1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the
3 internationalization features.)
5 Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
6 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
7 Free Software Foundation, Inc.
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 3, or (at your option)
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; if not, write to the Free Software
21 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
25 - structure the opcode space into opcode+flag.
26 - merge with glibc's regex.[ch].
27 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
28 need to modify the compiled regexp so that re_match can be reentrant.
29 - get rid of on_failure_jump_smart by doing the optimization in re_comp
30 rather than at run-time, so that re_match can be reentrant.
33 /* AIX requires this to be the first thing in the file. */
34 #if defined _AIX && !defined REGEX_MALLOC
42 #if defined STDC_HEADERS && !defined emacs
45 /* We need this for `regex.h', and perhaps for the Emacs include files. */
46 # include <sys/types.h>
49 /* Whether to use ISO C Amendment 1 wide char functions.
50 Those should not be used for Emacs since it uses its own. */
52 #define WIDE_CHAR_SUPPORT 1
54 #define WIDE_CHAR_SUPPORT \
55 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
58 /* For platform which support the ISO C amendement 1 functionality we
59 support user defined character classes. */
61 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
67 /* We have to keep the namespace clean. */
68 # define regfree(preg) __regfree (preg)
69 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
70 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
71 # define regerror(err_code, preg, errbuf, errbuf_size) \
72 __regerror(err_code, preg, errbuf, errbuf_size)
73 # define re_set_registers(bu, re, nu, st, en) \
74 __re_set_registers (bu, re, nu, st, en)
75 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
76 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
77 # define re_match(bufp, string, size, pos, regs) \
78 __re_match (bufp, string, size, pos, regs)
79 # define re_search(bufp, string, size, startpos, range, regs) \
80 __re_search (bufp, string, size, startpos, range, regs)
81 # define re_compile_pattern(pattern, length, bufp) \
82 __re_compile_pattern (pattern, length, bufp)
83 # define re_set_syntax(syntax) __re_set_syntax (syntax)
84 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
85 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
86 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
88 /* Make sure we call libc's function even if the user overrides them. */
89 # define btowc __btowc
90 # define iswctype __iswctype
91 # define wctype __wctype
93 # define WEAK_ALIAS(a,b) weak_alias (a, b)
95 /* We are also using some library internals. */
96 # include <locale/localeinfo.h>
97 # include <locale/elem-hash.h>
98 # include <langinfo.h>
100 # define WEAK_ALIAS(a,b)
103 /* This is for other GNU distributions with internationalized messages. */
104 #if HAVE_LIBINTL_H || defined _LIBC
105 # include <libintl.h>
107 # define gettext(msgid) (msgid)
111 /* This define is so xgettext can find the internationalizable
113 # define gettext_noop(String) String
116 /* The `emacs' switch turns on certain matching commands
117 that make sense only in Emacs. */
124 /* Make syntax table lookup grant data in gl_state. */
125 # define SYNTAX_ENTRY_VIA_PROPERTY
128 # include "character.h"
129 # include "category.h"
134 # define malloc xmalloc
138 # define realloc xrealloc
144 /* Converts the pointer to the char to BEG-based offset from the start. */
145 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
146 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
148 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
149 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
150 # define RE_STRING_CHAR(p, multibyte) \
151 (multibyte ? (STRING_CHAR (p)) : (*(p)))
152 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
153 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
155 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
157 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
159 /* Set C a (possibly converted to multibyte) character before P. P
160 points into a string which is the virtual concatenation of STR1
161 (which ends at END1) or STR2 (which ends at END2). */
162 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
164 if (target_multibyte) \
166 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
167 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
168 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
169 c = STRING_CHAR (dtemp); \
173 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
174 (c) = RE_CHAR_TO_MULTIBYTE (c); \
178 /* Set C a (possibly converted to multibyte) character at P, and set
179 LEN to the byte length of that character. */
180 # define GET_CHAR_AFTER(c, p, len) \
182 if (target_multibyte) \
183 (c) = STRING_CHAR_AND_LENGTH (p, len); \
188 (c) = RE_CHAR_TO_MULTIBYTE (c); \
192 #else /* not emacs */
194 /* If we are not linking with Emacs proper,
195 we can't use the relocating allocator
196 even if config.h says that we can. */
199 # if defined STDC_HEADERS || defined _LIBC
206 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
213 val
= (void *) malloc (size
);
216 write (2, "virtual memory exhausted\n", 25);
223 xrealloc (block
, size
)
228 /* We must call malloc explicitly when BLOCK is 0, since some
229 reallocs don't do this. */
231 val
= (void *) malloc (size
);
233 val
= (void *) realloc (block
, size
);
236 write (2, "virtual memory exhausted\n", 25);
245 # define malloc xmalloc
249 # define realloc xrealloc
251 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
252 If nothing else has been done, use the method below. */
253 # ifdef INHIBIT_STRING_HEADER
254 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
255 # if !defined bzero && !defined bcopy
256 # undef INHIBIT_STRING_HEADER
261 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
262 This is used in most programs--a few other programs avoid this
263 by defining INHIBIT_STRING_HEADER. */
264 # ifndef INHIBIT_STRING_HEADER
265 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
269 # define bzero(s, n) (memset (s, '\0', n), (s))
271 # define bzero(s, n) __bzero (s, n)
275 # include <strings.h>
277 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
280 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
285 /* Define the syntax stuff for \<, \>, etc. */
287 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
288 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
290 # define SWITCH_ENUM_CAST(x) (x)
292 /* Dummy macros for non-Emacs environments. */
293 # define BASE_LEADING_CODE_P(c) (0)
294 # define CHAR_CHARSET(c) 0
295 # define CHARSET_LEADING_CODE_BASE(c) 0
296 # define MAX_MULTIBYTE_LENGTH 1
297 # define RE_MULTIBYTE_P(x) 0
298 # define RE_TARGET_MULTIBYTE_P(x) 0
299 # define WORD_BOUNDARY_P(c1, c2) (0)
300 # define CHAR_HEAD_P(p) (1)
301 # define SINGLE_BYTE_CHAR_P(c) (1)
302 # define SAME_CHARSET_P(c1, c2) (1)
303 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
304 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
305 # define STRING_CHAR(p) (*(p))
306 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
307 # define CHAR_STRING(c, s) (*(s) = (c), 1)
308 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
309 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
310 # define RE_CHAR_TO_MULTIBYTE(c) (c)
311 # define RE_CHAR_TO_UNIBYTE(c) (c)
312 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
313 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
314 # define GET_CHAR_AFTER(c, p, len) \
316 # define MAKE_CHAR(charset, c1, c2) (c1)
317 # define BYTE8_TO_CHAR(c) (c)
318 # define CHAR_BYTE8_P(c) (0)
319 # define CHAR_LEADING_CODE(c) (c)
321 #endif /* not emacs */
324 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
325 # define RE_TRANSLATE_P(TBL) (TBL)
328 /* Get the interface, including the syntax bits. */
331 /* isalpha etc. are used for the character classes. */
336 /* 1 if C is an ASCII character. */
337 # define IS_REAL_ASCII(c) ((c) < 0200)
339 /* 1 if C is a unibyte character. */
340 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
342 /* The Emacs definitions should not be directly affected by locales. */
344 /* In Emacs, these are only used for single-byte characters. */
345 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
346 # define ISCNTRL(c) ((c) < ' ')
347 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
348 || ((c) >= 'a' && (c) <= 'f') \
349 || ((c) >= 'A' && (c) <= 'F'))
351 /* This is only used for single-byte characters. */
352 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
354 /* The rest must handle multibyte characters. */
356 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
357 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
360 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
361 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
364 # define ISALNUM(c) (IS_REAL_ASCII (c) \
365 ? (((c) >= 'a' && (c) <= 'z') \
366 || ((c) >= 'A' && (c) <= 'Z') \
367 || ((c) >= '0' && (c) <= '9')) \
368 : SYNTAX (c) == Sword)
370 # define ISALPHA(c) (IS_REAL_ASCII (c) \
371 ? (((c) >= 'a' && (c) <= 'z') \
372 || ((c) >= 'A' && (c) <= 'Z')) \
373 : SYNTAX (c) == Sword)
375 # define ISLOWER(c) (LOWERCASEP (c))
377 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
378 ? ((c) > ' ' && (c) < 0177 \
379 && !(((c) >= 'a' && (c) <= 'z') \
380 || ((c) >= 'A' && (c) <= 'Z') \
381 || ((c) >= '0' && (c) <= '9'))) \
382 : SYNTAX (c) != Sword)
384 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
386 # define ISUPPER(c) (UPPERCASEP (c))
388 # define ISWORD(c) (SYNTAX (c) == Sword)
390 #else /* not emacs */
392 /* Jim Meyering writes:
394 "... Some ctype macros are valid only for character codes that
395 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
396 using /bin/cc or gcc but without giving an ansi option). So, all
397 ctype uses should be through macros like ISPRINT... If
398 STDC_HEADERS is defined, then autoconf has verified that the ctype
399 macros don't need to be guarded with references to isascii. ...
400 Defining isascii to 1 should let any compiler worth its salt
401 eliminate the && through constant folding."
402 Solaris defines some of these symbols so we must undefine them first. */
405 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
406 # define ISASCII(c) 1
408 # define ISASCII(c) isascii(c)
411 /* 1 if C is an ASCII character. */
412 # define IS_REAL_ASCII(c) ((c) < 0200)
414 /* This distinction is not meaningful, except in Emacs. */
415 # define ISUNIBYTE(c) 1
418 # define ISBLANK(c) (ISASCII (c) && isblank (c))
420 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
423 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
425 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
429 # define ISPRINT(c) (ISASCII (c) && isprint (c))
430 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
431 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
432 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
433 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
434 # define ISLOWER(c) (ISASCII (c) && islower (c))
435 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
436 # define ISSPACE(c) (ISASCII (c) && isspace (c))
437 # define ISUPPER(c) (ISASCII (c) && isupper (c))
438 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
440 # define ISWORD(c) ISALPHA(c)
443 # define TOLOWER(c) _tolower(c)
445 # define TOLOWER(c) tolower(c)
448 /* How many characters in the character set. */
449 # define CHAR_SET_SIZE 256
453 extern char *re_syntax_table
;
455 # else /* not SYNTAX_TABLE */
457 static char re_syntax_table
[CHAR_SET_SIZE
];
468 bzero (re_syntax_table
, sizeof re_syntax_table
);
470 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
472 re_syntax_table
[c
] = Sword
;
474 re_syntax_table
['_'] = Ssymbol
;
479 # endif /* not SYNTAX_TABLE */
481 # define SYNTAX(c) re_syntax_table[(c)]
483 #endif /* not emacs */
486 # define NULL (void *)0
489 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
490 since ours (we hope) works properly with all combinations of
491 machines, compilers, `char' and `unsigned char' argument types.
492 (Per Bothner suggested the basic approach.) */
493 #undef SIGN_EXTEND_CHAR
495 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
496 #else /* not __STDC__ */
497 /* As in Harbison and Steele. */
498 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
501 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
502 use `alloca' instead of `malloc'. This is because using malloc in
503 re_search* or re_match* could cause memory leaks when C-g is used in
504 Emacs; also, malloc is slower and causes storage fragmentation. On
505 the other hand, malloc is more portable, and easier to debug.
507 Because we sometimes use alloca, some routines have to be macros,
508 not functions -- `alloca'-allocated space disappears at the end of the
509 function it is called in. */
513 # define REGEX_ALLOCATE malloc
514 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
515 # define REGEX_FREE free
517 #else /* not REGEX_MALLOC */
519 /* Emacs already defines alloca, sometimes. */
522 /* Make alloca work the best possible way. */
524 # define alloca __builtin_alloca
525 # else /* not __GNUC__ */
526 # ifdef HAVE_ALLOCA_H
528 # endif /* HAVE_ALLOCA_H */
529 # endif /* not __GNUC__ */
531 # endif /* not alloca */
533 # define REGEX_ALLOCATE alloca
535 /* Assumes a `char *destination' variable. */
536 # define REGEX_REALLOCATE(source, osize, nsize) \
537 (destination = (char *) alloca (nsize), \
538 memcpy (destination, source, osize))
540 /* No need to do anything to free, after alloca. */
541 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
543 #endif /* not REGEX_MALLOC */
545 /* Define how to allocate the failure stack. */
547 #if defined REL_ALLOC && defined REGEX_MALLOC
549 # define REGEX_ALLOCATE_STACK(size) \
550 r_alloc (&failure_stack_ptr, (size))
551 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
552 r_re_alloc (&failure_stack_ptr, (nsize))
553 # define REGEX_FREE_STACK(ptr) \
554 r_alloc_free (&failure_stack_ptr)
556 #else /* not using relocating allocator */
560 # define REGEX_ALLOCATE_STACK malloc
561 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
562 # define REGEX_FREE_STACK free
564 # else /* not REGEX_MALLOC */
566 # define REGEX_ALLOCATE_STACK alloca
568 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
569 REGEX_REALLOCATE (source, osize, nsize)
570 /* No need to explicitly free anything. */
571 # define REGEX_FREE_STACK(arg) ((void)0)
573 # endif /* not REGEX_MALLOC */
574 #endif /* not using relocating allocator */
577 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
578 `string1' or just past its end. This works if PTR is NULL, which is
580 #define FIRST_STRING_P(ptr) \
581 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
583 /* (Re)Allocate N items of type T using malloc, or fail. */
584 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
585 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
586 #define RETALLOC_IF(addr, n, t) \
587 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
588 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
590 #define BYTEWIDTH 8 /* In bits. */
592 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
596 #define MAX(a, b) ((a) > (b) ? (a) : (b))
597 #define MIN(a, b) ((a) < (b) ? (a) : (b))
599 /* Type of source-pattern and string chars. */
600 typedef const unsigned char re_char
;
602 typedef char boolean
;
606 static int re_match_2_internal
_RE_ARGS ((struct re_pattern_buffer
*bufp
,
607 re_char
*string1
, int size1
,
608 re_char
*string2
, int size2
,
610 struct re_registers
*regs
,
613 /* These are the command codes that appear in compiled regular
614 expressions. Some opcodes are followed by argument bytes. A
615 command code can specify any interpretation whatsoever for its
616 arguments. Zero bytes may appear in the compiled regular expression. */
622 /* Succeed right away--no more backtracking. */
625 /* Followed by one byte giving n, then by n literal bytes. */
628 /* Matches any (more or less) character. */
631 /* Matches any one char belonging to specified set. First
632 following byte is number of bitmap bytes. Then come bytes
633 for a bitmap saying which chars are in. Bits in each byte
634 are ordered low-bit-first. A character is in the set if its
635 bit is 1. A character too large to have a bit in the map is
636 automatically not in the set.
638 If the length byte has the 0x80 bit set, then that stuff
639 is followed by a range table:
640 2 bytes of flags for character sets (low 8 bits, high 8 bits)
641 See RANGE_TABLE_WORK_BITS below.
642 2 bytes, the number of pairs that follow (upto 32767)
643 pairs, each 2 multibyte characters,
644 each multibyte character represented as 3 bytes. */
647 /* Same parameters as charset, but match any character that is
648 not one of those specified. */
651 /* Start remembering the text that is matched, for storing in a
652 register. Followed by one byte with the register number, in
653 the range 0 to one less than the pattern buffer's re_nsub
657 /* Stop remembering the text that is matched and store it in a
658 memory register. Followed by one byte with the register
659 number, in the range 0 to one less than `re_nsub' in the
663 /* Match a duplicate of something remembered. Followed by one
664 byte containing the register number. */
667 /* Fail unless at beginning of line. */
670 /* Fail unless at end of line. */
673 /* Succeeds if at beginning of buffer (if emacs) or at beginning
674 of string to be matched (if not). */
677 /* Analogously, for end of buffer/string. */
680 /* Followed by two byte relative address to which to jump. */
683 /* Followed by two-byte relative address of place to resume at
684 in case of failure. */
687 /* Like on_failure_jump, but pushes a placeholder instead of the
688 current string position when executed. */
689 on_failure_keep_string_jump
,
691 /* Just like `on_failure_jump', except that it checks that we
692 don't get stuck in an infinite loop (matching an empty string
694 on_failure_jump_loop
,
696 /* Just like `on_failure_jump_loop', except that it checks for
697 a different kind of loop (the kind that shows up with non-greedy
698 operators). This operation has to be immediately preceded
700 on_failure_jump_nastyloop
,
702 /* A smart `on_failure_jump' used for greedy * and + operators.
703 It analyses the loop before which it is put and if the
704 loop does not require backtracking, it changes itself to
705 `on_failure_keep_string_jump' and short-circuits the loop,
706 else it just defaults to changing itself into `on_failure_jump'.
707 It assumes that it is pointing to just past a `jump'. */
708 on_failure_jump_smart
,
710 /* Followed by two-byte relative address and two-byte number n.
711 After matching N times, jump to the address upon failure.
712 Does not work if N starts at 0: use on_failure_jump_loop
716 /* Followed by two-byte relative address, and two-byte number n.
717 Jump to the address N times, then fail. */
720 /* Set the following two-byte relative address to the
721 subsequent two-byte number. The address *includes* the two
725 wordbeg
, /* Succeeds if at word beginning. */
726 wordend
, /* Succeeds if at word end. */
728 wordbound
, /* Succeeds if at a word boundary. */
729 notwordbound
, /* Succeeds if not at a word boundary. */
731 symbeg
, /* Succeeds if at symbol beginning. */
732 symend
, /* Succeeds if at symbol end. */
734 /* Matches any character whose syntax is specified. Followed by
735 a byte which contains a syntax code, e.g., Sword. */
738 /* Matches any character whose syntax is not that specified. */
742 ,before_dot
, /* Succeeds if before point. */
743 at_dot
, /* Succeeds if at point. */
744 after_dot
, /* Succeeds if after point. */
746 /* Matches any character whose category-set contains the specified
747 category. The operator is followed by a byte which contains a
748 category code (mnemonic ASCII character). */
751 /* Matches any character whose category-set does not contain the
752 specified category. The operator is followed by a byte which
753 contains the category code (mnemonic ASCII character). */
758 /* Common operations on the compiled pattern. */
760 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
762 #define STORE_NUMBER(destination, number) \
764 (destination)[0] = (number) & 0377; \
765 (destination)[1] = (number) >> 8; \
768 /* Same as STORE_NUMBER, except increment DESTINATION to
769 the byte after where the number is stored. Therefore, DESTINATION
770 must be an lvalue. */
772 #define STORE_NUMBER_AND_INCR(destination, number) \
774 STORE_NUMBER (destination, number); \
775 (destination) += 2; \
778 /* Put into DESTINATION a number stored in two contiguous bytes starting
781 #define EXTRACT_NUMBER(destination, source) \
783 (destination) = *(source) & 0377; \
784 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
788 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
790 extract_number (dest
, source
)
794 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
795 *dest
= *source
& 0377;
799 # ifndef EXTRACT_MACROS /* To debug the macros. */
800 # undef EXTRACT_NUMBER
801 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
802 # endif /* not EXTRACT_MACROS */
806 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
807 SOURCE must be an lvalue. */
809 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
811 EXTRACT_NUMBER (destination, source); \
816 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
819 extract_number_and_incr (destination
, source
)
823 extract_number (destination
, *source
);
827 # ifndef EXTRACT_MACROS
828 # undef EXTRACT_NUMBER_AND_INCR
829 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
830 extract_number_and_incr (&dest, &src)
831 # endif /* not EXTRACT_MACROS */
835 /* Store a multibyte character in three contiguous bytes starting
836 DESTINATION, and increment DESTINATION to the byte after where the
837 character is stored. Therefore, DESTINATION must be an lvalue. */
839 #define STORE_CHARACTER_AND_INCR(destination, character) \
841 (destination)[0] = (character) & 0377; \
842 (destination)[1] = ((character) >> 8) & 0377; \
843 (destination)[2] = (character) >> 16; \
844 (destination) += 3; \
847 /* Put into DESTINATION a character stored in three contiguous bytes
848 starting at SOURCE. */
850 #define EXTRACT_CHARACTER(destination, source) \
852 (destination) = ((source)[0] \
853 | ((source)[1] << 8) \
854 | ((source)[2] << 16)); \
858 /* Macros for charset. */
860 /* Size of bitmap of charset P in bytes. P is a start of charset,
861 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
862 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
864 /* Nonzero if charset P has range table. */
865 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
867 /* Return the address of range table of charset P. But not the start
868 of table itself, but the before where the number of ranges is
869 stored. `2 +' means to skip re_opcode_t and size of bitmap,
870 and the 2 bytes of flags at the start of the range table. */
871 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
873 /* Extract the bit flags that start a range table. */
874 #define CHARSET_RANGE_TABLE_BITS(p) \
875 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
876 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
878 /* Test if C is listed in the bitmap of charset P. */
879 #define CHARSET_LOOKUP_BITMAP(p, c) \
880 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
881 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
883 /* Return the address of end of RANGE_TABLE. COUNT is number of
884 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
885 is start of range and end of range. `* 3' is size of each start
887 #define CHARSET_RANGE_TABLE_END(range_table, count) \
888 ((range_table) + (count) * 2 * 3)
890 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
891 COUNT is number of ranges in RANGE_TABLE. */
892 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
895 re_wchar_t range_start, range_end; \
897 re_char *range_table_end \
898 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
900 for (p = (range_table); p < range_table_end; p += 2 * 3) \
902 EXTRACT_CHARACTER (range_start, p); \
903 EXTRACT_CHARACTER (range_end, p + 3); \
905 if (range_start <= (c) && (c) <= range_end) \
914 /* Test if C is in range table of CHARSET. The flag NOT is negated if
915 C is listed in it. */
916 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
919 /* Number of ranges in range table. */ \
921 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
923 EXTRACT_NUMBER_AND_INCR (count, range_table); \
924 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
928 /* If DEBUG is defined, Regex prints many voluminous messages about what
929 it is doing (if the variable `debug' is nonzero). If linked with the
930 main program in `iregex.c', you can enter patterns and strings
931 interactively. And if linked with the main program in `main.c' and
932 the other test files, you can run the already-written tests. */
936 /* We use standard I/O for debugging. */
939 /* It is useful to test things that ``must'' be true when debugging. */
942 static int debug
= -100000;
944 # define DEBUG_STATEMENT(e) e
945 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
946 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
947 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
948 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
949 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
950 if (debug > 0) print_partial_compiled_pattern (s, e)
951 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
952 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
955 /* Print the fastmap in human-readable form. */
958 print_fastmap (fastmap
)
961 unsigned was_a_range
= 0;
964 while (i
< (1 << BYTEWIDTH
))
970 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
986 /* Print a compiled pattern string in human-readable form, starting at
987 the START pointer into it and ending just before the pointer END. */
990 print_partial_compiled_pattern (start
, end
)
1000 fprintf (stderr
, "(null)\n");
1004 /* Loop over pattern commands. */
1007 fprintf (stderr
, "%d:\t", p
- start
);
1009 switch ((re_opcode_t
) *p
++)
1012 fprintf (stderr
, "/no_op");
1016 fprintf (stderr
, "/succeed");
1021 fprintf (stderr
, "/exactn/%d", mcnt
);
1024 fprintf (stderr
, "/%c", *p
++);
1030 fprintf (stderr
, "/start_memory/%d", *p
++);
1034 fprintf (stderr
, "/stop_memory/%d", *p
++);
1038 fprintf (stderr
, "/duplicate/%d", *p
++);
1042 fprintf (stderr
, "/anychar");
1048 register int c
, last
= -100;
1049 register int in_range
= 0;
1050 int length
= CHARSET_BITMAP_SIZE (p
- 1);
1051 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
1053 fprintf (stderr
, "/charset [%s",
1054 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
1057 fprintf (stderr
, " !extends past end of pattern! ");
1059 for (c
= 0; c
< 256; c
++)
1061 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
1063 /* Are we starting a range? */
1064 if (last
+ 1 == c
&& ! in_range
)
1066 fprintf (stderr
, "-");
1069 /* Have we broken a range? */
1070 else if (last
+ 1 != c
&& in_range
)
1072 fprintf (stderr
, "%c", last
);
1077 fprintf (stderr
, "%c", c
);
1083 fprintf (stderr
, "%c", last
);
1085 fprintf (stderr
, "]");
1089 if (has_range_table
)
1092 fprintf (stderr
, "has-range-table");
1094 /* ??? Should print the range table; for now, just skip it. */
1095 p
+= 2; /* skip range table bits */
1096 EXTRACT_NUMBER_AND_INCR (count
, p
);
1097 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1103 fprintf (stderr
, "/begline");
1107 fprintf (stderr
, "/endline");
1110 case on_failure_jump
:
1111 extract_number_and_incr (&mcnt
, &p
);
1112 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1115 case on_failure_keep_string_jump
:
1116 extract_number_and_incr (&mcnt
, &p
);
1117 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1120 case on_failure_jump_nastyloop
:
1121 extract_number_and_incr (&mcnt
, &p
);
1122 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1125 case on_failure_jump_loop
:
1126 extract_number_and_incr (&mcnt
, &p
);
1127 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1130 case on_failure_jump_smart
:
1131 extract_number_and_incr (&mcnt
, &p
);
1132 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1136 extract_number_and_incr (&mcnt
, &p
);
1137 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1141 extract_number_and_incr (&mcnt
, &p
);
1142 extract_number_and_incr (&mcnt2
, &p
);
1143 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1147 extract_number_and_incr (&mcnt
, &p
);
1148 extract_number_and_incr (&mcnt2
, &p
);
1149 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1153 extract_number_and_incr (&mcnt
, &p
);
1154 extract_number_and_incr (&mcnt2
, &p
);
1155 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1159 fprintf (stderr
, "/wordbound");
1163 fprintf (stderr
, "/notwordbound");
1167 fprintf (stderr
, "/wordbeg");
1171 fprintf (stderr
, "/wordend");
1175 fprintf (stderr
, "/symbeg");
1179 fprintf (stderr
, "/symend");
1183 fprintf (stderr
, "/syntaxspec");
1185 fprintf (stderr
, "/%d", mcnt
);
1189 fprintf (stderr
, "/notsyntaxspec");
1191 fprintf (stderr
, "/%d", mcnt
);
1196 fprintf (stderr
, "/before_dot");
1200 fprintf (stderr
, "/at_dot");
1204 fprintf (stderr
, "/after_dot");
1208 fprintf (stderr
, "/categoryspec");
1210 fprintf (stderr
, "/%d", mcnt
);
1213 case notcategoryspec
:
1214 fprintf (stderr
, "/notcategoryspec");
1216 fprintf (stderr
, "/%d", mcnt
);
1221 fprintf (stderr
, "/begbuf");
1225 fprintf (stderr
, "/endbuf");
1229 fprintf (stderr
, "?%d", *(p
-1));
1232 fprintf (stderr
, "\n");
1235 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1240 print_compiled_pattern (bufp
)
1241 struct re_pattern_buffer
*bufp
;
1243 re_char
*buffer
= bufp
->buffer
;
1245 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1246 printf ("%ld bytes used/%ld bytes allocated.\n",
1247 bufp
->used
, bufp
->allocated
);
1249 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1251 printf ("fastmap: ");
1252 print_fastmap (bufp
->fastmap
);
1255 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1256 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1257 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1258 printf ("no_sub: %d\t", bufp
->no_sub
);
1259 printf ("not_bol: %d\t", bufp
->not_bol
);
1260 printf ("not_eol: %d\t", bufp
->not_eol
);
1261 printf ("syntax: %lx\n", bufp
->syntax
);
1263 /* Perhaps we should print the translate table? */
1268 print_double_string (where
, string1
, size1
, string2
, size2
)
1281 if (FIRST_STRING_P (where
))
1283 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1284 putchar (string1
[this_char
]);
1289 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1290 putchar (string2
[this_char
]);
1294 #else /* not DEBUG */
1299 # define DEBUG_STATEMENT(e)
1300 # define DEBUG_PRINT1(x)
1301 # define DEBUG_PRINT2(x1, x2)
1302 # define DEBUG_PRINT3(x1, x2, x3)
1303 # define DEBUG_PRINT4(x1, x2, x3, x4)
1304 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1305 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1307 #endif /* not DEBUG */
1309 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1310 also be assigned to arbitrarily: each pattern buffer stores its own
1311 syntax, so it can be changed between regex compilations. */
1312 /* This has no initializer because initialized variables in Emacs
1313 become read-only after dumping. */
1314 reg_syntax_t re_syntax_options
;
1317 /* Specify the precise syntax of regexps for compilation. This provides
1318 for compatibility for various utilities which historically have
1319 different, incompatible syntaxes.
1321 The argument SYNTAX is a bit mask comprised of the various bits
1322 defined in regex.h. We return the old syntax. */
1325 re_set_syntax (syntax
)
1326 reg_syntax_t syntax
;
1328 reg_syntax_t ret
= re_syntax_options
;
1330 re_syntax_options
= syntax
;
1333 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1335 /* Regexp to use to replace spaces, or NULL meaning don't. */
1336 static re_char
*whitespace_regexp
;
1339 re_set_whitespace_regexp (regexp
)
1342 whitespace_regexp
= (re_char
*) regexp
;
1344 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1346 /* This table gives an error message for each of the error codes listed
1347 in regex.h. Obviously the order here has to be same as there.
1348 POSIX doesn't require that we do anything for REG_NOERROR,
1349 but why not be nice? */
1351 static const char *re_error_msgid
[] =
1353 gettext_noop ("Success"), /* REG_NOERROR */
1354 gettext_noop ("No match"), /* REG_NOMATCH */
1355 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1356 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1357 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1358 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1359 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1360 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1361 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1362 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1363 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1364 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1365 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1366 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1367 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1368 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1369 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1370 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1373 /* Avoiding alloca during matching, to placate r_alloc. */
1375 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1376 searching and matching functions should not call alloca. On some
1377 systems, alloca is implemented in terms of malloc, and if we're
1378 using the relocating allocator routines, then malloc could cause a
1379 relocation, which might (if the strings being searched are in the
1380 ralloc heap) shift the data out from underneath the regexp
1383 Here's another reason to avoid allocation: Emacs
1384 processes input from X in a signal handler; processing X input may
1385 call malloc; if input arrives while a matching routine is calling
1386 malloc, then we're scrod. But Emacs can't just block input while
1387 calling matching routines; then we don't notice interrupts when
1388 they come in. So, Emacs blocks input around all regexp calls
1389 except the matching calls, which it leaves unprotected, in the
1390 faith that they will not malloc. */
1392 /* Normally, this is fine. */
1393 #define MATCH_MAY_ALLOCATE
1395 /* The match routines may not allocate if (1) they would do it with malloc
1396 and (2) it's not safe for them to use malloc.
1397 Note that if REL_ALLOC is defined, matching would not use malloc for the
1398 failure stack, but we would still use it for the register vectors;
1399 so REL_ALLOC should not affect this. */
1400 #if defined REGEX_MALLOC && defined emacs
1401 # undef MATCH_MAY_ALLOCATE
1405 /* Failure stack declarations and macros; both re_compile_fastmap and
1406 re_match_2 use a failure stack. These have to be macros because of
1407 REGEX_ALLOCATE_STACK. */
1410 /* Approximate number of failure points for which to initially allocate space
1411 when matching. If this number is exceeded, we allocate more
1412 space, so it is not a hard limit. */
1413 #ifndef INIT_FAILURE_ALLOC
1414 # define INIT_FAILURE_ALLOC 20
1417 /* Roughly the maximum number of failure points on the stack. Would be
1418 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1419 This is a variable only so users of regex can assign to it; we never
1420 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1421 before using it, so it should probably be a byte-count instead. */
1422 # if defined MATCH_MAY_ALLOCATE
1423 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1424 whose default stack limit is 2mb. In order for a larger
1425 value to work reliably, you have to try to make it accord
1426 with the process stack limit. */
1427 size_t re_max_failures
= 40000;
1429 size_t re_max_failures
= 4000;
1432 union fail_stack_elt
1435 /* This should be the biggest `int' that's no bigger than a pointer. */
1439 typedef union fail_stack_elt fail_stack_elt_t
;
1443 fail_stack_elt_t
*stack
;
1445 size_t avail
; /* Offset of next open position. */
1446 size_t frame
; /* Offset of the cur constructed frame. */
1449 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1450 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1453 /* Define macros to initialize and free the failure stack.
1454 Do `return -2' if the alloc fails. */
1456 #ifdef MATCH_MAY_ALLOCATE
1457 # define INIT_FAIL_STACK() \
1459 fail_stack.stack = (fail_stack_elt_t *) \
1460 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1461 * sizeof (fail_stack_elt_t)); \
1463 if (fail_stack.stack == NULL) \
1466 fail_stack.size = INIT_FAILURE_ALLOC; \
1467 fail_stack.avail = 0; \
1468 fail_stack.frame = 0; \
1471 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1473 # define INIT_FAIL_STACK() \
1475 fail_stack.avail = 0; \
1476 fail_stack.frame = 0; \
1479 # define RESET_FAIL_STACK() ((void)0)
1483 /* Double the size of FAIL_STACK, up to a limit
1484 which allows approximately `re_max_failures' items.
1486 Return 1 if succeeds, and 0 if either ran out of memory
1487 allocating space for it or it was already too large.
1489 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1491 /* Factor to increase the failure stack size by
1492 when we increase it.
1493 This used to be 2, but 2 was too wasteful
1494 because the old discarded stacks added up to as much space
1495 were as ultimate, maximum-size stack. */
1496 #define FAIL_STACK_GROWTH_FACTOR 4
1498 #define GROW_FAIL_STACK(fail_stack) \
1499 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1500 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1502 : ((fail_stack).stack \
1503 = (fail_stack_elt_t *) \
1504 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1505 (fail_stack).size * sizeof (fail_stack_elt_t), \
1506 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1507 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1508 * FAIL_STACK_GROWTH_FACTOR))), \
1510 (fail_stack).stack == NULL \
1512 : ((fail_stack).size \
1513 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1514 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1515 * FAIL_STACK_GROWTH_FACTOR)) \
1516 / sizeof (fail_stack_elt_t)), \
1520 /* Push a pointer value onto the failure stack.
1521 Assumes the variable `fail_stack'. Probably should only
1522 be called from within `PUSH_FAILURE_POINT'. */
1523 #define PUSH_FAILURE_POINTER(item) \
1524 fail_stack.stack[fail_stack.avail++].pointer = (item)
1526 /* This pushes an integer-valued item onto the failure stack.
1527 Assumes the variable `fail_stack'. Probably should only
1528 be called from within `PUSH_FAILURE_POINT'. */
1529 #define PUSH_FAILURE_INT(item) \
1530 fail_stack.stack[fail_stack.avail++].integer = (item)
1532 /* Push a fail_stack_elt_t value onto the failure stack.
1533 Assumes the variable `fail_stack'. Probably should only
1534 be called from within `PUSH_FAILURE_POINT'. */
1535 #define PUSH_FAILURE_ELT(item) \
1536 fail_stack.stack[fail_stack.avail++] = (item)
1538 /* These three POP... operations complement the three PUSH... operations.
1539 All assume that `fail_stack' is nonempty. */
1540 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1541 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1542 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1544 /* Individual items aside from the registers. */
1545 #define NUM_NONREG_ITEMS 3
1547 /* Used to examine the stack (to detect infinite loops). */
1548 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1549 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1550 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1551 #define TOP_FAILURE_HANDLE() fail_stack.frame
1554 #define ENSURE_FAIL_STACK(space) \
1555 while (REMAINING_AVAIL_SLOTS <= space) { \
1556 if (!GROW_FAIL_STACK (fail_stack)) \
1558 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1559 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1562 /* Push register NUM onto the stack. */
1563 #define PUSH_FAILURE_REG(num) \
1565 char *destination; \
1566 ENSURE_FAIL_STACK(3); \
1567 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1568 num, regstart[num], regend[num]); \
1569 PUSH_FAILURE_POINTER (regstart[num]); \
1570 PUSH_FAILURE_POINTER (regend[num]); \
1571 PUSH_FAILURE_INT (num); \
1574 /* Change the counter's value to VAL, but make sure that it will
1575 be reset when backtracking. */
1576 #define PUSH_NUMBER(ptr,val) \
1578 char *destination; \
1580 ENSURE_FAIL_STACK(3); \
1581 EXTRACT_NUMBER (c, ptr); \
1582 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1583 PUSH_FAILURE_INT (c); \
1584 PUSH_FAILURE_POINTER (ptr); \
1585 PUSH_FAILURE_INT (-1); \
1586 STORE_NUMBER (ptr, val); \
1589 /* Pop a saved register off the stack. */
1590 #define POP_FAILURE_REG_OR_COUNT() \
1592 int reg = POP_FAILURE_INT (); \
1595 /* It's a counter. */ \
1596 /* Here, we discard `const', making re_match non-reentrant. */ \
1597 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1598 reg = POP_FAILURE_INT (); \
1599 STORE_NUMBER (ptr, reg); \
1600 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1604 regend[reg] = POP_FAILURE_POINTER (); \
1605 regstart[reg] = POP_FAILURE_POINTER (); \
1606 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1607 reg, regstart[reg], regend[reg]); \
1611 /* Check that we are not stuck in an infinite loop. */
1612 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1614 int failure = TOP_FAILURE_HANDLE (); \
1615 /* Check for infinite matching loops */ \
1616 while (failure > 0 \
1617 && (FAILURE_STR (failure) == string_place \
1618 || FAILURE_STR (failure) == NULL)) \
1620 assert (FAILURE_PAT (failure) >= bufp->buffer \
1621 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1622 if (FAILURE_PAT (failure) == pat_cur) \
1627 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1628 failure = NEXT_FAILURE_HANDLE(failure); \
1630 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1633 /* Push the information about the state we will need
1634 if we ever fail back to it.
1636 Requires variables fail_stack, regstart, regend and
1637 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1640 Does `return FAILURE_CODE' if runs out of memory. */
1642 #define PUSH_FAILURE_POINT(pattern, string_place) \
1644 char *destination; \
1645 /* Must be int, so when we don't save any registers, the arithmetic \
1646 of 0 + -1 isn't done as unsigned. */ \
1648 DEBUG_STATEMENT (nfailure_points_pushed++); \
1649 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1650 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1651 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1653 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1655 DEBUG_PRINT1 ("\n"); \
1657 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1658 PUSH_FAILURE_INT (fail_stack.frame); \
1660 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1661 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1662 DEBUG_PRINT1 ("'\n"); \
1663 PUSH_FAILURE_POINTER (string_place); \
1665 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1666 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1667 PUSH_FAILURE_POINTER (pattern); \
1669 /* Close the frame by moving the frame pointer past it. */ \
1670 fail_stack.frame = fail_stack.avail; \
1673 /* Estimate the size of data pushed by a typical failure stack entry.
1674 An estimate is all we need, because all we use this for
1675 is to choose a limit for how big to make the failure stack. */
1676 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1677 #define TYPICAL_FAILURE_SIZE 20
1679 /* How many items can still be added to the stack without overflowing it. */
1680 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1683 /* Pops what PUSH_FAIL_STACK pushes.
1685 We restore into the parameters, all of which should be lvalues:
1686 STR -- the saved data position.
1687 PAT -- the saved pattern position.
1688 REGSTART, REGEND -- arrays of string positions.
1690 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1691 `pend', `string1', `size1', `string2', and `size2'. */
1693 #define POP_FAILURE_POINT(str, pat) \
1695 assert (!FAIL_STACK_EMPTY ()); \
1697 /* Remove failure points and point to how many regs pushed. */ \
1698 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1699 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1700 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1702 /* Pop the saved registers. */ \
1703 while (fail_stack.frame < fail_stack.avail) \
1704 POP_FAILURE_REG_OR_COUNT (); \
1706 pat = POP_FAILURE_POINTER (); \
1707 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1708 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1710 /* If the saved string location is NULL, it came from an \
1711 on_failure_keep_string_jump opcode, and we want to throw away the \
1712 saved NULL, thus retaining our current position in the string. */ \
1713 str = POP_FAILURE_POINTER (); \
1714 DEBUG_PRINT2 (" Popping string %p: `", str); \
1715 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1716 DEBUG_PRINT1 ("'\n"); \
1718 fail_stack.frame = POP_FAILURE_INT (); \
1719 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1721 assert (fail_stack.avail >= 0); \
1722 assert (fail_stack.frame <= fail_stack.avail); \
1724 DEBUG_STATEMENT (nfailure_points_popped++); \
1725 } while (0) /* POP_FAILURE_POINT */
1729 /* Registers are set to a sentinel when they haven't yet matched. */
1730 #define REG_UNSET(e) ((e) == NULL)
1732 /* Subroutine declarations and macros for regex_compile. */
1734 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1735 reg_syntax_t syntax
,
1736 struct re_pattern_buffer
*bufp
));
1737 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1738 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1739 int arg1
, int arg2
));
1740 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1741 int arg
, unsigned char *end
));
1742 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1743 int arg1
, int arg2
, unsigned char *end
));
1744 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1746 reg_syntax_t syntax
));
1747 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1749 reg_syntax_t syntax
));
1750 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1751 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1752 char *fastmap
, const int multibyte
));
1754 /* Fetch the next character in the uncompiled pattern, with no
1756 #define PATFETCH(c) \
1759 if (p == pend) return REG_EEND; \
1760 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1765 /* If `translate' is non-null, return translate[D], else just D. We
1766 cast the subscript to translate because some data is declared as
1767 `char *', to avoid warnings when a string constant is passed. But
1768 when we use a character as a subscript we must make it unsigned. */
1770 # define TRANSLATE(d) \
1771 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1775 /* Macros for outputting the compiled pattern into `buffer'. */
1777 /* If the buffer isn't allocated when it comes in, use this. */
1778 #define INIT_BUF_SIZE 32
1780 /* Make sure we have at least N more bytes of space in buffer. */
1781 #define GET_BUFFER_SPACE(n) \
1782 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1785 /* Make sure we have one more byte of buffer space and then add C to it. */
1786 #define BUF_PUSH(c) \
1788 GET_BUFFER_SPACE (1); \
1789 *b++ = (unsigned char) (c); \
1793 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1794 #define BUF_PUSH_2(c1, c2) \
1796 GET_BUFFER_SPACE (2); \
1797 *b++ = (unsigned char) (c1); \
1798 *b++ = (unsigned char) (c2); \
1802 /* As with BUF_PUSH_2, except for three bytes. */
1803 #define BUF_PUSH_3(c1, c2, c3) \
1805 GET_BUFFER_SPACE (3); \
1806 *b++ = (unsigned char) (c1); \
1807 *b++ = (unsigned char) (c2); \
1808 *b++ = (unsigned char) (c3); \
1812 /* Store a jump with opcode OP at LOC to location TO. We store a
1813 relative address offset by the three bytes the jump itself occupies. */
1814 #define STORE_JUMP(op, loc, to) \
1815 store_op1 (op, loc, (to) - (loc) - 3)
1817 /* Likewise, for a two-argument jump. */
1818 #define STORE_JUMP2(op, loc, to, arg) \
1819 store_op2 (op, loc, (to) - (loc) - 3, arg)
1821 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1822 #define INSERT_JUMP(op, loc, to) \
1823 insert_op1 (op, loc, (to) - (loc) - 3, b)
1825 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1826 #define INSERT_JUMP2(op, loc, to, arg) \
1827 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1830 /* This is not an arbitrary limit: the arguments which represent offsets
1831 into the pattern are two bytes long. So if 2^15 bytes turns out to
1832 be too small, many things would have to change. */
1833 # define MAX_BUF_SIZE (1L << 15)
1835 #if 0 /* This is when we thought it could be 2^16 bytes. */
1836 /* Any other compiler which, like MSC, has allocation limit below 2^16
1837 bytes will have to use approach similar to what was done below for
1838 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1839 reallocating to 0 bytes. Such thing is not going to work too well.
1840 You have been warned!! */
1841 #if defined _MSC_VER && !defined WIN32
1842 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1843 # define MAX_BUF_SIZE 65500L
1845 # define MAX_BUF_SIZE (1L << 16)
1849 /* Extend the buffer by twice its current size via realloc and
1850 reset the pointers that pointed into the old block to point to the
1851 correct places in the new one. If extending the buffer results in it
1852 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1853 #if __BOUNDED_POINTERS__
1854 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1855 # define MOVE_BUFFER_POINTER(P) \
1856 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1857 SET_HIGH_BOUND (P), \
1858 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1859 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1862 SET_HIGH_BOUND (b); \
1863 SET_HIGH_BOUND (begalt); \
1864 if (fixup_alt_jump) \
1865 SET_HIGH_BOUND (fixup_alt_jump); \
1867 SET_HIGH_BOUND (laststart); \
1868 if (pending_exact) \
1869 SET_HIGH_BOUND (pending_exact); \
1872 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1873 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1875 #define EXTEND_BUFFER() \
1877 unsigned char *old_buffer = bufp->buffer; \
1878 if (bufp->allocated == MAX_BUF_SIZE) \
1880 bufp->allocated <<= 1; \
1881 if (bufp->allocated > MAX_BUF_SIZE) \
1882 bufp->allocated = MAX_BUF_SIZE; \
1883 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1884 if (bufp->buffer == NULL) \
1885 return REG_ESPACE; \
1886 /* If the buffer moved, move all the pointers into it. */ \
1887 if (old_buffer != bufp->buffer) \
1889 unsigned char *new_buffer = bufp->buffer; \
1890 MOVE_BUFFER_POINTER (b); \
1891 MOVE_BUFFER_POINTER (begalt); \
1892 if (fixup_alt_jump) \
1893 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1895 MOVE_BUFFER_POINTER (laststart); \
1896 if (pending_exact) \
1897 MOVE_BUFFER_POINTER (pending_exact); \
1899 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1903 /* Since we have one byte reserved for the register number argument to
1904 {start,stop}_memory, the maximum number of groups we can report
1905 things about is what fits in that byte. */
1906 #define MAX_REGNUM 255
1908 /* But patterns can have more than `MAX_REGNUM' registers. We just
1909 ignore the excess. */
1910 typedef int regnum_t
;
1913 /* Macros for the compile stack. */
1915 /* Since offsets can go either forwards or backwards, this type needs to
1916 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1917 /* int may be not enough when sizeof(int) == 2. */
1918 typedef long pattern_offset_t
;
1922 pattern_offset_t begalt_offset
;
1923 pattern_offset_t fixup_alt_jump
;
1924 pattern_offset_t laststart_offset
;
1926 } compile_stack_elt_t
;
1931 compile_stack_elt_t
*stack
;
1933 unsigned avail
; /* Offset of next open position. */
1934 } compile_stack_type
;
1937 #define INIT_COMPILE_STACK_SIZE 32
1939 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1940 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1942 /* The next available element. */
1943 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1945 /* Explicit quit checking is only used on NTemacs and whenever we
1946 use polling to process input events. */
1947 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1948 extern int immediate_quit
;
1949 # define IMMEDIATE_QUIT_CHECK \
1951 if (immediate_quit) QUIT; \
1954 # define IMMEDIATE_QUIT_CHECK ((void)0)
1957 /* Structure to manage work area for range table. */
1958 struct range_table_work_area
1960 int *table
; /* actual work area. */
1961 int allocated
; /* allocated size for work area in bytes. */
1962 int used
; /* actually used size in words. */
1963 int bits
; /* flag to record character classes */
1966 /* Make sure that WORK_AREA can hold more N multibyte characters.
1967 This is used only in set_image_of_range and set_image_of_range_1.
1968 It expects WORK_AREA to be a pointer.
1969 If it can't get the space, it returns from the surrounding function. */
1971 #define EXTEND_RANGE_TABLE(work_area, n) \
1973 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1975 extend_range_table_work_area (&work_area); \
1976 if ((work_area).table == 0) \
1977 return (REG_ESPACE); \
1981 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1982 (work_area).bits |= (bit)
1984 /* Bits used to implement the multibyte-part of the various character classes
1985 such as [:alnum:] in a charset's range table. */
1986 #define BIT_WORD 0x1
1987 #define BIT_LOWER 0x2
1988 #define BIT_PUNCT 0x4
1989 #define BIT_SPACE 0x8
1990 #define BIT_UPPER 0x10
1991 #define BIT_MULTIBYTE 0x20
1993 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1994 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1996 EXTEND_RANGE_TABLE ((work_area), 2); \
1997 (work_area).table[(work_area).used++] = (range_start); \
1998 (work_area).table[(work_area).used++] = (range_end); \
2001 /* Free allocated memory for WORK_AREA. */
2002 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2004 if ((work_area).table) \
2005 free ((work_area).table); \
2008 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2009 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2010 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2011 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2014 /* Set the bit for character C in a list. */
2015 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2020 /* Store characters in the range FROM to TO in the bitmap at B (for
2021 ASCII and unibyte characters) and WORK_AREA (for multibyte
2022 characters) while translating them and paying attention to the
2023 continuity of translated characters.
2025 Implementation note: It is better to implement these fairly big
2026 macros by a function, but it's not that easy because macros called
2027 in this macro assume various local variables already declared. */
2029 /* Both FROM and TO are ASCII characters. */
2031 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2035 for (C0 = (FROM); C0 <= (TO); C0++) \
2037 C1 = TRANSLATE (C0); \
2038 if (! ASCII_CHAR_P (C1)) \
2040 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2041 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2044 SET_LIST_BIT (C1); \
2049 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2051 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2053 int C0, C1, C2, I; \
2054 int USED = RANGE_TABLE_WORK_USED (work_area); \
2056 for (C0 = (FROM); C0 <= (TO); C0++) \
2058 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2059 if (CHAR_BYTE8_P (C1)) \
2060 SET_LIST_BIT (C0); \
2063 C2 = TRANSLATE (C1); \
2065 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2067 SET_LIST_BIT (C1); \
2068 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2070 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2071 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2073 if (C2 >= from - 1 && C2 <= to + 1) \
2075 if (C2 == from - 1) \
2076 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2077 else if (C2 == to + 1) \
2078 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2083 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2089 /* Both FROM and TO are multibyte characters. */
2091 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2093 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2095 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2096 for (C0 = (FROM); C0 <= (TO); C0++) \
2098 C1 = TRANSLATE (C0); \
2099 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2100 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2101 SET_LIST_BIT (C2); \
2102 if (C1 >= (FROM) && C1 <= (TO)) \
2104 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2106 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2107 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2109 if (C1 >= from - 1 && C1 <= to + 1) \
2111 if (C1 == from - 1) \
2112 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2113 else if (C1 == to + 1) \
2114 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2119 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2125 /* Get the next unsigned number in the uncompiled pattern. */
2126 #define GET_UNSIGNED_NUMBER(num) \
2129 FREE_STACK_RETURN (REG_EBRACE); \
2133 while ('0' <= c && c <= '9') \
2139 num = num * 10 + c - '0'; \
2140 if (num / 10 != prev) \
2141 FREE_STACK_RETURN (REG_BADBR); \
2143 FREE_STACK_RETURN (REG_EBRACE); \
2149 #if ! WIDE_CHAR_SUPPORT
2151 /* Map a string to the char class it names (if any). */
2156 const char *string
= str
;
2157 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2158 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2159 else if (STREQ (string
, "word")) return RECC_WORD
;
2160 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2161 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2162 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2163 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2164 else if (STREQ (string
, "print")) return RECC_PRINT
;
2165 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2166 else if (STREQ (string
, "space")) return RECC_SPACE
;
2167 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2168 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2169 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2170 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2171 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2172 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2173 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2177 /* True if CH is in the char class CC. */
2179 re_iswctype (ch
, cc
)
2185 case RECC_ALNUM
: return ISALNUM (ch
);
2186 case RECC_ALPHA
: return ISALPHA (ch
);
2187 case RECC_BLANK
: return ISBLANK (ch
);
2188 case RECC_CNTRL
: return ISCNTRL (ch
);
2189 case RECC_DIGIT
: return ISDIGIT (ch
);
2190 case RECC_GRAPH
: return ISGRAPH (ch
);
2191 case RECC_LOWER
: return ISLOWER (ch
);
2192 case RECC_PRINT
: return ISPRINT (ch
);
2193 case RECC_PUNCT
: return ISPUNCT (ch
);
2194 case RECC_SPACE
: return ISSPACE (ch
);
2195 case RECC_UPPER
: return ISUPPER (ch
);
2196 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2197 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2198 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2199 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2200 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2201 case RECC_WORD
: return ISWORD (ch
);
2202 case RECC_ERROR
: return false;
2208 /* Return a bit-pattern to use in the range-table bits to match multibyte
2209 chars of class CC. */
2211 re_wctype_to_bit (cc
)
2216 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2217 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2218 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2219 case RECC_LOWER
: return BIT_LOWER
;
2220 case RECC_UPPER
: return BIT_UPPER
;
2221 case RECC_PUNCT
: return BIT_PUNCT
;
2222 case RECC_SPACE
: return BIT_SPACE
;
2223 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2224 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2231 /* Filling in the work area of a range. */
2233 /* Actually extend the space in WORK_AREA. */
2236 extend_range_table_work_area (work_area
)
2237 struct range_table_work_area
*work_area
;
2239 work_area
->allocated
+= 16 * sizeof (int);
2240 if (work_area
->table
)
2242 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2245 = (int *) malloc (work_area
->allocated
);
2251 /* Carefully find the ranges of codes that are equivalent
2252 under case conversion to the range start..end when passed through
2253 TRANSLATE. Handle the case where non-letters can come in between
2254 two upper-case letters (which happens in Latin-1).
2255 Also handle the case of groups of more than 2 case-equivalent chars.
2257 The basic method is to look at consecutive characters and see
2258 if they can form a run that can be handled as one.
2260 Returns -1 if successful, REG_ESPACE if ran out of space. */
2263 set_image_of_range_1 (work_area
, start
, end
, translate
)
2264 RE_TRANSLATE_TYPE translate
;
2265 struct range_table_work_area
*work_area
;
2266 re_wchar_t start
, end
;
2268 /* `one_case' indicates a character, or a run of characters,
2269 each of which is an isolate (no case-equivalents).
2270 This includes all ASCII non-letters.
2272 `two_case' indicates a character, or a run of characters,
2273 each of which has two case-equivalent forms.
2274 This includes all ASCII letters.
2276 `strange' indicates a character that has more than one
2279 enum case_type
{one_case
, two_case
, strange
};
2281 /* Describe the run that is in progress,
2282 which the next character can try to extend.
2283 If run_type is strange, that means there really is no run.
2284 If run_type is one_case, then run_start...run_end is the run.
2285 If run_type is two_case, then the run is run_start...run_end,
2286 and the case-equivalents end at run_eqv_end. */
2288 enum case_type run_type
= strange
;
2289 int run_start
, run_end
, run_eqv_end
;
2291 Lisp_Object eqv_table
;
2293 if (!RE_TRANSLATE_P (translate
))
2295 EXTEND_RANGE_TABLE (work_area
, 2);
2296 work_area
->table
[work_area
->used
++] = (start
);
2297 work_area
->table
[work_area
->used
++] = (end
);
2301 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2303 for (; start
<= end
; start
++)
2305 enum case_type this_type
;
2306 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2307 int minchar
, maxchar
;
2309 /* Classify this character */
2311 this_type
= one_case
;
2312 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2313 this_type
= two_case
;
2315 this_type
= strange
;
2318 minchar
= start
, maxchar
= eqv
;
2320 minchar
= eqv
, maxchar
= start
;
2322 /* Can this character extend the run in progress? */
2323 if (this_type
== strange
|| this_type
!= run_type
2324 || !(minchar
== run_end
+ 1
2325 && (run_type
== two_case
2326 ? maxchar
== run_eqv_end
+ 1 : 1)))
2329 Record each of its equivalent ranges. */
2330 if (run_type
== one_case
)
2332 EXTEND_RANGE_TABLE (work_area
, 2);
2333 work_area
->table
[work_area
->used
++] = run_start
;
2334 work_area
->table
[work_area
->used
++] = run_end
;
2336 else if (run_type
== two_case
)
2338 EXTEND_RANGE_TABLE (work_area
, 4);
2339 work_area
->table
[work_area
->used
++] = run_start
;
2340 work_area
->table
[work_area
->used
++] = run_end
;
2341 work_area
->table
[work_area
->used
++]
2342 = RE_TRANSLATE (eqv_table
, run_start
);
2343 work_area
->table
[work_area
->used
++]
2344 = RE_TRANSLATE (eqv_table
, run_end
);
2349 if (this_type
== strange
)
2351 /* For a strange character, add each of its equivalents, one
2352 by one. Don't start a range. */
2355 EXTEND_RANGE_TABLE (work_area
, 2);
2356 work_area
->table
[work_area
->used
++] = eqv
;
2357 work_area
->table
[work_area
->used
++] = eqv
;
2358 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2360 while (eqv
!= start
);
2363 /* Add this char to the run, or start a new run. */
2364 else if (run_type
== strange
)
2366 /* Initialize a new range. */
2367 run_type
= this_type
;
2370 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2374 /* Extend a running range. */
2376 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2380 /* If a run is still in progress at the end, finish it now
2381 by recording its equivalent ranges. */
2382 if (run_type
== one_case
)
2384 EXTEND_RANGE_TABLE (work_area
, 2);
2385 work_area
->table
[work_area
->used
++] = run_start
;
2386 work_area
->table
[work_area
->used
++] = run_end
;
2388 else if (run_type
== two_case
)
2390 EXTEND_RANGE_TABLE (work_area
, 4);
2391 work_area
->table
[work_area
->used
++] = run_start
;
2392 work_area
->table
[work_area
->used
++] = run_end
;
2393 work_area
->table
[work_area
->used
++]
2394 = RE_TRANSLATE (eqv_table
, run_start
);
2395 work_area
->table
[work_area
->used
++]
2396 = RE_TRANSLATE (eqv_table
, run_end
);
2404 /* Record the image of the range start..end when passed through
2405 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2406 and is not even necessarily contiguous.
2407 Normally we approximate it with the smallest contiguous range that contains
2408 all the chars we need. However, for Latin-1 we go to extra effort
2411 This function is not called for ASCII ranges.
2413 Returns -1 if successful, REG_ESPACE if ran out of space. */
2416 set_image_of_range (work_area
, start
, end
, translate
)
2417 RE_TRANSLATE_TYPE translate
;
2418 struct range_table_work_area
*work_area
;
2419 re_wchar_t start
, end
;
2421 re_wchar_t cmin
, cmax
;
2424 /* For Latin-1 ranges, use set_image_of_range_1
2425 to get proper handling of ranges that include letters and nonletters.
2426 For a range that includes the whole of Latin-1, this is not necessary.
2427 For other character sets, we don't bother to get this right. */
2428 if (RE_TRANSLATE_P (translate
) && start
< 04400
2429 && !(start
< 04200 && end
>= 04377))
2436 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2446 EXTEND_RANGE_TABLE (work_area
, 2);
2447 work_area
->table
[work_area
->used
++] = (start
);
2448 work_area
->table
[work_area
->used
++] = (end
);
2450 cmin
= -1, cmax
= -1;
2452 if (RE_TRANSLATE_P (translate
))
2456 for (ch
= start
; ch
<= end
; ch
++)
2458 re_wchar_t c
= TRANSLATE (ch
);
2459 if (! (start
<= c
&& c
<= end
))
2465 cmin
= MIN (cmin
, c
);
2466 cmax
= MAX (cmax
, c
);
2473 EXTEND_RANGE_TABLE (work_area
, 2);
2474 work_area
->table
[work_area
->used
++] = (cmin
);
2475 work_area
->table
[work_area
->used
++] = (cmax
);
2483 #ifndef MATCH_MAY_ALLOCATE
2485 /* If we cannot allocate large objects within re_match_2_internal,
2486 we make the fail stack and register vectors global.
2487 The fail stack, we grow to the maximum size when a regexp
2489 The register vectors, we adjust in size each time we
2490 compile a regexp, according to the number of registers it needs. */
2492 static fail_stack_type fail_stack
;
2494 /* Size with which the following vectors are currently allocated.
2495 That is so we can make them bigger as needed,
2496 but never make them smaller. */
2497 static int regs_allocated_size
;
2499 static re_char
** regstart
, ** regend
;
2500 static re_char
**best_regstart
, **best_regend
;
2502 /* Make the register vectors big enough for NUM_REGS registers,
2503 but don't make them smaller. */
2506 regex_grow_registers (num_regs
)
2509 if (num_regs
> regs_allocated_size
)
2511 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2512 RETALLOC_IF (regend
, num_regs
, re_char
*);
2513 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2514 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2516 regs_allocated_size
= num_regs
;
2520 #endif /* not MATCH_MAY_ALLOCATE */
2522 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2526 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2527 Returns one of error codes defined in `regex.h', or zero for success.
2529 Assumes the `allocated' (and perhaps `buffer') and `translate'
2530 fields are set in BUFP on entry.
2532 If it succeeds, results are put in BUFP (if it returns an error, the
2533 contents of BUFP are undefined):
2534 `buffer' is the compiled pattern;
2535 `syntax' is set to SYNTAX;
2536 `used' is set to the length of the compiled pattern;
2537 `fastmap_accurate' is zero;
2538 `re_nsub' is the number of subexpressions in PATTERN;
2539 `not_bol' and `not_eol' are zero;
2541 The `fastmap' field is neither examined nor set. */
2543 /* Insert the `jump' from the end of last alternative to "here".
2544 The space for the jump has already been allocated. */
2545 #define FIXUP_ALT_JUMP() \
2547 if (fixup_alt_jump) \
2548 STORE_JUMP (jump, fixup_alt_jump, b); \
2552 /* Return, freeing storage we allocated. */
2553 #define FREE_STACK_RETURN(value) \
2555 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2556 free (compile_stack.stack); \
2560 static reg_errcode_t
2561 regex_compile (pattern
, size
, syntax
, bufp
)
2564 reg_syntax_t syntax
;
2565 struct re_pattern_buffer
*bufp
;
2567 /* We fetch characters from PATTERN here. */
2568 register re_wchar_t c
, c1
;
2570 /* A random temporary spot in PATTERN. */
2573 /* Points to the end of the buffer, where we should append. */
2574 register unsigned char *b
;
2576 /* Keeps track of unclosed groups. */
2577 compile_stack_type compile_stack
;
2579 /* Points to the current (ending) position in the pattern. */
2581 /* `const' makes AIX compiler fail. */
2582 unsigned char *p
= pattern
;
2584 re_char
*p
= pattern
;
2586 re_char
*pend
= pattern
+ size
;
2588 /* How to translate the characters in the pattern. */
2589 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2591 /* Address of the count-byte of the most recently inserted `exactn'
2592 command. This makes it possible to tell if a new exact-match
2593 character can be added to that command or if the character requires
2594 a new `exactn' command. */
2595 unsigned char *pending_exact
= 0;
2597 /* Address of start of the most recently finished expression.
2598 This tells, e.g., postfix * where to find the start of its
2599 operand. Reset at the beginning of groups and alternatives. */
2600 unsigned char *laststart
= 0;
2602 /* Address of beginning of regexp, or inside of last group. */
2603 unsigned char *begalt
;
2605 /* Place in the uncompiled pattern (i.e., the {) to
2606 which to go back if the interval is invalid. */
2607 re_char
*beg_interval
;
2609 /* Address of the place where a forward jump should go to the end of
2610 the containing expression. Each alternative of an `or' -- except the
2611 last -- ends with a forward jump of this sort. */
2612 unsigned char *fixup_alt_jump
= 0;
2614 /* Work area for range table of charset. */
2615 struct range_table_work_area range_table_work
;
2617 /* If the object matched can contain multibyte characters. */
2618 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2620 /* If a target of matching can contain multibyte characters. */
2621 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
2623 /* Nonzero if we have pushed down into a subpattern. */
2624 int in_subpattern
= 0;
2626 /* These hold the values of p, pattern, and pend from the main
2627 pattern when we have pushed into a subpattern. */
2629 re_char
*main_pattern
;
2634 DEBUG_PRINT1 ("\nCompiling pattern: ");
2637 unsigned debug_count
;
2639 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2640 putchar (pattern
[debug_count
]);
2645 /* Initialize the compile stack. */
2646 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2647 if (compile_stack
.stack
== NULL
)
2650 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2651 compile_stack
.avail
= 0;
2653 range_table_work
.table
= 0;
2654 range_table_work
.allocated
= 0;
2656 /* Initialize the pattern buffer. */
2657 bufp
->syntax
= syntax
;
2658 bufp
->fastmap_accurate
= 0;
2659 bufp
->not_bol
= bufp
->not_eol
= 0;
2660 bufp
->used_syntax
= 0;
2662 /* Set `used' to zero, so that if we return an error, the pattern
2663 printer (for debugging) will think there's no pattern. We reset it
2667 /* Always count groups, whether or not bufp->no_sub is set. */
2670 #if !defined emacs && !defined SYNTAX_TABLE
2671 /* Initialize the syntax table. */
2672 init_syntax_once ();
2675 if (bufp
->allocated
== 0)
2678 { /* If zero allocated, but buffer is non-null, try to realloc
2679 enough space. This loses if buffer's address is bogus, but
2680 that is the user's responsibility. */
2681 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2684 { /* Caller did not allocate a buffer. Do it for them. */
2685 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2687 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2689 bufp
->allocated
= INIT_BUF_SIZE
;
2692 begalt
= b
= bufp
->buffer
;
2694 /* Loop through the uncompiled pattern until we're at the end. */
2699 /* If this is the end of an included regexp,
2700 pop back to the main regexp and try again. */
2704 pattern
= main_pattern
;
2709 /* If this is the end of the main regexp, we are done. */
2721 /* If there's no special whitespace regexp, treat
2722 spaces normally. And don't try to do this recursively. */
2723 if (!whitespace_regexp
|| in_subpattern
)
2726 /* Peek past following spaces. */
2733 /* If the spaces are followed by a repetition op,
2734 treat them normally. */
2736 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2737 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2740 /* Replace the spaces with the whitespace regexp. */
2744 main_pattern
= pattern
;
2745 p
= pattern
= whitespace_regexp
;
2746 pend
= p
+ strlen (p
);
2752 if ( /* If at start of pattern, it's an operator. */
2754 /* If context independent, it's an operator. */
2755 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2756 /* Otherwise, depends on what's come before. */
2757 || at_begline_loc_p (pattern
, p
, syntax
))
2758 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2767 if ( /* If at end of pattern, it's an operator. */
2769 /* If context independent, it's an operator. */
2770 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2771 /* Otherwise, depends on what's next. */
2772 || at_endline_loc_p (p
, pend
, syntax
))
2773 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2782 if ((syntax
& RE_BK_PLUS_QM
)
2783 || (syntax
& RE_LIMITED_OPS
))
2787 /* If there is no previous pattern... */
2790 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2791 FREE_STACK_RETURN (REG_BADRPT
);
2792 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2797 /* 1 means zero (many) matches is allowed. */
2798 boolean zero_times_ok
= 0, many_times_ok
= 0;
2801 /* If there is a sequence of repetition chars, collapse it
2802 down to just one (the right one). We can't combine
2803 interval operators with these because of, e.g., `a{2}*',
2804 which should only match an even number of `a's. */
2808 if ((syntax
& RE_FRUGAL
)
2809 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2813 zero_times_ok
|= c
!= '+';
2814 many_times_ok
|= c
!= '?';
2820 || (!(syntax
& RE_BK_PLUS_QM
)
2821 && (*p
== '+' || *p
== '?')))
2823 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2826 FREE_STACK_RETURN (REG_EESCAPE
);
2827 if (p
[1] == '+' || p
[1] == '?')
2828 PATFETCH (c
); /* Gobble up the backslash. */
2834 /* If we get here, we found another repeat character. */
2838 /* Star, etc. applied to an empty pattern is equivalent
2839 to an empty pattern. */
2840 if (!laststart
|| laststart
== b
)
2843 /* Now we know whether or not zero matches is allowed
2844 and also whether or not two or more matches is allowed. */
2849 boolean simple
= skip_one_char (laststart
) == b
;
2850 unsigned int startoffset
= 0;
2852 /* Check if the loop can match the empty string. */
2853 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2854 ? on_failure_jump
: on_failure_jump_loop
;
2855 assert (skip_one_char (laststart
) <= b
);
2857 if (!zero_times_ok
&& simple
)
2858 { /* Since simple * loops can be made faster by using
2859 on_failure_keep_string_jump, we turn simple P+
2860 into PP* if P is simple. */
2861 unsigned char *p1
, *p2
;
2862 startoffset
= b
- laststart
;
2863 GET_BUFFER_SPACE (startoffset
);
2864 p1
= b
; p2
= laststart
;
2870 GET_BUFFER_SPACE (6);
2873 STORE_JUMP (ofj
, b
, b
+ 6);
2875 /* Simple * loops can use on_failure_keep_string_jump
2876 depending on what follows. But since we don't know
2877 that yet, we leave the decision up to
2878 on_failure_jump_smart. */
2879 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2880 laststart
+ startoffset
, b
+ 6);
2882 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2887 /* A simple ? pattern. */
2888 assert (zero_times_ok
);
2889 GET_BUFFER_SPACE (3);
2890 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2894 else /* not greedy */
2895 { /* I wish the greedy and non-greedy cases could be merged. */
2897 GET_BUFFER_SPACE (7); /* We might use less. */
2900 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2902 /* The non-greedy multiple match looks like
2903 a repeat..until: we only need a conditional jump
2904 at the end of the loop. */
2905 if (emptyp
) BUF_PUSH (no_op
);
2906 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2907 : on_failure_jump
, b
, laststart
);
2911 /* The repeat...until naturally matches one or more.
2912 To also match zero times, we need to first jump to
2913 the end of the loop (its conditional jump). */
2914 INSERT_JUMP (jump
, laststart
, b
);
2920 /* non-greedy a?? */
2921 INSERT_JUMP (jump
, laststart
, b
+ 3);
2923 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2940 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2942 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2944 /* Ensure that we have enough space to push a charset: the
2945 opcode, the length count, and the bitset; 34 bytes in all. */
2946 GET_BUFFER_SPACE (34);
2950 /* We test `*p == '^' twice, instead of using an if
2951 statement, so we only need one BUF_PUSH. */
2952 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2956 /* Remember the first position in the bracket expression. */
2959 /* Push the number of bytes in the bitmap. */
2960 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2962 /* Clear the whole map. */
2963 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2965 /* charset_not matches newline according to a syntax bit. */
2966 if ((re_opcode_t
) b
[-2] == charset_not
2967 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2968 SET_LIST_BIT ('\n');
2970 /* Read in characters and ranges, setting map bits. */
2973 boolean escaped_char
= false;
2974 const unsigned char *p2
= p
;
2977 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2979 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2980 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2981 So the translation is done later in a loop. Example:
2982 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2985 /* \ might escape characters inside [...] and [^...]. */
2986 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2988 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2991 escaped_char
= true;
2995 /* Could be the end of the bracket expression. If it's
2996 not (i.e., when the bracket expression is `[]' so
2997 far), the ']' character bit gets set way below. */
2998 if (c
== ']' && p2
!= p1
)
3002 /* See if we're at the beginning of a possible character
3005 if (!escaped_char
&&
3006 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
3008 /* Leave room for the null. */
3009 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
3010 const unsigned char *class_beg
;
3016 /* If pattern is `[[:'. */
3017 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3022 if ((c
== ':' && *p
== ']') || p
== pend
)
3024 if (c1
< CHAR_CLASS_MAX_LENGTH
)
3027 /* This is in any case an invalid class name. */
3032 /* If isn't a word bracketed by `[:' and `:]':
3033 undo the ending character, the letters, and
3034 leave the leading `:' and `[' (but set bits for
3036 if (c
== ':' && *p
== ']')
3041 cc
= re_wctype (str
);
3044 FREE_STACK_RETURN (REG_ECTYPE
);
3046 /* Throw away the ] at the end of the character
3050 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3053 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
3054 if (re_iswctype (btowc (ch
), cc
))
3057 if (c
< (1 << BYTEWIDTH
))
3061 /* Most character classes in a multibyte match
3062 just set a flag. Exceptions are is_blank,
3063 is_digit, is_cntrl, and is_xdigit, since
3064 they can only match ASCII characters. We
3065 don't need to handle them for multibyte.
3066 They are distinguished by a negative wctype. */
3068 /* Setup the gl_state object to its buffer-defined
3069 value. This hardcodes the buffer-global
3070 syntax-table for ASCII chars, while the other chars
3071 will obey syntax-table properties. It's not ideal,
3072 but it's the way it's been done until now. */
3073 SETUP_SYNTAX_TABLE (BEGV
, 0);
3075 for (ch
= 0; ch
< 256; ++ch
)
3077 c
= RE_CHAR_TO_MULTIBYTE (ch
);
3078 if (! CHAR_BYTE8_P (c
)
3079 && re_iswctype (c
, cc
))
3085 if (ASCII_CHAR_P (c1
))
3087 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
3091 SET_RANGE_TABLE_WORK_AREA_BIT
3092 (range_table_work
, re_wctype_to_bit (cc
));
3094 /* In most cases the matching rule for char classes
3095 only uses the syntax table for multibyte chars,
3096 so that the content of the syntax-table it is not
3097 hardcoded in the range_table. SPACE and WORD are
3098 the two exceptions. */
3099 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
3100 bufp
->used_syntax
= 1;
3102 /* Repeat the loop. */
3107 /* Go back to right after the "[:". */
3111 /* Because the `:' may starts the range, we
3112 can't simply set bit and repeat the loop.
3113 Instead, just set it to C and handle below. */
3118 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
3121 /* Discard the `-'. */
3124 /* Fetch the character which ends the range. */
3127 if (CHAR_BYTE8_P (c1
)
3128 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
3129 /* Treat the range from a multibyte character to
3130 raw-byte character as empty. */
3135 /* Range from C to C. */
3140 if (syntax
& RE_NO_EMPTY_RANGES
)
3141 FREE_STACK_RETURN (REG_ERANGEX
);
3142 /* Else, repeat the loop. */
3147 /* Set the range into bitmap */
3148 for (; c
<= c1
; c
++)
3151 if (ch
< (1 << BYTEWIDTH
))
3158 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
3160 if (CHAR_BYTE8_P (c1
))
3161 c
= BYTE8_TO_CHAR (128);
3165 if (CHAR_BYTE8_P (c
))
3167 c
= CHAR_TO_BYTE8 (c
);
3168 c1
= CHAR_TO_BYTE8 (c1
);
3169 for (; c
<= c1
; c
++)
3174 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
3178 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
3185 /* Discard any (non)matching list bytes that are all 0 at the
3186 end of the map. Decrease the map-length byte too. */
3187 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3191 /* Build real range table from work area. */
3192 if (RANGE_TABLE_WORK_USED (range_table_work
)
3193 || RANGE_TABLE_WORK_BITS (range_table_work
))
3196 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3198 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3199 bytes for flags, two for COUNT, and three bytes for
3201 GET_BUFFER_SPACE (4 + used
* 3);
3203 /* Indicate the existence of range table. */
3204 laststart
[1] |= 0x80;
3206 /* Store the character class flag bits into the range table.
3207 If not in emacs, these flag bits are always 0. */
3208 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3209 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3211 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3212 for (i
= 0; i
< used
; i
++)
3213 STORE_CHARACTER_AND_INCR
3214 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3221 if (syntax
& RE_NO_BK_PARENS
)
3228 if (syntax
& RE_NO_BK_PARENS
)
3235 if (syntax
& RE_NEWLINE_ALT
)
3242 if (syntax
& RE_NO_BK_VBAR
)
3249 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3250 goto handle_interval
;
3256 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3258 /* Do not translate the character after the \, so that we can
3259 distinguish, e.g., \B from \b, even if we normally would
3260 translate, e.g., B to b. */
3266 if (syntax
& RE_NO_BK_PARENS
)
3267 goto normal_backslash
;
3272 regnum_t regnum
= 0;
3275 /* Look for a special (?...) construct */
3276 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3278 PATFETCH (c
); /* Gobble up the '?'. */
3284 case ':': shy
= 1; break;
3286 /* An explicitly specified regnum must start
3289 FREE_STACK_RETURN (REG_BADPAT
);
3290 case '1': case '2': case '3': case '4':
3291 case '5': case '6': case '7': case '8': case '9':
3292 regnum
= 10*regnum
+ (c
- '0'); break;
3294 /* Only (?:...) is supported right now. */
3295 FREE_STACK_RETURN (REG_BADPAT
);
3302 regnum
= ++bufp
->re_nsub
;
3304 { /* It's actually not shy, but explicitly numbered. */
3306 if (regnum
> bufp
->re_nsub
)
3307 bufp
->re_nsub
= regnum
;
3308 else if (regnum
> bufp
->re_nsub
3309 /* Ideally, we'd want to check that the specified
3310 group can't have matched (i.e. all subgroups
3311 using the same regnum are in other branches of
3312 OR patterns), but we don't currently keep track
3313 of enough info to do that easily. */
3314 || group_in_compile_stack (compile_stack
, regnum
))
3315 FREE_STACK_RETURN (REG_BADPAT
);
3318 /* It's really shy. */
3319 regnum
= - bufp
->re_nsub
;
3321 if (COMPILE_STACK_FULL
)
3323 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3324 compile_stack_elt_t
);
3325 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3327 compile_stack
.size
<<= 1;
3330 /* These are the values to restore when we hit end of this
3331 group. They are all relative offsets, so that if the
3332 whole pattern moves because of realloc, they will still
3334 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3335 COMPILE_STACK_TOP
.fixup_alt_jump
3336 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3337 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3338 COMPILE_STACK_TOP
.regnum
= regnum
;
3340 /* Do not push a start_memory for groups beyond the last one
3341 we can represent in the compiled pattern. */
3342 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3343 BUF_PUSH_2 (start_memory
, regnum
);
3345 compile_stack
.avail
++;
3350 /* If we've reached MAX_REGNUM groups, then this open
3351 won't actually generate any code, so we'll have to
3352 clear pending_exact explicitly. */
3358 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3360 if (COMPILE_STACK_EMPTY
)
3362 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3363 goto normal_backslash
;
3365 FREE_STACK_RETURN (REG_ERPAREN
);
3371 /* See similar code for backslashed left paren above. */
3372 if (COMPILE_STACK_EMPTY
)
3374 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3377 FREE_STACK_RETURN (REG_ERPAREN
);
3380 /* Since we just checked for an empty stack above, this
3381 ``can't happen''. */
3382 assert (compile_stack
.avail
!= 0);
3384 /* We don't just want to restore into `regnum', because
3385 later groups should continue to be numbered higher,
3386 as in `(ab)c(de)' -- the second group is #2. */
3389 compile_stack
.avail
--;
3390 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3392 = COMPILE_STACK_TOP
.fixup_alt_jump
3393 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3395 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3396 regnum
= COMPILE_STACK_TOP
.regnum
;
3397 /* If we've reached MAX_REGNUM groups, then this open
3398 won't actually generate any code, so we'll have to
3399 clear pending_exact explicitly. */
3402 /* We're at the end of the group, so now we know how many
3403 groups were inside this one. */
3404 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3405 BUF_PUSH_2 (stop_memory
, regnum
);
3410 case '|': /* `\|'. */
3411 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3412 goto normal_backslash
;
3414 if (syntax
& RE_LIMITED_OPS
)
3417 /* Insert before the previous alternative a jump which
3418 jumps to this alternative if the former fails. */
3419 GET_BUFFER_SPACE (3);
3420 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3424 /* The alternative before this one has a jump after it
3425 which gets executed if it gets matched. Adjust that
3426 jump so it will jump to this alternative's analogous
3427 jump (put in below, which in turn will jump to the next
3428 (if any) alternative's such jump, etc.). The last such
3429 jump jumps to the correct final destination. A picture:
3435 If we are at `b', then fixup_alt_jump right now points to a
3436 three-byte space after `a'. We'll put in the jump, set
3437 fixup_alt_jump to right after `b', and leave behind three
3438 bytes which we'll fill in when we get to after `c'. */
3442 /* Mark and leave space for a jump after this alternative,
3443 to be filled in later either by next alternative or
3444 when know we're at the end of a series of alternatives. */
3446 GET_BUFFER_SPACE (3);
3455 /* If \{ is a literal. */
3456 if (!(syntax
& RE_INTERVALS
)
3457 /* If we're at `\{' and it's not the open-interval
3459 || (syntax
& RE_NO_BK_BRACES
))
3460 goto normal_backslash
;
3464 /* If got here, then the syntax allows intervals. */
3466 /* At least (most) this many matches must be made. */
3467 int lower_bound
= 0, upper_bound
= -1;
3471 GET_UNSIGNED_NUMBER (lower_bound
);
3474 GET_UNSIGNED_NUMBER (upper_bound
);
3476 /* Interval such as `{1}' => match exactly once. */
3477 upper_bound
= lower_bound
;
3479 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3480 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3481 FREE_STACK_RETURN (REG_BADBR
);
3483 if (!(syntax
& RE_NO_BK_BRACES
))
3486 FREE_STACK_RETURN (REG_BADBR
);
3488 FREE_STACK_RETURN (REG_EESCAPE
);
3493 FREE_STACK_RETURN (REG_BADBR
);
3495 /* We just parsed a valid interval. */
3497 /* If it's invalid to have no preceding re. */
3500 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3501 FREE_STACK_RETURN (REG_BADRPT
);
3502 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3505 goto unfetch_interval
;
3508 if (upper_bound
== 0)
3509 /* If the upper bound is zero, just drop the sub pattern
3512 else if (lower_bound
== 1 && upper_bound
== 1)
3513 /* Just match it once: nothing to do here. */
3516 /* Otherwise, we have a nontrivial interval. When
3517 we're all done, the pattern will look like:
3518 set_number_at <jump count> <upper bound>
3519 set_number_at <succeed_n count> <lower bound>
3520 succeed_n <after jump addr> <succeed_n count>
3522 jump_n <succeed_n addr> <jump count>
3523 (The upper bound and `jump_n' are omitted if
3524 `upper_bound' is 1, though.) */
3526 { /* If the upper bound is > 1, we need to insert
3527 more at the end of the loop. */
3528 unsigned int nbytes
= (upper_bound
< 0 ? 3
3529 : upper_bound
> 1 ? 5 : 0);
3530 unsigned int startoffset
= 0;
3532 GET_BUFFER_SPACE (20); /* We might use less. */
3534 if (lower_bound
== 0)
3536 /* A succeed_n that starts with 0 is really a
3537 a simple on_failure_jump_loop. */
3538 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3544 /* Initialize lower bound of the `succeed_n', even
3545 though it will be set during matching by its
3546 attendant `set_number_at' (inserted next),
3547 because `re_compile_fastmap' needs to know.
3548 Jump to the `jump_n' we might insert below. */
3549 INSERT_JUMP2 (succeed_n
, laststart
,
3554 /* Code to initialize the lower bound. Insert
3555 before the `succeed_n'. The `5' is the last two
3556 bytes of this `set_number_at', plus 3 bytes of
3557 the following `succeed_n'. */
3558 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3563 if (upper_bound
< 0)
3565 /* A negative upper bound stands for infinity,
3566 in which case it degenerates to a plain jump. */
3567 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3570 else if (upper_bound
> 1)
3571 { /* More than one repetition is allowed, so
3572 append a backward jump to the `succeed_n'
3573 that starts this interval.
3575 When we've reached this during matching,
3576 we'll have matched the interval once, so
3577 jump back only `upper_bound - 1' times. */
3578 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3582 /* The location we want to set is the second
3583 parameter of the `jump_n'; that is `b-2' as
3584 an absolute address. `laststart' will be
3585 the `set_number_at' we're about to insert;
3586 `laststart+3' the number to set, the source
3587 for the relative address. But we are
3588 inserting into the middle of the pattern --
3589 so everything is getting moved up by 5.
3590 Conclusion: (b - 2) - (laststart + 3) + 5,
3591 i.e., b - laststart.
3593 We insert this at the beginning of the loop
3594 so that if we fail during matching, we'll
3595 reinitialize the bounds. */
3596 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3597 upper_bound
- 1, b
);
3602 beg_interval
= NULL
;
3607 /* If an invalid interval, match the characters as literals. */
3608 assert (beg_interval
);
3610 beg_interval
= NULL
;
3612 /* normal_char and normal_backslash need `c'. */
3615 if (!(syntax
& RE_NO_BK_BRACES
))
3617 assert (p
> pattern
&& p
[-1] == '\\');
3618 goto normal_backslash
;
3624 /* There is no way to specify the before_dot and after_dot
3625 operators. rms says this is ok. --karl */
3633 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3639 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3645 BUF_PUSH_2 (categoryspec
, c
);
3651 BUF_PUSH_2 (notcategoryspec
, c
);
3657 if (syntax
& RE_NO_GNU_OPS
)
3660 BUF_PUSH_2 (syntaxspec
, Sword
);
3665 if (syntax
& RE_NO_GNU_OPS
)
3668 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3673 if (syntax
& RE_NO_GNU_OPS
)
3679 if (syntax
& RE_NO_GNU_OPS
)
3685 if (syntax
& RE_NO_GNU_OPS
)
3694 FREE_STACK_RETURN (REG_BADPAT
);
3698 if (syntax
& RE_NO_GNU_OPS
)
3700 BUF_PUSH (wordbound
);
3704 if (syntax
& RE_NO_GNU_OPS
)
3706 BUF_PUSH (notwordbound
);
3710 if (syntax
& RE_NO_GNU_OPS
)
3716 if (syntax
& RE_NO_GNU_OPS
)
3721 case '1': case '2': case '3': case '4': case '5':
3722 case '6': case '7': case '8': case '9':
3726 if (syntax
& RE_NO_BK_REFS
)
3727 goto normal_backslash
;
3731 if (reg
> bufp
->re_nsub
|| reg
< 1
3732 /* Can't back reference to a subexp before its end. */
3733 || group_in_compile_stack (compile_stack
, reg
))
3734 FREE_STACK_RETURN (REG_ESUBREG
);
3737 BUF_PUSH_2 (duplicate
, reg
);
3744 if (syntax
& RE_BK_PLUS_QM
)
3747 goto normal_backslash
;
3751 /* You might think it would be useful for \ to mean
3752 not to translate; but if we don't translate it
3753 it will never match anything. */
3760 /* Expects the character in `c'. */
3762 /* If no exactn currently being built. */
3765 /* If last exactn not at current position. */
3766 || pending_exact
+ *pending_exact
+ 1 != b
3768 /* We have only one byte following the exactn for the count. */
3769 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3771 /* If followed by a repetition operator. */
3772 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3773 || ((syntax
& RE_BK_PLUS_QM
)
3774 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3775 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3776 || ((syntax
& RE_INTERVALS
)
3777 && ((syntax
& RE_NO_BK_BRACES
)
3778 ? p
!= pend
&& *p
== '{'
3779 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3781 /* Start building a new exactn. */
3785 BUF_PUSH_2 (exactn
, 0);
3786 pending_exact
= b
- 1;
3789 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3796 len
= CHAR_STRING (c
, b
);
3801 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3802 if (! CHAR_BYTE8_P (c1
))
3804 re_wchar_t c2
= TRANSLATE (c1
);
3806 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3812 (*pending_exact
) += len
;
3817 } /* while p != pend */
3820 /* Through the pattern now. */
3824 if (!COMPILE_STACK_EMPTY
)
3825 FREE_STACK_RETURN (REG_EPAREN
);
3827 /* If we don't want backtracking, force success
3828 the first time we reach the end of the compiled pattern. */
3829 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3832 /* We have succeeded; set the length of the buffer. */
3833 bufp
->used
= b
- bufp
->buffer
;
3838 re_compile_fastmap (bufp
);
3839 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3840 print_compiled_pattern (bufp
);
3845 #ifndef MATCH_MAY_ALLOCATE
3846 /* Initialize the failure stack to the largest possible stack. This
3847 isn't necessary unless we're trying to avoid calling alloca in
3848 the search and match routines. */
3850 int num_regs
= bufp
->re_nsub
+ 1;
3852 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3854 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3856 if (! fail_stack
.stack
)
3858 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3859 * sizeof (fail_stack_elt_t
));
3862 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3864 * sizeof (fail_stack_elt_t
)));
3867 regex_grow_registers (num_regs
);
3869 #endif /* not MATCH_MAY_ALLOCATE */
3871 FREE_STACK_RETURN (REG_NOERROR
);
3872 } /* regex_compile */
3874 /* Subroutines for `regex_compile'. */
3876 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3879 store_op1 (op
, loc
, arg
)
3884 *loc
= (unsigned char) op
;
3885 STORE_NUMBER (loc
+ 1, arg
);
3889 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3892 store_op2 (op
, loc
, arg1
, arg2
)
3897 *loc
= (unsigned char) op
;
3898 STORE_NUMBER (loc
+ 1, arg1
);
3899 STORE_NUMBER (loc
+ 3, arg2
);
3903 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3904 for OP followed by two-byte integer parameter ARG. */
3907 insert_op1 (op
, loc
, arg
, end
)
3913 register unsigned char *pfrom
= end
;
3914 register unsigned char *pto
= end
+ 3;
3916 while (pfrom
!= loc
)
3919 store_op1 (op
, loc
, arg
);
3923 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3926 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3932 register unsigned char *pfrom
= end
;
3933 register unsigned char *pto
= end
+ 5;
3935 while (pfrom
!= loc
)
3938 store_op2 (op
, loc
, arg1
, arg2
);
3942 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3943 after an alternative or a begin-subexpression. We assume there is at
3944 least one character before the ^. */
3947 at_begline_loc_p (pattern
, p
, syntax
)
3948 re_char
*pattern
, *p
;
3949 reg_syntax_t syntax
;
3951 re_char
*prev
= p
- 2;
3952 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3955 /* After a subexpression? */
3956 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3957 /* After an alternative? */
3958 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3959 /* After a shy subexpression? */
3960 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3961 && prev
[-1] == '?' && prev
[-2] == '('
3962 && (syntax
& RE_NO_BK_PARENS
3963 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3967 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3968 at least one character after the $, i.e., `P < PEND'. */
3971 at_endline_loc_p (p
, pend
, syntax
)
3973 reg_syntax_t syntax
;
3976 boolean next_backslash
= *next
== '\\';
3977 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3980 /* Before a subexpression? */
3981 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3982 : next_backslash
&& next_next
&& *next_next
== ')')
3983 /* Before an alternative? */
3984 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3985 : next_backslash
&& next_next
&& *next_next
== '|');
3989 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3990 false if it's not. */
3993 group_in_compile_stack (compile_stack
, regnum
)
3994 compile_stack_type compile_stack
;
3999 for (this_element
= compile_stack
.avail
- 1;
4002 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
4009 If fastmap is non-NULL, go through the pattern and fill fastmap
4010 with all the possible leading chars. If fastmap is NULL, don't
4011 bother filling it up (obviously) and only return whether the
4012 pattern could potentially match the empty string.
4014 Return 1 if p..pend might match the empty string.
4015 Return 0 if p..pend matches at least one char.
4016 Return -1 if fastmap was not updated accurately. */
4019 analyse_first (p
, pend
, fastmap
, multibyte
)
4022 const int multibyte
;
4027 /* If all elements for base leading-codes in fastmap is set, this
4028 flag is set true. */
4029 boolean match_any_multibyte_characters
= false;
4033 /* The loop below works as follows:
4034 - It has a working-list kept in the PATTERN_STACK and which basically
4035 starts by only containing a pointer to the first operation.
4036 - If the opcode we're looking at is a match against some set of
4037 chars, then we add those chars to the fastmap and go on to the
4038 next work element from the worklist (done via `break').
4039 - If the opcode is a control operator on the other hand, we either
4040 ignore it (if it's meaningless at this point, such as `start_memory')
4041 or execute it (if it's a jump). If the jump has several destinations
4042 (i.e. `on_failure_jump'), then we push the other destination onto the
4044 We guarantee termination by ignoring backward jumps (more or less),
4045 so that `p' is monotonically increasing. More to the point, we
4046 never set `p' (or push) anything `<= p1'. */
4050 /* `p1' is used as a marker of how far back a `on_failure_jump'
4051 can go without being ignored. It is normally equal to `p'
4052 (which prevents any backward `on_failure_jump') except right
4053 after a plain `jump', to allow patterns such as:
4056 10: on_failure_jump 3
4057 as used for the *? operator. */
4060 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4067 /* If the first character has to match a backreference, that means
4068 that the group was empty (since it already matched). Since this
4069 is the only case that interests us here, we can assume that the
4070 backreference must match the empty string. */
4075 /* Following are the cases which match a character. These end
4081 /* If multibyte is nonzero, the first byte of each
4082 character is an ASCII or a leading code. Otherwise,
4083 each byte is a character. Thus, this works in both
4088 /* For the case of matching this unibyte regex
4089 against multibyte, we must set a leading code of
4090 the corresponding multibyte character. */
4091 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
4093 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
4100 /* We could put all the chars except for \n (and maybe \0)
4101 but we don't bother since it is generally not worth it. */
4102 if (!fastmap
) break;
4107 if (!fastmap
) break;
4109 /* Chars beyond end of bitmap are possible matches. */
4110 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
4111 j
< (1 << BYTEWIDTH
); j
++)
4117 if (!fastmap
) break;
4118 not = (re_opcode_t
) *(p
- 1) == charset_not
;
4119 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
4121 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
4125 if (/* Any leading code can possibly start a character
4126 which doesn't match the specified set of characters. */
4129 /* If we can match a character class, we can match any
4130 multibyte characters. */
4131 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
4132 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
4135 if (match_any_multibyte_characters
== false)
4137 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4138 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4140 match_any_multibyte_characters
= true;
4144 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
4145 && match_any_multibyte_characters
== false)
4147 /* Set fastmap[I] to 1 where I is a leading code of each
4148 multibyte characer in the range table. */
4150 unsigned char lc1
, lc2
;
4152 /* Make P points the range table. `+ 2' is to skip flag
4153 bits for a character class. */
4154 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
4156 /* Extract the number of ranges in range table into COUNT. */
4157 EXTRACT_NUMBER_AND_INCR (count
, p
);
4158 for (; count
> 0; count
--, p
+= 3)
4160 /* Extract the start and end of each range. */
4161 EXTRACT_CHARACTER (c
, p
);
4162 lc1
= CHAR_LEADING_CODE (c
);
4164 EXTRACT_CHARACTER (c
, p
);
4165 lc2
= CHAR_LEADING_CODE (c
);
4166 for (j
= lc1
; j
<= lc2
; j
++)
4175 if (!fastmap
) break;
4177 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4179 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4180 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4184 /* This match depends on text properties. These end with
4185 aborting optimizations. */
4189 case notcategoryspec
:
4190 if (!fastmap
) break;
4191 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4193 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
4194 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4197 /* Any leading code can possibly start a character which
4198 has or doesn't has the specified category. */
4199 if (match_any_multibyte_characters
== false)
4201 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4202 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4204 match_any_multibyte_characters
= true;
4208 /* All cases after this match the empty string. These end with
4230 EXTRACT_NUMBER_AND_INCR (j
, p
);
4232 /* Backward jumps can only go back to code that we've already
4233 visited. `re_compile' should make sure this is true. */
4236 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4238 case on_failure_jump
:
4239 case on_failure_keep_string_jump
:
4240 case on_failure_jump_loop
:
4241 case on_failure_jump_nastyloop
:
4242 case on_failure_jump_smart
:
4248 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4249 to jump back to "just after here". */
4252 case on_failure_jump
:
4253 case on_failure_keep_string_jump
:
4254 case on_failure_jump_nastyloop
:
4255 case on_failure_jump_loop
:
4256 case on_failure_jump_smart
:
4257 EXTRACT_NUMBER_AND_INCR (j
, p
);
4259 ; /* Backward jump to be ignored. */
4261 { /* We have to look down both arms.
4262 We first go down the "straight" path so as to minimize
4263 stack usage when going through alternatives. */
4264 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4272 /* This code simply does not properly handle forward jump_n. */
4273 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4275 /* jump_n can either jump or fall through. The (backward) jump
4276 case has already been handled, so we only need to look at the
4277 fallthrough case. */
4281 /* If N == 0, it should be an on_failure_jump_loop instead. */
4282 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4284 /* We only care about one iteration of the loop, so we don't
4285 need to consider the case where this behaves like an
4302 abort (); /* We have listed all the cases. */
4305 /* Getting here means we have found the possible starting
4306 characters for one path of the pattern -- and that the empty
4307 string does not match. We need not follow this path further. */
4311 /* We reached the end without matching anything. */
4314 } /* analyse_first */
4316 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4317 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4318 characters can start a string that matches the pattern. This fastmap
4319 is used by re_search to skip quickly over impossible starting points.
4321 Character codes above (1 << BYTEWIDTH) are not represented in the
4322 fastmap, but the leading codes are represented. Thus, the fastmap
4323 indicates which character sets could start a match.
4325 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4326 area as BUFP->fastmap.
4328 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4331 Returns 0 if we succeed, -2 if an internal error. */
4334 re_compile_fastmap (bufp
)
4335 struct re_pattern_buffer
*bufp
;
4337 char *fastmap
= bufp
->fastmap
;
4340 assert (fastmap
&& bufp
->buffer
);
4342 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4343 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4345 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4346 fastmap
, RE_MULTIBYTE_P (bufp
));
4347 bufp
->can_be_null
= (analysis
!= 0);
4349 } /* re_compile_fastmap */
4351 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4352 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4353 this memory for recording register information. STARTS and ENDS
4354 must be allocated using the malloc library routine, and must each
4355 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4357 If NUM_REGS == 0, then subsequent matches should allocate their own
4360 Unless this function is called, the first search or match using
4361 PATTERN_BUFFER will allocate its own register data, without
4362 freeing the old data. */
4365 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4366 struct re_pattern_buffer
*bufp
;
4367 struct re_registers
*regs
;
4369 regoff_t
*starts
, *ends
;
4373 bufp
->regs_allocated
= REGS_REALLOCATE
;
4374 regs
->num_regs
= num_regs
;
4375 regs
->start
= starts
;
4380 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4382 regs
->start
= regs
->end
= (regoff_t
*) 0;
4385 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4387 /* Searching routines. */
4389 /* Like re_search_2, below, but only one string is specified, and
4390 doesn't let you say where to stop matching. */
4393 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4394 struct re_pattern_buffer
*bufp
;
4396 int size
, startpos
, range
;
4397 struct re_registers
*regs
;
4399 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4402 WEAK_ALIAS (__re_search
, re_search
)
4404 /* Head address of virtual concatenation of string. */
4405 #define HEAD_ADDR_VSTRING(P) \
4406 (((P) >= size1 ? string2 : string1))
4408 /* End address of virtual concatenation of string. */
4409 #define STOP_ADDR_VSTRING(P) \
4410 (((P) >= size1 ? string2 + size2 : string1 + size1))
4412 /* Address of POS in the concatenation of virtual string. */
4413 #define POS_ADDR_VSTRING(POS) \
4414 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4416 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4417 virtual concatenation of STRING1 and STRING2, starting first at index
4418 STARTPOS, then at STARTPOS + 1, and so on.
4420 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4422 RANGE is how far to scan while trying to match. RANGE = 0 means try
4423 only at STARTPOS; in general, the last start tried is STARTPOS +
4426 In REGS, return the indices of the virtual concatenation of STRING1
4427 and STRING2 that matched the entire BUFP->buffer and its contained
4430 Do not consider matching one past the index STOP in the virtual
4431 concatenation of STRING1 and STRING2.
4433 We return either the position in the strings at which the match was
4434 found, -1 if no match, or -2 if error (such as failure
4438 re_search_2 (bufp
, str1
, size1
, str2
, size2
, startpos
, range
, regs
, stop
)
4439 struct re_pattern_buffer
*bufp
;
4440 const char *str1
, *str2
;
4444 struct re_registers
*regs
;
4448 re_char
*string1
= (re_char
*) str1
;
4449 re_char
*string2
= (re_char
*) str2
;
4450 register char *fastmap
= bufp
->fastmap
;
4451 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4452 int total_size
= size1
+ size2
;
4453 int endpos
= startpos
+ range
;
4454 boolean anchored_start
;
4455 /* Nonzero if we are searching multibyte string. */
4456 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4458 /* Check for out-of-range STARTPOS. */
4459 if (startpos
< 0 || startpos
> total_size
)
4462 /* Fix up RANGE if it might eventually take us outside
4463 the virtual concatenation of STRING1 and STRING2.
4464 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4466 range
= 0 - startpos
;
4467 else if (endpos
> total_size
)
4468 range
= total_size
- startpos
;
4470 /* If the search isn't to be a backwards one, don't waste time in a
4471 search for a pattern anchored at beginning of buffer. */
4472 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4481 /* In a forward search for something that starts with \=.
4482 don't keep searching past point. */
4483 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4485 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4491 /* Update the fastmap now if not correct already. */
4492 if (fastmap
&& !bufp
->fastmap_accurate
)
4493 re_compile_fastmap (bufp
);
4495 /* See whether the pattern is anchored. */
4496 anchored_start
= (bufp
->buffer
[0] == begline
);
4499 gl_state
.object
= re_match_object
;
4501 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4503 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4507 /* Loop through the string, looking for a place to start matching. */
4510 /* If the pattern is anchored,
4511 skip quickly past places we cannot match.
4512 We don't bother to treat startpos == 0 specially
4513 because that case doesn't repeat. */
4514 if (anchored_start
&& startpos
> 0)
4516 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4517 : string2
[startpos
- size1
- 1])
4522 /* If a fastmap is supplied, skip quickly over characters that
4523 cannot be the start of a match. If the pattern can match the
4524 null string, however, we don't need to skip characters; we want
4525 the first null string. */
4526 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4528 register re_char
*d
;
4529 register re_wchar_t buf_ch
;
4531 d
= POS_ADDR_VSTRING (startpos
);
4533 if (range
> 0) /* Searching forwards. */
4535 register int lim
= 0;
4538 if (startpos
< size1
&& startpos
+ range
>= size1
)
4539 lim
= range
- (size1
- startpos
);
4541 /* Written out as an if-else to avoid testing `translate'
4543 if (RE_TRANSLATE_P (translate
))
4550 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4551 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4552 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4555 range
-= buf_charlen
;
4561 register re_wchar_t ch
, translated
;
4564 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4565 translated
= RE_TRANSLATE (translate
, ch
);
4566 if (translated
!= ch
4567 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4569 if (fastmap
[buf_ch
])
4582 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4583 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4585 range
-= buf_charlen
;
4589 while (range
> lim
&& !fastmap
[*d
])
4595 startpos
+= irange
- range
;
4597 else /* Searching backwards. */
4601 buf_ch
= STRING_CHAR (d
);
4602 buf_ch
= TRANSLATE (buf_ch
);
4603 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4608 register re_wchar_t ch
, translated
;
4611 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4612 translated
= TRANSLATE (ch
);
4613 if (translated
!= ch
4614 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4616 if (! fastmap
[TRANSLATE (buf_ch
)])
4622 /* If can't match the null string, and that's all we have left, fail. */
4623 if (range
>= 0 && startpos
== total_size
&& fastmap
4624 && !bufp
->can_be_null
)
4627 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4628 startpos
, regs
, stop
);
4641 /* Update STARTPOS to the next character boundary. */
4644 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4645 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4646 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
4664 /* Update STARTPOS to the previous character boundary. */
4667 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4669 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4671 /* Find the head of multibyte form. */
4672 PREV_CHAR_BOUNDARY (p
, phead
);
4673 range
+= p0
- 1 - p
;
4677 startpos
-= p0
- 1 - p
;
4683 WEAK_ALIAS (__re_search_2
, re_search_2
)
4685 /* Declarations and macros for re_match_2. */
4687 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4689 RE_TRANSLATE_TYPE translate
,
4690 const int multibyte
));
4692 /* This converts PTR, a pointer into one of the search strings `string1'
4693 and `string2' into an offset from the beginning of that string. */
4694 #define POINTER_TO_OFFSET(ptr) \
4695 (FIRST_STRING_P (ptr) \
4696 ? ((regoff_t) ((ptr) - string1)) \
4697 : ((regoff_t) ((ptr) - string2 + size1)))
4699 /* Call before fetching a character with *d. This switches over to
4700 string2 if necessary.
4701 Check re_match_2_internal for a discussion of why end_match_2 might
4702 not be within string2 (but be equal to end_match_1 instead). */
4703 #define PREFETCH() \
4706 /* End of string2 => fail. */ \
4707 if (dend == end_match_2) \
4709 /* End of string1 => advance to string2. */ \
4711 dend = end_match_2; \
4714 /* Call before fetching a char with *d if you already checked other limits.
4715 This is meant for use in lookahead operations like wordend, etc..
4716 where we might need to look at parts of the string that might be
4717 outside of the LIMITs (i.e past `stop'). */
4718 #define PREFETCH_NOLIMIT() \
4722 dend = end_match_2; \
4725 /* Test if at very beginning or at very end of the virtual concatenation
4726 of `string1' and `string2'. If only one string, it's `string2'. */
4727 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4728 #define AT_STRINGS_END(d) ((d) == end2)
4731 /* Test if D points to a character which is word-constituent. We have
4732 two special cases to check for: if past the end of string1, look at
4733 the first character in string2; and if before the beginning of
4734 string2, look at the last character in string1. */
4735 #define WORDCHAR_P(d) \
4736 (SYNTAX ((d) == end1 ? *string2 \
4737 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4740 /* Disabled due to a compiler bug -- see comment at case wordbound */
4742 /* The comment at case wordbound is following one, but we don't use
4743 AT_WORD_BOUNDARY anymore to support multibyte form.
4745 The DEC Alpha C compiler 3.x generates incorrect code for the
4746 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4747 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4748 macro and introducing temporary variables works around the bug. */
4751 /* Test if the character before D and the one at D differ with respect
4752 to being word-constituent. */
4753 #define AT_WORD_BOUNDARY(d) \
4754 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4755 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4758 /* Free everything we malloc. */
4759 #ifdef MATCH_MAY_ALLOCATE
4760 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4761 # define FREE_VARIABLES() \
4763 REGEX_FREE_STACK (fail_stack.stack); \
4764 FREE_VAR (regstart); \
4765 FREE_VAR (regend); \
4766 FREE_VAR (best_regstart); \
4767 FREE_VAR (best_regend); \
4770 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4771 #endif /* not MATCH_MAY_ALLOCATE */
4774 /* Optimization routines. */
4776 /* If the operation is a match against one or more chars,
4777 return a pointer to the next operation, else return NULL. */
4782 switch (SWITCH_ENUM_CAST (*p
++))
4793 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4796 p
= CHARSET_RANGE_TABLE (p
- 1);
4797 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4798 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4801 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4808 case notcategoryspec
:
4820 /* Jump over non-matching operations. */
4822 skip_noops (p
, pend
)
4828 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4837 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4848 /* Non-zero if "p1 matches something" implies "p2 fails". */
4850 mutually_exclusive_p (bufp
, p1
, p2
)
4851 struct re_pattern_buffer
*bufp
;
4855 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4856 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4858 assert (p1
>= bufp
->buffer
&& p1
< pend
4859 && p2
>= bufp
->buffer
&& p2
<= pend
);
4861 /* Skip over open/close-group commands.
4862 If what follows this loop is a ...+ construct,
4863 look at what begins its body, since we will have to
4864 match at least one of that. */
4865 p2
= skip_noops (p2
, pend
);
4866 /* The same skip can be done for p1, except that this function
4867 is only used in the case where p1 is a simple match operator. */
4868 /* p1 = skip_noops (p1, pend); */
4870 assert (p1
>= bufp
->buffer
&& p1
< pend
4871 && p2
>= bufp
->buffer
&& p2
<= pend
);
4873 op2
= p2
== pend
? succeed
: *p2
;
4875 switch (SWITCH_ENUM_CAST (op2
))
4879 /* If we're at the end of the pattern, we can change. */
4880 if (skip_one_char (p1
))
4882 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4890 register re_wchar_t c
4891 = (re_opcode_t
) *p2
== endline
? '\n'
4892 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4894 if ((re_opcode_t
) *p1
== exactn
)
4896 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4898 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4903 else if ((re_opcode_t
) *p1
== charset
4904 || (re_opcode_t
) *p1
== charset_not
)
4906 int not = (re_opcode_t
) *p1
== charset_not
;
4908 /* Test if C is listed in charset (or charset_not)
4910 if (! multibyte
|| IS_REAL_ASCII (c
))
4912 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4913 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4916 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4917 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4919 /* `not' is equal to 1 if c would match, which means
4920 that we can't change to pop_failure_jump. */
4923 DEBUG_PRINT1 (" No match => fast loop.\n");
4927 else if ((re_opcode_t
) *p1
== anychar
4930 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4938 if ((re_opcode_t
) *p1
== exactn
)
4939 /* Reuse the code above. */
4940 return mutually_exclusive_p (bufp
, p2
, p1
);
4942 /* It is hard to list up all the character in charset
4943 P2 if it includes multibyte character. Give up in
4945 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4947 /* Now, we are sure that P2 has no range table.
4948 So, for the size of bitmap in P2, `p2[1]' is
4949 enough. But P1 may have range table, so the
4950 size of bitmap table of P1 is extracted by
4951 using macro `CHARSET_BITMAP_SIZE'.
4953 In a multibyte case, we know that all the character
4954 listed in P2 is ASCII. In a unibyte case, P1 has only a
4955 bitmap table. So, in both cases, it is enough to test
4956 only the bitmap table of P1. */
4958 if ((re_opcode_t
) *p1
== charset
)
4961 /* We win if the charset inside the loop
4962 has no overlap with the one after the loop. */
4965 && idx
< CHARSET_BITMAP_SIZE (p1
));
4967 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4971 || idx
== CHARSET_BITMAP_SIZE (p1
))
4973 DEBUG_PRINT1 (" No match => fast loop.\n");
4977 else if ((re_opcode_t
) *p1
== charset_not
)
4980 /* We win if the charset_not inside the loop lists
4981 every character listed in the charset after. */
4982 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4983 if (! (p2
[2 + idx
] == 0
4984 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4985 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4990 DEBUG_PRINT1 (" No match => fast loop.\n");
4999 switch (SWITCH_ENUM_CAST (*p1
))
5003 /* Reuse the code above. */
5004 return mutually_exclusive_p (bufp
, p2
, p1
);
5006 /* When we have two charset_not, it's very unlikely that
5007 they don't overlap. The union of the two sets of excluded
5008 chars should cover all possible chars, which, as a matter of
5009 fact, is virtually impossible in multibyte buffers. */
5015 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
5017 return ((re_opcode_t
) *p1
== syntaxspec
5018 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
5020 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
5023 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
5025 return ((re_opcode_t
) *p1
== notsyntaxspec
5026 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
5028 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
5031 return (((re_opcode_t
) *p1
== notsyntaxspec
5032 || (re_opcode_t
) *p1
== syntaxspec
)
5037 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
5038 case notcategoryspec
:
5039 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
5051 /* Matching routines. */
5053 #ifndef emacs /* Emacs never uses this. */
5054 /* re_match is like re_match_2 except it takes only a single string. */
5057 re_match (bufp
, string
, size
, pos
, regs
)
5058 struct re_pattern_buffer
*bufp
;
5061 struct re_registers
*regs
;
5063 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
5067 WEAK_ALIAS (__re_match
, re_match
)
5068 #endif /* not emacs */
5071 /* In Emacs, this is the string or buffer in which we
5072 are matching. It is used for looking up syntax properties. */
5073 Lisp_Object re_match_object
;
5076 /* re_match_2 matches the compiled pattern in BUFP against the
5077 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5078 and SIZE2, respectively). We start matching at POS, and stop
5081 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5082 store offsets for the substring each group matched in REGS. See the
5083 documentation for exactly how many groups we fill.
5085 We return -1 if no match, -2 if an internal error (such as the
5086 failure stack overflowing). Otherwise, we return the length of the
5087 matched substring. */
5090 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
5091 struct re_pattern_buffer
*bufp
;
5092 const char *string1
, *string2
;
5095 struct re_registers
*regs
;
5102 gl_state
.object
= re_match_object
;
5103 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
5104 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
5107 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
5108 (re_char
*) string2
, size2
,
5112 WEAK_ALIAS (__re_match_2
, re_match_2
)
5115 /* This is a separate function so that we can force an alloca cleanup
5118 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
5119 struct re_pattern_buffer
*bufp
;
5120 re_char
*string1
, *string2
;
5123 struct re_registers
*regs
;
5126 /* General temporaries. */
5131 /* Just past the end of the corresponding string. */
5132 re_char
*end1
, *end2
;
5134 /* Pointers into string1 and string2, just past the last characters in
5135 each to consider matching. */
5136 re_char
*end_match_1
, *end_match_2
;
5138 /* Where we are in the data, and the end of the current string. */
5141 /* Used sometimes to remember where we were before starting matching
5142 an operator so that we can go back in case of failure. This "atomic"
5143 behavior of matching opcodes is indispensable to the correctness
5144 of the on_failure_keep_string_jump optimization. */
5147 /* Where we are in the pattern, and the end of the pattern. */
5148 re_char
*p
= bufp
->buffer
;
5149 re_char
*pend
= p
+ bufp
->used
;
5151 /* We use this to map every character in the string. */
5152 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5154 /* Nonzero if BUFP is setup from a multibyte regex. */
5155 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
5157 /* Nonzero if STRING1/STRING2 are multibyte. */
5158 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
5160 /* Failure point stack. Each place that can handle a failure further
5161 down the line pushes a failure point on this stack. It consists of
5162 regstart, and regend for all registers corresponding to
5163 the subexpressions we're currently inside, plus the number of such
5164 registers, and, finally, two char *'s. The first char * is where
5165 to resume scanning the pattern; the second one is where to resume
5166 scanning the strings. */
5167 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5168 fail_stack_type fail_stack
;
5171 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5174 #if defined REL_ALLOC && defined REGEX_MALLOC
5175 /* This holds the pointer to the failure stack, when
5176 it is allocated relocatably. */
5177 fail_stack_elt_t
*failure_stack_ptr
;
5180 /* We fill all the registers internally, independent of what we
5181 return, for use in backreferences. The number here includes
5182 an element for register zero. */
5183 size_t num_regs
= bufp
->re_nsub
+ 1;
5185 /* Information on the contents of registers. These are pointers into
5186 the input strings; they record just what was matched (on this
5187 attempt) by a subexpression part of the pattern, that is, the
5188 regnum-th regstart pointer points to where in the pattern we began
5189 matching and the regnum-th regend points to right after where we
5190 stopped matching the regnum-th subexpression. (The zeroth register
5191 keeps track of what the whole pattern matches.) */
5192 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5193 re_char
**regstart
, **regend
;
5196 /* The following record the register info as found in the above
5197 variables when we find a match better than any we've seen before.
5198 This happens as we backtrack through the failure points, which in
5199 turn happens only if we have not yet matched the entire string. */
5200 unsigned best_regs_set
= false;
5201 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5202 re_char
**best_regstart
, **best_regend
;
5205 /* Logically, this is `best_regend[0]'. But we don't want to have to
5206 allocate space for that if we're not allocating space for anything
5207 else (see below). Also, we never need info about register 0 for
5208 any of the other register vectors, and it seems rather a kludge to
5209 treat `best_regend' differently than the rest. So we keep track of
5210 the end of the best match so far in a separate variable. We
5211 initialize this to NULL so that when we backtrack the first time
5212 and need to test it, it's not garbage. */
5213 re_char
*match_end
= NULL
;
5216 /* Counts the total number of registers pushed. */
5217 unsigned num_regs_pushed
= 0;
5220 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5224 #ifdef MATCH_MAY_ALLOCATE
5225 /* Do not bother to initialize all the register variables if there are
5226 no groups in the pattern, as it takes a fair amount of time. If
5227 there are groups, we include space for register 0 (the whole
5228 pattern), even though we never use it, since it simplifies the
5229 array indexing. We should fix this. */
5232 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5233 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5234 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5235 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5237 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5245 /* We must initialize all our variables to NULL, so that
5246 `FREE_VARIABLES' doesn't try to free them. */
5247 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5249 #endif /* MATCH_MAY_ALLOCATE */
5251 /* The starting position is bogus. */
5252 if (pos
< 0 || pos
> size1
+ size2
)
5258 /* Initialize subexpression text positions to -1 to mark ones that no
5259 start_memory/stop_memory has been seen for. Also initialize the
5260 register information struct. */
5261 for (reg
= 1; reg
< num_regs
; reg
++)
5262 regstart
[reg
] = regend
[reg
] = NULL
;
5264 /* We move `string1' into `string2' if the latter's empty -- but not if
5265 `string1' is null. */
5266 if (size2
== 0 && string1
!= NULL
)
5273 end1
= string1
+ size1
;
5274 end2
= string2
+ size2
;
5276 /* `p' scans through the pattern as `d' scans through the data.
5277 `dend' is the end of the input string that `d' points within. `d'
5278 is advanced into the following input string whenever necessary, but
5279 this happens before fetching; therefore, at the beginning of the
5280 loop, `d' can be pointing at the end of a string, but it cannot
5284 /* Only match within string2. */
5285 d
= string2
+ pos
- size1
;
5286 dend
= end_match_2
= string2
+ stop
- size1
;
5287 end_match_1
= end1
; /* Just to give it a value. */
5293 /* Only match within string1. */
5294 end_match_1
= string1
+ stop
;
5296 When we reach end_match_1, PREFETCH normally switches to string2.
5297 But in the present case, this means that just doing a PREFETCH
5298 makes us jump from `stop' to `gap' within the string.
5299 What we really want here is for the search to stop as
5300 soon as we hit end_match_1. That's why we set end_match_2
5301 to end_match_1 (since PREFETCH fails as soon as we hit
5303 end_match_2
= end_match_1
;
5306 { /* It's important to use this code when stop == size so that
5307 moving `d' from end1 to string2 will not prevent the d == dend
5308 check from catching the end of string. */
5310 end_match_2
= string2
+ stop
- size1
;
5316 DEBUG_PRINT1 ("The compiled pattern is: ");
5317 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5318 DEBUG_PRINT1 ("The string to match is: `");
5319 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5320 DEBUG_PRINT1 ("'\n");
5322 /* This loops over pattern commands. It exits by returning from the
5323 function if the match is complete, or it drops through if the match
5324 fails at this starting point in the input data. */
5327 DEBUG_PRINT2 ("\n%p: ", p
);
5330 { /* End of pattern means we might have succeeded. */
5331 DEBUG_PRINT1 ("end of pattern ... ");
5333 /* If we haven't matched the entire string, and we want the
5334 longest match, try backtracking. */
5335 if (d
!= end_match_2
)
5337 /* 1 if this match ends in the same string (string1 or string2)
5338 as the best previous match. */
5339 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5340 == FIRST_STRING_P (d
));
5341 /* 1 if this match is the best seen so far. */
5342 boolean best_match_p
;
5344 /* AIX compiler got confused when this was combined
5345 with the previous declaration. */
5347 best_match_p
= d
> match_end
;
5349 best_match_p
= !FIRST_STRING_P (d
);
5351 DEBUG_PRINT1 ("backtracking.\n");
5353 if (!FAIL_STACK_EMPTY ())
5354 { /* More failure points to try. */
5356 /* If exceeds best match so far, save it. */
5357 if (!best_regs_set
|| best_match_p
)
5359 best_regs_set
= true;
5362 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5364 for (reg
= 1; reg
< num_regs
; reg
++)
5366 best_regstart
[reg
] = regstart
[reg
];
5367 best_regend
[reg
] = regend
[reg
];
5373 /* If no failure points, don't restore garbage. And if
5374 last match is real best match, don't restore second
5376 else if (best_regs_set
&& !best_match_p
)
5379 /* Restore best match. It may happen that `dend ==
5380 end_match_1' while the restored d is in string2.
5381 For example, the pattern `x.*y.*z' against the
5382 strings `x-' and `y-z-', if the two strings are
5383 not consecutive in memory. */
5384 DEBUG_PRINT1 ("Restoring best registers.\n");
5387 dend
= ((d
>= string1
&& d
<= end1
)
5388 ? end_match_1
: end_match_2
);
5390 for (reg
= 1; reg
< num_regs
; reg
++)
5392 regstart
[reg
] = best_regstart
[reg
];
5393 regend
[reg
] = best_regend
[reg
];
5396 } /* d != end_match_2 */
5399 DEBUG_PRINT1 ("Accepting match.\n");
5401 /* If caller wants register contents data back, do it. */
5402 if (regs
&& !bufp
->no_sub
)
5404 /* Have the register data arrays been allocated? */
5405 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5406 { /* No. So allocate them with malloc. We need one
5407 extra element beyond `num_regs' for the `-1' marker
5409 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5410 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5411 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5412 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5417 bufp
->regs_allocated
= REGS_REALLOCATE
;
5419 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5420 { /* Yes. If we need more elements than were already
5421 allocated, reallocate them. If we need fewer, just
5423 if (regs
->num_regs
< num_regs
+ 1)
5425 regs
->num_regs
= num_regs
+ 1;
5426 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5427 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5428 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5437 /* These braces fend off a "empty body in an else-statement"
5438 warning under GCC when assert expands to nothing. */
5439 assert (bufp
->regs_allocated
== REGS_FIXED
);
5442 /* Convert the pointer data in `regstart' and `regend' to
5443 indices. Register zero has to be set differently,
5444 since we haven't kept track of any info for it. */
5445 if (regs
->num_regs
> 0)
5447 regs
->start
[0] = pos
;
5448 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5451 /* Go through the first `min (num_regs, regs->num_regs)'
5452 registers, since that is all we initialized. */
5453 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5455 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5456 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5460 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5462 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5466 /* If the regs structure we return has more elements than
5467 were in the pattern, set the extra elements to -1. If
5468 we (re)allocated the registers, this is the case,
5469 because we always allocate enough to have at least one
5471 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5472 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5473 } /* regs && !bufp->no_sub */
5475 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5476 nfailure_points_pushed
, nfailure_points_popped
,
5477 nfailure_points_pushed
- nfailure_points_popped
);
5478 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5480 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5482 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5488 /* Otherwise match next pattern command. */
5489 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5491 /* Ignore these. Used to ignore the n of succeed_n's which
5492 currently have n == 0. */
5494 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5498 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5501 /* Match the next n pattern characters exactly. The following
5502 byte in the pattern defines n, and the n bytes after that
5503 are the characters to match. */
5506 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5508 /* Remember the start point to rollback upon failure. */
5512 /* This is written out as an if-else so we don't waste time
5513 testing `translate' inside the loop. */
5514 if (RE_TRANSLATE_P (translate
))
5518 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5538 /* The cost of testing `translate' is comparatively small. */
5539 if (target_multibyte
)
5542 int pat_charlen
, buf_charlen
;
5547 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5550 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5553 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5555 if (TRANSLATE (buf_ch
) != pat_ch
)
5563 mcnt
-= pat_charlen
;
5569 int pat_charlen
, buf_charlen
;
5575 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5576 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5583 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5584 if (! CHAR_BYTE8_P (buf_ch
))
5586 buf_ch
= TRANSLATE (buf_ch
);
5587 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5593 if (buf_ch
!= pat_ch
)
5606 /* Match any character except possibly a newline or a null. */
5612 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5615 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5617 buf_ch
= TRANSLATE (buf_ch
);
5619 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5621 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5622 && buf_ch
== '\000'))
5625 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5634 register unsigned int c
;
5635 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5638 /* Start of actual range_table, or end of bitmap if there is no
5640 re_char
*range_table
;
5642 /* Nonzero if there is a range table. */
5643 int range_table_exists
;
5645 /* Number of ranges of range table. This is not included
5646 in the initial byte-length of the command. */
5649 /* Whether matching against a unibyte character. */
5650 boolean unibyte_char
= false;
5652 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5654 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5656 if (range_table_exists
)
5658 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5659 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5663 c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5664 if (target_multibyte
)
5669 c1
= RE_CHAR_TO_UNIBYTE (c
);
5672 unibyte_char
= true;
5678 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5680 if (! CHAR_BYTE8_P (c1
))
5682 c1
= TRANSLATE (c1
);
5683 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5686 unibyte_char
= true;
5691 unibyte_char
= true;
5694 if (unibyte_char
&& c
< (1 << BYTEWIDTH
))
5695 { /* Lookup bitmap. */
5696 /* Cast to `unsigned' instead of `unsigned char' in
5697 case the bit list is a full 32 bytes long. */
5698 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5699 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5703 else if (range_table_exists
)
5705 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5707 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5708 | (class_bits
& BIT_MULTIBYTE
)
5709 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5710 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5711 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5712 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5715 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5719 if (range_table_exists
)
5720 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5722 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5724 if (!not) goto fail
;
5731 /* The beginning of a group is represented by start_memory.
5732 The argument is the register number. The text
5733 matched within the group is recorded (in the internal
5734 registers data structure) under the register number. */
5736 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5738 /* In case we need to undo this operation (via backtracking). */
5739 PUSH_FAILURE_REG ((unsigned int)*p
);
5742 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5743 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5745 /* Move past the register number and inner group count. */
5750 /* The stop_memory opcode represents the end of a group. Its
5751 argument is the same as start_memory's: the register number. */
5753 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5755 assert (!REG_UNSET (regstart
[*p
]));
5756 /* Strictly speaking, there should be code such as:
5758 assert (REG_UNSET (regend[*p]));
5759 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5761 But the only info to be pushed is regend[*p] and it is known to
5762 be UNSET, so there really isn't anything to push.
5763 Not pushing anything, on the other hand deprives us from the
5764 guarantee that regend[*p] is UNSET since undoing this operation
5765 will not reset its value properly. This is not important since
5766 the value will only be read on the next start_memory or at
5767 the very end and both events can only happen if this stop_memory
5771 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5773 /* Move past the register number and the inner group count. */
5778 /* \<digit> has been turned into a `duplicate' command which is
5779 followed by the numeric value of <digit> as the register number. */
5782 register re_char
*d2
, *dend2
;
5783 int regno
= *p
++; /* Get which register to match against. */
5784 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5786 /* Can't back reference a group which we've never matched. */
5787 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5790 /* Where in input to try to start matching. */
5791 d2
= regstart
[regno
];
5793 /* Remember the start point to rollback upon failure. */
5796 /* Where to stop matching; if both the place to start and
5797 the place to stop matching are in the same string, then
5798 set to the place to stop, otherwise, for now have to use
5799 the end of the first string. */
5801 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5802 == FIRST_STRING_P (regend
[regno
]))
5803 ? regend
[regno
] : end_match_1
);
5806 /* If necessary, advance to next segment in register
5810 if (dend2
== end_match_2
) break;
5811 if (dend2
== regend
[regno
]) break;
5813 /* End of string1 => advance to string2. */
5815 dend2
= regend
[regno
];
5817 /* At end of register contents => success */
5818 if (d2
== dend2
) break;
5820 /* If necessary, advance to next segment in data. */
5823 /* How many characters left in this segment to match. */
5826 /* Want how many consecutive characters we can match in
5827 one shot, so, if necessary, adjust the count. */
5828 if (mcnt
> dend2
- d2
)
5831 /* Compare that many; failure if mismatch, else move
5833 if (RE_TRANSLATE_P (translate
)
5834 ? bcmp_translate (d
, d2
, mcnt
, translate
, target_multibyte
)
5835 : memcmp (d
, d2
, mcnt
))
5840 d
+= mcnt
, d2
+= mcnt
;
5846 /* begline matches the empty string at the beginning of the string
5847 (unless `not_bol' is set in `bufp'), and after newlines. */
5849 DEBUG_PRINT1 ("EXECUTING begline.\n");
5851 if (AT_STRINGS_BEG (d
))
5853 if (!bufp
->not_bol
) break;
5858 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5862 /* In all other cases, we fail. */
5866 /* endline is the dual of begline. */
5868 DEBUG_PRINT1 ("EXECUTING endline.\n");
5870 if (AT_STRINGS_END (d
))
5872 if (!bufp
->not_eol
) break;
5876 PREFETCH_NOLIMIT ();
5883 /* Match at the very beginning of the data. */
5885 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5886 if (AT_STRINGS_BEG (d
))
5891 /* Match at the very end of the data. */
5893 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5894 if (AT_STRINGS_END (d
))
5899 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5900 pushes NULL as the value for the string on the stack. Then
5901 `POP_FAILURE_POINT' will keep the current value for the
5902 string, instead of restoring it. To see why, consider
5903 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5904 then the . fails against the \n. But the next thing we want
5905 to do is match the \n against the \n; if we restored the
5906 string value, we would be back at the foo.
5908 Because this is used only in specific cases, we don't need to
5909 check all the things that `on_failure_jump' does, to make
5910 sure the right things get saved on the stack. Hence we don't
5911 share its code. The only reason to push anything on the
5912 stack at all is that otherwise we would have to change
5913 `anychar's code to do something besides goto fail in this
5914 case; that seems worse than this. */
5915 case on_failure_keep_string_jump
:
5916 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5917 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5920 PUSH_FAILURE_POINT (p
- 3, NULL
);
5923 /* A nasty loop is introduced by the non-greedy *? and +?.
5924 With such loops, the stack only ever contains one failure point
5925 at a time, so that a plain on_failure_jump_loop kind of
5926 cycle detection cannot work. Worse yet, such a detection
5927 can not only fail to detect a cycle, but it can also wrongly
5928 detect a cycle (between different instantiations of the same
5930 So the method used for those nasty loops is a little different:
5931 We use a special cycle-detection-stack-frame which is pushed
5932 when the on_failure_jump_nastyloop failure-point is *popped*.
5933 This special frame thus marks the beginning of one iteration
5934 through the loop and we can hence easily check right here
5935 whether something matched between the beginning and the end of
5937 case on_failure_jump_nastyloop
:
5938 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5939 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5942 assert ((re_opcode_t
)p
[-4] == no_op
);
5945 CHECK_INFINITE_LOOP (p
- 4, d
);
5947 /* If there's a cycle, just continue without pushing
5948 this failure point. The failure point is the "try again"
5949 option, which shouldn't be tried.
5950 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5951 PUSH_FAILURE_POINT (p
- 3, d
);
5955 /* Simple loop detecting on_failure_jump: just check on the
5956 failure stack if the same spot was already hit earlier. */
5957 case on_failure_jump_loop
:
5959 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5960 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5964 CHECK_INFINITE_LOOP (p
- 3, d
);
5966 /* If there's a cycle, get out of the loop, as if the matching
5967 had failed. We used to just `goto fail' here, but that was
5968 aborting the search a bit too early: we want to keep the
5969 empty-loop-match and keep matching after the loop.
5970 We want (x?)*y\1z to match both xxyz and xxyxz. */
5973 PUSH_FAILURE_POINT (p
- 3, d
);
5978 /* Uses of on_failure_jump:
5980 Each alternative starts with an on_failure_jump that points
5981 to the beginning of the next alternative. Each alternative
5982 except the last ends with a jump that in effect jumps past
5983 the rest of the alternatives. (They really jump to the
5984 ending jump of the following alternative, because tensioning
5985 these jumps is a hassle.)
5987 Repeats start with an on_failure_jump that points past both
5988 the repetition text and either the following jump or
5989 pop_failure_jump back to this on_failure_jump. */
5990 case on_failure_jump
:
5991 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5992 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5995 PUSH_FAILURE_POINT (p
-3, d
);
5998 /* This operation is used for greedy *.
5999 Compare the beginning of the repeat with what in the
6000 pattern follows its end. If we can establish that there
6001 is nothing that they would both match, i.e., that we
6002 would have to backtrack because of (as in, e.g., `a*a')
6003 then we can use a non-backtracking loop based on
6004 on_failure_keep_string_jump instead of on_failure_jump. */
6005 case on_failure_jump_smart
:
6006 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6007 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
6010 re_char
*p1
= p
; /* Next operation. */
6011 /* Here, we discard `const', making re_match non-reentrant. */
6012 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
6013 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
6015 p
-= 3; /* Reset so that we will re-execute the
6016 instruction once it's been changed. */
6018 EXTRACT_NUMBER (mcnt
, p2
- 2);
6020 /* Ensure this is a indeed the trivial kind of loop
6021 we are expecting. */
6022 assert (skip_one_char (p1
) == p2
- 3);
6023 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
6024 DEBUG_STATEMENT (debug
+= 2);
6025 if (mutually_exclusive_p (bufp
, p1
, p2
))
6027 /* Use a fast `on_failure_keep_string_jump' loop. */
6028 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
6029 *p3
= (unsigned char) on_failure_keep_string_jump
;
6030 STORE_NUMBER (p2
- 2, mcnt
+ 3);
6034 /* Default to a safe `on_failure_jump' loop. */
6035 DEBUG_PRINT1 (" smart default => slow loop.\n");
6036 *p3
= (unsigned char) on_failure_jump
;
6038 DEBUG_STATEMENT (debug
-= 2);
6042 /* Unconditionally jump (without popping any failure points). */
6045 IMMEDIATE_QUIT_CHECK
;
6046 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
6047 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
6048 p
+= mcnt
; /* Do the jump. */
6049 DEBUG_PRINT2 ("(to %p).\n", p
);
6053 /* Have to succeed matching what follows at least n times.
6054 After that, handle like `on_failure_jump'. */
6056 /* Signedness doesn't matter since we only compare MCNT to 0. */
6057 EXTRACT_NUMBER (mcnt
, p
+ 2);
6058 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
6060 /* Originally, mcnt is how many times we HAVE to succeed. */
6063 /* Here, we discard `const', making re_match non-reentrant. */
6064 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
6067 PUSH_NUMBER (p2
, mcnt
);
6070 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
6075 /* Signedness doesn't matter since we only compare MCNT to 0. */
6076 EXTRACT_NUMBER (mcnt
, p
+ 2);
6077 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
6079 /* Originally, this is how many times we CAN jump. */
6082 /* Here, we discard `const', making re_match non-reentrant. */
6083 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
6085 PUSH_NUMBER (p2
, mcnt
);
6086 goto unconditional_jump
;
6088 /* If don't have to jump any more, skip over the rest of command. */
6095 unsigned char *p2
; /* Location of the counter. */
6096 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
6098 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6099 /* Here, we discard `const', making re_match non-reentrant. */
6100 p2
= (unsigned char*) p
+ mcnt
;
6101 /* Signedness doesn't matter since we only copy MCNT's bits . */
6102 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6103 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
6104 PUSH_NUMBER (p2
, mcnt
);
6110 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
6111 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
6113 /* We SUCCEED (or FAIL) in one of the following cases: */
6115 /* Case 1: D is at the beginning or the end of string. */
6116 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
6120 /* C1 is the character before D, S1 is the syntax of C1, C2
6121 is the character at D, and S2 is the syntax of C2. */
6126 int offset
= PTR_TO_OFFSET (d
- 1);
6127 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6128 UPDATE_SYNTAX_TABLE (charpos
);
6130 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6133 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6135 PREFETCH_NOLIMIT ();
6136 GET_CHAR_AFTER (c2
, d
, dummy
);
6139 if (/* Case 2: Only one of S1 and S2 is Sword. */
6140 ((s1
== Sword
) != (s2
== Sword
))
6141 /* Case 3: Both of S1 and S2 are Sword, and macro
6142 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6143 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
6152 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6154 /* We FAIL in one of the following cases: */
6156 /* Case 1: D is at the end of string. */
6157 if (AT_STRINGS_END (d
))
6161 /* C1 is the character before D, S1 is the syntax of C1, C2
6162 is the character at D, and S2 is the syntax of C2. */
6167 int offset
= PTR_TO_OFFSET (d
);
6168 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6169 UPDATE_SYNTAX_TABLE (charpos
);
6172 GET_CHAR_AFTER (c2
, d
, dummy
);
6175 /* Case 2: S2 is not Sword. */
6179 /* Case 3: D is not at the beginning of string ... */
6180 if (!AT_STRINGS_BEG (d
))
6182 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6184 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6188 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6190 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6197 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6199 /* We FAIL in one of the following cases: */
6201 /* Case 1: D is at the beginning of string. */
6202 if (AT_STRINGS_BEG (d
))
6206 /* C1 is the character before D, S1 is the syntax of C1, C2
6207 is the character at D, and S2 is the syntax of C2. */
6212 int offset
= PTR_TO_OFFSET (d
) - 1;
6213 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6214 UPDATE_SYNTAX_TABLE (charpos
);
6216 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6219 /* Case 2: S1 is not Sword. */
6223 /* Case 3: D is not at the end of string ... */
6224 if (!AT_STRINGS_END (d
))
6226 PREFETCH_NOLIMIT ();
6227 GET_CHAR_AFTER (c2
, d
, dummy
);
6229 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6233 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6235 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6242 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6244 /* We FAIL in one of the following cases: */
6246 /* Case 1: D is at the end of string. */
6247 if (AT_STRINGS_END (d
))
6251 /* C1 is the character before D, S1 is the syntax of C1, C2
6252 is the character at D, and S2 is the syntax of C2. */
6256 int offset
= PTR_TO_OFFSET (d
);
6257 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6258 UPDATE_SYNTAX_TABLE (charpos
);
6261 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6264 /* Case 2: S2 is neither Sword nor Ssymbol. */
6265 if (s2
!= Sword
&& s2
!= Ssymbol
)
6268 /* Case 3: D is not at the beginning of string ... */
6269 if (!AT_STRINGS_BEG (d
))
6271 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6273 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6277 /* ... and S1 is Sword or Ssymbol. */
6278 if (s1
== Sword
|| s1
== Ssymbol
)
6285 DEBUG_PRINT1 ("EXECUTING symend.\n");
6287 /* We FAIL in one of the following cases: */
6289 /* Case 1: D is at the beginning of string. */
6290 if (AT_STRINGS_BEG (d
))
6294 /* C1 is the character before D, S1 is the syntax of C1, C2
6295 is the character at D, and S2 is the syntax of C2. */
6299 int offset
= PTR_TO_OFFSET (d
) - 1;
6300 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6301 UPDATE_SYNTAX_TABLE (charpos
);
6303 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6306 /* Case 2: S1 is neither Ssymbol nor Sword. */
6307 if (s1
!= Sword
&& s1
!= Ssymbol
)
6310 /* Case 3: D is not at the end of string ... */
6311 if (!AT_STRINGS_END (d
))
6313 PREFETCH_NOLIMIT ();
6314 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6316 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6320 /* ... and S2 is Sword or Ssymbol. */
6321 if (s2
== Sword
|| s2
== Ssymbol
)
6329 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6331 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
6335 int offset
= PTR_TO_OFFSET (d
);
6336 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6337 UPDATE_SYNTAX_TABLE (pos1
);
6344 GET_CHAR_AFTER (c
, d
, len
);
6345 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6353 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6354 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6359 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6360 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6365 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6366 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6371 case notcategoryspec
:
6372 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6374 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
6380 GET_CHAR_AFTER (c
, d
, len
);
6381 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6392 continue; /* Successfully executed one pattern command; keep going. */
6395 /* We goto here if a matching operation fails. */
6397 IMMEDIATE_QUIT_CHECK
;
6398 if (!FAIL_STACK_EMPTY ())
6401 /* A restart point is known. Restore to that state. */
6402 DEBUG_PRINT1 ("\nFAIL:\n");
6403 POP_FAILURE_POINT (str
, pat
);
6404 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
6406 case on_failure_keep_string_jump
:
6407 assert (str
== NULL
);
6408 goto continue_failure_jump
;
6410 case on_failure_jump_nastyloop
:
6411 assert ((re_opcode_t
)pat
[-2] == no_op
);
6412 PUSH_FAILURE_POINT (pat
- 2, str
);
6415 case on_failure_jump_loop
:
6416 case on_failure_jump
:
6419 continue_failure_jump
:
6420 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6425 /* A special frame used for nastyloops. */
6432 assert (p
>= bufp
->buffer
&& p
<= pend
);
6434 if (d
>= string1
&& d
<= end1
)
6438 break; /* Matching at this starting point really fails. */
6442 goto restore_best_regs
;
6446 return -1; /* Failure to match. */
6449 /* Subroutine definitions for re_match_2. */
6451 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6452 bytes; nonzero otherwise. */
6455 bcmp_translate (s1
, s2
, len
, translate
, target_multibyte
)
6458 RE_TRANSLATE_TYPE translate
;
6459 const int target_multibyte
;
6461 register re_char
*p1
= s1
, *p2
= s2
;
6462 re_char
*p1_end
= s1
+ len
;
6463 re_char
*p2_end
= s2
+ len
;
6465 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6466 different lengths, but relying on a single `len' would break this. -sm */
6467 while (p1
< p1_end
&& p2
< p2_end
)
6469 int p1_charlen
, p2_charlen
;
6470 re_wchar_t p1_ch
, p2_ch
;
6472 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6473 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6475 if (RE_TRANSLATE (translate
, p1_ch
)
6476 != RE_TRANSLATE (translate
, p2_ch
))
6479 p1
+= p1_charlen
, p2
+= p2_charlen
;
6482 if (p1
!= p1_end
|| p2
!= p2_end
)
6488 /* Entry points for GNU code. */
6490 /* re_compile_pattern is the GNU regular expression compiler: it
6491 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6492 Returns 0 if the pattern was valid, otherwise an error string.
6494 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6495 are set in BUFP on entry.
6497 We call regex_compile to do the actual compilation. */
6500 re_compile_pattern (pattern
, length
, bufp
)
6501 const char *pattern
;
6503 struct re_pattern_buffer
*bufp
;
6508 gl_state
.current_syntax_table
= current_buffer
->syntax_table
;
6511 /* GNU code is written to assume at least RE_NREGS registers will be set
6512 (and at least one extra will be -1). */
6513 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6515 /* And GNU code determines whether or not to get register information
6516 by passing null for the REGS argument to re_match, etc., not by
6520 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6524 return gettext (re_error_msgid
[(int) ret
]);
6526 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6528 /* Entry points compatible with 4.2 BSD regex library. We don't define
6529 them unless specifically requested. */
6531 #if defined _REGEX_RE_COMP || defined _LIBC
6533 /* BSD has one and only one pattern buffer. */
6534 static struct re_pattern_buffer re_comp_buf
;
6538 /* Make these definitions weak in libc, so POSIX programs can redefine
6539 these names if they don't use our functions, and still use
6540 regcomp/regexec below without link errors. */
6550 if (!re_comp_buf
.buffer
)
6551 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6552 return (char *) gettext ("No previous regular expression");
6556 if (!re_comp_buf
.buffer
)
6558 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6559 if (re_comp_buf
.buffer
== NULL
)
6560 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6561 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6562 re_comp_buf
.allocated
= 200;
6564 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6565 if (re_comp_buf
.fastmap
== NULL
)
6566 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6567 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6570 /* Since `re_exec' always passes NULL for the `regs' argument, we
6571 don't need to initialize the pattern buffer fields which affect it. */
6573 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6578 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6579 return (char *) gettext (re_error_msgid
[(int) ret
]);
6590 const int len
= strlen (s
);
6592 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6594 #endif /* _REGEX_RE_COMP */
6596 /* POSIX.2 functions. Don't define these for Emacs. */
6600 /* regcomp takes a regular expression as a string and compiles it.
6602 PREG is a regex_t *. We do not expect any fields to be initialized,
6603 since POSIX says we shouldn't. Thus, we set
6605 `buffer' to the compiled pattern;
6606 `used' to the length of the compiled pattern;
6607 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6608 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6609 RE_SYNTAX_POSIX_BASIC;
6610 `fastmap' to an allocated space for the fastmap;
6611 `fastmap_accurate' to zero;
6612 `re_nsub' to the number of subexpressions in PATTERN.
6614 PATTERN is the address of the pattern string.
6616 CFLAGS is a series of bits which affect compilation.
6618 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6619 use POSIX basic syntax.
6621 If REG_NEWLINE is set, then . and [^...] don't match newline.
6622 Also, regexec will try a match beginning after every newline.
6624 If REG_ICASE is set, then we considers upper- and lowercase
6625 versions of letters to be equivalent when matching.
6627 If REG_NOSUB is set, then when PREG is passed to regexec, that
6628 routine will report only success or failure, and nothing about the
6631 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6632 the return codes and their meanings.) */
6635 regcomp (preg
, pattern
, cflags
)
6636 regex_t
*__restrict preg
;
6637 const char *__restrict pattern
;
6642 = (cflags
& REG_EXTENDED
) ?
6643 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6645 /* regex_compile will allocate the space for the compiled pattern. */
6647 preg
->allocated
= 0;
6650 /* Try to allocate space for the fastmap. */
6651 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6653 if (cflags
& REG_ICASE
)
6658 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6659 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6660 if (preg
->translate
== NULL
)
6661 return (int) REG_ESPACE
;
6663 /* Map uppercase characters to corresponding lowercase ones. */
6664 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6665 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6668 preg
->translate
= NULL
;
6670 /* If REG_NEWLINE is set, newlines are treated differently. */
6671 if (cflags
& REG_NEWLINE
)
6672 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6673 syntax
&= ~RE_DOT_NEWLINE
;
6674 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6677 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6679 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6681 /* POSIX says a null character in the pattern terminates it, so we
6682 can use strlen here in compiling the pattern. */
6683 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6685 /* POSIX doesn't distinguish between an unmatched open-group and an
6686 unmatched close-group: both are REG_EPAREN. */
6687 if (ret
== REG_ERPAREN
)
6690 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6691 { /* Compute the fastmap now, since regexec cannot modify the pattern
6693 re_compile_fastmap (preg
);
6694 if (preg
->can_be_null
)
6695 { /* The fastmap can't be used anyway. */
6696 free (preg
->fastmap
);
6697 preg
->fastmap
= NULL
;
6702 WEAK_ALIAS (__regcomp
, regcomp
)
6705 /* regexec searches for a given pattern, specified by PREG, in the
6708 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6709 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6710 least NMATCH elements, and we set them to the offsets of the
6711 corresponding matched substrings.
6713 EFLAGS specifies `execution flags' which affect matching: if
6714 REG_NOTBOL is set, then ^ does not match at the beginning of the
6715 string; if REG_NOTEOL is set, then $ does not match at the end.
6717 We return 0 if we find a match and REG_NOMATCH if not. */
6720 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6721 const regex_t
*__restrict preg
;
6722 const char *__restrict string
;
6724 regmatch_t pmatch
[__restrict_arr
];
6728 struct re_registers regs
;
6729 regex_t private_preg
;
6730 int len
= strlen (string
);
6731 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6733 private_preg
= *preg
;
6735 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6736 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6738 /* The user has told us exactly how many registers to return
6739 information about, via `nmatch'. We have to pass that on to the
6740 matching routines. */
6741 private_preg
.regs_allocated
= REGS_FIXED
;
6745 regs
.num_regs
= nmatch
;
6746 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6747 if (regs
.start
== NULL
)
6748 return (int) REG_NOMATCH
;
6749 regs
.end
= regs
.start
+ nmatch
;
6752 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6753 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6754 was a little bit longer but still only matching the real part.
6755 This works because the `endline' will check for a '\n' and will find a
6756 '\0', correctly deciding that this is not the end of a line.
6757 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6758 a convenient '\0' there. For all we know, the string could be preceded
6759 by '\n' which would throw things off. */
6761 /* Perform the searching operation. */
6762 ret
= re_search (&private_preg
, string
, len
,
6763 /* start: */ 0, /* range: */ len
,
6764 want_reg_info
? ®s
: (struct re_registers
*) 0);
6766 /* Copy the register information to the POSIX structure. */
6773 for (r
= 0; r
< nmatch
; r
++)
6775 pmatch
[r
].rm_so
= regs
.start
[r
];
6776 pmatch
[r
].rm_eo
= regs
.end
[r
];
6780 /* If we needed the temporary register info, free the space now. */
6784 /* We want zero return to mean success, unlike `re_search'. */
6785 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6787 WEAK_ALIAS (__regexec
, regexec
)
6790 /* Returns a message corresponding to an error code, ERR_CODE, returned
6791 from either regcomp or regexec. We don't use PREG here.
6793 ERR_CODE was previously called ERRCODE, but that name causes an
6794 error with msvc8 compiler. */
6797 regerror (err_code
, preg
, errbuf
, errbuf_size
)
6799 const regex_t
*preg
;
6807 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6808 /* Only error codes returned by the rest of the code should be passed
6809 to this routine. If we are given anything else, or if other regex
6810 code generates an invalid error code, then the program has a bug.
6811 Dump core so we can fix it. */
6814 msg
= gettext (re_error_msgid
[err_code
]);
6816 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6818 if (errbuf_size
!= 0)
6820 if (msg_size
> errbuf_size
)
6822 strncpy (errbuf
, msg
, errbuf_size
- 1);
6823 errbuf
[errbuf_size
- 1] = 0;
6826 strcpy (errbuf
, msg
);
6831 WEAK_ALIAS (__regerror
, regerror
)
6834 /* Free dynamically allocated space used by PREG. */
6840 free (preg
->buffer
);
6841 preg
->buffer
= NULL
;
6843 preg
->allocated
= 0;
6846 free (preg
->fastmap
);
6847 preg
->fastmap
= NULL
;
6848 preg
->fastmap_accurate
= 0;
6850 free (preg
->translate
);
6851 preg
->translate
= NULL
;
6853 WEAK_ALIAS (__regfree
, regfree
)
6855 #endif /* not emacs */
6857 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6858 (do not change this comment) */