| blob | b39920a8af409ecdcdd79988ff523862b7e927b6 |
1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the
3 internationalization features.)
5 Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
6 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011
7 Free Software Foundation, Inc.
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 3, or (at your option)
12 any later version.
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; if not, write to the Free Software
21 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
22 USA. */
24 /* TODO:
25 - structure the opcode space into opcode+flag.
26 - merge with glibc's regex.[ch].
27 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
28 need to modify the compiled regexp so that re_match can be reentrant.
29 - get rid of on_failure_jump_smart by doing the optimization in re_comp
30 rather than at run-time, so that re_match can be reentrant.
31 */
33 /* AIX requires this to be the first thing in the file. */
34 #if defined _AIX && !defined REGEX_MALLOC
35 #pragma alloca
36 #endif
38 #ifdef HAVE_CONFIG_H
39 # include <config.h>
40 #endif
42 #if defined STDC_HEADERS && !defined emacs
43 # include <stddef.h>
44 #else
45 /* We need this for `regex.h', and perhaps for the Emacs include files. */
46 # include <sys/types.h>
47 #endif
49 /* Whether to use ISO C Amendment 1 wide char functions.
50 Those should not be used for Emacs since it uses its own. */
51 #if defined _LIBC
52 #define WIDE_CHAR_SUPPORT 1
53 #else
54 #define WIDE_CHAR_SUPPORT \
55 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
56 #endif
58 /* For platform which support the ISO C amendement 1 functionality we
59 support user defined character classes. */
60 #if WIDE_CHAR_SUPPORT
61 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
62 # include <wchar.h>
63 # include <wctype.h>
64 #endif
66 #ifdef _LIBC
67 /* We have to keep the namespace clean. */
68 # define regfree(preg) __regfree (preg)
69 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
70 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
71 # define regerror(err_code, preg, errbuf, errbuf_size) \
72 __regerror(err_code, preg, errbuf, errbuf_size)
73 # define re_set_registers(bu, re, nu, st, en) \
74 __re_set_registers (bu, re, nu, st, en)
75 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
76 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
77 # define re_match(bufp, string, size, pos, regs) \
78 __re_match (bufp, string, size, pos, regs)
79 # define re_search(bufp, string, size, startpos, range, regs) \
80 __re_search (bufp, string, size, startpos, range, regs)
81 # define re_compile_pattern(pattern, length, bufp) \
82 __re_compile_pattern (pattern, length, bufp)
83 # define re_set_syntax(syntax) __re_set_syntax (syntax)
84 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
85 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
86 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
88 /* Make sure we call libc's function even if the user overrides them. */
89 # define btowc __btowc
90 # define iswctype __iswctype
91 # define wctype __wctype
93 # define WEAK_ALIAS(a,b) weak_alias (a, b)
95 /* We are also using some library internals. */
96 # include <locale/localeinfo.h>
97 # include <locale/elem-hash.h>
98 # include <langinfo.h>
99 #else
100 # define WEAK_ALIAS(a,b)
101 #endif
103 /* This is for other GNU distributions with internationalized messages. */
104 #if HAVE_LIBINTL_H || defined _LIBC
105 # include <libintl.h>
106 #else
107 # define gettext(msgid) (msgid)
108 #endif
110 #ifndef gettext_noop
111 /* This define is so xgettext can find the internationalizable
112 strings. */
113 # define gettext_noop(String) String
114 #endif
116 /* The `emacs' switch turns on certain matching commands
117 that make sense only in Emacs. */
118 #ifdef emacs
120 # include <setjmp.h>
124 /* Make syntax table lookup grant data in gl_state. */
125 # define SYNTAX_ENTRY_VIA_PROPERTY
131 # ifdef malloc
132 # undef malloc
133 # endif
134 # define malloc xmalloc
135 # ifdef realloc
136 # undef realloc
137 # endif
138 # define realloc xrealloc
139 # ifdef free
140 # undef free
141 # endif
142 # define free xfree
144 /* Converts the pointer to the char to BEG-based offset from the start. */
145 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
146 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
148 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
149 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
150 # define RE_STRING_CHAR(p, multibyte) \
151 (multibyte ? (STRING_CHAR (p)) : (*(p)))
152 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
153 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
155 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
157 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
159 /* Set C a (possibly converted to multibyte) character before P. P
160 points into a string which is the virtual concatenation of STR1
161 (which ends at END1) or STR2 (which ends at END2). */
162 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
163 do { \
164 if (target_multibyte) \
165 { \
166 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
167 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
168 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
169 c = STRING_CHAR (dtemp); \
170 } \
171 else \
172 { \
173 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
174 (c) = RE_CHAR_TO_MULTIBYTE (c); \
175 } \
176 } while (0)
178 /* Set C a (possibly converted to multibyte) character at P, and set
179 LEN to the byte length of that character. */
180 # define GET_CHAR_AFTER(c, p, len) \
181 do { \
182 if (target_multibyte) \
183 (c) = STRING_CHAR_AND_LENGTH (p, len); \
184 else \
185 { \
186 (c) = *p; \
187 len = 1; \
188 (c) = RE_CHAR_TO_MULTIBYTE (c); \
189 } \
190 } while (0)
194 /* If we are not linking with Emacs proper,
195 we can't use the relocating allocator
196 even if config.h says that we can. */
197 # undef REL_ALLOC
199 # include <unistd.h>
201 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
205 {
209 {
212 }
214 }
218 {
220 /* We must call malloc explicitly when BLOCK is 0, since some
221 reallocs don't do this. */
224 else
227 {
230 }
232 }
234 # ifdef malloc
235 # undef malloc
236 # endif
237 # define malloc xmalloc
238 # ifdef realloc
239 # undef realloc
240 # endif
241 # define realloc xrealloc
243 /* This is the normal way of making sure we have memcpy, memcmp and memset. */
244 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
245 # include <string.h>
246 # else
247 # include <strings.h>
248 # ifndef memcmp
249 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
250 # endif
251 # ifndef memcpy
252 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
253 # endif
254 # endif
256 /* Define the syntax stuff for \<, \>, etc. */
258 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
261 # define SWITCH_ENUM_CAST(x) (x)
263 /* Dummy macros for non-Emacs environments. */
264 # define CHAR_CHARSET(c) 0
265 # define CHARSET_LEADING_CODE_BASE(c) 0
266 # define MAX_MULTIBYTE_LENGTH 1
267 # define RE_MULTIBYTE_P(x) 0
268 # define RE_TARGET_MULTIBYTE_P(x) 0
269 # define WORD_BOUNDARY_P(c1, c2) (0)
270 # define CHAR_HEAD_P(p) (1)
271 # define SINGLE_BYTE_CHAR_P(c) (1)
272 # define SAME_CHARSET_P(c1, c2) (1)
273 # define BYTES_BY_CHAR_HEAD(p) (1)
274 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
275 # define STRING_CHAR(p) (*(p))
276 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
277 # define CHAR_STRING(c, s) (*(s) = (c), 1)
278 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
279 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
280 # define RE_CHAR_TO_MULTIBYTE(c) (c)
281 # define RE_CHAR_TO_UNIBYTE(c) (c)
282 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
283 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
284 # define GET_CHAR_AFTER(c, p, len) \
285 (c = *p, len = 1)
286 # define MAKE_CHAR(charset, c1, c2) (c1)
287 # define BYTE8_TO_CHAR(c) (c)
288 # define CHAR_BYTE8_P(c) (0)
289 # define CHAR_LEADING_CODE(c) (c)
293 #ifndef RE_TRANSLATE
294 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
295 # define RE_TRANSLATE_P(TBL) (TBL)
296 #endif
297 \f
298 /* Get the interface, including the syntax bits. */
301 /* isalpha etc. are used for the character classes. */
302 #include <ctype.h>
304 #ifdef emacs
306 /* 1 if C is an ASCII character. */
307 # define IS_REAL_ASCII(c) ((c) < 0200)
309 /* 1 if C is a unibyte character. */
310 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
312 /* The Emacs definitions should not be directly affected by locales. */
314 /* In Emacs, these are only used for single-byte characters. */
321 /* This is only used for single-byte characters. */
324 /* The rest must handle multibyte characters. */
326 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
328 : 1)
330 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
332 : 1)
334 # define ISALNUM(c) (IS_REAL_ASCII (c) \
338 : SYNTAX (c) == Sword)
340 # define ISALPHA(c) (IS_REAL_ASCII (c) \
343 : SYNTAX (c) == Sword)
345 # define ISLOWER(c) (LOWERCASEP (c))
347 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
352 : SYNTAX (c) != Sword)
354 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
356 # define ISUPPER(c) (UPPERCASEP (c))
358 # define ISWORD(c) (SYNTAX (c) == Sword)
362 /* Jim Meyering writes:
364 "... Some ctype macros are valid only for character codes that
365 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
366 using /bin/cc or gcc but without giving an ansi option). So, all
367 ctype uses should be through macros like ISPRINT... If
368 STDC_HEADERS is defined, then autoconf has verified that the ctype
369 macros don't need to be guarded with references to isascii. ...
370 Defining isascii to 1 should let any compiler worth its salt
371 eliminate the && through constant folding."
372 Solaris defines some of these symbols so we must undefine them first. */
374 # undef ISASCII
375 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
376 # define ISASCII(c) 1
377 # else
378 # define ISASCII(c) isascii(c)
379 # endif
381 /* 1 if C is an ASCII character. */
382 # define IS_REAL_ASCII(c) ((c) < 0200)
384 /* This distinction is not meaningful, except in Emacs. */
385 # define ISUNIBYTE(c) 1
387 # ifdef isblank
388 # define ISBLANK(c) (ISASCII (c) && isblank (c))
389 # else
391 # endif
392 # ifdef isgraph
393 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
394 # else
395 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
396 # endif
398 # undef ISPRINT
399 # define ISPRINT(c) (ISASCII (c) && isprint (c))
400 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
401 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
402 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
403 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
404 # define ISLOWER(c) (ISASCII (c) && islower (c))
405 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
406 # define ISSPACE(c) (ISASCII (c) && isspace (c))
407 # define ISUPPER(c) (ISASCII (c) && isupper (c))
408 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
410 # define ISWORD(c) ISALPHA(c)
412 # ifdef _tolower
413 # define TOLOWER(c) _tolower(c)
414 # else
415 # define TOLOWER(c) tolower(c)
416 # endif
418 /* How many characters in the character set. */
419 # define CHAR_SET_SIZE 256
421 # ifdef SYNTAX_TABLE
429 static void
431 {
447 }
451 # define SYNTAX(c) re_syntax_table[(c)]
454 \f
455 #ifndef NULL
456 # define NULL (void *)0
457 #endif
459 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
460 since ours (we hope) works properly with all combinations of
461 machines, compilers, `char' and `unsigned char' argument types.
462 (Per Bothner suggested the basic approach.) */
463 #undef SIGN_EXTEND_CHAR
464 #if __STDC__
465 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
467 /* As in Harbison and Steele. */
468 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
469 #endif
470 \f
471 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
472 use `alloca' instead of `malloc'. This is because using malloc in
473 re_search* or re_match* could cause memory leaks when C-g is used in
474 Emacs; also, malloc is slower and causes storage fragmentation. On
475 the other hand, malloc is more portable, and easier to debug.
477 Because we sometimes use alloca, some routines have to be macros,
478 not functions -- `alloca'-allocated space disappears at the end of the
479 function it is called in. */
481 #ifdef REGEX_MALLOC
483 # define REGEX_ALLOCATE malloc
484 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
485 # define REGEX_FREE free
489 /* Emacs already defines alloca, sometimes. */
490 # ifndef alloca
492 /* Make alloca work the best possible way. */
493 # ifdef __GNUC__
494 # define alloca __builtin_alloca
496 # ifdef HAVE_ALLOCA_H
497 # include <alloca.h>
503 # define REGEX_ALLOCATE alloca
505 /* Assumes a `char *destination' variable. */
506 # define REGEX_REALLOCATE(source, osize, nsize) \
507 (destination = (char *) alloca (nsize), \
508 memcpy (destination, source, osize))
510 /* No need to do anything to free, after alloca. */
515 /* Define how to allocate the failure stack. */
517 #if defined REL_ALLOC && defined REGEX_MALLOC
519 # define REGEX_ALLOCATE_STACK(size) \
520 r_alloc (&failure_stack_ptr, (size))
521 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
522 r_re_alloc (&failure_stack_ptr, (nsize))
523 # define REGEX_FREE_STACK(ptr) \
524 r_alloc_free (&failure_stack_ptr)
528 # ifdef REGEX_MALLOC
530 # define REGEX_ALLOCATE_STACK malloc
531 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
532 # define REGEX_FREE_STACK free
536 # define REGEX_ALLOCATE_STACK alloca
538 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
539 REGEX_REALLOCATE (source, osize, nsize)
540 /* No need to explicitly free anything. */
541 # define REGEX_FREE_STACK(arg) ((void)0)
547 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
548 `string1' or just past its end. This works if PTR is NULL, which is
549 a good thing. */
550 #define FIRST_STRING_P(ptr) \
551 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
553 /* (Re)Allocate N items of type T using malloc, or fail. */
554 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
555 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
556 #define RETALLOC_IF(addr, n, t) \
557 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
558 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
562 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
564 #undef MAX
565 #undef MIN
566 #define MAX(a, b) ((a) > (b) ? (a) : (b))
567 #define MIN(a, b) ((a) < (b) ? (a) : (b))
569 /* Type of source-pattern and string chars. */
573 #define false 0
574 #define true 1
582 \f
583 /* These are the command codes that appear in compiled regular
584 expressions. Some opcodes are followed by argument bytes. A
585 command code can specify any interpretation whatsoever for its
586 arguments. Zero bytes may appear in the compiled regular expression. */
589 {
592 /* Succeed right away--no more backtracking. */
593 succeed,
595 /* Followed by one byte giving n, then by n literal bytes. */
596 exactn,
598 /* Matches any (more or less) character. */
599 anychar,
601 /* Matches any one char belonging to specified set. First
602 following byte is number of bitmap bytes. Then come bytes
603 for a bitmap saying which chars are in. Bits in each byte
604 are ordered low-bit-first. A character is in the set if its
605 bit is 1. A character too large to have a bit in the map is
606 automatically not in the set.
608 If the length byte has the 0x80 bit set, then that stuff
609 is followed by a range table:
610 2 bytes of flags for character sets (low 8 bits, high 8 bits)
611 See RANGE_TABLE_WORK_BITS below.
612 2 bytes, the number of pairs that follow (upto 32767)
613 pairs, each 2 multibyte characters,
614 each multibyte character represented as 3 bytes. */
615 charset,
617 /* Same parameters as charset, but match any character that is
618 not one of those specified. */
619 charset_not,
621 /* Start remembering the text that is matched, for storing in a
622 register. Followed by one byte with the register number, in
623 the range 0 to one less than the pattern buffer's re_nsub
624 field. */
625 start_memory,
627 /* Stop remembering the text that is matched and store it in a
628 memory register. Followed by one byte with the register
629 number, in the range 0 to one less than `re_nsub' in the
630 pattern buffer. */
631 stop_memory,
633 /* Match a duplicate of something remembered. Followed by one
634 byte containing the register number. */
635 duplicate,
637 /* Fail unless at beginning of line. */
638 begline,
640 /* Fail unless at end of line. */
641 endline,
643 /* Succeeds if at beginning of buffer (if emacs) or at beginning
644 of string to be matched (if not). */
645 begbuf,
647 /* Analogously, for end of buffer/string. */
648 endbuf,
650 /* Followed by two byte relative address to which to jump. */
651 jump,
653 /* Followed by two-byte relative address of place to resume at
654 in case of failure. */
655 on_failure_jump,
657 /* Like on_failure_jump, but pushes a placeholder instead of the
658 current string position when executed. */
659 on_failure_keep_string_jump,
661 /* Just like `on_failure_jump', except that it checks that we
662 don't get stuck in an infinite loop (matching an empty string
663 indefinitely). */
664 on_failure_jump_loop,
666 /* Just like `on_failure_jump_loop', except that it checks for
667 a different kind of loop (the kind that shows up with non-greedy
668 operators). This operation has to be immediately preceded
669 by a `no_op'. */
670 on_failure_jump_nastyloop,
672 /* A smart `on_failure_jump' used for greedy * and + operators.
673 It analyses the loop before which it is put and if the
674 loop does not require backtracking, it changes itself to
675 `on_failure_keep_string_jump' and short-circuits the loop,
676 else it just defaults to changing itself into `on_failure_jump'.
677 It assumes that it is pointing to just past a `jump'. */
678 on_failure_jump_smart,
680 /* Followed by two-byte relative address and two-byte number n.
681 After matching N times, jump to the address upon failure.
682 Does not work if N starts at 0: use on_failure_jump_loop
683 instead. */
684 succeed_n,
686 /* Followed by two-byte relative address, and two-byte number n.
687 Jump to the address N times, then fail. */
688 jump_n,
690 /* Set the following two-byte relative address to the
691 subsequent two-byte number. The address *includes* the two
692 bytes of number. */
693 set_number_at,
704 /* Matches any character whose syntax is specified. Followed by
705 a byte which contains a syntax code, e.g., Sword. */
706 syntaxspec,
708 /* Matches any character whose syntax is not that specified. */
709 notsyntaxspec
711 #ifdef emacs
716 /* Matches any character whose category-set contains the specified
717 category. The operator is followed by a byte which contains a
718 category code (mnemonic ASCII character). */
719 categoryspec,
721 /* Matches any character whose category-set does not contain the
722 specified category. The operator is followed by a byte which
723 contains the category code (mnemonic ASCII character). */
724 notcategoryspec
727 \f
728 /* Common operations on the compiled pattern. */
730 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
732 #define STORE_NUMBER(destination, number) \
733 do { \
734 (destination)[0] = (number) & 0377; \
735 (destination)[1] = (number) >> 8; \
736 } while (0)
738 /* Same as STORE_NUMBER, except increment DESTINATION to
739 the byte after where the number is stored. Therefore, DESTINATION
740 must be an lvalue. */
742 #define STORE_NUMBER_AND_INCR(destination, number) \
743 do { \
744 STORE_NUMBER (destination, number); \
745 (destination) += 2; \
746 } while (0)
748 /* Put into DESTINATION a number stored in two contiguous bytes starting
749 at SOURCE. */
751 #define EXTRACT_NUMBER(destination, source) \
752 do { \
753 (destination) = *(source) & 0377; \
754 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
755 } while (0)
757 #ifdef DEBUG
759 static void
763 {
767 }
770 # undef EXTRACT_NUMBER
771 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
776 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
777 SOURCE must be an lvalue. */
779 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
780 do { \
781 EXTRACT_NUMBER (destination, source); \
782 (source) += 2; \
783 } while (0)
785 #ifdef DEBUG
788 static void
792 {
795 }
797 # ifndef EXTRACT_MACROS
798 # undef EXTRACT_NUMBER_AND_INCR
799 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
800 extract_number_and_incr (&dest, &src)
804 \f
805 /* Store a multibyte character in three contiguous bytes starting
806 DESTINATION, and increment DESTINATION to the byte after where the
807 character is stored. Therefore, DESTINATION must be an lvalue. */
809 #define STORE_CHARACTER_AND_INCR(destination, character) \
810 do { \
811 (destination)[0] = (character) & 0377; \
812 (destination)[1] = ((character) >> 8) & 0377; \
813 (destination)[2] = (character) >> 16; \
814 (destination) += 3; \
815 } while (0)
817 /* Put into DESTINATION a character stored in three contiguous bytes
818 starting at SOURCE. */
820 #define EXTRACT_CHARACTER(destination, source) \
821 do { \
822 (destination) = ((source)[0] \
823 | ((source)[1] << 8) \
824 | ((source)[2] << 16)); \
825 } while (0)
828 /* Macros for charset. */
830 /* Size of bitmap of charset P in bytes. P is a start of charset,
831 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
832 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
834 /* Nonzero if charset P has range table. */
835 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
837 /* Return the address of range table of charset P. But not the start
838 of table itself, but the before where the number of ranges is
839 stored. `2 +' means to skip re_opcode_t and size of bitmap,
840 and the 2 bytes of flags at the start of the range table. */
841 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
843 /* Extract the bit flags that start a range table. */
844 #define CHARSET_RANGE_TABLE_BITS(p) \
845 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
846 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
848 /* Test if C is listed in the bitmap of charset P. */
849 #define CHARSET_LOOKUP_BITMAP(p, c) \
850 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
851 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
853 /* Return the address of end of RANGE_TABLE. COUNT is number of
854 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
855 is start of range and end of range. `* 3' is size of each start
856 and end. */
857 #define CHARSET_RANGE_TABLE_END(range_table, count) \
858 ((range_table) + (count) * 2 * 3)
860 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
861 COUNT is number of ranges in RANGE_TABLE. */
862 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
863 do \
864 { \
865 re_wchar_t range_start, range_end; \
866 re_char *p; \
867 re_char *range_table_end \
868 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
869 \
870 for (p = (range_table); p < range_table_end; p += 2 * 3) \
871 { \
872 EXTRACT_CHARACTER (range_start, p); \
873 EXTRACT_CHARACTER (range_end, p + 3); \
874 \
875 if (range_start <= (c) && (c) <= range_end) \
876 { \
877 (not) = !(not); \
878 break; \
879 } \
880 } \
881 } \
882 while (0)
884 /* Test if C is in range table of CHARSET. The flag NOT is negated if
885 C is listed in it. */
886 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
887 do \
888 { \
890 int count; \
891 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
892 \
893 EXTRACT_NUMBER_AND_INCR (count, range_table); \
894 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
895 } \
896 while (0)
897 \f
898 /* If DEBUG is defined, Regex prints many voluminous messages about what
899 it is doing (if the variable `debug' is nonzero). If linked with the
900 main program in `iregex.c', you can enter patterns and strings
901 interactively. And if linked with the main program in `main.c' and
902 the other test files, you can run the already-written tests. */
904 #ifdef DEBUG
906 /* We use standard I/O for debugging. */
907 # include <stdio.h>
909 /* It is useful to test things that ``must'' be true when debugging. */
910 # include <assert.h>
914 # define DEBUG_STATEMENT(e) e
915 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
916 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
917 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
918 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
919 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
920 if (debug > 0) print_partial_compiled_pattern (s, e)
921 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
922 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
925 /* Print the fastmap in human-readable form. */
927 void
930 {
935 {
937 {
941 {
943 i++;
944 }
946 {
949 }
950 }
951 }
953 }
956 /* Print a compiled pattern string in human-readable form, starting at
957 the START pointer into it and ending just before the pointer END. */
959 void
963 {
969 {
972 }
974 /* Loop over pattern commands. */
976 {
980 {
992 do
993 {
995 }
1017 {
1032 {
1033 /* Are we starting a range? */
1035 {
1038 }
1039 /* Have we broken a range? */
1041 {
1044 }
1050 }
1060 {
1064 /* ??? Should print the range table; for now, just skip it. */
1068 }
1069 }
1164 # ifdef emacs
1200 }
1203 }
1206 }
1209 void
1212 {
1220 {
1223 }
1233 /* Perhaps we should print the translate table? */
1234 }
1237 void
1244 {
1249 else
1250 {
1252 {
1257 }
1261 }
1262 }
1266 # undef assert
1267 # define assert(e)
1269 # define DEBUG_STATEMENT(e)
1270 # define DEBUG_PRINT1(x)
1271 # define DEBUG_PRINT2(x1, x2)
1272 # define DEBUG_PRINT3(x1, x2, x3)
1273 # define DEBUG_PRINT4(x1, x2, x3, x4)
1274 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1275 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1278 \f
1279 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1280 also be assigned to arbitrarily: each pattern buffer stores its own
1281 syntax, so it can be changed between regex compilations. */
1282 /* This has no initializer because initialized variables in Emacs
1283 become read-only after dumping. */
1284 reg_syntax_t re_syntax_options;
1287 /* Specify the precise syntax of regexps for compilation. This provides
1288 for compatibility for various utilities which historically have
1289 different, incompatible syntaxes.
1291 The argument SYNTAX is a bit mask comprised of the various bits
1292 defined in regex.h. We return the old syntax. */
1294 reg_syntax_t
1296 {
1301 }
1304 /* Regexp to use to replace spaces, or NULL meaning don't. */
1307 void
1309 {
1311 }
1313 \f
1314 /* This table gives an error message for each of the error codes listed
1315 in regex.h. Obviously the order here has to be same as there.
1316 POSIX doesn't require that we do anything for REG_NOERROR,
1317 but why not be nice? */
1320 {
1339 };
1340 \f
1341 /* Avoiding alloca during matching, to placate r_alloc. */
1343 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1344 searching and matching functions should not call alloca. On some
1345 systems, alloca is implemented in terms of malloc, and if we're
1346 using the relocating allocator routines, then malloc could cause a
1347 relocation, which might (if the strings being searched are in the
1348 ralloc heap) shift the data out from underneath the regexp
1349 routines.
1351 Here's another reason to avoid allocation: Emacs
1352 processes input from X in a signal handler; processing X input may
1353 call malloc; if input arrives while a matching routine is calling
1354 malloc, then we're scrod. But Emacs can't just block input while
1355 calling matching routines; then we don't notice interrupts when
1356 they come in. So, Emacs blocks input around all regexp calls
1357 except the matching calls, which it leaves unprotected, in the
1358 faith that they will not malloc. */
1360 /* Normally, this is fine. */
1361 #define MATCH_MAY_ALLOCATE
1363 /* The match routines may not allocate if (1) they would do it with malloc
1364 and (2) it's not safe for them to use malloc.
1365 Note that if REL_ALLOC is defined, matching would not use malloc for the
1366 failure stack, but we would still use it for the register vectors;
1367 so REL_ALLOC should not affect this. */
1368 #if defined REGEX_MALLOC && defined emacs
1369 # undef MATCH_MAY_ALLOCATE
1370 #endif
1372 \f
1373 /* Failure stack declarations and macros; both re_compile_fastmap and
1374 re_match_2 use a failure stack. These have to be macros because of
1375 REGEX_ALLOCATE_STACK. */
1378 /* Approximate number of failure points for which to initially allocate space
1379 when matching. If this number is exceeded, we allocate more
1380 space, so it is not a hard limit. */
1381 #ifndef INIT_FAILURE_ALLOC
1382 # define INIT_FAILURE_ALLOC 20
1383 #endif
1385 /* Roughly the maximum number of failure points on the stack. Would be
1386 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1387 This is a variable only so users of regex can assign to it; we never
1388 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1389 before using it, so it should probably be a byte-count instead. */
1390 # if defined MATCH_MAY_ALLOCATE
1391 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1392 whose default stack limit is 2mb. In order for a larger
1393 value to work reliably, you have to try to make it accord
1394 with the process stack limit. */
1396 # else
1398 # endif
1400 union fail_stack_elt
1401 {
1403 /* This should be the biggest `int' that's no bigger than a pointer. */
1405 };
1410 {
1417 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1418 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1421 /* Define macros to initialize and free the failure stack.
1422 Do `return -2' if the alloc fails. */
1424 #ifdef MATCH_MAY_ALLOCATE
1425 # define INIT_FAIL_STACK() \
1426 do { \
1427 fail_stack.stack = (fail_stack_elt_t *) \
1428 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1429 * sizeof (fail_stack_elt_t)); \
1430 \
1431 if (fail_stack.stack == NULL) \
1432 return -2; \
1433 \
1434 fail_stack.size = INIT_FAILURE_ALLOC; \
1435 fail_stack.avail = 0; \
1436 fail_stack.frame = 0; \
1437 } while (0)
1439 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1440 #else
1441 # define INIT_FAIL_STACK() \
1442 do { \
1443 fail_stack.avail = 0; \
1444 fail_stack.frame = 0; \
1445 } while (0)
1447 # define RESET_FAIL_STACK() ((void)0)
1448 #endif
1451 /* Double the size of FAIL_STACK, up to a limit
1452 which allows approximately `re_max_failures' items.
1454 Return 1 if succeeds, and 0 if either ran out of memory
1455 allocating space for it or it was already too large.
1457 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1459 /* Factor to increase the failure stack size by
1460 when we increase it.
1461 This used to be 2, but 2 was too wasteful
1462 because the old discarded stacks added up to as much space
1463 were as ultimate, maximum-size stack. */
1464 #define FAIL_STACK_GROWTH_FACTOR 4
1466 #define GROW_FAIL_STACK(fail_stack) \
1467 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1468 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1469 ? 0 \
1470 : ((fail_stack).stack \
1471 = (fail_stack_elt_t *) \
1472 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1473 (fail_stack).size * sizeof (fail_stack_elt_t), \
1474 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1475 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1476 * FAIL_STACK_GROWTH_FACTOR))), \
1477 \
1478 (fail_stack).stack == NULL \
1479 ? 0 \
1480 : ((fail_stack).size \
1481 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1482 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1483 * FAIL_STACK_GROWTH_FACTOR)) \
1484 / sizeof (fail_stack_elt_t)), \
1485 1)))
1488 /* Push a pointer value onto the failure stack.
1489 Assumes the variable `fail_stack'. Probably should only
1490 be called from within `PUSH_FAILURE_POINT'. */
1491 #define PUSH_FAILURE_POINTER(item) \
1492 fail_stack.stack[fail_stack.avail++].pointer = (item)
1494 /* This pushes an integer-valued item onto the failure stack.
1495 Assumes the variable `fail_stack'. Probably should only
1496 be called from within `PUSH_FAILURE_POINT'. */
1497 #define PUSH_FAILURE_INT(item) \
1498 fail_stack.stack[fail_stack.avail++].integer = (item)
1500 /* Push a fail_stack_elt_t value onto the failure stack.
1501 Assumes the variable `fail_stack'. Probably should only
1502 be called from within `PUSH_FAILURE_POINT'. */
1503 #define PUSH_FAILURE_ELT(item) \
1504 fail_stack.stack[fail_stack.avail++] = (item)
1506 /* These three POP... operations complement the three PUSH... operations.
1507 All assume that `fail_stack' is nonempty. */
1508 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1509 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1510 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1512 /* Individual items aside from the registers. */
1513 #define NUM_NONREG_ITEMS 3
1515 /* Used to examine the stack (to detect infinite loops). */
1516 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1517 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1518 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1519 #define TOP_FAILURE_HANDLE() fail_stack.frame
1522 #define ENSURE_FAIL_STACK(space) \
1523 while (REMAINING_AVAIL_SLOTS <= space) { \
1524 if (!GROW_FAIL_STACK (fail_stack)) \
1525 return -2; \
1528 }
1530 /* Push register NUM onto the stack. */
1531 #define PUSH_FAILURE_REG(num) \
1532 do { \
1533 char *destination; \
1534 ENSURE_FAIL_STACK(3); \
1536 num, regstart[num], regend[num]); \
1537 PUSH_FAILURE_POINTER (regstart[num]); \
1538 PUSH_FAILURE_POINTER (regend[num]); \
1539 PUSH_FAILURE_INT (num); \
1540 } while (0)
1542 /* Change the counter's value to VAL, but make sure that it will
1543 be reset when backtracking. */
1544 #define PUSH_NUMBER(ptr,val) \
1545 do { \
1546 char *destination; \
1547 int c; \
1548 ENSURE_FAIL_STACK(3); \
1549 EXTRACT_NUMBER (c, ptr); \
1551 PUSH_FAILURE_INT (c); \
1552 PUSH_FAILURE_POINTER (ptr); \
1553 PUSH_FAILURE_INT (-1); \
1554 STORE_NUMBER (ptr, val); \
1555 } while (0)
1557 /* Pop a saved register off the stack. */
1558 #define POP_FAILURE_REG_OR_COUNT() \
1559 do { \
1560 int reg = POP_FAILURE_INT (); \
1561 if (reg == -1) \
1562 { \
1565 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1566 reg = POP_FAILURE_INT (); \
1567 STORE_NUMBER (ptr, reg); \
1569 } \
1570 else \
1571 { \
1572 regend[reg] = POP_FAILURE_POINTER (); \
1573 regstart[reg] = POP_FAILURE_POINTER (); \
1575 reg, regstart[reg], regend[reg]); \
1576 } \
1577 } while (0)
1579 /* Check that we are not stuck in an infinite loop. */
1580 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1581 do { \
1582 int failure = TOP_FAILURE_HANDLE (); \
1584 while (failure > 0 \
1585 && (FAILURE_STR (failure) == string_place \
1586 || FAILURE_STR (failure) == NULL)) \
1587 { \
1588 assert (FAILURE_PAT (failure) >= bufp->buffer \
1589 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1590 if (FAILURE_PAT (failure) == pat_cur) \
1591 { \
1592 cycle = 1; \
1593 break; \
1594 } \
1596 failure = NEXT_FAILURE_HANDLE(failure); \
1597 } \
1599 } while (0)
1601 /* Push the information about the state we will need
1602 if we ever fail back to it.
1604 Requires variables fail_stack, regstart, regend and
1605 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1606 declared.
1608 Does `return FAILURE_CODE' if runs out of memory. */
1610 #define PUSH_FAILURE_POINT(pattern, string_place) \
1611 do { \
1612 char *destination; \
1613 /* Must be int, so when we don't save any registers, the arithmetic \
1615 \
1616 DEBUG_STATEMENT (nfailure_points_pushed++); \
1620 \
1621 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1622 \
1624 \
1626 PUSH_FAILURE_INT (fail_stack.frame); \
1627 \
1629 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1631 PUSH_FAILURE_POINTER (string_place); \
1632 \
1634 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1635 PUSH_FAILURE_POINTER (pattern); \
1636 \
1638 fail_stack.frame = fail_stack.avail; \
1639 } while (0)
1641 /* Estimate the size of data pushed by a typical failure stack entry.
1642 An estimate is all we need, because all we use this for
1643 is to choose a limit for how big to make the failure stack. */
1644 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1645 #define TYPICAL_FAILURE_SIZE 20
1647 /* How many items can still be added to the stack without overflowing it. */
1648 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1651 /* Pops what PUSH_FAIL_STACK pushes.
1653 We restore into the parameters, all of which should be lvalues:
1654 STR -- the saved data position.
1655 PAT -- the saved pattern position.
1656 REGSTART, REGEND -- arrays of string positions.
1658 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1659 `pend', `string1', `size1', `string2', and `size2'. */
1661 #define POP_FAILURE_POINT(str, pat) \
1662 do { \
1663 assert (!FAIL_STACK_EMPTY ()); \
1664 \
1669 \
1671 while (fail_stack.frame < fail_stack.avail) \
1672 POP_FAILURE_REG_OR_COUNT (); \
1673 \
1674 pat = POP_FAILURE_POINTER (); \
1676 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1677 \
1678 /* If the saved string location is NULL, it came from an \
1679 on_failure_keep_string_jump opcode, and we want to throw away the \
1681 str = POP_FAILURE_POINTER (); \
1683 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1685 \
1686 fail_stack.frame = POP_FAILURE_INT (); \
1688 \
1689 assert (fail_stack.avail >= 0); \
1690 assert (fail_stack.frame <= fail_stack.avail); \
1691 \
1692 DEBUG_STATEMENT (nfailure_points_popped++); \
1696 \f
1697 /* Registers are set to a sentinel when they haven't yet matched. */
1698 #define REG_UNSET(e) ((e) == NULL)
1699 \f
1700 /* Subroutine declarations and macros for regex_compile. */
1703 reg_syntax_t syntax,
1714 reg_syntax_t syntax));
1717 reg_syntax_t syntax));
1722 /* Fetch the next character in the uncompiled pattern, with no
1723 translation. */
1724 #define PATFETCH(c) \
1725 do { \
1726 int len; \
1727 if (p == pend) return REG_EEND; \
1728 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1729 p += len; \
1730 } while (0)
1733 /* If `translate' is non-null, return translate[D], else just D. We
1734 cast the subscript to translate because some data is declared as
1735 `char *', to avoid warnings when a string constant is passed. But
1736 when we use a character as a subscript we must make it unsigned. */
1737 #ifndef TRANSLATE
1738 # define TRANSLATE(d) \
1739 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1740 #endif
1743 /* Macros for outputting the compiled pattern into `buffer'. */
1745 /* If the buffer isn't allocated when it comes in, use this. */
1746 #define INIT_BUF_SIZE 32
1748 /* Make sure we have at least N more bytes of space in buffer. */
1749 #define GET_BUFFER_SPACE(n) \
1750 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1751 EXTEND_BUFFER ()
1753 /* Make sure we have one more byte of buffer space and then add C to it. */
1754 #define BUF_PUSH(c) \
1755 do { \
1756 GET_BUFFER_SPACE (1); \
1757 *b++ = (unsigned char) (c); \
1758 } while (0)
1761 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1762 #define BUF_PUSH_2(c1, c2) \
1763 do { \
1764 GET_BUFFER_SPACE (2); \
1765 *b++ = (unsigned char) (c1); \
1766 *b++ = (unsigned char) (c2); \
1767 } while (0)
1770 /* As with BUF_PUSH_2, except for three bytes. */
1771 #define BUF_PUSH_3(c1, c2, c3) \
1772 do { \
1773 GET_BUFFER_SPACE (3); \
1774 *b++ = (unsigned char) (c1); \
1775 *b++ = (unsigned char) (c2); \
1776 *b++ = (unsigned char) (c3); \
1777 } while (0)
1780 /* Store a jump with opcode OP at LOC to location TO. We store a
1781 relative address offset by the three bytes the jump itself occupies. */
1782 #define STORE_JUMP(op, loc, to) \
1783 store_op1 (op, loc, (to) - (loc) - 3)
1785 /* Likewise, for a two-argument jump. */
1786 #define STORE_JUMP2(op, loc, to, arg) \
1787 store_op2 (op, loc, (to) - (loc) - 3, arg)
1789 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1790 #define INSERT_JUMP(op, loc, to) \
1791 insert_op1 (op, loc, (to) - (loc) - 3, b)
1793 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1794 #define INSERT_JUMP2(op, loc, to, arg) \
1795 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1798 /* This is not an arbitrary limit: the arguments which represent offsets
1799 into the pattern are two bytes long. So if 2^15 bytes turns out to
1800 be too small, many things would have to change. */
1801 # define MAX_BUF_SIZE (1L << 15)
1804 /* Any other compiler which, like MSC, has allocation limit below 2^16
1805 bytes will have to use approach similar to what was done below for
1806 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1807 reallocating to 0 bytes. Such thing is not going to work too well.
1808 You have been warned!! */
1809 #if defined _MSC_VER && !defined WIN32
1810 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1811 # define MAX_BUF_SIZE 65500L
1812 #else
1813 # define MAX_BUF_SIZE (1L << 16)
1814 #endif
1817 /* Extend the buffer by twice its current size via realloc and
1818 reset the pointers that pointed into the old block to point to the
1819 correct places in the new one. If extending the buffer results in it
1820 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1821 #if __BOUNDED_POINTERS__
1822 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1823 # define MOVE_BUFFER_POINTER(P) \
1824 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1825 SET_HIGH_BOUND (P), \
1826 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1827 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1828 else \
1829 { \
1830 SET_HIGH_BOUND (b); \
1831 SET_HIGH_BOUND (begalt); \
1832 if (fixup_alt_jump) \
1833 SET_HIGH_BOUND (fixup_alt_jump); \
1834 if (laststart) \
1835 SET_HIGH_BOUND (laststart); \
1836 if (pending_exact) \
1837 SET_HIGH_BOUND (pending_exact); \
1838 }
1839 #else
1840 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1841 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1842 #endif
1843 #define EXTEND_BUFFER() \
1844 do { \
1845 unsigned char *old_buffer = bufp->buffer; \
1846 if (bufp->allocated == MAX_BUF_SIZE) \
1847 return REG_ESIZE; \
1848 bufp->allocated <<= 1; \
1849 if (bufp->allocated > MAX_BUF_SIZE) \
1850 bufp->allocated = MAX_BUF_SIZE; \
1851 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1852 if (bufp->buffer == NULL) \
1853 return REG_ESPACE; \
1855 if (old_buffer != bufp->buffer) \
1856 { \
1857 unsigned char *new_buffer = bufp->buffer; \
1858 MOVE_BUFFER_POINTER (b); \
1859 MOVE_BUFFER_POINTER (begalt); \
1860 if (fixup_alt_jump) \
1861 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1862 if (laststart) \
1863 MOVE_BUFFER_POINTER (laststart); \
1864 if (pending_exact) \
1865 MOVE_BUFFER_POINTER (pending_exact); \
1866 } \
1867 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1868 } while (0)
1871 /* Since we have one byte reserved for the register number argument to
1872 {start,stop}_memory, the maximum number of groups we can report
1873 things about is what fits in that byte. */
1874 #define MAX_REGNUM 255
1876 /* But patterns can have more than `MAX_REGNUM' registers. We just
1877 ignore the excess. */
1881 /* Macros for the compile stack. */
1883 /* Since offsets can go either forwards or backwards, this type needs to
1884 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1885 /* int may be not enough when sizeof(int) == 2. */
1889 {
1890 pattern_offset_t begalt_offset;
1891 pattern_offset_t fixup_alt_jump;
1892 pattern_offset_t laststart_offset;
1893 regnum_t regnum;
1898 {
1905 #define INIT_COMPILE_STACK_SIZE 32
1907 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1908 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1910 /* The next available element. */
1911 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1913 /* Explicit quit checking is only used on NTemacs and whenever we
1914 use polling to process input events. */
1915 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1917 # define IMMEDIATE_QUIT_CHECK \
1918 do { \
1919 if (immediate_quit) QUIT; \
1920 } while (0)
1921 #else
1922 # define IMMEDIATE_QUIT_CHECK ((void)0)
1923 #endif
1924 \f
1925 /* Structure to manage work area for range table. */
1926 struct range_table_work_area
1927 {
1932 };
1934 /* Make sure that WORK_AREA can hold more N multibyte characters.
1935 This is used only in set_image_of_range and set_image_of_range_1.
1936 It expects WORK_AREA to be a pointer.
1937 If it can't get the space, it returns from the surrounding function. */
1939 #define EXTEND_RANGE_TABLE(work_area, n) \
1940 do { \
1941 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1942 { \
1943 extend_range_table_work_area (&work_area); \
1944 if ((work_area).table == 0) \
1945 return (REG_ESPACE); \
1946 } \
1947 } while (0)
1949 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1950 (work_area).bits |= (bit)
1952 /* Bits used to implement the multibyte-part of the various character classes
1953 such as [:alnum:] in a charset's range table. */
1954 #define BIT_WORD 0x1
1955 #define BIT_LOWER 0x2
1956 #define BIT_PUNCT 0x4
1957 #define BIT_SPACE 0x8
1958 #define BIT_UPPER 0x10
1959 #define BIT_MULTIBYTE 0x20
1961 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1962 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1963 do { \
1964 EXTEND_RANGE_TABLE ((work_area), 2); \
1965 (work_area).table[(work_area).used++] = (range_start); \
1966 (work_area).table[(work_area).used++] = (range_end); \
1967 } while (0)
1969 /* Free allocated memory for WORK_AREA. */
1970 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1971 do { \
1972 if ((work_area).table) \
1973 free ((work_area).table); \
1974 } while (0)
1976 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1977 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1978 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1979 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1980 \f
1982 /* Set the bit for character C in a list. */
1983 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1986 #ifdef emacs
1988 /* Store characters in the range FROM to TO in the bitmap at B (for
1989 ASCII and unibyte characters) and WORK_AREA (for multibyte
1990 characters) while translating them and paying attention to the
1991 continuity of translated characters.
1993 Implementation note: It is better to implement these fairly big
1994 macros by a function, but it's not that easy because macros called
1995 in this macro assume various local variables already declared. */
1997 /* Both FROM and TO are ASCII characters. */
1999 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2000 do { \
2001 int C0, C1; \
2002 \
2003 for (C0 = (FROM); C0 <= (TO); C0++) \
2004 { \
2005 C1 = TRANSLATE (C0); \
2006 if (! ASCII_CHAR_P (C1)) \
2007 { \
2008 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2009 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2010 C1 = C0; \
2011 } \
2012 SET_LIST_BIT (C1); \
2013 } \
2014 } while (0)
2017 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2019 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2020 do { \
2021 int C0, C1, C2, I; \
2022 int USED = RANGE_TABLE_WORK_USED (work_area); \
2023 \
2024 for (C0 = (FROM); C0 <= (TO); C0++) \
2025 { \
2026 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2027 if (CHAR_BYTE8_P (C1)) \
2028 SET_LIST_BIT (C0); \
2029 else \
2030 { \
2031 C2 = TRANSLATE (C1); \
2032 if (C2 == C1 \
2033 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2034 C1 = C0; \
2035 SET_LIST_BIT (C1); \
2036 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2037 { \
2038 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2039 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2040 \
2041 if (C2 >= from - 1 && C2 <= to + 1) \
2042 { \
2043 if (C2 == from - 1) \
2044 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2045 else if (C2 == to + 1) \
2046 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2047 break; \
2048 } \
2049 } \
2050 if (I < USED) \
2051 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2052 } \
2053 } \
2054 } while (0)
2057 /* Both FROM and TO are multibyte characters. */
2059 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2060 do { \
2061 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2062 \
2063 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2064 for (C0 = (FROM); C0 <= (TO); C0++) \
2065 { \
2066 C1 = TRANSLATE (C0); \
2067 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2068 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2069 SET_LIST_BIT (C2); \
2070 if (C1 >= (FROM) && C1 <= (TO)) \
2071 continue; \
2072 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2073 { \
2074 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2075 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2076 \
2077 if (C1 >= from - 1 && C1 <= to + 1) \
2078 { \
2079 if (C1 == from - 1) \
2080 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2081 else if (C1 == to + 1) \
2082 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2083 break; \
2084 } \
2085 } \
2086 if (I < USED) \
2087 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2088 } \
2089 } while (0)
2093 /* Get the next unsigned number in the uncompiled pattern. */
2094 #define GET_UNSIGNED_NUMBER(num) \
2095 do { \
2096 if (p == pend) \
2097 FREE_STACK_RETURN (REG_EBRACE); \
2098 else \
2099 { \
2100 PATFETCH (c); \
2102 { \
2103 int prev; \
2104 if (num < 0) \
2105 num = 0; \
2106 prev = num; \
2108 if (num / 10 != prev) \
2109 FREE_STACK_RETURN (REG_BADBR); \
2110 if (p == pend) \
2111 FREE_STACK_RETURN (REG_EBRACE); \
2112 PATFETCH (c); \
2113 } \
2114 } \
2115 } while (0)
2116 \f
2117 #if ! WIDE_CHAR_SUPPORT
2119 /* Map a string to the char class it names (if any). */
2120 re_wctype_t
2122 {
2142 }
2144 /* True if CH is in the char class CC. */
2145 boolean
2147 {
2149 {
2170 }
2171 }
2173 /* Return a bit-pattern to use in the range-table bits to match multibyte
2174 chars of class CC. */
2175 static int
2177 {
2179 {
2191 }
2192 }
2193 #endif
2194 \f
2195 /* Filling in the work area of a range. */
2197 /* Actually extend the space in WORK_AREA. */
2199 static void
2201 {
2204 work_area->table
2206 else
2207 work_area->table
2209 }
2211 #if 0
2212 #ifdef emacs
2214 /* Carefully find the ranges of codes that are equivalent
2215 under case conversion to the range start..end when passed through
2216 TRANSLATE. Handle the case where non-letters can come in between
2217 two upper-case letters (which happens in Latin-1).
2218 Also handle the case of groups of more than 2 case-equivalent chars.
2220 The basic method is to look at consecutive characters and see
2221 if they can form a run that can be handled as one.
2223 Returns -1 if successful, REG_ESPACE if ran out of space. */
2225 static int
2227 RE_TRANSLATE_TYPE translate;
2230 {
2231 /* `one_case' indicates a character, or a run of characters,
2232 each of which is an isolate (no case-equivalents).
2233 This includes all ASCII non-letters.
2235 `two_case' indicates a character, or a run of characters,
2236 each of which has two case-equivalent forms.
2237 This includes all ASCII letters.
2239 `strange' indicates a character that has more than one
2240 case-equivalent. */
2244 /* Describe the run that is in progress,
2245 which the next character can try to extend.
2246 If run_type is strange, that means there really is no run.
2247 If run_type is one_case, then run_start...run_end is the run.
2248 If run_type is two_case, then the run is run_start...run_end,
2249 and the case-equivalents end at run_eqv_end. */
2254 Lisp_Object eqv_table;
2257 {
2262 }
2267 {
2272 /* Classify this character */
2277 else
2282 else
2285 /* Can this character extend the run in progress? */
2290 {
2291 /* No, end the run.
2292 Record each of its equivalent ranges. */
2294 {
2298 }
2300 {
2308 }
2310 }
2313 {
2314 /* For a strange character, add each of its equivalents, one
2315 by one. Don't start a range. */
2316 do
2317 {
2322 }
2324 }
2326 /* Add this char to the run, or start a new run. */
2328 {
2329 /* Initialize a new range. */
2334 }
2335 else
2336 {
2337 /* Extend a running range. */
2340 }
2341 }
2343 /* If a run is still in progress at the end, finish it now
2344 by recording its equivalent ranges. */
2346 {
2350 }
2352 {
2360 }
2363 }
2367 /* Record the image of the range start..end when passed through
2368 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2369 and is not even necessarily contiguous.
2370 Normally we approximate it with the smallest contiguous range that contains
2371 all the chars we need. However, for Latin-1 we go to extra effort
2372 to do a better job.
2374 This function is not called for ASCII ranges.
2376 Returns -1 if successful, REG_ESPACE if ran out of space. */
2378 static int
2380 RE_TRANSLATE_TYPE translate;
2383 {
2386 #ifdef emacs
2387 /* For Latin-1 ranges, use set_image_of_range_1
2388 to get proper handling of ranges that include letters and nonletters.
2389 For a range that includes the whole of Latin-1, this is not necessary.
2390 For other character sets, we don't bother to get this right. */
2393 {
2406 }
2407 #endif
2416 {
2420 {
2423 {
2426 else
2427 {
2430 }
2431 }
2432 }
2435 {
2439 }
2440 }
2443 }
2445 \f
2446 #ifndef MATCH_MAY_ALLOCATE
2448 /* If we cannot allocate large objects within re_match_2_internal,
2449 we make the fail stack and register vectors global.
2450 The fail stack, we grow to the maximum size when a regexp
2451 is compiled.
2452 The register vectors, we adjust in size each time we
2453 compile a regexp, according to the number of registers it needs. */
2457 /* Size with which the following vectors are currently allocated.
2458 That is so we can make them bigger as needed,
2459 but never make them smaller. */
2465 /* Make the register vectors big enough for NUM_REGS registers,
2466 but don't make them smaller. */
2468 static
2471 {
2473 {
2480 }
2481 }
2484 \f
2486 compile_stack,
2487 regnum_t regnum));
2489 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2490 Returns one of error codes defined in `regex.h', or zero for success.
2492 Assumes the `allocated' (and perhaps `buffer') and `translate'
2493 fields are set in BUFP on entry.
2495 If it succeeds, results are put in BUFP (if it returns an error, the
2496 contents of BUFP are undefined):
2497 `buffer' is the compiled pattern;
2498 `syntax' is set to SYNTAX;
2499 `used' is set to the length of the compiled pattern;
2500 `fastmap_accurate' is zero;
2501 `re_nsub' is the number of subexpressions in PATTERN;
2502 `not_bol' and `not_eol' are zero;
2504 The `fastmap' field is neither examined nor set. */
2506 /* Insert the `jump' from the end of last alternative to "here".
2507 The space for the jump has already been allocated. */
2508 #define FIXUP_ALT_JUMP() \
2509 do { \
2510 if (fixup_alt_jump) \
2511 STORE_JUMP (jump, fixup_alt_jump, b); \
2512 } while (0)
2515 /* Return, freeing storage we allocated. */
2516 #define FREE_STACK_RETURN(value) \
2517 do { \
2518 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2519 free (compile_stack.stack); \
2520 return value; \
2521 } while (0)
2523 static reg_errcode_t
2524 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2525 {
2526 /* We fetch characters from PATTERN here. */
2529 /* A random temporary spot in PATTERN. */
2532 /* Points to the end of the buffer, where we should append. */
2535 /* Keeps track of unclosed groups. */
2536 compile_stack_type compile_stack;
2538 /* Points to the current (ending) position in the pattern. */
2539 #ifdef AIX
2540 /* `const' makes AIX compiler fail. */
2542 #else
2544 #endif
2547 /* How to translate the characters in the pattern. */
2550 /* Address of the count-byte of the most recently inserted `exactn'
2551 command. This makes it possible to tell if a new exact-match
2552 character can be added to that command or if the character requires
2553 a new `exactn' command. */
2556 /* Address of start of the most recently finished expression.
2557 This tells, e.g., postfix * where to find the start of its
2558 operand. Reset at the beginning of groups and alternatives. */
2561 /* Address of beginning of regexp, or inside of last group. */
2564 /* Place in the uncompiled pattern (i.e., the {) to
2565 which to go back if the interval is invalid. */
2568 /* Address of the place where a forward jump should go to the end of
2569 the containing expression. Each alternative of an `or' -- except the
2570 last -- ends with a forward jump of this sort. */
2573 /* Work area for range table of charset. */
2576 /* If the object matched can contain multibyte characters. */
2579 /* If a target of matching can contain multibyte characters. */
2582 /* Nonzero if we have pushed down into a subpattern. */
2585 /* These hold the values of p, pattern, and pend from the main
2586 pattern when we have pushed into a subpattern. */
2591 #ifdef DEBUG
2592 debug++;
2595 {
2601 }
2604 /* Initialize the compile stack. */
2615 /* Initialize the pattern buffer. */
2621 /* Set `used' to zero, so that if we return an error, the pattern
2622 printer (for debugging) will think there's no pattern. We reset it
2623 at the end. */
2626 /* Always count groups, whether or not bufp->no_sub is set. */
2629 #if !defined emacs && !defined SYNTAX_TABLE
2630 /* Initialize the syntax table. */
2632 #endif
2635 {
2638 enough space. This loses if buffer's address is bogus, but
2639 that is the user's responsibility. */
2641 }
2642 else
2645 }
2649 }
2653 /* Loop through the uncompiled pattern until we're at the end. */
2655 {
2657 {
2658 /* If this is the end of an included regexp,
2659 pop back to the main regexp and try again. */
2661 {
2667 }
2668 /* If this is the end of the main regexp, we are done. */
2670 }
2675 {
2677 {
2680 /* If there's no special whitespace regexp, treat
2681 spaces normally. And don't try to do this recursively. */
2685 /* Peek past following spaces. */
2687 {
2690 p1++;
2691 }
2692 /* If the spaces are followed by a repetition op,
2693 treat them normally. */
2699 /* Replace the spaces with the whitespace regexp. */
2707 }
2710 {
2713 /* If context independent, it's an operator. */
2715 /* Otherwise, depends on what's come before. */
2718 else
2720 }
2725 {
2727 p == pend
2728 /* If context independent, it's an operator. */
2730 /* Otherwise, depends on what's next. */
2733 else
2735 }
2744 handle_plus:
2746 /* If there is no previous pattern... */
2748 {
2753 }
2755 {
2756 /* 1 means zero (many) matches is allowed. */
2760 /* If there is a sequence of repetition chars, collapse it
2761 down to just one (the right one). We can't combine
2762 interval operators with these because of, e.g., `a{2}*',
2763 which should only match an even number of `a's. */
2766 {
2770 else
2771 {
2774 }
2781 ;
2783 {
2788 else
2790 }
2791 else
2793 /* If we get here, we found another repeat character. */
2795 }
2797 /* Star, etc. applied to an empty pattern is equivalent
2798 to an empty pattern. */
2802 /* Now we know whether or not zero matches is allowed
2803 and also whether or not two or more matches is allowed. */
2805 {
2807 {
2810 re_opcode_t ofj =
2811 /* Check if the loop can match the empty string. */
2818 on_failure_keep_string_jump, we turn simple P+
2819 into PP* if P is simple. */
2827 }
2831 /* A + loop. */
2833 else
2834 /* Simple * loops can use on_failure_keep_string_jump
2835 depending on what follows. But since we don't know
2836 that yet, we leave the decision up to
2837 on_failure_jump_smart. */
2843 }
2844 else
2845 {
2846 /* A simple ? pattern. */
2851 }
2852 }
2858 {
2861 /* The non-greedy multiple match looks like
2862 a repeat..until: we only need a conditional jump
2863 at the end of the loop. */
2869 {
2870 /* The repeat...until naturally matches one or more.
2871 To also match zero times, we need to first jump to
2872 the end of the loop (its conditional jump). */
2875 }
2876 }
2877 else
2878 {
2879 /* non-greedy a?? */
2884 }
2885 }
2886 }
2898 {
2903 /* Ensure that we have enough space to push a charset: the
2904 opcode, the length count, and the bitset; 34 bytes in all. */
2909 /* We test `*p == '^' twice, instead of using an if
2910 statement, so we only need one BUF_PUSH. */
2913 p++;
2915 /* Remember the first position in the bracket expression. */
2918 /* Push the number of bytes in the bitmap. */
2921 /* Clear the whole map. */
2924 /* charset_not matches newline according to a syntax bit. */
2929 /* Read in characters and ranges, setting map bits. */
2931 {
2938 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2939 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2940 So the translation is done later in a loop. Example:
2941 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2944 /* \ might escape characters inside [...] and [^...]. */
2946 {
2951 }
2952 else
2953 {
2954 /* Could be the end of the bracket expression. If it's
2955 not (i.e., when the bracket expression is `[]' so
2956 far), the ']' character bit gets set way below. */
2959 }
2961 /* See if we're at the beginning of a possible character
2962 class. */
2966 {
2967 /* Leave room for the null. */
2975 /* If pattern is `[[:'. */
2979 {
2985 else
2986 /* This is in any case an invalid class name. */
2988 }
2991 /* If isn't a word bracketed by `[:' and `:]':
2992 undo the ending character, the letters, and
2993 leave the leading `:' and `[' (but set bits for
2994 them). */
2996 {
2997 re_wctype_t cc;
3005 /* Throw away the ] at the end of the character
3006 class. */
3011 #ifndef emacs
3014 {
3018 }
3020 /* Most character classes in a multibyte match
3021 just set a flag. Exceptions are is_blank,
3022 is_digit, is_cntrl, and is_xdigit, since
3023 they can only match ASCII characters. We
3024 don't need to handle them for multibyte.
3025 They are distinguished by a negative wctype. */
3027 /* Setup the gl_state object to its buffer-defined
3028 value. This hardcodes the buffer-global
3029 syntax-table for ASCII chars, while the other chars
3030 will obey syntax-table properties. It's not ideal,
3031 but it's the way it's been done until now. */
3035 {
3039 {
3048 }
3049 }
3050 SET_RANGE_TABLE_WORK_AREA_BIT
3053 /* In most cases the matching rule for char classes
3054 only uses the syntax table for multibyte chars,
3055 so that the content of the syntax-table it is not
3056 hardcoded in the range_table. SPACE and WORD are
3057 the two exceptions. */
3061 /* Repeat the loop. */
3063 }
3064 else
3065 {
3066 /* Go back to right after the "[:". */
3070 /* Because the `:' may starts the range, we
3071 can't simply set bit and repeat the loop.
3072 Instead, just set it to C and handle below. */
3074 }
3075 }
3078 {
3080 /* Discard the `-'. */
3083 /* Fetch the character which ends the range. */
3085 #ifdef emacs
3088 /* Treat the range from a multibyte character to
3089 raw-byte character as empty. */
3092 }
3093 else
3094 /* Range from C to C. */
3098 {
3101 /* Else, repeat the loop. */
3102 }
3103 else
3104 {
3105 #ifndef emacs
3106 /* Set the range into bitmap */
3108 {
3112 }
3115 {
3121 }
3123 {
3125 {
3130 }
3132 {
3134 }
3135 else
3136 {
3138 }
3139 }
3141 }
3142 }
3144 /* Discard any (non)matching list bytes that are all 0 at the
3145 end of the map. Decrease the map-length byte too. */
3150 /* Build real range table from work area. */
3153 {
3157 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3158 bytes for flags, two for COUNT, and three bytes for
3159 each character. */
3162 /* Indicate the existence of range table. */
3165 /* Store the character class flag bits into the range table.
3166 If not in emacs, these flag bits are always 0. */
3172 STORE_CHARACTER_AND_INCR
3174 }
3175 }
3182 else
3189 else
3196 else
3203 else
3210 else
3217 /* Do not translate the character after the \, so that we can
3218 distinguish, e.g., \B from \b, even if we normally would
3219 translate, e.g., B to b. */
3223 {
3228 handle_open:
3229 {
3233 {
3234 /* Look for a special (?...) construct */
3236 {
3239 {
3242 {
3245 /* An explicitly specified regnum must start
3246 with non-0. */
3253 /* Only (?:...) is supported right now. */
3255 }
3256 }
3257 }
3258 }
3268 /* Ideally, we'd want to check that the specified
3269 group can't have matched (i.e. all subgroups
3270 using the same regnum are in other branches of
3271 OR patterns), but we don't currently keep track
3272 of enough info to do that easily. */
3275 }
3276 else
3277 /* It's really shy. */
3281 {
3283 compile_stack_elt_t);
3287 }
3289 /* These are the values to restore when we hit end of this
3290 group. They are all relative offsets, so that if the
3291 whole pattern moves because of realloc, they will still
3292 be valid. */
3294 COMPILE_STACK_TOP.fixup_alt_jump
3299 /* Do not push a start_memory for groups beyond the last one
3300 we can represent in the compiled pattern. */
3309 /* If we've reached MAX_REGNUM groups, then this open
3310 won't actually generate any code, so we'll have to
3311 clear pending_exact explicitly. */
3314 }
3320 {
3323 else
3325 }
3327 handle_close:
3330 /* See similar code for backslashed left paren above. */
3332 {
3335 else
3337 }
3339 /* Since we just checked for an empty stack above, this
3340 ``can't happen''. */
3342 {
3343 /* We don't just want to restore into `regnum', because
3344 later groups should continue to be numbered higher,
3345 as in `(ab)c(de)' -- the second group is #2. */
3346 regnum_t regnum;
3350 fixup_alt_jump
3356 /* If we've reached MAX_REGNUM groups, then this open
3357 won't actually generate any code, so we'll have to
3358 clear pending_exact explicitly. */
3361 /* We're at the end of the group, so now we know how many
3362 groups were inside this one. */
3365 }
3372 handle_alt:
3376 /* Insert before the previous alternative a jump which
3377 jumps to this alternative if the former fails. */
3383 /* The alternative before this one has a jump after it
3384 which gets executed if it gets matched. Adjust that
3385 jump so it will jump to this alternative's analogous
3386 jump (put in below, which in turn will jump to the next
3387 (if any) alternative's such jump, etc.). The last such
3388 jump jumps to the correct final destination. A picture:
3389 _____ _____
3390 | | | |
3391 | v | v
3392 a | b | c
3394 If we are at `b', then fixup_alt_jump right now points to a
3395 three-byte space after `a'. We'll put in the jump, set
3396 fixup_alt_jump to right after `b', and leave behind three
3397 bytes which we'll fill in when we get to after `c'. */
3401 /* Mark and leave space for a jump after this alternative,
3402 to be filled in later either by next alternative or
3403 when know we're at the end of a series of alternatives. */
3414 /* If \{ is a literal. */
3416 /* If we're at `\{' and it's not the open-interval
3417 operator. */
3421 handle_interval:
3422 {
3423 /* If got here, then the syntax allows intervals. */
3425 /* At least (most) this many matches must be made. */
3434 else
3435 /* Interval such as `{1}' => match exactly once. */
3443 {
3449 }
3454 /* We just parsed a valid interval. */
3456 /* If it's invalid to have no preceding re. */
3458 {
3463 else
3465 }
3468 /* If the upper bound is zero, just drop the sub pattern
3469 altogether. */
3472 /* Just match it once: nothing to do here. */
3473 ;
3475 /* Otherwise, we have a nontrivial interval. When
3476 we're all done, the pattern will look like:
3477 set_number_at <jump count> <upper bound>
3478 set_number_at <succeed_n count> <lower bound>
3479 succeed_n <after jump addr> <succeed_n count>
3480 <body of loop>
3481 jump_n <succeed_n addr> <jump count>
3482 (The upper bound and `jump_n' are omitted if
3483 `upper_bound' is 1, though.) */
3484 else
3486 more at the end of the loop. */
3494 {
3495 /* A succeed_n that starts with 0 is really a
3496 a simple on_failure_jump_loop. */
3500 }
3501 else
3502 {
3503 /* Initialize lower bound of the `succeed_n', even
3504 though it will be set during matching by its
3505 attendant `set_number_at' (inserted next),
3506 because `re_compile_fastmap' needs to know.
3507 Jump to the `jump_n' we might insert below. */
3510 lower_bound);
3513 /* Code to initialize the lower bound. Insert
3514 before the `succeed_n'. The `5' is the last two
3515 bytes of this `set_number_at', plus 3 bytes of
3516 the following `succeed_n'. */
3520 }
3523 {
3524 /* A negative upper bound stands for infinity,
3525 in which case it degenerates to a plain jump. */
3528 }
3531 append a backward jump to the `succeed_n'
3532 that starts this interval.
3534 When we've reached this during matching,
3535 we'll have matched the interval once, so
3536 jump back only `upper_bound - 1' times. */
3541 /* The location we want to set is the second
3542 parameter of the `jump_n'; that is `b-2' as
3543 an absolute address. `laststart' will be
3544 the `set_number_at' we're about to insert;
3545 `laststart+3' the number to set, the source
3546 for the relative address. But we are
3547 inserting into the middle of the pattern --
3548 so everything is getting moved up by 5.
3549 Conclusion: (b - 2) - (laststart + 3) + 5,
3550 i.e., b - laststart.
3552 We insert this at the beginning of the loop
3553 so that if we fail during matching, we'll
3554 reinitialize the bounds. */
3558 }
3559 }
3562 }
3565 unfetch_interval:
3566 /* If an invalid interval, match the characters as literals. */
3571 /* normal_char and normal_backslash need `c'. */
3575 {
3578 }
3579 else
3582 #ifdef emacs
3583 /* There is no way to specify the before_dot and after_dot
3584 operators. rms says this is ok. --karl */
3652 else
3682 {
3683 regnum_t reg;
3691 /* Can't back reference to a subexp before its end. */
3697 }
3705 else
3709 normal_backslash:
3710 /* You might think it would be useful for \ to mean
3711 not to translate; but if we don't translate it
3712 it will never match anything. */
3714 }
3719 /* Expects the character in `c'. */
3720 normal_char:
3721 /* If no exactn currently being built. */
3724 /* If last exactn not at current position. */
3727 /* We have only one byte following the exactn for the count. */
3730 /* If followed by a repetition operator. */
3739 {
3740 /* Start building a new exactn. */
3746 }
3749 {
3753 {
3757 }
3758 else
3759 {
3762 {
3767 }
3770 }
3772 }
3779 /* Through the pattern now. */
3786 /* If we don't want backtracking, force success
3787 the first time we reach the end of the compiled pattern. */
3791 /* We have succeeded; set the length of the buffer. */
3794 #ifdef DEBUG
3796 {
3800 }
3801 debug--;
3804 #ifndef MATCH_MAY_ALLOCATE
3805 /* Initialize the failure stack to the largest possible stack. This
3806 isn't necessary unless we're trying to avoid calling alloca in
3807 the search and match routines. */
3808 {
3812 {
3816 fail_stack.stack
3819 else
3820 fail_stack.stack
3824 }
3827 }
3832 \f
3833 /* Subroutines for `regex_compile'. */
3835 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3837 static void
3839 {
3842 }
3845 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3847 static void
3849 {
3853 }
3856 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3857 for OP followed by two-byte integer parameter ARG. */
3859 static void
3861 {
3869 }
3872 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3874 static void
3876 {
3884 }
3887 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3888 after an alternative or a begin-subexpression. We assume there is at
3889 least one character before the ^. */
3891 static boolean
3893 {
3897 return
3898 /* After a subexpression? */
3900 /* After an alternative? */
3902 /* After a shy subexpression? */
3907 }
3910 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3911 at least one character after the $, i.e., `P < PEND'. */
3913 static boolean
3915 {
3920 return
3921 /* Before a subexpression? */
3924 /* Before an alternative? */
3927 }
3930 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3931 false if it's not. */
3933 static boolean
3935 {
3940 this_element--)
3945 }
3946 \f
3947 /* analyse_first.
3948 If fastmap is non-NULL, go through the pattern and fill fastmap
3949 with all the possible leading chars. If fastmap is NULL, don't
3950 bother filling it up (obviously) and only return whether the
3951 pattern could potentially match the empty string.
3953 Return 1 if p..pend might match the empty string.
3954 Return 0 if p..pend matches at least one char.
3955 Return -1 if fastmap was not updated accurately. */
3957 static int
3959 {
3963 /* If all elements for base leading-codes in fastmap is set, this
3964 flag is set true. */
3969 /* The loop below works as follows:
3970 - It has a working-list kept in the PATTERN_STACK and which basically
3971 starts by only containing a pointer to the first operation.
3972 - If the opcode we're looking at is a match against some set of
3973 chars, then we add those chars to the fastmap and go on to the
3974 next work element from the worklist (done via `break').
3975 - If the opcode is a control operator on the other hand, we either
3976 ignore it (if it's meaningless at this point, such as `start_memory')
3977 or execute it (if it's a jump). If the jump has several destinations
3978 (i.e. `on_failure_jump'), then we push the other destination onto the
3979 worklist.
3980 We guarantee termination by ignoring backward jumps (more or less),
3981 so that `p' is monotonically increasing. More to the point, we
3982 never set `p' (or push) anything `<= p1'. */
3985 {
3986 /* `p1' is used as a marker of how far back a `on_failure_jump'
3987 can go without being ignored. It is normally equal to `p'
3988 (which prevents any backward `on_failure_jump') except right
3989 after a plain `jump', to allow patterns such as:
3990 0: jump 10
3991 3..9: <body>
3992 10: on_failure_jump 3
3993 as used for the *? operator. */
3997 {
4003 /* If the first character has to match a backreference, that means
4004 that the group was empty (since it already matched). Since this
4005 is the only case that interests us here, we can assume that the
4006 backreference must match the empty string. */
4007 p++;
4011 /* Following are the cases which match a character. These end
4012 with `break'. */
4016 {
4017 /* If multibyte is nonzero, the first byte of each
4018 character is an ASCII or a leading code. Otherwise,
4019 each byte is a character. Thus, this works in both
4020 cases. */
4023 {
4024 /* For the case of matching this unibyte regex
4025 against multibyte, we must set a leading code of
4026 the corresponding multibyte character. */
4030 }
4031 }
4036 /* We could put all the chars except for \n (and maybe \0)
4037 but we don't bother since it is generally not worth it. */
4044 {
4045 /* Chars beyond end of bitmap are possible matches. */
4049 }
4051 /* Fallthrough */
4060 #ifdef emacs
4062 which doesn't match the specified set of characters. */
4063 not
4064 ||
4065 /* If we can match a character class, we can match any
4066 multibyte characters. */
4070 {
4072 {
4077 }
4078 }
4082 {
4083 /* Set fastmap[I] to 1 where I is a leading code of each
4084 multibyte character in the range table. */
4088 /* Make P points the range table. `+ 2' is to skip flag
4089 bits for a character class. */
4092 /* Extract the number of ranges in range table into COUNT. */
4095 {
4096 /* Extract the start and end of each range. */
4104 }
4105 }
4106 #endif
4112 #ifndef emacs
4120 /* This match depends on text properties. These end with
4121 aborting optimizations. */
4133 /* Any leading code can possibly start a character which
4134 has or doesn't has the specified category. */
4136 {
4141 }
4144 /* All cases after this match the empty string. These end with
4145 `continue'. */
4168 /* Backward jumps can only go back to code that we've already
4169 visited. `re_compile' should make sure this is true. */
4173 {
4179 p++;
4183 };
4184 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4185 to jump back to "just after here". */
4186 /* Fallthrough */
4196 else
4198 We first go down the "straight" path so as to minimize
4199 stack usage when going through alternatives. */
4203 }
4208 /* This code simply does not properly handle forward jump_n. */
4211 /* jump_n can either jump or fall through. The (backward) jump
4212 case has already been handled, so we only need to look at the
4213 fallthrough case. */
4217 /* If N == 0, it should be an on_failure_jump_loop instead. */
4220 /* We only care about one iteration of the loop, so we don't
4221 need to consider the case where this behaves like an
4222 on_failure_jump. */
4241 /* Getting here means we have found the possible starting
4242 characters for one path of the pattern -- and that the empty
4243 string does not match. We need not follow this path further. */
4247 /* We reached the end without matching anything. */
4251 \f
4252 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4253 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4254 characters can start a string that matches the pattern. This fastmap
4255 is used by re_search to skip quickly over impossible starting points.
4257 Character codes above (1 << BYTEWIDTH) are not represented in the
4258 fastmap, but the leading codes are represented. Thus, the fastmap
4259 indicates which character sets could start a match.
4261 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4262 area as BUFP->fastmap.
4264 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4265 the pattern buffer.
4267 Returns 0 if we succeed, -2 if an internal error. */
4269 int
4271 {
4285 \f
4286 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4287 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4288 this memory for recording register information. STARTS and ENDS
4289 must be allocated using the malloc library routine, and must each
4290 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4292 If NUM_REGS == 0, then subsequent matches should allocate their own
4293 register data.
4295 Unless this function is called, the first search or match using
4296 PATTERN_BUFFER will allocate its own register data, without
4297 freeing the old data. */
4299 void
4300 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4301 {
4303 {
4308 }
4309 else
4310 {
4314 }
4315 }
4317 \f
4318 /* Searching routines. */
4320 /* Like re_search_2, below, but only one string is specified, and
4321 doesn't let you say where to stop matching. */
4323 int
4324 re_search (struct re_pattern_buffer *bufp, const char *string, int size, int startpos, int range, struct re_registers *regs)
4325 {
4328 }
4331 /* Head address of virtual concatenation of string. */
4332 #define HEAD_ADDR_VSTRING(P) \
4333 (((P) >= size1 ? string2 : string1))
4335 /* End address of virtual concatenation of string. */
4336 #define STOP_ADDR_VSTRING(P) \
4337 (((P) >= size1 ? string2 + size2 : string1 + size1))
4339 /* Address of POS in the concatenation of virtual string. */
4340 #define POS_ADDR_VSTRING(POS) \
4341 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4343 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4344 virtual concatenation of STRING1 and STRING2, starting first at index
4345 STARTPOS, then at STARTPOS + 1, and so on.
4347 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4349 RANGE is how far to scan while trying to match. RANGE = 0 means try
4350 only at STARTPOS; in general, the last start tried is STARTPOS +
4351 RANGE.
4353 In REGS, return the indices of the virtual concatenation of STRING1
4354 and STRING2 that matched the entire BUFP->buffer and its contained
4355 subexpressions.
4357 Do not consider matching one past the index STOP in the virtual
4358 concatenation of STRING1 and STRING2.
4360 We return either the position in the strings at which the match was
4361 found, -1 if no match, or -2 if error (such as failure
4362 stack overflow). */
4364 int
4365 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, int size1, const char *str2, int size2, int startpos, int range, struct re_registers *regs, int stop)
4366 {
4374 boolean anchored_start;
4375 /* Nonzero if we are searching multibyte string. */
4378 /* Check for out-of-range STARTPOS. */
4382 /* Fix up RANGE if it might eventually take us outside
4383 the virtual concatenation of STRING1 and STRING2.
4384 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4390 /* If the search isn't to be a backwards one, don't waste time in a
4391 search for a pattern anchored at beginning of buffer. */
4393 {
4396 else
4398 }
4400 #ifdef emacs
4401 /* In a forward search for something that starts with \=.
4402 don't keep searching past point. */
4404 {
4408 }
4411 /* Update the fastmap now if not correct already. */
4415 /* See whether the pattern is anchored. */
4418 #ifdef emacs
4420 {
4424 }
4425 #endif
4427 /* Loop through the string, looking for a place to start matching. */
4429 {
4430 /* If the pattern is anchored,
4431 skip quickly past places we cannot match.
4432 We don't bother to treat startpos == 0 specially
4433 because that case doesn't repeat. */
4435 {
4440 }
4442 /* If a fastmap is supplied, skip quickly over characters that
4443 cannot be the start of a match. If the pattern can match the
4444 null string, however, we don't need to skip characters; we want
4445 the first null string. */
4447 {
4454 {
4461 /* Written out as an if-else to avoid testing `translate'
4462 inside the loop. */
4464 {
4467 {
4477 }
4478 else
4480 {
4491 d++;
4492 range--;
4493 }
4494 }
4495 else
4496 {
4499 {
4507 }
4508 else
4510 {
4511 d++;
4512 range--;
4513 }
4514 }
4516 }
4518 {
4520 {
4525 }
4526 else
4527 {
4538 }
4539 }
4540 }
4542 /* If can't match the null string, and that's all we have left, fail. */
4556 advance:
4560 {
4561 /* Update STARTPOS to the next character boundary. */
4563 {
4572 }
4573 else
4574 {
4575 range--;
4576 startpos++;
4577 }
4578 }
4579 else
4580 {
4581 range++;
4582 startpos--;
4584 /* Update STARTPOS to the previous character boundary. */
4586 {
4591 /* Find the head of multibyte form. */
4598 }
4599 }
4600 }
4604 \f
4605 /* Declarations and macros for re_match_2. */
4609 RE_TRANSLATE_TYPE translate,
4612 /* This converts PTR, a pointer into one of the search strings `string1'
4613 and `string2' into an offset from the beginning of that string. */
4614 #define POINTER_TO_OFFSET(ptr) \
4615 (FIRST_STRING_P (ptr) \
4616 ? ((regoff_t) ((ptr) - string1)) \
4617 : ((regoff_t) ((ptr) - string2 + size1)))
4619 /* Call before fetching a character with *d. This switches over to
4620 string2 if necessary.
4621 Check re_match_2_internal for a discussion of why end_match_2 might
4622 not be within string2 (but be equal to end_match_1 instead). */
4623 #define PREFETCH() \
4624 while (d == dend) \
4625 { \
4627 if (dend == end_match_2) \
4628 goto fail; \
4630 d = string2; \
4631 dend = end_match_2; \
4632 }
4634 /* Call before fetching a char with *d if you already checked other limits.
4635 This is meant for use in lookahead operations like wordend, etc..
4636 where we might need to look at parts of the string that might be
4637 outside of the LIMITs (i.e past `stop'). */
4638 #define PREFETCH_NOLIMIT() \
4639 if (d == end1) \
4640 { \
4641 d = string2; \
4642 dend = end_match_2; \
4643 } \
4645 /* Test if at very beginning or at very end of the virtual concatenation
4646 of `string1' and `string2'. If only one string, it's `string2'. */
4647 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4648 #define AT_STRINGS_END(d) ((d) == end2)
4651 /* Test if D points to a character which is word-constituent. We have
4652 two special cases to check for: if past the end of string1, look at
4653 the first character in string2; and if before the beginning of
4654 string2, look at the last character in string1. */
4655 #define WORDCHAR_P(d) \
4656 (SYNTAX ((d) == end1 ? *string2 \
4657 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4658 == Sword)
4660 /* Disabled due to a compiler bug -- see comment at case wordbound */
4662 /* The comment at case wordbound is following one, but we don't use
4663 AT_WORD_BOUNDARY anymore to support multibyte form.
4665 The DEC Alpha C compiler 3.x generates incorrect code for the
4666 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4667 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4668 macro and introducing temporary variables works around the bug. */
4670 #if 0
4671 /* Test if the character before D and the one at D differ with respect
4672 to being word-constituent. */
4673 #define AT_WORD_BOUNDARY(d) \
4674 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4675 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4676 #endif
4678 /* Free everything we malloc. */
4679 #ifdef MATCH_MAY_ALLOCATE
4680 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4681 # define FREE_VARIABLES() \
4682 do { \
4683 REGEX_FREE_STACK (fail_stack.stack); \
4684 FREE_VAR (regstart); \
4685 FREE_VAR (regend); \
4686 FREE_VAR (best_regstart); \
4687 FREE_VAR (best_regend); \
4688 } while (0)
4689 #else
4693 \f
4694 /* Optimization routines. */
4696 /* If the operation is a match against one or more chars,
4697 return a pointer to the next operation, else return NULL. */
4700 {
4702 {
4713 {
4718 }
4719 else
4725 #ifdef emacs
4729 p++;
4734 }
4736 }
4739 /* Jump over non-matching operations. */
4742 {
4745 {
4747 {
4760 }
4761 }
4764 }
4766 /* Non-zero if "p1 matches something" implies "p2 fails". */
4767 static int
4769 {
4770 re_opcode_t op2;
4777 /* Skip over open/close-group commands.
4778 If what follows this loop is a ...+ construct,
4779 look at what begins its body, since we will have to
4780 match at least one of that. */
4782 /* The same skip can be done for p1, except that this function
4783 is only used in the case where p1 is a simple match operator. */
4784 /* p1 = skip_noops (p1, pend); */
4792 {
4795 /* If we're at the end of the pattern, we can change. */
4797 {
4800 }
4805 {
4806 register re_wchar_t c
4811 {
4813 {
4816 }
4817 }
4821 {
4824 /* Test if C is listed in charset (or charset_not)
4825 at `p1'. */
4827 {
4831 }
4835 /* `not' is equal to 1 if c would match, which means
4836 that we can't change to pop_failure_jump. */
4838 {
4841 }
4842 }
4845 {
4848 }
4849 }
4853 {
4855 /* Reuse the code above. */
4858 /* It is hard to list up all the character in charset
4859 P2 if it includes multibyte character. Give up in
4860 such case. */
4862 {
4863 /* Now, we are sure that P2 has no range table.
4864 So, for the size of bitmap in P2, `p2[1]' is
4865 enough. But P1 may have range table, so the
4866 size of bitmap table of P1 is extracted by
4867 using macro `CHARSET_BITMAP_SIZE'.
4869 In a multibyte case, we know that all the character
4870 listed in P2 is ASCII. In a unibyte case, P1 has only a
4871 bitmap table. So, in both cases, it is enough to test
4872 only the bitmap table of P1. */
4875 {
4877 /* We win if the charset inside the loop
4878 has no overlap with the one after the loop. */
4882 idx++)
4888 {
4891 }
4892 }
4894 {
4896 /* We win if the charset_not inside the loop lists
4897 every character listed in the charset after. */
4905 {
4908 }
4909 }
4910 }
4911 }
4916 {
4919 /* Reuse the code above. */
4922 /* When we have two charset_not, it's very unlikely that
4923 they don't overlap. The union of the two sets of excluded
4924 chars should cover all possible chars, which, as a matter of
4925 fact, is virtually impossible in multibyte buffers. */
4927 }
4951 #ifdef emacs
4959 ;
4960 }
4962 /* Safe default. */
4964 }
4966 \f
4967 /* Matching routines. */
4970 /* re_match is like re_match_2 except it takes only a single string. */
4972 int
4975 {
4979 }
4983 #ifdef emacs
4984 /* In Emacs, this is the string or buffer in which we
4985 are matching. It is used for looking up syntax properties. */
4986 Lisp_Object re_match_object;
4987 #endif
4989 /* re_match_2 matches the compiled pattern in BUFP against the
4990 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4991 and SIZE2, respectively). We start matching at POS, and stop
4992 matching at STOP.
4994 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4995 store offsets for the substring each group matched in REGS. See the
4996 documentation for exactly how many groups we fill.
4998 We return -1 if no match, -2 if an internal error (such as the
4999 failure stack overflowing). Otherwise, we return the length of the
5000 matched substring. */
5002 int
5003 re_match_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, const char *string2, int size2, int pos, struct re_registers *regs, int stop)
5004 {
5007 #ifdef emacs
5012 #endif
5018 }
5022 /* This is a separate function so that we can force an alloca cleanup
5023 afterwards. */
5024 static int
5025 re_match_2_internal (struct re_pattern_buffer *bufp, const re_char *string1, int size1, const re_char *string2, int size2, int pos, struct re_registers *regs, int stop)
5026 {
5027 /* General temporaries. */
5032 /* Just past the end of the corresponding string. */
5035 /* Pointers into string1 and string2, just past the last characters in
5036 each to consider matching. */
5039 /* Where we are in the data, and the end of the current string. */
5042 /* Used sometimes to remember where we were before starting matching
5043 an operator so that we can go back in case of failure. This "atomic"
5044 behavior of matching opcodes is indispensable to the correctness
5045 of the on_failure_keep_string_jump optimization. */
5048 /* Where we are in the pattern, and the end of the pattern. */
5052 /* We use this to map every character in the string. */
5055 /* Nonzero if BUFP is setup from a multibyte regex. */
5058 /* Nonzero if STRING1/STRING2 are multibyte. */
5061 /* Failure point stack. Each place that can handle a failure further
5062 down the line pushes a failure point on this stack. It consists of
5063 regstart, and regend for all registers corresponding to
5064 the subexpressions we're currently inside, plus the number of such
5065 registers, and, finally, two char *'s. The first char * is where
5066 to resume scanning the pattern; the second one is where to resume
5067 scanning the strings. */
5069 fail_stack_type fail_stack;
5070 #endif
5071 #ifdef DEBUG
5073 #endif
5075 #if defined REL_ALLOC && defined REGEX_MALLOC
5076 /* This holds the pointer to the failure stack, when
5077 it is allocated relocatably. */
5079 #endif
5081 /* We fill all the registers internally, independent of what we
5082 return, for use in backreferences. The number here includes
5083 an element for register zero. */
5086 /* Information on the contents of registers. These are pointers into
5087 the input strings; they record just what was matched (on this
5088 attempt) by a subexpression part of the pattern, that is, the
5089 regnum-th regstart pointer points to where in the pattern we began
5090 matching and the regnum-th regend points to right after where we
5091 stopped matching the regnum-th subexpression. (The zeroth register
5092 keeps track of what the whole pattern matches.) */
5095 #endif
5097 /* The following record the register info as found in the above
5098 variables when we find a match better than any we've seen before.
5099 This happens as we backtrack through the failure points, which in
5100 turn happens only if we have not yet matched the entire string. */
5104 #endif
5106 /* Logically, this is `best_regend[0]'. But we don't want to have to
5107 allocate space for that if we're not allocating space for anything
5108 else (see below). Also, we never need info about register 0 for
5109 any of the other register vectors, and it seems rather a kludge to
5110 treat `best_regend' differently than the rest. So we keep track of
5111 the end of the best match so far in a separate variable. We
5112 initialize this to NULL so that when we backtrack the first time
5113 and need to test it, it's not garbage. */
5116 #ifdef DEBUG
5117 /* Counts the total number of registers pushed. */
5119 #endif
5125 #ifdef MATCH_MAY_ALLOCATE
5126 /* Do not bother to initialize all the register variables if there are
5127 no groups in the pattern, as it takes a fair amount of time. If
5128 there are groups, we include space for register 0 (the whole
5129 pattern), even though we never use it, since it simplifies the
5130 array indexing. We should fix this. */
5132 {
5139 {
5142 }
5143 }
5144 else
5145 {
5146 /* We must initialize all our variables to NULL, so that
5147 `FREE_VARIABLES' doesn't try to free them. */
5149 }
5152 /* The starting position is bogus. */
5154 {
5157 }
5159 /* Initialize subexpression text positions to -1 to mark ones that no
5160 start_memory/stop_memory has been seen for. Also initialize the
5161 register information struct. */
5165 /* We move `string1' into `string2' if the latter's empty -- but not if
5166 `string1' is null. */
5168 {
5173 }
5177 /* `p' scans through the pattern as `d' scans through the data.
5178 `dend' is the end of the input string that `d' points within. `d'
5179 is advanced into the following input string whenever necessary, but
5180 this happens before fetching; therefore, at the beginning of the
5181 loop, `d' can be pointing at the end of a string, but it cannot
5182 equal `string2'. */
5184 {
5185 /* Only match within string2. */
5189 }
5190 else
5191 {
5193 {
5194 /* Only match within string1. */
5196 /* BEWARE!
5197 When we reach end_match_1, PREFETCH normally switches to string2.
5198 But in the present case, this means that just doing a PREFETCH
5199 makes us jump from `stop' to `gap' within the string.
5200 What we really want here is for the search to stop as
5201 soon as we hit end_match_1. That's why we set end_match_2
5202 to end_match_1 (since PREFETCH fails as soon as we hit
5203 end_match_2). */
5205 }
5206 else
5208 moving `d' from end1 to string2 will not prevent the d == dend
5209 check from catching the end of string. */
5212 }
5215 }
5223 /* This loops over pattern commands. It exits by returning from the
5224 function if the match is complete, or it drops through if the match
5225 fails at this starting point in the input data. */
5227 {
5234 /* If we haven't matched the entire string, and we want the
5235 longest match, try backtracking. */
5237 {
5238 /* 1 if this match ends in the same string (string1 or string2)
5239 as the best previous match. */
5242 /* 1 if this match is the best seen so far. */
5243 boolean best_match_p;
5245 /* AIX compiler got confused when this was combined
5246 with the previous declaration. */
5249 else
5257 /* If exceeds best match so far, save it. */
5259 {
5266 {
5269 }
5270 }
5272 }
5274 /* If no failure points, don't restore garbage. And if
5275 last match is real best match, don't restore second
5276 best one. */
5278 {
5279 restore_best_regs:
5280 /* Restore best match. It may happen that `dend ==
5281 end_match_1' while the restored d is in string2.
5282 For example, the pattern `x.*y.*z' against the
5283 strings `x-' and `y-z-', if the two strings are
5284 not consecutive in memory. */
5292 {
5295 }
5296 }
5299 succeed_label:
5302 /* If caller wants register contents data back, do it. */
5304 {
5305 /* Have the register data arrays been allocated? */
5308 extra element beyond `num_regs' for the `-1' marker
5309 GNU code uses. */
5314 {
5317 }
5319 }
5322 allocated, reallocate them. If we need fewer, just
5323 leave it alone. */
5325 {
5330 {
5333 }
5334 }
5335 }
5336 else
5337 {
5338 /* These braces fend off a "empty body in an else-statement"
5339 warning under GCC when assert expands to nothing. */
5341 }
5343 /* Convert the pointer data in `regstart' and `regend' to
5344 indices. Register zero has to be set differently,
5345 since we haven't kept track of any info for it. */
5347 {
5350 }
5352 /* Go through the first `min (num_regs, regs->num_regs)'
5353 registers, since that is all we initialized. */
5355 {
5358 else
5359 {
5364 }
5365 }
5367 /* If the regs structure we return has more elements than
5368 were in the pattern, set the extra elements to -1. If
5369 we (re)allocated the registers, this is the case,
5370 because we always allocate enough to have at least one
5371 -1 at the end. */
5387 }
5389 /* Otherwise match next pattern command. */
5391 {
5392 /* Ignore these. Used to ignore the n of succeed_n's which
5393 currently have n == 0. */
5402 /* Match the next n pattern characters exactly. The following
5403 byte in the pattern defines n, and the n bytes after that
5404 are the characters to match. */
5409 /* Remember the start point to rollback upon failure. */
5412 #ifndef emacs
5413 /* This is written out as an if-else so we don't waste time
5414 testing `translate' inside the loop. */
5416 do
5417 {
5420 {
5423 }
5424 d++;
5425 }
5427 else
5428 do
5429 {
5432 {
5435 }
5436 }
5439 /* The cost of testing `translate' is comparatively small. */
5441 do
5442 {
5449 else
5450 {
5453 }
5457 {
5460 }
5465 }
5467 else
5468 do
5469 {
5475 {
5478 }
5479 else
5480 {
5483 }
5486 {
5491 }
5492 else
5495 {
5498 }
5500 d++;
5501 }
5503 #endif
5507 /* Match any character except possibly a newline or a null. */
5509 {
5511 re_wchar_t buf_ch;
5517 target_multibyte);
5528 }
5534 {
5539 /* Start of actual range_table, or end of bitmap if there is no
5540 range table. */
5543 /* Nonzero if there is a range table. */
5546 /* Number of ranges of range table. This is not included
5547 in the initial byte-length of the command. */
5550 /* Whether matching against a unibyte character. */
5558 {
5561 }
5566 {
5572 {
5575 }
5576 }
5577 else
5578 {
5582 {
5586 {
5589 }
5590 }
5591 else
5593 }
5597 /* Cast to `unsigned' instead of `unsigned char' in
5598 case the bit list is a full 32 bytes long. */
5602 }
5603 #ifdef emacs
5605 {
5615 else
5617 }
5622 else
5629 }
5632 /* The beginning of a group is represented by start_memory.
5633 The argument is the register number. The text
5634 matched within the group is recorded (in the internal
5635 registers data structure) under the register number. */
5639 /* In case we need to undo this operation (via backtracking). */
5646 /* Move past the register number and inner group count. */
5651 /* The stop_memory opcode represents the end of a group. Its
5652 argument is the same as start_memory's: the register number. */
5657 /* Strictly speaking, there should be code such as:
5659 assert (REG_UNSET (regend[*p]));
5660 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5662 But the only info to be pushed is regend[*p] and it is known to
5663 be UNSET, so there really isn't anything to push.
5664 Not pushing anything, on the other hand deprives us from the
5665 guarantee that regend[*p] is UNSET since undoing this operation
5666 will not reset its value properly. This is not important since
5667 the value will only be read on the next start_memory or at
5668 the very end and both events can only happen if this stop_memory
5669 is *not* undone. */
5674 /* Move past the register number and the inner group count. */
5679 /* \<digit> has been turned into a `duplicate' command which is
5680 followed by the numeric value of <digit> as the register number. */
5682 {
5687 /* Can't back reference a group which we've never matched. */
5691 /* Where in input to try to start matching. */
5694 /* Remember the start point to rollback upon failure. */
5697 /* Where to stop matching; if both the place to start and
5698 the place to stop matching are in the same string, then
5699 set to the place to stop, otherwise, for now have to use
5700 the end of the first string. */
5706 {
5707 /* If necessary, advance to next segment in register
5708 contents. */
5710 {
5714 /* End of string1 => advance to string2. */
5717 }
5718 /* At end of register contents => success */
5721 /* If necessary, advance to next segment in data. */
5724 /* How many characters left in this segment to match. */
5727 /* Want how many consecutive characters we can match in
5728 one shot, so, if necessary, adjust the count. */
5732 /* Compare that many; failure if mismatch, else move
5733 past them. */
5737 {
5740 }
5742 }
5743 }
5747 /* begline matches the empty string at the beginning of the string
5748 (unless `not_bol' is set in `bufp'), and after newlines. */
5753 {
5755 }
5756 else
5757 {
5762 }
5763 /* In all other cases, we fail. */
5767 /* endline is the dual of begline. */
5772 {
5774 }
5775 else
5776 {
5780 }
5784 /* Match at the very beginning of the data. */
5792 /* Match at the very end of the data. */
5800 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5801 pushes NULL as the value for the string on the stack. Then
5802 `POP_FAILURE_POINT' will keep the current value for the
5803 string, instead of restoring it. To see why, consider
5804 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5805 then the . fails against the \n. But the next thing we want
5806 to do is match the \n against the \n; if we restored the
5807 string value, we would be back at the foo.
5809 Because this is used only in specific cases, we don't need to
5810 check all the things that `on_failure_jump' does, to make
5811 sure the right things get saved on the stack. Hence we don't
5812 share its code. The only reason to push anything on the
5813 stack at all is that otherwise we would have to change
5814 `anychar's code to do something besides goto fail in this
5815 case; that seems worse than this. */
5824 /* A nasty loop is introduced by the non-greedy *? and +?.
5825 With such loops, the stack only ever contains one failure point
5826 at a time, so that a plain on_failure_jump_loop kind of
5827 cycle detection cannot work. Worse yet, such a detection
5828 can not only fail to detect a cycle, but it can also wrongly
5829 detect a cycle (between different instantiations of the same
5830 loop).
5831 So the method used for those nasty loops is a little different:
5832 We use a special cycle-detection-stack-frame which is pushed
5833 when the on_failure_jump_nastyloop failure-point is *popped*.
5834 This special frame thus marks the beginning of one iteration
5835 through the loop and we can hence easily check right here
5836 whether something matched between the beginning and the end of
5837 the loop. */
5844 {
5848 /* If there's a cycle, just continue without pushing
5849 this failure point. The failure point is the "try again"
5850 option, which shouldn't be tried.
5851 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5853 }
5856 /* Simple loop detecting on_failure_jump: just check on the
5857 failure stack if the same spot was already hit earlier. */
5859 on_failure:
5863 {
5867 /* If there's a cycle, get out of the loop, as if the matching
5868 had failed. We used to just `goto fail' here, but that was
5869 aborting the search a bit too early: we want to keep the
5870 empty-loop-match and keep matching after the loop.
5871 We want (x?)*y\1z to match both xxyz and xxyxz. */
5873 else
5875 }
5879 /* Uses of on_failure_jump:
5881 Each alternative starts with an on_failure_jump that points
5882 to the beginning of the next alternative. Each alternative
5883 except the last ends with a jump that in effect jumps past
5884 the rest of the alternatives. (They really jump to the
5885 ending jump of the following alternative, because tensioning
5886 these jumps is a hassle.)
5888 Repeats start with an on_failure_jump that points past both
5889 the repetition text and either the following jump or
5890 pop_failure_jump back to this on_failure_jump. */
5899 /* This operation is used for greedy *.
5900 Compare the beginning of the repeat with what in the
5901 pattern follows its end. If we can establish that there
5902 is nothing that they would both match, i.e., that we
5903 would have to backtrack because of (as in, e.g., `a*a')
5904 then we can use a non-backtracking loop based on
5905 on_failure_keep_string_jump instead of on_failure_jump. */
5910 {
5912 /* Here, we discard `const', making re_match non-reentrant. */
5917 instruction once it's been changed. */
5921 /* Ensure this is a indeed the trivial kind of loop
5922 we are expecting. */
5927 {
5928 /* Use a fast `on_failure_keep_string_jump' loop. */
5932 }
5933 else
5934 {
5935 /* Default to a safe `on_failure_jump' loop. */
5938 }
5940 }
5943 /* Unconditionally jump (without popping any failure points). */
5945 unconditional_jump:
5946 IMMEDIATE_QUIT_CHECK;
5954 /* Have to succeed matching what follows at least n times.
5955 After that, handle like `on_failure_jump'. */
5957 /* Signedness doesn't matter since we only compare MCNT to 0. */
5961 /* Originally, mcnt is how many times we HAVE to succeed. */
5963 {
5964 /* Here, we discard `const', making re_match non-reentrant. */
5966 mcnt--;
5969 }
5970 else
5971 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5976 /* Signedness doesn't matter since we only compare MCNT to 0. */
5980 /* Originally, this is how many times we CAN jump. */
5982 {
5983 /* Here, we discard `const', making re_match non-reentrant. */
5985 mcnt--;
5988 }
5989 /* If don't have to jump any more, skip over the rest of command. */
5990 else
5995 {
6000 /* Here, we discard `const', making re_match non-reentrant. */
6002 /* Signedness doesn't matter since we only copy MCNT's bits . */
6007 }
6014 /* We SUCCEED (or FAIL) in one of the following cases: */
6016 /* Case 1: D is at the beginning or the end of string. */
6019 else
6020 {
6021 /* C1 is the character before D, S1 is the syntax of C1, C2
6022 is the character at D, and S2 is the syntax of C2. */
6026 #ifdef emacs
6030 #endif
6033 #ifdef emacs
6035 #endif
6042 /* Case 3: Both of S1 and S2 are Sword, and macro
6043 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6046 }
6049 else
6055 /* We FAIL in one of the following cases: */
6057 /* Case 1: D is at the end of string. */
6060 else
6061 {
6062 /* C1 is the character before D, S1 is the syntax of C1, C2
6063 is the character at D, and S2 is the syntax of C2. */
6067 #ifdef emacs
6071 #endif
6076 /* Case 2: S2 is not Sword. */
6080 /* Case 3: D is not at the beginning of string ... */
6082 {
6084 #ifdef emacs
6086 #endif
6089 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6090 returns 0. */
6093 }
6094 }
6100 /* We FAIL in one of the following cases: */
6102 /* Case 1: D is at the beginning of string. */
6105 else
6106 {
6107 /* C1 is the character before D, S1 is the syntax of C1, C2
6108 is the character at D, and S2 is the syntax of C2. */
6112 #ifdef emacs
6116 #endif
6120 /* Case 2: S1 is not Sword. */
6124 /* Case 3: D is not at the end of string ... */
6126 {
6129 #ifdef emacs
6131 #endif
6134 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6135 returns 0. */
6138 }
6139 }
6145 /* We FAIL in one of the following cases: */
6147 /* Case 1: D is at the end of string. */
6150 else
6151 {
6152 /* C1 is the character before D, S1 is the syntax of C1, C2
6153 is the character at D, and S2 is the syntax of C2. */
6156 #ifdef emacs
6160 #endif
6165 /* Case 2: S2 is neither Sword nor Ssymbol. */
6169 /* Case 3: D is not at the beginning of string ... */
6171 {
6173 #ifdef emacs
6175 #endif
6178 /* ... and S1 is Sword or Ssymbol. */
6181 }
6182 }
6188 /* We FAIL in one of the following cases: */
6190 /* Case 1: D is at the beginning of string. */
6193 else
6194 {
6195 /* C1 is the character before D, S1 is the syntax of C1, C2
6196 is the character at D, and S2 is the syntax of C2. */
6199 #ifdef emacs
6203 #endif
6207 /* Case 2: S1 is neither Ssymbol nor Sword. */
6211 /* Case 3: D is not at the end of string ... */
6213 {
6216 #ifdef emacs
6218 #endif
6221 /* ... and S2 is Sword or Ssymbol. */
6224 }
6225 }
6234 #ifdef emacs
6235 {
6239 }
6240 #endif
6241 {
6243 re_wchar_t c;
6249 }
6252 #ifdef emacs
6277 {
6279 re_wchar_t c;
6285 }
6292 }
6296 /* We goto here if a matching operation fails. */
6297 fail:
6298 IMMEDIATE_QUIT_CHECK;
6300 {
6302 /* A restart point is known. Restore to that state. */
6306 {
6314 /* Fallthrough */
6320 continue_failure_jump:
6326 /* A special frame used for nastyloops. */
6331 }
6337 }
6338 else
6349 \f
6350 /* Subroutine definitions for re_match_2. */
6352 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6353 bytes; nonzero otherwise. */
6355 static int
6358 {
6363 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6364 different lengths, but relying on a single `len' would break this. -sm */
6366 {
6378 }
6384 }
6385 \f
6386 /* Entry points for GNU code. */
6388 /* re_compile_pattern is the GNU regular expression compiler: it
6389 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6390 Returns 0 if the pattern was valid, otherwise an error string.
6392 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6393 are set in BUFP on entry.
6395 We call regex_compile to do the actual compilation. */
6399 {
6400 reg_errcode_t ret;
6402 /* GNU code is written to assume at least RE_NREGS registers will be set
6403 (and at least one extra will be -1). */
6406 /* And GNU code determines whether or not to get register information
6407 by passing null for the REGS argument to re_match, etc., not by
6408 setting no_sub. */
6416 }
6418 \f
6419 /* Entry points compatible with 4.2 BSD regex library. We don't define
6420 them unless specifically requested. */
6422 #if defined _REGEX_RE_COMP || defined _LIBC
6424 /* BSD has one and only one pattern buffer. */
6428 # ifdef _LIBC
6429 /* Make these definitions weak in libc, so POSIX programs can redefine
6430 these names if they don't use our functions, and still use
6431 regcomp/regexec below without link errors. */
6432 weak_function
6433 # endif
6436 {
6437 reg_errcode_t ret;
6440 {
6442 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6445 }
6448 {
6451 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6457 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6459 }
6461 /* Since `re_exec' always passes NULL for the `regs' argument, we
6462 don't need to initialize the pattern buffer fields which affect it. */
6469 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6471 }
6474 int
6475 # ifdef _LIBC
6476 weak_function
6477 # endif
6480 {
6482 return
6484 }
6486 \f
6487 /* POSIX.2 functions. Don't define these for Emacs. */
6489 #ifndef emacs
6491 /* regcomp takes a regular expression as a string and compiles it.
6493 PREG is a regex_t *. We do not expect any fields to be initialized,
6494 since POSIX says we shouldn't. Thus, we set
6496 `buffer' to the compiled pattern;
6497 `used' to the length of the compiled pattern;
6498 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6499 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6500 RE_SYNTAX_POSIX_BASIC;
6501 `fastmap' to an allocated space for the fastmap;
6502 `fastmap_accurate' to zero;
6503 `re_nsub' to the number of subexpressions in PATTERN.
6505 PATTERN is the address of the pattern string.
6507 CFLAGS is a series of bits which affect compilation.
6509 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6510 use POSIX basic syntax.
6512 If REG_NEWLINE is set, then . and [^...] don't match newline.
6513 Also, regexec will try a match beginning after every newline.
6515 If REG_ICASE is set, then we considers upper- and lowercase
6516 versions of letters to be equivalent when matching.
6518 If REG_NOSUB is set, then when PREG is passed to regexec, that
6519 routine will report only success or failure, and nothing about the
6520 registers.
6522 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6523 the return codes and their meanings.) */
6525 int
6528 {
6529 reg_errcode_t ret;
6530 reg_syntax_t syntax
6534 /* regex_compile will allocate the space for the compiled pattern. */
6539 /* Try to allocate space for the fastmap. */
6543 {
6546 preg->translate
6552 /* Map uppercase characters to corresponding lowercase ones. */
6555 }
6556 else
6559 /* If REG_NEWLINE is set, newlines are treated differently. */
6564 }
6565 else
6570 /* POSIX says a null character in the pattern terminates it, so we
6571 can use strlen here in compiling the pattern. */
6574 /* POSIX doesn't distinguish between an unmatched open-group and an
6575 unmatched close-group: both are REG_EPAREN. */
6581 buffer. */
6587 }
6588 }
6590 }
6594 /* regexec searches for a given pattern, specified by PREG, in the
6595 string STRING.
6597 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6598 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6599 least NMATCH elements, and we set them to the offsets of the
6600 corresponding matched substrings.
6602 EFLAGS specifies `execution flags' which affect matching: if
6603 REG_NOTBOL is set, then ^ does not match at the beginning of the
6604 string; if REG_NOTEOL is set, then $ does not match at the end.
6606 We return 0 if we find a match and REG_NOMATCH if not. */
6608 int
6611 {
6614 regex_t private_preg;
6623 /* The user has told us exactly how many registers to return
6624 information about, via `nmatch'. We have to pass that on to the
6625 matching routines. */
6629 {
6635 }
6637 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6638 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6639 was a little bit longer but still only matching the real part.
6640 This works because the `endline' will check for a '\n' and will find a
6641 '\0', correctly deciding that this is not the end of a line.
6642 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6643 a convenient '\0' there. For all we know, the string could be preceded
6644 by '\n' which would throw things off. */
6646 /* Perform the searching operation. */
6651 /* Copy the register information to the POSIX structure. */
6653 {
6655 {
6659 {
6662 }
6663 }
6665 /* If we needed the temporary register info, free the space now. */
6667 }
6669 /* We want zero return to mean success, unlike `re_search'. */
6671 }
6675 /* Returns a message corresponding to an error code, ERR_CODE, returned
6676 from either regcomp or regexec. We don't use PREG here.
6678 ERR_CODE was previously called ERRCODE, but that name causes an
6679 error with msvc8 compiler. */
6681 size_t
6683 {
6689 /* Only error codes returned by the rest of the code should be passed
6690 to this routine. If we are given anything else, or if other regex
6691 code generates an invalid error code, then the program has a bug.
6692 Dump core so we can fix it. */
6700 {
6702 {
6705 }
6706 else
6708 }
6711 }
6715 /* Free dynamically allocated space used by PREG. */
6717 void
6719 {
6732 }