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) 1993-2011 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3, or (at your option)
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
23 - structure the opcode space into opcode+flag.
24 - merge with glibc's regex.[ch].
25 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
26 need to modify the compiled regexp so that re_match can be reentrant.
27 - get rid of on_failure_jump_smart by doing the optimization in re_comp
28 rather than at run-time, so that re_match can be reentrant.
31 /* AIX requires this to be the first thing in the file. */
32 #if defined _AIX && !defined REGEX_MALLOC
40 #if defined STDC_HEADERS && !defined emacs
43 /* We need this for `regex.h', and perhaps for the Emacs include files. */
44 # include <sys/types.h>
47 /* Whether to use ISO C Amendment 1 wide char functions.
48 Those should not be used for Emacs since it uses its own. */
50 #define WIDE_CHAR_SUPPORT 1
52 #define WIDE_CHAR_SUPPORT \
53 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
56 /* For platform which support the ISO C amendement 1 functionality we
57 support user defined character classes. */
59 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
65 /* We have to keep the namespace clean. */
66 # define regfree(preg) __regfree (preg)
67 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
68 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
69 # define regerror(err_code, preg, errbuf, errbuf_size) \
70 __regerror(err_code, preg, errbuf, errbuf_size)
71 # define re_set_registers(bu, re, nu, st, en) \
72 __re_set_registers (bu, re, nu, st, en)
73 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
74 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
75 # define re_match(bufp, string, size, pos, regs) \
76 __re_match (bufp, string, size, pos, regs)
77 # define re_search(bufp, string, size, startpos, range, regs) \
78 __re_search (bufp, string, size, startpos, range, regs)
79 # define re_compile_pattern(pattern, length, bufp) \
80 __re_compile_pattern (pattern, length, bufp)
81 # define re_set_syntax(syntax) __re_set_syntax (syntax)
82 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
83 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
84 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
86 /* Make sure we call libc's function even if the user overrides them. */
87 # define btowc __btowc
88 # define iswctype __iswctype
89 # define wctype __wctype
91 # define WEAK_ALIAS(a,b) weak_alias (a, b)
93 /* We are also using some library internals. */
94 # include <locale/localeinfo.h>
95 # include <locale/elem-hash.h>
96 # include <langinfo.h>
98 # define WEAK_ALIAS(a,b)
101 /* This is for other GNU distributions with internationalized messages. */
102 #if HAVE_LIBINTL_H || defined _LIBC
103 # include <libintl.h>
105 # define gettext(msgid) (msgid)
109 /* This define is so xgettext can find the internationalizable
111 # define gettext_noop(String) String
114 /* The `emacs' switch turns on certain matching commands
115 that make sense only in Emacs. */
122 /* Make syntax table lookup grant data in gl_state. */
123 # define SYNTAX_ENTRY_VIA_PROPERTY
126 # include "character.h"
127 # include "category.h"
132 # define malloc xmalloc
136 # define realloc xrealloc
142 /* Converts the pointer to the char to BEG-based offset from the start. */
143 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
144 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
146 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
147 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
148 # define RE_STRING_CHAR(p, multibyte) \
149 (multibyte ? (STRING_CHAR (p)) : (*(p)))
150 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
151 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
153 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
155 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
157 /* Set C a (possibly converted to multibyte) character before P. P
158 points into a string which is the virtual concatenation of STR1
159 (which ends at END1) or STR2 (which ends at END2). */
160 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
162 if (target_multibyte) \
164 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
165 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
166 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
167 c = STRING_CHAR (dtemp); \
171 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
172 (c) = RE_CHAR_TO_MULTIBYTE (c); \
176 /* Set C a (possibly converted to multibyte) character at P, and set
177 LEN to the byte length of that character. */
178 # define GET_CHAR_AFTER(c, p, len) \
180 if (target_multibyte) \
181 (c) = STRING_CHAR_AND_LENGTH (p, len); \
186 (c) = RE_CHAR_TO_MULTIBYTE (c); \
190 #else /* not emacs */
192 /* If we are not linking with Emacs proper,
193 we can't use the relocating allocator
194 even if config.h says that we can. */
199 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
202 xmalloc (size_t size
)
205 val
= (void *) malloc (size
);
208 write (2, "virtual memory exhausted\n", 25);
215 xrealloc (void *block
, size_t size
)
218 /* We must call malloc explicitly when BLOCK is 0, since some
219 reallocs don't do this. */
221 val
= (void *) malloc (size
);
223 val
= (void *) realloc (block
, size
);
226 write (2, "virtual memory exhausted\n", 25);
235 # define malloc xmalloc
239 # define realloc xrealloc
241 /* This is the normal way of making sure we have memcpy, memcmp and memset. */
242 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
245 # include <strings.h>
247 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
250 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
254 /* Define the syntax stuff for \<, \>, etc. */
256 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
257 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
259 # define SWITCH_ENUM_CAST(x) (x)
261 /* Dummy macros for non-Emacs environments. */
262 # define CHAR_CHARSET(c) 0
263 # define CHARSET_LEADING_CODE_BASE(c) 0
264 # define MAX_MULTIBYTE_LENGTH 1
265 # define RE_MULTIBYTE_P(x) 0
266 # define RE_TARGET_MULTIBYTE_P(x) 0
267 # define WORD_BOUNDARY_P(c1, c2) (0)
268 # define CHAR_HEAD_P(p) (1)
269 # define SINGLE_BYTE_CHAR_P(c) (1)
270 # define SAME_CHARSET_P(c1, c2) (1)
271 # define BYTES_BY_CHAR_HEAD(p) (1)
272 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
273 # define STRING_CHAR(p) (*(p))
274 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
275 # define CHAR_STRING(c, s) (*(s) = (c), 1)
276 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
277 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
278 # define RE_CHAR_TO_MULTIBYTE(c) (c)
279 # define RE_CHAR_TO_UNIBYTE(c) (c)
280 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
281 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
282 # define GET_CHAR_AFTER(c, p, len) \
284 # define MAKE_CHAR(charset, c1, c2) (c1)
285 # define BYTE8_TO_CHAR(c) (c)
286 # define CHAR_BYTE8_P(c) (0)
287 # define CHAR_LEADING_CODE(c) (c)
289 #endif /* not emacs */
292 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
293 # define RE_TRANSLATE_P(TBL) (TBL)
296 /* Get the interface, including the syntax bits. */
299 /* isalpha etc. are used for the character classes. */
304 /* 1 if C is an ASCII character. */
305 # define IS_REAL_ASCII(c) ((c) < 0200)
307 /* 1 if C is a unibyte character. */
308 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
310 /* The Emacs definitions should not be directly affected by locales. */
312 /* In Emacs, these are only used for single-byte characters. */
313 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
314 # define ISCNTRL(c) ((c) < ' ')
315 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
316 || ((c) >= 'a' && (c) <= 'f') \
317 || ((c) >= 'A' && (c) <= 'F'))
319 /* This is only used for single-byte characters. */
320 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
322 /* The rest must handle multibyte characters. */
324 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
325 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
328 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
329 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
332 # define ISALNUM(c) (IS_REAL_ASCII (c) \
333 ? (((c) >= 'a' && (c) <= 'z') \
334 || ((c) >= 'A' && (c) <= 'Z') \
335 || ((c) >= '0' && (c) <= '9')) \
336 : SYNTAX (c) == Sword)
338 # define ISALPHA(c) (IS_REAL_ASCII (c) \
339 ? (((c) >= 'a' && (c) <= 'z') \
340 || ((c) >= 'A' && (c) <= 'Z')) \
341 : SYNTAX (c) == Sword)
343 # define ISLOWER(c) lowercasep (c)
345 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
346 ? ((c) > ' ' && (c) < 0177 \
347 && !(((c) >= 'a' && (c) <= 'z') \
348 || ((c) >= 'A' && (c) <= 'Z') \
349 || ((c) >= '0' && (c) <= '9'))) \
350 : SYNTAX (c) != Sword)
352 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
354 # define ISUPPER(c) uppercasep (c)
356 # define ISWORD(c) (SYNTAX (c) == Sword)
358 #else /* not emacs */
360 /* Jim Meyering writes:
362 "... Some ctype macros are valid only for character codes that
363 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
364 using /bin/cc or gcc but without giving an ansi option). So, all
365 ctype uses should be through macros like ISPRINT... If
366 STDC_HEADERS is defined, then autoconf has verified that the ctype
367 macros don't need to be guarded with references to isascii. ...
368 Defining isascii to 1 should let any compiler worth its salt
369 eliminate the && through constant folding."
370 Solaris defines some of these symbols so we must undefine them first. */
373 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
374 # define ISASCII(c) 1
376 # define ISASCII(c) isascii(c)
379 /* 1 if C is an ASCII character. */
380 # define IS_REAL_ASCII(c) ((c) < 0200)
382 /* This distinction is not meaningful, except in Emacs. */
383 # define ISUNIBYTE(c) 1
386 # define ISBLANK(c) (ISASCII (c) && isblank (c))
388 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
391 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
393 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
397 # define ISPRINT(c) (ISASCII (c) && isprint (c))
398 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
399 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
400 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
401 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
402 # define ISLOWER(c) (ISASCII (c) && islower (c))
403 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
404 # define ISSPACE(c) (ISASCII (c) && isspace (c))
405 # define ISUPPER(c) (ISASCII (c) && isupper (c))
406 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
408 # define ISWORD(c) ISALPHA(c)
411 # define TOLOWER(c) _tolower(c)
413 # define TOLOWER(c) tolower(c)
416 /* How many characters in the character set. */
417 # define CHAR_SET_SIZE 256
421 extern char *re_syntax_table
;
423 # else /* not SYNTAX_TABLE */
425 static char re_syntax_table
[CHAR_SET_SIZE
];
428 init_syntax_once (void)
436 memset (re_syntax_table
, 0, sizeof re_syntax_table
);
438 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
440 re_syntax_table
[c
] = Sword
;
442 re_syntax_table
['_'] = Ssymbol
;
447 # endif /* not SYNTAX_TABLE */
449 # define SYNTAX(c) re_syntax_table[(c)]
451 #endif /* not emacs */
454 # define NULL (void *)0
457 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
458 since ours (we hope) works properly with all combinations of
459 machines, compilers, `char' and `unsigned char' argument types.
460 (Per Bothner suggested the basic approach.) */
461 #undef SIGN_EXTEND_CHAR
463 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
464 #else /* not __STDC__ */
465 /* As in Harbison and Steele. */
466 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
469 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
470 use `alloca' instead of `malloc'. This is because using malloc in
471 re_search* or re_match* could cause memory leaks when C-g is used in
472 Emacs; also, malloc is slower and causes storage fragmentation. On
473 the other hand, malloc is more portable, and easier to debug.
475 Because we sometimes use alloca, some routines have to be macros,
476 not functions -- `alloca'-allocated space disappears at the end of the
477 function it is called in. */
481 # define REGEX_ALLOCATE malloc
482 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
483 # define REGEX_FREE free
485 #else /* not REGEX_MALLOC */
487 /* Emacs already defines alloca, sometimes. */
490 /* Make alloca work the best possible way. */
492 # define alloca __builtin_alloca
493 # else /* not __GNUC__ */
494 # ifdef HAVE_ALLOCA_H
496 # endif /* HAVE_ALLOCA_H */
497 # endif /* not __GNUC__ */
499 # endif /* not alloca */
501 # define REGEX_ALLOCATE alloca
503 /* Assumes a `char *destination' variable. */
504 # define REGEX_REALLOCATE(source, osize, nsize) \
505 (destination = (char *) alloca (nsize), \
506 memcpy (destination, source, osize))
508 /* No need to do anything to free, after alloca. */
509 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
511 #endif /* not REGEX_MALLOC */
513 /* Define how to allocate the failure stack. */
515 #if defined REL_ALLOC && defined REGEX_MALLOC
517 # define REGEX_ALLOCATE_STACK(size) \
518 r_alloc (&failure_stack_ptr, (size))
519 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
520 r_re_alloc (&failure_stack_ptr, (nsize))
521 # define REGEX_FREE_STACK(ptr) \
522 r_alloc_free (&failure_stack_ptr)
524 #else /* not using relocating allocator */
528 # define REGEX_ALLOCATE_STACK malloc
529 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
530 # define REGEX_FREE_STACK free
532 # else /* not REGEX_MALLOC */
534 # define REGEX_ALLOCATE_STACK alloca
536 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
537 REGEX_REALLOCATE (source, osize, nsize)
538 /* No need to explicitly free anything. */
539 # define REGEX_FREE_STACK(arg) ((void)0)
541 # endif /* not REGEX_MALLOC */
542 #endif /* not using relocating allocator */
545 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
546 `string1' or just past its end. This works if PTR is NULL, which is
548 #define FIRST_STRING_P(ptr) \
549 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
551 /* (Re)Allocate N items of type T using malloc, or fail. */
552 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
553 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
554 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
556 #define BYTEWIDTH 8 /* In bits. */
558 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
562 #define MAX(a, b) ((a) > (b) ? (a) : (b))
563 #define MIN(a, b) ((a) < (b) ? (a) : (b))
565 /* Type of source-pattern and string chars. */
566 typedef const unsigned char re_char
;
568 typedef char boolean
;
572 static int re_match_2_internal
_RE_ARGS ((struct re_pattern_buffer
*bufp
,
573 re_char
*string1
, int size1
,
574 re_char
*string2
, int size2
,
576 struct re_registers
*regs
,
579 /* These are the command codes that appear in compiled regular
580 expressions. Some opcodes are followed by argument bytes. A
581 command code can specify any interpretation whatsoever for its
582 arguments. Zero bytes may appear in the compiled regular expression. */
588 /* Succeed right away--no more backtracking. */
591 /* Followed by one byte giving n, then by n literal bytes. */
594 /* Matches any (more or less) character. */
597 /* Matches any one char belonging to specified set. First
598 following byte is number of bitmap bytes. Then come bytes
599 for a bitmap saying which chars are in. Bits in each byte
600 are ordered low-bit-first. A character is in the set if its
601 bit is 1. A character too large to have a bit in the map is
602 automatically not in the set.
604 If the length byte has the 0x80 bit set, then that stuff
605 is followed by a range table:
606 2 bytes of flags for character sets (low 8 bits, high 8 bits)
607 See RANGE_TABLE_WORK_BITS below.
608 2 bytes, the number of pairs that follow (upto 32767)
609 pairs, each 2 multibyte characters,
610 each multibyte character represented as 3 bytes. */
613 /* Same parameters as charset, but match any character that is
614 not one of those specified. */
617 /* Start remembering the text that is matched, for storing in a
618 register. Followed by one byte with the register number, in
619 the range 0 to one less than the pattern buffer's re_nsub
623 /* Stop remembering the text that is matched and store it in a
624 memory register. Followed by one byte with the register
625 number, in the range 0 to one less than `re_nsub' in the
629 /* Match a duplicate of something remembered. Followed by one
630 byte containing the register number. */
633 /* Fail unless at beginning of line. */
636 /* Fail unless at end of line. */
639 /* Succeeds if at beginning of buffer (if emacs) or at beginning
640 of string to be matched (if not). */
643 /* Analogously, for end of buffer/string. */
646 /* Followed by two byte relative address to which to jump. */
649 /* Followed by two-byte relative address of place to resume at
650 in case of failure. */
653 /* Like on_failure_jump, but pushes a placeholder instead of the
654 current string position when executed. */
655 on_failure_keep_string_jump
,
657 /* Just like `on_failure_jump', except that it checks that we
658 don't get stuck in an infinite loop (matching an empty string
660 on_failure_jump_loop
,
662 /* Just like `on_failure_jump_loop', except that it checks for
663 a different kind of loop (the kind that shows up with non-greedy
664 operators). This operation has to be immediately preceded
666 on_failure_jump_nastyloop
,
668 /* A smart `on_failure_jump' used for greedy * and + operators.
669 It analyses the loop before which it is put and if the
670 loop does not require backtracking, it changes itself to
671 `on_failure_keep_string_jump' and short-circuits the loop,
672 else it just defaults to changing itself into `on_failure_jump'.
673 It assumes that it is pointing to just past a `jump'. */
674 on_failure_jump_smart
,
676 /* Followed by two-byte relative address and two-byte number n.
677 After matching N times, jump to the address upon failure.
678 Does not work if N starts at 0: use on_failure_jump_loop
682 /* Followed by two-byte relative address, and two-byte number n.
683 Jump to the address N times, then fail. */
686 /* Set the following two-byte relative address to the
687 subsequent two-byte number. The address *includes* the two
691 wordbeg
, /* Succeeds if at word beginning. */
692 wordend
, /* Succeeds if at word end. */
694 wordbound
, /* Succeeds if at a word boundary. */
695 notwordbound
, /* Succeeds if not at a word boundary. */
697 symbeg
, /* Succeeds if at symbol beginning. */
698 symend
, /* Succeeds if at symbol end. */
700 /* Matches any character whose syntax is specified. Followed by
701 a byte which contains a syntax code, e.g., Sword. */
704 /* Matches any character whose syntax is not that specified. */
708 ,before_dot
, /* Succeeds if before point. */
709 at_dot
, /* Succeeds if at point. */
710 after_dot
, /* Succeeds if after point. */
712 /* Matches any character whose category-set contains the specified
713 category. The operator is followed by a byte which contains a
714 category code (mnemonic ASCII character). */
717 /* Matches any character whose category-set does not contain the
718 specified category. The operator is followed by a byte which
719 contains the category code (mnemonic ASCII character). */
724 /* Common operations on the compiled pattern. */
726 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
728 #define STORE_NUMBER(destination, number) \
730 (destination)[0] = (number) & 0377; \
731 (destination)[1] = (number) >> 8; \
734 /* Same as STORE_NUMBER, except increment DESTINATION to
735 the byte after where the number is stored. Therefore, DESTINATION
736 must be an lvalue. */
738 #define STORE_NUMBER_AND_INCR(destination, number) \
740 STORE_NUMBER (destination, number); \
741 (destination) += 2; \
744 /* Put into DESTINATION a number stored in two contiguous bytes starting
747 #define EXTRACT_NUMBER(destination, source) \
749 (destination) = *(source) & 0377; \
750 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
754 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
756 extract_number (dest
, source
)
760 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
761 *dest
= *source
& 0377;
765 # ifndef EXTRACT_MACROS /* To debug the macros. */
766 # undef EXTRACT_NUMBER
767 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
768 # endif /* not EXTRACT_MACROS */
772 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
773 SOURCE must be an lvalue. */
775 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
777 EXTRACT_NUMBER (destination, source); \
782 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
785 extract_number_and_incr (destination
, source
)
789 extract_number (destination
, *source
);
793 # ifndef EXTRACT_MACROS
794 # undef EXTRACT_NUMBER_AND_INCR
795 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
796 extract_number_and_incr (&dest, &src)
797 # endif /* not EXTRACT_MACROS */
801 /* Store a multibyte character in three contiguous bytes starting
802 DESTINATION, and increment DESTINATION to the byte after where the
803 character is stored. Therefore, DESTINATION must be an lvalue. */
805 #define STORE_CHARACTER_AND_INCR(destination, character) \
807 (destination)[0] = (character) & 0377; \
808 (destination)[1] = ((character) >> 8) & 0377; \
809 (destination)[2] = (character) >> 16; \
810 (destination) += 3; \
813 /* Put into DESTINATION a character stored in three contiguous bytes
814 starting at SOURCE. */
816 #define EXTRACT_CHARACTER(destination, source) \
818 (destination) = ((source)[0] \
819 | ((source)[1] << 8) \
820 | ((source)[2] << 16)); \
824 /* Macros for charset. */
826 /* Size of bitmap of charset P in bytes. P is a start of charset,
827 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
828 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
830 /* Nonzero if charset P has range table. */
831 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
833 /* Return the address of range table of charset P. But not the start
834 of table itself, but the before where the number of ranges is
835 stored. `2 +' means to skip re_opcode_t and size of bitmap,
836 and the 2 bytes of flags at the start of the range table. */
837 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
839 /* Extract the bit flags that start a range table. */
840 #define CHARSET_RANGE_TABLE_BITS(p) \
841 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
842 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
844 /* Return the address of end of RANGE_TABLE. COUNT is number of
845 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
846 is start of range and end of range. `* 3' is size of each start
848 #define CHARSET_RANGE_TABLE_END(range_table, count) \
849 ((range_table) + (count) * 2 * 3)
851 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
852 COUNT is number of ranges in RANGE_TABLE. */
853 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
856 re_wchar_t range_start, range_end; \
858 re_char *range_table_end \
859 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
861 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
863 EXTRACT_CHARACTER (range_start, rtp); \
864 EXTRACT_CHARACTER (range_end, rtp + 3); \
866 if (range_start <= (c) && (c) <= range_end) \
875 /* Test if C is in range table of CHARSET. The flag NOT is negated if
876 C is listed in it. */
877 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
880 /* Number of ranges in range table. */ \
882 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
884 EXTRACT_NUMBER_AND_INCR (count, range_table); \
885 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
889 /* If DEBUG is defined, Regex prints many voluminous messages about what
890 it is doing (if the variable `debug' is nonzero). If linked with the
891 main program in `iregex.c', you can enter patterns and strings
892 interactively. And if linked with the main program in `main.c' and
893 the other test files, you can run the already-written tests. */
897 /* We use standard I/O for debugging. */
900 /* It is useful to test things that ``must'' be true when debugging. */
903 static int debug
= -100000;
905 # define DEBUG_STATEMENT(e) e
906 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
907 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
908 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
909 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
910 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
911 if (debug > 0) print_partial_compiled_pattern (s, e)
912 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
913 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
916 /* Print the fastmap in human-readable form. */
919 print_fastmap (fastmap
)
922 unsigned was_a_range
= 0;
925 while (i
< (1 << BYTEWIDTH
))
931 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
947 /* Print a compiled pattern string in human-readable form, starting at
948 the START pointer into it and ending just before the pointer END. */
951 print_partial_compiled_pattern (start
, end
)
961 fprintf (stderr
, "(null)\n");
965 /* Loop over pattern commands. */
968 fprintf (stderr
, "%d:\t", p
- start
);
970 switch ((re_opcode_t
) *p
++)
973 fprintf (stderr
, "/no_op");
977 fprintf (stderr
, "/succeed");
982 fprintf (stderr
, "/exactn/%d", mcnt
);
985 fprintf (stderr
, "/%c", *p
++);
991 fprintf (stderr
, "/start_memory/%d", *p
++);
995 fprintf (stderr
, "/stop_memory/%d", *p
++);
999 fprintf (stderr
, "/duplicate/%d", *p
++);
1003 fprintf (stderr
, "/anychar");
1009 register int c
, last
= -100;
1010 register int in_range
= 0;
1011 int length
= CHARSET_BITMAP_SIZE (p
- 1);
1012 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
1014 fprintf (stderr
, "/charset [%s",
1015 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
1018 fprintf (stderr
, " !extends past end of pattern! ");
1020 for (c
= 0; c
< 256; c
++)
1022 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
1024 /* Are we starting a range? */
1025 if (last
+ 1 == c
&& ! in_range
)
1027 fprintf (stderr
, "-");
1030 /* Have we broken a range? */
1031 else if (last
+ 1 != c
&& in_range
)
1033 fprintf (stderr
, "%c", last
);
1038 fprintf (stderr
, "%c", c
);
1044 fprintf (stderr
, "%c", last
);
1046 fprintf (stderr
, "]");
1050 if (has_range_table
)
1053 fprintf (stderr
, "has-range-table");
1055 /* ??? Should print the range table; for now, just skip it. */
1056 p
+= 2; /* skip range table bits */
1057 EXTRACT_NUMBER_AND_INCR (count
, p
);
1058 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1064 fprintf (stderr
, "/begline");
1068 fprintf (stderr
, "/endline");
1071 case on_failure_jump
:
1072 extract_number_and_incr (&mcnt
, &p
);
1073 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1076 case on_failure_keep_string_jump
:
1077 extract_number_and_incr (&mcnt
, &p
);
1078 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1081 case on_failure_jump_nastyloop
:
1082 extract_number_and_incr (&mcnt
, &p
);
1083 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1086 case on_failure_jump_loop
:
1087 extract_number_and_incr (&mcnt
, &p
);
1088 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1091 case on_failure_jump_smart
:
1092 extract_number_and_incr (&mcnt
, &p
);
1093 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1097 extract_number_and_incr (&mcnt
, &p
);
1098 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1102 extract_number_and_incr (&mcnt
, &p
);
1103 extract_number_and_incr (&mcnt2
, &p
);
1104 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1108 extract_number_and_incr (&mcnt
, &p
);
1109 extract_number_and_incr (&mcnt2
, &p
);
1110 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1114 extract_number_and_incr (&mcnt
, &p
);
1115 extract_number_and_incr (&mcnt2
, &p
);
1116 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1120 fprintf (stderr
, "/wordbound");
1124 fprintf (stderr
, "/notwordbound");
1128 fprintf (stderr
, "/wordbeg");
1132 fprintf (stderr
, "/wordend");
1136 fprintf (stderr
, "/symbeg");
1140 fprintf (stderr
, "/symend");
1144 fprintf (stderr
, "/syntaxspec");
1146 fprintf (stderr
, "/%d", mcnt
);
1150 fprintf (stderr
, "/notsyntaxspec");
1152 fprintf (stderr
, "/%d", mcnt
);
1157 fprintf (stderr
, "/before_dot");
1161 fprintf (stderr
, "/at_dot");
1165 fprintf (stderr
, "/after_dot");
1169 fprintf (stderr
, "/categoryspec");
1171 fprintf (stderr
, "/%d", mcnt
);
1174 case notcategoryspec
:
1175 fprintf (stderr
, "/notcategoryspec");
1177 fprintf (stderr
, "/%d", mcnt
);
1182 fprintf (stderr
, "/begbuf");
1186 fprintf (stderr
, "/endbuf");
1190 fprintf (stderr
, "?%d", *(p
-1));
1193 fprintf (stderr
, "\n");
1196 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1201 print_compiled_pattern (bufp
)
1202 struct re_pattern_buffer
*bufp
;
1204 re_char
*buffer
= bufp
->buffer
;
1206 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1207 printf ("%ld bytes used/%ld bytes allocated.\n",
1208 bufp
->used
, bufp
->allocated
);
1210 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1212 printf ("fastmap: ");
1213 print_fastmap (bufp
->fastmap
);
1216 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1217 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1218 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1219 printf ("no_sub: %d\t", bufp
->no_sub
);
1220 printf ("not_bol: %d\t", bufp
->not_bol
);
1221 printf ("not_eol: %d\t", bufp
->not_eol
);
1222 printf ("syntax: %lx\n", bufp
->syntax
);
1224 /* Perhaps we should print the translate table? */
1229 print_double_string (where
, string1
, size1
, string2
, size2
)
1242 if (FIRST_STRING_P (where
))
1244 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1245 putchar (string1
[this_char
]);
1250 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1251 putchar (string2
[this_char
]);
1255 #else /* not DEBUG */
1260 # define DEBUG_STATEMENT(e)
1261 # define DEBUG_PRINT1(x)
1262 # define DEBUG_PRINT2(x1, x2)
1263 # define DEBUG_PRINT3(x1, x2, x3)
1264 # define DEBUG_PRINT4(x1, x2, x3, x4)
1265 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1266 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1268 #endif /* not DEBUG */
1270 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1271 also be assigned to arbitrarily: each pattern buffer stores its own
1272 syntax, so it can be changed between regex compilations. */
1273 /* This has no initializer because initialized variables in Emacs
1274 become read-only after dumping. */
1275 reg_syntax_t re_syntax_options
;
1278 /* Specify the precise syntax of regexps for compilation. This provides
1279 for compatibility for various utilities which historically have
1280 different, incompatible syntaxes.
1282 The argument SYNTAX is a bit mask comprised of the various bits
1283 defined in regex.h. We return the old syntax. */
1286 re_set_syntax (reg_syntax_t syntax
)
1288 reg_syntax_t ret
= re_syntax_options
;
1290 re_syntax_options
= syntax
;
1293 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1295 /* Regexp to use to replace spaces, or NULL meaning don't. */
1296 static re_char
*whitespace_regexp
;
1299 re_set_whitespace_regexp (const char *regexp
)
1301 whitespace_regexp
= (re_char
*) regexp
;
1303 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1305 /* This table gives an error message for each of the error codes listed
1306 in regex.h. Obviously the order here has to be same as there.
1307 POSIX doesn't require that we do anything for REG_NOERROR,
1308 but why not be nice? */
1310 static const char *re_error_msgid
[] =
1312 gettext_noop ("Success"), /* REG_NOERROR */
1313 gettext_noop ("No match"), /* REG_NOMATCH */
1314 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1315 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1316 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1317 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1318 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1319 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1320 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1321 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1322 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1323 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1324 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1325 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1326 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1327 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1328 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1329 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1332 /* Avoiding alloca during matching, to placate r_alloc. */
1334 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1335 searching and matching functions should not call alloca. On some
1336 systems, alloca is implemented in terms of malloc, and if we're
1337 using the relocating allocator routines, then malloc could cause a
1338 relocation, which might (if the strings being searched are in the
1339 ralloc heap) shift the data out from underneath the regexp
1342 Here's another reason to avoid allocation: Emacs
1343 processes input from X in a signal handler; processing X input may
1344 call malloc; if input arrives while a matching routine is calling
1345 malloc, then we're scrod. But Emacs can't just block input while
1346 calling matching routines; then we don't notice interrupts when
1347 they come in. So, Emacs blocks input around all regexp calls
1348 except the matching calls, which it leaves unprotected, in the
1349 faith that they will not malloc. */
1351 /* Normally, this is fine. */
1352 #define MATCH_MAY_ALLOCATE
1354 /* The match routines may not allocate if (1) they would do it with malloc
1355 and (2) it's not safe for them to use malloc.
1356 Note that if REL_ALLOC is defined, matching would not use malloc for the
1357 failure stack, but we would still use it for the register vectors;
1358 so REL_ALLOC should not affect this. */
1359 #if defined REGEX_MALLOC && defined emacs
1360 # undef MATCH_MAY_ALLOCATE
1364 /* Failure stack declarations and macros; both re_compile_fastmap and
1365 re_match_2 use a failure stack. These have to be macros because of
1366 REGEX_ALLOCATE_STACK. */
1369 /* Approximate number of failure points for which to initially allocate space
1370 when matching. If this number is exceeded, we allocate more
1371 space, so it is not a hard limit. */
1372 #ifndef INIT_FAILURE_ALLOC
1373 # define INIT_FAILURE_ALLOC 20
1376 /* Roughly the maximum number of failure points on the stack. Would be
1377 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1378 This is a variable only so users of regex can assign to it; we never
1379 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1380 before using it, so it should probably be a byte-count instead. */
1381 # if defined MATCH_MAY_ALLOCATE
1382 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1383 whose default stack limit is 2mb. In order for a larger
1384 value to work reliably, you have to try to make it accord
1385 with the process stack limit. */
1386 size_t re_max_failures
= 40000;
1388 size_t re_max_failures
= 4000;
1391 union fail_stack_elt
1394 /* This should be the biggest `int' that's no bigger than a pointer. */
1398 typedef union fail_stack_elt fail_stack_elt_t
;
1402 fail_stack_elt_t
*stack
;
1404 size_t avail
; /* Offset of next open position. */
1405 size_t frame
; /* Offset of the cur constructed frame. */
1408 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1411 /* Define macros to initialize and free the failure stack.
1412 Do `return -2' if the alloc fails. */
1414 #ifdef MATCH_MAY_ALLOCATE
1415 # define INIT_FAIL_STACK() \
1417 fail_stack.stack = (fail_stack_elt_t *) \
1418 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1419 * sizeof (fail_stack_elt_t)); \
1421 if (fail_stack.stack == NULL) \
1424 fail_stack.size = INIT_FAILURE_ALLOC; \
1425 fail_stack.avail = 0; \
1426 fail_stack.frame = 0; \
1429 # define INIT_FAIL_STACK() \
1431 fail_stack.avail = 0; \
1432 fail_stack.frame = 0; \
1435 # define RETALLOC_IF(addr, n, t) \
1436 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1440 /* Double the size of FAIL_STACK, up to a limit
1441 which allows approximately `re_max_failures' items.
1443 Return 1 if succeeds, and 0 if either ran out of memory
1444 allocating space for it or it was already too large.
1446 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1448 /* Factor to increase the failure stack size by
1449 when we increase it.
1450 This used to be 2, but 2 was too wasteful
1451 because the old discarded stacks added up to as much space
1452 were as ultimate, maximum-size stack. */
1453 #define FAIL_STACK_GROWTH_FACTOR 4
1455 #define GROW_FAIL_STACK(fail_stack) \
1456 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1457 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1459 : ((fail_stack).stack \
1460 = (fail_stack_elt_t *) \
1461 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1462 (fail_stack).size * sizeof (fail_stack_elt_t), \
1463 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1464 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1465 * FAIL_STACK_GROWTH_FACTOR))), \
1467 (fail_stack).stack == NULL \
1469 : ((fail_stack).size \
1470 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1471 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1472 * FAIL_STACK_GROWTH_FACTOR)) \
1473 / sizeof (fail_stack_elt_t)), \
1477 /* Push a pointer value onto the failure stack.
1478 Assumes the variable `fail_stack'. Probably should only
1479 be called from within `PUSH_FAILURE_POINT'. */
1480 #define PUSH_FAILURE_POINTER(item) \
1481 fail_stack.stack[fail_stack.avail++].pointer = (item)
1483 /* This pushes an integer-valued item onto the failure stack.
1484 Assumes the variable `fail_stack'. Probably should only
1485 be called from within `PUSH_FAILURE_POINT'. */
1486 #define PUSH_FAILURE_INT(item) \
1487 fail_stack.stack[fail_stack.avail++].integer = (item)
1489 /* These POP... operations complement the PUSH... operations.
1490 All assume that `fail_stack' is nonempty. */
1491 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1492 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1494 /* Individual items aside from the registers. */
1495 #define NUM_NONREG_ITEMS 3
1497 /* Used to examine the stack (to detect infinite loops). */
1498 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1499 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1500 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1501 #define TOP_FAILURE_HANDLE() fail_stack.frame
1504 #define ENSURE_FAIL_STACK(space) \
1505 while (REMAINING_AVAIL_SLOTS <= space) { \
1506 if (!GROW_FAIL_STACK (fail_stack)) \
1508 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1509 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1512 /* Push register NUM onto the stack. */
1513 #define PUSH_FAILURE_REG(num) \
1515 char *destination; \
1516 ENSURE_FAIL_STACK(3); \
1517 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1518 num, regstart[num], regend[num]); \
1519 PUSH_FAILURE_POINTER (regstart[num]); \
1520 PUSH_FAILURE_POINTER (regend[num]); \
1521 PUSH_FAILURE_INT (num); \
1524 /* Change the counter's value to VAL, but make sure that it will
1525 be reset when backtracking. */
1526 #define PUSH_NUMBER(ptr,val) \
1528 char *destination; \
1530 ENSURE_FAIL_STACK(3); \
1531 EXTRACT_NUMBER (c, ptr); \
1532 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1533 PUSH_FAILURE_INT (c); \
1534 PUSH_FAILURE_POINTER (ptr); \
1535 PUSH_FAILURE_INT (-1); \
1536 STORE_NUMBER (ptr, val); \
1539 /* Pop a saved register off the stack. */
1540 #define POP_FAILURE_REG_OR_COUNT() \
1542 int pfreg = POP_FAILURE_INT (); \
1545 /* It's a counter. */ \
1546 /* Here, we discard `const', making re_match non-reentrant. */ \
1547 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1548 pfreg = POP_FAILURE_INT (); \
1549 STORE_NUMBER (ptr, pfreg); \
1550 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, pfreg); \
1554 regend[pfreg] = POP_FAILURE_POINTER (); \
1555 regstart[pfreg] = POP_FAILURE_POINTER (); \
1556 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1557 pfreg, regstart[pfreg], regend[pfreg]); \
1561 /* Check that we are not stuck in an infinite loop. */
1562 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1564 int failure = TOP_FAILURE_HANDLE (); \
1565 /* Check for infinite matching loops */ \
1566 while (failure > 0 \
1567 && (FAILURE_STR (failure) == string_place \
1568 || FAILURE_STR (failure) == NULL)) \
1570 assert (FAILURE_PAT (failure) >= bufp->buffer \
1571 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1572 if (FAILURE_PAT (failure) == pat_cur) \
1577 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1578 failure = NEXT_FAILURE_HANDLE(failure); \
1580 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1583 /* Push the information about the state we will need
1584 if we ever fail back to it.
1586 Requires variables fail_stack, regstart, regend and
1587 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1590 Does `return FAILURE_CODE' if runs out of memory. */
1592 #define PUSH_FAILURE_POINT(pattern, string_place) \
1594 char *destination; \
1595 /* Must be int, so when we don't save any registers, the arithmetic \
1596 of 0 + -1 isn't done as unsigned. */ \
1598 DEBUG_STATEMENT (nfailure_points_pushed++); \
1599 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1600 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1601 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1603 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1605 DEBUG_PRINT1 ("\n"); \
1607 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1608 PUSH_FAILURE_INT (fail_stack.frame); \
1610 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1611 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1612 DEBUG_PRINT1 ("'\n"); \
1613 PUSH_FAILURE_POINTER (string_place); \
1615 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1616 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1617 PUSH_FAILURE_POINTER (pattern); \
1619 /* Close the frame by moving the frame pointer past it. */ \
1620 fail_stack.frame = fail_stack.avail; \
1623 /* Estimate the size of data pushed by a typical failure stack entry.
1624 An estimate is all we need, because all we use this for
1625 is to choose a limit for how big to make the failure stack. */
1626 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1627 #define TYPICAL_FAILURE_SIZE 20
1629 /* How many items can still be added to the stack without overflowing it. */
1630 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1633 /* Pops what PUSH_FAIL_STACK pushes.
1635 We restore into the parameters, all of which should be lvalues:
1636 STR -- the saved data position.
1637 PAT -- the saved pattern position.
1638 REGSTART, REGEND -- arrays of string positions.
1640 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1641 `pend', `string1', `size1', `string2', and `size2'. */
1643 #define POP_FAILURE_POINT(str, pat) \
1645 assert (!FAIL_STACK_EMPTY ()); \
1647 /* Remove failure points and point to how many regs pushed. */ \
1648 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1649 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1650 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1652 /* Pop the saved registers. */ \
1653 while (fail_stack.frame < fail_stack.avail) \
1654 POP_FAILURE_REG_OR_COUNT (); \
1656 pat = POP_FAILURE_POINTER (); \
1657 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1658 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1660 /* If the saved string location is NULL, it came from an \
1661 on_failure_keep_string_jump opcode, and we want to throw away the \
1662 saved NULL, thus retaining our current position in the string. */ \
1663 str = POP_FAILURE_POINTER (); \
1664 DEBUG_PRINT2 (" Popping string %p: `", str); \
1665 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1666 DEBUG_PRINT1 ("'\n"); \
1668 fail_stack.frame = POP_FAILURE_INT (); \
1669 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1671 assert (fail_stack.avail >= 0); \
1672 assert (fail_stack.frame <= fail_stack.avail); \
1674 DEBUG_STATEMENT (nfailure_points_popped++); \
1675 } while (0) /* POP_FAILURE_POINT */
1679 /* Registers are set to a sentinel when they haven't yet matched. */
1680 #define REG_UNSET(e) ((e) == NULL)
1682 /* Subroutine declarations and macros for regex_compile. */
1684 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1685 reg_syntax_t syntax
,
1686 struct re_pattern_buffer
*bufp
));
1687 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1688 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1689 int arg1
, int arg2
));
1690 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1691 int arg
, unsigned char *end
));
1692 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1693 int arg1
, int arg2
, unsigned char *end
));
1694 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1696 reg_syntax_t syntax
));
1697 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1699 reg_syntax_t syntax
));
1700 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1701 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1702 char *fastmap
, const int multibyte
));
1704 /* Fetch the next character in the uncompiled pattern, with no
1706 #define PATFETCH(c) \
1709 if (p == pend) return REG_EEND; \
1710 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1715 /* If `translate' is non-null, return translate[D], else just D. We
1716 cast the subscript to translate because some data is declared as
1717 `char *', to avoid warnings when a string constant is passed. But
1718 when we use a character as a subscript we must make it unsigned. */
1720 # define TRANSLATE(d) \
1721 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1725 /* Macros for outputting the compiled pattern into `buffer'. */
1727 /* If the buffer isn't allocated when it comes in, use this. */
1728 #define INIT_BUF_SIZE 32
1730 /* Make sure we have at least N more bytes of space in buffer. */
1731 #define GET_BUFFER_SPACE(n) \
1732 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1735 /* Make sure we have one more byte of buffer space and then add C to it. */
1736 #define BUF_PUSH(c) \
1738 GET_BUFFER_SPACE (1); \
1739 *b++ = (unsigned char) (c); \
1743 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1744 #define BUF_PUSH_2(c1, c2) \
1746 GET_BUFFER_SPACE (2); \
1747 *b++ = (unsigned char) (c1); \
1748 *b++ = (unsigned char) (c2); \
1752 /* Store a jump with opcode OP at LOC to location TO. We store a
1753 relative address offset by the three bytes the jump itself occupies. */
1754 #define STORE_JUMP(op, loc, to) \
1755 store_op1 (op, loc, (to) - (loc) - 3)
1757 /* Likewise, for a two-argument jump. */
1758 #define STORE_JUMP2(op, loc, to, arg) \
1759 store_op2 (op, loc, (to) - (loc) - 3, arg)
1761 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1762 #define INSERT_JUMP(op, loc, to) \
1763 insert_op1 (op, loc, (to) - (loc) - 3, b)
1765 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1766 #define INSERT_JUMP2(op, loc, to, arg) \
1767 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1770 /* This is not an arbitrary limit: the arguments which represent offsets
1771 into the pattern are two bytes long. So if 2^15 bytes turns out to
1772 be too small, many things would have to change. */
1773 # define MAX_BUF_SIZE (1L << 15)
1775 #if 0 /* This is when we thought it could be 2^16 bytes. */
1776 /* Any other compiler which, like MSC, has allocation limit below 2^16
1777 bytes will have to use approach similar to what was done below for
1778 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1779 reallocating to 0 bytes. Such thing is not going to work too well.
1780 You have been warned!! */
1781 #if defined _MSC_VER && !defined WIN32
1782 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1783 # define MAX_BUF_SIZE 65500L
1785 # define MAX_BUF_SIZE (1L << 16)
1789 /* Extend the buffer by twice its current size via realloc and
1790 reset the pointers that pointed into the old block to point to the
1791 correct places in the new one. If extending the buffer results in it
1792 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1793 #if __BOUNDED_POINTERS__
1794 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1795 # define MOVE_BUFFER_POINTER(P) \
1796 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1797 SET_HIGH_BOUND (P), \
1798 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1799 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1802 SET_HIGH_BOUND (b); \
1803 SET_HIGH_BOUND (begalt); \
1804 if (fixup_alt_jump) \
1805 SET_HIGH_BOUND (fixup_alt_jump); \
1807 SET_HIGH_BOUND (laststart); \
1808 if (pending_exact) \
1809 SET_HIGH_BOUND (pending_exact); \
1812 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1813 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1815 #define EXTEND_BUFFER() \
1817 unsigned char *old_buffer = bufp->buffer; \
1818 if (bufp->allocated == MAX_BUF_SIZE) \
1820 bufp->allocated <<= 1; \
1821 if (bufp->allocated > MAX_BUF_SIZE) \
1822 bufp->allocated = MAX_BUF_SIZE; \
1823 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1824 if (bufp->buffer == NULL) \
1825 return REG_ESPACE; \
1826 /* If the buffer moved, move all the pointers into it. */ \
1827 if (old_buffer != bufp->buffer) \
1829 unsigned char *new_buffer = bufp->buffer; \
1830 MOVE_BUFFER_POINTER (b); \
1831 MOVE_BUFFER_POINTER (begalt); \
1832 if (fixup_alt_jump) \
1833 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1835 MOVE_BUFFER_POINTER (laststart); \
1836 if (pending_exact) \
1837 MOVE_BUFFER_POINTER (pending_exact); \
1839 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1843 /* Since we have one byte reserved for the register number argument to
1844 {start,stop}_memory, the maximum number of groups we can report
1845 things about is what fits in that byte. */
1846 #define MAX_REGNUM 255
1848 /* But patterns can have more than `MAX_REGNUM' registers. We just
1849 ignore the excess. */
1850 typedef int regnum_t
;
1853 /* Macros for the compile stack. */
1855 /* Since offsets can go either forwards or backwards, this type needs to
1856 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1857 /* int may be not enough when sizeof(int) == 2. */
1858 typedef long pattern_offset_t
;
1862 pattern_offset_t begalt_offset
;
1863 pattern_offset_t fixup_alt_jump
;
1864 pattern_offset_t laststart_offset
;
1866 } compile_stack_elt_t
;
1871 compile_stack_elt_t
*stack
;
1873 unsigned avail
; /* Offset of next open position. */
1874 } compile_stack_type
;
1877 #define INIT_COMPILE_STACK_SIZE 32
1879 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1880 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1882 /* The next available element. */
1883 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1885 /* Explicit quit checking is only used on NTemacs and whenever we
1886 use polling to process input events. */
1887 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1888 extern int immediate_quit
;
1889 # define IMMEDIATE_QUIT_CHECK \
1891 if (immediate_quit) QUIT; \
1894 # define IMMEDIATE_QUIT_CHECK ((void)0)
1897 /* Structure to manage work area for range table. */
1898 struct range_table_work_area
1900 int *table
; /* actual work area. */
1901 int allocated
; /* allocated size for work area in bytes. */
1902 int used
; /* actually used size in words. */
1903 int bits
; /* flag to record character classes */
1906 /* Make sure that WORK_AREA can hold more N multibyte characters.
1907 This is used only in set_image_of_range and set_image_of_range_1.
1908 It expects WORK_AREA to be a pointer.
1909 If it can't get the space, it returns from the surrounding function. */
1911 #define EXTEND_RANGE_TABLE(work_area, n) \
1913 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1915 extend_range_table_work_area (&work_area); \
1916 if ((work_area).table == 0) \
1917 return (REG_ESPACE); \
1921 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1922 (work_area).bits |= (bit)
1924 /* Bits used to implement the multibyte-part of the various character classes
1925 such as [:alnum:] in a charset's range table. */
1926 #define BIT_WORD 0x1
1927 #define BIT_LOWER 0x2
1928 #define BIT_PUNCT 0x4
1929 #define BIT_SPACE 0x8
1930 #define BIT_UPPER 0x10
1931 #define BIT_MULTIBYTE 0x20
1933 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1934 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1936 EXTEND_RANGE_TABLE ((work_area), 2); \
1937 (work_area).table[(work_area).used++] = (range_start); \
1938 (work_area).table[(work_area).used++] = (range_end); \
1941 /* Free allocated memory for WORK_AREA. */
1942 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1944 if ((work_area).table) \
1945 free ((work_area).table); \
1948 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1949 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1950 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1951 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1954 /* Set the bit for character C in a list. */
1955 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1960 /* Store characters in the range FROM to TO in the bitmap at B (for
1961 ASCII and unibyte characters) and WORK_AREA (for multibyte
1962 characters) while translating them and paying attention to the
1963 continuity of translated characters.
1965 Implementation note: It is better to implement these fairly big
1966 macros by a function, but it's not that easy because macros called
1967 in this macro assume various local variables already declared. */
1969 /* Both FROM and TO are ASCII characters. */
1971 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1975 for (C0 = (FROM); C0 <= (TO); C0++) \
1977 C1 = TRANSLATE (C0); \
1978 if (! ASCII_CHAR_P (C1)) \
1980 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1981 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1984 SET_LIST_BIT (C1); \
1989 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1991 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1993 int C0, C1, C2, I; \
1994 int USED = RANGE_TABLE_WORK_USED (work_area); \
1996 for (C0 = (FROM); C0 <= (TO); C0++) \
1998 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1999 if (CHAR_BYTE8_P (C1)) \
2000 SET_LIST_BIT (C0); \
2003 C2 = TRANSLATE (C1); \
2005 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2007 SET_LIST_BIT (C1); \
2008 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2010 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2011 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2013 if (C2 >= from - 1 && C2 <= to + 1) \
2015 if (C2 == from - 1) \
2016 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2017 else if (C2 == to + 1) \
2018 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2023 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2029 /* Both FROM and TO are multibyte characters. */
2031 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2033 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2035 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2036 for (C0 = (FROM); C0 <= (TO); C0++) \
2038 C1 = TRANSLATE (C0); \
2039 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2040 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2041 SET_LIST_BIT (C2); \
2042 if (C1 >= (FROM) && C1 <= (TO)) \
2044 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2046 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2047 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2049 if (C1 >= from - 1 && C1 <= to + 1) \
2051 if (C1 == from - 1) \
2052 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2053 else if (C1 == to + 1) \
2054 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2059 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2065 /* Get the next unsigned number in the uncompiled pattern. */
2066 #define GET_UNSIGNED_NUMBER(num) \
2069 FREE_STACK_RETURN (REG_EBRACE); \
2073 while ('0' <= c && c <= '9') \
2079 num = num * 10 + c - '0'; \
2080 if (num / 10 != prev) \
2081 FREE_STACK_RETURN (REG_BADBR); \
2083 FREE_STACK_RETURN (REG_EBRACE); \
2089 #if ! WIDE_CHAR_SUPPORT
2091 /* Map a string to the char class it names (if any). */
2093 re_wctype (const re_char
*str
)
2095 const char *string
= (const char *) str
;
2096 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2097 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2098 else if (STREQ (string
, "word")) return RECC_WORD
;
2099 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2100 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2101 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2102 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2103 else if (STREQ (string
, "print")) return RECC_PRINT
;
2104 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2105 else if (STREQ (string
, "space")) return RECC_SPACE
;
2106 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2107 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2108 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2109 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2110 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2111 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2112 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2116 /* True if CH is in the char class CC. */
2118 re_iswctype (int ch
, re_wctype_t cc
)
2122 case RECC_ALNUM
: return ISALNUM (ch
);
2123 case RECC_ALPHA
: return ISALPHA (ch
);
2124 case RECC_BLANK
: return ISBLANK (ch
);
2125 case RECC_CNTRL
: return ISCNTRL (ch
);
2126 case RECC_DIGIT
: return ISDIGIT (ch
);
2127 case RECC_GRAPH
: return ISGRAPH (ch
);
2128 case RECC_LOWER
: return ISLOWER (ch
);
2129 case RECC_PRINT
: return ISPRINT (ch
);
2130 case RECC_PUNCT
: return ISPUNCT (ch
);
2131 case RECC_SPACE
: return ISSPACE (ch
);
2132 case RECC_UPPER
: return ISUPPER (ch
);
2133 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2134 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2135 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2136 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2137 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2138 case RECC_WORD
: return ISWORD (ch
);
2139 case RECC_ERROR
: return false;
2145 /* Return a bit-pattern to use in the range-table bits to match multibyte
2146 chars of class CC. */
2148 re_wctype_to_bit (re_wctype_t cc
)
2152 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2153 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2154 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2155 case RECC_LOWER
: return BIT_LOWER
;
2156 case RECC_UPPER
: return BIT_UPPER
;
2157 case RECC_PUNCT
: return BIT_PUNCT
;
2158 case RECC_SPACE
: return BIT_SPACE
;
2159 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2160 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2167 /* Filling in the work area of a range. */
2169 /* Actually extend the space in WORK_AREA. */
2172 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2174 work_area
->allocated
+= 16 * sizeof (int);
2175 if (work_area
->table
)
2177 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2180 = (int *) malloc (work_area
->allocated
);
2186 /* Carefully find the ranges of codes that are equivalent
2187 under case conversion to the range start..end when passed through
2188 TRANSLATE. Handle the case where non-letters can come in between
2189 two upper-case letters (which happens in Latin-1).
2190 Also handle the case of groups of more than 2 case-equivalent chars.
2192 The basic method is to look at consecutive characters and see
2193 if they can form a run that can be handled as one.
2195 Returns -1 if successful, REG_ESPACE if ran out of space. */
2198 set_image_of_range_1 (work_area
, start
, end
, translate
)
2199 RE_TRANSLATE_TYPE translate
;
2200 struct range_table_work_area
*work_area
;
2201 re_wchar_t start
, end
;
2203 /* `one_case' indicates a character, or a run of characters,
2204 each of which is an isolate (no case-equivalents).
2205 This includes all ASCII non-letters.
2207 `two_case' indicates a character, or a run of characters,
2208 each of which has two case-equivalent forms.
2209 This includes all ASCII letters.
2211 `strange' indicates a character that has more than one
2214 enum case_type
{one_case
, two_case
, strange
};
2216 /* Describe the run that is in progress,
2217 which the next character can try to extend.
2218 If run_type is strange, that means there really is no run.
2219 If run_type is one_case, then run_start...run_end is the run.
2220 If run_type is two_case, then the run is run_start...run_end,
2221 and the case-equivalents end at run_eqv_end. */
2223 enum case_type run_type
= strange
;
2224 int run_start
, run_end
, run_eqv_end
;
2226 Lisp_Object eqv_table
;
2228 if (!RE_TRANSLATE_P (translate
))
2230 EXTEND_RANGE_TABLE (work_area
, 2);
2231 work_area
->table
[work_area
->used
++] = (start
);
2232 work_area
->table
[work_area
->used
++] = (end
);
2236 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2238 for (; start
<= end
; start
++)
2240 enum case_type this_type
;
2241 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2242 int minchar
, maxchar
;
2244 /* Classify this character */
2246 this_type
= one_case
;
2247 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2248 this_type
= two_case
;
2250 this_type
= strange
;
2253 minchar
= start
, maxchar
= eqv
;
2255 minchar
= eqv
, maxchar
= start
;
2257 /* Can this character extend the run in progress? */
2258 if (this_type
== strange
|| this_type
!= run_type
2259 || !(minchar
== run_end
+ 1
2260 && (run_type
== two_case
2261 ? maxchar
== run_eqv_end
+ 1 : 1)))
2264 Record each of its equivalent ranges. */
2265 if (run_type
== one_case
)
2267 EXTEND_RANGE_TABLE (work_area
, 2);
2268 work_area
->table
[work_area
->used
++] = run_start
;
2269 work_area
->table
[work_area
->used
++] = run_end
;
2271 else if (run_type
== two_case
)
2273 EXTEND_RANGE_TABLE (work_area
, 4);
2274 work_area
->table
[work_area
->used
++] = run_start
;
2275 work_area
->table
[work_area
->used
++] = run_end
;
2276 work_area
->table
[work_area
->used
++]
2277 = RE_TRANSLATE (eqv_table
, run_start
);
2278 work_area
->table
[work_area
->used
++]
2279 = RE_TRANSLATE (eqv_table
, run_end
);
2284 if (this_type
== strange
)
2286 /* For a strange character, add each of its equivalents, one
2287 by one. Don't start a range. */
2290 EXTEND_RANGE_TABLE (work_area
, 2);
2291 work_area
->table
[work_area
->used
++] = eqv
;
2292 work_area
->table
[work_area
->used
++] = eqv
;
2293 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2295 while (eqv
!= start
);
2298 /* Add this char to the run, or start a new run. */
2299 else if (run_type
== strange
)
2301 /* Initialize a new range. */
2302 run_type
= this_type
;
2305 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2309 /* Extend a running range. */
2311 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2315 /* If a run is still in progress at the end, finish it now
2316 by recording its equivalent ranges. */
2317 if (run_type
== one_case
)
2319 EXTEND_RANGE_TABLE (work_area
, 2);
2320 work_area
->table
[work_area
->used
++] = run_start
;
2321 work_area
->table
[work_area
->used
++] = run_end
;
2323 else if (run_type
== two_case
)
2325 EXTEND_RANGE_TABLE (work_area
, 4);
2326 work_area
->table
[work_area
->used
++] = run_start
;
2327 work_area
->table
[work_area
->used
++] = run_end
;
2328 work_area
->table
[work_area
->used
++]
2329 = RE_TRANSLATE (eqv_table
, run_start
);
2330 work_area
->table
[work_area
->used
++]
2331 = RE_TRANSLATE (eqv_table
, run_end
);
2339 /* Record the image of the range start..end when passed through
2340 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2341 and is not even necessarily contiguous.
2342 Normally we approximate it with the smallest contiguous range that contains
2343 all the chars we need. However, for Latin-1 we go to extra effort
2346 This function is not called for ASCII ranges.
2348 Returns -1 if successful, REG_ESPACE if ran out of space. */
2351 set_image_of_range (work_area
, start
, end
, translate
)
2352 RE_TRANSLATE_TYPE translate
;
2353 struct range_table_work_area
*work_area
;
2354 re_wchar_t start
, end
;
2356 re_wchar_t cmin
, cmax
;
2359 /* For Latin-1 ranges, use set_image_of_range_1
2360 to get proper handling of ranges that include letters and nonletters.
2361 For a range that includes the whole of Latin-1, this is not necessary.
2362 For other character sets, we don't bother to get this right. */
2363 if (RE_TRANSLATE_P (translate
) && start
< 04400
2364 && !(start
< 04200 && end
>= 04377))
2371 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2381 EXTEND_RANGE_TABLE (work_area
, 2);
2382 work_area
->table
[work_area
->used
++] = (start
);
2383 work_area
->table
[work_area
->used
++] = (end
);
2385 cmin
= -1, cmax
= -1;
2387 if (RE_TRANSLATE_P (translate
))
2391 for (ch
= start
; ch
<= end
; ch
++)
2393 re_wchar_t c
= TRANSLATE (ch
);
2394 if (! (start
<= c
&& c
<= end
))
2400 cmin
= MIN (cmin
, c
);
2401 cmax
= MAX (cmax
, c
);
2408 EXTEND_RANGE_TABLE (work_area
, 2);
2409 work_area
->table
[work_area
->used
++] = (cmin
);
2410 work_area
->table
[work_area
->used
++] = (cmax
);
2418 #ifndef MATCH_MAY_ALLOCATE
2420 /* If we cannot allocate large objects within re_match_2_internal,
2421 we make the fail stack and register vectors global.
2422 The fail stack, we grow to the maximum size when a regexp
2424 The register vectors, we adjust in size each time we
2425 compile a regexp, according to the number of registers it needs. */
2427 static fail_stack_type fail_stack
;
2429 /* Size with which the following vectors are currently allocated.
2430 That is so we can make them bigger as needed,
2431 but never make them smaller. */
2432 static int regs_allocated_size
;
2434 static re_char
** regstart
, ** regend
;
2435 static re_char
**best_regstart
, **best_regend
;
2437 /* Make the register vectors big enough for NUM_REGS registers,
2438 but don't make them smaller. */
2441 regex_grow_registers (num_regs
)
2444 if (num_regs
> regs_allocated_size
)
2446 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2447 RETALLOC_IF (regend
, num_regs
, re_char
*);
2448 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2449 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2451 regs_allocated_size
= num_regs
;
2455 #endif /* not MATCH_MAY_ALLOCATE */
2457 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2461 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2462 Returns one of error codes defined in `regex.h', or zero for success.
2464 Assumes the `allocated' (and perhaps `buffer') and `translate'
2465 fields are set in BUFP on entry.
2467 If it succeeds, results are put in BUFP (if it returns an error, the
2468 contents of BUFP are undefined):
2469 `buffer' is the compiled pattern;
2470 `syntax' is set to SYNTAX;
2471 `used' is set to the length of the compiled pattern;
2472 `fastmap_accurate' is zero;
2473 `re_nsub' is the number of subexpressions in PATTERN;
2474 `not_bol' and `not_eol' are zero;
2476 The `fastmap' field is neither examined nor set. */
2478 /* Insert the `jump' from the end of last alternative to "here".
2479 The space for the jump has already been allocated. */
2480 #define FIXUP_ALT_JUMP() \
2482 if (fixup_alt_jump) \
2483 STORE_JUMP (jump, fixup_alt_jump, b); \
2487 /* Return, freeing storage we allocated. */
2488 #define FREE_STACK_RETURN(value) \
2490 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2491 free (compile_stack.stack); \
2495 static reg_errcode_t
2496 regex_compile (const re_char
*pattern
, size_t size
, reg_syntax_t syntax
, struct re_pattern_buffer
*bufp
)
2498 /* We fetch characters from PATTERN here. */
2499 register re_wchar_t c
, c1
;
2501 /* Points to the end of the buffer, where we should append. */
2502 register unsigned char *b
;
2504 /* Keeps track of unclosed groups. */
2505 compile_stack_type compile_stack
;
2507 /* Points to the current (ending) position in the pattern. */
2509 /* `const' makes AIX compiler fail. */
2510 unsigned char *p
= pattern
;
2512 re_char
*p
= pattern
;
2514 re_char
*pend
= pattern
+ size
;
2516 /* How to translate the characters in the pattern. */
2517 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2519 /* Address of the count-byte of the most recently inserted `exactn'
2520 command. This makes it possible to tell if a new exact-match
2521 character can be added to that command or if the character requires
2522 a new `exactn' command. */
2523 unsigned char *pending_exact
= 0;
2525 /* Address of start of the most recently finished expression.
2526 This tells, e.g., postfix * where to find the start of its
2527 operand. Reset at the beginning of groups and alternatives. */
2528 unsigned char *laststart
= 0;
2530 /* Address of beginning of regexp, or inside of last group. */
2531 unsigned char *begalt
;
2533 /* Place in the uncompiled pattern (i.e., the {) to
2534 which to go back if the interval is invalid. */
2535 re_char
*beg_interval
;
2537 /* Address of the place where a forward jump should go to the end of
2538 the containing expression. Each alternative of an `or' -- except the
2539 last -- ends with a forward jump of this sort. */
2540 unsigned char *fixup_alt_jump
= 0;
2542 /* Work area for range table of charset. */
2543 struct range_table_work_area range_table_work
;
2545 /* If the object matched can contain multibyte characters. */
2546 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2548 /* Nonzero if we have pushed down into a subpattern. */
2549 int in_subpattern
= 0;
2551 /* These hold the values of p, pattern, and pend from the main
2552 pattern when we have pushed into a subpattern. */
2553 re_char
*main_p
IF_LINT (= NULL
);
2554 re_char
*main_pattern
IF_LINT (= NULL
);
2555 re_char
*main_pend
IF_LINT (= NULL
);
2559 DEBUG_PRINT1 ("\nCompiling pattern: ");
2562 unsigned debug_count
;
2564 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2565 putchar (pattern
[debug_count
]);
2570 /* Initialize the compile stack. */
2571 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2572 if (compile_stack
.stack
== NULL
)
2575 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2576 compile_stack
.avail
= 0;
2578 range_table_work
.table
= 0;
2579 range_table_work
.allocated
= 0;
2581 /* Initialize the pattern buffer. */
2582 bufp
->syntax
= syntax
;
2583 bufp
->fastmap_accurate
= 0;
2584 bufp
->not_bol
= bufp
->not_eol
= 0;
2585 bufp
->used_syntax
= 0;
2587 /* Set `used' to zero, so that if we return an error, the pattern
2588 printer (for debugging) will think there's no pattern. We reset it
2592 /* Always count groups, whether or not bufp->no_sub is set. */
2595 #if !defined emacs && !defined SYNTAX_TABLE
2596 /* Initialize the syntax table. */
2597 init_syntax_once ();
2600 if (bufp
->allocated
== 0)
2603 { /* If zero allocated, but buffer is non-null, try to realloc
2604 enough space. This loses if buffer's address is bogus, but
2605 that is the user's responsibility. */
2606 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2609 { /* Caller did not allocate a buffer. Do it for them. */
2610 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2612 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2614 bufp
->allocated
= INIT_BUF_SIZE
;
2617 begalt
= b
= bufp
->buffer
;
2619 /* Loop through the uncompiled pattern until we're at the end. */
2624 /* If this is the end of an included regexp,
2625 pop back to the main regexp and try again. */
2629 pattern
= main_pattern
;
2634 /* If this is the end of the main regexp, we are done. */
2646 /* If there's no special whitespace regexp, treat
2647 spaces normally. And don't try to do this recursively. */
2648 if (!whitespace_regexp
|| in_subpattern
)
2651 /* Peek past following spaces. */
2658 /* If the spaces are followed by a repetition op,
2659 treat them normally. */
2661 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2662 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2665 /* Replace the spaces with the whitespace regexp. */
2669 main_pattern
= pattern
;
2670 p
= pattern
= whitespace_regexp
;
2671 pend
= p
+ strlen ((const char *) p
);
2677 if ( /* If at start of pattern, it's an operator. */
2679 /* If context independent, it's an operator. */
2680 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2681 /* Otherwise, depends on what's come before. */
2682 || at_begline_loc_p (pattern
, p
, syntax
))
2683 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2692 if ( /* If at end of pattern, it's an operator. */
2694 /* If context independent, it's an operator. */
2695 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2696 /* Otherwise, depends on what's next. */
2697 || at_endline_loc_p (p
, pend
, syntax
))
2698 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2707 if ((syntax
& RE_BK_PLUS_QM
)
2708 || (syntax
& RE_LIMITED_OPS
))
2712 /* If there is no previous pattern... */
2715 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2716 FREE_STACK_RETURN (REG_BADRPT
);
2717 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2722 /* 1 means zero (many) matches is allowed. */
2723 boolean zero_times_ok
= 0, many_times_ok
= 0;
2726 /* If there is a sequence of repetition chars, collapse it
2727 down to just one (the right one). We can't combine
2728 interval operators with these because of, e.g., `a{2}*',
2729 which should only match an even number of `a's. */
2733 if ((syntax
& RE_FRUGAL
)
2734 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2738 zero_times_ok
|= c
!= '+';
2739 many_times_ok
|= c
!= '?';
2745 || (!(syntax
& RE_BK_PLUS_QM
)
2746 && (*p
== '+' || *p
== '?')))
2748 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2751 FREE_STACK_RETURN (REG_EESCAPE
);
2752 if (p
[1] == '+' || p
[1] == '?')
2753 PATFETCH (c
); /* Gobble up the backslash. */
2759 /* If we get here, we found another repeat character. */
2763 /* Star, etc. applied to an empty pattern is equivalent
2764 to an empty pattern. */
2765 if (!laststart
|| laststart
== b
)
2768 /* Now we know whether or not zero matches is allowed
2769 and also whether or not two or more matches is allowed. */
2774 boolean simple
= skip_one_char (laststart
) == b
;
2775 unsigned int startoffset
= 0;
2777 /* Check if the loop can match the empty string. */
2778 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2779 ? on_failure_jump
: on_failure_jump_loop
;
2780 assert (skip_one_char (laststart
) <= b
);
2782 if (!zero_times_ok
&& simple
)
2783 { /* Since simple * loops can be made faster by using
2784 on_failure_keep_string_jump, we turn simple P+
2785 into PP* if P is simple. */
2786 unsigned char *p1
, *p2
;
2787 startoffset
= b
- laststart
;
2788 GET_BUFFER_SPACE (startoffset
);
2789 p1
= b
; p2
= laststart
;
2795 GET_BUFFER_SPACE (6);
2798 STORE_JUMP (ofj
, b
, b
+ 6);
2800 /* Simple * loops can use on_failure_keep_string_jump
2801 depending on what follows. But since we don't know
2802 that yet, we leave the decision up to
2803 on_failure_jump_smart. */
2804 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2805 laststart
+ startoffset
, b
+ 6);
2807 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2812 /* A simple ? pattern. */
2813 assert (zero_times_ok
);
2814 GET_BUFFER_SPACE (3);
2815 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2819 else /* not greedy */
2820 { /* I wish the greedy and non-greedy cases could be merged. */
2822 GET_BUFFER_SPACE (7); /* We might use less. */
2825 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2827 /* The non-greedy multiple match looks like
2828 a repeat..until: we only need a conditional jump
2829 at the end of the loop. */
2830 if (emptyp
) BUF_PUSH (no_op
);
2831 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2832 : on_failure_jump
, b
, laststart
);
2836 /* The repeat...until naturally matches one or more.
2837 To also match zero times, we need to first jump to
2838 the end of the loop (its conditional jump). */
2839 INSERT_JUMP (jump
, laststart
, b
);
2845 /* non-greedy a?? */
2846 INSERT_JUMP (jump
, laststart
, b
+ 3);
2848 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2867 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2869 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2871 /* Ensure that we have enough space to push a charset: the
2872 opcode, the length count, and the bitset; 34 bytes in all. */
2873 GET_BUFFER_SPACE (34);
2877 /* We test `*p == '^' twice, instead of using an if
2878 statement, so we only need one BUF_PUSH. */
2879 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2883 /* Remember the first position in the bracket expression. */
2886 /* Push the number of bytes in the bitmap. */
2887 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2889 /* Clear the whole map. */
2890 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2892 /* charset_not matches newline according to a syntax bit. */
2893 if ((re_opcode_t
) b
[-2] == charset_not
2894 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2895 SET_LIST_BIT ('\n');
2897 /* Read in characters and ranges, setting map bits. */
2900 boolean escaped_char
= false;
2901 const unsigned char *p2
= p
;
2904 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2906 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2907 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2908 So the translation is done later in a loop. Example:
2909 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2912 /* \ might escape characters inside [...] and [^...]. */
2913 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2915 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2918 escaped_char
= true;
2922 /* Could be the end of the bracket expression. If it's
2923 not (i.e., when the bracket expression is `[]' so
2924 far), the ']' character bit gets set way below. */
2925 if (c
== ']' && p2
!= p1
)
2929 /* See if we're at the beginning of a possible character
2932 if (!escaped_char
&&
2933 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2935 /* Leave room for the null. */
2936 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2937 const unsigned char *class_beg
;
2943 /* If pattern is `[[:'. */
2944 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2949 if ((c
== ':' && *p
== ']') || p
== pend
)
2951 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2954 /* This is in any case an invalid class name. */
2959 /* If isn't a word bracketed by `[:' and `:]':
2960 undo the ending character, the letters, and
2961 leave the leading `:' and `[' (but set bits for
2963 if (c
== ':' && *p
== ']')
2965 re_wctype_t cc
= re_wctype (str
);
2968 FREE_STACK_RETURN (REG_ECTYPE
);
2970 /* Throw away the ] at the end of the character
2974 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2977 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2978 if (re_iswctype (btowc (ch
), cc
))
2981 if (c
< (1 << BYTEWIDTH
))
2985 /* Most character classes in a multibyte match
2986 just set a flag. Exceptions are is_blank,
2987 is_digit, is_cntrl, and is_xdigit, since
2988 they can only match ASCII characters. We
2989 don't need to handle them for multibyte.
2990 They are distinguished by a negative wctype. */
2992 /* Setup the gl_state object to its buffer-defined
2993 value. This hardcodes the buffer-global
2994 syntax-table for ASCII chars, while the other chars
2995 will obey syntax-table properties. It's not ideal,
2996 but it's the way it's been done until now. */
2997 SETUP_BUFFER_SYNTAX_TABLE ();
2999 for (ch
= 0; ch
< 256; ++ch
)
3001 c
= RE_CHAR_TO_MULTIBYTE (ch
);
3002 if (! CHAR_BYTE8_P (c
)
3003 && re_iswctype (c
, cc
))
3009 if (ASCII_CHAR_P (c1
))
3011 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
3015 SET_RANGE_TABLE_WORK_AREA_BIT
3016 (range_table_work
, re_wctype_to_bit (cc
));
3018 /* In most cases the matching rule for char classes
3019 only uses the syntax table for multibyte chars,
3020 so that the content of the syntax-table it is not
3021 hardcoded in the range_table. SPACE and WORD are
3022 the two exceptions. */
3023 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
3024 bufp
->used_syntax
= 1;
3026 /* Repeat the loop. */
3031 /* Go back to right after the "[:". */
3035 /* Because the `:' may starts the range, we
3036 can't simply set bit and repeat the loop.
3037 Instead, just set it to C and handle below. */
3042 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
3045 /* Discard the `-'. */
3048 /* Fetch the character which ends the range. */
3051 if (CHAR_BYTE8_P (c1
)
3052 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
3053 /* Treat the range from a multibyte character to
3054 raw-byte character as empty. */
3059 /* Range from C to C. */
3064 if (syntax
& RE_NO_EMPTY_RANGES
)
3065 FREE_STACK_RETURN (REG_ERANGEX
);
3066 /* Else, repeat the loop. */
3071 /* Set the range into bitmap */
3072 for (; c
<= c1
; c
++)
3075 if (ch
< (1 << BYTEWIDTH
))
3082 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
3084 if (CHAR_BYTE8_P (c1
))
3085 c
= BYTE8_TO_CHAR (128);
3089 if (CHAR_BYTE8_P (c
))
3091 c
= CHAR_TO_BYTE8 (c
);
3092 c1
= CHAR_TO_BYTE8 (c1
);
3093 for (; c
<= c1
; c
++)
3098 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
3102 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
3109 /* Discard any (non)matching list bytes that are all 0 at the
3110 end of the map. Decrease the map-length byte too. */
3111 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3115 /* Build real range table from work area. */
3116 if (RANGE_TABLE_WORK_USED (range_table_work
)
3117 || RANGE_TABLE_WORK_BITS (range_table_work
))
3120 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3122 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3123 bytes for flags, two for COUNT, and three bytes for
3125 GET_BUFFER_SPACE (4 + used
* 3);
3127 /* Indicate the existence of range table. */
3128 laststart
[1] |= 0x80;
3130 /* Store the character class flag bits into the range table.
3131 If not in emacs, these flag bits are always 0. */
3132 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3133 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3135 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3136 for (i
= 0; i
< used
; i
++)
3137 STORE_CHARACTER_AND_INCR
3138 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3145 if (syntax
& RE_NO_BK_PARENS
)
3152 if (syntax
& RE_NO_BK_PARENS
)
3159 if (syntax
& RE_NEWLINE_ALT
)
3166 if (syntax
& RE_NO_BK_VBAR
)
3173 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3174 goto handle_interval
;
3180 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3182 /* Do not translate the character after the \, so that we can
3183 distinguish, e.g., \B from \b, even if we normally would
3184 translate, e.g., B to b. */
3190 if (syntax
& RE_NO_BK_PARENS
)
3191 goto normal_backslash
;
3196 regnum_t regnum
= 0;
3199 /* Look for a special (?...) construct */
3200 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3202 PATFETCH (c
); /* Gobble up the '?'. */
3208 case ':': shy
= 1; break;
3210 /* An explicitly specified regnum must start
3213 FREE_STACK_RETURN (REG_BADPAT
);
3214 case '1': case '2': case '3': case '4':
3215 case '5': case '6': case '7': case '8': case '9':
3216 regnum
= 10*regnum
+ (c
- '0'); break;
3218 /* Only (?:...) is supported right now. */
3219 FREE_STACK_RETURN (REG_BADPAT
);
3226 regnum
= ++bufp
->re_nsub
;
3228 { /* It's actually not shy, but explicitly numbered. */
3230 if (regnum
> bufp
->re_nsub
)
3231 bufp
->re_nsub
= regnum
;
3232 else if (regnum
> bufp
->re_nsub
3233 /* Ideally, we'd want to check that the specified
3234 group can't have matched (i.e. all subgroups
3235 using the same regnum are in other branches of
3236 OR patterns), but we don't currently keep track
3237 of enough info to do that easily. */
3238 || group_in_compile_stack (compile_stack
, regnum
))
3239 FREE_STACK_RETURN (REG_BADPAT
);
3242 /* It's really shy. */
3243 regnum
= - bufp
->re_nsub
;
3245 if (COMPILE_STACK_FULL
)
3247 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3248 compile_stack_elt_t
);
3249 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3251 compile_stack
.size
<<= 1;
3254 /* These are the values to restore when we hit end of this
3255 group. They are all relative offsets, so that if the
3256 whole pattern moves because of realloc, they will still
3258 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3259 COMPILE_STACK_TOP
.fixup_alt_jump
3260 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3261 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3262 COMPILE_STACK_TOP
.regnum
= regnum
;
3264 /* Do not push a start_memory for groups beyond the last one
3265 we can represent in the compiled pattern. */
3266 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3267 BUF_PUSH_2 (start_memory
, regnum
);
3269 compile_stack
.avail
++;
3274 /* If we've reached MAX_REGNUM groups, then this open
3275 won't actually generate any code, so we'll have to
3276 clear pending_exact explicitly. */
3282 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3284 if (COMPILE_STACK_EMPTY
)
3286 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3287 goto normal_backslash
;
3289 FREE_STACK_RETURN (REG_ERPAREN
);
3295 /* See similar code for backslashed left paren above. */
3296 if (COMPILE_STACK_EMPTY
)
3298 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3301 FREE_STACK_RETURN (REG_ERPAREN
);
3304 /* Since we just checked for an empty stack above, this
3305 ``can't happen''. */
3306 assert (compile_stack
.avail
!= 0);
3308 /* We don't just want to restore into `regnum', because
3309 later groups should continue to be numbered higher,
3310 as in `(ab)c(de)' -- the second group is #2. */
3313 compile_stack
.avail
--;
3314 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3316 = COMPILE_STACK_TOP
.fixup_alt_jump
3317 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3319 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3320 regnum
= COMPILE_STACK_TOP
.regnum
;
3321 /* If we've reached MAX_REGNUM groups, then this open
3322 won't actually generate any code, so we'll have to
3323 clear pending_exact explicitly. */
3326 /* We're at the end of the group, so now we know how many
3327 groups were inside this one. */
3328 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3329 BUF_PUSH_2 (stop_memory
, regnum
);
3334 case '|': /* `\|'. */
3335 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3336 goto normal_backslash
;
3338 if (syntax
& RE_LIMITED_OPS
)
3341 /* Insert before the previous alternative a jump which
3342 jumps to this alternative if the former fails. */
3343 GET_BUFFER_SPACE (3);
3344 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3348 /* The alternative before this one has a jump after it
3349 which gets executed if it gets matched. Adjust that
3350 jump so it will jump to this alternative's analogous
3351 jump (put in below, which in turn will jump to the next
3352 (if any) alternative's such jump, etc.). The last such
3353 jump jumps to the correct final destination. A picture:
3359 If we are at `b', then fixup_alt_jump right now points to a
3360 three-byte space after `a'. We'll put in the jump, set
3361 fixup_alt_jump to right after `b', and leave behind three
3362 bytes which we'll fill in when we get to after `c'. */
3366 /* Mark and leave space for a jump after this alternative,
3367 to be filled in later either by next alternative or
3368 when know we're at the end of a series of alternatives. */
3370 GET_BUFFER_SPACE (3);
3379 /* If \{ is a literal. */
3380 if (!(syntax
& RE_INTERVALS
)
3381 /* If we're at `\{' and it's not the open-interval
3383 || (syntax
& RE_NO_BK_BRACES
))
3384 goto normal_backslash
;
3388 /* If got here, then the syntax allows intervals. */
3390 /* At least (most) this many matches must be made. */
3391 int lower_bound
= 0, upper_bound
= -1;
3395 GET_UNSIGNED_NUMBER (lower_bound
);
3398 GET_UNSIGNED_NUMBER (upper_bound
);
3400 /* Interval such as `{1}' => match exactly once. */
3401 upper_bound
= lower_bound
;
3403 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3404 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3405 FREE_STACK_RETURN (REG_BADBR
);
3407 if (!(syntax
& RE_NO_BK_BRACES
))
3410 FREE_STACK_RETURN (REG_BADBR
);
3412 FREE_STACK_RETURN (REG_EESCAPE
);
3417 FREE_STACK_RETURN (REG_BADBR
);
3419 /* We just parsed a valid interval. */
3421 /* If it's invalid to have no preceding re. */
3424 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3425 FREE_STACK_RETURN (REG_BADRPT
);
3426 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3429 goto unfetch_interval
;
3432 if (upper_bound
== 0)
3433 /* If the upper bound is zero, just drop the sub pattern
3436 else if (lower_bound
== 1 && upper_bound
== 1)
3437 /* Just match it once: nothing to do here. */
3440 /* Otherwise, we have a nontrivial interval. When
3441 we're all done, the pattern will look like:
3442 set_number_at <jump count> <upper bound>
3443 set_number_at <succeed_n count> <lower bound>
3444 succeed_n <after jump addr> <succeed_n count>
3446 jump_n <succeed_n addr> <jump count>
3447 (The upper bound and `jump_n' are omitted if
3448 `upper_bound' is 1, though.) */
3450 { /* If the upper bound is > 1, we need to insert
3451 more at the end of the loop. */
3452 unsigned int nbytes
= (upper_bound
< 0 ? 3
3453 : upper_bound
> 1 ? 5 : 0);
3454 unsigned int startoffset
= 0;
3456 GET_BUFFER_SPACE (20); /* We might use less. */
3458 if (lower_bound
== 0)
3460 /* A succeed_n that starts with 0 is really a
3461 a simple on_failure_jump_loop. */
3462 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3468 /* Initialize lower bound of the `succeed_n', even
3469 though it will be set during matching by its
3470 attendant `set_number_at' (inserted next),
3471 because `re_compile_fastmap' needs to know.
3472 Jump to the `jump_n' we might insert below. */
3473 INSERT_JUMP2 (succeed_n
, laststart
,
3478 /* Code to initialize the lower bound. Insert
3479 before the `succeed_n'. The `5' is the last two
3480 bytes of this `set_number_at', plus 3 bytes of
3481 the following `succeed_n'. */
3482 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3487 if (upper_bound
< 0)
3489 /* A negative upper bound stands for infinity,
3490 in which case it degenerates to a plain jump. */
3491 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3494 else if (upper_bound
> 1)
3495 { /* More than one repetition is allowed, so
3496 append a backward jump to the `succeed_n'
3497 that starts this interval.
3499 When we've reached this during matching,
3500 we'll have matched the interval once, so
3501 jump back only `upper_bound - 1' times. */
3502 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3506 /* The location we want to set is the second
3507 parameter of the `jump_n'; that is `b-2' as
3508 an absolute address. `laststart' will be
3509 the `set_number_at' we're about to insert;
3510 `laststart+3' the number to set, the source
3511 for the relative address. But we are
3512 inserting into the middle of the pattern --
3513 so everything is getting moved up by 5.
3514 Conclusion: (b - 2) - (laststart + 3) + 5,
3515 i.e., b - laststart.
3517 We insert this at the beginning of the loop
3518 so that if we fail during matching, we'll
3519 reinitialize the bounds. */
3520 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3521 upper_bound
- 1, b
);
3526 beg_interval
= NULL
;
3531 /* If an invalid interval, match the characters as literals. */
3532 assert (beg_interval
);
3534 beg_interval
= NULL
;
3536 /* normal_char and normal_backslash need `c'. */
3539 if (!(syntax
& RE_NO_BK_BRACES
))
3541 assert (p
> pattern
&& p
[-1] == '\\');
3542 goto normal_backslash
;
3548 /* There is no way to specify the before_dot and after_dot
3549 operators. rms says this is ok. --karl */
3557 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3563 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3569 BUF_PUSH_2 (categoryspec
, c
);
3575 BUF_PUSH_2 (notcategoryspec
, c
);
3581 if (syntax
& RE_NO_GNU_OPS
)
3584 BUF_PUSH_2 (syntaxspec
, Sword
);
3589 if (syntax
& RE_NO_GNU_OPS
)
3592 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3597 if (syntax
& RE_NO_GNU_OPS
)
3603 if (syntax
& RE_NO_GNU_OPS
)
3609 if (syntax
& RE_NO_GNU_OPS
)
3618 FREE_STACK_RETURN (REG_BADPAT
);
3622 if (syntax
& RE_NO_GNU_OPS
)
3624 BUF_PUSH (wordbound
);
3628 if (syntax
& RE_NO_GNU_OPS
)
3630 BUF_PUSH (notwordbound
);
3634 if (syntax
& RE_NO_GNU_OPS
)
3640 if (syntax
& RE_NO_GNU_OPS
)
3645 case '1': case '2': case '3': case '4': case '5':
3646 case '6': case '7': case '8': case '9':
3650 if (syntax
& RE_NO_BK_REFS
)
3651 goto normal_backslash
;
3655 if (reg
> bufp
->re_nsub
|| reg
< 1
3656 /* Can't back reference to a subexp before its end. */
3657 || group_in_compile_stack (compile_stack
, reg
))
3658 FREE_STACK_RETURN (REG_ESUBREG
);
3661 BUF_PUSH_2 (duplicate
, reg
);
3668 if (syntax
& RE_BK_PLUS_QM
)
3671 goto normal_backslash
;
3675 /* You might think it would be useful for \ to mean
3676 not to translate; but if we don't translate it
3677 it will never match anything. */
3684 /* Expects the character in `c'. */
3686 /* If no exactn currently being built. */
3689 /* If last exactn not at current position. */
3690 || pending_exact
+ *pending_exact
+ 1 != b
3692 /* We have only one byte following the exactn for the count. */
3693 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3695 /* If followed by a repetition operator. */
3696 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3697 || ((syntax
& RE_BK_PLUS_QM
)
3698 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3699 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3700 || ((syntax
& RE_INTERVALS
)
3701 && ((syntax
& RE_NO_BK_BRACES
)
3702 ? p
!= pend
&& *p
== '{'
3703 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3705 /* Start building a new exactn. */
3709 BUF_PUSH_2 (exactn
, 0);
3710 pending_exact
= b
- 1;
3713 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3720 len
= CHAR_STRING (c
, b
);
3725 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3726 if (! CHAR_BYTE8_P (c1
))
3728 re_wchar_t c2
= TRANSLATE (c1
);
3730 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3736 (*pending_exact
) += len
;
3741 } /* while p != pend */
3744 /* Through the pattern now. */
3748 if (!COMPILE_STACK_EMPTY
)
3749 FREE_STACK_RETURN (REG_EPAREN
);
3751 /* If we don't want backtracking, force success
3752 the first time we reach the end of the compiled pattern. */
3753 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3756 /* We have succeeded; set the length of the buffer. */
3757 bufp
->used
= b
- bufp
->buffer
;
3762 re_compile_fastmap (bufp
);
3763 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3764 print_compiled_pattern (bufp
);
3769 #ifndef MATCH_MAY_ALLOCATE
3770 /* Initialize the failure stack to the largest possible stack. This
3771 isn't necessary unless we're trying to avoid calling alloca in
3772 the search and match routines. */
3774 int num_regs
= bufp
->re_nsub
+ 1;
3776 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3778 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3780 if (! fail_stack
.stack
)
3782 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3783 * sizeof (fail_stack_elt_t
));
3786 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3788 * sizeof (fail_stack_elt_t
)));
3791 regex_grow_registers (num_regs
);
3793 #endif /* not MATCH_MAY_ALLOCATE */
3795 FREE_STACK_RETURN (REG_NOERROR
);
3796 } /* regex_compile */
3798 /* Subroutines for `regex_compile'. */
3800 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3803 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3805 *loc
= (unsigned char) op
;
3806 STORE_NUMBER (loc
+ 1, arg
);
3810 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3813 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3815 *loc
= (unsigned char) op
;
3816 STORE_NUMBER (loc
+ 1, arg1
);
3817 STORE_NUMBER (loc
+ 3, arg2
);
3821 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3822 for OP followed by two-byte integer parameter ARG. */
3825 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3827 register unsigned char *pfrom
= end
;
3828 register unsigned char *pto
= end
+ 3;
3830 while (pfrom
!= loc
)
3833 store_op1 (op
, loc
, arg
);
3837 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3840 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3842 register unsigned char *pfrom
= end
;
3843 register unsigned char *pto
= end
+ 5;
3845 while (pfrom
!= loc
)
3848 store_op2 (op
, loc
, arg1
, arg2
);
3852 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3853 after an alternative or a begin-subexpression. We assume there is at
3854 least one character before the ^. */
3857 at_begline_loc_p (const re_char
*pattern
, const re_char
*p
, reg_syntax_t syntax
)
3859 re_char
*prev
= p
- 2;
3860 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3863 /* After a subexpression? */
3864 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3865 /* After an alternative? */
3866 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3867 /* After a shy subexpression? */
3868 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3869 && prev
[-1] == '?' && prev
[-2] == '('
3870 && (syntax
& RE_NO_BK_PARENS
3871 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3875 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3876 at least one character after the $, i.e., `P < PEND'. */
3879 at_endline_loc_p (const re_char
*p
, const re_char
*pend
, reg_syntax_t syntax
)
3882 boolean next_backslash
= *next
== '\\';
3883 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3886 /* Before a subexpression? */
3887 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3888 : next_backslash
&& next_next
&& *next_next
== ')')
3889 /* Before an alternative? */
3890 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3891 : next_backslash
&& next_next
&& *next_next
== '|');
3895 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3896 false if it's not. */
3899 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3903 for (this_element
= compile_stack
.avail
- 1;
3906 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3913 If fastmap is non-NULL, go through the pattern and fill fastmap
3914 with all the possible leading chars. If fastmap is NULL, don't
3915 bother filling it up (obviously) and only return whether the
3916 pattern could potentially match the empty string.
3918 Return 1 if p..pend might match the empty string.
3919 Return 0 if p..pend matches at least one char.
3920 Return -1 if fastmap was not updated accurately. */
3923 analyse_first (const re_char
*p
, const re_char
*pend
, char *fastmap
, const int multibyte
)
3928 /* If all elements for base leading-codes in fastmap is set, this
3929 flag is set true. */
3930 boolean match_any_multibyte_characters
= false;
3934 /* The loop below works as follows:
3935 - It has a working-list kept in the PATTERN_STACK and which basically
3936 starts by only containing a pointer to the first operation.
3937 - If the opcode we're looking at is a match against some set of
3938 chars, then we add those chars to the fastmap and go on to the
3939 next work element from the worklist (done via `break').
3940 - If the opcode is a control operator on the other hand, we either
3941 ignore it (if it's meaningless at this point, such as `start_memory')
3942 or execute it (if it's a jump). If the jump has several destinations
3943 (i.e. `on_failure_jump'), then we push the other destination onto the
3945 We guarantee termination by ignoring backward jumps (more or less),
3946 so that `p' is monotonically increasing. More to the point, we
3947 never set `p' (or push) anything `<= p1'. */
3951 /* `p1' is used as a marker of how far back a `on_failure_jump'
3952 can go without being ignored. It is normally equal to `p'
3953 (which prevents any backward `on_failure_jump') except right
3954 after a plain `jump', to allow patterns such as:
3957 10: on_failure_jump 3
3958 as used for the *? operator. */
3961 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3967 /* If the first character has to match a backreference, that means
3968 that the group was empty (since it already matched). Since this
3969 is the only case that interests us here, we can assume that the
3970 backreference must match the empty string. */
3975 /* Following are the cases which match a character. These end
3981 /* If multibyte is nonzero, the first byte of each
3982 character is an ASCII or a leading code. Otherwise,
3983 each byte is a character. Thus, this works in both
3988 /* For the case of matching this unibyte regex
3989 against multibyte, we must set a leading code of
3990 the corresponding multibyte character. */
3991 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
3993 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
4000 /* We could put all the chars except for \n (and maybe \0)
4001 but we don't bother since it is generally not worth it. */
4002 if (!fastmap
) break;
4007 if (!fastmap
) break;
4009 /* Chars beyond end of bitmap are possible matches. */
4010 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
4011 j
< (1 << BYTEWIDTH
); j
++)
4017 if (!fastmap
) break;
4018 not = (re_opcode_t
) *(p
- 1) == charset_not
;
4019 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
4021 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
4025 if (/* Any leading code can possibly start a character
4026 which doesn't match the specified set of characters. */
4029 /* If we can match a character class, we can match any
4030 multibyte characters. */
4031 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
4032 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
4035 if (match_any_multibyte_characters
== false)
4037 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4038 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4040 match_any_multibyte_characters
= true;
4044 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
4045 && match_any_multibyte_characters
== false)
4047 /* Set fastmap[I] to 1 where I is a leading code of each
4048 multibyte character in the range table. */
4050 unsigned char lc1
, lc2
;
4052 /* Make P points the range table. `+ 2' is to skip flag
4053 bits for a character class. */
4054 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
4056 /* Extract the number of ranges in range table into COUNT. */
4057 EXTRACT_NUMBER_AND_INCR (count
, p
);
4058 for (; count
> 0; count
--, p
+= 3)
4060 /* Extract the start and end of each range. */
4061 EXTRACT_CHARACTER (c
, p
);
4062 lc1
= CHAR_LEADING_CODE (c
);
4064 EXTRACT_CHARACTER (c
, p
);
4065 lc2
= CHAR_LEADING_CODE (c
);
4066 for (j
= lc1
; j
<= lc2
; j
++)
4075 if (!fastmap
) break;
4077 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4079 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4080 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4084 /* This match depends on text properties. These end with
4085 aborting optimizations. */
4089 case notcategoryspec
:
4090 if (!fastmap
) break;
4091 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4093 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
4094 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4097 /* Any leading code can possibly start a character which
4098 has or doesn't has the specified category. */
4099 if (match_any_multibyte_characters
== false)
4101 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4102 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4104 match_any_multibyte_characters
= true;
4108 /* All cases after this match the empty string. These end with
4130 EXTRACT_NUMBER_AND_INCR (j
, p
);
4132 /* Backward jumps can only go back to code that we've already
4133 visited. `re_compile' should make sure this is true. */
4136 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4138 case on_failure_jump
:
4139 case on_failure_keep_string_jump
:
4140 case on_failure_jump_loop
:
4141 case on_failure_jump_nastyloop
:
4142 case on_failure_jump_smart
:
4148 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4149 to jump back to "just after here". */
4152 case on_failure_jump
:
4153 case on_failure_keep_string_jump
:
4154 case on_failure_jump_nastyloop
:
4155 case on_failure_jump_loop
:
4156 case on_failure_jump_smart
:
4157 EXTRACT_NUMBER_AND_INCR (j
, p
);
4159 ; /* Backward jump to be ignored. */
4161 { /* We have to look down both arms.
4162 We first go down the "straight" path so as to minimize
4163 stack usage when going through alternatives. */
4164 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4172 /* This code simply does not properly handle forward jump_n. */
4173 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4175 /* jump_n can either jump or fall through. The (backward) jump
4176 case has already been handled, so we only need to look at the
4177 fallthrough case. */
4181 /* If N == 0, it should be an on_failure_jump_loop instead. */
4182 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4184 /* We only care about one iteration of the loop, so we don't
4185 need to consider the case where this behaves like an
4202 abort (); /* We have listed all the cases. */
4205 /* Getting here means we have found the possible starting
4206 characters for one path of the pattern -- and that the empty
4207 string does not match. We need not follow this path further. */
4211 /* We reached the end without matching anything. */
4214 } /* analyse_first */
4216 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4217 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4218 characters can start a string that matches the pattern. This fastmap
4219 is used by re_search to skip quickly over impossible starting points.
4221 Character codes above (1 << BYTEWIDTH) are not represented in the
4222 fastmap, but the leading codes are represented. Thus, the fastmap
4223 indicates which character sets could start a match.
4225 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4226 area as BUFP->fastmap.
4228 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4231 Returns 0 if we succeed, -2 if an internal error. */
4234 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4236 char *fastmap
= bufp
->fastmap
;
4239 assert (fastmap
&& bufp
->buffer
);
4241 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4242 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4244 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4245 fastmap
, RE_MULTIBYTE_P (bufp
));
4246 bufp
->can_be_null
= (analysis
!= 0);
4248 } /* re_compile_fastmap */
4250 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4251 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4252 this memory for recording register information. STARTS and ENDS
4253 must be allocated using the malloc library routine, and must each
4254 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4256 If NUM_REGS == 0, then subsequent matches should allocate their own
4259 Unless this function is called, the first search or match using
4260 PATTERN_BUFFER will allocate its own register data, without
4261 freeing the old data. */
4264 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4268 bufp
->regs_allocated
= REGS_REALLOCATE
;
4269 regs
->num_regs
= num_regs
;
4270 regs
->start
= starts
;
4275 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4277 regs
->start
= regs
->end
= (regoff_t
*) 0;
4280 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4282 /* Searching routines. */
4284 /* Like re_search_2, below, but only one string is specified, and
4285 doesn't let you say where to stop matching. */
4288 re_search (struct re_pattern_buffer
*bufp
, const char *string
, int size
, int startpos
, int range
, struct re_registers
*regs
)
4290 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4293 WEAK_ALIAS (__re_search
, re_search
)
4295 /* Head address of virtual concatenation of string. */
4296 #define HEAD_ADDR_VSTRING(P) \
4297 (((P) >= size1 ? string2 : string1))
4299 /* Address of POS in the concatenation of virtual string. */
4300 #define POS_ADDR_VSTRING(POS) \
4301 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4303 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4304 virtual concatenation of STRING1 and STRING2, starting first at index
4305 STARTPOS, then at STARTPOS + 1, and so on.
4307 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4309 RANGE is how far to scan while trying to match. RANGE = 0 means try
4310 only at STARTPOS; in general, the last start tried is STARTPOS +
4313 In REGS, return the indices of the virtual concatenation of STRING1
4314 and STRING2 that matched the entire BUFP->buffer and its contained
4317 Do not consider matching one past the index STOP in the virtual
4318 concatenation of STRING1 and STRING2.
4320 We return either the position in the strings at which the match was
4321 found, -1 if no match, or -2 if error (such as failure
4325 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, int size1
, const char *str2
, int size2
, int startpos
, int range
, struct re_registers
*regs
, int stop
)
4328 re_char
*string1
= (re_char
*) str1
;
4329 re_char
*string2
= (re_char
*) str2
;
4330 register char *fastmap
= bufp
->fastmap
;
4331 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4332 int total_size
= size1
+ size2
;
4333 int endpos
= startpos
+ range
;
4334 boolean anchored_start
;
4335 /* Nonzero if we are searching multibyte string. */
4336 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4338 /* Check for out-of-range STARTPOS. */
4339 if (startpos
< 0 || startpos
> total_size
)
4342 /* Fix up RANGE if it might eventually take us outside
4343 the virtual concatenation of STRING1 and STRING2.
4344 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4346 range
= 0 - startpos
;
4347 else if (endpos
> total_size
)
4348 range
= total_size
- startpos
;
4350 /* If the search isn't to be a backwards one, don't waste time in a
4351 search for a pattern anchored at beginning of buffer. */
4352 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4361 /* In a forward search for something that starts with \=.
4362 don't keep searching past point. */
4363 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4365 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4371 /* Update the fastmap now if not correct already. */
4372 if (fastmap
&& !bufp
->fastmap_accurate
)
4373 re_compile_fastmap (bufp
);
4375 /* See whether the pattern is anchored. */
4376 anchored_start
= (bufp
->buffer
[0] == begline
);
4379 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4381 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4383 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4387 /* Loop through the string, looking for a place to start matching. */
4390 /* If the pattern is anchored,
4391 skip quickly past places we cannot match.
4392 We don't bother to treat startpos == 0 specially
4393 because that case doesn't repeat. */
4394 if (anchored_start
&& startpos
> 0)
4396 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4397 : string2
[startpos
- size1
- 1])
4402 /* If a fastmap is supplied, skip quickly over characters that
4403 cannot be the start of a match. If the pattern can match the
4404 null string, however, we don't need to skip characters; we want
4405 the first null string. */
4406 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4408 register re_char
*d
;
4409 register re_wchar_t buf_ch
;
4411 d
= POS_ADDR_VSTRING (startpos
);
4413 if (range
> 0) /* Searching forwards. */
4415 register int lim
= 0;
4418 if (startpos
< size1
&& startpos
+ range
>= size1
)
4419 lim
= range
- (size1
- startpos
);
4421 /* Written out as an if-else to avoid testing `translate'
4423 if (RE_TRANSLATE_P (translate
))
4430 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4431 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4432 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4435 range
-= buf_charlen
;
4441 register re_wchar_t ch
, translated
;
4444 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4445 translated
= RE_TRANSLATE (translate
, ch
);
4446 if (translated
!= ch
4447 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4449 if (fastmap
[buf_ch
])
4462 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4463 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4465 range
-= buf_charlen
;
4469 while (range
> lim
&& !fastmap
[*d
])
4475 startpos
+= irange
- range
;
4477 else /* Searching backwards. */
4481 buf_ch
= STRING_CHAR (d
);
4482 buf_ch
= TRANSLATE (buf_ch
);
4483 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4488 register re_wchar_t ch
, translated
;
4491 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4492 translated
= TRANSLATE (ch
);
4493 if (translated
!= ch
4494 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4496 if (! fastmap
[TRANSLATE (buf_ch
)])
4502 /* If can't match the null string, and that's all we have left, fail. */
4503 if (range
>= 0 && startpos
== total_size
&& fastmap
4504 && !bufp
->can_be_null
)
4507 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4508 startpos
, regs
, stop
);
4521 /* Update STARTPOS to the next character boundary. */
4524 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4525 int len
= BYTES_BY_CHAR_HEAD (*p
);
4543 /* Update STARTPOS to the previous character boundary. */
4546 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4548 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4550 /* Find the head of multibyte form. */
4551 PREV_CHAR_BOUNDARY (p
, phead
);
4552 range
+= p0
- 1 - p
;
4556 startpos
-= p0
- 1 - p
;
4562 WEAK_ALIAS (__re_search_2
, re_search_2
)
4564 /* Declarations and macros for re_match_2. */
4566 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4568 RE_TRANSLATE_TYPE translate
,
4569 const int multibyte
));
4571 /* This converts PTR, a pointer into one of the search strings `string1'
4572 and `string2' into an offset from the beginning of that string. */
4573 #define POINTER_TO_OFFSET(ptr) \
4574 (FIRST_STRING_P (ptr) \
4575 ? ((regoff_t) ((ptr) - string1)) \
4576 : ((regoff_t) ((ptr) - string2 + size1)))
4578 /* Call before fetching a character with *d. This switches over to
4579 string2 if necessary.
4580 Check re_match_2_internal for a discussion of why end_match_2 might
4581 not be within string2 (but be equal to end_match_1 instead). */
4582 #define PREFETCH() \
4585 /* End of string2 => fail. */ \
4586 if (dend == end_match_2) \
4588 /* End of string1 => advance to string2. */ \
4590 dend = end_match_2; \
4593 /* Call before fetching a char with *d if you already checked other limits.
4594 This is meant for use in lookahead operations like wordend, etc..
4595 where we might need to look at parts of the string that might be
4596 outside of the LIMITs (i.e past `stop'). */
4597 #define PREFETCH_NOLIMIT() \
4601 dend = end_match_2; \
4604 /* Test if at very beginning or at very end of the virtual concatenation
4605 of `string1' and `string2'. If only one string, it's `string2'. */
4606 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4607 #define AT_STRINGS_END(d) ((d) == end2)
4609 /* Disabled due to a compiler bug -- see comment at case wordbound */
4611 /* The comment at case wordbound is following one, but we don't use
4612 AT_WORD_BOUNDARY anymore to support multibyte form.
4614 The DEC Alpha C compiler 3.x generates incorrect code for the
4615 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4616 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4617 macro and introducing temporary variables works around the bug. */
4620 /* Test if D points to a character which is word-constituent. We have
4621 two special cases to check for: if past the end of string1, look at
4622 the first character in string2; and if before the beginning of
4623 string2, look at the last character in string1. */
4624 #define WORDCHAR_P(d) \
4625 (SYNTAX ((d) == end1 ? *string2 \
4626 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4629 /* Test if the character before D and the one at D differ with respect
4630 to being word-constituent. */
4631 #define AT_WORD_BOUNDARY(d) \
4632 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4633 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4636 /* Free everything we malloc. */
4637 #ifdef MATCH_MAY_ALLOCATE
4638 # define FREE_VAR(var) \
4646 # define FREE_VARIABLES() \
4648 REGEX_FREE_STACK (fail_stack.stack); \
4649 FREE_VAR (regstart); \
4650 FREE_VAR (regend); \
4651 FREE_VAR (best_regstart); \
4652 FREE_VAR (best_regend); \
4655 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4656 #endif /* not MATCH_MAY_ALLOCATE */
4659 /* Optimization routines. */
4661 /* If the operation is a match against one or more chars,
4662 return a pointer to the next operation, else return NULL. */
4664 skip_one_char (const re_char
*p
)
4666 switch (SWITCH_ENUM_CAST (*p
++))
4677 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4680 p
= CHARSET_RANGE_TABLE (p
- 1);
4681 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4682 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4685 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4692 case notcategoryspec
:
4704 /* Jump over non-matching operations. */
4706 skip_noops (const re_char
*p
, const re_char
*pend
)
4711 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4720 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4731 /* Non-zero if "p1 matches something" implies "p2 fails". */
4733 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, const re_char
*p1
, const re_char
*p2
)
4736 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4737 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4739 assert (p1
>= bufp
->buffer
&& p1
< pend
4740 && p2
>= bufp
->buffer
&& p2
<= pend
);
4742 /* Skip over open/close-group commands.
4743 If what follows this loop is a ...+ construct,
4744 look at what begins its body, since we will have to
4745 match at least one of that. */
4746 p2
= skip_noops (p2
, pend
);
4747 /* The same skip can be done for p1, except that this function
4748 is only used in the case where p1 is a simple match operator. */
4749 /* p1 = skip_noops (p1, pend); */
4751 assert (p1
>= bufp
->buffer
&& p1
< pend
4752 && p2
>= bufp
->buffer
&& p2
<= pend
);
4754 op2
= p2
== pend
? succeed
: *p2
;
4756 switch (SWITCH_ENUM_CAST (op2
))
4760 /* If we're at the end of the pattern, we can change. */
4761 if (skip_one_char (p1
))
4763 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4771 register re_wchar_t c
4772 = (re_opcode_t
) *p2
== endline
? '\n'
4773 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4775 if ((re_opcode_t
) *p1
== exactn
)
4777 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4779 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4784 else if ((re_opcode_t
) *p1
== charset
4785 || (re_opcode_t
) *p1
== charset_not
)
4787 int not = (re_opcode_t
) *p1
== charset_not
;
4789 /* Test if C is listed in charset (or charset_not)
4791 if (! multibyte
|| IS_REAL_ASCII (c
))
4793 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4794 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4797 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4798 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4800 /* `not' is equal to 1 if c would match, which means
4801 that we can't change to pop_failure_jump. */
4804 DEBUG_PRINT1 (" No match => fast loop.\n");
4808 else if ((re_opcode_t
) *p1
== anychar
4811 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4819 if ((re_opcode_t
) *p1
== exactn
)
4820 /* Reuse the code above. */
4821 return mutually_exclusive_p (bufp
, p2
, p1
);
4823 /* It is hard to list up all the character in charset
4824 P2 if it includes multibyte character. Give up in
4826 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4828 /* Now, we are sure that P2 has no range table.
4829 So, for the size of bitmap in P2, `p2[1]' is
4830 enough. But P1 may have range table, so the
4831 size of bitmap table of P1 is extracted by
4832 using macro `CHARSET_BITMAP_SIZE'.
4834 In a multibyte case, we know that all the character
4835 listed in P2 is ASCII. In a unibyte case, P1 has only a
4836 bitmap table. So, in both cases, it is enough to test
4837 only the bitmap table of P1. */
4839 if ((re_opcode_t
) *p1
== charset
)
4842 /* We win if the charset inside the loop
4843 has no overlap with the one after the loop. */
4846 && idx
< CHARSET_BITMAP_SIZE (p1
));
4848 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4852 || idx
== CHARSET_BITMAP_SIZE (p1
))
4854 DEBUG_PRINT1 (" No match => fast loop.\n");
4858 else if ((re_opcode_t
) *p1
== charset_not
)
4861 /* We win if the charset_not inside the loop lists
4862 every character listed in the charset after. */
4863 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4864 if (! (p2
[2 + idx
] == 0
4865 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4866 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4871 DEBUG_PRINT1 (" No match => fast loop.\n");
4880 switch (SWITCH_ENUM_CAST (*p1
))
4884 /* Reuse the code above. */
4885 return mutually_exclusive_p (bufp
, p2
, p1
);
4887 /* When we have two charset_not, it's very unlikely that
4888 they don't overlap. The union of the two sets of excluded
4889 chars should cover all possible chars, which, as a matter of
4890 fact, is virtually impossible in multibyte buffers. */
4896 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4898 return ((re_opcode_t
) *p1
== syntaxspec
4899 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4901 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4904 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4906 return ((re_opcode_t
) *p1
== notsyntaxspec
4907 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4909 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4912 return (((re_opcode_t
) *p1
== notsyntaxspec
4913 || (re_opcode_t
) *p1
== syntaxspec
)
4918 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4919 case notcategoryspec
:
4920 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4932 /* Matching routines. */
4934 #ifndef emacs /* Emacs never uses this. */
4935 /* re_match is like re_match_2 except it takes only a single string. */
4938 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
4939 int size
, int pos
, struct re_registers
*regs
)
4941 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
4945 WEAK_ALIAS (__re_match
, re_match
)
4946 #endif /* not emacs */
4949 /* In Emacs, this is the string or buffer in which we
4950 are matching. It is used for looking up syntax properties. */
4951 Lisp_Object re_match_object
;
4954 /* re_match_2 matches the compiled pattern in BUFP against the
4955 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4956 and SIZE2, respectively). We start matching at POS, and stop
4959 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4960 store offsets for the substring each group matched in REGS. See the
4961 documentation for exactly how many groups we fill.
4963 We return -1 if no match, -2 if an internal error (such as the
4964 failure stack overflowing). Otherwise, we return the length of the
4965 matched substring. */
4968 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
, int size1
, const char *string2
, int size2
, int pos
, struct re_registers
*regs
, int stop
)
4974 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4975 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4976 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4979 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4980 (re_char
*) string2
, size2
,
4984 WEAK_ALIAS (__re_match_2
, re_match_2
)
4987 /* This is a separate function so that we can force an alloca cleanup
4990 re_match_2_internal (struct re_pattern_buffer
*bufp
, const re_char
*string1
, int size1
, const re_char
*string2
, int size2
, int pos
, struct re_registers
*regs
, int stop
)
4992 /* General temporaries. */
4996 /* Just past the end of the corresponding string. */
4997 re_char
*end1
, *end2
;
4999 /* Pointers into string1 and string2, just past the last characters in
5000 each to consider matching. */
5001 re_char
*end_match_1
, *end_match_2
;
5003 /* Where we are in the data, and the end of the current string. */
5006 /* Used sometimes to remember where we were before starting matching
5007 an operator so that we can go back in case of failure. This "atomic"
5008 behavior of matching opcodes is indispensable to the correctness
5009 of the on_failure_keep_string_jump optimization. */
5012 /* Where we are in the pattern, and the end of the pattern. */
5013 re_char
*p
= bufp
->buffer
;
5014 re_char
*pend
= p
+ bufp
->used
;
5016 /* We use this to map every character in the string. */
5017 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5019 /* Nonzero if BUFP is setup from a multibyte regex. */
5020 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
5022 /* Nonzero if STRING1/STRING2 are multibyte. */
5023 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
5025 /* Failure point stack. Each place that can handle a failure further
5026 down the line pushes a failure point on this stack. It consists of
5027 regstart, and regend for all registers corresponding to
5028 the subexpressions we're currently inside, plus the number of such
5029 registers, and, finally, two char *'s. The first char * is where
5030 to resume scanning the pattern; the second one is where to resume
5031 scanning the strings. */
5032 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5033 fail_stack_type fail_stack
;
5036 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5039 #if defined REL_ALLOC && defined REGEX_MALLOC
5040 /* This holds the pointer to the failure stack, when
5041 it is allocated relocatably. */
5042 fail_stack_elt_t
*failure_stack_ptr
;
5045 /* We fill all the registers internally, independent of what we
5046 return, for use in backreferences. The number here includes
5047 an element for register zero. */
5048 size_t num_regs
= bufp
->re_nsub
+ 1;
5050 /* Information on the contents of registers. These are pointers into
5051 the input strings; they record just what was matched (on this
5052 attempt) by a subexpression part of the pattern, that is, the
5053 regnum-th regstart pointer points to where in the pattern we began
5054 matching and the regnum-th regend points to right after where we
5055 stopped matching the regnum-th subexpression. (The zeroth register
5056 keeps track of what the whole pattern matches.) */
5057 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5058 re_char
**regstart
, **regend
;
5061 /* The following record the register info as found in the above
5062 variables when we find a match better than any we've seen before.
5063 This happens as we backtrack through the failure points, which in
5064 turn happens only if we have not yet matched the entire string. */
5065 unsigned best_regs_set
= false;
5066 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5067 re_char
**best_regstart
, **best_regend
;
5070 /* Logically, this is `best_regend[0]'. But we don't want to have to
5071 allocate space for that if we're not allocating space for anything
5072 else (see below). Also, we never need info about register 0 for
5073 any of the other register vectors, and it seems rather a kludge to
5074 treat `best_regend' differently than the rest. So we keep track of
5075 the end of the best match so far in a separate variable. We
5076 initialize this to NULL so that when we backtrack the first time
5077 and need to test it, it's not garbage. */
5078 re_char
*match_end
= NULL
;
5081 /* Counts the total number of registers pushed. */
5082 unsigned num_regs_pushed
= 0;
5085 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5089 #ifdef MATCH_MAY_ALLOCATE
5090 /* Do not bother to initialize all the register variables if there are
5091 no groups in the pattern, as it takes a fair amount of time. If
5092 there are groups, we include space for register 0 (the whole
5093 pattern), even though we never use it, since it simplifies the
5094 array indexing. We should fix this. */
5097 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5098 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5099 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5100 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5102 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5110 /* We must initialize all our variables to NULL, so that
5111 `FREE_VARIABLES' doesn't try to free them. */
5112 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5114 #endif /* MATCH_MAY_ALLOCATE */
5116 /* The starting position is bogus. */
5117 if (pos
< 0 || pos
> size1
+ size2
)
5123 /* Initialize subexpression text positions to -1 to mark ones that no
5124 start_memory/stop_memory has been seen for. Also initialize the
5125 register information struct. */
5126 for (reg
= 1; reg
< num_regs
; reg
++)
5127 regstart
[reg
] = regend
[reg
] = NULL
;
5129 /* We move `string1' into `string2' if the latter's empty -- but not if
5130 `string1' is null. */
5131 if (size2
== 0 && string1
!= NULL
)
5138 end1
= string1
+ size1
;
5139 end2
= string2
+ size2
;
5141 /* `p' scans through the pattern as `d' scans through the data.
5142 `dend' is the end of the input string that `d' points within. `d'
5143 is advanced into the following input string whenever necessary, but
5144 this happens before fetching; therefore, at the beginning of the
5145 loop, `d' can be pointing at the end of a string, but it cannot
5149 /* Only match within string2. */
5150 d
= string2
+ pos
- size1
;
5151 dend
= end_match_2
= string2
+ stop
- size1
;
5152 end_match_1
= end1
; /* Just to give it a value. */
5158 /* Only match within string1. */
5159 end_match_1
= string1
+ stop
;
5161 When we reach end_match_1, PREFETCH normally switches to string2.
5162 But in the present case, this means that just doing a PREFETCH
5163 makes us jump from `stop' to `gap' within the string.
5164 What we really want here is for the search to stop as
5165 soon as we hit end_match_1. That's why we set end_match_2
5166 to end_match_1 (since PREFETCH fails as soon as we hit
5168 end_match_2
= end_match_1
;
5171 { /* It's important to use this code when stop == size so that
5172 moving `d' from end1 to string2 will not prevent the d == dend
5173 check from catching the end of string. */
5175 end_match_2
= string2
+ stop
- size1
;
5181 DEBUG_PRINT1 ("The compiled pattern is: ");
5182 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5183 DEBUG_PRINT1 ("The string to match is: `");
5184 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5185 DEBUG_PRINT1 ("'\n");
5187 /* This loops over pattern commands. It exits by returning from the
5188 function if the match is complete, or it drops through if the match
5189 fails at this starting point in the input data. */
5192 DEBUG_PRINT2 ("\n%p: ", p
);
5195 { /* End of pattern means we might have succeeded. */
5196 DEBUG_PRINT1 ("end of pattern ... ");
5198 /* If we haven't matched the entire string, and we want the
5199 longest match, try backtracking. */
5200 if (d
!= end_match_2
)
5202 /* 1 if this match ends in the same string (string1 or string2)
5203 as the best previous match. */
5204 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5205 == FIRST_STRING_P (d
));
5206 /* 1 if this match is the best seen so far. */
5207 boolean best_match_p
;
5209 /* AIX compiler got confused when this was combined
5210 with the previous declaration. */
5212 best_match_p
= d
> match_end
;
5214 best_match_p
= !FIRST_STRING_P (d
);
5216 DEBUG_PRINT1 ("backtracking.\n");
5218 if (!FAIL_STACK_EMPTY ())
5219 { /* More failure points to try. */
5221 /* If exceeds best match so far, save it. */
5222 if (!best_regs_set
|| best_match_p
)
5224 best_regs_set
= true;
5227 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5229 for (reg
= 1; reg
< num_regs
; reg
++)
5231 best_regstart
[reg
] = regstart
[reg
];
5232 best_regend
[reg
] = regend
[reg
];
5238 /* If no failure points, don't restore garbage. And if
5239 last match is real best match, don't restore second
5241 else if (best_regs_set
&& !best_match_p
)
5244 /* Restore best match. It may happen that `dend ==
5245 end_match_1' while the restored d is in string2.
5246 For example, the pattern `x.*y.*z' against the
5247 strings `x-' and `y-z-', if the two strings are
5248 not consecutive in memory. */
5249 DEBUG_PRINT1 ("Restoring best registers.\n");
5252 dend
= ((d
>= string1
&& d
<= end1
)
5253 ? end_match_1
: end_match_2
);
5255 for (reg
= 1; reg
< num_regs
; reg
++)
5257 regstart
[reg
] = best_regstart
[reg
];
5258 regend
[reg
] = best_regend
[reg
];
5261 } /* d != end_match_2 */
5264 DEBUG_PRINT1 ("Accepting match.\n");
5266 /* If caller wants register contents data back, do it. */
5267 if (regs
&& !bufp
->no_sub
)
5269 /* Have the register data arrays been allocated? */
5270 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5271 { /* No. So allocate them with malloc. We need one
5272 extra element beyond `num_regs' for the `-1' marker
5274 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5275 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5276 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5277 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5282 bufp
->regs_allocated
= REGS_REALLOCATE
;
5284 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5285 { /* Yes. If we need more elements than were already
5286 allocated, reallocate them. If we need fewer, just
5288 if (regs
->num_regs
< num_regs
+ 1)
5290 regs
->num_regs
= num_regs
+ 1;
5291 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5292 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5293 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5302 /* These braces fend off a "empty body in an else-statement"
5303 warning under GCC when assert expands to nothing. */
5304 assert (bufp
->regs_allocated
== REGS_FIXED
);
5307 /* Convert the pointer data in `regstart' and `regend' to
5308 indices. Register zero has to be set differently,
5309 since we haven't kept track of any info for it. */
5310 if (regs
->num_regs
> 0)
5312 regs
->start
[0] = pos
;
5313 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5316 /* Go through the first `min (num_regs, regs->num_regs)'
5317 registers, since that is all we initialized. */
5318 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5320 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5321 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5325 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5327 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5331 /* If the regs structure we return has more elements than
5332 were in the pattern, set the extra elements to -1. If
5333 we (re)allocated the registers, this is the case,
5334 because we always allocate enough to have at least one
5336 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5337 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5338 } /* regs && !bufp->no_sub */
5340 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5341 nfailure_points_pushed
, nfailure_points_popped
,
5342 nfailure_points_pushed
- nfailure_points_popped
);
5343 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5345 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5347 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5353 /* Otherwise match next pattern command. */
5354 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5356 /* Ignore these. Used to ignore the n of succeed_n's which
5357 currently have n == 0. */
5359 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5363 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5366 /* Match the next n pattern characters exactly. The following
5367 byte in the pattern defines n, and the n bytes after that
5368 are the characters to match. */
5371 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5373 /* Remember the start point to rollback upon failure. */
5377 /* This is written out as an if-else so we don't waste time
5378 testing `translate' inside the loop. */
5379 if (RE_TRANSLATE_P (translate
))
5383 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5403 /* The cost of testing `translate' is comparatively small. */
5404 if (target_multibyte
)
5407 int pat_charlen
, buf_charlen
;
5412 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5415 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5418 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5420 if (TRANSLATE (buf_ch
) != pat_ch
)
5428 mcnt
-= pat_charlen
;
5440 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5441 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5448 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5449 if (! CHAR_BYTE8_P (buf_ch
))
5451 buf_ch
= TRANSLATE (buf_ch
);
5452 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5458 if (buf_ch
!= pat_ch
)
5471 /* Match any character except possibly a newline or a null. */
5477 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5480 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5482 buf_ch
= TRANSLATE (buf_ch
);
5484 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5486 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5487 && buf_ch
== '\000'))
5490 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5499 register unsigned int c
;
5500 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5503 /* Start of actual range_table, or end of bitmap if there is no
5505 re_char
*range_table
IF_LINT (= NULL
);
5507 /* Nonzero if there is a range table. */
5508 int range_table_exists
;
5510 /* Number of ranges of range table. This is not included
5511 in the initial byte-length of the command. */
5514 /* Whether matching against a unibyte character. */
5515 boolean unibyte_char
= false;
5517 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5519 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5521 if (range_table_exists
)
5523 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5524 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5528 c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5529 if (target_multibyte
)
5534 c1
= RE_CHAR_TO_UNIBYTE (c
);
5537 unibyte_char
= true;
5543 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5545 if (! CHAR_BYTE8_P (c1
))
5547 c1
= TRANSLATE (c1
);
5548 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5551 unibyte_char
= true;
5556 unibyte_char
= true;
5559 if (unibyte_char
&& c
< (1 << BYTEWIDTH
))
5560 { /* Lookup bitmap. */
5561 /* Cast to `unsigned' instead of `unsigned char' in
5562 case the bit list is a full 32 bytes long. */
5563 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5564 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5568 else if (range_table_exists
)
5570 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5572 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5573 | (class_bits
& BIT_MULTIBYTE
)
5574 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5575 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5576 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5577 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5580 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5584 if (range_table_exists
)
5585 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5587 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5589 if (!not) goto fail
;
5596 /* The beginning of a group is represented by start_memory.
5597 The argument is the register number. The text
5598 matched within the group is recorded (in the internal
5599 registers data structure) under the register number. */
5601 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5603 /* In case we need to undo this operation (via backtracking). */
5604 PUSH_FAILURE_REG ((unsigned int)*p
);
5607 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5608 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5610 /* Move past the register number and inner group count. */
5615 /* The stop_memory opcode represents the end of a group. Its
5616 argument is the same as start_memory's: the register number. */
5618 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5620 assert (!REG_UNSET (regstart
[*p
]));
5621 /* Strictly speaking, there should be code such as:
5623 assert (REG_UNSET (regend[*p]));
5624 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5626 But the only info to be pushed is regend[*p] and it is known to
5627 be UNSET, so there really isn't anything to push.
5628 Not pushing anything, on the other hand deprives us from the
5629 guarantee that regend[*p] is UNSET since undoing this operation
5630 will not reset its value properly. This is not important since
5631 the value will only be read on the next start_memory or at
5632 the very end and both events can only happen if this stop_memory
5636 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5638 /* Move past the register number and the inner group count. */
5643 /* \<digit> has been turned into a `duplicate' command which is
5644 followed by the numeric value of <digit> as the register number. */
5647 register re_char
*d2
, *dend2
;
5648 int regno
= *p
++; /* Get which register to match against. */
5649 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5651 /* Can't back reference a group which we've never matched. */
5652 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5655 /* Where in input to try to start matching. */
5656 d2
= regstart
[regno
];
5658 /* Remember the start point to rollback upon failure. */
5661 /* Where to stop matching; if both the place to start and
5662 the place to stop matching are in the same string, then
5663 set to the place to stop, otherwise, for now have to use
5664 the end of the first string. */
5666 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5667 == FIRST_STRING_P (regend
[regno
]))
5668 ? regend
[regno
] : end_match_1
);
5671 /* If necessary, advance to next segment in register
5675 if (dend2
== end_match_2
) break;
5676 if (dend2
== regend
[regno
]) break;
5678 /* End of string1 => advance to string2. */
5680 dend2
= regend
[regno
];
5682 /* At end of register contents => success */
5683 if (d2
== dend2
) break;
5685 /* If necessary, advance to next segment in data. */
5688 /* How many characters left in this segment to match. */
5691 /* Want how many consecutive characters we can match in
5692 one shot, so, if necessary, adjust the count. */
5693 if (mcnt
> dend2
- d2
)
5696 /* Compare that many; failure if mismatch, else move
5698 if (RE_TRANSLATE_P (translate
)
5699 ? bcmp_translate (d
, d2
, mcnt
, translate
, target_multibyte
)
5700 : memcmp (d
, d2
, mcnt
))
5705 d
+= mcnt
, d2
+= mcnt
;
5711 /* begline matches the empty string at the beginning of the string
5712 (unless `not_bol' is set in `bufp'), and after newlines. */
5714 DEBUG_PRINT1 ("EXECUTING begline.\n");
5716 if (AT_STRINGS_BEG (d
))
5718 if (!bufp
->not_bol
) break;
5723 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5727 /* In all other cases, we fail. */
5731 /* endline is the dual of begline. */
5733 DEBUG_PRINT1 ("EXECUTING endline.\n");
5735 if (AT_STRINGS_END (d
))
5737 if (!bufp
->not_eol
) break;
5741 PREFETCH_NOLIMIT ();
5748 /* Match at the very beginning of the data. */
5750 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5751 if (AT_STRINGS_BEG (d
))
5756 /* Match at the very end of the data. */
5758 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5759 if (AT_STRINGS_END (d
))
5764 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5765 pushes NULL as the value for the string on the stack. Then
5766 `POP_FAILURE_POINT' will keep the current value for the
5767 string, instead of restoring it. To see why, consider
5768 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5769 then the . fails against the \n. But the next thing we want
5770 to do is match the \n against the \n; if we restored the
5771 string value, we would be back at the foo.
5773 Because this is used only in specific cases, we don't need to
5774 check all the things that `on_failure_jump' does, to make
5775 sure the right things get saved on the stack. Hence we don't
5776 share its code. The only reason to push anything on the
5777 stack at all is that otherwise we would have to change
5778 `anychar's code to do something besides goto fail in this
5779 case; that seems worse than this. */
5780 case on_failure_keep_string_jump
:
5781 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5782 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5785 PUSH_FAILURE_POINT (p
- 3, NULL
);
5788 /* A nasty loop is introduced by the non-greedy *? and +?.
5789 With such loops, the stack only ever contains one failure point
5790 at a time, so that a plain on_failure_jump_loop kind of
5791 cycle detection cannot work. Worse yet, such a detection
5792 can not only fail to detect a cycle, but it can also wrongly
5793 detect a cycle (between different instantiations of the same
5795 So the method used for those nasty loops is a little different:
5796 We use a special cycle-detection-stack-frame which is pushed
5797 when the on_failure_jump_nastyloop failure-point is *popped*.
5798 This special frame thus marks the beginning of one iteration
5799 through the loop and we can hence easily check right here
5800 whether something matched between the beginning and the end of
5802 case on_failure_jump_nastyloop
:
5803 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5804 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5807 assert ((re_opcode_t
)p
[-4] == no_op
);
5810 CHECK_INFINITE_LOOP (p
- 4, d
);
5812 /* If there's a cycle, just continue without pushing
5813 this failure point. The failure point is the "try again"
5814 option, which shouldn't be tried.
5815 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5816 PUSH_FAILURE_POINT (p
- 3, d
);
5820 /* Simple loop detecting on_failure_jump: just check on the
5821 failure stack if the same spot was already hit earlier. */
5822 case on_failure_jump_loop
:
5824 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5825 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5829 CHECK_INFINITE_LOOP (p
- 3, d
);
5831 /* If there's a cycle, get out of the loop, as if the matching
5832 had failed. We used to just `goto fail' here, but that was
5833 aborting the search a bit too early: we want to keep the
5834 empty-loop-match and keep matching after the loop.
5835 We want (x?)*y\1z to match both xxyz and xxyxz. */
5838 PUSH_FAILURE_POINT (p
- 3, d
);
5843 /* Uses of on_failure_jump:
5845 Each alternative starts with an on_failure_jump that points
5846 to the beginning of the next alternative. Each alternative
5847 except the last ends with a jump that in effect jumps past
5848 the rest of the alternatives. (They really jump to the
5849 ending jump of the following alternative, because tensioning
5850 these jumps is a hassle.)
5852 Repeats start with an on_failure_jump that points past both
5853 the repetition text and either the following jump or
5854 pop_failure_jump back to this on_failure_jump. */
5855 case on_failure_jump
:
5856 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5857 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5860 PUSH_FAILURE_POINT (p
-3, d
);
5863 /* This operation is used for greedy *.
5864 Compare the beginning of the repeat with what in the
5865 pattern follows its end. If we can establish that there
5866 is nothing that they would both match, i.e., that we
5867 would have to backtrack because of (as in, e.g., `a*a')
5868 then we can use a non-backtracking loop based on
5869 on_failure_keep_string_jump instead of on_failure_jump. */
5870 case on_failure_jump_smart
:
5871 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5872 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5875 re_char
*p1
= p
; /* Next operation. */
5876 /* Here, we discard `const', making re_match non-reentrant. */
5877 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5878 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5880 p
-= 3; /* Reset so that we will re-execute the
5881 instruction once it's been changed. */
5883 EXTRACT_NUMBER (mcnt
, p2
- 2);
5885 /* Ensure this is a indeed the trivial kind of loop
5886 we are expecting. */
5887 assert (skip_one_char (p1
) == p2
- 3);
5888 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5889 DEBUG_STATEMENT (debug
+= 2);
5890 if (mutually_exclusive_p (bufp
, p1
, p2
))
5892 /* Use a fast `on_failure_keep_string_jump' loop. */
5893 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5894 *p3
= (unsigned char) on_failure_keep_string_jump
;
5895 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5899 /* Default to a safe `on_failure_jump' loop. */
5900 DEBUG_PRINT1 (" smart default => slow loop.\n");
5901 *p3
= (unsigned char) on_failure_jump
;
5903 DEBUG_STATEMENT (debug
-= 2);
5907 /* Unconditionally jump (without popping any failure points). */
5910 IMMEDIATE_QUIT_CHECK
;
5911 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5912 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5913 p
+= mcnt
; /* Do the jump. */
5914 DEBUG_PRINT2 ("(to %p).\n", p
);
5918 /* Have to succeed matching what follows at least n times.
5919 After that, handle like `on_failure_jump'. */
5921 /* Signedness doesn't matter since we only compare MCNT to 0. */
5922 EXTRACT_NUMBER (mcnt
, p
+ 2);
5923 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5925 /* Originally, mcnt is how many times we HAVE to succeed. */
5928 /* Here, we discard `const', making re_match non-reentrant. */
5929 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5932 PUSH_NUMBER (p2
, mcnt
);
5935 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5940 /* Signedness doesn't matter since we only compare MCNT to 0. */
5941 EXTRACT_NUMBER (mcnt
, p
+ 2);
5942 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5944 /* Originally, this is how many times we CAN jump. */
5947 /* Here, we discard `const', making re_match non-reentrant. */
5948 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5950 PUSH_NUMBER (p2
, mcnt
);
5951 goto unconditional_jump
;
5953 /* If don't have to jump any more, skip over the rest of command. */
5960 unsigned char *p2
; /* Location of the counter. */
5961 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5963 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5964 /* Here, we discard `const', making re_match non-reentrant. */
5965 p2
= (unsigned char*) p
+ mcnt
;
5966 /* Signedness doesn't matter since we only copy MCNT's bits . */
5967 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5968 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
5969 PUSH_NUMBER (p2
, mcnt
);
5976 boolean
not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5977 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5979 /* We SUCCEED (or FAIL) in one of the following cases: */
5981 /* Case 1: D is at the beginning or the end of string. */
5982 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5986 /* C1 is the character before D, S1 is the syntax of C1, C2
5987 is the character at D, and S2 is the syntax of C2. */
5992 int offset
= PTR_TO_OFFSET (d
- 1);
5993 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5994 UPDATE_SYNTAX_TABLE (charpos
);
5996 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5999 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6001 PREFETCH_NOLIMIT ();
6002 GET_CHAR_AFTER (c2
, d
, dummy
);
6005 if (/* Case 2: Only one of S1 and S2 is Sword. */
6006 ((s1
== Sword
) != (s2
== Sword
))
6007 /* Case 3: Both of S1 and S2 are Sword, and macro
6008 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6009 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
6019 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6021 /* We FAIL in one of the following cases: */
6023 /* Case 1: D is at the end of string. */
6024 if (AT_STRINGS_END (d
))
6028 /* C1 is the character before D, S1 is the syntax of C1, C2
6029 is the character at D, and S2 is the syntax of C2. */
6034 int offset
= PTR_TO_OFFSET (d
);
6035 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6036 UPDATE_SYNTAX_TABLE (charpos
);
6039 GET_CHAR_AFTER (c2
, d
, dummy
);
6042 /* Case 2: S2 is not Sword. */
6046 /* Case 3: D is not at the beginning of string ... */
6047 if (!AT_STRINGS_BEG (d
))
6049 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6051 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6055 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6057 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6064 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6066 /* We FAIL in one of the following cases: */
6068 /* Case 1: D is at the beginning of string. */
6069 if (AT_STRINGS_BEG (d
))
6073 /* C1 is the character before D, S1 is the syntax of C1, C2
6074 is the character at D, and S2 is the syntax of C2. */
6079 int offset
= PTR_TO_OFFSET (d
) - 1;
6080 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6081 UPDATE_SYNTAX_TABLE (charpos
);
6083 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6086 /* Case 2: S1 is not Sword. */
6090 /* Case 3: D is not at the end of string ... */
6091 if (!AT_STRINGS_END (d
))
6093 PREFETCH_NOLIMIT ();
6094 GET_CHAR_AFTER (c2
, d
, dummy
);
6096 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6100 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6102 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6109 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6111 /* We FAIL in one of the following cases: */
6113 /* Case 1: D is at the end of string. */
6114 if (AT_STRINGS_END (d
))
6118 /* C1 is the character before D, S1 is the syntax of C1, C2
6119 is the character at D, and S2 is the syntax of C2. */
6123 int offset
= PTR_TO_OFFSET (d
);
6124 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6125 UPDATE_SYNTAX_TABLE (charpos
);
6128 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6131 /* Case 2: S2 is neither Sword nor Ssymbol. */
6132 if (s2
!= Sword
&& s2
!= Ssymbol
)
6135 /* Case 3: D is not at the beginning of string ... */
6136 if (!AT_STRINGS_BEG (d
))
6138 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6140 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6144 /* ... and S1 is Sword or Ssymbol. */
6145 if (s1
== Sword
|| s1
== Ssymbol
)
6152 DEBUG_PRINT1 ("EXECUTING symend.\n");
6154 /* We FAIL in one of the following cases: */
6156 /* Case 1: D is at the beginning of string. */
6157 if (AT_STRINGS_BEG (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. */
6166 int offset
= PTR_TO_OFFSET (d
) - 1;
6167 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6168 UPDATE_SYNTAX_TABLE (charpos
);
6170 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6173 /* Case 2: S1 is neither Ssymbol nor Sword. */
6174 if (s1
!= Sword
&& s1
!= Ssymbol
)
6177 /* Case 3: D is not at the end of string ... */
6178 if (!AT_STRINGS_END (d
))
6180 PREFETCH_NOLIMIT ();
6181 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6183 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6187 /* ... and S2 is Sword or Ssymbol. */
6188 if (s2
== Sword
|| s2
== Ssymbol
)
6197 boolean
not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6199 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
6203 int offset
= PTR_TO_OFFSET (d
);
6204 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6205 UPDATE_SYNTAX_TABLE (pos1
);
6212 GET_CHAR_AFTER (c
, d
, len
);
6213 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6222 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6223 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6228 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6229 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6234 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6235 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6240 case notcategoryspec
:
6242 boolean
not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6244 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n",
6245 not?"not":"", mcnt
);
6251 GET_CHAR_AFTER (c
, d
, len
);
6252 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6264 continue; /* Successfully executed one pattern command; keep going. */
6267 /* We goto here if a matching operation fails. */
6269 IMMEDIATE_QUIT_CHECK
;
6270 if (!FAIL_STACK_EMPTY ())
6273 /* A restart point is known. Restore to that state. */
6274 DEBUG_PRINT1 ("\nFAIL:\n");
6275 POP_FAILURE_POINT (str
, pat
);
6276 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
6278 case on_failure_keep_string_jump
:
6279 assert (str
== NULL
);
6280 goto continue_failure_jump
;
6282 case on_failure_jump_nastyloop
:
6283 assert ((re_opcode_t
)pat
[-2] == no_op
);
6284 PUSH_FAILURE_POINT (pat
- 2, str
);
6287 case on_failure_jump_loop
:
6288 case on_failure_jump
:
6291 continue_failure_jump
:
6292 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6297 /* A special frame used for nastyloops. */
6304 assert (p
>= bufp
->buffer
&& p
<= pend
);
6306 if (d
>= string1
&& d
<= end1
)
6310 break; /* Matching at this starting point really fails. */
6314 goto restore_best_regs
;
6318 return -1; /* Failure to match. */
6321 /* Subroutine definitions for re_match_2. */
6323 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6324 bytes; nonzero otherwise. */
6327 bcmp_translate (const re_char
*s1
, const re_char
*s2
, register int len
,
6328 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6330 register re_char
*p1
= s1
, *p2
= s2
;
6331 re_char
*p1_end
= s1
+ len
;
6332 re_char
*p2_end
= s2
+ len
;
6334 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6335 different lengths, but relying on a single `len' would break this. -sm */
6336 while (p1
< p1_end
&& p2
< p2_end
)
6338 int p1_charlen
, p2_charlen
;
6339 re_wchar_t p1_ch
, p2_ch
;
6341 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6342 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6344 if (RE_TRANSLATE (translate
, p1_ch
)
6345 != RE_TRANSLATE (translate
, p2_ch
))
6348 p1
+= p1_charlen
, p2
+= p2_charlen
;
6351 if (p1
!= p1_end
|| p2
!= p2_end
)
6357 /* Entry points for GNU code. */
6359 /* re_compile_pattern is the GNU regular expression compiler: it
6360 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6361 Returns 0 if the pattern was valid, otherwise an error string.
6363 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6364 are set in BUFP on entry.
6366 We call regex_compile to do the actual compilation. */
6369 re_compile_pattern (const char *pattern
, size_t length
, struct re_pattern_buffer
*bufp
)
6373 /* GNU code is written to assume at least RE_NREGS registers will be set
6374 (and at least one extra will be -1). */
6375 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6377 /* And GNU code determines whether or not to get register information
6378 by passing null for the REGS argument to re_match, etc., not by
6382 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6386 return gettext (re_error_msgid
[(int) ret
]);
6388 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6390 /* Entry points compatible with 4.2 BSD regex library. We don't define
6391 them unless specifically requested. */
6393 #if defined _REGEX_RE_COMP || defined _LIBC
6395 /* BSD has one and only one pattern buffer. */
6396 static struct re_pattern_buffer re_comp_buf
;
6400 /* Make these definitions weak in libc, so POSIX programs can redefine
6401 these names if they don't use our functions, and still use
6402 regcomp/regexec below without link errors. */
6412 if (!re_comp_buf
.buffer
)
6413 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6414 return (char *) gettext ("No previous regular expression");
6418 if (!re_comp_buf
.buffer
)
6420 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6421 if (re_comp_buf
.buffer
== NULL
)
6422 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6423 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6424 re_comp_buf
.allocated
= 200;
6426 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6427 if (re_comp_buf
.fastmap
== NULL
)
6428 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6429 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6432 /* Since `re_exec' always passes NULL for the `regs' argument, we
6433 don't need to initialize the pattern buffer fields which affect it. */
6435 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6440 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6441 return (char *) gettext (re_error_msgid
[(int) ret
]);
6452 const int len
= strlen (s
);
6454 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6456 #endif /* _REGEX_RE_COMP */
6458 /* POSIX.2 functions. Don't define these for Emacs. */
6462 /* regcomp takes a regular expression as a string and compiles it.
6464 PREG is a regex_t *. We do not expect any fields to be initialized,
6465 since POSIX says we shouldn't. Thus, we set
6467 `buffer' to the compiled pattern;
6468 `used' to the length of the compiled pattern;
6469 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6470 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6471 RE_SYNTAX_POSIX_BASIC;
6472 `fastmap' to an allocated space for the fastmap;
6473 `fastmap_accurate' to zero;
6474 `re_nsub' to the number of subexpressions in PATTERN.
6476 PATTERN is the address of the pattern string.
6478 CFLAGS is a series of bits which affect compilation.
6480 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6481 use POSIX basic syntax.
6483 If REG_NEWLINE is set, then . and [^...] don't match newline.
6484 Also, regexec will try a match beginning after every newline.
6486 If REG_ICASE is set, then we considers upper- and lowercase
6487 versions of letters to be equivalent when matching.
6489 If REG_NOSUB is set, then when PREG is passed to regexec, that
6490 routine will report only success or failure, and nothing about the
6493 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6494 the return codes and their meanings.) */
6497 regcomp (regex_t
*__restrict preg
, const char *__restrict pattern
,
6502 = (cflags
& REG_EXTENDED
) ?
6503 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6505 /* regex_compile will allocate the space for the compiled pattern. */
6507 preg
->allocated
= 0;
6510 /* Try to allocate space for the fastmap. */
6511 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6513 if (cflags
& REG_ICASE
)
6518 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6519 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6520 if (preg
->translate
== NULL
)
6521 return (int) REG_ESPACE
;
6523 /* Map uppercase characters to corresponding lowercase ones. */
6524 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6525 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6528 preg
->translate
= NULL
;
6530 /* If REG_NEWLINE is set, newlines are treated differently. */
6531 if (cflags
& REG_NEWLINE
)
6532 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6533 syntax
&= ~RE_DOT_NEWLINE
;
6534 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6537 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6539 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6541 /* POSIX says a null character in the pattern terminates it, so we
6542 can use strlen here in compiling the pattern. */
6543 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6545 /* POSIX doesn't distinguish between an unmatched open-group and an
6546 unmatched close-group: both are REG_EPAREN. */
6547 if (ret
== REG_ERPAREN
)
6550 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6551 { /* Compute the fastmap now, since regexec cannot modify the pattern
6553 re_compile_fastmap (preg
);
6554 if (preg
->can_be_null
)
6555 { /* The fastmap can't be used anyway. */
6556 free (preg
->fastmap
);
6557 preg
->fastmap
= NULL
;
6562 WEAK_ALIAS (__regcomp
, regcomp
)
6565 /* regexec searches for a given pattern, specified by PREG, in the
6568 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6569 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6570 least NMATCH elements, and we set them to the offsets of the
6571 corresponding matched substrings.
6573 EFLAGS specifies `execution flags' which affect matching: if
6574 REG_NOTBOL is set, then ^ does not match at the beginning of the
6575 string; if REG_NOTEOL is set, then $ does not match at the end.
6577 We return 0 if we find a match and REG_NOMATCH if not. */
6580 regexec (const regex_t
*__restrict preg
, const char *__restrict string
,
6581 size_t nmatch
, regmatch_t pmatch
[__restrict_arr
], int eflags
)
6584 struct re_registers regs
;
6585 regex_t private_preg
;
6586 int len
= strlen (string
);
6587 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6589 private_preg
= *preg
;
6591 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6592 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6594 /* The user has told us exactly how many registers to return
6595 information about, via `nmatch'. We have to pass that on to the
6596 matching routines. */
6597 private_preg
.regs_allocated
= REGS_FIXED
;
6601 regs
.num_regs
= nmatch
;
6602 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6603 if (regs
.start
== NULL
)
6604 return (int) REG_NOMATCH
;
6605 regs
.end
= regs
.start
+ nmatch
;
6608 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6609 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6610 was a little bit longer but still only matching the real part.
6611 This works because the `endline' will check for a '\n' and will find a
6612 '\0', correctly deciding that this is not the end of a line.
6613 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6614 a convenient '\0' there. For all we know, the string could be preceded
6615 by '\n' which would throw things off. */
6617 /* Perform the searching operation. */
6618 ret
= re_search (&private_preg
, string
, len
,
6619 /* start: */ 0, /* range: */ len
,
6620 want_reg_info
? ®s
: (struct re_registers
*) 0);
6622 /* Copy the register information to the POSIX structure. */
6629 for (r
= 0; r
< nmatch
; r
++)
6631 pmatch
[r
].rm_so
= regs
.start
[r
];
6632 pmatch
[r
].rm_eo
= regs
.end
[r
];
6636 /* If we needed the temporary register info, free the space now. */
6640 /* We want zero return to mean success, unlike `re_search'. */
6641 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6643 WEAK_ALIAS (__regexec
, regexec
)
6646 /* Returns a message corresponding to an error code, ERR_CODE, returned
6647 from either regcomp or regexec. We don't use PREG here.
6649 ERR_CODE was previously called ERRCODE, but that name causes an
6650 error with msvc8 compiler. */
6653 regerror (int err_code
, const regex_t
*preg
, char *errbuf
, size_t errbuf_size
)
6659 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6660 /* Only error codes returned by the rest of the code should be passed
6661 to this routine. If we are given anything else, or if other regex
6662 code generates an invalid error code, then the program has a bug.
6663 Dump core so we can fix it. */
6666 msg
= gettext (re_error_msgid
[err_code
]);
6668 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6670 if (errbuf_size
!= 0)
6672 if (msg_size
> errbuf_size
)
6674 strncpy (errbuf
, msg
, errbuf_size
- 1);
6675 errbuf
[errbuf_size
- 1] = 0;
6678 strcpy (errbuf
, msg
);
6683 WEAK_ALIAS (__regerror
, regerror
)
6686 /* Free dynamically allocated space used by PREG. */
6689 regfree (regex_t
*preg
)
6691 free (preg
->buffer
);
6692 preg
->buffer
= NULL
;
6694 preg
->allocated
= 0;
6697 free (preg
->fastmap
);
6698 preg
->fastmap
= NULL
;
6699 preg
->fastmap_accurate
= 0;
6701 free (preg
->translate
);
6702 preg
->translate
= NULL
;
6704 WEAK_ALIAS (__regfree
, regfree
)
6706 #endif /* not emacs */