but it cannot
5124 /* Only match within string2. */
5125 d
= string2
+ pos
- size1
;
5126 dend
= end_match_2
= string2
+ stop
- size1
;
5127 end_match_1
= end1
; /* Just to give it a value. */
5133 /* Only match within string1. */
5134 end_match_1
= string1
+ stop
;
5136 When we reach end_match_1, PREFETCH normally switches to string2.
5137 But in the present case, this means that just doing a PREFETCH
5138 makes us jump from `stop' to `gap' within the string.
5139 What we really want here is for the search to stop as
5140 soon as we hit end_match_1. That's why we set end_match_2
5141 to end_match_1 (since PREFETCH fails as soon as we hit
5143 end_match_2
= end_match_1
;
5146 { /* It's important to use this code when stop == size so that
5147 moving `d' from end1 to string2 will not prevent the d == dend
5148 check from catching the end of string. */
5150 end_match_2
= string2
+ stop
- size1
;
5156 DEBUG_PRINT1 ("The compiled pattern is: ");
5157 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5158 DEBUG_PRINT1 ("The string to match is: `");
5159 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5160 DEBUG_PRINT1 ("'\n");
5162 /* This loops over pattern commands. It exits by returning from the
5163 function if the match is complete, or it drops through if the match
5164 fails at this starting point in the input data. */
5167 DEBUG_PRINT2 ("\n%p: ", p
);
5170 { /* End of pattern means we might have succeeded. */
5171 DEBUG_PRINT1 ("end of pattern ... ");
5173 /* If we haven't matched the entire string, and we want the
5174 longest match, try backtracking. */
5175 if (d
!= end_match_2
)
5177 /* 1 if this match ends in the same string (string1 or string2)
5178 as the best previous match. */
5179 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5180 == FIRST_STRING_P (d
));
5181 /* 1 if this match is the best seen so far. */
5182 boolean best_match_p
;
5184 /* AIX compiler got confused when this was combined
5185 with the previous declaration. */
5187 best_match_p
= d
> match_end
;
5189 best_match_p
= !FIRST_STRING_P (d
);
5191 DEBUG_PRINT1 ("backtracking.\n");
5193 if (!FAIL_STACK_EMPTY ())
5194 { /* More failure points to try. */
5196 /* If exceeds best match so far, save it. */
5197 if (!best_regs_set
|| best_match_p
)
5199 best_regs_set
= true;
5202 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5204 for (reg
= 1; reg
< num_regs
; reg
++)
5206 best_regstart
[reg
] = regstart
[reg
];
5207 best_regend
[reg
] = regend
[reg
];
5213 /* If no failure points, don't restore garbage. And if
5214 last match is real best match, don't restore second
5216 else if (best_regs_set
&& !best_match_p
)
5219 /* Restore best match. It may happen that `dend ==
5220 end_match_1' while the restored d is in string2.
5221 For example, the pattern `x.*y.*z' against the
5222 strings `x-' and `y-z-', if the two strings are
5223 not consecutive in memory. */
5224 DEBUG_PRINT1 ("Restoring best registers.\n");
5227 dend
= ((d
>= string1
&& d
<= end1
)
5228 ? end_match_1
: end_match_2
);
5230 for (reg
= 1; reg
< num_regs
; reg
++)
5232 regstart
[reg
] = best_regstart
[reg
];
5233 regend
[reg
] = best_regend
[reg
];
5236 } /* d != end_match_2 */
5239 DEBUG_PRINT1 ("Accepting match.\n");
5241 /* If caller wants register contents data back, do it. */
5242 if (regs
&& !bufp
->no_sub
)
5244 /* Have the register data arrays been allocated? */
5245 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5246 { /* No. So allocate them with malloc. We need one
5247 extra element beyond `num_regs' for the `-1' marker
5249 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5250 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5251 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5252 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5257 bufp
->regs_allocated
= REGS_REALLOCATE
;
5259 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5260 { /* Yes. If we need more elements than were already
5261 allocated, reallocate them. If we need fewer, just
5263 if (regs
->num_regs
< num_regs
+ 1)
5265 regs
->num_regs
= num_regs
+ 1;
5266 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5267 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5268 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5277 /* These braces fend off a "empty body in an else-statement"
5278 warning under GCC when assert expands to nothing. */
5279 assert (bufp
->regs_allocated
== REGS_FIXED
);
5282 /* Convert the pointer data in `regstart' and `regend' to
5283 indices. Register zero has to be set differently,
5284 since we haven't kept track of any info for it. */
5285 if (regs
->num_regs
> 0)
5287 regs
->start
[0] = pos
;
5288 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5291 /* Go through the first `min (num_regs, regs->num_regs)'
5292 registers, since that is all we initialized. */
5293 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5295 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5296 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5300 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5302 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5306 /* If the regs structure we return has more elements than
5307 were in the pattern, set the extra elements to -1. If
5308 we (re)allocated the registers, this is the case,
5309 because we always allocate enough to have at least one
5311 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5312 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5313 } /* regs && !bufp->no_sub */
5315 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5316 nfailure_points_pushed
, nfailure_points_popped
,
5317 nfailure_points_pushed
- nfailure_points_popped
);
5318 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5320 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5322 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5328 /* Otherwise match next pattern command. */
5329 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5331 /* Ignore these. Used to ignore the n of succeed_n's which
5332 currently have n == 0. */
5334 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5338 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5341 /* Match the next n pattern characters exactly. The following
5342 byte in the pattern defines n, and the n bytes after that
5343 are the characters to match. */
5346 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5348 /* Remember the start point to rollback upon failure. */
5352 /* This is written out as an if-else so we don't waste time
5353 testing `translate' inside the loop. */
5354 if (RE_TRANSLATE_P (translate
))
5358 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5378 /* The cost of testing `translate' is comparatively small. */
5382 int pat_charlen
, buf_charlen
;
5383 unsigned int pat_ch
, buf_ch
;
5386 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
5387 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5389 if (TRANSLATE (buf_ch
) != pat_ch
)
5397 mcnt
-= pat_charlen
;
5403 unsigned int buf_ch
;
5407 TRANSLATE_VIA_MULTIBYTE (buf_ch
);
5419 /* Match any character except possibly a newline or a null. */
5425 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5428 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5429 buf_ch
= TRANSLATE (buf_ch
);
5431 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5433 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5434 && buf_ch
== '\000'))
5437 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5446 register unsigned int c
;
5447 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5450 /* Start of actual range_table, or end of bitmap if there is no
5452 re_char
*range_table
;
5454 /* Nonzero if there is a range table. */
5455 int range_table_exists
;
5457 /* Number of ranges of range table. This is not included
5458 in the initial byte-length of the command. */
5461 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5463 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5465 if (range_table_exists
)
5467 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5468 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5472 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5473 TRANSLATE_VIA_MULTIBYTE (c
); /* The character to match. */
5475 if (! multibyte
|| IS_REAL_ASCII (c
))
5476 { /* Lookup bitmap. */
5477 /* Cast to `unsigned' instead of `unsigned char' in
5478 case the bit list is a full 32 bytes long. */
5479 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5480 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5484 else if (range_table_exists
)
5486 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5488 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5489 | (class_bits
& BIT_MULTIBYTE
)
5490 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5491 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5492 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5493 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5496 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5500 if (range_table_exists
)
5501 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5503 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5505 if (!not) goto fail
;
5512 /* The beginning of a group is represented by start_memory.
5513 The argument is the register number. The text
5514 matched within the group is recorded (in the internal
5515 registers data structure) under the register number. */
5517 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5519 /* In case we need to undo this operation (via backtracking). */
5520 PUSH_FAILURE_REG ((unsigned int)*p
);
5523 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5524 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5526 /* Move past the register number and inner group count. */
5531 /* The stop_memory opcode represents the end of a group. Its
5532 argument is the same as start_memory's: the register number. */
5534 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5536 assert (!REG_UNSET (regstart
[*p
]));
5537 /* Strictly speaking, there should be code such as:
5539 assert (REG_UNSET (regend[*p]));
5540 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5542 But the only info to be pushed is regend[*p] and it is known to
5543 be UNSET, so there really isn't anything to push.
5544 Not pushing anything, on the other hand deprives us from the
5545 guarantee that regend[*p] is UNSET since undoing this operation
5546 will not reset its value properly. This is not important since
5547 the value will only be read on the next start_memory or at
5548 the very end and both events can only happen if this stop_memory
5552 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5554 /* Move past the register number and the inner group count. */
5559 /* \<digit> has been turned into a `duplicate' command which is
5560 followed by the numeric value of <digit> as the register number. */
5563 register re_char
*d2
, *dend2
;
5564 int regno
= *p
++; /* Get which register to match against. */
5565 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5567 /* Can't back reference a group which we've never matched. */
5568 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5571 /* Where in input to try to start matching. */
5572 d2
= regstart
[regno
];
5574 /* Remember the start point to rollback upon failure. */
5577 /* Where to stop matching; if both the place to start and
5578 the place to stop matching are in the same string, then
5579 set to the place to stop, otherwise, for now have to use
5580 the end of the first string. */
5582 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5583 == FIRST_STRING_P (regend
[regno
]))
5584 ? regend
[regno
] : end_match_1
);
5587 /* If necessary, advance to next segment in register
5591 if (dend2
== end_match_2
) break;
5592 if (dend2
== regend
[regno
]) break;
5594 /* End of string1 => advance to string2. */
5596 dend2
= regend
[regno
];
5598 /* At end of register contents => success */
5599 if (d2
== dend2
) break;
5601 /* If necessary, advance to next segment in data. */
5604 /* How many characters left in this segment to match. */
5607 /* Want how many consecutive characters we can match in
5608 one shot, so, if necessary, adjust the count. */
5609 if (mcnt
> dend2
- d2
)
5612 /* Compare that many; failure if mismatch, else move
5614 if (RE_TRANSLATE_P (translate
)
5615 ? bcmp_translate (d
, d2
, mcnt
, translate
, multibyte
)
5616 : memcmp (d
, d2
, mcnt
))
5621 d
+= mcnt
, d2
+= mcnt
;
5627 /* begline matches the empty string at the beginning of the string
5628 (unless `not_bol' is set in `bufp'), and after newlines. */
5630 DEBUG_PRINT1 ("EXECUTING begline.\n");
5632 if (AT_STRINGS_BEG (d
))
5634 if (!bufp
->not_bol
) break;
5639 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5643 /* In all other cases, we fail. */
5647 /* endline is the dual of begline. */
5649 DEBUG_PRINT1 ("EXECUTING endline.\n");
5651 if (AT_STRINGS_END (d
))
5653 if (!bufp
->not_eol
) break;
5657 PREFETCH_NOLIMIT ();
5664 /* Match at the very beginning of the data. */
5666 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5667 if (AT_STRINGS_BEG (d
))
5672 /* Match at the very end of the data. */
5674 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5675 if (AT_STRINGS_END (d
))
5680 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5681 pushes NULL as the value for the string on the stack. Then
5682 `POP_FAILURE_POINT' will keep the current value for the
5683 string, instead of restoring it. To see why, consider
5684 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5685 then the . fails against the \n. But the next thing we want
5686 to do is match the \n against the \n; if we restored the
5687 string value, we would be back at the foo.
5689 Because this is used only in specific cases, we don't need to
5690 check all the things that `on_failure_jump' does, to make
5691 sure the right things get saved on the stack. Hence we don't
5692 share its code. The only reason to push anything on the
5693 stack at all is that otherwise we would have to change
5694 `anychar's code to do something besides goto fail in this
5695 case; that seems worse than this. */
5696 case on_failure_keep_string_jump
:
5697 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5698 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5701 PUSH_FAILURE_POINT (p
- 3, NULL
);
5704 /* A nasty loop is introduced by the non-greedy *? and +?.
5705 With such loops, the stack only ever contains one failure point
5706 at a time, so that a plain on_failure_jump_loop kind of
5707 cycle detection cannot work. Worse yet, such a detection
5708 can not only fail to detect a cycle, but it can also wrongly
5709 detect a cycle (between different instantiations of the same
5711 So the method used for those nasty loops is a little different:
5712 We use a special cycle-detection-stack-frame which is pushed
5713 when the on_failure_jump_nastyloop failure-point is *popped*.
5714 This special frame thus marks the beginning of one iteration
5715 through the loop and we can hence easily check right here
5716 whether something matched between the beginning and the end of
5718 case on_failure_jump_nastyloop
:
5719 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5720 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5723 assert ((re_opcode_t
)p
[-4] == no_op
);
5726 CHECK_INFINITE_LOOP (p
- 4, d
);
5728 /* If there's a cycle, just continue without pushing
5729 this failure point. The failure point is the "try again"
5730 option, which shouldn't be tried.
5731 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5732 PUSH_FAILURE_POINT (p
- 3, d
);
5736 /* Simple loop detecting on_failure_jump: just check on the
5737 failure stack if the same spot was already hit earlier. */
5738 case on_failure_jump_loop
:
5740 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5741 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5745 CHECK_INFINITE_LOOP (p
- 3, d
);
5747 /* If there's a cycle, get out of the loop, as if the matching
5748 had failed. We used to just `goto fail' here, but that was
5749 aborting the search a bit too early: we want to keep the
5750 empty-loop-match and keep matching after the loop.
5751 We want (x?)*y\1z to match both xxyz and xxyxz. */
5754 PUSH_FAILURE_POINT (p
- 3, d
);
5759 /* Uses of on_failure_jump:
5761 Each alternative starts with an on_failure_jump that points
5762 to the beginning of the next alternative. Each alternative
5763 except the last ends with a jump that in effect jumps past
5764 the rest of the alternatives. (They really jump to the
5765 ending jump of the following alternative, because tensioning
5766 these jumps is a hassle.)
5768 Repeats start with an on_failure_jump that points past both
5769 the repetition text and either the following jump or
5770 pop_failure_jump back to this on_failure_jump. */
5771 case on_failure_jump
:
5772 IMMEDIATE_QUIT_CHECK
;
5773 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5774 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5777 PUSH_FAILURE_POINT (p
-3, d
);
5780 /* This operation is used for greedy *.
5781 Compare the beginning of the repeat with what in the
5782 pattern follows its end. If we can establish that there
5783 is nothing that they would both match, i.e., that we
5784 would have to backtrack because of (as in, e.g., `a*a')
5785 then we can use a non-backtracking loop based on
5786 on_failure_keep_string_jump instead of on_failure_jump. */
5787 case on_failure_jump_smart
:
5788 IMMEDIATE_QUIT_CHECK
;
5789 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5790 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5793 re_char
*p1
= p
; /* Next operation. */
5794 /* Here, we discard `const', making re_match non-reentrant. */
5795 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5796 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5798 p
-= 3; /* Reset so that we will re-execute the
5799 instruction once it's been changed. */
5801 EXTRACT_NUMBER (mcnt
, p2
- 2);
5803 /* Ensure this is a indeed the trivial kind of loop
5804 we are expecting. */
5805 assert (skip_one_char (p1
) == p2
- 3);
5806 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5807 DEBUG_STATEMENT (debug
+= 2);
5808 if (mutually_exclusive_p (bufp
, p1
, p2
))
5810 /* Use a fast `on_failure_keep_string_jump' loop. */
5811 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5812 *p3
= (unsigned char) on_failure_keep_string_jump
;
5813 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5817 /* Default to a safe `on_failure_jump' loop. */
5818 DEBUG_PRINT1 (" smart default => slow loop.\n");
5819 *p3
= (unsigned char) on_failure_jump
;
5821 DEBUG_STATEMENT (debug
-= 2);
5825 /* Unconditionally jump (without popping any failure points). */
5828 IMMEDIATE_QUIT_CHECK
;
5829 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5830 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5831 p
+= mcnt
; /* Do the jump. */
5832 DEBUG_PRINT2 ("(to %p).\n", p
);
5836 /* Have to succeed matching what follows at least n times.
5837 After that, handle like `on_failure_jump'. */
5839 /* Signedness doesn't matter since we only compare MCNT to 0. */
5840 EXTRACT_NUMBER (mcnt
, p
+ 2);
5841 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5843 /* Originally, mcnt is how many times we HAVE to succeed. */
5846 /* Here, we discard `const', making re_match non-reentrant. */
5847 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5850 PUSH_NUMBER (p2
, mcnt
);
5853 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5858 /* Signedness doesn't matter since we only compare MCNT to 0. */
5859 EXTRACT_NUMBER (mcnt
, p
+ 2);
5860 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5862 /* Originally, this is how many times we CAN jump. */
5865 /* Here, we discard `const', making re_match non-reentrant. */
5866 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5868 PUSH_NUMBER (p2
, mcnt
);
5869 goto unconditional_jump
;
5871 /* If don't have to jump any more, skip over the rest of command. */
5878 unsigned char *p2
; /* Location of the counter. */
5879 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5881 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5882 /* Here, we discard `const', making re_match non-reentrant. */
5883 p2
= (unsigned char*) p
+ mcnt
;
5884 /* Signedness doesn't matter since we only copy MCNT's bits . */
5885 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5886 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
5887 PUSH_NUMBER (p2
, mcnt
);
5893 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5894 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5896 /* We SUCCEED (or FAIL) in one of the following cases: */
5898 /* Case 1: D is at the beginning or the end of string. */
5899 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5903 /* C1 is the character before D, S1 is the syntax of C1, C2
5904 is the character at D, and S2 is the syntax of C2. */
5909 int offset
= PTR_TO_OFFSET (d
- 1);
5910 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5911 UPDATE_SYNTAX_TABLE (charpos
);
5913 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5916 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5918 PREFETCH_NOLIMIT ();
5919 GET_CHAR_AFTER (c2
, d
, dummy
);
5922 if (/* Case 2: Only one of S1 and S2 is Sword. */
5923 ((s1
== Sword
) != (s2
== Sword
))
5924 /* Case 3: Both of S1 and S2 are Sword, and macro
5925 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5926 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5935 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5937 /* We FAIL in one of the following cases: */
5939 /* Case 1: D is at the end of string. */
5940 if (AT_STRINGS_END (d
))
5944 /* C1 is the character before D, S1 is the syntax of C1, C2
5945 is the character at D, and S2 is the syntax of C2. */
5950 int offset
= PTR_TO_OFFSET (d
);
5951 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5952 UPDATE_SYNTAX_TABLE (charpos
);
5955 GET_CHAR_AFTER (c2
, d
, dummy
);
5958 /* Case 2: S2 is not Sword. */
5962 /* Case 3: D is not at the beginning of string ... */
5963 if (!AT_STRINGS_BEG (d
))
5965 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5967 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5971 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5973 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5980 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5982 /* We FAIL in one of the following cases: */
5984 /* Case 1: D is at the beginning of string. */
5985 if (AT_STRINGS_BEG (d
))
5989 /* C1 is the character before D, S1 is the syntax of C1, C2
5990 is the character at D, and S2 is the syntax of C2. */
5995 int offset
= PTR_TO_OFFSET (d
) - 1;
5996 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5997 UPDATE_SYNTAX_TABLE (charpos
);
5999 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6002 /* Case 2: S1 is not Sword. */
6006 /* Case 3: D is not at the end of string ... */
6007 if (!AT_STRINGS_END (d
))
6009 PREFETCH_NOLIMIT ();
6010 GET_CHAR_AFTER (c2
, d
, dummy
);
6012 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6016 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6018 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6025 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6027 /* We FAIL in one of the following cases: */
6029 /* Case 1: D is at the end of string. */
6030 if (AT_STRINGS_END (d
))
6034 /* C1 is the character before D, S1 is the syntax of C1, C2
6035 is the character at D, and S2 is the syntax of C2. */
6039 int offset
= PTR_TO_OFFSET (d
);
6040 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6041 UPDATE_SYNTAX_TABLE (charpos
);
6044 c2
= RE_STRING_CHAR (d
, dend
- d
);
6047 /* Case 2: S2 is neither Sword nor Ssymbol. */
6048 if (s2
!= Sword
&& s2
!= Ssymbol
)
6051 /* Case 3: D is not at the beginning of string ... */
6052 if (!AT_STRINGS_BEG (d
))
6054 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6056 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6060 /* ... and S1 is Sword or Ssymbol. */
6061 if (s1
== Sword
|| s1
== Ssymbol
)
6068 DEBUG_PRINT1 ("EXECUTING symend.\n");
6070 /* We FAIL in one of the following cases: */
6072 /* Case 1: D is at the beginning of string. */
6073 if (AT_STRINGS_BEG (d
))
6077 /* C1 is the character before D, S1 is the syntax of C1, C2
6078 is the character at D, and S2 is the syntax of C2. */
6082 int offset
= PTR_TO_OFFSET (d
) - 1;
6083 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6084 UPDATE_SYNTAX_TABLE (charpos
);
6086 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6089 /* Case 2: S1 is neither Ssymbol nor Sword. */
6090 if (s1
!= Sword
&& s1
!= Ssymbol
)
6093 /* Case 3: D is not at the end of string ... */
6094 if (!AT_STRINGS_END (d
))
6096 PREFETCH_NOLIMIT ();
6097 c2
= RE_STRING_CHAR (d
, dend
- d
);
6099 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6103 /* ... and S2 is Sword or Ssymbol. */
6104 if (s2
== Sword
|| s2
== Ssymbol
)
6112 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6114 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
6118 int offset
= PTR_TO_OFFSET (d
);
6119 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6120 UPDATE_SYNTAX_TABLE (pos1
);
6127 GET_CHAR_AFTER (c
, d
, len
);
6128 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6136 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6137 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6142 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6143 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6148 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6149 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6154 case notcategoryspec
:
6155 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6157 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
6163 GET_CHAR_AFTER (c
, d
, len
);
6164 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6175 continue; /* Successfully executed one pattern command; keep going. */
6178 /* We goto here if a matching operation fails. */
6180 IMMEDIATE_QUIT_CHECK
;
6181 if (!FAIL_STACK_EMPTY ())
6184 /* A restart point is known. Restore to that state. */
6185 DEBUG_PRINT1 ("\nFAIL:\n");
6186 POP_FAILURE_POINT (str
, pat
);
6187 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
6189 case on_failure_keep_string_jump
:
6190 assert (str
== NULL
);
6191 goto continue_failure_jump
;
6193 case on_failure_jump_nastyloop
:
6194 assert ((re_opcode_t
)pat
[-2] == no_op
);
6195 PUSH_FAILURE_POINT (pat
- 2, str
);
6198 case on_failure_jump_loop
:
6199 case on_failure_jump
:
6202 continue_failure_jump
:
6203 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6208 /* A special frame used for nastyloops. */
6215 assert (p
>= bufp
->buffer
&& p
<= pend
);
6217 if (d
>= string1
&& d
<= end1
)
6221 break; /* Matching at this starting point really fails. */
6225 goto restore_best_regs
;
6229 return -1; /* Failure to match. */
6232 /* Subroutine definitions for re_match_2. */
6234 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6235 bytes; nonzero otherwise. */
6238 bcmp_translate (s1
, s2
, len
, translate
, multibyte
)
6241 RE_TRANSLATE_TYPE translate
;
6242 const int multibyte
;
6244 register re_char
*p1
= s1
, *p2
= s2
;
6245 re_char
*p1_end
= s1
+ len
;
6246 re_char
*p2_end
= s2
+ len
;
6248 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6249 different lengths, but relying on a single `len' would break this. -sm */
6250 while (p1
< p1_end
&& p2
< p2_end
)
6252 int p1_charlen
, p2_charlen
;
6253 re_wchar_t p1_ch
, p2_ch
;
6255 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6256 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6258 if (RE_TRANSLATE (translate
, p1_ch
)
6259 != RE_TRANSLATE (translate
, p2_ch
))
6262 p1
+= p1_charlen
, p2
+= p2_charlen
;
6265 if (p1
!= p1_end
|| p2
!= p2_end
)
6271 /* Entry points for GNU code. */
6273 /* re_compile_pattern is the GNU regular expression compiler: it
6274 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6275 Returns 0 if the pattern was valid, otherwise an error string.
6277 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6278 are set in BUFP on entry.
6280 We call regex_compile to do the actual compilation. */
6283 re_compile_pattern (pattern
, length
, bufp
)
6284 const char *pattern
;
6286 struct re_pattern_buffer
*bufp
;
6290 /* GNU code is written to assume at least RE_NREGS registers will be set
6291 (and at least one extra will be -1). */
6292 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6294 /* And GNU code determines whether or not to get register information
6295 by passing null for the REGS argument to re_match, etc., not by
6299 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6303 return gettext (re_error_msgid
[(int) ret
]);
6305 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6307 /* Entry points compatible with 4.2 BSD regex library. We don't define
6308 them unless specifically requested. */
6310 #if defined _REGEX_RE_COMP || defined _LIBC
6312 /* BSD has one and only one pattern buffer. */
6313 static struct re_pattern_buffer re_comp_buf
;
6317 /* Make these definitions weak in libc, so POSIX programs can redefine
6318 these names if they don't use our functions, and still use
6319 regcomp/regexec below without link errors. */
6329 if (!re_comp_buf
.buffer
)
6330 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6331 return (char *) gettext ("No previous regular expression");
6335 if (!re_comp_buf
.buffer
)
6337 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6338 if (re_comp_buf
.buffer
== NULL
)
6339 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6340 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6341 re_comp_buf
.allocated
= 200;
6343 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6344 if (re_comp_buf
.fastmap
== NULL
)
6345 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6346 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6349 /* Since `re_exec' always passes NULL for the `regs' argument, we
6350 don't need to initialize the pattern buffer fields which affect it. */
6352 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6357 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6358 return (char *) gettext (re_error_msgid
[(int) ret
]);
6369 const int len
= strlen (s
);
6371 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6373 #endif /* _REGEX_RE_COMP */
6375 /* POSIX.2 functions. Don't define these for Emacs. */
6379 /* regcomp takes a regular expression as a string and compiles it.
6381 PREG is a regex_t *. We do not expect any fields to be initialized,
6382 since POSIX says we shouldn't. Thus, we set
6384 `buffer' to the compiled pattern;
6385 `used' to the length of the compiled pattern;
6386 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6387 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6388 RE_SYNTAX_POSIX_BASIC;
6389 `fastmap' to an allocated space for the fastmap;
6390 `fastmap_accurate' to zero;
6391 `re_nsub' to the number of subexpressions in PATTERN.
6393 PATTERN is the address of the pattern string.
6395 CFLAGS is a series of bits which affect compilation.
6397 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6398 use POSIX basic syntax.
6400 If REG_NEWLINE is set, then . and [^...] don't match newline.
6401 Also, regexec will try a match beginning after every newline.
6403 If REG_ICASE is set, then we considers upper- and lowercase
6404 versions of letters to be equivalent when matching.
6406 If REG_NOSUB is set, then when PREG is passed to regexec, that
6407 routine will report only success or failure, and nothing about the
6410 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6411 the return codes and their meanings.) */
6414 regcomp (preg
, pattern
, cflags
)
6415 regex_t
*__restrict preg
;
6416 const char *__restrict pattern
;
6421 = (cflags
& REG_EXTENDED
) ?
6422 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6424 /* regex_compile will allocate the space for the compiled pattern. */
6426 preg
->allocated
= 0;
6429 /* Try to allocate space for the fastmap. */
6430 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6432 if (cflags
& REG_ICASE
)
6437 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6438 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6439 if (preg
->translate
== NULL
)
6440 return (int) REG_ESPACE
;
6442 /* Map uppercase characters to corresponding lowercase ones. */
6443 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6444 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6447 preg
->translate
= NULL
;
6449 /* If REG_NEWLINE is set, newlines are treated differently. */
6450 if (cflags
& REG_NEWLINE
)
6451 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6452 syntax
&= ~RE_DOT_NEWLINE
;
6453 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6456 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6458 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6460 /* POSIX says a null character in the pattern terminates it, so we
6461 can use strlen here in compiling the pattern. */
6462 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6464 /* POSIX doesn't distinguish between an unmatched open-group and an
6465 unmatched close-group: both are REG_EPAREN. */
6466 if (ret
== REG_ERPAREN
)
6469 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6470 { /* Compute the fastmap now, since regexec cannot modify the pattern
6472 re_compile_fastmap (preg
);
6473 if (preg
->can_be_null
)
6474 { /* The fastmap can't be used anyway. */
6475 free (preg
->fastmap
);
6476 preg
->fastmap
= NULL
;
6481 WEAK_ALIAS (__regcomp
, regcomp
)
6484 /* regexec searches for a given pattern, specified by PREG, in the
6487 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6488 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6489 least NMATCH elements, and we set them to the offsets of the
6490 corresponding matched substrings.
6492 EFLAGS specifies `execution flags' which affect matching: if
6493 REG_NOTBOL is set, then ^ does not match at the beginning of the
6494 string; if REG_NOTEOL is set, then $ does not match at the end.
6496 We return 0 if we find a match and REG_NOMATCH if not. */
6499 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6500 const regex_t
*__restrict preg
;
6501 const char *__restrict string
;
6503 regmatch_t pmatch
[__restrict_arr
];
6507 struct re_registers regs
;
6508 regex_t private_preg
;
6509 int len
= strlen (string
);
6510 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6512 private_preg
= *preg
;
6514 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6515 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6517 /* The user has told us exactly how many registers to return
6518 information about, via `nmatch'. We have to pass that on to the
6519 matching routines. */
6520 private_preg
.regs_allocated
= REGS_FIXED
;
6524 regs
.num_regs
= nmatch
;
6525 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6526 if (regs
.start
== NULL
)
6527 return (int) REG_NOMATCH
;
6528 regs
.end
= regs
.start
+ nmatch
;
6531 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6532 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6533 was a little bit longer but still only matching the real part.
6534 This works because the `endline' will check for a '\n' and will find a
6535 '\0', correctly deciding that this is not the end of a line.
6536 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6537 a convenient '\0' there. For all we know, the string could be preceded
6538 by '\n' which would throw things off. */
6540 /* Perform the searching operation. */
6541 ret
= re_search (&private_preg
, string
, len
,
6542 /* start: */ 0, /* range: */ len
,
6543 want_reg_info
? ®s
: (struct re_registers
*) 0);
6545 /* Copy the register information to the POSIX structure. */
6552 for (r
= 0; r
< nmatch
; r
++)
6554 pmatch
[r
].rm_so
= regs
.start
[r
];
6555 pmatch
[r
].rm_eo
= regs
.end
[r
];
6559 /* If we needed the temporary register info, free the space now. */
6563 /* We want zero return to mean success, unlike `re_search'. */
6564 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6566 WEAK_ALIAS (__regexec
, regexec
)
6569 /* Returns a message corresponding to an error code, ERRCODE, returned
6570 from either regcomp or regexec. We don't use PREG here. */
6573 regerror (errcode
, preg
, errbuf
, errbuf_size
)
6575 const regex_t
*preg
;
6583 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6584 /* Only error codes returned by the rest of the code should be passed
6585 to this routine. If we are given anything else, or if other regex
6586 code generates an invalid error code, then the program has a bug.
6587 Dump core so we can fix it. */
6590 msg
= gettext (re_error_msgid
[errcode
]);
6592 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6594 if (errbuf_size
!= 0)
6596 if (msg_size
> errbuf_size
)
6598 strncpy (errbuf
, msg
, errbuf_size
- 1);
6599 errbuf
[errbuf_size
- 1] = 0;
6602 strcpy (errbuf
, msg
);
6607 WEAK_ALIAS (__regerror
, regerror
)
6610 /* Free dynamically allocated space used by PREG. */
6616 if (preg
->buffer
!= NULL
)
6617 free (preg
->buffer
);
6618 preg
->buffer
= NULL
;
6620 preg
->allocated
= 0;
6623 if (preg
->fastmap
!= NULL
)
6624 free (preg
->fastmap
);
6625 preg
->fastmap
= NULL
;
6626 preg
->fastmap_accurate
= 0;
6628 if (preg
->translate
!= NULL
)
6629 free (preg
->translate
);
6630 preg
->translate
= NULL
;
6632 WEAK_ALIAS (__regfree
, regfree
)
6634 #endif /* not emacs */
6636 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6637 (do not change this comment) */