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) 1993-2014 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, see <http://www.gnu.org/licenses/>. */
21 - structure the opcode space into opcode+flag.
22 - merge with glibc's regex.[ch].
23 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
24 need to modify the compiled regexp so that re_match can be reentrant.
25 - get rid of on_failure_jump_smart by doing the optimization in re_comp
26 rather than at run-time, so that re_match can be reentrant.
29 /* AIX requires this to be the first thing in the file. */
30 #if defined _AIX && !defined REGEX_MALLOC
34 /* Ignore some GCC warnings for now. This section should go away
35 once the Emacs and Gnulib regex code is merged. */
36 #if 4 < __GNUC__ + (5 <= __GNUC_MINOR__) || defined __clang__
37 # pragma GCC diagnostic ignored "-Wstrict-overflow"
39 # pragma GCC diagnostic ignored "-Wunused-function"
40 # pragma GCC diagnostic ignored "-Wunused-macros"
41 # pragma GCC diagnostic ignored "-Wunused-result"
42 # pragma GCC diagnostic ignored "-Wunused-variable"
46 #if 4 < __GNUC__ + (6 <= __GNUC_MINOR__) && ! defined __clang__
47 # pragma GCC diagnostic ignored "-Wunused-but-set-variable"
55 /* We need this for `regex.h', and perhaps for the Emacs include files. */
56 # include <sys/types.h>
59 /* Whether to use ISO C Amendment 1 wide char functions.
60 Those should not be used for Emacs since it uses its own. */
62 #define WIDE_CHAR_SUPPORT 1
64 #define WIDE_CHAR_SUPPORT \
65 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
68 /* For platform which support the ISO C amendment 1 functionality we
69 support user defined character classes. */
71 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
77 /* We have to keep the namespace clean. */
78 # define regfree(preg) __regfree (preg)
79 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
80 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
81 # define regerror(err_code, preg, errbuf, errbuf_size) \
82 __regerror (err_code, preg, errbuf, errbuf_size)
83 # define re_set_registers(bu, re, nu, st, en) \
84 __re_set_registers (bu, re, nu, st, en)
85 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
86 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
87 # define re_match(bufp, string, size, pos, regs) \
88 __re_match (bufp, string, size, pos, regs)
89 # define re_search(bufp, string, size, startpos, range, regs) \
90 __re_search (bufp, string, size, startpos, range, regs)
91 # define re_compile_pattern(pattern, length, bufp) \
92 __re_compile_pattern (pattern, length, bufp)
93 # define re_set_syntax(syntax) __re_set_syntax (syntax)
94 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
95 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
96 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
98 /* Make sure we call libc's function even if the user overrides them. */
99 # define btowc __btowc
100 # define iswctype __iswctype
101 # define wctype __wctype
103 # define WEAK_ALIAS(a,b) weak_alias (a, b)
105 /* We are also using some library internals. */
106 # include <locale/localeinfo.h>
107 # include <locale/elem-hash.h>
108 # include <langinfo.h>
110 # define WEAK_ALIAS(a,b)
113 /* This is for other GNU distributions with internationalized messages. */
114 #if HAVE_LIBINTL_H || defined _LIBC
115 # include <libintl.h>
117 # define gettext(msgid) (msgid)
121 /* This define is so xgettext can find the internationalizable
123 # define gettext_noop(String) String
126 /* The `emacs' switch turns on certain matching commands
127 that make sense only in Emacs. */
131 # include "character.h"
135 # include "category.h"
137 /* Make syntax table lookup grant data in gl_state. */
138 # define SYNTAX(c) syntax_property (c, 1)
143 # define malloc xmalloc
147 # define realloc xrealloc
153 /* Converts the pointer to the char to BEG-based offset from the start. */
154 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
155 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
157 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
158 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
159 # define RE_STRING_CHAR(p, multibyte) \
160 (multibyte ? (STRING_CHAR (p)) : (*(p)))
161 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
162 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
164 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
166 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
168 /* Set C a (possibly converted to multibyte) character before P. P
169 points into a string which is the virtual concatenation of STR1
170 (which ends at END1) or STR2 (which ends at END2). */
171 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
173 if (target_multibyte) \
175 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
176 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
177 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
178 c = STRING_CHAR (dtemp); \
182 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
183 (c) = RE_CHAR_TO_MULTIBYTE (c); \
187 /* Set C a (possibly converted to multibyte) character at P, and set
188 LEN to the byte length of that character. */
189 # define GET_CHAR_AFTER(c, p, len) \
191 if (target_multibyte) \
192 (c) = STRING_CHAR_AND_LENGTH (p, len); \
197 (c) = RE_CHAR_TO_MULTIBYTE (c); \
201 #else /* not emacs */
203 /* If we are not linking with Emacs proper,
204 we can't use the relocating allocator
205 even if config.h says that we can. */
210 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
213 xmalloc (size_t size
)
215 void *val
= malloc (size
);
218 write (2, "virtual memory exhausted\n", 25);
225 xrealloc (void *block
, size_t size
)
228 /* We must call malloc explicitly when BLOCK is 0, since some
229 reallocs don't do this. */
233 val
= realloc (block
, size
);
236 write (2, "virtual memory exhausted\n", 25);
245 # define malloc xmalloc
249 # define realloc xrealloc
251 # include <stdbool.h>
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 /* Dummy macros for non-Emacs environments. */
260 # define MAX_MULTIBYTE_LENGTH 1
261 # define RE_MULTIBYTE_P(x) 0
262 # define RE_TARGET_MULTIBYTE_P(x) 0
263 # define WORD_BOUNDARY_P(c1, c2) (0)
264 # define BYTES_BY_CHAR_HEAD(p) (1)
265 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
266 # define STRING_CHAR(p) (*(p))
267 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
268 # define CHAR_STRING(c, s) (*(s) = (c), 1)
269 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
270 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
271 # define RE_CHAR_TO_MULTIBYTE(c) (c)
272 # define RE_CHAR_TO_UNIBYTE(c) (c)
273 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
274 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
275 # define GET_CHAR_AFTER(c, p, len) \
277 # define CHAR_BYTE8_P(c) (0)
278 # define CHAR_LEADING_CODE(c) (c)
280 #endif /* not emacs */
283 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
284 # define RE_TRANSLATE_P(TBL) (TBL)
287 /* Get the interface, including the syntax bits. */
290 /* isalpha etc. are used for the character classes. */
295 /* 1 if C is an ASCII character. */
296 # define IS_REAL_ASCII(c) ((c) < 0200)
298 /* 1 if C is a unibyte character. */
299 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
301 /* The Emacs definitions should not be directly affected by locales. */
303 /* In Emacs, these are only used for single-byte characters. */
304 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
305 # define ISCNTRL(c) ((c) < ' ')
306 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
307 || ((c) >= 'a' && (c) <= 'f') \
308 || ((c) >= 'A' && (c) <= 'F'))
310 /* This is only used for single-byte characters. */
311 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
313 /* The rest must handle multibyte characters. */
315 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
316 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
319 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
320 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
323 # define ISALNUM(c) (IS_REAL_ASCII (c) \
324 ? (((c) >= 'a' && (c) <= 'z') \
325 || ((c) >= 'A' && (c) <= 'Z') \
326 || ((c) >= '0' && (c) <= '9')) \
327 : SYNTAX (c) == Sword)
329 # define ISALPHA(c) (IS_REAL_ASCII (c) \
330 ? (((c) >= 'a' && (c) <= 'z') \
331 || ((c) >= 'A' && (c) <= 'Z')) \
332 : SYNTAX (c) == Sword)
334 # define ISLOWER(c) lowercasep (c)
336 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
337 ? ((c) > ' ' && (c) < 0177 \
338 && !(((c) >= 'a' && (c) <= 'z') \
339 || ((c) >= 'A' && (c) <= 'Z') \
340 || ((c) >= '0' && (c) <= '9'))) \
341 : SYNTAX (c) != Sword)
343 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
345 # define ISUPPER(c) uppercasep (c)
347 # define ISWORD(c) (SYNTAX (c) == Sword)
349 #else /* not emacs */
351 /* 1 if C is an ASCII character. */
352 # define IS_REAL_ASCII(c) ((c) < 0200)
354 /* This distinction is not meaningful, except in Emacs. */
355 # define ISUNIBYTE(c) 1
358 # define ISBLANK(c) isblank (c)
360 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
363 # define ISGRAPH(c) isgraph (c)
365 # define ISGRAPH(c) (isprint (c) && !isspace (c))
368 /* Solaris defines ISPRINT so we must undefine it first. */
370 # define ISPRINT(c) isprint (c)
371 # define ISDIGIT(c) isdigit (c)
372 # define ISALNUM(c) isalnum (c)
373 # define ISALPHA(c) isalpha (c)
374 # define ISCNTRL(c) iscntrl (c)
375 # define ISLOWER(c) islower (c)
376 # define ISPUNCT(c) ispunct (c)
377 # define ISSPACE(c) isspace (c)
378 # define ISUPPER(c) isupper (c)
379 # define ISXDIGIT(c) isxdigit (c)
381 # define ISWORD(c) ISALPHA (c)
384 # define TOLOWER(c) _tolower (c)
386 # define TOLOWER(c) tolower (c)
389 /* How many characters in the character set. */
390 # define CHAR_SET_SIZE 256
394 extern char *re_syntax_table
;
396 # else /* not SYNTAX_TABLE */
398 static char re_syntax_table
[CHAR_SET_SIZE
];
401 init_syntax_once (void)
409 memset (re_syntax_table
, 0, sizeof re_syntax_table
);
411 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
413 re_syntax_table
[c
] = Sword
;
415 re_syntax_table
['_'] = Ssymbol
;
420 # endif /* not SYNTAX_TABLE */
422 # define SYNTAX(c) re_syntax_table[(c)]
424 #endif /* not emacs */
426 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
428 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
429 use `alloca' instead of `malloc'. This is because using malloc in
430 re_search* or re_match* could cause memory leaks when C-g is used in
431 Emacs; also, malloc is slower and causes storage fragmentation. On
432 the other hand, malloc is more portable, and easier to debug.
434 Because we sometimes use alloca, some routines have to be macros,
435 not functions -- `alloca'-allocated space disappears at the end of the
436 function it is called in. */
440 # define REGEX_ALLOCATE malloc
441 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
442 # define REGEX_FREE free
444 #else /* not REGEX_MALLOC */
446 /* Emacs already defines alloca, sometimes. */
449 /* Make alloca work the best possible way. */
451 # define alloca __builtin_alloca
452 # else /* not __GNUC__ */
453 # ifdef HAVE_ALLOCA_H
455 # endif /* HAVE_ALLOCA_H */
456 # endif /* not __GNUC__ */
458 # endif /* not alloca */
461 # define REGEX_USE_SAFE_ALLOCA USE_SAFE_ALLOCA
462 # define REGEX_SAFE_FREE() SAFE_FREE ()
463 # define REGEX_ALLOCATE SAFE_ALLOCA
465 # define REGEX_ALLOCATE alloca
468 /* Assumes a `char *destination' variable. */
469 # define REGEX_REALLOCATE(source, osize, nsize) \
470 (destination = REGEX_ALLOCATE (nsize), \
471 memcpy (destination, source, osize))
473 /* No need to do anything to free, after alloca. */
474 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
476 #endif /* not REGEX_MALLOC */
478 #ifndef REGEX_USE_SAFE_ALLOCA
479 # define REGEX_USE_SAFE_ALLOCA ((void) 0)
480 # define REGEX_SAFE_FREE() ((void) 0)
483 /* Define how to allocate the failure stack. */
485 #if defined REL_ALLOC && defined REGEX_MALLOC
487 # define REGEX_ALLOCATE_STACK(size) \
488 r_alloc (&failure_stack_ptr, (size))
489 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
490 r_re_alloc (&failure_stack_ptr, (nsize))
491 # define REGEX_FREE_STACK(ptr) \
492 r_alloc_free (&failure_stack_ptr)
494 #else /* not using relocating allocator */
496 # define REGEX_ALLOCATE_STACK(size) REGEX_ALLOCATE (size)
497 # define REGEX_REALLOCATE_STACK(source, o, n) REGEX_REALLOCATE (source, o, n)
498 # define REGEX_FREE_STACK(ptr) REGEX_FREE (ptr)
500 #endif /* not using relocating allocator */
503 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
504 `string1' or just past its end. This works if PTR is NULL, which is
506 #define FIRST_STRING_P(ptr) \
507 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
509 /* (Re)Allocate N items of type T using malloc, or fail. */
510 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
511 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
512 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
514 #define BYTEWIDTH 8 /* In bits. */
516 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
520 #define MAX(a, b) ((a) > (b) ? (a) : (b))
521 #define MIN(a, b) ((a) < (b) ? (a) : (b))
523 /* Type of source-pattern and string chars. */
525 typedef unsigned char re_char
;
526 typedef const re_char const_re_char
;
528 typedef const unsigned char re_char
;
529 typedef re_char const_re_char
;
532 typedef char boolean
;
534 static regoff_t
re_match_2_internal (struct re_pattern_buffer
*bufp
,
535 re_char
*string1
, size_t size1
,
536 re_char
*string2
, size_t size2
,
538 struct re_registers
*regs
,
541 /* These are the command codes that appear in compiled regular
542 expressions. Some opcodes are followed by argument bytes. A
543 command code can specify any interpretation whatsoever for its
544 arguments. Zero bytes may appear in the compiled regular expression. */
550 /* Succeed right away--no more backtracking. */
553 /* Followed by one byte giving n, then by n literal bytes. */
556 /* Matches any (more or less) character. */
559 /* Matches any one char belonging to specified set. First
560 following byte is number of bitmap bytes. Then come bytes
561 for a bitmap saying which chars are in. Bits in each byte
562 are ordered low-bit-first. A character is in the set if its
563 bit is 1. A character too large to have a bit in the map is
564 automatically not in the set.
566 If the length byte has the 0x80 bit set, then that stuff
567 is followed by a range table:
568 2 bytes of flags for character sets (low 8 bits, high 8 bits)
569 See RANGE_TABLE_WORK_BITS below.
570 2 bytes, the number of pairs that follow (upto 32767)
571 pairs, each 2 multibyte characters,
572 each multibyte character represented as 3 bytes. */
575 /* Same parameters as charset, but match any character that is
576 not one of those specified. */
579 /* Start remembering the text that is matched, for storing in a
580 register. Followed by one byte with the register number, in
581 the range 0 to one less than the pattern buffer's re_nsub
585 /* Stop remembering the text that is matched and store it in a
586 memory register. Followed by one byte with the register
587 number, in the range 0 to one less than `re_nsub' in the
591 /* Match a duplicate of something remembered. Followed by one
592 byte containing the register number. */
595 /* Fail unless at beginning of line. */
598 /* Fail unless at end of line. */
601 /* Succeeds if at beginning of buffer (if emacs) or at beginning
602 of string to be matched (if not). */
605 /* Analogously, for end of buffer/string. */
608 /* Followed by two byte relative address to which to jump. */
611 /* Followed by two-byte relative address of place to resume at
612 in case of failure. */
615 /* Like on_failure_jump, but pushes a placeholder instead of the
616 current string position when executed. */
617 on_failure_keep_string_jump
,
619 /* Just like `on_failure_jump', except that it checks that we
620 don't get stuck in an infinite loop (matching an empty string
622 on_failure_jump_loop
,
624 /* Just like `on_failure_jump_loop', except that it checks for
625 a different kind of loop (the kind that shows up with non-greedy
626 operators). This operation has to be immediately preceded
628 on_failure_jump_nastyloop
,
630 /* A smart `on_failure_jump' used for greedy * and + operators.
631 It analyzes the loop before which it is put and if the
632 loop does not require backtracking, it changes itself to
633 `on_failure_keep_string_jump' and short-circuits the loop,
634 else it just defaults to changing itself into `on_failure_jump'.
635 It assumes that it is pointing to just past a `jump'. */
636 on_failure_jump_smart
,
638 /* Followed by two-byte relative address and two-byte number n.
639 After matching N times, jump to the address upon failure.
640 Does not work if N starts at 0: use on_failure_jump_loop
644 /* Followed by two-byte relative address, and two-byte number n.
645 Jump to the address N times, then fail. */
648 /* Set the following two-byte relative address to the
649 subsequent two-byte number. The address *includes* the two
653 wordbeg
, /* Succeeds if at word beginning. */
654 wordend
, /* Succeeds if at word end. */
656 wordbound
, /* Succeeds if at a word boundary. */
657 notwordbound
, /* Succeeds if not at a word boundary. */
659 symbeg
, /* Succeeds if at symbol beginning. */
660 symend
, /* Succeeds if at symbol end. */
662 /* Matches any character whose syntax is specified. Followed by
663 a byte which contains a syntax code, e.g., Sword. */
666 /* Matches any character whose syntax is not that specified. */
670 ,before_dot
, /* Succeeds if before point. */
671 at_dot
, /* Succeeds if at point. */
672 after_dot
, /* Succeeds if after point. */
674 /* Matches any character whose category-set contains the specified
675 category. The operator is followed by a byte which contains a
676 category code (mnemonic ASCII character). */
679 /* Matches any character whose category-set does not contain the
680 specified category. The operator is followed by a byte which
681 contains the category code (mnemonic ASCII character). */
686 /* Common operations on the compiled pattern. */
688 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
690 #define STORE_NUMBER(destination, number) \
692 (destination)[0] = (number) & 0377; \
693 (destination)[1] = (number) >> 8; \
696 /* Same as STORE_NUMBER, except increment DESTINATION to
697 the byte after where the number is stored. Therefore, DESTINATION
698 must be an lvalue. */
700 #define STORE_NUMBER_AND_INCR(destination, number) \
702 STORE_NUMBER (destination, number); \
703 (destination) += 2; \
706 /* Put into DESTINATION a number stored in two contiguous bytes starting
709 #define EXTRACT_NUMBER(destination, source) \
710 ((destination) = extract_number (source))
713 extract_number (re_char
*source
)
715 unsigned leading_byte
= SIGN_EXTEND_CHAR (source
[1]);
716 return (leading_byte
<< 8) + source
[0];
719 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
720 SOURCE must be an lvalue. */
722 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
723 ((destination) = extract_number_and_incr (&source))
726 extract_number_and_incr (re_char
**source
)
728 int num
= extract_number (*source
);
733 /* Store a multibyte character in three contiguous bytes starting
734 DESTINATION, and increment DESTINATION to the byte after where the
735 character is stored. Therefore, DESTINATION must be an lvalue. */
737 #define STORE_CHARACTER_AND_INCR(destination, character) \
739 (destination)[0] = (character) & 0377; \
740 (destination)[1] = ((character) >> 8) & 0377; \
741 (destination)[2] = (character) >> 16; \
742 (destination) += 3; \
745 /* Put into DESTINATION a character stored in three contiguous bytes
746 starting at SOURCE. */
748 #define EXTRACT_CHARACTER(destination, source) \
750 (destination) = ((source)[0] \
751 | ((source)[1] << 8) \
752 | ((source)[2] << 16)); \
756 /* Macros for charset. */
758 /* Size of bitmap of charset P in bytes. P is a start of charset,
759 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
760 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
762 /* Nonzero if charset P has range table. */
763 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
765 /* Return the address of range table of charset P. But not the start
766 of table itself, but the before where the number of ranges is
767 stored. `2 +' means to skip re_opcode_t and size of bitmap,
768 and the 2 bytes of flags at the start of the range table. */
769 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
772 /* Extract the bit flags that start a range table. */
773 #define CHARSET_RANGE_TABLE_BITS(p) \
774 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
775 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
778 /* Return the address of end of RANGE_TABLE. COUNT is number of
779 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
780 is start of range and end of range. `* 3' is size of each start
782 #define CHARSET_RANGE_TABLE_END(range_table, count) \
783 ((range_table) + (count) * 2 * 3)
785 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
786 COUNT is number of ranges in RANGE_TABLE. */
787 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
790 re_wchar_t range_start, range_end; \
792 re_char *range_table_end \
793 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
795 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
797 EXTRACT_CHARACTER (range_start, rtp); \
798 EXTRACT_CHARACTER (range_end, rtp + 3); \
800 if (range_start <= (c) && (c) <= range_end) \
809 /* Test if C is in range table of CHARSET. The flag NOT is negated if
810 C is listed in it. */
811 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
814 /* Number of ranges in range table. */ \
816 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
818 EXTRACT_NUMBER_AND_INCR (count, range_table); \
819 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
823 /* If DEBUG is defined, Regex prints many voluminous messages about what
824 it is doing (if the variable `debug' is nonzero). If linked with the
825 main program in `iregex.c', you can enter patterns and strings
826 interactively. And if linked with the main program in `main.c' and
827 the other test files, you can run the already-written tests. */
831 /* We use standard I/O for debugging. */
834 /* It is useful to test things that ``must'' be true when debugging. */
837 static int debug
= -100000;
839 # define DEBUG_STATEMENT(e) e
840 # define DEBUG_PRINT(...) if (debug > 0) printf (__VA_ARGS__)
841 # define DEBUG_COMPILES_ARGUMENTS
842 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
843 if (debug > 0) print_partial_compiled_pattern (s, e)
844 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
845 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
848 /* Print the fastmap in human-readable form. */
851 print_fastmap (char *fastmap
)
853 unsigned was_a_range
= 0;
856 while (i
< (1 << BYTEWIDTH
))
862 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
878 /* Print a compiled pattern string in human-readable form, starting at
879 the START pointer into it and ending just before the pointer END. */
882 print_partial_compiled_pattern (re_char
*start
, re_char
*end
)
890 fprintf (stderr
, "(null)\n");
894 /* Loop over pattern commands. */
897 fprintf (stderr
, "%td:\t", p
- start
);
899 switch ((re_opcode_t
) *p
++)
902 fprintf (stderr
, "/no_op");
906 fprintf (stderr
, "/succeed");
911 fprintf (stderr
, "/exactn/%d", mcnt
);
914 fprintf (stderr
, "/%c", *p
++);
920 fprintf (stderr
, "/start_memory/%d", *p
++);
924 fprintf (stderr
, "/stop_memory/%d", *p
++);
928 fprintf (stderr
, "/duplicate/%d", *p
++);
932 fprintf (stderr
, "/anychar");
938 register int c
, last
= -100;
939 register int in_range
= 0;
940 int length
= CHARSET_BITMAP_SIZE (p
- 1);
941 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
943 fprintf (stderr
, "/charset [%s",
944 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
947 fprintf (stderr
, " !extends past end of pattern! ");
949 for (c
= 0; c
< 256; c
++)
951 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
953 /* Are we starting a range? */
954 if (last
+ 1 == c
&& ! in_range
)
956 fprintf (stderr
, "-");
959 /* Have we broken a range? */
960 else if (last
+ 1 != c
&& in_range
)
962 fprintf (stderr
, "%c", last
);
967 fprintf (stderr
, "%c", c
);
973 fprintf (stderr
, "%c", last
);
975 fprintf (stderr
, "]");
982 fprintf (stderr
, "has-range-table");
984 /* ??? Should print the range table; for now, just skip it. */
985 p
+= 2; /* skip range table bits */
986 EXTRACT_NUMBER_AND_INCR (count
, p
);
987 p
= CHARSET_RANGE_TABLE_END (p
, count
);
993 fprintf (stderr
, "/begline");
997 fprintf (stderr
, "/endline");
1000 case on_failure_jump
:
1001 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1002 fprintf (stderr
, "/on_failure_jump to %td", p
+ mcnt
- start
);
1005 case on_failure_keep_string_jump
:
1006 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1007 fprintf (stderr
, "/on_failure_keep_string_jump to %td",
1011 case on_failure_jump_nastyloop
:
1012 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1013 fprintf (stderr
, "/on_failure_jump_nastyloop to %td",
1017 case on_failure_jump_loop
:
1018 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1019 fprintf (stderr
, "/on_failure_jump_loop to %td",
1023 case on_failure_jump_smart
:
1024 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1025 fprintf (stderr
, "/on_failure_jump_smart to %td",
1030 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1031 fprintf (stderr
, "/jump to %td", p
+ mcnt
- start
);
1035 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1036 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1037 fprintf (stderr
, "/succeed_n to %td, %d times",
1038 p
- 2 + mcnt
- start
, mcnt2
);
1042 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1043 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1044 fprintf (stderr
, "/jump_n to %td, %d times",
1045 p
- 2 + mcnt
- start
, mcnt2
);
1049 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1050 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1051 fprintf (stderr
, "/set_number_at location %td to %d",
1052 p
- 2 + mcnt
- start
, mcnt2
);
1056 fprintf (stderr
, "/wordbound");
1060 fprintf (stderr
, "/notwordbound");
1064 fprintf (stderr
, "/wordbeg");
1068 fprintf (stderr
, "/wordend");
1072 fprintf (stderr
, "/symbeg");
1076 fprintf (stderr
, "/symend");
1080 fprintf (stderr
, "/syntaxspec");
1082 fprintf (stderr
, "/%d", mcnt
);
1086 fprintf (stderr
, "/notsyntaxspec");
1088 fprintf (stderr
, "/%d", mcnt
);
1093 fprintf (stderr
, "/before_dot");
1097 fprintf (stderr
, "/at_dot");
1101 fprintf (stderr
, "/after_dot");
1105 fprintf (stderr
, "/categoryspec");
1107 fprintf (stderr
, "/%d", mcnt
);
1110 case notcategoryspec
:
1111 fprintf (stderr
, "/notcategoryspec");
1113 fprintf (stderr
, "/%d", mcnt
);
1118 fprintf (stderr
, "/begbuf");
1122 fprintf (stderr
, "/endbuf");
1126 fprintf (stderr
, "?%d", *(p
-1));
1129 fprintf (stderr
, "\n");
1132 fprintf (stderr
, "%td:\tend of pattern.\n", p
- start
);
1137 print_compiled_pattern (struct re_pattern_buffer
*bufp
)
1139 re_char
*buffer
= bufp
->buffer
;
1141 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1142 printf ("%ld bytes used/%ld bytes allocated.\n",
1143 bufp
->used
, bufp
->allocated
);
1145 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1147 printf ("fastmap: ");
1148 print_fastmap (bufp
->fastmap
);
1151 printf ("re_nsub: %zu\t", bufp
->re_nsub
);
1152 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1153 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1154 printf ("no_sub: %d\t", bufp
->no_sub
);
1155 printf ("not_bol: %d\t", bufp
->not_bol
);
1156 printf ("not_eol: %d\t", bufp
->not_eol
);
1157 printf ("syntax: %lx\n", bufp
->syntax
);
1159 /* Perhaps we should print the translate table? */
1164 print_double_string (re_char
*where
, re_char
*string1
, ssize_t size1
,
1165 re_char
*string2
, ssize_t size2
)
1173 if (FIRST_STRING_P (where
))
1175 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1176 putchar (string1
[this_char
]);
1181 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1182 putchar (string2
[this_char
]);
1186 #else /* not DEBUG */
1191 # define DEBUG_STATEMENT(e)
1192 # define DEBUG_PRINT(...)
1193 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1194 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1196 #endif /* not DEBUG */
1198 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1200 # define IF_LINT(Code) Code
1202 # define IF_LINT(Code) /* empty */
1205 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1206 also be assigned to arbitrarily: each pattern buffer stores its own
1207 syntax, so it can be changed between regex compilations. */
1208 /* This has no initializer because initialized variables in Emacs
1209 become read-only after dumping. */
1210 reg_syntax_t re_syntax_options
;
1213 /* Specify the precise syntax of regexps for compilation. This provides
1214 for compatibility for various utilities which historically have
1215 different, incompatible syntaxes.
1217 The argument SYNTAX is a bit mask comprised of the various bits
1218 defined in regex.h. We return the old syntax. */
1221 re_set_syntax (reg_syntax_t syntax
)
1223 reg_syntax_t ret
= re_syntax_options
;
1225 re_syntax_options
= syntax
;
1228 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1230 /* Regexp to use to replace spaces, or NULL meaning don't. */
1231 static const_re_char
*whitespace_regexp
;
1234 re_set_whitespace_regexp (const char *regexp
)
1236 whitespace_regexp
= (const_re_char
*) regexp
;
1238 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1240 /* This table gives an error message for each of the error codes listed
1241 in regex.h. Obviously the order here has to be same as there.
1242 POSIX doesn't require that we do anything for REG_NOERROR,
1243 but why not be nice? */
1245 static const char *re_error_msgid
[] =
1247 gettext_noop ("Success"), /* REG_NOERROR */
1248 gettext_noop ("No match"), /* REG_NOMATCH */
1249 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1250 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1251 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1252 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1253 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1254 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1255 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1256 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1257 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1258 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1259 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1260 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1261 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1262 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1263 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1264 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1267 /* Avoiding alloca during matching, to placate r_alloc. */
1269 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1270 searching and matching functions should not call alloca. On some
1271 systems, alloca is implemented in terms of malloc, and if we're
1272 using the relocating allocator routines, then malloc could cause a
1273 relocation, which might (if the strings being searched are in the
1274 ralloc heap) shift the data out from underneath the regexp
1277 Here's another reason to avoid allocation: Emacs
1278 processes input from X in a signal handler; processing X input may
1279 call malloc; if input arrives while a matching routine is calling
1280 malloc, then we're scrod. But Emacs can't just block input while
1281 calling matching routines; then we don't notice interrupts when
1282 they come in. So, Emacs blocks input around all regexp calls
1283 except the matching calls, which it leaves unprotected, in the
1284 faith that they will not malloc. */
1286 /* Normally, this is fine. */
1287 #define MATCH_MAY_ALLOCATE
1289 /* The match routines may not allocate if (1) they would do it with malloc
1290 and (2) it's not safe for them to use malloc.
1291 Note that if REL_ALLOC is defined, matching would not use malloc for the
1292 failure stack, but we would still use it for the register vectors;
1293 so REL_ALLOC should not affect this. */
1294 #if defined REGEX_MALLOC && defined emacs
1295 # undef MATCH_MAY_ALLOCATE
1299 /* Failure stack declarations and macros; both re_compile_fastmap and
1300 re_match_2 use a failure stack. These have to be macros because of
1301 REGEX_ALLOCATE_STACK. */
1304 /* Approximate number of failure points for which to initially allocate space
1305 when matching. If this number is exceeded, we allocate more
1306 space, so it is not a hard limit. */
1307 #ifndef INIT_FAILURE_ALLOC
1308 # define INIT_FAILURE_ALLOC 20
1311 /* Roughly the maximum number of failure points on the stack. Would be
1312 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1313 This is a variable only so users of regex can assign to it; we never
1314 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1315 before using it, so it should probably be a byte-count instead. */
1316 # if defined MATCH_MAY_ALLOCATE
1317 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1318 whose default stack limit is 2mb. In order for a larger
1319 value to work reliably, you have to try to make it accord
1320 with the process stack limit. */
1321 size_t re_max_failures
= 40000;
1323 size_t re_max_failures
= 4000;
1326 union fail_stack_elt
1329 /* This should be the biggest `int' that's no bigger than a pointer. */
1333 typedef union fail_stack_elt fail_stack_elt_t
;
1337 fail_stack_elt_t
*stack
;
1339 size_t avail
; /* Offset of next open position. */
1340 size_t frame
; /* Offset of the cur constructed frame. */
1343 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1346 /* Define macros to initialize and free the failure stack.
1347 Do `return -2' if the alloc fails. */
1349 #ifdef MATCH_MAY_ALLOCATE
1350 # define INIT_FAIL_STACK() \
1352 fail_stack.stack = \
1353 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1354 * sizeof (fail_stack_elt_t)); \
1356 if (fail_stack.stack == NULL) \
1359 fail_stack.size = INIT_FAILURE_ALLOC; \
1360 fail_stack.avail = 0; \
1361 fail_stack.frame = 0; \
1364 # define INIT_FAIL_STACK() \
1366 fail_stack.avail = 0; \
1367 fail_stack.frame = 0; \
1370 # define RETALLOC_IF(addr, n, t) \
1371 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1375 /* Double the size of FAIL_STACK, up to a limit
1376 which allows approximately `re_max_failures' items.
1378 Return 1 if succeeds, and 0 if either ran out of memory
1379 allocating space for it or it was already too large.
1381 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1383 /* Factor to increase the failure stack size by
1384 when we increase it.
1385 This used to be 2, but 2 was too wasteful
1386 because the old discarded stacks added up to as much space
1387 were as ultimate, maximum-size stack. */
1388 #define FAIL_STACK_GROWTH_FACTOR 4
1390 #define GROW_FAIL_STACK(fail_stack) \
1391 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1392 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1394 : ((fail_stack).stack \
1395 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1396 (fail_stack).size * sizeof (fail_stack_elt_t), \
1397 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1398 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1399 * FAIL_STACK_GROWTH_FACTOR))), \
1401 (fail_stack).stack == NULL \
1403 : ((fail_stack).size \
1404 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1405 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1406 * FAIL_STACK_GROWTH_FACTOR)) \
1407 / sizeof (fail_stack_elt_t)), \
1411 /* Push a pointer value onto the failure stack.
1412 Assumes the variable `fail_stack'. Probably should only
1413 be called from within `PUSH_FAILURE_POINT'. */
1414 #define PUSH_FAILURE_POINTER(item) \
1415 fail_stack.stack[fail_stack.avail++].pointer = (item)
1417 /* This pushes an integer-valued item onto the failure stack.
1418 Assumes the variable `fail_stack'. Probably should only
1419 be called from within `PUSH_FAILURE_POINT'. */
1420 #define PUSH_FAILURE_INT(item) \
1421 fail_stack.stack[fail_stack.avail++].integer = (item)
1423 /* These POP... operations complement the PUSH... operations.
1424 All assume that `fail_stack' is nonempty. */
1425 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1426 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1428 /* Individual items aside from the registers. */
1429 #define NUM_NONREG_ITEMS 3
1431 /* Used to examine the stack (to detect infinite loops). */
1432 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1433 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1434 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1435 #define TOP_FAILURE_HANDLE() fail_stack.frame
1438 #define ENSURE_FAIL_STACK(space) \
1439 while (REMAINING_AVAIL_SLOTS <= space) { \
1440 if (!GROW_FAIL_STACK (fail_stack)) \
1442 DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\
1443 DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\
1446 /* Push register NUM onto the stack. */
1447 #define PUSH_FAILURE_REG(num) \
1449 char *destination; \
1451 ENSURE_FAIL_STACK(3); \
1452 DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \
1453 n, regstart[n], regend[n]); \
1454 PUSH_FAILURE_POINTER (regstart[n]); \
1455 PUSH_FAILURE_POINTER (regend[n]); \
1456 PUSH_FAILURE_INT (n); \
1459 /* Change the counter's value to VAL, but make sure that it will
1460 be reset when backtracking. */
1461 #define PUSH_NUMBER(ptr,val) \
1463 char *destination; \
1465 ENSURE_FAIL_STACK(3); \
1466 EXTRACT_NUMBER (c, ptr); \
1467 DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \
1468 PUSH_FAILURE_INT (c); \
1469 PUSH_FAILURE_POINTER (ptr); \
1470 PUSH_FAILURE_INT (-1); \
1471 STORE_NUMBER (ptr, val); \
1474 /* Pop a saved register off the stack. */
1475 #define POP_FAILURE_REG_OR_COUNT() \
1477 long pfreg = POP_FAILURE_INT (); \
1480 /* It's a counter. */ \
1481 /* Here, we discard `const', making re_match non-reentrant. */ \
1482 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1483 pfreg = POP_FAILURE_INT (); \
1484 STORE_NUMBER (ptr, pfreg); \
1485 DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \
1489 regend[pfreg] = POP_FAILURE_POINTER (); \
1490 regstart[pfreg] = POP_FAILURE_POINTER (); \
1491 DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \
1492 pfreg, regstart[pfreg], regend[pfreg]); \
1496 /* Check that we are not stuck in an infinite loop. */
1497 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1499 ssize_t failure = TOP_FAILURE_HANDLE (); \
1500 /* Check for infinite matching loops */ \
1501 while (failure > 0 \
1502 && (FAILURE_STR (failure) == string_place \
1503 || FAILURE_STR (failure) == NULL)) \
1505 assert (FAILURE_PAT (failure) >= bufp->buffer \
1506 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1507 if (FAILURE_PAT (failure) == pat_cur) \
1512 DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1513 failure = NEXT_FAILURE_HANDLE(failure); \
1515 DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \
1518 /* Push the information about the state we will need
1519 if we ever fail back to it.
1521 Requires variables fail_stack, regstart, regend and
1522 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1525 Does `return FAILURE_CODE' if runs out of memory. */
1527 #define PUSH_FAILURE_POINT(pattern, string_place) \
1529 char *destination; \
1530 /* Must be int, so when we don't save any registers, the arithmetic \
1531 of 0 + -1 isn't done as unsigned. */ \
1533 DEBUG_STATEMENT (nfailure_points_pushed++); \
1534 DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \
1535 DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \
1536 DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\
1538 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1540 DEBUG_PRINT ("\n"); \
1542 DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \
1543 PUSH_FAILURE_INT (fail_stack.frame); \
1545 DEBUG_PRINT (" Push string %p: `", string_place); \
1546 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1547 DEBUG_PRINT ("'\n"); \
1548 PUSH_FAILURE_POINTER (string_place); \
1550 DEBUG_PRINT (" Push pattern %p: ", pattern); \
1551 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1552 PUSH_FAILURE_POINTER (pattern); \
1554 /* Close the frame by moving the frame pointer past it. */ \
1555 fail_stack.frame = fail_stack.avail; \
1558 /* Estimate the size of data pushed by a typical failure stack entry.
1559 An estimate is all we need, because all we use this for
1560 is to choose a limit for how big to make the failure stack. */
1561 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1562 #define TYPICAL_FAILURE_SIZE 20
1564 /* How many items can still be added to the stack without overflowing it. */
1565 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1568 /* Pops what PUSH_FAIL_STACK pushes.
1570 We restore into the parameters, all of which should be lvalues:
1571 STR -- the saved data position.
1572 PAT -- the saved pattern position.
1573 REGSTART, REGEND -- arrays of string positions.
1575 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1576 `pend', `string1', `size1', `string2', and `size2'. */
1578 #define POP_FAILURE_POINT(str, pat) \
1580 assert (!FAIL_STACK_EMPTY ()); \
1582 /* Remove failure points and point to how many regs pushed. */ \
1583 DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \
1584 DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \
1585 DEBUG_PRINT (" size: %zd\n", fail_stack.size); \
1587 /* Pop the saved registers. */ \
1588 while (fail_stack.frame < fail_stack.avail) \
1589 POP_FAILURE_REG_OR_COUNT (); \
1591 pat = POP_FAILURE_POINTER (); \
1592 DEBUG_PRINT (" Popping pattern %p: ", pat); \
1593 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1595 /* If the saved string location is NULL, it came from an \
1596 on_failure_keep_string_jump opcode, and we want to throw away the \
1597 saved NULL, thus retaining our current position in the string. */ \
1598 str = POP_FAILURE_POINTER (); \
1599 DEBUG_PRINT (" Popping string %p: `", str); \
1600 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1601 DEBUG_PRINT ("'\n"); \
1603 fail_stack.frame = POP_FAILURE_INT (); \
1604 DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \
1606 assert (fail_stack.avail >= 0); \
1607 assert (fail_stack.frame <= fail_stack.avail); \
1609 DEBUG_STATEMENT (nfailure_points_popped++); \
1610 } while (0) /* POP_FAILURE_POINT */
1614 /* Registers are set to a sentinel when they haven't yet matched. */
1615 #define REG_UNSET(e) ((e) == NULL)
1617 /* Subroutine declarations and macros for regex_compile. */
1619 static reg_errcode_t
regex_compile (re_char
*pattern
, size_t size
,
1620 reg_syntax_t syntax
,
1621 struct re_pattern_buffer
*bufp
);
1622 static void store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
);
1623 static void store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
);
1624 static void insert_op1 (re_opcode_t op
, unsigned char *loc
,
1625 int arg
, unsigned char *end
);
1626 static void insert_op2 (re_opcode_t op
, unsigned char *loc
,
1627 int arg1
, int arg2
, unsigned char *end
);
1628 static boolean
at_begline_loc_p (re_char
*pattern
, re_char
*p
,
1629 reg_syntax_t syntax
);
1630 static boolean
at_endline_loc_p (re_char
*p
, re_char
*pend
,
1631 reg_syntax_t syntax
);
1632 static re_char
*skip_one_char (re_char
*p
);
1633 static int analyse_first (re_char
*p
, re_char
*pend
,
1634 char *fastmap
, const int multibyte
);
1636 /* Fetch the next character in the uncompiled pattern, with no
1638 #define PATFETCH(c) \
1641 if (p == pend) return REG_EEND; \
1642 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1647 /* If `translate' is non-null, return translate[D], else just D. We
1648 cast the subscript to translate because some data is declared as
1649 `char *', to avoid warnings when a string constant is passed. But
1650 when we use a character as a subscript we must make it unsigned. */
1652 # define TRANSLATE(d) \
1653 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1657 /* Macros for outputting the compiled pattern into `buffer'. */
1659 /* If the buffer isn't allocated when it comes in, use this. */
1660 #define INIT_BUF_SIZE 32
1662 /* Make sure we have at least N more bytes of space in buffer. */
1663 #define GET_BUFFER_SPACE(n) \
1664 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1667 /* Make sure we have one more byte of buffer space and then add C to it. */
1668 #define BUF_PUSH(c) \
1670 GET_BUFFER_SPACE (1); \
1671 *b++ = (unsigned char) (c); \
1675 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1676 #define BUF_PUSH_2(c1, c2) \
1678 GET_BUFFER_SPACE (2); \
1679 *b++ = (unsigned char) (c1); \
1680 *b++ = (unsigned char) (c2); \
1684 /* Store a jump with opcode OP at LOC to location TO. We store a
1685 relative address offset by the three bytes the jump itself occupies. */
1686 #define STORE_JUMP(op, loc, to) \
1687 store_op1 (op, loc, (to) - (loc) - 3)
1689 /* Likewise, for a two-argument jump. */
1690 #define STORE_JUMP2(op, loc, to, arg) \
1691 store_op2 (op, loc, (to) - (loc) - 3, arg)
1693 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1694 #define INSERT_JUMP(op, loc, to) \
1695 insert_op1 (op, loc, (to) - (loc) - 3, b)
1697 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1698 #define INSERT_JUMP2(op, loc, to, arg) \
1699 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1702 /* This is not an arbitrary limit: the arguments which represent offsets
1703 into the pattern are two bytes long. So if 2^15 bytes turns out to
1704 be too small, many things would have to change. */
1705 # define MAX_BUF_SIZE (1L << 15)
1707 /* Extend the buffer by twice its current size via realloc and
1708 reset the pointers that pointed into the old block to point to the
1709 correct places in the new one. If extending the buffer results in it
1710 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1711 #if __BOUNDED_POINTERS__
1712 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1713 # define MOVE_BUFFER_POINTER(P) \
1714 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1715 SET_HIGH_BOUND (P), \
1716 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1717 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1720 SET_HIGH_BOUND (b); \
1721 SET_HIGH_BOUND (begalt); \
1722 if (fixup_alt_jump) \
1723 SET_HIGH_BOUND (fixup_alt_jump); \
1725 SET_HIGH_BOUND (laststart); \
1726 if (pending_exact) \
1727 SET_HIGH_BOUND (pending_exact); \
1730 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1731 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1733 #define EXTEND_BUFFER() \
1735 unsigned char *old_buffer = bufp->buffer; \
1736 if (bufp->allocated == MAX_BUF_SIZE) \
1738 bufp->allocated <<= 1; \
1739 if (bufp->allocated > MAX_BUF_SIZE) \
1740 bufp->allocated = MAX_BUF_SIZE; \
1741 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1742 if (bufp->buffer == NULL) \
1743 return REG_ESPACE; \
1744 /* If the buffer moved, move all the pointers into it. */ \
1745 if (old_buffer != bufp->buffer) \
1747 unsigned char *new_buffer = bufp->buffer; \
1748 MOVE_BUFFER_POINTER (b); \
1749 MOVE_BUFFER_POINTER (begalt); \
1750 if (fixup_alt_jump) \
1751 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1753 MOVE_BUFFER_POINTER (laststart); \
1754 if (pending_exact) \
1755 MOVE_BUFFER_POINTER (pending_exact); \
1757 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1761 /* Since we have one byte reserved for the register number argument to
1762 {start,stop}_memory, the maximum number of groups we can report
1763 things about is what fits in that byte. */
1764 #define MAX_REGNUM 255
1766 /* But patterns can have more than `MAX_REGNUM' registers. We just
1767 ignore the excess. */
1768 typedef int regnum_t
;
1771 /* Macros for the compile stack. */
1773 /* Since offsets can go either forwards or backwards, this type needs to
1774 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1775 /* int may be not enough when sizeof(int) == 2. */
1776 typedef long pattern_offset_t
;
1780 pattern_offset_t begalt_offset
;
1781 pattern_offset_t fixup_alt_jump
;
1782 pattern_offset_t laststart_offset
;
1784 } compile_stack_elt_t
;
1789 compile_stack_elt_t
*stack
;
1791 size_t avail
; /* Offset of next open position. */
1792 } compile_stack_type
;
1795 #define INIT_COMPILE_STACK_SIZE 32
1797 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1798 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1800 /* The next available element. */
1801 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1803 /* Explicit quit checking is needed for Emacs, which uses polling to
1804 process input events. */
1806 # define IMMEDIATE_QUIT_CHECK \
1808 if (immediate_quit) QUIT; \
1811 # define IMMEDIATE_QUIT_CHECK ((void)0)
1814 /* Structure to manage work area for range table. */
1815 struct range_table_work_area
1817 int *table
; /* actual work area. */
1818 int allocated
; /* allocated size for work area in bytes. */
1819 int used
; /* actually used size in words. */
1820 int bits
; /* flag to record character classes */
1825 /* Make sure that WORK_AREA can hold more N multibyte characters.
1826 This is used only in set_image_of_range and set_image_of_range_1.
1827 It expects WORK_AREA to be a pointer.
1828 If it can't get the space, it returns from the surrounding function. */
1830 #define EXTEND_RANGE_TABLE(work_area, n) \
1832 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1834 extend_range_table_work_area (&work_area); \
1835 if ((work_area).table == 0) \
1836 return (REG_ESPACE); \
1840 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1841 (work_area).bits |= (bit)
1843 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1844 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1846 EXTEND_RANGE_TABLE ((work_area), 2); \
1847 (work_area).table[(work_area).used++] = (range_start); \
1848 (work_area).table[(work_area).used++] = (range_end); \
1853 /* Free allocated memory for WORK_AREA. */
1854 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1856 if ((work_area).table) \
1857 free ((work_area).table); \
1860 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1861 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1862 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1863 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1865 /* Bits used to implement the multibyte-part of the various character classes
1866 such as [:alnum:] in a charset's range table. */
1867 #define BIT_WORD 0x1
1868 #define BIT_LOWER 0x2
1869 #define BIT_PUNCT 0x4
1870 #define BIT_SPACE 0x8
1871 #define BIT_UPPER 0x10
1872 #define BIT_MULTIBYTE 0x20
1875 /* Set the bit for character C in a list. */
1876 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1881 /* Store characters in the range FROM to TO in the bitmap at B (for
1882 ASCII and unibyte characters) and WORK_AREA (for multibyte
1883 characters) while translating them and paying attention to the
1884 continuity of translated characters.
1886 Implementation note: It is better to implement these fairly big
1887 macros by a function, but it's not that easy because macros called
1888 in this macro assume various local variables already declared. */
1890 /* Both FROM and TO are ASCII characters. */
1892 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1896 for (C0 = (FROM); C0 <= (TO); C0++) \
1898 C1 = TRANSLATE (C0); \
1899 if (! ASCII_CHAR_P (C1)) \
1901 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1902 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1905 SET_LIST_BIT (C1); \
1910 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1912 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1914 int C0, C1, C2, I; \
1915 int USED = RANGE_TABLE_WORK_USED (work_area); \
1917 for (C0 = (FROM); C0 <= (TO); C0++) \
1919 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1920 if (CHAR_BYTE8_P (C1)) \
1921 SET_LIST_BIT (C0); \
1924 C2 = TRANSLATE (C1); \
1926 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1928 SET_LIST_BIT (C1); \
1929 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1931 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1932 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1934 if (C2 >= from - 1 && C2 <= to + 1) \
1936 if (C2 == from - 1) \
1937 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1938 else if (C2 == to + 1) \
1939 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1944 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1950 /* Both FROM and TO are multibyte characters. */
1952 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1954 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1956 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1957 for (C0 = (FROM); C0 <= (TO); C0++) \
1959 C1 = TRANSLATE (C0); \
1960 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1961 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1962 SET_LIST_BIT (C2); \
1963 if (C1 >= (FROM) && C1 <= (TO)) \
1965 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1967 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1968 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1970 if (C1 >= from - 1 && C1 <= to + 1) \
1972 if (C1 == from - 1) \
1973 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1974 else if (C1 == to + 1) \
1975 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1980 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1986 /* Get the next unsigned number in the uncompiled pattern. */
1987 #define GET_INTERVAL_COUNT(num) \
1990 FREE_STACK_RETURN (REG_EBRACE); \
1994 while ('0' <= c && c <= '9') \
1998 if (RE_DUP_MAX / 10 - (RE_DUP_MAX % 10 < c - '0') < num) \
1999 FREE_STACK_RETURN (REG_BADBR); \
2000 num = num * 10 + c - '0'; \
2002 FREE_STACK_RETURN (REG_EBRACE); \
2008 #if ! WIDE_CHAR_SUPPORT
2010 /* Map a string to the char class it names (if any). */
2012 re_wctype (const_re_char
*str
)
2014 const char *string
= (const char *) str
;
2015 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2016 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2017 else if (STREQ (string
, "word")) return RECC_WORD
;
2018 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2019 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2020 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2021 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2022 else if (STREQ (string
, "print")) return RECC_PRINT
;
2023 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2024 else if (STREQ (string
, "space")) return RECC_SPACE
;
2025 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2026 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2027 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2028 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2029 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2030 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2031 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2035 /* True if CH is in the char class CC. */
2037 re_iswctype (int ch
, re_wctype_t cc
)
2041 case RECC_ALNUM
: return ISALNUM (ch
) != 0;
2042 case RECC_ALPHA
: return ISALPHA (ch
) != 0;
2043 case RECC_BLANK
: return ISBLANK (ch
) != 0;
2044 case RECC_CNTRL
: return ISCNTRL (ch
) != 0;
2045 case RECC_DIGIT
: return ISDIGIT (ch
) != 0;
2046 case RECC_GRAPH
: return ISGRAPH (ch
) != 0;
2047 case RECC_LOWER
: return ISLOWER (ch
) != 0;
2048 case RECC_PRINT
: return ISPRINT (ch
) != 0;
2049 case RECC_PUNCT
: return ISPUNCT (ch
) != 0;
2050 case RECC_SPACE
: return ISSPACE (ch
) != 0;
2051 case RECC_UPPER
: return ISUPPER (ch
) != 0;
2052 case RECC_XDIGIT
: return ISXDIGIT (ch
) != 0;
2053 case RECC_ASCII
: return IS_REAL_ASCII (ch
) != 0;
2054 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2055 case RECC_UNIBYTE
: return ISUNIBYTE (ch
) != 0;
2056 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2057 case RECC_WORD
: return ISWORD (ch
) != 0;
2058 case RECC_ERROR
: return false;
2064 /* Return a bit-pattern to use in the range-table bits to match multibyte
2065 chars of class CC. */
2067 re_wctype_to_bit (re_wctype_t cc
)
2071 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2072 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2073 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2074 case RECC_LOWER
: return BIT_LOWER
;
2075 case RECC_UPPER
: return BIT_UPPER
;
2076 case RECC_PUNCT
: return BIT_PUNCT
;
2077 case RECC_SPACE
: return BIT_SPACE
;
2078 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2079 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2086 /* Filling in the work area of a range. */
2088 /* Actually extend the space in WORK_AREA. */
2091 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2093 work_area
->allocated
+= 16 * sizeof (int);
2094 work_area
->table
= realloc (work_area
->table
, work_area
->allocated
);
2100 /* Carefully find the ranges of codes that are equivalent
2101 under case conversion to the range start..end when passed through
2102 TRANSLATE. Handle the case where non-letters can come in between
2103 two upper-case letters (which happens in Latin-1).
2104 Also handle the case of groups of more than 2 case-equivalent chars.
2106 The basic method is to look at consecutive characters and see
2107 if they can form a run that can be handled as one.
2109 Returns -1 if successful, REG_ESPACE if ran out of space. */
2112 set_image_of_range_1 (struct range_table_work_area
*work_area
,
2113 re_wchar_t start
, re_wchar_t end
,
2114 RE_TRANSLATE_TYPE translate
)
2116 /* `one_case' indicates a character, or a run of characters,
2117 each of which is an isolate (no case-equivalents).
2118 This includes all ASCII non-letters.
2120 `two_case' indicates a character, or a run of characters,
2121 each of which has two case-equivalent forms.
2122 This includes all ASCII letters.
2124 `strange' indicates a character that has more than one
2127 enum case_type
{one_case
, two_case
, strange
};
2129 /* Describe the run that is in progress,
2130 which the next character can try to extend.
2131 If run_type is strange, that means there really is no run.
2132 If run_type is one_case, then run_start...run_end is the run.
2133 If run_type is two_case, then the run is run_start...run_end,
2134 and the case-equivalents end at run_eqv_end. */
2136 enum case_type run_type
= strange
;
2137 int run_start
, run_end
, run_eqv_end
;
2139 Lisp_Object eqv_table
;
2141 if (!RE_TRANSLATE_P (translate
))
2143 EXTEND_RANGE_TABLE (work_area
, 2);
2144 work_area
->table
[work_area
->used
++] = (start
);
2145 work_area
->table
[work_area
->used
++] = (end
);
2149 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2151 for (; start
<= end
; start
++)
2153 enum case_type this_type
;
2154 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2155 int minchar
, maxchar
;
2157 /* Classify this character */
2159 this_type
= one_case
;
2160 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2161 this_type
= two_case
;
2163 this_type
= strange
;
2166 minchar
= start
, maxchar
= eqv
;
2168 minchar
= eqv
, maxchar
= start
;
2170 /* Can this character extend the run in progress? */
2171 if (this_type
== strange
|| this_type
!= run_type
2172 || !(minchar
== run_end
+ 1
2173 && (run_type
== two_case
2174 ? maxchar
== run_eqv_end
+ 1 : 1)))
2177 Record each of its equivalent ranges. */
2178 if (run_type
== one_case
)
2180 EXTEND_RANGE_TABLE (work_area
, 2);
2181 work_area
->table
[work_area
->used
++] = run_start
;
2182 work_area
->table
[work_area
->used
++] = run_end
;
2184 else if (run_type
== two_case
)
2186 EXTEND_RANGE_TABLE (work_area
, 4);
2187 work_area
->table
[work_area
->used
++] = run_start
;
2188 work_area
->table
[work_area
->used
++] = run_end
;
2189 work_area
->table
[work_area
->used
++]
2190 = RE_TRANSLATE (eqv_table
, run_start
);
2191 work_area
->table
[work_area
->used
++]
2192 = RE_TRANSLATE (eqv_table
, run_end
);
2197 if (this_type
== strange
)
2199 /* For a strange character, add each of its equivalents, one
2200 by one. Don't start a range. */
2203 EXTEND_RANGE_TABLE (work_area
, 2);
2204 work_area
->table
[work_area
->used
++] = eqv
;
2205 work_area
->table
[work_area
->used
++] = eqv
;
2206 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2208 while (eqv
!= start
);
2211 /* Add this char to the run, or start a new run. */
2212 else if (run_type
== strange
)
2214 /* Initialize a new range. */
2215 run_type
= this_type
;
2218 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2222 /* Extend a running range. */
2224 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2228 /* If a run is still in progress at the end, finish it now
2229 by recording its equivalent ranges. */
2230 if (run_type
== one_case
)
2232 EXTEND_RANGE_TABLE (work_area
, 2);
2233 work_area
->table
[work_area
->used
++] = run_start
;
2234 work_area
->table
[work_area
->used
++] = run_end
;
2236 else if (run_type
== two_case
)
2238 EXTEND_RANGE_TABLE (work_area
, 4);
2239 work_area
->table
[work_area
->used
++] = run_start
;
2240 work_area
->table
[work_area
->used
++] = run_end
;
2241 work_area
->table
[work_area
->used
++]
2242 = RE_TRANSLATE (eqv_table
, run_start
);
2243 work_area
->table
[work_area
->used
++]
2244 = RE_TRANSLATE (eqv_table
, run_end
);
2252 /* Record the image of the range start..end when passed through
2253 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2254 and is not even necessarily contiguous.
2255 Normally we approximate it with the smallest contiguous range that contains
2256 all the chars we need. However, for Latin-1 we go to extra effort
2259 This function is not called for ASCII ranges.
2261 Returns -1 if successful, REG_ESPACE if ran out of space. */
2264 set_image_of_range (struct range_table_work_area
*work_area
,
2265 re_wchar_t start
, re_wchar_t end
,
2266 RE_TRANSLATE_TYPE translate
)
2268 re_wchar_t cmin
, cmax
;
2271 /* For Latin-1 ranges, use set_image_of_range_1
2272 to get proper handling of ranges that include letters and nonletters.
2273 For a range that includes the whole of Latin-1, this is not necessary.
2274 For other character sets, we don't bother to get this right. */
2275 if (RE_TRANSLATE_P (translate
) && start
< 04400
2276 && !(start
< 04200 && end
>= 04377))
2283 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2293 EXTEND_RANGE_TABLE (work_area
, 2);
2294 work_area
->table
[work_area
->used
++] = (start
);
2295 work_area
->table
[work_area
->used
++] = (end
);
2297 cmin
= -1, cmax
= -1;
2299 if (RE_TRANSLATE_P (translate
))
2303 for (ch
= start
; ch
<= end
; ch
++)
2305 re_wchar_t c
= TRANSLATE (ch
);
2306 if (! (start
<= c
&& c
<= end
))
2312 cmin
= MIN (cmin
, c
);
2313 cmax
= MAX (cmax
, c
);
2320 EXTEND_RANGE_TABLE (work_area
, 2);
2321 work_area
->table
[work_area
->used
++] = (cmin
);
2322 work_area
->table
[work_area
->used
++] = (cmax
);
2330 #ifndef MATCH_MAY_ALLOCATE
2332 /* If we cannot allocate large objects within re_match_2_internal,
2333 we make the fail stack and register vectors global.
2334 The fail stack, we grow to the maximum size when a regexp
2336 The register vectors, we adjust in size each time we
2337 compile a regexp, according to the number of registers it needs. */
2339 static fail_stack_type fail_stack
;
2341 /* Size with which the following vectors are currently allocated.
2342 That is so we can make them bigger as needed,
2343 but never make them smaller. */
2344 static int regs_allocated_size
;
2346 static re_char
** regstart
, ** regend
;
2347 static re_char
**best_regstart
, **best_regend
;
2349 /* Make the register vectors big enough for NUM_REGS registers,
2350 but don't make them smaller. */
2353 regex_grow_registers (int num_regs
)
2355 if (num_regs
> regs_allocated_size
)
2357 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2358 RETALLOC_IF (regend
, num_regs
, re_char
*);
2359 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2360 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2362 regs_allocated_size
= num_regs
;
2366 #endif /* not MATCH_MAY_ALLOCATE */
2368 static boolean
group_in_compile_stack (compile_stack_type compile_stack
,
2371 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2372 Returns one of error codes defined in `regex.h', or zero for success.
2374 Assumes the `allocated' (and perhaps `buffer') and `translate'
2375 fields are set in BUFP on entry.
2377 If it succeeds, results are put in BUFP (if it returns an error, the
2378 contents of BUFP are undefined):
2379 `buffer' is the compiled pattern;
2380 `syntax' is set to SYNTAX;
2381 `used' is set to the length of the compiled pattern;
2382 `fastmap_accurate' is zero;
2383 `re_nsub' is the number of subexpressions in PATTERN;
2384 `not_bol' and `not_eol' are zero;
2386 The `fastmap' field is neither examined nor set. */
2388 /* Insert the `jump' from the end of last alternative to "here".
2389 The space for the jump has already been allocated. */
2390 #define FIXUP_ALT_JUMP() \
2392 if (fixup_alt_jump) \
2393 STORE_JUMP (jump, fixup_alt_jump, b); \
2397 /* Return, freeing storage we allocated. */
2398 #define FREE_STACK_RETURN(value) \
2400 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2401 free (compile_stack.stack); \
2405 static reg_errcode_t
2406 regex_compile (const_re_char
*pattern
, size_t size
, reg_syntax_t syntax
,
2407 struct re_pattern_buffer
*bufp
)
2409 /* We fetch characters from PATTERN here. */
2410 register re_wchar_t c
, c1
;
2412 /* Points to the end of the buffer, where we should append. */
2413 register unsigned char *b
;
2415 /* Keeps track of unclosed groups. */
2416 compile_stack_type compile_stack
;
2418 /* Points to the current (ending) position in the pattern. */
2420 /* `const' makes AIX compiler fail. */
2421 unsigned char *p
= pattern
;
2423 re_char
*p
= pattern
;
2425 re_char
*pend
= pattern
+ size
;
2427 /* How to translate the characters in the pattern. */
2428 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2430 /* Address of the count-byte of the most recently inserted `exactn'
2431 command. This makes it possible to tell if a new exact-match
2432 character can be added to that command or if the character requires
2433 a new `exactn' command. */
2434 unsigned char *pending_exact
= 0;
2436 /* Address of start of the most recently finished expression.
2437 This tells, e.g., postfix * where to find the start of its
2438 operand. Reset at the beginning of groups and alternatives. */
2439 unsigned char *laststart
= 0;
2441 /* Address of beginning of regexp, or inside of last group. */
2442 unsigned char *begalt
;
2444 /* Place in the uncompiled pattern (i.e., the {) to
2445 which to go back if the interval is invalid. */
2446 re_char
*beg_interval
;
2448 /* Address of the place where a forward jump should go to the end of
2449 the containing expression. Each alternative of an `or' -- except the
2450 last -- ends with a forward jump of this sort. */
2451 unsigned char *fixup_alt_jump
= 0;
2453 /* Work area for range table of charset. */
2454 struct range_table_work_area range_table_work
;
2456 /* If the object matched can contain multibyte characters. */
2457 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2459 /* Nonzero if we have pushed down into a subpattern. */
2460 int in_subpattern
= 0;
2462 /* These hold the values of p, pattern, and pend from the main
2463 pattern when we have pushed into a subpattern. */
2464 re_char
*main_p
IF_LINT (= NULL
);
2465 re_char
*main_pattern
IF_LINT (= NULL
);
2466 re_char
*main_pend
IF_LINT (= NULL
);
2470 DEBUG_PRINT ("\nCompiling pattern: ");
2473 unsigned debug_count
;
2475 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2476 putchar (pattern
[debug_count
]);
2481 /* Initialize the compile stack. */
2482 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2483 if (compile_stack
.stack
== NULL
)
2486 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2487 compile_stack
.avail
= 0;
2489 range_table_work
.table
= 0;
2490 range_table_work
.allocated
= 0;
2492 /* Initialize the pattern buffer. */
2493 bufp
->syntax
= syntax
;
2494 bufp
->fastmap_accurate
= 0;
2495 bufp
->not_bol
= bufp
->not_eol
= 0;
2496 bufp
->used_syntax
= 0;
2498 /* Set `used' to zero, so that if we return an error, the pattern
2499 printer (for debugging) will think there's no pattern. We reset it
2503 /* Always count groups, whether or not bufp->no_sub is set. */
2506 #if !defined emacs && !defined SYNTAX_TABLE
2507 /* Initialize the syntax table. */
2508 init_syntax_once ();
2511 if (bufp
->allocated
== 0)
2514 { /* If zero allocated, but buffer is non-null, try to realloc
2515 enough space. This loses if buffer's address is bogus, but
2516 that is the user's responsibility. */
2517 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2520 { /* Caller did not allocate a buffer. Do it for them. */
2521 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2523 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2525 bufp
->allocated
= INIT_BUF_SIZE
;
2528 begalt
= b
= bufp
->buffer
;
2530 /* Loop through the uncompiled pattern until we're at the end. */
2535 /* If this is the end of an included regexp,
2536 pop back to the main regexp and try again. */
2540 pattern
= main_pattern
;
2545 /* If this is the end of the main regexp, we are done. */
2557 /* If there's no special whitespace regexp, treat
2558 spaces normally. And don't try to do this recursively. */
2559 if (!whitespace_regexp
|| in_subpattern
)
2562 /* Peek past following spaces. */
2569 /* If the spaces are followed by a repetition op,
2570 treat them normally. */
2572 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2573 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2576 /* Replace the spaces with the whitespace regexp. */
2580 main_pattern
= pattern
;
2581 p
= pattern
= whitespace_regexp
;
2582 pend
= p
+ strlen ((const char *) p
);
2588 if ( /* If at start of pattern, it's an operator. */
2590 /* If context independent, it's an operator. */
2591 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2592 /* Otherwise, depends on what's come before. */
2593 || at_begline_loc_p (pattern
, p
, syntax
))
2594 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2603 if ( /* If at end of pattern, it's an operator. */
2605 /* If context independent, it's an operator. */
2606 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2607 /* Otherwise, depends on what's next. */
2608 || at_endline_loc_p (p
, pend
, syntax
))
2609 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2618 if ((syntax
& RE_BK_PLUS_QM
)
2619 || (syntax
& RE_LIMITED_OPS
))
2623 /* If there is no previous pattern... */
2626 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2627 FREE_STACK_RETURN (REG_BADRPT
);
2628 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2633 /* 1 means zero (many) matches is allowed. */
2634 boolean zero_times_ok
= 0, many_times_ok
= 0;
2637 /* If there is a sequence of repetition chars, collapse it
2638 down to just one (the right one). We can't combine
2639 interval operators with these because of, e.g., `a{2}*',
2640 which should only match an even number of `a's. */
2644 if ((syntax
& RE_FRUGAL
)
2645 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2649 zero_times_ok
|= c
!= '+';
2650 many_times_ok
|= c
!= '?';
2656 || (!(syntax
& RE_BK_PLUS_QM
)
2657 && (*p
== '+' || *p
== '?')))
2659 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2662 FREE_STACK_RETURN (REG_EESCAPE
);
2663 if (p
[1] == '+' || p
[1] == '?')
2664 PATFETCH (c
); /* Gobble up the backslash. */
2670 /* If we get here, we found another repeat character. */
2674 /* Star, etc. applied to an empty pattern is equivalent
2675 to an empty pattern. */
2676 if (!laststart
|| laststart
== b
)
2679 /* Now we know whether or not zero matches is allowed
2680 and also whether or not two or more matches is allowed. */
2685 boolean simple
= skip_one_char (laststart
) == b
;
2686 size_t startoffset
= 0;
2688 /* Check if the loop can match the empty string. */
2689 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2690 ? on_failure_jump
: on_failure_jump_loop
;
2691 assert (skip_one_char (laststart
) <= b
);
2693 if (!zero_times_ok
&& simple
)
2694 { /* Since simple * loops can be made faster by using
2695 on_failure_keep_string_jump, we turn simple P+
2696 into PP* if P is simple. */
2697 unsigned char *p1
, *p2
;
2698 startoffset
= b
- laststart
;
2699 GET_BUFFER_SPACE (startoffset
);
2700 p1
= b
; p2
= laststart
;
2706 GET_BUFFER_SPACE (6);
2709 STORE_JUMP (ofj
, b
, b
+ 6);
2711 /* Simple * loops can use on_failure_keep_string_jump
2712 depending on what follows. But since we don't know
2713 that yet, we leave the decision up to
2714 on_failure_jump_smart. */
2715 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2716 laststart
+ startoffset
, b
+ 6);
2718 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2723 /* A simple ? pattern. */
2724 assert (zero_times_ok
);
2725 GET_BUFFER_SPACE (3);
2726 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2730 else /* not greedy */
2731 { /* I wish the greedy and non-greedy cases could be merged. */
2733 GET_BUFFER_SPACE (7); /* We might use less. */
2736 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2738 /* The non-greedy multiple match looks like
2739 a repeat..until: we only need a conditional jump
2740 at the end of the loop. */
2741 if (emptyp
) BUF_PUSH (no_op
);
2742 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2743 : on_failure_jump
, b
, laststart
);
2747 /* The repeat...until naturally matches one or more.
2748 To also match zero times, we need to first jump to
2749 the end of the loop (its conditional jump). */
2750 INSERT_JUMP (jump
, laststart
, b
);
2756 /* non-greedy a?? */
2757 INSERT_JUMP (jump
, laststart
, b
+ 3);
2759 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2778 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2780 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2782 /* Ensure that we have enough space to push a charset: the
2783 opcode, the length count, and the bitset; 34 bytes in all. */
2784 GET_BUFFER_SPACE (34);
2788 /* We test `*p == '^' twice, instead of using an if
2789 statement, so we only need one BUF_PUSH. */
2790 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2794 /* Remember the first position in the bracket expression. */
2797 /* Push the number of bytes in the bitmap. */
2798 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2800 /* Clear the whole map. */
2801 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2803 /* charset_not matches newline according to a syntax bit. */
2804 if ((re_opcode_t
) b
[-2] == charset_not
2805 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2806 SET_LIST_BIT ('\n');
2808 /* Read in characters and ranges, setting map bits. */
2811 boolean escaped_char
= false;
2812 const unsigned char *p2
= p
;
2815 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2817 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2818 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2819 So the translation is done later in a loop. Example:
2820 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2823 /* \ might escape characters inside [...] and [^...]. */
2824 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2826 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2829 escaped_char
= true;
2833 /* Could be the end of the bracket expression. If it's
2834 not (i.e., when the bracket expression is `[]' so
2835 far), the ']' character bit gets set way below. */
2836 if (c
== ']' && p2
!= p1
)
2840 /* See if we're at the beginning of a possible character
2843 if (!escaped_char
&&
2844 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2846 /* Leave room for the null. */
2847 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2848 const unsigned char *class_beg
;
2854 /* If pattern is `[[:'. */
2855 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2860 if ((c
== ':' && *p
== ']') || p
== pend
)
2862 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2865 /* This is in any case an invalid class name. */
2870 /* If isn't a word bracketed by `[:' and `:]':
2871 undo the ending character, the letters, and
2872 leave the leading `:' and `[' (but set bits for
2874 if (c
== ':' && *p
== ']')
2876 re_wctype_t cc
= re_wctype (str
);
2879 FREE_STACK_RETURN (REG_ECTYPE
);
2881 /* Throw away the ] at the end of the character
2885 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2888 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2889 if (re_iswctype (btowc (ch
), cc
))
2892 if (c
< (1 << BYTEWIDTH
))
2896 /* Most character classes in a multibyte match
2897 just set a flag. Exceptions are is_blank,
2898 is_digit, is_cntrl, and is_xdigit, since
2899 they can only match ASCII characters. We
2900 don't need to handle them for multibyte.
2901 They are distinguished by a negative wctype. */
2903 /* Setup the gl_state object to its buffer-defined
2904 value. This hardcodes the buffer-global
2905 syntax-table for ASCII chars, while the other chars
2906 will obey syntax-table properties. It's not ideal,
2907 but it's the way it's been done until now. */
2908 SETUP_BUFFER_SYNTAX_TABLE ();
2910 for (ch
= 0; ch
< 256; ++ch
)
2912 c
= RE_CHAR_TO_MULTIBYTE (ch
);
2913 if (! CHAR_BYTE8_P (c
)
2914 && re_iswctype (c
, cc
))
2920 if (ASCII_CHAR_P (c1
))
2922 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
2926 SET_RANGE_TABLE_WORK_AREA_BIT
2927 (range_table_work
, re_wctype_to_bit (cc
));
2929 /* In most cases the matching rule for char classes
2930 only uses the syntax table for multibyte chars,
2931 so that the content of the syntax-table it is not
2932 hardcoded in the range_table. SPACE and WORD are
2933 the two exceptions. */
2934 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2935 bufp
->used_syntax
= 1;
2937 /* Repeat the loop. */
2942 /* Go back to right after the "[:". */
2946 /* Because the `:' may starts the range, we
2947 can't simply set bit and repeat the loop.
2948 Instead, just set it to C and handle below. */
2953 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2956 /* Discard the `-'. */
2959 /* Fetch the character which ends the range. */
2962 if (CHAR_BYTE8_P (c1
)
2963 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
2964 /* Treat the range from a multibyte character to
2965 raw-byte character as empty. */
2970 /* Range from C to C. */
2975 if (syntax
& RE_NO_EMPTY_RANGES
)
2976 FREE_STACK_RETURN (REG_ERANGEX
);
2977 /* Else, repeat the loop. */
2982 /* Set the range into bitmap */
2983 for (; c
<= c1
; c
++)
2986 if (ch
< (1 << BYTEWIDTH
))
2993 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
2995 if (CHAR_BYTE8_P (c1
))
2996 c
= BYTE8_TO_CHAR (128);
3000 if (CHAR_BYTE8_P (c
))
3002 c
= CHAR_TO_BYTE8 (c
);
3003 c1
= CHAR_TO_BYTE8 (c1
);
3004 for (; c
<= c1
; c
++)
3009 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
3013 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
3020 /* Discard any (non)matching list bytes that are all 0 at the
3021 end of the map. Decrease the map-length byte too. */
3022 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3026 /* Build real range table from work area. */
3027 if (RANGE_TABLE_WORK_USED (range_table_work
)
3028 || RANGE_TABLE_WORK_BITS (range_table_work
))
3031 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3033 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3034 bytes for flags, two for COUNT, and three bytes for
3036 GET_BUFFER_SPACE (4 + used
* 3);
3038 /* Indicate the existence of range table. */
3039 laststart
[1] |= 0x80;
3041 /* Store the character class flag bits into the range table.
3042 If not in emacs, these flag bits are always 0. */
3043 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3044 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3046 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3047 for (i
= 0; i
< used
; i
++)
3048 STORE_CHARACTER_AND_INCR
3049 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3056 if (syntax
& RE_NO_BK_PARENS
)
3063 if (syntax
& RE_NO_BK_PARENS
)
3070 if (syntax
& RE_NEWLINE_ALT
)
3077 if (syntax
& RE_NO_BK_VBAR
)
3084 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3085 goto handle_interval
;
3091 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3093 /* Do not translate the character after the \, so that we can
3094 distinguish, e.g., \B from \b, even if we normally would
3095 translate, e.g., B to b. */
3101 if (syntax
& RE_NO_BK_PARENS
)
3102 goto normal_backslash
;
3107 regnum_t regnum
= 0;
3110 /* Look for a special (?...) construct */
3111 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3113 PATFETCH (c
); /* Gobble up the '?'. */
3119 case ':': shy
= 1; break;
3121 /* An explicitly specified regnum must start
3124 FREE_STACK_RETURN (REG_BADPAT
);
3125 case '1': case '2': case '3': case '4':
3126 case '5': case '6': case '7': case '8': case '9':
3127 regnum
= 10*regnum
+ (c
- '0'); break;
3129 /* Only (?:...) is supported right now. */
3130 FREE_STACK_RETURN (REG_BADPAT
);
3137 regnum
= ++bufp
->re_nsub
;
3139 { /* It's actually not shy, but explicitly numbered. */
3141 if (regnum
> bufp
->re_nsub
)
3142 bufp
->re_nsub
= regnum
;
3143 else if (regnum
> bufp
->re_nsub
3144 /* Ideally, we'd want to check that the specified
3145 group can't have matched (i.e. all subgroups
3146 using the same regnum are in other branches of
3147 OR patterns), but we don't currently keep track
3148 of enough info to do that easily. */
3149 || group_in_compile_stack (compile_stack
, regnum
))
3150 FREE_STACK_RETURN (REG_BADPAT
);
3153 /* It's really shy. */
3154 regnum
= - bufp
->re_nsub
;
3156 if (COMPILE_STACK_FULL
)
3158 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3159 compile_stack_elt_t
);
3160 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3162 compile_stack
.size
<<= 1;
3165 /* These are the values to restore when we hit end of this
3166 group. They are all relative offsets, so that if the
3167 whole pattern moves because of realloc, they will still
3169 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3170 COMPILE_STACK_TOP
.fixup_alt_jump
3171 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3172 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3173 COMPILE_STACK_TOP
.regnum
= regnum
;
3175 /* Do not push a start_memory for groups beyond the last one
3176 we can represent in the compiled pattern. */
3177 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3178 BUF_PUSH_2 (start_memory
, regnum
);
3180 compile_stack
.avail
++;
3185 /* If we've reached MAX_REGNUM groups, then this open
3186 won't actually generate any code, so we'll have to
3187 clear pending_exact explicitly. */
3193 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3195 if (COMPILE_STACK_EMPTY
)
3197 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3198 goto normal_backslash
;
3200 FREE_STACK_RETURN (REG_ERPAREN
);
3206 /* See similar code for backslashed left paren above. */
3207 if (COMPILE_STACK_EMPTY
)
3209 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3212 FREE_STACK_RETURN (REG_ERPAREN
);
3215 /* Since we just checked for an empty stack above, this
3216 ``can't happen''. */
3217 assert (compile_stack
.avail
!= 0);
3219 /* We don't just want to restore into `regnum', because
3220 later groups should continue to be numbered higher,
3221 as in `(ab)c(de)' -- the second group is #2. */
3224 compile_stack
.avail
--;
3225 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3227 = COMPILE_STACK_TOP
.fixup_alt_jump
3228 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3230 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3231 regnum
= COMPILE_STACK_TOP
.regnum
;
3232 /* If we've reached MAX_REGNUM groups, then this open
3233 won't actually generate any code, so we'll have to
3234 clear pending_exact explicitly. */
3237 /* We're at the end of the group, so now we know how many
3238 groups were inside this one. */
3239 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3240 BUF_PUSH_2 (stop_memory
, regnum
);
3245 case '|': /* `\|'. */
3246 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3247 goto normal_backslash
;
3249 if (syntax
& RE_LIMITED_OPS
)
3252 /* Insert before the previous alternative a jump which
3253 jumps to this alternative if the former fails. */
3254 GET_BUFFER_SPACE (3);
3255 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3259 /* The alternative before this one has a jump after it
3260 which gets executed if it gets matched. Adjust that
3261 jump so it will jump to this alternative's analogous
3262 jump (put in below, which in turn will jump to the next
3263 (if any) alternative's such jump, etc.). The last such
3264 jump jumps to the correct final destination. A picture:
3270 If we are at `b', then fixup_alt_jump right now points to a
3271 three-byte space after `a'. We'll put in the jump, set
3272 fixup_alt_jump to right after `b', and leave behind three
3273 bytes which we'll fill in when we get to after `c'. */
3277 /* Mark and leave space for a jump after this alternative,
3278 to be filled in later either by next alternative or
3279 when know we're at the end of a series of alternatives. */
3281 GET_BUFFER_SPACE (3);
3290 /* If \{ is a literal. */
3291 if (!(syntax
& RE_INTERVALS
)
3292 /* If we're at `\{' and it's not the open-interval
3294 || (syntax
& RE_NO_BK_BRACES
))
3295 goto normal_backslash
;
3299 /* If got here, then the syntax allows intervals. */
3301 /* At least (most) this many matches must be made. */
3302 int lower_bound
= 0, upper_bound
= -1;
3306 GET_INTERVAL_COUNT (lower_bound
);
3309 GET_INTERVAL_COUNT (upper_bound
);
3311 /* Interval such as `{1}' => match exactly once. */
3312 upper_bound
= lower_bound
;
3315 || (0 <= upper_bound
&& upper_bound
< lower_bound
))
3316 FREE_STACK_RETURN (REG_BADBR
);
3318 if (!(syntax
& RE_NO_BK_BRACES
))
3321 FREE_STACK_RETURN (REG_BADBR
);
3323 FREE_STACK_RETURN (REG_EESCAPE
);
3328 FREE_STACK_RETURN (REG_BADBR
);
3330 /* We just parsed a valid interval. */
3332 /* If it's invalid to have no preceding re. */
3335 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3336 FREE_STACK_RETURN (REG_BADRPT
);
3337 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3340 goto unfetch_interval
;
3343 if (upper_bound
== 0)
3344 /* If the upper bound is zero, just drop the sub pattern
3347 else if (lower_bound
== 1 && upper_bound
== 1)
3348 /* Just match it once: nothing to do here. */
3351 /* Otherwise, we have a nontrivial interval. When
3352 we're all done, the pattern will look like:
3353 set_number_at <jump count> <upper bound>
3354 set_number_at <succeed_n count> <lower bound>
3355 succeed_n <after jump addr> <succeed_n count>
3357 jump_n <succeed_n addr> <jump count>
3358 (The upper bound and `jump_n' are omitted if
3359 `upper_bound' is 1, though.) */
3361 { /* If the upper bound is > 1, we need to insert
3362 more at the end of the loop. */
3363 unsigned int nbytes
= (upper_bound
< 0 ? 3
3364 : upper_bound
> 1 ? 5 : 0);
3365 unsigned int startoffset
= 0;
3367 GET_BUFFER_SPACE (20); /* We might use less. */
3369 if (lower_bound
== 0)
3371 /* A succeed_n that starts with 0 is really a
3372 a simple on_failure_jump_loop. */
3373 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3379 /* Initialize lower bound of the `succeed_n', even
3380 though it will be set during matching by its
3381 attendant `set_number_at' (inserted next),
3382 because `re_compile_fastmap' needs to know.
3383 Jump to the `jump_n' we might insert below. */
3384 INSERT_JUMP2 (succeed_n
, laststart
,
3389 /* Code to initialize the lower bound. Insert
3390 before the `succeed_n'. The `5' is the last two
3391 bytes of this `set_number_at', plus 3 bytes of
3392 the following `succeed_n'. */
3393 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3398 if (upper_bound
< 0)
3400 /* A negative upper bound stands for infinity,
3401 in which case it degenerates to a plain jump. */
3402 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3405 else if (upper_bound
> 1)
3406 { /* More than one repetition is allowed, so
3407 append a backward jump to the `succeed_n'
3408 that starts this interval.
3410 When we've reached this during matching,
3411 we'll have matched the interval once, so
3412 jump back only `upper_bound - 1' times. */
3413 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3417 /* The location we want to set is the second
3418 parameter of the `jump_n'; that is `b-2' as
3419 an absolute address. `laststart' will be
3420 the `set_number_at' we're about to insert;
3421 `laststart+3' the number to set, the source
3422 for the relative address. But we are
3423 inserting into the middle of the pattern --
3424 so everything is getting moved up by 5.
3425 Conclusion: (b - 2) - (laststart + 3) + 5,
3426 i.e., b - laststart.
3428 We insert this at the beginning of the loop
3429 so that if we fail during matching, we'll
3430 reinitialize the bounds. */
3431 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3432 upper_bound
- 1, b
);
3437 beg_interval
= NULL
;
3442 /* If an invalid interval, match the characters as literals. */
3443 assert (beg_interval
);
3445 beg_interval
= NULL
;
3447 /* normal_char and normal_backslash need `c'. */
3450 if (!(syntax
& RE_NO_BK_BRACES
))
3452 assert (p
> pattern
&& p
[-1] == '\\');
3453 goto normal_backslash
;
3459 /* There is no way to specify the before_dot and after_dot
3460 operators. rms says this is ok. --karl */
3469 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3475 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3481 BUF_PUSH_2 (categoryspec
, c
);
3487 BUF_PUSH_2 (notcategoryspec
, c
);
3493 if (syntax
& RE_NO_GNU_OPS
)
3496 BUF_PUSH_2 (syntaxspec
, Sword
);
3501 if (syntax
& RE_NO_GNU_OPS
)
3504 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3509 if (syntax
& RE_NO_GNU_OPS
)
3516 if (syntax
& RE_NO_GNU_OPS
)
3523 if (syntax
& RE_NO_GNU_OPS
)
3532 FREE_STACK_RETURN (REG_BADPAT
);
3536 if (syntax
& RE_NO_GNU_OPS
)
3538 BUF_PUSH (wordbound
);
3542 if (syntax
& RE_NO_GNU_OPS
)
3544 BUF_PUSH (notwordbound
);
3548 if (syntax
& RE_NO_GNU_OPS
)
3554 if (syntax
& RE_NO_GNU_OPS
)
3559 case '1': case '2': case '3': case '4': case '5':
3560 case '6': case '7': case '8': case '9':
3564 if (syntax
& RE_NO_BK_REFS
)
3565 goto normal_backslash
;
3569 if (reg
> bufp
->re_nsub
|| reg
< 1
3570 /* Can't back reference to a subexp before its end. */
3571 || group_in_compile_stack (compile_stack
, reg
))
3572 FREE_STACK_RETURN (REG_ESUBREG
);
3575 BUF_PUSH_2 (duplicate
, reg
);
3582 if (syntax
& RE_BK_PLUS_QM
)
3585 goto normal_backslash
;
3589 /* You might think it would be useful for \ to mean
3590 not to translate; but if we don't translate it
3591 it will never match anything. */
3598 /* Expects the character in `c'. */
3600 /* If no exactn currently being built. */
3603 /* If last exactn not at current position. */
3604 || pending_exact
+ *pending_exact
+ 1 != b
3606 /* We have only one byte following the exactn for the count. */
3607 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3609 /* If followed by a repetition operator. */
3610 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3611 || ((syntax
& RE_BK_PLUS_QM
)
3612 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3613 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3614 || ((syntax
& RE_INTERVALS
)
3615 && ((syntax
& RE_NO_BK_BRACES
)
3616 ? p
!= pend
&& *p
== '{'
3617 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3619 /* Start building a new exactn. */
3623 BUF_PUSH_2 (exactn
, 0);
3624 pending_exact
= b
- 1;
3627 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3634 len
= CHAR_STRING (c
, b
);
3639 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3640 if (! CHAR_BYTE8_P (c1
))
3642 re_wchar_t c2
= TRANSLATE (c1
);
3644 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3650 (*pending_exact
) += len
;
3655 } /* while p != pend */
3658 /* Through the pattern now. */
3662 if (!COMPILE_STACK_EMPTY
)
3663 FREE_STACK_RETURN (REG_EPAREN
);
3665 /* If we don't want backtracking, force success
3666 the first time we reach the end of the compiled pattern. */
3667 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3670 /* We have succeeded; set the length of the buffer. */
3671 bufp
->used
= b
- bufp
->buffer
;
3676 re_compile_fastmap (bufp
);
3677 DEBUG_PRINT ("\nCompiled pattern: \n");
3678 print_compiled_pattern (bufp
);
3683 #ifndef MATCH_MAY_ALLOCATE
3684 /* Initialize the failure stack to the largest possible stack. This
3685 isn't necessary unless we're trying to avoid calling alloca in
3686 the search and match routines. */
3688 int num_regs
= bufp
->re_nsub
+ 1;
3690 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3692 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3693 falk_stack
.stack
= realloc (fail_stack
.stack
,
3694 fail_stack
.size
* sizeof *falk_stack
.stack
);
3697 regex_grow_registers (num_regs
);
3699 #endif /* not MATCH_MAY_ALLOCATE */
3701 FREE_STACK_RETURN (REG_NOERROR
);
3702 } /* regex_compile */
3704 /* Subroutines for `regex_compile'. */
3706 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3709 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3711 *loc
= (unsigned char) op
;
3712 STORE_NUMBER (loc
+ 1, arg
);
3716 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3719 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3721 *loc
= (unsigned char) op
;
3722 STORE_NUMBER (loc
+ 1, arg1
);
3723 STORE_NUMBER (loc
+ 3, arg2
);
3727 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3728 for OP followed by two-byte integer parameter ARG. */
3731 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3733 register unsigned char *pfrom
= end
;
3734 register unsigned char *pto
= end
+ 3;
3736 while (pfrom
!= loc
)
3739 store_op1 (op
, loc
, arg
);
3743 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3746 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3748 register unsigned char *pfrom
= end
;
3749 register unsigned char *pto
= end
+ 5;
3751 while (pfrom
!= loc
)
3754 store_op2 (op
, loc
, arg1
, arg2
);
3758 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3759 after an alternative or a begin-subexpression. We assume there is at
3760 least one character before the ^. */
3763 at_begline_loc_p (const_re_char
*pattern
, const_re_char
*p
, reg_syntax_t syntax
)
3765 re_char
*prev
= p
- 2;
3766 boolean odd_backslashes
;
3768 /* After a subexpression? */
3770 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3772 /* After an alternative? */
3773 else if (*prev
== '|')
3774 odd_backslashes
= (syntax
& RE_NO_BK_VBAR
) == 0;
3776 /* After a shy subexpression? */
3777 else if (*prev
== ':' && (syntax
& RE_SHY_GROUPS
))
3779 /* Skip over optional regnum. */
3780 while (prev
- 1 >= pattern
&& prev
[-1] >= '0' && prev
[-1] <= '9')
3783 if (!(prev
- 2 >= pattern
3784 && prev
[-1] == '?' && prev
[-2] == '('))
3787 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3792 /* Count the number of preceding backslashes. */
3794 while (prev
- 1 >= pattern
&& prev
[-1] == '\\')
3796 return (p
- prev
) & odd_backslashes
;
3800 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3801 at least one character after the $, i.e., `P < PEND'. */
3804 at_endline_loc_p (const_re_char
*p
, const_re_char
*pend
, reg_syntax_t syntax
)
3807 boolean next_backslash
= *next
== '\\';
3808 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3811 /* Before a subexpression? */
3812 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3813 : next_backslash
&& next_next
&& *next_next
== ')')
3814 /* Before an alternative? */
3815 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3816 : next_backslash
&& next_next
&& *next_next
== '|');
3820 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3821 false if it's not. */
3824 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3826 ssize_t this_element
;
3828 for (this_element
= compile_stack
.avail
- 1;
3831 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3838 If fastmap is non-NULL, go through the pattern and fill fastmap
3839 with all the possible leading chars. If fastmap is NULL, don't
3840 bother filling it up (obviously) and only return whether the
3841 pattern could potentially match the empty string.
3843 Return 1 if p..pend might match the empty string.
3844 Return 0 if p..pend matches at least one char.
3845 Return -1 if fastmap was not updated accurately. */
3848 analyse_first (const_re_char
*p
, const_re_char
*pend
, char *fastmap
,
3849 const int multibyte
)
3854 /* If all elements for base leading-codes in fastmap is set, this
3855 flag is set true. */
3856 boolean match_any_multibyte_characters
= false;
3860 /* The loop below works as follows:
3861 - It has a working-list kept in the PATTERN_STACK and which basically
3862 starts by only containing a pointer to the first operation.
3863 - If the opcode we're looking at is a match against some set of
3864 chars, then we add those chars to the fastmap and go on to the
3865 next work element from the worklist (done via `break').
3866 - If the opcode is a control operator on the other hand, we either
3867 ignore it (if it's meaningless at this point, such as `start_memory')
3868 or execute it (if it's a jump). If the jump has several destinations
3869 (i.e. `on_failure_jump'), then we push the other destination onto the
3871 We guarantee termination by ignoring backward jumps (more or less),
3872 so that `p' is monotonically increasing. More to the point, we
3873 never set `p' (or push) anything `<= p1'. */
3877 /* `p1' is used as a marker of how far back a `on_failure_jump'
3878 can go without being ignored. It is normally equal to `p'
3879 (which prevents any backward `on_failure_jump') except right
3880 after a plain `jump', to allow patterns such as:
3883 10: on_failure_jump 3
3884 as used for the *? operator. */
3893 /* If the first character has to match a backreference, that means
3894 that the group was empty (since it already matched). Since this
3895 is the only case that interests us here, we can assume that the
3896 backreference must match the empty string. */
3901 /* Following are the cases which match a character. These end
3907 /* If multibyte is nonzero, the first byte of each
3908 character is an ASCII or a leading code. Otherwise,
3909 each byte is a character. Thus, this works in both
3914 /* For the case of matching this unibyte regex
3915 against multibyte, we must set a leading code of
3916 the corresponding multibyte character. */
3917 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
3919 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
3926 /* We could put all the chars except for \n (and maybe \0)
3927 but we don't bother since it is generally not worth it. */
3928 if (!fastmap
) break;
3933 if (!fastmap
) break;
3935 /* Chars beyond end of bitmap are possible matches. */
3936 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3937 j
< (1 << BYTEWIDTH
); j
++)
3943 if (!fastmap
) break;
3944 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3945 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3947 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3951 if (/* Any leading code can possibly start a character
3952 which doesn't match the specified set of characters. */
3955 /* If we can match a character class, we can match any
3956 multibyte characters. */
3957 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3958 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3961 if (match_any_multibyte_characters
== false)
3963 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3964 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3966 match_any_multibyte_characters
= true;
3970 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3971 && match_any_multibyte_characters
== false)
3973 /* Set fastmap[I] to 1 where I is a leading code of each
3974 multibyte character in the range table. */
3976 unsigned char lc1
, lc2
;
3978 /* Make P points the range table. `+ 2' is to skip flag
3979 bits for a character class. */
3980 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3982 /* Extract the number of ranges in range table into COUNT. */
3983 EXTRACT_NUMBER_AND_INCR (count
, p
);
3984 for (; count
> 0; count
--, p
+= 3)
3986 /* Extract the start and end of each range. */
3987 EXTRACT_CHARACTER (c
, p
);
3988 lc1
= CHAR_LEADING_CODE (c
);
3990 EXTRACT_CHARACTER (c
, p
);
3991 lc2
= CHAR_LEADING_CODE (c
);
3992 for (j
= lc1
; j
<= lc2
; j
++)
4001 if (!fastmap
) break;
4003 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4005 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4006 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4010 /* This match depends on text properties. These end with
4011 aborting optimizations. */
4015 case notcategoryspec
:
4016 if (!fastmap
) break;
4017 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4019 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
4020 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4023 /* Any leading code can possibly start a character which
4024 has or doesn't has the specified category. */
4025 if (match_any_multibyte_characters
== false)
4027 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4028 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4030 match_any_multibyte_characters
= true;
4034 /* All cases after this match the empty string. These end with
4056 EXTRACT_NUMBER_AND_INCR (j
, p
);
4058 /* Backward jumps can only go back to code that we've already
4059 visited. `re_compile' should make sure this is true. */
4064 case on_failure_jump
:
4065 case on_failure_keep_string_jump
:
4066 case on_failure_jump_loop
:
4067 case on_failure_jump_nastyloop
:
4068 case on_failure_jump_smart
:
4074 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4075 to jump back to "just after here". */
4078 case on_failure_jump
:
4079 case on_failure_keep_string_jump
:
4080 case on_failure_jump_nastyloop
:
4081 case on_failure_jump_loop
:
4082 case on_failure_jump_smart
:
4083 EXTRACT_NUMBER_AND_INCR (j
, p
);
4085 ; /* Backward jump to be ignored. */
4087 { /* We have to look down both arms.
4088 We first go down the "straight" path so as to minimize
4089 stack usage when going through alternatives. */
4090 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4098 /* This code simply does not properly handle forward jump_n. */
4099 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4101 /* jump_n can either jump or fall through. The (backward) jump
4102 case has already been handled, so we only need to look at the
4103 fallthrough case. */
4107 /* If N == 0, it should be an on_failure_jump_loop instead. */
4108 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4110 /* We only care about one iteration of the loop, so we don't
4111 need to consider the case where this behaves like an
4128 abort (); /* We have listed all the cases. */
4131 /* Getting here means we have found the possible starting
4132 characters for one path of the pattern -- and that the empty
4133 string does not match. We need not follow this path further. */
4137 /* We reached the end without matching anything. */
4140 } /* analyse_first */
4142 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4143 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4144 characters can start a string that matches the pattern. This fastmap
4145 is used by re_search to skip quickly over impossible starting points.
4147 Character codes above (1 << BYTEWIDTH) are not represented in the
4148 fastmap, but the leading codes are represented. Thus, the fastmap
4149 indicates which character sets could start a match.
4151 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4152 area as BUFP->fastmap.
4154 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4157 Returns 0 if we succeed, -2 if an internal error. */
4160 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4162 char *fastmap
= bufp
->fastmap
;
4165 assert (fastmap
&& bufp
->buffer
);
4167 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4168 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4170 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4171 fastmap
, RE_MULTIBYTE_P (bufp
));
4172 bufp
->can_be_null
= (analysis
!= 0);
4174 } /* re_compile_fastmap */
4176 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4177 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4178 this memory for recording register information. STARTS and ENDS
4179 must be allocated using the malloc library routine, and must each
4180 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4182 If NUM_REGS == 0, then subsequent matches should allocate their own
4185 Unless this function is called, the first search or match using
4186 PATTERN_BUFFER will allocate its own register data, without
4187 freeing the old data. */
4190 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4194 bufp
->regs_allocated
= REGS_REALLOCATE
;
4195 regs
->num_regs
= num_regs
;
4196 regs
->start
= starts
;
4201 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4203 regs
->start
= regs
->end
= 0;
4206 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4208 /* Searching routines. */
4210 /* Like re_search_2, below, but only one string is specified, and
4211 doesn't let you say where to stop matching. */
4214 re_search (struct re_pattern_buffer
*bufp
, const char *string
, size_t size
,
4215 ssize_t startpos
, ssize_t range
, struct re_registers
*regs
)
4217 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4220 WEAK_ALIAS (__re_search
, re_search
)
4222 /* Head address of virtual concatenation of string. */
4223 #define HEAD_ADDR_VSTRING(P) \
4224 (((P) >= size1 ? string2 : string1))
4226 /* Address of POS in the concatenation of virtual string. */
4227 #define POS_ADDR_VSTRING(POS) \
4228 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4230 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4231 virtual concatenation of STRING1 and STRING2, starting first at index
4232 STARTPOS, then at STARTPOS + 1, and so on.
4234 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4236 RANGE is how far to scan while trying to match. RANGE = 0 means try
4237 only at STARTPOS; in general, the last start tried is STARTPOS +
4240 In REGS, return the indices of the virtual concatenation of STRING1
4241 and STRING2 that matched the entire BUFP->buffer and its contained
4244 Do not consider matching one past the index STOP in the virtual
4245 concatenation of STRING1 and STRING2.
4247 We return either the position in the strings at which the match was
4248 found, -1 if no match, or -2 if error (such as failure
4252 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, size_t size1
,
4253 const char *str2
, size_t size2
, ssize_t startpos
, ssize_t range
,
4254 struct re_registers
*regs
, ssize_t stop
)
4257 re_char
*string1
= (re_char
*) str1
;
4258 re_char
*string2
= (re_char
*) str2
;
4259 register char *fastmap
= bufp
->fastmap
;
4260 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4261 size_t total_size
= size1
+ size2
;
4262 ssize_t endpos
= startpos
+ range
;
4263 boolean anchored_start
;
4264 /* Nonzero if we are searching multibyte string. */
4265 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4267 /* Check for out-of-range STARTPOS. */
4268 if (startpos
< 0 || startpos
> total_size
)
4271 /* Fix up RANGE if it might eventually take us outside
4272 the virtual concatenation of STRING1 and STRING2.
4273 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4275 range
= 0 - startpos
;
4276 else if (endpos
> total_size
)
4277 range
= total_size
- startpos
;
4279 /* If the search isn't to be a backwards one, don't waste time in a
4280 search for a pattern anchored at beginning of buffer. */
4281 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4290 /* In a forward search for something that starts with \=.
4291 don't keep searching past point. */
4292 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4294 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4300 /* Update the fastmap now if not correct already. */
4301 if (fastmap
&& !bufp
->fastmap_accurate
)
4302 re_compile_fastmap (bufp
);
4304 /* See whether the pattern is anchored. */
4305 anchored_start
= (bufp
->buffer
[0] == begline
);
4308 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4310 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4312 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4316 /* Loop through the string, looking for a place to start matching. */
4319 /* If the pattern is anchored,
4320 skip quickly past places we cannot match.
4321 We don't bother to treat startpos == 0 specially
4322 because that case doesn't repeat. */
4323 if (anchored_start
&& startpos
> 0)
4325 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4326 : string2
[startpos
- size1
- 1])
4331 /* If a fastmap is supplied, skip quickly over characters that
4332 cannot be the start of a match. If the pattern can match the
4333 null string, however, we don't need to skip characters; we want
4334 the first null string. */
4335 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4337 register re_char
*d
;
4338 register re_wchar_t buf_ch
;
4340 d
= POS_ADDR_VSTRING (startpos
);
4342 if (range
> 0) /* Searching forwards. */
4344 ssize_t irange
= range
, lim
= 0;
4346 if (startpos
< size1
&& startpos
+ range
>= size1
)
4347 lim
= range
- (size1
- startpos
);
4349 /* Written out as an if-else to avoid testing `translate'
4351 if (RE_TRANSLATE_P (translate
))
4358 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4359 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4360 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4363 range
-= buf_charlen
;
4369 register re_wchar_t ch
, translated
;
4372 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4373 translated
= RE_TRANSLATE (translate
, ch
);
4374 if (translated
!= ch
4375 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4377 if (fastmap
[buf_ch
])
4390 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4391 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4393 range
-= buf_charlen
;
4397 while (range
> lim
&& !fastmap
[*d
])
4403 startpos
+= irange
- range
;
4405 else /* Searching backwards. */
4409 buf_ch
= STRING_CHAR (d
);
4410 buf_ch
= TRANSLATE (buf_ch
);
4411 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4416 register re_wchar_t ch
, translated
;
4419 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4420 translated
= TRANSLATE (ch
);
4421 if (translated
!= ch
4422 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4424 if (! fastmap
[TRANSLATE (buf_ch
)])
4430 /* If can't match the null string, and that's all we have left, fail. */
4431 if (range
>= 0 && startpos
== total_size
&& fastmap
4432 && !bufp
->can_be_null
)
4435 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4436 startpos
, regs
, stop
);
4449 /* Update STARTPOS to the next character boundary. */
4452 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4453 int len
= BYTES_BY_CHAR_HEAD (*p
);
4471 /* Update STARTPOS to the previous character boundary. */
4474 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4476 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4478 /* Find the head of multibyte form. */
4479 PREV_CHAR_BOUNDARY (p
, phead
);
4480 range
+= p0
- 1 - p
;
4484 startpos
-= p0
- 1 - p
;
4490 WEAK_ALIAS (__re_search_2
, re_search_2
)
4492 /* Declarations and macros for re_match_2. */
4494 static int bcmp_translate (re_char
*s1
, re_char
*s2
,
4495 register ssize_t len
,
4496 RE_TRANSLATE_TYPE translate
,
4497 const int multibyte
);
4499 /* This converts PTR, a pointer into one of the search strings `string1'
4500 and `string2' into an offset from the beginning of that string. */
4501 #define POINTER_TO_OFFSET(ptr) \
4502 (FIRST_STRING_P (ptr) \
4504 : (ptr) - string2 + (ptrdiff_t) size1)
4506 /* Call before fetching a character with *d. This switches over to
4507 string2 if necessary.
4508 Check re_match_2_internal for a discussion of why end_match_2 might
4509 not be within string2 (but be equal to end_match_1 instead). */
4510 #define PREFETCH() \
4513 /* End of string2 => fail. */ \
4514 if (dend == end_match_2) \
4516 /* End of string1 => advance to string2. */ \
4518 dend = end_match_2; \
4521 /* Call before fetching a char with *d if you already checked other limits.
4522 This is meant for use in lookahead operations like wordend, etc..
4523 where we might need to look at parts of the string that might be
4524 outside of the LIMITs (i.e past `stop'). */
4525 #define PREFETCH_NOLIMIT() \
4529 dend = end_match_2; \
4532 /* Test if at very beginning or at very end of the virtual concatenation
4533 of `string1' and `string2'. If only one string, it's `string2'. */
4534 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4535 #define AT_STRINGS_END(d) ((d) == end2)
4537 /* Disabled due to a compiler bug -- see comment at case wordbound */
4539 /* The comment at case wordbound is following one, but we don't use
4540 AT_WORD_BOUNDARY anymore to support multibyte form.
4542 The DEC Alpha C compiler 3.x generates incorrect code for the
4543 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4544 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4545 macro and introducing temporary variables works around the bug. */
4548 /* Test if D points to a character which is word-constituent. We have
4549 two special cases to check for: if past the end of string1, look at
4550 the first character in string2; and if before the beginning of
4551 string2, look at the last character in string1. */
4552 #define WORDCHAR_P(d) \
4553 (SYNTAX ((d) == end1 ? *string2 \
4554 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4557 /* Test if the character before D and the one at D differ with respect
4558 to being word-constituent. */
4559 #define AT_WORD_BOUNDARY(d) \
4560 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4561 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4564 /* Free everything we malloc. */
4565 #ifdef MATCH_MAY_ALLOCATE
4566 # define FREE_VAR(var) \
4574 # define FREE_VARIABLES() \
4576 REGEX_FREE_STACK (fail_stack.stack); \
4577 FREE_VAR (regstart); \
4578 FREE_VAR (regend); \
4579 FREE_VAR (best_regstart); \
4580 FREE_VAR (best_regend); \
4581 REGEX_SAFE_FREE (); \
4584 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4585 #endif /* not MATCH_MAY_ALLOCATE */
4588 /* Optimization routines. */
4590 /* If the operation is a match against one or more chars,
4591 return a pointer to the next operation, else return NULL. */
4593 skip_one_char (const_re_char
*p
)
4606 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4609 p
= CHARSET_RANGE_TABLE (p
- 1);
4610 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4611 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4614 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4621 case notcategoryspec
:
4633 /* Jump over non-matching operations. */
4635 skip_noops (const_re_char
*p
, const_re_char
*pend
)
4649 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4660 /* Non-zero if "p1 matches something" implies "p2 fails". */
4662 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, const_re_char
*p1
,
4666 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4667 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4669 assert (p1
>= bufp
->buffer
&& p1
< pend
4670 && p2
>= bufp
->buffer
&& p2
<= pend
);
4672 /* Skip over open/close-group commands.
4673 If what follows this loop is a ...+ construct,
4674 look at what begins its body, since we will have to
4675 match at least one of that. */
4676 p2
= skip_noops (p2
, pend
);
4677 /* The same skip can be done for p1, except that this function
4678 is only used in the case where p1 is a simple match operator. */
4679 /* p1 = skip_noops (p1, pend); */
4681 assert (p1
>= bufp
->buffer
&& p1
< pend
4682 && p2
>= bufp
->buffer
&& p2
<= pend
);
4684 op2
= p2
== pend
? succeed
: *p2
;
4690 /* If we're at the end of the pattern, we can change. */
4691 if (skip_one_char (p1
))
4693 DEBUG_PRINT (" End of pattern: fast loop.\n");
4701 register re_wchar_t c
4702 = (re_opcode_t
) *p2
== endline
? '\n'
4703 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4705 if ((re_opcode_t
) *p1
== exactn
)
4707 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4709 DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4714 else if ((re_opcode_t
) *p1
== charset
4715 || (re_opcode_t
) *p1
== charset_not
)
4717 int not = (re_opcode_t
) *p1
== charset_not
;
4719 /* Test if C is listed in charset (or charset_not)
4721 if (! multibyte
|| IS_REAL_ASCII (c
))
4723 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4724 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4727 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4728 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4730 /* `not' is equal to 1 if c would match, which means
4731 that we can't change to pop_failure_jump. */
4734 DEBUG_PRINT (" No match => fast loop.\n");
4738 else if ((re_opcode_t
) *p1
== anychar
4741 DEBUG_PRINT (" . != \\n => fast loop.\n");
4749 if ((re_opcode_t
) *p1
== exactn
)
4750 /* Reuse the code above. */
4751 return mutually_exclusive_p (bufp
, p2
, p1
);
4753 /* It is hard to list up all the character in charset
4754 P2 if it includes multibyte character. Give up in
4756 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4758 /* Now, we are sure that P2 has no range table.
4759 So, for the size of bitmap in P2, `p2[1]' is
4760 enough. But P1 may have range table, so the
4761 size of bitmap table of P1 is extracted by
4762 using macro `CHARSET_BITMAP_SIZE'.
4764 In a multibyte case, we know that all the character
4765 listed in P2 is ASCII. In a unibyte case, P1 has only a
4766 bitmap table. So, in both cases, it is enough to test
4767 only the bitmap table of P1. */
4769 if ((re_opcode_t
) *p1
== charset
)
4772 /* We win if the charset inside the loop
4773 has no overlap with the one after the loop. */
4776 && idx
< CHARSET_BITMAP_SIZE (p1
));
4778 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4782 || idx
== CHARSET_BITMAP_SIZE (p1
))
4784 DEBUG_PRINT (" No match => fast loop.\n");
4788 else if ((re_opcode_t
) *p1
== charset_not
)
4791 /* We win if the charset_not inside the loop lists
4792 every character listed in the charset after. */
4793 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4794 if (! (p2
[2 + idx
] == 0
4795 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4796 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4801 DEBUG_PRINT (" No match => fast loop.\n");
4814 /* Reuse the code above. */
4815 return mutually_exclusive_p (bufp
, p2
, p1
);
4817 /* When we have two charset_not, it's very unlikely that
4818 they don't overlap. The union of the two sets of excluded
4819 chars should cover all possible chars, which, as a matter of
4820 fact, is virtually impossible in multibyte buffers. */
4826 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4828 return ((re_opcode_t
) *p1
== syntaxspec
4829 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4831 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4834 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4836 return ((re_opcode_t
) *p1
== notsyntaxspec
4837 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4839 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4842 return (((re_opcode_t
) *p1
== notsyntaxspec
4843 || (re_opcode_t
) *p1
== syntaxspec
)
4848 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4849 case notcategoryspec
:
4850 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4862 /* Matching routines. */
4864 #ifndef emacs /* Emacs never uses this. */
4865 /* re_match is like re_match_2 except it takes only a single string. */
4868 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
4869 size_t size
, ssize_t pos
, struct re_registers
*regs
)
4871 regoff_t result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
,
4872 size
, pos
, regs
, size
);
4875 WEAK_ALIAS (__re_match
, re_match
)
4876 #endif /* not emacs */
4879 /* In Emacs, this is the string or buffer in which we
4880 are matching. It is used for looking up syntax properties. */
4881 Lisp_Object re_match_object
;
4884 /* re_match_2 matches the compiled pattern in BUFP against the
4885 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4886 and SIZE2, respectively). We start matching at POS, and stop
4889 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4890 store offsets for the substring each group matched in REGS. See the
4891 documentation for exactly how many groups we fill.
4893 We return -1 if no match, -2 if an internal error (such as the
4894 failure stack overflowing). Otherwise, we return the length of the
4895 matched substring. */
4898 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
,
4899 size_t size1
, const char *string2
, size_t size2
, ssize_t pos
,
4900 struct re_registers
*regs
, ssize_t stop
)
4906 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4907 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4908 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4911 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4912 (re_char
*) string2
, size2
,
4916 WEAK_ALIAS (__re_match_2
, re_match_2
)
4919 /* This is a separate function so that we can force an alloca cleanup
4922 re_match_2_internal (struct re_pattern_buffer
*bufp
, const_re_char
*string1
,
4923 size_t size1
, const_re_char
*string2
, size_t size2
,
4924 ssize_t pos
, struct re_registers
*regs
, ssize_t stop
)
4926 /* General temporaries. */
4930 /* Just past the end of the corresponding string. */
4931 re_char
*end1
, *end2
;
4933 /* Pointers into string1 and string2, just past the last characters in
4934 each to consider matching. */
4935 re_char
*end_match_1
, *end_match_2
;
4937 /* Where we are in the data, and the end of the current string. */
4940 /* Used sometimes to remember where we were before starting matching
4941 an operator so that we can go back in case of failure. This "atomic"
4942 behavior of matching opcodes is indispensable to the correctness
4943 of the on_failure_keep_string_jump optimization. */
4946 /* Where we are in the pattern, and the end of the pattern. */
4947 re_char
*p
= bufp
->buffer
;
4948 re_char
*pend
= p
+ bufp
->used
;
4950 /* We use this to map every character in the string. */
4951 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4953 /* Nonzero if BUFP is setup from a multibyte regex. */
4954 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4956 /* Nonzero if STRING1/STRING2 are multibyte. */
4957 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4959 /* Failure point stack. Each place that can handle a failure further
4960 down the line pushes a failure point on this stack. It consists of
4961 regstart, and regend for all registers corresponding to
4962 the subexpressions we're currently inside, plus the number of such
4963 registers, and, finally, two char *'s. The first char * is where
4964 to resume scanning the pattern; the second one is where to resume
4965 scanning the strings. */
4966 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4967 fail_stack_type fail_stack
;
4969 #ifdef DEBUG_COMPILES_ARGUMENTS
4970 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4973 #if defined REL_ALLOC && defined REGEX_MALLOC
4974 /* This holds the pointer to the failure stack, when
4975 it is allocated relocatably. */
4976 fail_stack_elt_t
*failure_stack_ptr
;
4979 /* We fill all the registers internally, independent of what we
4980 return, for use in backreferences. The number here includes
4981 an element for register zero. */
4982 size_t num_regs
= bufp
->re_nsub
+ 1;
4984 /* Information on the contents of registers. These are pointers into
4985 the input strings; they record just what was matched (on this
4986 attempt) by a subexpression part of the pattern, that is, the
4987 regnum-th regstart pointer points to where in the pattern we began
4988 matching and the regnum-th regend points to right after where we
4989 stopped matching the regnum-th subexpression. (The zeroth register
4990 keeps track of what the whole pattern matches.) */
4991 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4992 re_char
**regstart
, **regend
;
4995 /* The following record the register info as found in the above
4996 variables when we find a match better than any we've seen before.
4997 This happens as we backtrack through the failure points, which in
4998 turn happens only if we have not yet matched the entire string. */
4999 unsigned best_regs_set
= false;
5000 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5001 re_char
**best_regstart
, **best_regend
;
5004 /* Logically, this is `best_regend[0]'. But we don't want to have to
5005 allocate space for that if we're not allocating space for anything
5006 else (see below). Also, we never need info about register 0 for
5007 any of the other register vectors, and it seems rather a kludge to
5008 treat `best_regend' differently than the rest. So we keep track of
5009 the end of the best match so far in a separate variable. We
5010 initialize this to NULL so that when we backtrack the first time
5011 and need to test it, it's not garbage. */
5012 re_char
*match_end
= NULL
;
5014 #ifdef DEBUG_COMPILES_ARGUMENTS
5015 /* Counts the total number of registers pushed. */
5016 unsigned num_regs_pushed
= 0;
5019 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5021 REGEX_USE_SAFE_ALLOCA
;
5025 #ifdef MATCH_MAY_ALLOCATE
5026 /* Do not bother to initialize all the register variables if there are
5027 no groups in the pattern, as it takes a fair amount of time. If
5028 there are groups, we include space for register 0 (the whole
5029 pattern), even though we never use it, since it simplifies the
5030 array indexing. We should fix this. */
5033 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5034 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5035 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5036 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5038 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5046 /* We must initialize all our variables to NULL, so that
5047 `FREE_VARIABLES' doesn't try to free them. */
5048 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5050 #endif /* MATCH_MAY_ALLOCATE */
5052 /* The starting position is bogus. */
5053 if (pos
< 0 || pos
> size1
+ size2
)
5059 /* Initialize subexpression text positions to -1 to mark ones that no
5060 start_memory/stop_memory has been seen for. Also initialize the
5061 register information struct. */
5062 for (reg
= 1; reg
< num_regs
; reg
++)
5063 regstart
[reg
] = regend
[reg
] = NULL
;
5065 /* We move `string1' into `string2' if the latter's empty -- but not if
5066 `string1' is null. */
5067 if (size2
== 0 && string1
!= NULL
)
5074 end1
= string1
+ size1
;
5075 end2
= string2
+ size2
;
5077 /* `p' scans through the pattern as `d' scans through the data.
5078 `dend' is the end of the input string that `d' points within. `d'
5079 is advanced into the following input string whenever necessary, but
5080 this happens before fetching; therefore, at the beginning of the
5081 loop, `d' can be pointing at the end of a string, but it cannot
5085 /* Only match within string2. */
5086 d
= string2
+ pos
- size1
;
5087 dend
= end_match_2
= string2
+ stop
- size1
;
5088 end_match_1
= end1
; /* Just to give it a value. */
5094 /* Only match within string1. */
5095 end_match_1
= string1
+ stop
;
5097 When we reach end_match_1, PREFETCH normally switches to string2.
5098 But in the present case, this means that just doing a PREFETCH
5099 makes us jump from `stop' to `gap' within the string.
5100 What we really want here is for the search to stop as
5101 soon as we hit end_match_1. That's why we set end_match_2
5102 to end_match_1 (since PREFETCH fails as soon as we hit
5104 end_match_2
= end_match_1
;
5107 { /* It's important to use this code when stop == size so that
5108 moving `d' from end1 to string2 will not prevent the d == dend
5109 check from catching the end of string. */
5111 end_match_2
= string2
+ stop
- size1
;
5117 DEBUG_PRINT ("The compiled pattern is: ");
5118 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5119 DEBUG_PRINT ("The string to match is: `");
5120 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5121 DEBUG_PRINT ("'\n");
5123 /* This loops over pattern commands. It exits by returning from the
5124 function if the match is complete, or it drops through if the match
5125 fails at this starting point in the input data. */
5128 DEBUG_PRINT ("\n%p: ", p
);
5134 /* End of pattern means we might have succeeded. */
5135 DEBUG_PRINT ("end of pattern ... ");
5137 /* If we haven't matched the entire string, and we want the
5138 longest match, try backtracking. */
5139 if (d
!= end_match_2
)
5141 /* 1 if this match ends in the same string (string1 or string2)
5142 as the best previous match. */
5143 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5144 == FIRST_STRING_P (d
));
5145 /* 1 if this match is the best seen so far. */
5146 boolean best_match_p
;
5148 /* AIX compiler got confused when this was combined
5149 with the previous declaration. */
5151 best_match_p
= d
> match_end
;
5153 best_match_p
= !FIRST_STRING_P (d
);
5155 DEBUG_PRINT ("backtracking.\n");
5157 if (!FAIL_STACK_EMPTY ())
5158 { /* More failure points to try. */
5160 /* If exceeds best match so far, save it. */
5161 if (!best_regs_set
|| best_match_p
)
5163 best_regs_set
= true;
5166 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5168 for (reg
= 1; reg
< num_regs
; reg
++)
5170 best_regstart
[reg
] = regstart
[reg
];
5171 best_regend
[reg
] = regend
[reg
];
5177 /* If no failure points, don't restore garbage. And if
5178 last match is real best match, don't restore second
5180 else if (best_regs_set
&& !best_match_p
)
5183 /* Restore best match. It may happen that `dend ==
5184 end_match_1' while the restored d is in string2.
5185 For example, the pattern `x.*y.*z' against the
5186 strings `x-' and `y-z-', if the two strings are
5187 not consecutive in memory. */
5188 DEBUG_PRINT ("Restoring best registers.\n");
5191 dend
= ((d
>= string1
&& d
<= end1
)
5192 ? end_match_1
: end_match_2
);
5194 for (reg
= 1; reg
< num_regs
; reg
++)
5196 regstart
[reg
] = best_regstart
[reg
];
5197 regend
[reg
] = best_regend
[reg
];
5200 } /* d != end_match_2 */
5203 DEBUG_PRINT ("Accepting match.\n");
5205 /* If caller wants register contents data back, do it. */
5206 if (regs
&& !bufp
->no_sub
)
5208 /* Have the register data arrays been allocated? */
5209 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5210 { /* No. So allocate them with malloc. We need one
5211 extra element beyond `num_regs' for the `-1' marker
5213 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5214 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5215 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5216 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5221 bufp
->regs_allocated
= REGS_REALLOCATE
;
5223 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5224 { /* Yes. If we need more elements than were already
5225 allocated, reallocate them. If we need fewer, just
5227 if (regs
->num_regs
< num_regs
+ 1)
5229 regs
->num_regs
= num_regs
+ 1;
5230 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5231 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5232 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5241 /* These braces fend off a "empty body in an else-statement"
5242 warning under GCC when assert expands to nothing. */
5243 assert (bufp
->regs_allocated
== REGS_FIXED
);
5246 /* Convert the pointer data in `regstart' and `regend' to
5247 indices. Register zero has to be set differently,
5248 since we haven't kept track of any info for it. */
5249 if (regs
->num_regs
> 0)
5251 regs
->start
[0] = pos
;
5252 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5255 /* Go through the first `min (num_regs, regs->num_regs)'
5256 registers, since that is all we initialized. */
5257 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5259 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5260 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5263 regs
->start
[reg
] = POINTER_TO_OFFSET (regstart
[reg
]);
5264 regs
->end
[reg
] = POINTER_TO_OFFSET (regend
[reg
]);
5268 /* If the regs structure we return has more elements than
5269 were in the pattern, set the extra elements to -1. If
5270 we (re)allocated the registers, this is the case,
5271 because we always allocate enough to have at least one
5273 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5274 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5275 } /* regs && !bufp->no_sub */
5277 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5278 nfailure_points_pushed
, nfailure_points_popped
,
5279 nfailure_points_pushed
- nfailure_points_popped
);
5280 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed
);
5282 dcnt
= POINTER_TO_OFFSET (d
) - pos
;
5284 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt
);
5290 /* Otherwise match next pattern command. */
5293 /* Ignore these. Used to ignore the n of succeed_n's which
5294 currently have n == 0. */
5296 DEBUG_PRINT ("EXECUTING no_op.\n");
5300 DEBUG_PRINT ("EXECUTING succeed.\n");
5303 /* Match the next n pattern characters exactly. The following
5304 byte in the pattern defines n, and the n bytes after that
5305 are the characters to match. */
5308 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt
);
5310 /* Remember the start point to rollback upon failure. */
5314 /* This is written out as an if-else so we don't waste time
5315 testing `translate' inside the loop. */
5316 if (RE_TRANSLATE_P (translate
))
5320 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5340 /* The cost of testing `translate' is comparatively small. */
5341 if (target_multibyte
)
5344 int pat_charlen
, buf_charlen
;
5349 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5352 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5355 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5357 if (TRANSLATE (buf_ch
) != pat_ch
)
5365 mcnt
-= pat_charlen
;
5377 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5378 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5385 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5386 if (! CHAR_BYTE8_P (buf_ch
))
5388 buf_ch
= TRANSLATE (buf_ch
);
5389 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5395 if (buf_ch
!= pat_ch
)
5408 /* Match any character except possibly a newline or a null. */
5414 DEBUG_PRINT ("EXECUTING anychar.\n");
5417 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5419 buf_ch
= TRANSLATE (buf_ch
);
5421 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5423 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5424 && buf_ch
== '\000'))
5427 DEBUG_PRINT (" Matched `%d'.\n", *d
);
5436 register unsigned int c
;
5437 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5440 /* Start of actual range_table, or end of bitmap if there is no
5442 re_char
*range_table
IF_LINT (= NULL
);
5444 /* Nonzero if there is a range table. */
5445 int range_table_exists
;
5447 /* Number of ranges of range table. This is not included
5448 in the initial byte-length of the command. */
5451 /* Whether matching against a unibyte character. */
5452 boolean unibyte_char
= false;
5454 DEBUG_PRINT ("EXECUTING charset%s.\n", not ? "_not" : "");
5456 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5458 if (range_table_exists
)
5460 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5461 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5465 c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5466 if (target_multibyte
)
5471 c1
= RE_CHAR_TO_UNIBYTE (c
);
5474 unibyte_char
= true;
5480 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5482 if (! CHAR_BYTE8_P (c1
))
5484 c1
= TRANSLATE (c1
);
5485 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5488 unibyte_char
= true;
5493 unibyte_char
= true;
5496 if (unibyte_char
&& c
< (1 << BYTEWIDTH
))
5497 { /* Lookup bitmap. */
5498 /* Cast to `unsigned' instead of `unsigned char' in
5499 case the bit list is a full 32 bytes long. */
5500 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5501 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5505 else if (range_table_exists
)
5507 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5509 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5510 | (class_bits
& BIT_MULTIBYTE
)
5511 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5512 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5513 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5514 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5517 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5521 if (range_table_exists
)
5522 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5524 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5526 if (!not) goto fail
;
5533 /* The beginning of a group is represented by start_memory.
5534 The argument is the register number. The text
5535 matched within the group is recorded (in the internal
5536 registers data structure) under the register number. */
5538 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p
);
5540 /* In case we need to undo this operation (via backtracking). */
5541 PUSH_FAILURE_REG (*p
);
5544 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5545 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart
[*p
]));
5547 /* Move past the register number and inner group count. */
5552 /* The stop_memory opcode represents the end of a group. Its
5553 argument is the same as start_memory's: the register number. */
5555 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p
);
5557 assert (!REG_UNSET (regstart
[*p
]));
5558 /* Strictly speaking, there should be code such as:
5560 assert (REG_UNSET (regend[*p]));
5561 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5563 But the only info to be pushed is regend[*p] and it is known to
5564 be UNSET, so there really isn't anything to push.
5565 Not pushing anything, on the other hand deprives us from the
5566 guarantee that regend[*p] is UNSET since undoing this operation
5567 will not reset its value properly. This is not important since
5568 the value will only be read on the next start_memory or at
5569 the very end and both events can only happen if this stop_memory
5573 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend
[*p
]));
5575 /* Move past the register number and the inner group count. */
5580 /* \<digit> has been turned into a `duplicate' command which is
5581 followed by the numeric value of <digit> as the register number. */
5584 register re_char
*d2
, *dend2
;
5585 int regno
= *p
++; /* Get which register to match against. */
5586 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno
);
5588 /* Can't back reference a group which we've never matched. */
5589 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5592 /* Where in input to try to start matching. */
5593 d2
= regstart
[regno
];
5595 /* Remember the start point to rollback upon failure. */
5598 /* Where to stop matching; if both the place to start and
5599 the place to stop matching are in the same string, then
5600 set to the place to stop, otherwise, for now have to use
5601 the end of the first string. */
5603 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5604 == FIRST_STRING_P (regend
[regno
]))
5605 ? regend
[regno
] : end_match_1
);
5610 /* If necessary, advance to next segment in register
5614 if (dend2
== end_match_2
) break;
5615 if (dend2
== regend
[regno
]) break;
5617 /* End of string1 => advance to string2. */
5619 dend2
= regend
[regno
];
5621 /* At end of register contents => success */
5622 if (d2
== dend2
) break;
5624 /* If necessary, advance to next segment in data. */
5627 /* How many characters left in this segment to match. */
5630 /* Want how many consecutive characters we can match in
5631 one shot, so, if necessary, adjust the count. */
5632 if (dcnt
> dend2
- d2
)
5635 /* Compare that many; failure if mismatch, else move
5637 if (RE_TRANSLATE_P (translate
)
5638 ? bcmp_translate (d
, d2
, dcnt
, translate
, target_multibyte
)
5639 : memcmp (d
, d2
, dcnt
))
5644 d
+= dcnt
, d2
+= dcnt
;
5650 /* begline matches the empty string at the beginning of the string
5651 (unless `not_bol' is set in `bufp'), and after newlines. */
5653 DEBUG_PRINT ("EXECUTING begline.\n");
5655 if (AT_STRINGS_BEG (d
))
5657 if (!bufp
->not_bol
) break;
5662 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5666 /* In all other cases, we fail. */
5670 /* endline is the dual of begline. */
5672 DEBUG_PRINT ("EXECUTING endline.\n");
5674 if (AT_STRINGS_END (d
))
5676 if (!bufp
->not_eol
) break;
5680 PREFETCH_NOLIMIT ();
5687 /* Match at the very beginning of the data. */
5689 DEBUG_PRINT ("EXECUTING begbuf.\n");
5690 if (AT_STRINGS_BEG (d
))
5695 /* Match at the very end of the data. */
5697 DEBUG_PRINT ("EXECUTING endbuf.\n");
5698 if (AT_STRINGS_END (d
))
5703 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5704 pushes NULL as the value for the string on the stack. Then
5705 `POP_FAILURE_POINT' will keep the current value for the
5706 string, instead of restoring it. To see why, consider
5707 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5708 then the . fails against the \n. But the next thing we want
5709 to do is match the \n against the \n; if we restored the
5710 string value, we would be back at the foo.
5712 Because this is used only in specific cases, we don't need to
5713 check all the things that `on_failure_jump' does, to make
5714 sure the right things get saved on the stack. Hence we don't
5715 share its code. The only reason to push anything on the
5716 stack at all is that otherwise we would have to change
5717 `anychar's code to do something besides goto fail in this
5718 case; that seems worse than this. */
5719 case on_failure_keep_string_jump
:
5720 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5721 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5724 PUSH_FAILURE_POINT (p
- 3, NULL
);
5727 /* A nasty loop is introduced by the non-greedy *? and +?.
5728 With such loops, the stack only ever contains one failure point
5729 at a time, so that a plain on_failure_jump_loop kind of
5730 cycle detection cannot work. Worse yet, such a detection
5731 can not only fail to detect a cycle, but it can also wrongly
5732 detect a cycle (between different instantiations of the same
5734 So the method used for those nasty loops is a little different:
5735 We use a special cycle-detection-stack-frame which is pushed
5736 when the on_failure_jump_nastyloop failure-point is *popped*.
5737 This special frame thus marks the beginning of one iteration
5738 through the loop and we can hence easily check right here
5739 whether something matched between the beginning and the end of
5741 case on_failure_jump_nastyloop
:
5742 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5743 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5746 assert ((re_opcode_t
)p
[-4] == no_op
);
5749 CHECK_INFINITE_LOOP (p
- 4, d
);
5751 /* If there's a cycle, just continue without pushing
5752 this failure point. The failure point is the "try again"
5753 option, which shouldn't be tried.
5754 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5755 PUSH_FAILURE_POINT (p
- 3, d
);
5759 /* Simple loop detecting on_failure_jump: just check on the
5760 failure stack if the same spot was already hit earlier. */
5761 case on_failure_jump_loop
:
5763 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5764 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5768 CHECK_INFINITE_LOOP (p
- 3, d
);
5770 /* If there's a cycle, get out of the loop, as if the matching
5771 had failed. We used to just `goto fail' here, but that was
5772 aborting the search a bit too early: we want to keep the
5773 empty-loop-match and keep matching after the loop.
5774 We want (x?)*y\1z to match both xxyz and xxyxz. */
5777 PUSH_FAILURE_POINT (p
- 3, d
);
5782 /* Uses of on_failure_jump:
5784 Each alternative starts with an on_failure_jump that points
5785 to the beginning of the next alternative. Each alternative
5786 except the last ends with a jump that in effect jumps past
5787 the rest of the alternatives. (They really jump to the
5788 ending jump of the following alternative, because tensioning
5789 these jumps is a hassle.)
5791 Repeats start with an on_failure_jump that points past both
5792 the repetition text and either the following jump or
5793 pop_failure_jump back to this on_failure_jump. */
5794 case on_failure_jump
:
5795 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5796 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5799 PUSH_FAILURE_POINT (p
-3, d
);
5802 /* This operation is used for greedy *.
5803 Compare the beginning of the repeat with what in the
5804 pattern follows its end. If we can establish that there
5805 is nothing that they would both match, i.e., that we
5806 would have to backtrack because of (as in, e.g., `a*a')
5807 then we can use a non-backtracking loop based on
5808 on_failure_keep_string_jump instead of on_failure_jump. */
5809 case on_failure_jump_smart
:
5810 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5811 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5814 re_char
*p1
= p
; /* Next operation. */
5815 /* Here, we discard `const', making re_match non-reentrant. */
5816 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5817 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5819 p
-= 3; /* Reset so that we will re-execute the
5820 instruction once it's been changed. */
5822 EXTRACT_NUMBER (mcnt
, p2
- 2);
5824 /* Ensure this is a indeed the trivial kind of loop
5825 we are expecting. */
5826 assert (skip_one_char (p1
) == p2
- 3);
5827 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5828 DEBUG_STATEMENT (debug
+= 2);
5829 if (mutually_exclusive_p (bufp
, p1
, p2
))
5831 /* Use a fast `on_failure_keep_string_jump' loop. */
5832 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5833 *p3
= (unsigned char) on_failure_keep_string_jump
;
5834 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5838 /* Default to a safe `on_failure_jump' loop. */
5839 DEBUG_PRINT (" smart default => slow loop.\n");
5840 *p3
= (unsigned char) on_failure_jump
;
5842 DEBUG_STATEMENT (debug
-= 2);
5846 /* Unconditionally jump (without popping any failure points). */
5849 IMMEDIATE_QUIT_CHECK
;
5850 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5851 DEBUG_PRINT ("EXECUTING jump %d ", mcnt
);
5852 p
+= mcnt
; /* Do the jump. */
5853 DEBUG_PRINT ("(to %p).\n", p
);
5857 /* Have to succeed matching what follows at least n times.
5858 After that, handle like `on_failure_jump'. */
5860 /* Signedness doesn't matter since we only compare MCNT to 0. */
5861 EXTRACT_NUMBER (mcnt
, p
+ 2);
5862 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt
);
5864 /* Originally, mcnt is how many times we HAVE to succeed. */
5867 /* Here, we discard `const', making re_match non-reentrant. */
5868 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5871 PUSH_NUMBER (p2
, mcnt
);
5874 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5879 /* Signedness doesn't matter since we only compare MCNT to 0. */
5880 EXTRACT_NUMBER (mcnt
, p
+ 2);
5881 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt
);
5883 /* Originally, this is how many times we CAN jump. */
5886 /* Here, we discard `const', making re_match non-reentrant. */
5887 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5889 PUSH_NUMBER (p2
, mcnt
);
5890 goto unconditional_jump
;
5892 /* If don't have to jump any more, skip over the rest of command. */
5899 unsigned char *p2
; /* Location of the counter. */
5900 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5902 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5903 /* Here, we discard `const', making re_match non-reentrant. */
5904 p2
= (unsigned char*) p
+ mcnt
;
5905 /* Signedness doesn't matter since we only copy MCNT's bits. */
5906 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5907 DEBUG_PRINT (" Setting %p to %d.\n", p2
, mcnt
);
5908 PUSH_NUMBER (p2
, mcnt
);
5915 boolean
not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5916 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5918 /* We SUCCEED (or FAIL) in one of the following cases: */
5920 /* Case 1: D is at the beginning or the end of string. */
5921 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5925 /* C1 is the character before D, S1 is the syntax of C1, C2
5926 is the character at D, and S2 is the syntax of C2. */
5931 ssize_t offset
= PTR_TO_OFFSET (d
- 1);
5932 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5933 UPDATE_SYNTAX_TABLE (charpos
);
5935 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5938 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5940 PREFETCH_NOLIMIT ();
5941 GET_CHAR_AFTER (c2
, d
, dummy
);
5944 if (/* Case 2: Only one of S1 and S2 is Sword. */
5945 ((s1
== Sword
) != (s2
== Sword
))
5946 /* Case 3: Both of S1 and S2 are Sword, and macro
5947 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5948 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5958 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5960 /* We FAIL in one of the following cases: */
5962 /* Case 1: D is at the end of string. */
5963 if (AT_STRINGS_END (d
))
5967 /* C1 is the character before D, S1 is the syntax of C1, C2
5968 is the character at D, and S2 is the syntax of C2. */
5973 ssize_t offset
= PTR_TO_OFFSET (d
);
5974 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5975 UPDATE_SYNTAX_TABLE (charpos
);
5978 GET_CHAR_AFTER (c2
, d
, dummy
);
5981 /* Case 2: S2 is not Sword. */
5985 /* Case 3: D is not at the beginning of string ... */
5986 if (!AT_STRINGS_BEG (d
))
5988 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5990 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5994 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5996 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6003 DEBUG_PRINT ("EXECUTING wordend.\n");
6005 /* We FAIL in one of the following cases: */
6007 /* Case 1: D is at the beginning of string. */
6008 if (AT_STRINGS_BEG (d
))
6012 /* C1 is the character before D, S1 is the syntax of C1, C2
6013 is the character at D, and S2 is the syntax of C2. */
6018 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6019 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6020 UPDATE_SYNTAX_TABLE (charpos
);
6022 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6025 /* Case 2: S1 is not Sword. */
6029 /* Case 3: D is not at the end of string ... */
6030 if (!AT_STRINGS_END (d
))
6032 PREFETCH_NOLIMIT ();
6033 GET_CHAR_AFTER (c2
, d
, dummy
);
6035 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6039 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6041 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6048 DEBUG_PRINT ("EXECUTING symbeg.\n");
6050 /* We FAIL in one of the following cases: */
6052 /* Case 1: D is at the end of string. */
6053 if (AT_STRINGS_END (d
))
6057 /* C1 is the character before D, S1 is the syntax of C1, C2
6058 is the character at D, and S2 is the syntax of C2. */
6062 ssize_t offset
= PTR_TO_OFFSET (d
);
6063 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6064 UPDATE_SYNTAX_TABLE (charpos
);
6067 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6070 /* Case 2: S2 is neither Sword nor Ssymbol. */
6071 if (s2
!= Sword
&& s2
!= Ssymbol
)
6074 /* Case 3: D is not at the beginning of string ... */
6075 if (!AT_STRINGS_BEG (d
))
6077 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6079 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6083 /* ... and S1 is Sword or Ssymbol. */
6084 if (s1
== Sword
|| s1
== Ssymbol
)
6091 DEBUG_PRINT ("EXECUTING symend.\n");
6093 /* We FAIL in one of the following cases: */
6095 /* Case 1: D is at the beginning of string. */
6096 if (AT_STRINGS_BEG (d
))
6100 /* C1 is the character before D, S1 is the syntax of C1, C2
6101 is the character at D, and S2 is the syntax of C2. */
6105 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6106 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6107 UPDATE_SYNTAX_TABLE (charpos
);
6109 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6112 /* Case 2: S1 is neither Ssymbol nor Sword. */
6113 if (s1
!= Sword
&& s1
!= Ssymbol
)
6116 /* Case 3: D is not at the end of string ... */
6117 if (!AT_STRINGS_END (d
))
6119 PREFETCH_NOLIMIT ();
6120 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6122 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6126 /* ... and S2 is Sword or Ssymbol. */
6127 if (s2
== Sword
|| s2
== Ssymbol
)
6136 boolean
not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6138 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6143 ssize_t offset
= PTR_TO_OFFSET (d
);
6144 ssize_t pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6145 UPDATE_SYNTAX_TABLE (pos1
);
6152 GET_CHAR_AFTER (c
, d
, len
);
6153 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6162 DEBUG_PRINT ("EXECUTING before_dot.\n");
6163 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6168 DEBUG_PRINT ("EXECUTING at_dot.\n");
6169 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6174 DEBUG_PRINT ("EXECUTING after_dot.\n");
6175 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6180 case notcategoryspec
:
6182 boolean
not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6184 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6185 not ? "not" : "", mcnt
);
6191 GET_CHAR_AFTER (c
, d
, len
);
6192 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6204 continue; /* Successfully executed one pattern command; keep going. */
6207 /* We goto here if a matching operation fails. */
6209 IMMEDIATE_QUIT_CHECK
;
6210 if (!FAIL_STACK_EMPTY ())
6213 /* A restart point is known. Restore to that state. */
6214 DEBUG_PRINT ("\nFAIL:\n");
6215 POP_FAILURE_POINT (str
, pat
);
6218 case on_failure_keep_string_jump
:
6219 assert (str
== NULL
);
6220 goto continue_failure_jump
;
6222 case on_failure_jump_nastyloop
:
6223 assert ((re_opcode_t
)pat
[-2] == no_op
);
6224 PUSH_FAILURE_POINT (pat
- 2, str
);
6227 case on_failure_jump_loop
:
6228 case on_failure_jump
:
6231 continue_failure_jump
:
6232 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6237 /* A special frame used for nastyloops. */
6244 assert (p
>= bufp
->buffer
&& p
<= pend
);
6246 if (d
>= string1
&& d
<= end1
)
6250 break; /* Matching at this starting point really fails. */
6254 goto restore_best_regs
;
6258 return -1; /* Failure to match. */
6261 /* Subroutine definitions for re_match_2. */
6263 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6264 bytes; nonzero otherwise. */
6267 bcmp_translate (const_re_char
*s1
, const_re_char
*s2
, register ssize_t len
,
6268 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6270 register re_char
*p1
= s1
, *p2
= s2
;
6271 re_char
*p1_end
= s1
+ len
;
6272 re_char
*p2_end
= s2
+ len
;
6274 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6275 different lengths, but relying on a single `len' would break this. -sm */
6276 while (p1
< p1_end
&& p2
< p2_end
)
6278 int p1_charlen
, p2_charlen
;
6279 re_wchar_t p1_ch
, p2_ch
;
6281 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6282 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6284 if (RE_TRANSLATE (translate
, p1_ch
)
6285 != RE_TRANSLATE (translate
, p2_ch
))
6288 p1
+= p1_charlen
, p2
+= p2_charlen
;
6291 if (p1
!= p1_end
|| p2
!= p2_end
)
6297 /* Entry points for GNU code. */
6299 /* re_compile_pattern is the GNU regular expression compiler: it
6300 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6301 Returns 0 if the pattern was valid, otherwise an error string.
6303 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6304 are set in BUFP on entry.
6306 We call regex_compile to do the actual compilation. */
6309 re_compile_pattern (const char *pattern
, size_t length
,
6310 struct re_pattern_buffer
*bufp
)
6314 /* GNU code is written to assume at least RE_NREGS registers will be set
6315 (and at least one extra will be -1). */
6316 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6318 /* And GNU code determines whether or not to get register information
6319 by passing null for the REGS argument to re_match, etc., not by
6323 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6327 return gettext (re_error_msgid
[(int) ret
]);
6329 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6331 /* Entry points compatible with 4.2 BSD regex library. We don't define
6332 them unless specifically requested. */
6334 #if defined _REGEX_RE_COMP || defined _LIBC
6336 /* BSD has one and only one pattern buffer. */
6337 static struct re_pattern_buffer re_comp_buf
;
6341 /* Make these definitions weak in libc, so POSIX programs can redefine
6342 these names if they don't use our functions, and still use
6343 regcomp/regexec below without link errors. */
6346 re_comp (const char *s
)
6352 if (!re_comp_buf
.buffer
)
6353 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6354 return (char *) gettext ("No previous regular expression");
6358 if (!re_comp_buf
.buffer
)
6360 re_comp_buf
.buffer
= malloc (200);
6361 if (re_comp_buf
.buffer
== NULL
)
6362 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6363 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6364 re_comp_buf
.allocated
= 200;
6366 re_comp_buf
.fastmap
= malloc (1 << BYTEWIDTH
);
6367 if (re_comp_buf
.fastmap
== NULL
)
6368 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6369 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6372 /* Since `re_exec' always passes NULL for the `regs' argument, we
6373 don't need to initialize the pattern buffer fields which affect it. */
6375 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6380 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6381 return (char *) gettext (re_error_msgid
[(int) ret
]);
6389 re_exec (const char *s
)
6391 const size_t len
= strlen (s
);
6392 return re_search (&re_comp_buf
, s
, len
, 0, len
, 0) >= 0;
6394 #endif /* _REGEX_RE_COMP */
6396 /* POSIX.2 functions. Don't define these for Emacs. */
6400 /* regcomp takes a regular expression as a string and compiles it.
6402 PREG is a regex_t *. We do not expect any fields to be initialized,
6403 since POSIX says we shouldn't. Thus, we set
6405 `buffer' to the compiled pattern;
6406 `used' to the length of the compiled pattern;
6407 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6408 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6409 RE_SYNTAX_POSIX_BASIC;
6410 `fastmap' to an allocated space for the fastmap;
6411 `fastmap_accurate' to zero;
6412 `re_nsub' to the number of subexpressions in PATTERN.
6414 PATTERN is the address of the pattern string.
6416 CFLAGS is a series of bits which affect compilation.
6418 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6419 use POSIX basic syntax.
6421 If REG_NEWLINE is set, then . and [^...] don't match newline.
6422 Also, regexec will try a match beginning after every newline.
6424 If REG_ICASE is set, then we considers upper- and lowercase
6425 versions of letters to be equivalent when matching.
6427 If REG_NOSUB is set, then when PREG is passed to regexec, that
6428 routine will report only success or failure, and nothing about the
6431 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6432 the return codes and their meanings.) */
6435 regcomp (regex_t
*_Restrict_ preg
, const char *_Restrict_ pattern
,
6440 = (cflags
& REG_EXTENDED
) ?
6441 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6443 /* regex_compile will allocate the space for the compiled pattern. */
6445 preg
->allocated
= 0;
6448 /* Try to allocate space for the fastmap. */
6449 preg
->fastmap
= malloc (1 << BYTEWIDTH
);
6451 if (cflags
& REG_ICASE
)
6455 preg
->translate
= malloc (CHAR_SET_SIZE
* sizeof *preg
->translate
);
6456 if (preg
->translate
== NULL
)
6457 return (int) REG_ESPACE
;
6459 /* Map uppercase characters to corresponding lowercase ones. */
6460 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6461 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6464 preg
->translate
= NULL
;
6466 /* If REG_NEWLINE is set, newlines are treated differently. */
6467 if (cflags
& REG_NEWLINE
)
6468 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6469 syntax
&= ~RE_DOT_NEWLINE
;
6470 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6473 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6475 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6477 /* POSIX says a null character in the pattern terminates it, so we
6478 can use strlen here in compiling the pattern. */
6479 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6481 /* POSIX doesn't distinguish between an unmatched open-group and an
6482 unmatched close-group: both are REG_EPAREN. */
6483 if (ret
== REG_ERPAREN
)
6486 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6487 { /* Compute the fastmap now, since regexec cannot modify the pattern
6489 re_compile_fastmap (preg
);
6490 if (preg
->can_be_null
)
6491 { /* The fastmap can't be used anyway. */
6492 free (preg
->fastmap
);
6493 preg
->fastmap
= NULL
;
6498 WEAK_ALIAS (__regcomp
, regcomp
)
6501 /* regexec searches for a given pattern, specified by PREG, in the
6504 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6505 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6506 least NMATCH elements, and we set them to the offsets of the
6507 corresponding matched substrings.
6509 EFLAGS specifies `execution flags' which affect matching: if
6510 REG_NOTBOL is set, then ^ does not match at the beginning of the
6511 string; if REG_NOTEOL is set, then $ does not match at the end.
6513 We return 0 if we find a match and REG_NOMATCH if not. */
6516 regexec (const regex_t
*_Restrict_ preg
, const char *_Restrict_ string
,
6517 size_t nmatch
, regmatch_t pmatch
[_Restrict_arr_
], int eflags
)
6520 struct re_registers regs
;
6521 regex_t private_preg
;
6522 size_t len
= strlen (string
);
6523 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6525 private_preg
= *preg
;
6527 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6528 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6530 /* The user has told us exactly how many registers to return
6531 information about, via `nmatch'. We have to pass that on to the
6532 matching routines. */
6533 private_preg
.regs_allocated
= REGS_FIXED
;
6537 regs
.num_regs
= nmatch
;
6538 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6539 if (regs
.start
== NULL
)
6541 regs
.end
= regs
.start
+ nmatch
;
6544 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6545 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6546 was a little bit longer but still only matching the real part.
6547 This works because the `endline' will check for a '\n' and will find a
6548 '\0', correctly deciding that this is not the end of a line.
6549 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6550 a convenient '\0' there. For all we know, the string could be preceded
6551 by '\n' which would throw things off. */
6553 /* Perform the searching operation. */
6554 ret
= re_search (&private_preg
, string
, len
,
6555 /* start: */ 0, /* range: */ len
,
6556 want_reg_info
? ®s
: 0);
6558 /* Copy the register information to the POSIX structure. */
6565 for (r
= 0; r
< nmatch
; r
++)
6567 pmatch
[r
].rm_so
= regs
.start
[r
];
6568 pmatch
[r
].rm_eo
= regs
.end
[r
];
6572 /* If we needed the temporary register info, free the space now. */
6576 /* We want zero return to mean success, unlike `re_search'. */
6577 return ret
>= 0 ? REG_NOERROR
: REG_NOMATCH
;
6579 WEAK_ALIAS (__regexec
, regexec
)
6582 /* Returns a message corresponding to an error code, ERR_CODE, returned
6583 from either regcomp or regexec. We don't use PREG here.
6585 ERR_CODE was previously called ERRCODE, but that name causes an
6586 error with msvc8 compiler. */
6589 regerror (int err_code
, const regex_t
*preg
, char *errbuf
, size_t errbuf_size
)
6595 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6596 /* Only error codes returned by the rest of the code should be passed
6597 to this routine. If we are given anything else, or if other regex
6598 code generates an invalid error code, then the program has a bug.
6599 Dump core so we can fix it. */
6602 msg
= gettext (re_error_msgid
[err_code
]);
6604 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6606 if (errbuf_size
!= 0)
6608 if (msg_size
> errbuf_size
)
6610 memcpy (errbuf
, msg
, errbuf_size
- 1);
6611 errbuf
[errbuf_size
- 1] = 0;
6614 strcpy (errbuf
, msg
);
6619 WEAK_ALIAS (__regerror
, regerror
)
6622 /* Free dynamically allocated space used by PREG. */
6625 regfree (regex_t
*preg
)
6627 free (preg
->buffer
);
6628 preg
->buffer
= NULL
;
6630 preg
->allocated
= 0;
6633 free (preg
->fastmap
);
6634 preg
->fastmap
= NULL
;
6635 preg
->fastmap_accurate
= 0;
6637 free (preg
->translate
);
6638 preg
->translate
= NULL
;
6640 WEAK_ALIAS (__regfree
, regfree
)
6642 #endif /* not emacs */