| blob | 17158552a95e67ee1fc2924917873b49fae572d3 |
1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the
3 internationalization features.)
5 Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
6 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
7 Free Software Foundation, Inc.
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 3, or (at your option)
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 # if defined STDC_HEADERS || defined _LIBC
200 # include <stdlib.h>
201 # else
204 # endif
206 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
211 {
215 {
218 }
220 }
226 {
228 /* We must call malloc explicitly when BLOCK is 0, since some
229 reallocs don't do this. */
232 else
235 {
238 }
240 }
242 # ifdef malloc
243 # undef malloc
244 # endif
245 # define malloc xmalloc
246 # ifdef realloc
247 # undef realloc
248 # endif
249 # define realloc xrealloc
251 /* This is the normal way of making sure we have memcpy, memcmp and memset. */
252 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
253 # include <string.h>
254 # else
255 # include <strings.h>
256 # ifndef memcmp
257 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
258 # endif
259 # ifndef memcpy
260 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
261 # endif
262 # endif
264 /* Define the syntax stuff for \<, \>, etc. */
266 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
269 # define SWITCH_ENUM_CAST(x) (x)
271 /* Dummy macros for non-Emacs environments. */
272 # define CHAR_CHARSET(c) 0
273 # define CHARSET_LEADING_CODE_BASE(c) 0
274 # define MAX_MULTIBYTE_LENGTH 1
275 # define RE_MULTIBYTE_P(x) 0
276 # define RE_TARGET_MULTIBYTE_P(x) 0
277 # define WORD_BOUNDARY_P(c1, c2) (0)
278 # define CHAR_HEAD_P(p) (1)
279 # define SINGLE_BYTE_CHAR_P(c) (1)
280 # define SAME_CHARSET_P(c1, c2) (1)
281 # define BYTES_BY_CHAR_HEAD(p) (1)
282 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
283 # define STRING_CHAR(p) (*(p))
284 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
285 # define CHAR_STRING(c, s) (*(s) = (c), 1)
286 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
287 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
288 # define RE_CHAR_TO_MULTIBYTE(c) (c)
289 # define RE_CHAR_TO_UNIBYTE(c) (c)
290 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
291 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
292 # define GET_CHAR_AFTER(c, p, len) \
293 (c = *p, len = 1)
294 # define MAKE_CHAR(charset, c1, c2) (c1)
295 # define BYTE8_TO_CHAR(c) (c)
296 # define CHAR_BYTE8_P(c) (0)
297 # define CHAR_LEADING_CODE(c) (c)
301 #ifndef RE_TRANSLATE
302 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
303 # define RE_TRANSLATE_P(TBL) (TBL)
304 #endif
305 \f
306 /* Get the interface, including the syntax bits. */
309 /* isalpha etc. are used for the character classes. */
310 #include <ctype.h>
312 #ifdef emacs
314 /* 1 if C is an ASCII character. */
315 # define IS_REAL_ASCII(c) ((c) < 0200)
317 /* 1 if C is a unibyte character. */
318 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
320 /* The Emacs definitions should not be directly affected by locales. */
322 /* In Emacs, these are only used for single-byte characters. */
329 /* This is only used for single-byte characters. */
332 /* The rest must handle multibyte characters. */
334 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
336 : 1)
338 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
340 : 1)
342 # define ISALNUM(c) (IS_REAL_ASCII (c) \
346 : SYNTAX (c) == Sword)
348 # define ISALPHA(c) (IS_REAL_ASCII (c) \
351 : SYNTAX (c) == Sword)
353 # define ISLOWER(c) (LOWERCASEP (c))
355 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
360 : SYNTAX (c) != Sword)
362 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
364 # define ISUPPER(c) (UPPERCASEP (c))
366 # define ISWORD(c) (SYNTAX (c) == Sword)
370 /* Jim Meyering writes:
372 "... Some ctype macros are valid only for character codes that
373 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
374 using /bin/cc or gcc but without giving an ansi option). So, all
375 ctype uses should be through macros like ISPRINT... If
376 STDC_HEADERS is defined, then autoconf has verified that the ctype
377 macros don't need to be guarded with references to isascii. ...
378 Defining isascii to 1 should let any compiler worth its salt
379 eliminate the && through constant folding."
380 Solaris defines some of these symbols so we must undefine them first. */
382 # undef ISASCII
383 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
384 # define ISASCII(c) 1
385 # else
386 # define ISASCII(c) isascii(c)
387 # endif
389 /* 1 if C is an ASCII character. */
390 # define IS_REAL_ASCII(c) ((c) < 0200)
392 /* This distinction is not meaningful, except in Emacs. */
393 # define ISUNIBYTE(c) 1
395 # ifdef isblank
396 # define ISBLANK(c) (ISASCII (c) && isblank (c))
397 # else
399 # endif
400 # ifdef isgraph
401 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
402 # else
403 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
404 # endif
406 # undef ISPRINT
407 # define ISPRINT(c) (ISASCII (c) && isprint (c))
408 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
409 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
410 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
411 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
412 # define ISLOWER(c) (ISASCII (c) && islower (c))
413 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
414 # define ISSPACE(c) (ISASCII (c) && isspace (c))
415 # define ISUPPER(c) (ISASCII (c) && isupper (c))
416 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
418 # define ISWORD(c) ISALPHA(c)
420 # ifdef _tolower
421 # define TOLOWER(c) _tolower(c)
422 # else
423 # define TOLOWER(c) tolower(c)
424 # endif
426 /* How many characters in the character set. */
427 # define CHAR_SET_SIZE 256
429 # ifdef SYNTAX_TABLE
437 static void
439 {
455 }
459 # define SYNTAX(c) re_syntax_table[(c)]
462 \f
463 #ifndef NULL
464 # define NULL (void *)0
465 #endif
467 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
468 since ours (we hope) works properly with all combinations of
469 machines, compilers, `char' and `unsigned char' argument types.
470 (Per Bothner suggested the basic approach.) */
471 #undef SIGN_EXTEND_CHAR
472 #if __STDC__
473 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
475 /* As in Harbison and Steele. */
476 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
477 #endif
478 \f
479 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
480 use `alloca' instead of `malloc'. This is because using malloc in
481 re_search* or re_match* could cause memory leaks when C-g is used in
482 Emacs; also, malloc is slower and causes storage fragmentation. On
483 the other hand, malloc is more portable, and easier to debug.
485 Because we sometimes use alloca, some routines have to be macros,
486 not functions -- `alloca'-allocated space disappears at the end of the
487 function it is called in. */
489 #ifdef REGEX_MALLOC
491 # define REGEX_ALLOCATE malloc
492 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
493 # define REGEX_FREE free
497 /* Emacs already defines alloca, sometimes. */
498 # ifndef alloca
500 /* Make alloca work the best possible way. */
501 # ifdef __GNUC__
502 # define alloca __builtin_alloca
504 # ifdef HAVE_ALLOCA_H
505 # include <alloca.h>
511 # define REGEX_ALLOCATE alloca
513 /* Assumes a `char *destination' variable. */
514 # define REGEX_REALLOCATE(source, osize, nsize) \
515 (destination = (char *) alloca (nsize), \
516 memcpy (destination, source, osize))
518 /* No need to do anything to free, after alloca. */
523 /* Define how to allocate the failure stack. */
525 #if defined REL_ALLOC && defined REGEX_MALLOC
527 # define REGEX_ALLOCATE_STACK(size) \
528 r_alloc (&failure_stack_ptr, (size))
529 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
530 r_re_alloc (&failure_stack_ptr, (nsize))
531 # define REGEX_FREE_STACK(ptr) \
532 r_alloc_free (&failure_stack_ptr)
536 # ifdef REGEX_MALLOC
538 # define REGEX_ALLOCATE_STACK malloc
539 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
540 # define REGEX_FREE_STACK free
544 # define REGEX_ALLOCATE_STACK alloca
546 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
547 REGEX_REALLOCATE (source, osize, nsize)
548 /* No need to explicitly free anything. */
549 # define REGEX_FREE_STACK(arg) ((void)0)
555 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
556 `string1' or just past its end. This works if PTR is NULL, which is
557 a good thing. */
558 #define FIRST_STRING_P(ptr) \
559 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
561 /* (Re)Allocate N items of type T using malloc, or fail. */
562 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
563 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
564 #define RETALLOC_IF(addr, n, t) \
565 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
566 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
570 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
572 #undef MAX
573 #undef MIN
574 #define MAX(a, b) ((a) > (b) ? (a) : (b))
575 #define MIN(a, b) ((a) < (b) ? (a) : (b))
577 /* Type of source-pattern and string chars. */
581 #define false 0
582 #define true 1
590 \f
591 /* These are the command codes that appear in compiled regular
592 expressions. Some opcodes are followed by argument bytes. A
593 command code can specify any interpretation whatsoever for its
594 arguments. Zero bytes may appear in the compiled regular expression. */
597 {
600 /* Succeed right away--no more backtracking. */
601 succeed,
603 /* Followed by one byte giving n, then by n literal bytes. */
604 exactn,
606 /* Matches any (more or less) character. */
607 anychar,
609 /* Matches any one char belonging to specified set. First
610 following byte is number of bitmap bytes. Then come bytes
611 for a bitmap saying which chars are in. Bits in each byte
612 are ordered low-bit-first. A character is in the set if its
613 bit is 1. A character too large to have a bit in the map is
614 automatically not in the set.
616 If the length byte has the 0x80 bit set, then that stuff
617 is followed by a range table:
618 2 bytes of flags for character sets (low 8 bits, high 8 bits)
619 See RANGE_TABLE_WORK_BITS below.
620 2 bytes, the number of pairs that follow (upto 32767)
621 pairs, each 2 multibyte characters,
622 each multibyte character represented as 3 bytes. */
623 charset,
625 /* Same parameters as charset, but match any character that is
626 not one of those specified. */
627 charset_not,
629 /* Start remembering the text that is matched, for storing in a
630 register. Followed by one byte with the register number, in
631 the range 0 to one less than the pattern buffer's re_nsub
632 field. */
633 start_memory,
635 /* Stop remembering the text that is matched and store it in a
636 memory register. Followed by one byte with the register
637 number, in the range 0 to one less than `re_nsub' in the
638 pattern buffer. */
639 stop_memory,
641 /* Match a duplicate of something remembered. Followed by one
642 byte containing the register number. */
643 duplicate,
645 /* Fail unless at beginning of line. */
646 begline,
648 /* Fail unless at end of line. */
649 endline,
651 /* Succeeds if at beginning of buffer (if emacs) or at beginning
652 of string to be matched (if not). */
653 begbuf,
655 /* Analogously, for end of buffer/string. */
656 endbuf,
658 /* Followed by two byte relative address to which to jump. */
659 jump,
661 /* Followed by two-byte relative address of place to resume at
662 in case of failure. */
663 on_failure_jump,
665 /* Like on_failure_jump, but pushes a placeholder instead of the
666 current string position when executed. */
667 on_failure_keep_string_jump,
669 /* Just like `on_failure_jump', except that it checks that we
670 don't get stuck in an infinite loop (matching an empty string
671 indefinitely). */
672 on_failure_jump_loop,
674 /* Just like `on_failure_jump_loop', except that it checks for
675 a different kind of loop (the kind that shows up with non-greedy
676 operators). This operation has to be immediately preceded
677 by a `no_op'. */
678 on_failure_jump_nastyloop,
680 /* A smart `on_failure_jump' used for greedy * and + operators.
681 It analyses the loop before which it is put and if the
682 loop does not require backtracking, it changes itself to
683 `on_failure_keep_string_jump' and short-circuits the loop,
684 else it just defaults to changing itself into `on_failure_jump'.
685 It assumes that it is pointing to just past a `jump'. */
686 on_failure_jump_smart,
688 /* Followed by two-byte relative address and two-byte number n.
689 After matching N times, jump to the address upon failure.
690 Does not work if N starts at 0: use on_failure_jump_loop
691 instead. */
692 succeed_n,
694 /* Followed by two-byte relative address, and two-byte number n.
695 Jump to the address N times, then fail. */
696 jump_n,
698 /* Set the following two-byte relative address to the
699 subsequent two-byte number. The address *includes* the two
700 bytes of number. */
701 set_number_at,
712 /* Matches any character whose syntax is specified. Followed by
713 a byte which contains a syntax code, e.g., Sword. */
714 syntaxspec,
716 /* Matches any character whose syntax is not that specified. */
717 notsyntaxspec
719 #ifdef emacs
724 /* Matches any character whose category-set contains the specified
725 category. The operator is followed by a byte which contains a
726 category code (mnemonic ASCII character). */
727 categoryspec,
729 /* Matches any character whose category-set does not contain the
730 specified category. The operator is followed by a byte which
731 contains the category code (mnemonic ASCII character). */
732 notcategoryspec
735 \f
736 /* Common operations on the compiled pattern. */
738 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
740 #define STORE_NUMBER(destination, number) \
741 do { \
742 (destination)[0] = (number) & 0377; \
743 (destination)[1] = (number) >> 8; \
744 } while (0)
746 /* Same as STORE_NUMBER, except increment DESTINATION to
747 the byte after where the number is stored. Therefore, DESTINATION
748 must be an lvalue. */
750 #define STORE_NUMBER_AND_INCR(destination, number) \
751 do { \
752 STORE_NUMBER (destination, number); \
753 (destination) += 2; \
754 } while (0)
756 /* Put into DESTINATION a number stored in two contiguous bytes starting
757 at SOURCE. */
759 #define EXTRACT_NUMBER(destination, source) \
760 do { \
761 (destination) = *(source) & 0377; \
762 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
763 } while (0)
765 #ifdef DEBUG
767 static void
771 {
775 }
778 # undef EXTRACT_NUMBER
779 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
784 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
785 SOURCE must be an lvalue. */
787 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
788 do { \
789 EXTRACT_NUMBER (destination, source); \
790 (source) += 2; \
791 } while (0)
793 #ifdef DEBUG
796 static void
800 {
803 }
805 # ifndef EXTRACT_MACROS
806 # undef EXTRACT_NUMBER_AND_INCR
807 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
808 extract_number_and_incr (&dest, &src)
812 \f
813 /* Store a multibyte character in three contiguous bytes starting
814 DESTINATION, and increment DESTINATION to the byte after where the
815 character is stored. Therefore, DESTINATION must be an lvalue. */
817 #define STORE_CHARACTER_AND_INCR(destination, character) \
818 do { \
819 (destination)[0] = (character) & 0377; \
820 (destination)[1] = ((character) >> 8) & 0377; \
821 (destination)[2] = (character) >> 16; \
822 (destination) += 3; \
823 } while (0)
825 /* Put into DESTINATION a character stored in three contiguous bytes
826 starting at SOURCE. */
828 #define EXTRACT_CHARACTER(destination, source) \
829 do { \
830 (destination) = ((source)[0] \
831 | ((source)[1] << 8) \
832 | ((source)[2] << 16)); \
833 } while (0)
836 /* Macros for charset. */
838 /* Size of bitmap of charset P in bytes. P is a start of charset,
839 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
840 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
842 /* Nonzero if charset P has range table. */
843 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
845 /* Return the address of range table of charset P. But not the start
846 of table itself, but the before where the number of ranges is
847 stored. `2 +' means to skip re_opcode_t and size of bitmap,
848 and the 2 bytes of flags at the start of the range table. */
849 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
851 /* Extract the bit flags that start a range table. */
852 #define CHARSET_RANGE_TABLE_BITS(p) \
853 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
854 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
856 /* Test if C is listed in the bitmap of charset P. */
857 #define CHARSET_LOOKUP_BITMAP(p, c) \
858 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
859 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
861 /* Return the address of end of RANGE_TABLE. COUNT is number of
862 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
863 is start of range and end of range. `* 3' is size of each start
864 and end. */
865 #define CHARSET_RANGE_TABLE_END(range_table, count) \
866 ((range_table) + (count) * 2 * 3)
868 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
869 COUNT is number of ranges in RANGE_TABLE. */
870 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
871 do \
872 { \
873 re_wchar_t range_start, range_end; \
874 re_char *p; \
875 re_char *range_table_end \
876 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
877 \
878 for (p = (range_table); p < range_table_end; p += 2 * 3) \
879 { \
880 EXTRACT_CHARACTER (range_start, p); \
881 EXTRACT_CHARACTER (range_end, p + 3); \
882 \
883 if (range_start <= (c) && (c) <= range_end) \
884 { \
885 (not) = !(not); \
886 break; \
887 } \
888 } \
889 } \
890 while (0)
892 /* Test if C is in range table of CHARSET. The flag NOT is negated if
893 C is listed in it. */
894 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
895 do \
896 { \
898 int count; \
899 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
900 \
901 EXTRACT_NUMBER_AND_INCR (count, range_table); \
902 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
903 } \
904 while (0)
905 \f
906 /* If DEBUG is defined, Regex prints many voluminous messages about what
907 it is doing (if the variable `debug' is nonzero). If linked with the
908 main program in `iregex.c', you can enter patterns and strings
909 interactively. And if linked with the main program in `main.c' and
910 the other test files, you can run the already-written tests. */
912 #ifdef DEBUG
914 /* We use standard I/O for debugging. */
915 # include <stdio.h>
917 /* It is useful to test things that ``must'' be true when debugging. */
918 # include <assert.h>
922 # define DEBUG_STATEMENT(e) e
923 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
924 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
925 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
926 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
927 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
928 if (debug > 0) print_partial_compiled_pattern (s, e)
929 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
930 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
933 /* Print the fastmap in human-readable form. */
935 void
938 {
943 {
945 {
949 {
951 i++;
952 }
954 {
957 }
958 }
959 }
961 }
964 /* Print a compiled pattern string in human-readable form, starting at
965 the START pointer into it and ending just before the pointer END. */
967 void
971 {
977 {
980 }
982 /* Loop over pattern commands. */
984 {
988 {
1000 do
1001 {
1003 }
1025 {
1040 {
1041 /* Are we starting a range? */
1043 {
1046 }
1047 /* Have we broken a range? */
1049 {
1052 }
1058 }
1068 {
1072 /* ??? Should print the range table; for now, just skip it. */
1076 }
1077 }
1172 # ifdef emacs
1208 }
1211 }
1214 }
1217 void
1220 {
1228 {
1231 }
1241 /* Perhaps we should print the translate table? */
1242 }
1245 void
1252 {
1257 else
1258 {
1260 {
1265 }
1269 }
1270 }
1274 # undef assert
1275 # define assert(e)
1277 # define DEBUG_STATEMENT(e)
1278 # define DEBUG_PRINT1(x)
1279 # define DEBUG_PRINT2(x1, x2)
1280 # define DEBUG_PRINT3(x1, x2, x3)
1281 # define DEBUG_PRINT4(x1, x2, x3, x4)
1282 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1283 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1286 \f
1287 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1288 also be assigned to arbitrarily: each pattern buffer stores its own
1289 syntax, so it can be changed between regex compilations. */
1290 /* This has no initializer because initialized variables in Emacs
1291 become read-only after dumping. */
1292 reg_syntax_t re_syntax_options;
1295 /* Specify the precise syntax of regexps for compilation. This provides
1296 for compatibility for various utilities which historically have
1297 different, incompatible syntaxes.
1299 The argument SYNTAX is a bit mask comprised of the various bits
1300 defined in regex.h. We return the old syntax. */
1302 reg_syntax_t
1304 {
1309 }
1312 /* Regexp to use to replace spaces, or NULL meaning don't. */
1315 void
1317 {
1319 }
1321 \f
1322 /* This table gives an error message for each of the error codes listed
1323 in regex.h. Obviously the order here has to be same as there.
1324 POSIX doesn't require that we do anything for REG_NOERROR,
1325 but why not be nice? */
1328 {
1347 };
1348 \f
1349 /* Avoiding alloca during matching, to placate r_alloc. */
1351 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1352 searching and matching functions should not call alloca. On some
1353 systems, alloca is implemented in terms of malloc, and if we're
1354 using the relocating allocator routines, then malloc could cause a
1355 relocation, which might (if the strings being searched are in the
1356 ralloc heap) shift the data out from underneath the regexp
1357 routines.
1359 Here's another reason to avoid allocation: Emacs
1360 processes input from X in a signal handler; processing X input may
1361 call malloc; if input arrives while a matching routine is calling
1362 malloc, then we're scrod. But Emacs can't just block input while
1363 calling matching routines; then we don't notice interrupts when
1364 they come in. So, Emacs blocks input around all regexp calls
1365 except the matching calls, which it leaves unprotected, in the
1366 faith that they will not malloc. */
1368 /* Normally, this is fine. */
1369 #define MATCH_MAY_ALLOCATE
1371 /* The match routines may not allocate if (1) they would do it with malloc
1372 and (2) it's not safe for them to use malloc.
1373 Note that if REL_ALLOC is defined, matching would not use malloc for the
1374 failure stack, but we would still use it for the register vectors;
1375 so REL_ALLOC should not affect this. */
1376 #if defined REGEX_MALLOC && defined emacs
1377 # undef MATCH_MAY_ALLOCATE
1378 #endif
1380 \f
1381 /* Failure stack declarations and macros; both re_compile_fastmap and
1382 re_match_2 use a failure stack. These have to be macros because of
1383 REGEX_ALLOCATE_STACK. */
1386 /* Approximate number of failure points for which to initially allocate space
1387 when matching. If this number is exceeded, we allocate more
1388 space, so it is not a hard limit. */
1389 #ifndef INIT_FAILURE_ALLOC
1390 # define INIT_FAILURE_ALLOC 20
1391 #endif
1393 /* Roughly the maximum number of failure points on the stack. Would be
1394 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1395 This is a variable only so users of regex can assign to it; we never
1396 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1397 before using it, so it should probably be a byte-count instead. */
1398 # if defined MATCH_MAY_ALLOCATE
1399 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1400 whose default stack limit is 2mb. In order for a larger
1401 value to work reliably, you have to try to make it accord
1402 with the process stack limit. */
1404 # else
1406 # endif
1408 union fail_stack_elt
1409 {
1411 /* This should be the biggest `int' that's no bigger than a pointer. */
1413 };
1418 {
1425 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1426 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1429 /* Define macros to initialize and free the failure stack.
1430 Do `return -2' if the alloc fails. */
1432 #ifdef MATCH_MAY_ALLOCATE
1433 # define INIT_FAIL_STACK() \
1434 do { \
1435 fail_stack.stack = (fail_stack_elt_t *) \
1436 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1437 * sizeof (fail_stack_elt_t)); \
1438 \
1439 if (fail_stack.stack == NULL) \
1440 return -2; \
1441 \
1442 fail_stack.size = INIT_FAILURE_ALLOC; \
1443 fail_stack.avail = 0; \
1444 fail_stack.frame = 0; \
1445 } while (0)
1447 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1448 #else
1449 # define INIT_FAIL_STACK() \
1450 do { \
1451 fail_stack.avail = 0; \
1452 fail_stack.frame = 0; \
1453 } while (0)
1455 # define RESET_FAIL_STACK() ((void)0)
1456 #endif
1459 /* Double the size of FAIL_STACK, up to a limit
1460 which allows approximately `re_max_failures' items.
1462 Return 1 if succeeds, and 0 if either ran out of memory
1463 allocating space for it or it was already too large.
1465 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1467 /* Factor to increase the failure stack size by
1468 when we increase it.
1469 This used to be 2, but 2 was too wasteful
1470 because the old discarded stacks added up to as much space
1471 were as ultimate, maximum-size stack. */
1472 #define FAIL_STACK_GROWTH_FACTOR 4
1474 #define GROW_FAIL_STACK(fail_stack) \
1475 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1476 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1477 ? 0 \
1478 : ((fail_stack).stack \
1479 = (fail_stack_elt_t *) \
1480 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1481 (fail_stack).size * sizeof (fail_stack_elt_t), \
1482 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1483 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1484 * FAIL_STACK_GROWTH_FACTOR))), \
1485 \
1486 (fail_stack).stack == NULL \
1487 ? 0 \
1488 : ((fail_stack).size \
1489 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1490 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1491 * FAIL_STACK_GROWTH_FACTOR)) \
1492 / sizeof (fail_stack_elt_t)), \
1493 1)))
1496 /* Push a pointer value onto the failure stack.
1497 Assumes the variable `fail_stack'. Probably should only
1498 be called from within `PUSH_FAILURE_POINT'. */
1499 #define PUSH_FAILURE_POINTER(item) \
1500 fail_stack.stack[fail_stack.avail++].pointer = (item)
1502 /* This pushes an integer-valued item onto the failure stack.
1503 Assumes the variable `fail_stack'. Probably should only
1504 be called from within `PUSH_FAILURE_POINT'. */
1505 #define PUSH_FAILURE_INT(item) \
1506 fail_stack.stack[fail_stack.avail++].integer = (item)
1508 /* Push a fail_stack_elt_t value onto the failure stack.
1509 Assumes the variable `fail_stack'. Probably should only
1510 be called from within `PUSH_FAILURE_POINT'. */
1511 #define PUSH_FAILURE_ELT(item) \
1512 fail_stack.stack[fail_stack.avail++] = (item)
1514 /* These three POP... operations complement the three PUSH... operations.
1515 All assume that `fail_stack' is nonempty. */
1516 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1517 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1518 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1520 /* Individual items aside from the registers. */
1521 #define NUM_NONREG_ITEMS 3
1523 /* Used to examine the stack (to detect infinite loops). */
1524 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1525 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1526 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1527 #define TOP_FAILURE_HANDLE() fail_stack.frame
1530 #define ENSURE_FAIL_STACK(space) \
1531 while (REMAINING_AVAIL_SLOTS <= space) { \
1532 if (!GROW_FAIL_STACK (fail_stack)) \
1533 return -2; \
1536 }
1538 /* Push register NUM onto the stack. */
1539 #define PUSH_FAILURE_REG(num) \
1540 do { \
1541 char *destination; \
1542 ENSURE_FAIL_STACK(3); \
1544 num, regstart[num], regend[num]); \
1545 PUSH_FAILURE_POINTER (regstart[num]); \
1546 PUSH_FAILURE_POINTER (regend[num]); \
1547 PUSH_FAILURE_INT (num); \
1548 } while (0)
1550 /* Change the counter's value to VAL, but make sure that it will
1551 be reset when backtracking. */
1552 #define PUSH_NUMBER(ptr,val) \
1553 do { \
1554 char *destination; \
1555 int c; \
1556 ENSURE_FAIL_STACK(3); \
1557 EXTRACT_NUMBER (c, ptr); \
1559 PUSH_FAILURE_INT (c); \
1560 PUSH_FAILURE_POINTER (ptr); \
1561 PUSH_FAILURE_INT (-1); \
1562 STORE_NUMBER (ptr, val); \
1563 } while (0)
1565 /* Pop a saved register off the stack. */
1566 #define POP_FAILURE_REG_OR_COUNT() \
1567 do { \
1568 int reg = POP_FAILURE_INT (); \
1569 if (reg == -1) \
1570 { \
1573 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1574 reg = POP_FAILURE_INT (); \
1575 STORE_NUMBER (ptr, reg); \
1577 } \
1578 else \
1579 { \
1580 regend[reg] = POP_FAILURE_POINTER (); \
1581 regstart[reg] = POP_FAILURE_POINTER (); \
1583 reg, regstart[reg], regend[reg]); \
1584 } \
1585 } while (0)
1587 /* Check that we are not stuck in an infinite loop. */
1588 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1589 do { \
1590 int failure = TOP_FAILURE_HANDLE (); \
1592 while (failure > 0 \
1593 && (FAILURE_STR (failure) == string_place \
1594 || FAILURE_STR (failure) == NULL)) \
1595 { \
1596 assert (FAILURE_PAT (failure) >= bufp->buffer \
1597 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1598 if (FAILURE_PAT (failure) == pat_cur) \
1599 { \
1600 cycle = 1; \
1601 break; \
1602 } \
1604 failure = NEXT_FAILURE_HANDLE(failure); \
1605 } \
1607 } while (0)
1609 /* Push the information about the state we will need
1610 if we ever fail back to it.
1612 Requires variables fail_stack, regstart, regend and
1613 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1614 declared.
1616 Does `return FAILURE_CODE' if runs out of memory. */
1618 #define PUSH_FAILURE_POINT(pattern, string_place) \
1619 do { \
1620 char *destination; \
1621 /* Must be int, so when we don't save any registers, the arithmetic \
1623 \
1624 DEBUG_STATEMENT (nfailure_points_pushed++); \
1628 \
1629 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1630 \
1632 \
1634 PUSH_FAILURE_INT (fail_stack.frame); \
1635 \
1637 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1639 PUSH_FAILURE_POINTER (string_place); \
1640 \
1642 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1643 PUSH_FAILURE_POINTER (pattern); \
1644 \
1646 fail_stack.frame = fail_stack.avail; \
1647 } while (0)
1649 /* Estimate the size of data pushed by a typical failure stack entry.
1650 An estimate is all we need, because all we use this for
1651 is to choose a limit for how big to make the failure stack. */
1652 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1653 #define TYPICAL_FAILURE_SIZE 20
1655 /* How many items can still be added to the stack without overflowing it. */
1656 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1659 /* Pops what PUSH_FAIL_STACK pushes.
1661 We restore into the parameters, all of which should be lvalues:
1662 STR -- the saved data position.
1663 PAT -- the saved pattern position.
1664 REGSTART, REGEND -- arrays of string positions.
1666 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1667 `pend', `string1', `size1', `string2', and `size2'. */
1669 #define POP_FAILURE_POINT(str, pat) \
1670 do { \
1671 assert (!FAIL_STACK_EMPTY ()); \
1672 \
1677 \
1679 while (fail_stack.frame < fail_stack.avail) \
1680 POP_FAILURE_REG_OR_COUNT (); \
1681 \
1682 pat = POP_FAILURE_POINTER (); \
1684 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1685 \
1686 /* If the saved string location is NULL, it came from an \
1687 on_failure_keep_string_jump opcode, and we want to throw away the \
1689 str = POP_FAILURE_POINTER (); \
1691 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1693 \
1694 fail_stack.frame = POP_FAILURE_INT (); \
1696 \
1697 assert (fail_stack.avail >= 0); \
1698 assert (fail_stack.frame <= fail_stack.avail); \
1699 \
1700 DEBUG_STATEMENT (nfailure_points_popped++); \
1704 \f
1705 /* Registers are set to a sentinel when they haven't yet matched. */
1706 #define REG_UNSET(e) ((e) == NULL)
1707 \f
1708 /* Subroutine declarations and macros for regex_compile. */
1711 reg_syntax_t syntax,
1722 reg_syntax_t syntax));
1725 reg_syntax_t syntax));
1730 /* Fetch the next character in the uncompiled pattern, with no
1731 translation. */
1732 #define PATFETCH(c) \
1733 do { \
1734 int len; \
1735 if (p == pend) return REG_EEND; \
1736 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1737 p += len; \
1738 } while (0)
1741 /* If `translate' is non-null, return translate[D], else just D. We
1742 cast the subscript to translate because some data is declared as
1743 `char *', to avoid warnings when a string constant is passed. But
1744 when we use a character as a subscript we must make it unsigned. */
1745 #ifndef TRANSLATE
1746 # define TRANSLATE(d) \
1747 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1748 #endif
1751 /* Macros for outputting the compiled pattern into `buffer'. */
1753 /* If the buffer isn't allocated when it comes in, use this. */
1754 #define INIT_BUF_SIZE 32
1756 /* Make sure we have at least N more bytes of space in buffer. */
1757 #define GET_BUFFER_SPACE(n) \
1758 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1759 EXTEND_BUFFER ()
1761 /* Make sure we have one more byte of buffer space and then add C to it. */
1762 #define BUF_PUSH(c) \
1763 do { \
1764 GET_BUFFER_SPACE (1); \
1765 *b++ = (unsigned char) (c); \
1766 } while (0)
1769 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1770 #define BUF_PUSH_2(c1, c2) \
1771 do { \
1772 GET_BUFFER_SPACE (2); \
1773 *b++ = (unsigned char) (c1); \
1774 *b++ = (unsigned char) (c2); \
1775 } while (0)
1778 /* As with BUF_PUSH_2, except for three bytes. */
1779 #define BUF_PUSH_3(c1, c2, c3) \
1780 do { \
1781 GET_BUFFER_SPACE (3); \
1782 *b++ = (unsigned char) (c1); \
1783 *b++ = (unsigned char) (c2); \
1784 *b++ = (unsigned char) (c3); \
1785 } while (0)
1788 /* Store a jump with opcode OP at LOC to location TO. We store a
1789 relative address offset by the three bytes the jump itself occupies. */
1790 #define STORE_JUMP(op, loc, to) \
1791 store_op1 (op, loc, (to) - (loc) - 3)
1793 /* Likewise, for a two-argument jump. */
1794 #define STORE_JUMP2(op, loc, to, arg) \
1795 store_op2 (op, loc, (to) - (loc) - 3, arg)
1797 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1798 #define INSERT_JUMP(op, loc, to) \
1799 insert_op1 (op, loc, (to) - (loc) - 3, b)
1801 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1802 #define INSERT_JUMP2(op, loc, to, arg) \
1803 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1806 /* This is not an arbitrary limit: the arguments which represent offsets
1807 into the pattern are two bytes long. So if 2^15 bytes turns out to
1808 be too small, many things would have to change. */
1809 # define MAX_BUF_SIZE (1L << 15)
1812 /* Any other compiler which, like MSC, has allocation limit below 2^16
1813 bytes will have to use approach similar to what was done below for
1814 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1815 reallocating to 0 bytes. Such thing is not going to work too well.
1816 You have been warned!! */
1817 #if defined _MSC_VER && !defined WIN32
1818 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1819 # define MAX_BUF_SIZE 65500L
1820 #else
1821 # define MAX_BUF_SIZE (1L << 16)
1822 #endif
1825 /* Extend the buffer by twice its current size via realloc and
1826 reset the pointers that pointed into the old block to point to the
1827 correct places in the new one. If extending the buffer results in it
1828 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1829 #if __BOUNDED_POINTERS__
1830 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1831 # define MOVE_BUFFER_POINTER(P) \
1832 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1833 SET_HIGH_BOUND (P), \
1834 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1835 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1836 else \
1837 { \
1838 SET_HIGH_BOUND (b); \
1839 SET_HIGH_BOUND (begalt); \
1840 if (fixup_alt_jump) \
1841 SET_HIGH_BOUND (fixup_alt_jump); \
1842 if (laststart) \
1843 SET_HIGH_BOUND (laststart); \
1844 if (pending_exact) \
1845 SET_HIGH_BOUND (pending_exact); \
1846 }
1847 #else
1848 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1849 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1850 #endif
1851 #define EXTEND_BUFFER() \
1852 do { \
1853 unsigned char *old_buffer = bufp->buffer; \
1854 if (bufp->allocated == MAX_BUF_SIZE) \
1855 return REG_ESIZE; \
1856 bufp->allocated <<= 1; \
1857 if (bufp->allocated > MAX_BUF_SIZE) \
1858 bufp->allocated = MAX_BUF_SIZE; \
1859 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1860 if (bufp->buffer == NULL) \
1861 return REG_ESPACE; \
1863 if (old_buffer != bufp->buffer) \
1864 { \
1865 unsigned char *new_buffer = bufp->buffer; \
1866 MOVE_BUFFER_POINTER (b); \
1867 MOVE_BUFFER_POINTER (begalt); \
1868 if (fixup_alt_jump) \
1869 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1870 if (laststart) \
1871 MOVE_BUFFER_POINTER (laststart); \
1872 if (pending_exact) \
1873 MOVE_BUFFER_POINTER (pending_exact); \
1874 } \
1875 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1876 } while (0)
1879 /* Since we have one byte reserved for the register number argument to
1880 {start,stop}_memory, the maximum number of groups we can report
1881 things about is what fits in that byte. */
1882 #define MAX_REGNUM 255
1884 /* But patterns can have more than `MAX_REGNUM' registers. We just
1885 ignore the excess. */
1889 /* Macros for the compile stack. */
1891 /* Since offsets can go either forwards or backwards, this type needs to
1892 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1893 /* int may be not enough when sizeof(int) == 2. */
1897 {
1898 pattern_offset_t begalt_offset;
1899 pattern_offset_t fixup_alt_jump;
1900 pattern_offset_t laststart_offset;
1901 regnum_t regnum;
1906 {
1913 #define INIT_COMPILE_STACK_SIZE 32
1915 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1916 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1918 /* The next available element. */
1919 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1921 /* Explicit quit checking is only used on NTemacs and whenever we
1922 use polling to process input events. */
1923 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1925 # define IMMEDIATE_QUIT_CHECK \
1926 do { \
1927 if (immediate_quit) QUIT; \
1928 } while (0)
1929 #else
1930 # define IMMEDIATE_QUIT_CHECK ((void)0)
1931 #endif
1932 \f
1933 /* Structure to manage work area for range table. */
1934 struct range_table_work_area
1935 {
1940 };
1942 /* Make sure that WORK_AREA can hold more N multibyte characters.
1943 This is used only in set_image_of_range and set_image_of_range_1.
1944 It expects WORK_AREA to be a pointer.
1945 If it can't get the space, it returns from the surrounding function. */
1947 #define EXTEND_RANGE_TABLE(work_area, n) \
1948 do { \
1949 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1950 { \
1951 extend_range_table_work_area (&work_area); \
1952 if ((work_area).table == 0) \
1953 return (REG_ESPACE); \
1954 } \
1955 } while (0)
1957 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1958 (work_area).bits |= (bit)
1960 /* Bits used to implement the multibyte-part of the various character classes
1961 such as [:alnum:] in a charset's range table. */
1962 #define BIT_WORD 0x1
1963 #define BIT_LOWER 0x2
1964 #define BIT_PUNCT 0x4
1965 #define BIT_SPACE 0x8
1966 #define BIT_UPPER 0x10
1967 #define BIT_MULTIBYTE 0x20
1969 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1970 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1971 do { \
1972 EXTEND_RANGE_TABLE ((work_area), 2); \
1973 (work_area).table[(work_area).used++] = (range_start); \
1974 (work_area).table[(work_area).used++] = (range_end); \
1975 } while (0)
1977 /* Free allocated memory for WORK_AREA. */
1978 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1979 do { \
1980 if ((work_area).table) \
1981 free ((work_area).table); \
1982 } while (0)
1984 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1985 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1986 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1987 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1988 \f
1990 /* Set the bit for character C in a list. */
1991 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1994 #ifdef emacs
1996 /* Store characters in the range FROM to TO in the bitmap at B (for
1997 ASCII and unibyte characters) and WORK_AREA (for multibyte
1998 characters) while translating them and paying attention to the
1999 continuity of translated characters.
2001 Implementation note: It is better to implement these fairly big
2002 macros by a function, but it's not that easy because macros called
2003 in this macro assume various local variables already declared. */
2005 /* Both FROM and TO are ASCII characters. */
2007 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2008 do { \
2009 int C0, C1; \
2010 \
2011 for (C0 = (FROM); C0 <= (TO); C0++) \
2012 { \
2013 C1 = TRANSLATE (C0); \
2014 if (! ASCII_CHAR_P (C1)) \
2015 { \
2016 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2017 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2018 C1 = C0; \
2019 } \
2020 SET_LIST_BIT (C1); \
2021 } \
2022 } while (0)
2025 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2027 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2028 do { \
2029 int C0, C1, C2, I; \
2030 int USED = RANGE_TABLE_WORK_USED (work_area); \
2031 \
2032 for (C0 = (FROM); C0 <= (TO); C0++) \
2033 { \
2034 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2035 if (CHAR_BYTE8_P (C1)) \
2036 SET_LIST_BIT (C0); \
2037 else \
2038 { \
2039 C2 = TRANSLATE (C1); \
2040 if (C2 == C1 \
2041 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2042 C1 = C0; \
2043 SET_LIST_BIT (C1); \
2044 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2045 { \
2046 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2047 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2048 \
2049 if (C2 >= from - 1 && C2 <= to + 1) \
2050 { \
2051 if (C2 == from - 1) \
2052 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2053 else if (C2 == to + 1) \
2054 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2055 break; \
2056 } \
2057 } \
2058 if (I < USED) \
2059 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2060 } \
2061 } \
2062 } while (0)
2065 /* Both FROM and TO are multibyte characters. */
2067 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2068 do { \
2069 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2070 \
2071 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2072 for (C0 = (FROM); C0 <= (TO); C0++) \
2073 { \
2074 C1 = TRANSLATE (C0); \
2075 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2076 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2077 SET_LIST_BIT (C2); \
2078 if (C1 >= (FROM) && C1 <= (TO)) \
2079 continue; \
2080 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2081 { \
2082 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2083 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2084 \
2085 if (C1 >= from - 1 && C1 <= to + 1) \
2086 { \
2087 if (C1 == from - 1) \
2088 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2089 else if (C1 == to + 1) \
2090 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2091 break; \
2092 } \
2093 } \
2094 if (I < USED) \
2095 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2096 } \
2097 } while (0)
2101 /* Get the next unsigned number in the uncompiled pattern. */
2102 #define GET_UNSIGNED_NUMBER(num) \
2103 do { \
2104 if (p == pend) \
2105 FREE_STACK_RETURN (REG_EBRACE); \
2106 else \
2107 { \
2108 PATFETCH (c); \
2110 { \
2111 int prev; \
2112 if (num < 0) \
2113 num = 0; \
2114 prev = num; \
2116 if (num / 10 != prev) \
2117 FREE_STACK_RETURN (REG_BADBR); \
2118 if (p == pend) \
2119 FREE_STACK_RETURN (REG_EBRACE); \
2120 PATFETCH (c); \
2121 } \
2122 } \
2123 } while (0)
2124 \f
2125 #if ! WIDE_CHAR_SUPPORT
2127 /* Map a string to the char class it names (if any). */
2128 re_wctype_t
2130 {
2150 }
2152 /* True if CH is in the char class CC. */
2153 boolean
2155 {
2157 {
2178 }
2179 }
2181 /* Return a bit-pattern to use in the range-table bits to match multibyte
2182 chars of class CC. */
2183 static int
2185 {
2187 {
2199 }
2200 }
2201 #endif
2202 \f
2203 /* Filling in the work area of a range. */
2205 /* Actually extend the space in WORK_AREA. */
2207 static void
2209 {
2212 work_area->table
2214 else
2215 work_area->table
2217 }
2219 #if 0
2220 #ifdef emacs
2222 /* Carefully find the ranges of codes that are equivalent
2223 under case conversion to the range start..end when passed through
2224 TRANSLATE. Handle the case where non-letters can come in between
2225 two upper-case letters (which happens in Latin-1).
2226 Also handle the case of groups of more than 2 case-equivalent chars.
2228 The basic method is to look at consecutive characters and see
2229 if they can form a run that can be handled as one.
2231 Returns -1 if successful, REG_ESPACE if ran out of space. */
2233 static int
2235 RE_TRANSLATE_TYPE translate;
2238 {
2239 /* `one_case' indicates a character, or a run of characters,
2240 each of which is an isolate (no case-equivalents).
2241 This includes all ASCII non-letters.
2243 `two_case' indicates a character, or a run of characters,
2244 each of which has two case-equivalent forms.
2245 This includes all ASCII letters.
2247 `strange' indicates a character that has more than one
2248 case-equivalent. */
2252 /* Describe the run that is in progress,
2253 which the next character can try to extend.
2254 If run_type is strange, that means there really is no run.
2255 If run_type is one_case, then run_start...run_end is the run.
2256 If run_type is two_case, then the run is run_start...run_end,
2257 and the case-equivalents end at run_eqv_end. */
2262 Lisp_Object eqv_table;
2265 {
2270 }
2275 {
2280 /* Classify this character */
2285 else
2290 else
2293 /* Can this character extend the run in progress? */
2298 {
2299 /* No, end the run.
2300 Record each of its equivalent ranges. */
2302 {
2306 }
2308 {
2316 }
2318 }
2321 {
2322 /* For a strange character, add each of its equivalents, one
2323 by one. Don't start a range. */
2324 do
2325 {
2330 }
2332 }
2334 /* Add this char to the run, or start a new run. */
2336 {
2337 /* Initialize a new range. */
2342 }
2343 else
2344 {
2345 /* Extend a running range. */
2348 }
2349 }
2351 /* If a run is still in progress at the end, finish it now
2352 by recording its equivalent ranges. */
2354 {
2358 }
2360 {
2368 }
2371 }
2375 /* Record the image of the range start..end when passed through
2376 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2377 and is not even necessarily contiguous.
2378 Normally we approximate it with the smallest contiguous range that contains
2379 all the chars we need. However, for Latin-1 we go to extra effort
2380 to do a better job.
2382 This function is not called for ASCII ranges.
2384 Returns -1 if successful, REG_ESPACE if ran out of space. */
2386 static int
2388 RE_TRANSLATE_TYPE translate;
2391 {
2394 #ifdef emacs
2395 /* For Latin-1 ranges, use set_image_of_range_1
2396 to get proper handling of ranges that include letters and nonletters.
2397 For a range that includes the whole of Latin-1, this is not necessary.
2398 For other character sets, we don't bother to get this right. */
2401 {
2414 }
2415 #endif
2424 {
2428 {
2431 {
2434 else
2435 {
2438 }
2439 }
2440 }
2443 {
2447 }
2448 }
2451 }
2453 \f
2454 #ifndef MATCH_MAY_ALLOCATE
2456 /* If we cannot allocate large objects within re_match_2_internal,
2457 we make the fail stack and register vectors global.
2458 The fail stack, we grow to the maximum size when a regexp
2459 is compiled.
2460 The register vectors, we adjust in size each time we
2461 compile a regexp, according to the number of registers it needs. */
2465 /* Size with which the following vectors are currently allocated.
2466 That is so we can make them bigger as needed,
2467 but never make them smaller. */
2473 /* Make the register vectors big enough for NUM_REGS registers,
2474 but don't make them smaller. */
2476 static
2479 {
2481 {
2488 }
2489 }
2492 \f
2494 compile_stack,
2495 regnum_t regnum));
2497 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2498 Returns one of error codes defined in `regex.h', or zero for success.
2500 Assumes the `allocated' (and perhaps `buffer') and `translate'
2501 fields are set in BUFP on entry.
2503 If it succeeds, results are put in BUFP (if it returns an error, the
2504 contents of BUFP are undefined):
2505 `buffer' is the compiled pattern;
2506 `syntax' is set to SYNTAX;
2507 `used' is set to the length of the compiled pattern;
2508 `fastmap_accurate' is zero;
2509 `re_nsub' is the number of subexpressions in PATTERN;
2510 `not_bol' and `not_eol' are zero;
2512 The `fastmap' field is neither examined nor set. */
2514 /* Insert the `jump' from the end of last alternative to "here".
2515 The space for the jump has already been allocated. */
2516 #define FIXUP_ALT_JUMP() \
2517 do { \
2518 if (fixup_alt_jump) \
2519 STORE_JUMP (jump, fixup_alt_jump, b); \
2520 } while (0)
2523 /* Return, freeing storage we allocated. */
2524 #define FREE_STACK_RETURN(value) \
2525 do { \
2526 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2527 free (compile_stack.stack); \
2528 return value; \
2529 } while (0)
2531 static reg_errcode_t
2532 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2533 {
2534 /* We fetch characters from PATTERN here. */
2537 /* A random temporary spot in PATTERN. */
2540 /* Points to the end of the buffer, where we should append. */
2543 /* Keeps track of unclosed groups. */
2544 compile_stack_type compile_stack;
2546 /* Points to the current (ending) position in the pattern. */
2547 #ifdef AIX
2548 /* `const' makes AIX compiler fail. */
2550 #else
2552 #endif
2555 /* How to translate the characters in the pattern. */
2558 /* Address of the count-byte of the most recently inserted `exactn'
2559 command. This makes it possible to tell if a new exact-match
2560 character can be added to that command or if the character requires
2561 a new `exactn' command. */
2564 /* Address of start of the most recently finished expression.
2565 This tells, e.g., postfix * where to find the start of its
2566 operand. Reset at the beginning of groups and alternatives. */
2569 /* Address of beginning of regexp, or inside of last group. */
2572 /* Place in the uncompiled pattern (i.e., the {) to
2573 which to go back if the interval is invalid. */
2576 /* Address of the place where a forward jump should go to the end of
2577 the containing expression. Each alternative of an `or' -- except the
2578 last -- ends with a forward jump of this sort. */
2581 /* Work area for range table of charset. */
2584 /* If the object matched can contain multibyte characters. */
2587 /* If a target of matching can contain multibyte characters. */
2590 /* Nonzero if we have pushed down into a subpattern. */
2593 /* These hold the values of p, pattern, and pend from the main
2594 pattern when we have pushed into a subpattern. */
2599 #ifdef DEBUG
2600 debug++;
2603 {
2609 }
2612 /* Initialize the compile stack. */
2623 /* Initialize the pattern buffer. */
2629 /* Set `used' to zero, so that if we return an error, the pattern
2630 printer (for debugging) will think there's no pattern. We reset it
2631 at the end. */
2634 /* Always count groups, whether or not bufp->no_sub is set. */
2637 #if !defined emacs && !defined SYNTAX_TABLE
2638 /* Initialize the syntax table. */
2640 #endif
2643 {
2646 enough space. This loses if buffer's address is bogus, but
2647 that is the user's responsibility. */
2649 }
2650 else
2653 }
2657 }
2661 /* Loop through the uncompiled pattern until we're at the end. */
2663 {
2665 {
2666 /* If this is the end of an included regexp,
2667 pop back to the main regexp and try again. */
2669 {
2675 }
2676 /* If this is the end of the main regexp, we are done. */
2678 }
2683 {
2685 {
2688 /* If there's no special whitespace regexp, treat
2689 spaces normally. And don't try to do this recursively. */
2693 /* Peek past following spaces. */
2695 {
2698 p1++;
2699 }
2700 /* If the spaces are followed by a repetition op,
2701 treat them normally. */
2707 /* Replace the spaces with the whitespace regexp. */
2715 }
2718 {
2721 /* If context independent, it's an operator. */
2723 /* Otherwise, depends on what's come before. */
2726 else
2728 }
2733 {
2735 p == pend
2736 /* If context independent, it's an operator. */
2738 /* Otherwise, depends on what's next. */
2741 else
2743 }
2752 handle_plus:
2754 /* If there is no previous pattern... */
2756 {
2761 }
2763 {
2764 /* 1 means zero (many) matches is allowed. */
2768 /* If there is a sequence of repetition chars, collapse it
2769 down to just one (the right one). We can't combine
2770 interval operators with these because of, e.g., `a{2}*',
2771 which should only match an even number of `a's. */
2774 {
2778 else
2779 {
2782 }
2789 ;
2791 {
2796 else
2798 }
2799 else
2801 /* If we get here, we found another repeat character. */
2803 }
2805 /* Star, etc. applied to an empty pattern is equivalent
2806 to an empty pattern. */
2810 /* Now we know whether or not zero matches is allowed
2811 and also whether or not two or more matches is allowed. */
2813 {
2815 {
2818 re_opcode_t ofj =
2819 /* Check if the loop can match the empty string. */
2826 on_failure_keep_string_jump, we turn simple P+
2827 into PP* if P is simple. */
2835 }
2839 /* A + loop. */
2841 else
2842 /* Simple * loops can use on_failure_keep_string_jump
2843 depending on what follows. But since we don't know
2844 that yet, we leave the decision up to
2845 on_failure_jump_smart. */
2851 }
2852 else
2853 {
2854 /* A simple ? pattern. */
2859 }
2860 }
2866 {
2869 /* The non-greedy multiple match looks like
2870 a repeat..until: we only need a conditional jump
2871 at the end of the loop. */
2877 {
2878 /* The repeat...until naturally matches one or more.
2879 To also match zero times, we need to first jump to
2880 the end of the loop (its conditional jump). */
2883 }
2884 }
2885 else
2886 {
2887 /* non-greedy a?? */
2892 }
2893 }
2894 }
2906 {
2911 /* Ensure that we have enough space to push a charset: the
2912 opcode, the length count, and the bitset; 34 bytes in all. */
2917 /* We test `*p == '^' twice, instead of using an if
2918 statement, so we only need one BUF_PUSH. */
2921 p++;
2923 /* Remember the first position in the bracket expression. */
2926 /* Push the number of bytes in the bitmap. */
2929 /* Clear the whole map. */
2932 /* charset_not matches newline according to a syntax bit. */
2937 /* Read in characters and ranges, setting map bits. */
2939 {
2946 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2947 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2948 So the translation is done later in a loop. Example:
2949 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2952 /* \ might escape characters inside [...] and [^...]. */
2954 {
2959 }
2960 else
2961 {
2962 /* Could be the end of the bracket expression. If it's
2963 not (i.e., when the bracket expression is `[]' so
2964 far), the ']' character bit gets set way below. */
2967 }
2969 /* See if we're at the beginning of a possible character
2970 class. */
2974 {
2975 /* Leave room for the null. */
2983 /* If pattern is `[[:'. */
2987 {
2993 else
2994 /* This is in any case an invalid class name. */
2996 }
2999 /* If isn't a word bracketed by `[:' and `:]':
3000 undo the ending character, the letters, and
3001 leave the leading `:' and `[' (but set bits for
3002 them). */
3004 {
3005 re_wctype_t cc;
3013 /* Throw away the ] at the end of the character
3014 class. */
3019 #ifndef emacs
3022 {
3026 }
3028 /* Most character classes in a multibyte match
3029 just set a flag. Exceptions are is_blank,
3030 is_digit, is_cntrl, and is_xdigit, since
3031 they can only match ASCII characters. We
3032 don't need to handle them for multibyte.
3033 They are distinguished by a negative wctype. */
3035 /* Setup the gl_state object to its buffer-defined
3036 value. This hardcodes the buffer-global
3037 syntax-table for ASCII chars, while the other chars
3038 will obey syntax-table properties. It's not ideal,
3039 but it's the way it's been done until now. */
3043 {
3047 {
3056 }
3057 }
3058 SET_RANGE_TABLE_WORK_AREA_BIT
3061 /* In most cases the matching rule for char classes
3062 only uses the syntax table for multibyte chars,
3063 so that the content of the syntax-table it is not
3064 hardcoded in the range_table. SPACE and WORD are
3065 the two exceptions. */
3069 /* Repeat the loop. */
3071 }
3072 else
3073 {
3074 /* Go back to right after the "[:". */
3078 /* Because the `:' may starts the range, we
3079 can't simply set bit and repeat the loop.
3080 Instead, just set it to C and handle below. */
3082 }
3083 }
3086 {
3088 /* Discard the `-'. */
3091 /* Fetch the character which ends the range. */
3093 #ifdef emacs
3096 /* Treat the range from a multibyte character to
3097 raw-byte character as empty. */
3100 }
3101 else
3102 /* Range from C to C. */
3106 {
3109 /* Else, repeat the loop. */
3110 }
3111 else
3112 {
3113 #ifndef emacs
3114 /* Set the range into bitmap */
3116 {
3120 }
3123 {
3129 }
3131 {
3133 {
3138 }
3140 {
3142 }
3143 else
3144 {
3146 }
3147 }
3149 }
3150 }
3152 /* Discard any (non)matching list bytes that are all 0 at the
3153 end of the map. Decrease the map-length byte too. */
3158 /* Build real range table from work area. */
3161 {
3165 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3166 bytes for flags, two for COUNT, and three bytes for
3167 each character. */
3170 /* Indicate the existence of range table. */
3173 /* Store the character class flag bits into the range table.
3174 If not in emacs, these flag bits are always 0. */
3180 STORE_CHARACTER_AND_INCR
3182 }
3183 }
3190 else
3197 else
3204 else
3211 else
3218 else
3225 /* Do not translate the character after the \, so that we can
3226 distinguish, e.g., \B from \b, even if we normally would
3227 translate, e.g., B to b. */
3231 {
3236 handle_open:
3237 {
3241 {
3242 /* Look for a special (?...) construct */
3244 {
3247 {
3250 {
3253 /* An explicitly specified regnum must start
3254 with non-0. */
3261 /* Only (?:...) is supported right now. */
3263 }
3264 }
3265 }
3266 }
3276 /* Ideally, we'd want to check that the specified
3277 group can't have matched (i.e. all subgroups
3278 using the same regnum are in other branches of
3279 OR patterns), but we don't currently keep track
3280 of enough info to do that easily. */
3283 }
3284 else
3285 /* It's really shy. */
3289 {
3291 compile_stack_elt_t);
3295 }
3297 /* These are the values to restore when we hit end of this
3298 group. They are all relative offsets, so that if the
3299 whole pattern moves because of realloc, they will still
3300 be valid. */
3302 COMPILE_STACK_TOP.fixup_alt_jump
3307 /* Do not push a start_memory for groups beyond the last one
3308 we can represent in the compiled pattern. */
3317 /* If we've reached MAX_REGNUM groups, then this open
3318 won't actually generate any code, so we'll have to
3319 clear pending_exact explicitly. */
3322 }
3328 {
3331 else
3333 }
3335 handle_close:
3338 /* See similar code for backslashed left paren above. */
3340 {
3343 else
3345 }
3347 /* Since we just checked for an empty stack above, this
3348 ``can't happen''. */
3350 {
3351 /* We don't just want to restore into `regnum', because
3352 later groups should continue to be numbered higher,
3353 as in `(ab)c(de)' -- the second group is #2. */
3354 regnum_t regnum;
3358 fixup_alt_jump
3364 /* If we've reached MAX_REGNUM groups, then this open
3365 won't actually generate any code, so we'll have to
3366 clear pending_exact explicitly. */
3369 /* We're at the end of the group, so now we know how many
3370 groups were inside this one. */
3373 }
3380 handle_alt:
3384 /* Insert before the previous alternative a jump which
3385 jumps to this alternative if the former fails. */
3391 /* The alternative before this one has a jump after it
3392 which gets executed if it gets matched. Adjust that
3393 jump so it will jump to this alternative's analogous
3394 jump (put in below, which in turn will jump to the next
3395 (if any) alternative's such jump, etc.). The last such
3396 jump jumps to the correct final destination. A picture:
3397 _____ _____
3398 | | | |
3399 | v | v
3400 a | b | c
3402 If we are at `b', then fixup_alt_jump right now points to a
3403 three-byte space after `a'. We'll put in the jump, set
3404 fixup_alt_jump to right after `b', and leave behind three
3405 bytes which we'll fill in when we get to after `c'. */
3409 /* Mark and leave space for a jump after this alternative,
3410 to be filled in later either by next alternative or
3411 when know we're at the end of a series of alternatives. */
3422 /* If \{ is a literal. */
3424 /* If we're at `\{' and it's not the open-interval
3425 operator. */
3429 handle_interval:
3430 {
3431 /* If got here, then the syntax allows intervals. */
3433 /* At least (most) this many matches must be made. */
3442 else
3443 /* Interval such as `{1}' => match exactly once. */
3451 {
3457 }
3462 /* We just parsed a valid interval. */
3464 /* If it's invalid to have no preceding re. */
3466 {
3471 else
3473 }
3476 /* If the upper bound is zero, just drop the sub pattern
3477 altogether. */
3480 /* Just match it once: nothing to do here. */
3481 ;
3483 /* Otherwise, we have a nontrivial interval. When
3484 we're all done, the pattern will look like:
3485 set_number_at <jump count> <upper bound>
3486 set_number_at <succeed_n count> <lower bound>
3487 succeed_n <after jump addr> <succeed_n count>
3488 <body of loop>
3489 jump_n <succeed_n addr> <jump count>
3490 (The upper bound and `jump_n' are omitted if
3491 `upper_bound' is 1, though.) */
3492 else
3494 more at the end of the loop. */
3502 {
3503 /* A succeed_n that starts with 0 is really a
3504 a simple on_failure_jump_loop. */
3508 }
3509 else
3510 {
3511 /* Initialize lower bound of the `succeed_n', even
3512 though it will be set during matching by its
3513 attendant `set_number_at' (inserted next),
3514 because `re_compile_fastmap' needs to know.
3515 Jump to the `jump_n' we might insert below. */
3518 lower_bound);
3521 /* Code to initialize the lower bound. Insert
3522 before the `succeed_n'. The `5' is the last two
3523 bytes of this `set_number_at', plus 3 bytes of
3524 the following `succeed_n'. */
3528 }
3531 {
3532 /* A negative upper bound stands for infinity,
3533 in which case it degenerates to a plain jump. */
3536 }
3539 append a backward jump to the `succeed_n'
3540 that starts this interval.
3542 When we've reached this during matching,
3543 we'll have matched the interval once, so
3544 jump back only `upper_bound - 1' times. */
3549 /* The location we want to set is the second
3550 parameter of the `jump_n'; that is `b-2' as
3551 an absolute address. `laststart' will be
3552 the `set_number_at' we're about to insert;
3553 `laststart+3' the number to set, the source
3554 for the relative address. But we are
3555 inserting into the middle of the pattern --
3556 so everything is getting moved up by 5.
3557 Conclusion: (b - 2) - (laststart + 3) + 5,
3558 i.e., b - laststart.
3560 We insert this at the beginning of the loop
3561 so that if we fail during matching, we'll
3562 reinitialize the bounds. */
3566 }
3567 }
3570 }
3573 unfetch_interval:
3574 /* If an invalid interval, match the characters as literals. */
3579 /* normal_char and normal_backslash need `c'. */
3583 {
3586 }
3587 else
3590 #ifdef emacs
3591 /* There is no way to specify the before_dot and after_dot
3592 operators. rms says this is ok. --karl */
3660 else
3690 {
3691 regnum_t reg;
3699 /* Can't back reference to a subexp before its end. */
3705 }
3713 else
3717 normal_backslash:
3718 /* You might think it would be useful for \ to mean
3719 not to translate; but if we don't translate it
3720 it will never match anything. */
3722 }
3727 /* Expects the character in `c'. */
3728 normal_char:
3729 /* If no exactn currently being built. */
3732 /* If last exactn not at current position. */
3735 /* We have only one byte following the exactn for the count. */
3738 /* If followed by a repetition operator. */
3747 {
3748 /* Start building a new exactn. */
3754 }
3757 {
3761 {
3765 }
3766 else
3767 {
3770 {
3775 }
3778 }
3780 }
3787 /* Through the pattern now. */
3794 /* If we don't want backtracking, force success
3795 the first time we reach the end of the compiled pattern. */
3799 /* We have succeeded; set the length of the buffer. */
3802 #ifdef DEBUG
3804 {
3808 }
3809 debug--;
3812 #ifndef MATCH_MAY_ALLOCATE
3813 /* Initialize the failure stack to the largest possible stack. This
3814 isn't necessary unless we're trying to avoid calling alloca in
3815 the search and match routines. */
3816 {
3820 {
3824 fail_stack.stack
3827 else
3828 fail_stack.stack
3832 }
3835 }
3840 \f
3841 /* Subroutines for `regex_compile'. */
3843 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3845 static void
3847 {
3850 }
3853 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3855 static void
3857 {
3861 }
3864 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3865 for OP followed by two-byte integer parameter ARG. */
3867 static void
3869 {
3877 }
3880 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3882 static void
3884 {
3892 }
3895 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3896 after an alternative or a begin-subexpression. We assume there is at
3897 least one character before the ^. */
3899 static boolean
3901 {
3905 return
3906 /* After a subexpression? */
3908 /* After an alternative? */
3910 /* After a shy subexpression? */
3915 }
3918 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3919 at least one character after the $, i.e., `P < PEND'. */
3921 static boolean
3923 {
3928 return
3929 /* Before a subexpression? */
3932 /* Before an alternative? */
3935 }
3938 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3939 false if it's not. */
3941 static boolean
3943 {
3948 this_element--)
3953 }
3954 \f
3955 /* analyse_first.
3956 If fastmap is non-NULL, go through the pattern and fill fastmap
3957 with all the possible leading chars. If fastmap is NULL, don't
3958 bother filling it up (obviously) and only return whether the
3959 pattern could potentially match the empty string.
3961 Return 1 if p..pend might match the empty string.
3962 Return 0 if p..pend matches at least one char.
3963 Return -1 if fastmap was not updated accurately. */
3965 static int
3967 {
3971 /* If all elements for base leading-codes in fastmap is set, this
3972 flag is set true. */
3977 /* The loop below works as follows:
3978 - It has a working-list kept in the PATTERN_STACK and which basically
3979 starts by only containing a pointer to the first operation.
3980 - If the opcode we're looking at is a match against some set of
3981 chars, then we add those chars to the fastmap and go on to the
3982 next work element from the worklist (done via `break').
3983 - If the opcode is a control operator on the other hand, we either
3984 ignore it (if it's meaningless at this point, such as `start_memory')
3985 or execute it (if it's a jump). If the jump has several destinations
3986 (i.e. `on_failure_jump'), then we push the other destination onto the
3987 worklist.
3988 We guarantee termination by ignoring backward jumps (more or less),
3989 so that `p' is monotonically increasing. More to the point, we
3990 never set `p' (or push) anything `<= p1'. */
3993 {
3994 /* `p1' is used as a marker of how far back a `on_failure_jump'
3995 can go without being ignored. It is normally equal to `p'
3996 (which prevents any backward `on_failure_jump') except right
3997 after a plain `jump', to allow patterns such as:
3998 0: jump 10
3999 3..9: <body>
4000 10: on_failure_jump 3
4001 as used for the *? operator. */
4005 {
4011 /* If the first character has to match a backreference, that means
4012 that the group was empty (since it already matched). Since this
4013 is the only case that interests us here, we can assume that the
4014 backreference must match the empty string. */
4015 p++;
4019 /* Following are the cases which match a character. These end
4020 with `break'. */
4024 {
4025 /* If multibyte is nonzero, the first byte of each
4026 character is an ASCII or a leading code. Otherwise,
4027 each byte is a character. Thus, this works in both
4028 cases. */
4031 {
4032 /* For the case of matching this unibyte regex
4033 against multibyte, we must set a leading code of
4034 the corresponding multibyte character. */
4038 }
4039 }
4044 /* We could put all the chars except for \n (and maybe \0)
4045 but we don't bother since it is generally not worth it. */
4052 {
4053 /* Chars beyond end of bitmap are possible matches. */
4057 }
4059 /* Fallthrough */
4068 #ifdef emacs
4070 which doesn't match the specified set of characters. */
4071 not
4072 ||
4073 /* If we can match a character class, we can match any
4074 multibyte characters. */
4078 {
4080 {
4085 }
4086 }
4090 {
4091 /* Set fastmap[I] to 1 where I is a leading code of each
4092 multibyte character in the range table. */
4096 /* Make P points the range table. `+ 2' is to skip flag
4097 bits for a character class. */
4100 /* Extract the number of ranges in range table into COUNT. */
4103 {
4104 /* Extract the start and end of each range. */
4112 }
4113 }
4114 #endif
4120 #ifndef emacs
4128 /* This match depends on text properties. These end with
4129 aborting optimizations. */
4141 /* Any leading code can possibly start a character which
4142 has or doesn't has the specified category. */
4144 {
4149 }
4152 /* All cases after this match the empty string. These end with
4153 `continue'. */
4176 /* Backward jumps can only go back to code that we've already
4177 visited. `re_compile' should make sure this is true. */
4181 {
4187 p++;
4191 };
4192 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4193 to jump back to "just after here". */
4194 /* Fallthrough */
4204 else
4206 We first go down the "straight" path so as to minimize
4207 stack usage when going through alternatives. */
4211 }
4216 /* This code simply does not properly handle forward jump_n. */
4219 /* jump_n can either jump or fall through. The (backward) jump
4220 case has already been handled, so we only need to look at the
4221 fallthrough case. */
4225 /* If N == 0, it should be an on_failure_jump_loop instead. */
4228 /* We only care about one iteration of the loop, so we don't
4229 need to consider the case where this behaves like an
4230 on_failure_jump. */
4249 /* Getting here means we have found the possible starting
4250 characters for one path of the pattern -- and that the empty
4251 string does not match. We need not follow this path further. */
4255 /* We reached the end without matching anything. */
4259 \f
4260 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4261 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4262 characters can start a string that matches the pattern. This fastmap
4263 is used by re_search to skip quickly over impossible starting points.
4265 Character codes above (1 << BYTEWIDTH) are not represented in the
4266 fastmap, but the leading codes are represented. Thus, the fastmap
4267 indicates which character sets could start a match.
4269 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4270 area as BUFP->fastmap.
4272 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4273 the pattern buffer.
4275 Returns 0 if we succeed, -2 if an internal error. */
4277 int
4279 {
4293 \f
4294 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4295 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4296 this memory for recording register information. STARTS and ENDS
4297 must be allocated using the malloc library routine, and must each
4298 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4300 If NUM_REGS == 0, then subsequent matches should allocate their own
4301 register data.
4303 Unless this function is called, the first search or match using
4304 PATTERN_BUFFER will allocate its own register data, without
4305 freeing the old data. */
4307 void
4308 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4309 {
4311 {
4316 }
4317 else
4318 {
4322 }
4323 }
4325 \f
4326 /* Searching routines. */
4328 /* Like re_search_2, below, but only one string is specified, and
4329 doesn't let you say where to stop matching. */
4331 int
4332 re_search (struct re_pattern_buffer *bufp, const char *string, int size, int startpos, int range, struct re_registers *regs)
4333 {
4336 }
4339 /* Head address of virtual concatenation of string. */
4340 #define HEAD_ADDR_VSTRING(P) \
4341 (((P) >= size1 ? string2 : string1))
4343 /* End address of virtual concatenation of string. */
4344 #define STOP_ADDR_VSTRING(P) \
4345 (((P) >= size1 ? string2 + size2 : string1 + size1))
4347 /* Address of POS in the concatenation of virtual string. */
4348 #define POS_ADDR_VSTRING(POS) \
4349 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4351 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4352 virtual concatenation of STRING1 and STRING2, starting first at index
4353 STARTPOS, then at STARTPOS + 1, and so on.
4355 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4357 RANGE is how far to scan while trying to match. RANGE = 0 means try
4358 only at STARTPOS; in general, the last start tried is STARTPOS +
4359 RANGE.
4361 In REGS, return the indices of the virtual concatenation of STRING1
4362 and STRING2 that matched the entire BUFP->buffer and its contained
4363 subexpressions.
4365 Do not consider matching one past the index STOP in the virtual
4366 concatenation of STRING1 and STRING2.
4368 We return either the position in the strings at which the match was
4369 found, -1 if no match, or -2 if error (such as failure
4370 stack overflow). */
4372 int
4373 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)
4374 {
4382 boolean anchored_start;
4383 /* Nonzero if we are searching multibyte string. */
4386 /* Check for out-of-range STARTPOS. */
4390 /* Fix up RANGE if it might eventually take us outside
4391 the virtual concatenation of STRING1 and STRING2.
4392 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4398 /* If the search isn't to be a backwards one, don't waste time in a
4399 search for a pattern anchored at beginning of buffer. */
4401 {
4404 else
4406 }
4408 #ifdef emacs
4409 /* In a forward search for something that starts with \=.
4410 don't keep searching past point. */
4412 {
4416 }
4419 /* Update the fastmap now if not correct already. */
4423 /* See whether the pattern is anchored. */
4426 #ifdef emacs
4428 {
4432 }
4433 #endif
4435 /* Loop through the string, looking for a place to start matching. */
4437 {
4438 /* If the pattern is anchored,
4439 skip quickly past places we cannot match.
4440 We don't bother to treat startpos == 0 specially
4441 because that case doesn't repeat. */
4443 {
4448 }
4450 /* If a fastmap is supplied, skip quickly over characters that
4451 cannot be the start of a match. If the pattern can match the
4452 null string, however, we don't need to skip characters; we want
4453 the first null string. */
4455 {
4462 {
4469 /* Written out as an if-else to avoid testing `translate'
4470 inside the loop. */
4472 {
4475 {
4485 }
4486 else
4488 {
4499 d++;
4500 range--;
4501 }
4502 }
4503 else
4504 {
4507 {
4515 }
4516 else
4518 {
4519 d++;
4520 range--;
4521 }
4522 }
4524 }
4526 {
4528 {
4533 }
4534 else
4535 {
4546 }
4547 }
4548 }
4550 /* If can't match the null string, and that's all we have left, fail. */
4564 advance:
4568 {
4569 /* Update STARTPOS to the next character boundary. */
4571 {
4580 }
4581 else
4582 {
4583 range--;
4584 startpos++;
4585 }
4586 }
4587 else
4588 {
4589 range++;
4590 startpos--;
4592 /* Update STARTPOS to the previous character boundary. */
4594 {
4599 /* Find the head of multibyte form. */
4606 }
4607 }
4608 }
4612 \f
4613 /* Declarations and macros for re_match_2. */
4617 RE_TRANSLATE_TYPE translate,
4620 /* This converts PTR, a pointer into one of the search strings `string1'
4621 and `string2' into an offset from the beginning of that string. */
4622 #define POINTER_TO_OFFSET(ptr) \
4623 (FIRST_STRING_P (ptr) \
4624 ? ((regoff_t) ((ptr) - string1)) \
4625 : ((regoff_t) ((ptr) - string2 + size1)))
4627 /* Call before fetching a character with *d. This switches over to
4628 string2 if necessary.
4629 Check re_match_2_internal for a discussion of why end_match_2 might
4630 not be within string2 (but be equal to end_match_1 instead). */
4631 #define PREFETCH() \
4632 while (d == dend) \
4633 { \
4635 if (dend == end_match_2) \
4636 goto fail; \
4638 d = string2; \
4639 dend = end_match_2; \
4640 }
4642 /* Call before fetching a char with *d if you already checked other limits.
4643 This is meant for use in lookahead operations like wordend, etc..
4644 where we might need to look at parts of the string that might be
4645 outside of the LIMITs (i.e past `stop'). */
4646 #define PREFETCH_NOLIMIT() \
4647 if (d == end1) \
4648 { \
4649 d = string2; \
4650 dend = end_match_2; \
4651 } \
4653 /* Test if at very beginning or at very end of the virtual concatenation
4654 of `string1' and `string2'. If only one string, it's `string2'. */
4655 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4656 #define AT_STRINGS_END(d) ((d) == end2)
4659 /* Test if D points to a character which is word-constituent. We have
4660 two special cases to check for: if past the end of string1, look at
4661 the first character in string2; and if before the beginning of
4662 string2, look at the last character in string1. */
4663 #define WORDCHAR_P(d) \
4664 (SYNTAX ((d) == end1 ? *string2 \
4665 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4666 == Sword)
4668 /* Disabled due to a compiler bug -- see comment at case wordbound */
4670 /* The comment at case wordbound is following one, but we don't use
4671 AT_WORD_BOUNDARY anymore to support multibyte form.
4673 The DEC Alpha C compiler 3.x generates incorrect code for the
4674 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4675 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4676 macro and introducing temporary variables works around the bug. */
4678 #if 0
4679 /* Test if the character before D and the one at D differ with respect
4680 to being word-constituent. */
4681 #define AT_WORD_BOUNDARY(d) \
4682 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4683 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4684 #endif
4686 /* Free everything we malloc. */
4687 #ifdef MATCH_MAY_ALLOCATE
4688 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4689 # define FREE_VARIABLES() \
4690 do { \
4691 REGEX_FREE_STACK (fail_stack.stack); \
4692 FREE_VAR (regstart); \
4693 FREE_VAR (regend); \
4694 FREE_VAR (best_regstart); \
4695 FREE_VAR (best_regend); \
4696 } while (0)
4697 #else
4701 \f
4702 /* Optimization routines. */
4704 /* If the operation is a match against one or more chars,
4705 return a pointer to the next operation, else return NULL. */
4708 {
4710 {
4721 {
4726 }
4727 else
4733 #ifdef emacs
4737 p++;
4742 }
4744 }
4747 /* Jump over non-matching operations. */
4750 {
4753 {
4755 {
4768 }
4769 }
4772 }
4774 /* Non-zero if "p1 matches something" implies "p2 fails". */
4775 static int
4777 {
4778 re_opcode_t op2;
4785 /* Skip over open/close-group commands.
4786 If what follows this loop is a ...+ construct,
4787 look at what begins its body, since we will have to
4788 match at least one of that. */
4790 /* The same skip can be done for p1, except that this function
4791 is only used in the case where p1 is a simple match operator. */
4792 /* p1 = skip_noops (p1, pend); */
4800 {
4803 /* If we're at the end of the pattern, we can change. */
4805 {
4808 }
4813 {
4814 register re_wchar_t c
4819 {
4821 {
4824 }
4825 }
4829 {
4832 /* Test if C is listed in charset (or charset_not)
4833 at `p1'. */
4835 {
4839 }
4843 /* `not' is equal to 1 if c would match, which means
4844 that we can't change to pop_failure_jump. */
4846 {
4849 }
4850 }
4853 {
4856 }
4857 }
4861 {
4863 /* Reuse the code above. */
4866 /* It is hard to list up all the character in charset
4867 P2 if it includes multibyte character. Give up in
4868 such case. */
4870 {
4871 /* Now, we are sure that P2 has no range table.
4872 So, for the size of bitmap in P2, `p2[1]' is
4873 enough. But P1 may have range table, so the
4874 size of bitmap table of P1 is extracted by
4875 using macro `CHARSET_BITMAP_SIZE'.
4877 In a multibyte case, we know that all the character
4878 listed in P2 is ASCII. In a unibyte case, P1 has only a
4879 bitmap table. So, in both cases, it is enough to test
4880 only the bitmap table of P1. */
4883 {
4885 /* We win if the charset inside the loop
4886 has no overlap with the one after the loop. */
4890 idx++)
4896 {
4899 }
4900 }
4902 {
4904 /* We win if the charset_not inside the loop lists
4905 every character listed in the charset after. */
4913 {
4916 }
4917 }
4918 }
4919 }
4924 {
4927 /* Reuse the code above. */
4930 /* When we have two charset_not, it's very unlikely that
4931 they don't overlap. The union of the two sets of excluded
4932 chars should cover all possible chars, which, as a matter of
4933 fact, is virtually impossible in multibyte buffers. */
4935 }
4959 #ifdef emacs
4967 ;
4968 }
4970 /* Safe default. */
4972 }
4974 \f
4975 /* Matching routines. */
4978 /* re_match is like re_match_2 except it takes only a single string. */
4980 int
4986 {
4990 }
4994 #ifdef emacs
4995 /* In Emacs, this is the string or buffer in which we
4996 are matching. It is used for looking up syntax properties. */
4997 Lisp_Object re_match_object;
4998 #endif
5000 /* re_match_2 matches the compiled pattern in BUFP against the
5001 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5002 and SIZE2, respectively). We start matching at POS, and stop
5003 matching at STOP.
5005 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5006 store offsets for the substring each group matched in REGS. See the
5007 documentation for exactly how many groups we fill.
5009 We return -1 if no match, -2 if an internal error (such as the
5010 failure stack overflowing). Otherwise, we return the length of the
5011 matched substring. */
5013 int
5014 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)
5015 {
5018 #ifdef emacs
5023 #endif
5029 }
5033 /* This is a separate function so that we can force an alloca cleanup
5034 afterwards. */
5035 static int
5036 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)
5037 {
5038 /* General temporaries. */
5043 /* Just past the end of the corresponding string. */
5046 /* Pointers into string1 and string2, just past the last characters in
5047 each to consider matching. */
5050 /* Where we are in the data, and the end of the current string. */
5053 /* Used sometimes to remember where we were before starting matching
5054 an operator so that we can go back in case of failure. This "atomic"
5055 behavior of matching opcodes is indispensable to the correctness
5056 of the on_failure_keep_string_jump optimization. */
5059 /* Where we are in the pattern, and the end of the pattern. */
5063 /* We use this to map every character in the string. */
5066 /* Nonzero if BUFP is setup from a multibyte regex. */
5069 /* Nonzero if STRING1/STRING2 are multibyte. */
5072 /* Failure point stack. Each place that can handle a failure further
5073 down the line pushes a failure point on this stack. It consists of
5074 regstart, and regend for all registers corresponding to
5075 the subexpressions we're currently inside, plus the number of such
5076 registers, and, finally, two char *'s. The first char * is where
5077 to resume scanning the pattern; the second one is where to resume
5078 scanning the strings. */
5080 fail_stack_type fail_stack;
5081 #endif
5082 #ifdef DEBUG
5084 #endif
5086 #if defined REL_ALLOC && defined REGEX_MALLOC
5087 /* This holds the pointer to the failure stack, when
5088 it is allocated relocatably. */
5090 #endif
5092 /* We fill all the registers internally, independent of what we
5093 return, for use in backreferences. The number here includes
5094 an element for register zero. */
5097 /* Information on the contents of registers. These are pointers into
5098 the input strings; they record just what was matched (on this
5099 attempt) by a subexpression part of the pattern, that is, the
5100 regnum-th regstart pointer points to where in the pattern we began
5101 matching and the regnum-th regend points to right after where we
5102 stopped matching the regnum-th subexpression. (The zeroth register
5103 keeps track of what the whole pattern matches.) */
5106 #endif
5108 /* The following record the register info as found in the above
5109 variables when we find a match better than any we've seen before.
5110 This happens as we backtrack through the failure points, which in
5111 turn happens only if we have not yet matched the entire string. */
5115 #endif
5117 /* Logically, this is `best_regend[0]'. But we don't want to have to
5118 allocate space for that if we're not allocating space for anything
5119 else (see below). Also, we never need info about register 0 for
5120 any of the other register vectors, and it seems rather a kludge to
5121 treat `best_regend' differently than the rest. So we keep track of
5122 the end of the best match so far in a separate variable. We
5123 initialize this to NULL so that when we backtrack the first time
5124 and need to test it, it's not garbage. */
5127 #ifdef DEBUG
5128 /* Counts the total number of registers pushed. */
5130 #endif
5136 #ifdef MATCH_MAY_ALLOCATE
5137 /* Do not bother to initialize all the register variables if there are
5138 no groups in the pattern, as it takes a fair amount of time. If
5139 there are groups, we include space for register 0 (the whole
5140 pattern), even though we never use it, since it simplifies the
5141 array indexing. We should fix this. */
5143 {
5150 {
5153 }
5154 }
5155 else
5156 {
5157 /* We must initialize all our variables to NULL, so that
5158 `FREE_VARIABLES' doesn't try to free them. */
5160 }
5163 /* The starting position is bogus. */
5165 {
5168 }
5170 /* Initialize subexpression text positions to -1 to mark ones that no
5171 start_memory/stop_memory has been seen for. Also initialize the
5172 register information struct. */
5176 /* We move `string1' into `string2' if the latter's empty -- but not if
5177 `string1' is null. */
5179 {
5184 }
5188 /* `p' scans through the pattern as `d' scans through the data.
5189 `dend' is the end of the input string that `d' points within. `d'
5190 is advanced into the following input string whenever necessary, but
5191 this happens before fetching; therefore, at the beginning of the
5192 loop, `d' can be pointing at the end of a string, but it cannot
5193 equal `string2'. */
5195 {
5196 /* Only match within string2. */
5200 }
5201 else
5202 {
5204 {
5205 /* Only match within string1. */
5207 /* BEWARE!
5208 When we reach end_match_1, PREFETCH normally switches to string2.
5209 But in the present case, this means that just doing a PREFETCH
5210 makes us jump from `stop' to `gap' within the string.
5211 What we really want here is for the search to stop as
5212 soon as we hit end_match_1. That's why we set end_match_2
5213 to end_match_1 (since PREFETCH fails as soon as we hit
5214 end_match_2). */
5216 }
5217 else
5219 moving `d' from end1 to string2 will not prevent the d == dend
5220 check from catching the end of string. */
5223 }
5226 }
5234 /* This loops over pattern commands. It exits by returning from the
5235 function if the match is complete, or it drops through if the match
5236 fails at this starting point in the input data. */
5238 {
5245 /* If we haven't matched the entire string, and we want the
5246 longest match, try backtracking. */
5248 {
5249 /* 1 if this match ends in the same string (string1 or string2)
5250 as the best previous match. */
5253 /* 1 if this match is the best seen so far. */
5254 boolean best_match_p;
5256 /* AIX compiler got confused when this was combined
5257 with the previous declaration. */
5260 else
5268 /* If exceeds best match so far, save it. */
5270 {
5277 {
5280 }
5281 }
5283 }
5285 /* If no failure points, don't restore garbage. And if
5286 last match is real best match, don't restore second
5287 best one. */
5289 {
5290 restore_best_regs:
5291 /* Restore best match. It may happen that `dend ==
5292 end_match_1' while the restored d is in string2.
5293 For example, the pattern `x.*y.*z' against the
5294 strings `x-' and `y-z-', if the two strings are
5295 not consecutive in memory. */
5303 {
5306 }
5307 }
5310 succeed_label:
5313 /* If caller wants register contents data back, do it. */
5315 {
5316 /* Have the register data arrays been allocated? */
5319 extra element beyond `num_regs' for the `-1' marker
5320 GNU code uses. */
5325 {
5328 }
5330 }
5333 allocated, reallocate them. If we need fewer, just
5334 leave it alone. */
5336 {
5341 {
5344 }
5345 }
5346 }
5347 else
5348 {
5349 /* These braces fend off a "empty body in an else-statement"
5350 warning under GCC when assert expands to nothing. */
5352 }
5354 /* Convert the pointer data in `regstart' and `regend' to
5355 indices. Register zero has to be set differently,
5356 since we haven't kept track of any info for it. */
5358 {
5361 }
5363 /* Go through the first `min (num_regs, regs->num_regs)'
5364 registers, since that is all we initialized. */
5366 {
5369 else
5370 {
5375 }
5376 }
5378 /* If the regs structure we return has more elements than
5379 were in the pattern, set the extra elements to -1. If
5380 we (re)allocated the registers, this is the case,
5381 because we always allocate enough to have at least one
5382 -1 at the end. */
5398 }
5400 /* Otherwise match next pattern command. */
5402 {
5403 /* Ignore these. Used to ignore the n of succeed_n's which
5404 currently have n == 0. */
5413 /* Match the next n pattern characters exactly. The following
5414 byte in the pattern defines n, and the n bytes after that
5415 are the characters to match. */
5420 /* Remember the start point to rollback upon failure. */
5423 #ifndef emacs
5424 /* This is written out as an if-else so we don't waste time
5425 testing `translate' inside the loop. */
5427 do
5428 {
5431 {
5434 }
5435 d++;
5436 }
5438 else
5439 do
5440 {
5443 {
5446 }
5447 }
5450 /* The cost of testing `translate' is comparatively small. */
5452 do
5453 {
5460 else
5461 {
5464 }
5468 {
5471 }
5476 }
5478 else
5479 do
5480 {
5486 {
5489 }
5490 else
5491 {
5494 }
5497 {
5502 }
5503 else
5506 {
5509 }
5511 d++;
5512 }
5514 #endif
5518 /* Match any character except possibly a newline or a null. */
5520 {
5522 re_wchar_t buf_ch;
5528 target_multibyte);
5539 }
5545 {
5550 /* Start of actual range_table, or end of bitmap if there is no
5551 range table. */
5554 /* Nonzero if there is a range table. */
5557 /* Number of ranges of range table. This is not included
5558 in the initial byte-length of the command. */
5561 /* Whether matching against a unibyte character. */
5569 {
5572 }
5577 {
5583 {
5586 }
5587 }
5588 else
5589 {
5593 {
5597 {
5600 }
5601 }
5602 else
5604 }
5608 /* Cast to `unsigned' instead of `unsigned char' in
5609 case the bit list is a full 32 bytes long. */
5613 }
5614 #ifdef emacs
5616 {
5626 else
5628 }
5633 else
5640 }
5643 /* The beginning of a group is represented by start_memory.
5644 The argument is the register number. The text
5645 matched within the group is recorded (in the internal
5646 registers data structure) under the register number. */
5650 /* In case we need to undo this operation (via backtracking). */
5657 /* Move past the register number and inner group count. */
5662 /* The stop_memory opcode represents the end of a group. Its
5663 argument is the same as start_memory's: the register number. */
5668 /* Strictly speaking, there should be code such as:
5670 assert (REG_UNSET (regend[*p]));
5671 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5673 But the only info to be pushed is regend[*p] and it is known to
5674 be UNSET, so there really isn't anything to push.
5675 Not pushing anything, on the other hand deprives us from the
5676 guarantee that regend[*p] is UNSET since undoing this operation
5677 will not reset its value properly. This is not important since
5678 the value will only be read on the next start_memory or at
5679 the very end and both events can only happen if this stop_memory
5680 is *not* undone. */
5685 /* Move past the register number and the inner group count. */
5690 /* \<digit> has been turned into a `duplicate' command which is
5691 followed by the numeric value of <digit> as the register number. */
5693 {
5698 /* Can't back reference a group which we've never matched. */
5702 /* Where in input to try to start matching. */
5705 /* Remember the start point to rollback upon failure. */
5708 /* Where to stop matching; if both the place to start and
5709 the place to stop matching are in the same string, then
5710 set to the place to stop, otherwise, for now have to use
5711 the end of the first string. */
5717 {
5718 /* If necessary, advance to next segment in register
5719 contents. */
5721 {
5725 /* End of string1 => advance to string2. */
5728 }
5729 /* At end of register contents => success */
5732 /* If necessary, advance to next segment in data. */
5735 /* How many characters left in this segment to match. */
5738 /* Want how many consecutive characters we can match in
5739 one shot, so, if necessary, adjust the count. */
5743 /* Compare that many; failure if mismatch, else move
5744 past them. */
5748 {
5751 }
5753 }
5754 }
5758 /* begline matches the empty string at the beginning of the string
5759 (unless `not_bol' is set in `bufp'), and after newlines. */
5764 {
5766 }
5767 else
5768 {
5773 }
5774 /* In all other cases, we fail. */
5778 /* endline is the dual of begline. */
5783 {
5785 }
5786 else
5787 {
5791 }
5795 /* Match at the very beginning of the data. */
5803 /* Match at the very end of the data. */
5811 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5812 pushes NULL as the value for the string on the stack. Then
5813 `POP_FAILURE_POINT' will keep the current value for the
5814 string, instead of restoring it. To see why, consider
5815 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5816 then the . fails against the \n. But the next thing we want
5817 to do is match the \n against the \n; if we restored the
5818 string value, we would be back at the foo.
5820 Because this is used only in specific cases, we don't need to
5821 check all the things that `on_failure_jump' does, to make
5822 sure the right things get saved on the stack. Hence we don't
5823 share its code. The only reason to push anything on the
5824 stack at all is that otherwise we would have to change
5825 `anychar's code to do something besides goto fail in this
5826 case; that seems worse than this. */
5835 /* A nasty loop is introduced by the non-greedy *? and +?.
5836 With such loops, the stack only ever contains one failure point
5837 at a time, so that a plain on_failure_jump_loop kind of
5838 cycle detection cannot work. Worse yet, such a detection
5839 can not only fail to detect a cycle, but it can also wrongly
5840 detect a cycle (between different instantiations of the same
5841 loop).
5842 So the method used for those nasty loops is a little different:
5843 We use a special cycle-detection-stack-frame which is pushed
5844 when the on_failure_jump_nastyloop failure-point is *popped*.
5845 This special frame thus marks the beginning of one iteration
5846 through the loop and we can hence easily check right here
5847 whether something matched between the beginning and the end of
5848 the loop. */
5855 {
5859 /* If there's a cycle, just continue without pushing
5860 this failure point. The failure point is the "try again"
5861 option, which shouldn't be tried.
5862 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5864 }
5867 /* Simple loop detecting on_failure_jump: just check on the
5868 failure stack if the same spot was already hit earlier. */
5870 on_failure:
5874 {
5878 /* If there's a cycle, get out of the loop, as if the matching
5879 had failed. We used to just `goto fail' here, but that was
5880 aborting the search a bit too early: we want to keep the
5881 empty-loop-match and keep matching after the loop.
5882 We want (x?)*y\1z to match both xxyz and xxyxz. */
5884 else
5886 }
5890 /* Uses of on_failure_jump:
5892 Each alternative starts with an on_failure_jump that points
5893 to the beginning of the next alternative. Each alternative
5894 except the last ends with a jump that in effect jumps past
5895 the rest of the alternatives. (They really jump to the
5896 ending jump of the following alternative, because tensioning
5897 these jumps is a hassle.)
5899 Repeats start with an on_failure_jump that points past both
5900 the repetition text and either the following jump or
5901 pop_failure_jump back to this on_failure_jump. */
5910 /* This operation is used for greedy *.
5911 Compare the beginning of the repeat with what in the
5912 pattern follows its end. If we can establish that there
5913 is nothing that they would both match, i.e., that we
5914 would have to backtrack because of (as in, e.g., `a*a')
5915 then we can use a non-backtracking loop based on
5916 on_failure_keep_string_jump instead of on_failure_jump. */
5921 {
5923 /* Here, we discard `const', making re_match non-reentrant. */
5928 instruction once it's been changed. */
5932 /* Ensure this is a indeed the trivial kind of loop
5933 we are expecting. */
5938 {
5939 /* Use a fast `on_failure_keep_string_jump' loop. */
5943 }
5944 else
5945 {
5946 /* Default to a safe `on_failure_jump' loop. */
5949 }
5951 }
5954 /* Unconditionally jump (without popping any failure points). */
5956 unconditional_jump:
5957 IMMEDIATE_QUIT_CHECK;
5965 /* Have to succeed matching what follows at least n times.
5966 After that, handle like `on_failure_jump'. */
5968 /* Signedness doesn't matter since we only compare MCNT to 0. */
5972 /* Originally, mcnt is how many times we HAVE to succeed. */
5974 {
5975 /* Here, we discard `const', making re_match non-reentrant. */
5977 mcnt--;
5980 }
5981 else
5982 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5987 /* Signedness doesn't matter since we only compare MCNT to 0. */
5991 /* Originally, this is how many times we CAN jump. */
5993 {
5994 /* Here, we discard `const', making re_match non-reentrant. */
5996 mcnt--;
5999 }
6000 /* If don't have to jump any more, skip over the rest of command. */
6001 else
6006 {
6011 /* Here, we discard `const', making re_match non-reentrant. */
6013 /* Signedness doesn't matter since we only copy MCNT's bits . */
6018 }
6025 /* We SUCCEED (or FAIL) in one of the following cases: */
6027 /* Case 1: D is at the beginning or the end of string. */
6030 else
6031 {
6032 /* C1 is the character before D, S1 is the syntax of C1, C2
6033 is the character at D, and S2 is the syntax of C2. */
6037 #ifdef emacs
6041 #endif
6044 #ifdef emacs
6046 #endif
6053 /* Case 3: Both of S1 and S2 are Sword, and macro
6054 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6057 }
6060 else
6066 /* We FAIL in one of the following cases: */
6068 /* Case 1: D is at the end of string. */
6071 else
6072 {
6073 /* C1 is the character before D, S1 is the syntax of C1, C2
6074 is the character at D, and S2 is the syntax of C2. */
6078 #ifdef emacs
6082 #endif
6087 /* Case 2: S2 is not Sword. */
6091 /* Case 3: D is not at the beginning of string ... */
6093 {
6095 #ifdef emacs
6097 #endif
6100 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6101 returns 0. */
6104 }
6105 }
6111 /* We FAIL in one of the following cases: */
6113 /* Case 1: D is at the beginning of string. */
6116 else
6117 {
6118 /* C1 is the character before D, S1 is the syntax of C1, C2
6119 is the character at D, and S2 is the syntax of C2. */
6123 #ifdef emacs
6127 #endif
6131 /* Case 2: S1 is not Sword. */
6135 /* Case 3: D is not at the end of string ... */
6137 {
6140 #ifdef emacs
6142 #endif
6145 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6146 returns 0. */
6149 }
6150 }
6156 /* We FAIL in one of the following cases: */
6158 /* Case 1: D is at the end of string. */
6161 else
6162 {
6163 /* C1 is the character before D, S1 is the syntax of C1, C2
6164 is the character at D, and S2 is the syntax of C2. */
6167 #ifdef emacs
6171 #endif
6176 /* Case 2: S2 is neither Sword nor Ssymbol. */
6180 /* Case 3: D is not at the beginning of string ... */
6182 {
6184 #ifdef emacs
6186 #endif
6189 /* ... and S1 is Sword or Ssymbol. */
6192 }
6193 }
6199 /* We FAIL in one of the following cases: */
6201 /* Case 1: D is at the beginning of string. */
6204 else
6205 {
6206 /* C1 is the character before D, S1 is the syntax of C1, C2
6207 is the character at D, and S2 is the syntax of C2. */
6210 #ifdef emacs
6214 #endif
6218 /* Case 2: S1 is neither Ssymbol nor Sword. */
6222 /* Case 3: D is not at the end of string ... */
6224 {
6227 #ifdef emacs
6229 #endif
6232 /* ... and S2 is Sword or Ssymbol. */
6235 }
6236 }
6245 #ifdef emacs
6246 {
6250 }
6251 #endif
6252 {
6254 re_wchar_t c;
6260 }
6263 #ifdef emacs
6288 {
6290 re_wchar_t c;
6296 }
6303 }
6307 /* We goto here if a matching operation fails. */
6308 fail:
6309 IMMEDIATE_QUIT_CHECK;
6311 {
6313 /* A restart point is known. Restore to that state. */
6317 {
6325 /* Fallthrough */
6331 continue_failure_jump:
6337 /* A special frame used for nastyloops. */
6342 }
6348 }
6349 else
6360 \f
6361 /* Subroutine definitions for re_match_2. */
6363 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6364 bytes; nonzero otherwise. */
6366 static int
6369 {
6374 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6375 different lengths, but relying on a single `len' would break this. -sm */
6377 {
6389 }
6395 }
6396 \f
6397 /* Entry points for GNU code. */
6399 /* re_compile_pattern is the GNU regular expression compiler: it
6400 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6401 Returns 0 if the pattern was valid, otherwise an error string.
6403 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6404 are set in BUFP on entry.
6406 We call regex_compile to do the actual compilation. */
6410 {
6411 reg_errcode_t ret;
6413 /* GNU code is written to assume at least RE_NREGS registers will be set
6414 (and at least one extra will be -1). */
6417 /* And GNU code determines whether or not to get register information
6418 by passing null for the REGS argument to re_match, etc., not by
6419 setting no_sub. */
6427 }
6429 \f
6430 /* Entry points compatible with 4.2 BSD regex library. We don't define
6431 them unless specifically requested. */
6433 #if defined _REGEX_RE_COMP || defined _LIBC
6435 /* BSD has one and only one pattern buffer. */
6439 # ifdef _LIBC
6440 /* Make these definitions weak in libc, so POSIX programs can redefine
6441 these names if they don't use our functions, and still use
6442 regcomp/regexec below without link errors. */
6443 weak_function
6444 # endif
6447 {
6448 reg_errcode_t ret;
6451 {
6453 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6456 }
6459 {
6462 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6468 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6470 }
6472 /* Since `re_exec' always passes NULL for the `regs' argument, we
6473 don't need to initialize the pattern buffer fields which affect it. */
6480 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6482 }
6485 int
6486 # ifdef _LIBC
6487 weak_function
6488 # endif
6491 {
6493 return
6495 }
6497 \f
6498 /* POSIX.2 functions. Don't define these for Emacs. */
6500 #ifndef emacs
6502 /* regcomp takes a regular expression as a string and compiles it.
6504 PREG is a regex_t *. We do not expect any fields to be initialized,
6505 since POSIX says we shouldn't. Thus, we set
6507 `buffer' to the compiled pattern;
6508 `used' to the length of the compiled pattern;
6509 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6510 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6511 RE_SYNTAX_POSIX_BASIC;
6512 `fastmap' to an allocated space for the fastmap;
6513 `fastmap_accurate' to zero;
6514 `re_nsub' to the number of subexpressions in PATTERN.
6516 PATTERN is the address of the pattern string.
6518 CFLAGS is a series of bits which affect compilation.
6520 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6521 use POSIX basic syntax.
6523 If REG_NEWLINE is set, then . and [^...] don't match newline.
6524 Also, regexec will try a match beginning after every newline.
6526 If REG_ICASE is set, then we considers upper- and lowercase
6527 versions of letters to be equivalent when matching.
6529 If REG_NOSUB is set, then when PREG is passed to regexec, that
6530 routine will report only success or failure, and nothing about the
6531 registers.
6533 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6534 the return codes and their meanings.) */
6536 int
6541 {
6542 reg_errcode_t ret;
6543 reg_syntax_t syntax
6547 /* regex_compile will allocate the space for the compiled pattern. */
6552 /* Try to allocate space for the fastmap. */
6556 {
6559 preg->translate
6565 /* Map uppercase characters to corresponding lowercase ones. */
6568 }
6569 else
6572 /* If REG_NEWLINE is set, newlines are treated differently. */
6577 }
6578 else
6583 /* POSIX says a null character in the pattern terminates it, so we
6584 can use strlen here in compiling the pattern. */
6587 /* POSIX doesn't distinguish between an unmatched open-group and an
6588 unmatched close-group: both are REG_EPAREN. */
6594 buffer. */
6600 }
6601 }
6603 }
6607 /* regexec searches for a given pattern, specified by PREG, in the
6608 string STRING.
6610 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6611 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6612 least NMATCH elements, and we set them to the offsets of the
6613 corresponding matched substrings.
6615 EFLAGS specifies `execution flags' which affect matching: if
6616 REG_NOTBOL is set, then ^ does not match at the beginning of the
6617 string; if REG_NOTEOL is set, then $ does not match at the end.
6619 We return 0 if we find a match and REG_NOMATCH if not. */
6621 int
6628 {
6631 regex_t private_preg;
6640 /* The user has told us exactly how many registers to return
6641 information about, via `nmatch'. We have to pass that on to the
6642 matching routines. */
6646 {
6652 }
6654 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6655 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6656 was a little bit longer but still only matching the real part.
6657 This works because the `endline' will check for a '\n' and will find a
6658 '\0', correctly deciding that this is not the end of a line.
6659 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6660 a convenient '\0' there. For all we know, the string could be preceded
6661 by '\n' which would throw things off. */
6663 /* Perform the searching operation. */
6668 /* Copy the register information to the POSIX structure. */
6670 {
6672 {
6676 {
6679 }
6680 }
6682 /* If we needed the temporary register info, free the space now. */
6684 }
6686 /* We want zero return to mean success, unlike `re_search'. */
6688 }
6692 /* Returns a message corresponding to an error code, ERR_CODE, returned
6693 from either regcomp or regexec. We don't use PREG here.
6695 ERR_CODE was previously called ERRCODE, but that name causes an
6696 error with msvc8 compiler. */
6698 size_t
6704 {
6710 /* Only error codes returned by the rest of the code should be passed
6711 to this routine. If we are given anything else, or if other regex
6712 code generates an invalid error code, then the program has a bug.
6713 Dump core so we can fix it. */
6721 {
6723 {
6726 }
6727 else
6729 }
6732 }
6736 /* Free dynamically allocated space used by PREG. */
6738 void
6741 {
6754 }
6759 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6760 (do not change this comment) */