| blob | 10ab857b00a1333dfb2f9f1098d9246443b7dad9 |
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 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
252 If nothing else has been done, use the method below. */
253 # ifdef INHIBIT_STRING_HEADER
254 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
255 # if !defined bzero && !defined bcopy
256 # undef INHIBIT_STRING_HEADER
257 # endif
258 # endif
259 # endif
261 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
262 This is used in most programs--a few other programs avoid this
263 by defining INHIBIT_STRING_HEADER. */
264 # ifndef INHIBIT_STRING_HEADER
265 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
266 # include <string.h>
267 # ifndef bzero
268 # ifndef _LIBC
270 # else
271 # define bzero(s, n) __bzero (s, n)
272 # endif
273 # endif
274 # else
275 # include <strings.h>
276 # ifndef memcmp
277 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
278 # endif
279 # ifndef memcpy
280 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
281 # endif
282 # endif
283 # endif
285 /* Define the syntax stuff for \<, \>, etc. */
287 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
290 # define SWITCH_ENUM_CAST(x) (x)
292 /* Dummy macros for non-Emacs environments. */
293 # define CHAR_CHARSET(c) 0
294 # define CHARSET_LEADING_CODE_BASE(c) 0
295 # define MAX_MULTIBYTE_LENGTH 1
296 # define RE_MULTIBYTE_P(x) 0
297 # define RE_TARGET_MULTIBYTE_P(x) 0
298 # define WORD_BOUNDARY_P(c1, c2) (0)
299 # define CHAR_HEAD_P(p) (1)
300 # define SINGLE_BYTE_CHAR_P(c) (1)
301 # define SAME_CHARSET_P(c1, c2) (1)
302 # define BYTES_BY_CHAR_HEAD(p) (1)
303 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
304 # define STRING_CHAR(p) (*(p))
305 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
306 # define CHAR_STRING(c, s) (*(s) = (c), 1)
307 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
308 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
309 # define RE_CHAR_TO_MULTIBYTE(c) (c)
310 # define RE_CHAR_TO_UNIBYTE(c) (c)
311 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
312 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
313 # define GET_CHAR_AFTER(c, p, len) \
314 (c = *p, len = 1)
315 # define MAKE_CHAR(charset, c1, c2) (c1)
316 # define BYTE8_TO_CHAR(c) (c)
317 # define CHAR_BYTE8_P(c) (0)
318 # define CHAR_LEADING_CODE(c) (c)
322 #ifndef RE_TRANSLATE
323 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
324 # define RE_TRANSLATE_P(TBL) (TBL)
325 #endif
326 \f
327 /* Get the interface, including the syntax bits. */
330 /* isalpha etc. are used for the character classes. */
331 #include <ctype.h>
333 #ifdef emacs
335 /* 1 if C is an ASCII character. */
336 # define IS_REAL_ASCII(c) ((c) < 0200)
338 /* 1 if C is a unibyte character. */
339 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
341 /* The Emacs definitions should not be directly affected by locales. */
343 /* In Emacs, these are only used for single-byte characters. */
350 /* This is only used for single-byte characters. */
353 /* The rest must handle multibyte characters. */
355 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
357 : 1)
359 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
361 : 1)
363 # define ISALNUM(c) (IS_REAL_ASCII (c) \
367 : SYNTAX (c) == Sword)
369 # define ISALPHA(c) (IS_REAL_ASCII (c) \
372 : SYNTAX (c) == Sword)
374 # define ISLOWER(c) (LOWERCASEP (c))
376 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
381 : SYNTAX (c) != Sword)
383 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
385 # define ISUPPER(c) (UPPERCASEP (c))
387 # define ISWORD(c) (SYNTAX (c) == Sword)
391 /* Jim Meyering writes:
393 "... Some ctype macros are valid only for character codes that
394 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
395 using /bin/cc or gcc but without giving an ansi option). So, all
396 ctype uses should be through macros like ISPRINT... If
397 STDC_HEADERS is defined, then autoconf has verified that the ctype
398 macros don't need to be guarded with references to isascii. ...
399 Defining isascii to 1 should let any compiler worth its salt
400 eliminate the && through constant folding."
401 Solaris defines some of these symbols so we must undefine them first. */
403 # undef ISASCII
404 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
405 # define ISASCII(c) 1
406 # else
407 # define ISASCII(c) isascii(c)
408 # endif
410 /* 1 if C is an ASCII character. */
411 # define IS_REAL_ASCII(c) ((c) < 0200)
413 /* This distinction is not meaningful, except in Emacs. */
414 # define ISUNIBYTE(c) 1
416 # ifdef isblank
417 # define ISBLANK(c) (ISASCII (c) && isblank (c))
418 # else
420 # endif
421 # ifdef isgraph
422 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
423 # else
424 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
425 # endif
427 # undef ISPRINT
428 # define ISPRINT(c) (ISASCII (c) && isprint (c))
429 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
430 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
431 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
432 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
433 # define ISLOWER(c) (ISASCII (c) && islower (c))
434 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
435 # define ISSPACE(c) (ISASCII (c) && isspace (c))
436 # define ISUPPER(c) (ISASCII (c) && isupper (c))
437 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
439 # define ISWORD(c) ISALPHA(c)
441 # ifdef _tolower
442 # define TOLOWER(c) _tolower(c)
443 # else
444 # define TOLOWER(c) tolower(c)
445 # endif
447 /* How many characters in the character set. */
448 # define CHAR_SET_SIZE 256
450 # ifdef SYNTAX_TABLE
458 static void
460 {
476 }
480 # define SYNTAX(c) re_syntax_table[(c)]
483 \f
484 #ifndef NULL
485 # define NULL (void *)0
486 #endif
488 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
489 since ours (we hope) works properly with all combinations of
490 machines, compilers, `char' and `unsigned char' argument types.
491 (Per Bothner suggested the basic approach.) */
492 #undef SIGN_EXTEND_CHAR
493 #if __STDC__
494 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
496 /* As in Harbison and Steele. */
497 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
498 #endif
499 \f
500 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
501 use `alloca' instead of `malloc'. This is because using malloc in
502 re_search* or re_match* could cause memory leaks when C-g is used in
503 Emacs; also, malloc is slower and causes storage fragmentation. On
504 the other hand, malloc is more portable, and easier to debug.
506 Because we sometimes use alloca, some routines have to be macros,
507 not functions -- `alloca'-allocated space disappears at the end of the
508 function it is called in. */
510 #ifdef REGEX_MALLOC
512 # define REGEX_ALLOCATE malloc
513 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
514 # define REGEX_FREE free
518 /* Emacs already defines alloca, sometimes. */
519 # ifndef alloca
521 /* Make alloca work the best possible way. */
522 # ifdef __GNUC__
523 # define alloca __builtin_alloca
525 # ifdef HAVE_ALLOCA_H
526 # include <alloca.h>
532 # define REGEX_ALLOCATE alloca
534 /* Assumes a `char *destination' variable. */
535 # define REGEX_REALLOCATE(source, osize, nsize) \
536 (destination = (char *) alloca (nsize), \
537 memcpy (destination, source, osize))
539 /* No need to do anything to free, after alloca. */
544 /* Define how to allocate the failure stack. */
546 #if defined REL_ALLOC && defined REGEX_MALLOC
548 # define REGEX_ALLOCATE_STACK(size) \
549 r_alloc (&failure_stack_ptr, (size))
550 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
551 r_re_alloc (&failure_stack_ptr, (nsize))
552 # define REGEX_FREE_STACK(ptr) \
553 r_alloc_free (&failure_stack_ptr)
557 # ifdef REGEX_MALLOC
559 # define REGEX_ALLOCATE_STACK malloc
560 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
561 # define REGEX_FREE_STACK free
565 # define REGEX_ALLOCATE_STACK alloca
567 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
568 REGEX_REALLOCATE (source, osize, nsize)
569 /* No need to explicitly free anything. */
570 # define REGEX_FREE_STACK(arg) ((void)0)
576 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
577 `string1' or just past its end. This works if PTR is NULL, which is
578 a good thing. */
579 #define FIRST_STRING_P(ptr) \
580 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
582 /* (Re)Allocate N items of type T using malloc, or fail. */
583 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
584 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
585 #define RETALLOC_IF(addr, n, t) \
586 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
587 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
591 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
593 #undef MAX
594 #undef MIN
595 #define MAX(a, b) ((a) > (b) ? (a) : (b))
596 #define MIN(a, b) ((a) < (b) ? (a) : (b))
598 /* Type of source-pattern and string chars. */
602 #define false 0
603 #define true 1
611 \f
612 /* These are the command codes that appear in compiled regular
613 expressions. Some opcodes are followed by argument bytes. A
614 command code can specify any interpretation whatsoever for its
615 arguments. Zero bytes may appear in the compiled regular expression. */
618 {
621 /* Succeed right away--no more backtracking. */
622 succeed,
624 /* Followed by one byte giving n, then by n literal bytes. */
625 exactn,
627 /* Matches any (more or less) character. */
628 anychar,
630 /* Matches any one char belonging to specified set. First
631 following byte is number of bitmap bytes. Then come bytes
632 for a bitmap saying which chars are in. Bits in each byte
633 are ordered low-bit-first. A character is in the set if its
634 bit is 1. A character too large to have a bit in the map is
635 automatically not in the set.
637 If the length byte has the 0x80 bit set, then that stuff
638 is followed by a range table:
639 2 bytes of flags for character sets (low 8 bits, high 8 bits)
640 See RANGE_TABLE_WORK_BITS below.
641 2 bytes, the number of pairs that follow (upto 32767)
642 pairs, each 2 multibyte characters,
643 each multibyte character represented as 3 bytes. */
644 charset,
646 /* Same parameters as charset, but match any character that is
647 not one of those specified. */
648 charset_not,
650 /* Start remembering the text that is matched, for storing in a
651 register. Followed by one byte with the register number, in
652 the range 0 to one less than the pattern buffer's re_nsub
653 field. */
654 start_memory,
656 /* Stop remembering the text that is matched and store it in a
657 memory register. Followed by one byte with the register
658 number, in the range 0 to one less than `re_nsub' in the
659 pattern buffer. */
660 stop_memory,
662 /* Match a duplicate of something remembered. Followed by one
663 byte containing the register number. */
664 duplicate,
666 /* Fail unless at beginning of line. */
667 begline,
669 /* Fail unless at end of line. */
670 endline,
672 /* Succeeds if at beginning of buffer (if emacs) or at beginning
673 of string to be matched (if not). */
674 begbuf,
676 /* Analogously, for end of buffer/string. */
677 endbuf,
679 /* Followed by two byte relative address to which to jump. */
680 jump,
682 /* Followed by two-byte relative address of place to resume at
683 in case of failure. */
684 on_failure_jump,
686 /* Like on_failure_jump, but pushes a placeholder instead of the
687 current string position when executed. */
688 on_failure_keep_string_jump,
690 /* Just like `on_failure_jump', except that it checks that we
691 don't get stuck in an infinite loop (matching an empty string
692 indefinitely). */
693 on_failure_jump_loop,
695 /* Just like `on_failure_jump_loop', except that it checks for
696 a different kind of loop (the kind that shows up with non-greedy
697 operators). This operation has to be immediately preceded
698 by a `no_op'. */
699 on_failure_jump_nastyloop,
701 /* A smart `on_failure_jump' used for greedy * and + operators.
702 It analyses the loop before which it is put and if the
703 loop does not require backtracking, it changes itself to
704 `on_failure_keep_string_jump' and short-circuits the loop,
705 else it just defaults to changing itself into `on_failure_jump'.
706 It assumes that it is pointing to just past a `jump'. */
707 on_failure_jump_smart,
709 /* Followed by two-byte relative address and two-byte number n.
710 After matching N times, jump to the address upon failure.
711 Does not work if N starts at 0: use on_failure_jump_loop
712 instead. */
713 succeed_n,
715 /* Followed by two-byte relative address, and two-byte number n.
716 Jump to the address N times, then fail. */
717 jump_n,
719 /* Set the following two-byte relative address to the
720 subsequent two-byte number. The address *includes* the two
721 bytes of number. */
722 set_number_at,
733 /* Matches any character whose syntax is specified. Followed by
734 a byte which contains a syntax code, e.g., Sword. */
735 syntaxspec,
737 /* Matches any character whose syntax is not that specified. */
738 notsyntaxspec
740 #ifdef emacs
745 /* Matches any character whose category-set contains the specified
746 category. The operator is followed by a byte which contains a
747 category code (mnemonic ASCII character). */
748 categoryspec,
750 /* Matches any character whose category-set does not contain the
751 specified category. The operator is followed by a byte which
752 contains the category code (mnemonic ASCII character). */
753 notcategoryspec
756 \f
757 /* Common operations on the compiled pattern. */
759 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
761 #define STORE_NUMBER(destination, number) \
762 do { \
763 (destination)[0] = (number) & 0377; \
764 (destination)[1] = (number) >> 8; \
765 } while (0)
767 /* Same as STORE_NUMBER, except increment DESTINATION to
768 the byte after where the number is stored. Therefore, DESTINATION
769 must be an lvalue. */
771 #define STORE_NUMBER_AND_INCR(destination, number) \
772 do { \
773 STORE_NUMBER (destination, number); \
774 (destination) += 2; \
775 } while (0)
777 /* Put into DESTINATION a number stored in two contiguous bytes starting
778 at SOURCE. */
780 #define EXTRACT_NUMBER(destination, source) \
781 do { \
782 (destination) = *(source) & 0377; \
783 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
784 } while (0)
786 #ifdef DEBUG
788 static void
792 {
796 }
799 # undef EXTRACT_NUMBER
800 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
805 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
806 SOURCE must be an lvalue. */
808 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
809 do { \
810 EXTRACT_NUMBER (destination, source); \
811 (source) += 2; \
812 } while (0)
814 #ifdef DEBUG
817 static void
821 {
824 }
826 # ifndef EXTRACT_MACROS
827 # undef EXTRACT_NUMBER_AND_INCR
828 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
829 extract_number_and_incr (&dest, &src)
833 \f
834 /* Store a multibyte character in three contiguous bytes starting
835 DESTINATION, and increment DESTINATION to the byte after where the
836 character is stored. Therefore, DESTINATION must be an lvalue. */
838 #define STORE_CHARACTER_AND_INCR(destination, character) \
839 do { \
840 (destination)[0] = (character) & 0377; \
841 (destination)[1] = ((character) >> 8) & 0377; \
842 (destination)[2] = (character) >> 16; \
843 (destination) += 3; \
844 } while (0)
846 /* Put into DESTINATION a character stored in three contiguous bytes
847 starting at SOURCE. */
849 #define EXTRACT_CHARACTER(destination, source) \
850 do { \
851 (destination) = ((source)[0] \
852 | ((source)[1] << 8) \
853 | ((source)[2] << 16)); \
854 } while (0)
857 /* Macros for charset. */
859 /* Size of bitmap of charset P in bytes. P is a start of charset,
860 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
861 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
863 /* Nonzero if charset P has range table. */
864 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
866 /* Return the address of range table of charset P. But not the start
867 of table itself, but the before where the number of ranges is
868 stored. `2 +' means to skip re_opcode_t and size of bitmap,
869 and the 2 bytes of flags at the start of the range table. */
870 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
872 /* Extract the bit flags that start a range table. */
873 #define CHARSET_RANGE_TABLE_BITS(p) \
874 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
875 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
877 /* Test if C is listed in the bitmap of charset P. */
878 #define CHARSET_LOOKUP_BITMAP(p, c) \
879 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
880 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
882 /* Return the address of end of RANGE_TABLE. COUNT is number of
883 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
884 is start of range and end of range. `* 3' is size of each start
885 and end. */
886 #define CHARSET_RANGE_TABLE_END(range_table, count) \
887 ((range_table) + (count) * 2 * 3)
889 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
890 COUNT is number of ranges in RANGE_TABLE. */
891 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
892 do \
893 { \
894 re_wchar_t range_start, range_end; \
895 re_char *p; \
896 re_char *range_table_end \
897 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
898 \
899 for (p = (range_table); p < range_table_end; p += 2 * 3) \
900 { \
901 EXTRACT_CHARACTER (range_start, p); \
902 EXTRACT_CHARACTER (range_end, p + 3); \
903 \
904 if (range_start <= (c) && (c) <= range_end) \
905 { \
906 (not) = !(not); \
907 break; \
908 } \
909 } \
910 } \
911 while (0)
913 /* Test if C is in range table of CHARSET. The flag NOT is negated if
914 C is listed in it. */
915 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
916 do \
917 { \
919 int count; \
920 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
921 \
922 EXTRACT_NUMBER_AND_INCR (count, range_table); \
923 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
924 } \
925 while (0)
926 \f
927 /* If DEBUG is defined, Regex prints many voluminous messages about what
928 it is doing (if the variable `debug' is nonzero). If linked with the
929 main program in `iregex.c', you can enter patterns and strings
930 interactively. And if linked with the main program in `main.c' and
931 the other test files, you can run the already-written tests. */
933 #ifdef DEBUG
935 /* We use standard I/O for debugging. */
936 # include <stdio.h>
938 /* It is useful to test things that ``must'' be true when debugging. */
939 # include <assert.h>
943 # define DEBUG_STATEMENT(e) e
944 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
945 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
946 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
947 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
948 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
949 if (debug > 0) print_partial_compiled_pattern (s, e)
950 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
951 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
954 /* Print the fastmap in human-readable form. */
956 void
959 {
964 {
966 {
970 {
972 i++;
973 }
975 {
978 }
979 }
980 }
982 }
985 /* Print a compiled pattern string in human-readable form, starting at
986 the START pointer into it and ending just before the pointer END. */
988 void
992 {
998 {
1001 }
1003 /* Loop over pattern commands. */
1005 {
1009 {
1021 do
1022 {
1024 }
1046 {
1061 {
1062 /* Are we starting a range? */
1064 {
1067 }
1068 /* Have we broken a range? */
1070 {
1073 }
1079 }
1089 {
1093 /* ??? Should print the range table; for now, just skip it. */
1097 }
1098 }
1193 # ifdef emacs
1229 }
1232 }
1235 }
1238 void
1241 {
1249 {
1252 }
1262 /* Perhaps we should print the translate table? */
1263 }
1266 void
1273 {
1278 else
1279 {
1281 {
1286 }
1290 }
1291 }
1295 # undef assert
1296 # define assert(e)
1298 # define DEBUG_STATEMENT(e)
1299 # define DEBUG_PRINT1(x)
1300 # define DEBUG_PRINT2(x1, x2)
1301 # define DEBUG_PRINT3(x1, x2, x3)
1302 # define DEBUG_PRINT4(x1, x2, x3, x4)
1303 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1304 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1307 \f
1308 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1309 also be assigned to arbitrarily: each pattern buffer stores its own
1310 syntax, so it can be changed between regex compilations. */
1311 /* This has no initializer because initialized variables in Emacs
1312 become read-only after dumping. */
1313 reg_syntax_t re_syntax_options;
1316 /* Specify the precise syntax of regexps for compilation. This provides
1317 for compatibility for various utilities which historically have
1318 different, incompatible syntaxes.
1320 The argument SYNTAX is a bit mask comprised of the various bits
1321 defined in regex.h. We return the old syntax. */
1323 reg_syntax_t
1325 {
1330 }
1333 /* Regexp to use to replace spaces, or NULL meaning don't. */
1336 void
1338 {
1340 }
1342 \f
1343 /* This table gives an error message for each of the error codes listed
1344 in regex.h. Obviously the order here has to be same as there.
1345 POSIX doesn't require that we do anything for REG_NOERROR,
1346 but why not be nice? */
1349 {
1368 };
1369 \f
1370 /* Avoiding alloca during matching, to placate r_alloc. */
1372 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1373 searching and matching functions should not call alloca. On some
1374 systems, alloca is implemented in terms of malloc, and if we're
1375 using the relocating allocator routines, then malloc could cause a
1376 relocation, which might (if the strings being searched are in the
1377 ralloc heap) shift the data out from underneath the regexp
1378 routines.
1380 Here's another reason to avoid allocation: Emacs
1381 processes input from X in a signal handler; processing X input may
1382 call malloc; if input arrives while a matching routine is calling
1383 malloc, then we're scrod. But Emacs can't just block input while
1384 calling matching routines; then we don't notice interrupts when
1385 they come in. So, Emacs blocks input around all regexp calls
1386 except the matching calls, which it leaves unprotected, in the
1387 faith that they will not malloc. */
1389 /* Normally, this is fine. */
1390 #define MATCH_MAY_ALLOCATE
1392 /* The match routines may not allocate if (1) they would do it with malloc
1393 and (2) it's not safe for them to use malloc.
1394 Note that if REL_ALLOC is defined, matching would not use malloc for the
1395 failure stack, but we would still use it for the register vectors;
1396 so REL_ALLOC should not affect this. */
1397 #if defined REGEX_MALLOC && defined emacs
1398 # undef MATCH_MAY_ALLOCATE
1399 #endif
1401 \f
1402 /* Failure stack declarations and macros; both re_compile_fastmap and
1403 re_match_2 use a failure stack. These have to be macros because of
1404 REGEX_ALLOCATE_STACK. */
1407 /* Approximate number of failure points for which to initially allocate space
1408 when matching. If this number is exceeded, we allocate more
1409 space, so it is not a hard limit. */
1410 #ifndef INIT_FAILURE_ALLOC
1411 # define INIT_FAILURE_ALLOC 20
1412 #endif
1414 /* Roughly the maximum number of failure points on the stack. Would be
1415 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1416 This is a variable only so users of regex can assign to it; we never
1417 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1418 before using it, so it should probably be a byte-count instead. */
1419 # if defined MATCH_MAY_ALLOCATE
1420 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1421 whose default stack limit is 2mb. In order for a larger
1422 value to work reliably, you have to try to make it accord
1423 with the process stack limit. */
1425 # else
1427 # endif
1429 union fail_stack_elt
1430 {
1432 /* This should be the biggest `int' that's no bigger than a pointer. */
1434 };
1439 {
1446 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1447 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1450 /* Define macros to initialize and free the failure stack.
1451 Do `return -2' if the alloc fails. */
1453 #ifdef MATCH_MAY_ALLOCATE
1454 # define INIT_FAIL_STACK() \
1455 do { \
1456 fail_stack.stack = (fail_stack_elt_t *) \
1457 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1458 * sizeof (fail_stack_elt_t)); \
1459 \
1460 if (fail_stack.stack == NULL) \
1461 return -2; \
1462 \
1463 fail_stack.size = INIT_FAILURE_ALLOC; \
1464 fail_stack.avail = 0; \
1465 fail_stack.frame = 0; \
1466 } while (0)
1468 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1469 #else
1470 # define INIT_FAIL_STACK() \
1471 do { \
1472 fail_stack.avail = 0; \
1473 fail_stack.frame = 0; \
1474 } while (0)
1476 # define RESET_FAIL_STACK() ((void)0)
1477 #endif
1480 /* Double the size of FAIL_STACK, up to a limit
1481 which allows approximately `re_max_failures' items.
1483 Return 1 if succeeds, and 0 if either ran out of memory
1484 allocating space for it or it was already too large.
1486 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1488 /* Factor to increase the failure stack size by
1489 when we increase it.
1490 This used to be 2, but 2 was too wasteful
1491 because the old discarded stacks added up to as much space
1492 were as ultimate, maximum-size stack. */
1493 #define FAIL_STACK_GROWTH_FACTOR 4
1495 #define GROW_FAIL_STACK(fail_stack) \
1496 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1497 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1498 ? 0 \
1499 : ((fail_stack).stack \
1500 = (fail_stack_elt_t *) \
1501 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1502 (fail_stack).size * sizeof (fail_stack_elt_t), \
1503 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1504 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1505 * FAIL_STACK_GROWTH_FACTOR))), \
1506 \
1507 (fail_stack).stack == NULL \
1508 ? 0 \
1509 : ((fail_stack).size \
1510 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1511 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1512 * FAIL_STACK_GROWTH_FACTOR)) \
1513 / sizeof (fail_stack_elt_t)), \
1514 1)))
1517 /* Push a pointer value onto the failure stack.
1518 Assumes the variable `fail_stack'. Probably should only
1519 be called from within `PUSH_FAILURE_POINT'. */
1520 #define PUSH_FAILURE_POINTER(item) \
1521 fail_stack.stack[fail_stack.avail++].pointer = (item)
1523 /* This pushes an integer-valued item onto the failure stack.
1524 Assumes the variable `fail_stack'. Probably should only
1525 be called from within `PUSH_FAILURE_POINT'. */
1526 #define PUSH_FAILURE_INT(item) \
1527 fail_stack.stack[fail_stack.avail++].integer = (item)
1529 /* Push a fail_stack_elt_t value onto the failure stack.
1530 Assumes the variable `fail_stack'. Probably should only
1531 be called from within `PUSH_FAILURE_POINT'. */
1532 #define PUSH_FAILURE_ELT(item) \
1533 fail_stack.stack[fail_stack.avail++] = (item)
1535 /* These three POP... operations complement the three PUSH... operations.
1536 All assume that `fail_stack' is nonempty. */
1537 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1538 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1539 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1541 /* Individual items aside from the registers. */
1542 #define NUM_NONREG_ITEMS 3
1544 /* Used to examine the stack (to detect infinite loops). */
1545 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1546 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1547 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1548 #define TOP_FAILURE_HANDLE() fail_stack.frame
1551 #define ENSURE_FAIL_STACK(space) \
1552 while (REMAINING_AVAIL_SLOTS <= space) { \
1553 if (!GROW_FAIL_STACK (fail_stack)) \
1554 return -2; \
1557 }
1559 /* Push register NUM onto the stack. */
1560 #define PUSH_FAILURE_REG(num) \
1561 do { \
1562 char *destination; \
1563 ENSURE_FAIL_STACK(3); \
1565 num, regstart[num], regend[num]); \
1566 PUSH_FAILURE_POINTER (regstart[num]); \
1567 PUSH_FAILURE_POINTER (regend[num]); \
1568 PUSH_FAILURE_INT (num); \
1569 } while (0)
1571 /* Change the counter's value to VAL, but make sure that it will
1572 be reset when backtracking. */
1573 #define PUSH_NUMBER(ptr,val) \
1574 do { \
1575 char *destination; \
1576 int c; \
1577 ENSURE_FAIL_STACK(3); \
1578 EXTRACT_NUMBER (c, ptr); \
1580 PUSH_FAILURE_INT (c); \
1581 PUSH_FAILURE_POINTER (ptr); \
1582 PUSH_FAILURE_INT (-1); \
1583 STORE_NUMBER (ptr, val); \
1584 } while (0)
1586 /* Pop a saved register off the stack. */
1587 #define POP_FAILURE_REG_OR_COUNT() \
1588 do { \
1589 int reg = POP_FAILURE_INT (); \
1590 if (reg == -1) \
1591 { \
1594 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1595 reg = POP_FAILURE_INT (); \
1596 STORE_NUMBER (ptr, reg); \
1598 } \
1599 else \
1600 { \
1601 regend[reg] = POP_FAILURE_POINTER (); \
1602 regstart[reg] = POP_FAILURE_POINTER (); \
1604 reg, regstart[reg], regend[reg]); \
1605 } \
1606 } while (0)
1608 /* Check that we are not stuck in an infinite loop. */
1609 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1610 do { \
1611 int failure = TOP_FAILURE_HANDLE (); \
1613 while (failure > 0 \
1614 && (FAILURE_STR (failure) == string_place \
1615 || FAILURE_STR (failure) == NULL)) \
1616 { \
1617 assert (FAILURE_PAT (failure) >= bufp->buffer \
1618 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1619 if (FAILURE_PAT (failure) == pat_cur) \
1620 { \
1621 cycle = 1; \
1622 break; \
1623 } \
1625 failure = NEXT_FAILURE_HANDLE(failure); \
1626 } \
1628 } while (0)
1630 /* Push the information about the state we will need
1631 if we ever fail back to it.
1633 Requires variables fail_stack, regstart, regend and
1634 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1635 declared.
1637 Does `return FAILURE_CODE' if runs out of memory. */
1639 #define PUSH_FAILURE_POINT(pattern, string_place) \
1640 do { \
1641 char *destination; \
1642 /* Must be int, so when we don't save any registers, the arithmetic \
1644 \
1645 DEBUG_STATEMENT (nfailure_points_pushed++); \
1649 \
1650 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1651 \
1653 \
1655 PUSH_FAILURE_INT (fail_stack.frame); \
1656 \
1658 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1660 PUSH_FAILURE_POINTER (string_place); \
1661 \
1663 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1664 PUSH_FAILURE_POINTER (pattern); \
1665 \
1667 fail_stack.frame = fail_stack.avail; \
1668 } while (0)
1670 /* Estimate the size of data pushed by a typical failure stack entry.
1671 An estimate is all we need, because all we use this for
1672 is to choose a limit for how big to make the failure stack. */
1673 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1674 #define TYPICAL_FAILURE_SIZE 20
1676 /* How many items can still be added to the stack without overflowing it. */
1677 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1680 /* Pops what PUSH_FAIL_STACK pushes.
1682 We restore into the parameters, all of which should be lvalues:
1683 STR -- the saved data position.
1684 PAT -- the saved pattern position.
1685 REGSTART, REGEND -- arrays of string positions.
1687 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1688 `pend', `string1', `size1', `string2', and `size2'. */
1690 #define POP_FAILURE_POINT(str, pat) \
1691 do { \
1692 assert (!FAIL_STACK_EMPTY ()); \
1693 \
1698 \
1700 while (fail_stack.frame < fail_stack.avail) \
1701 POP_FAILURE_REG_OR_COUNT (); \
1702 \
1703 pat = POP_FAILURE_POINTER (); \
1705 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1706 \
1707 /* If the saved string location is NULL, it came from an \
1708 on_failure_keep_string_jump opcode, and we want to throw away the \
1710 str = POP_FAILURE_POINTER (); \
1712 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1714 \
1715 fail_stack.frame = POP_FAILURE_INT (); \
1717 \
1718 assert (fail_stack.avail >= 0); \
1719 assert (fail_stack.frame <= fail_stack.avail); \
1720 \
1721 DEBUG_STATEMENT (nfailure_points_popped++); \
1725 \f
1726 /* Registers are set to a sentinel when they haven't yet matched. */
1727 #define REG_UNSET(e) ((e) == NULL)
1728 \f
1729 /* Subroutine declarations and macros for regex_compile. */
1732 reg_syntax_t syntax,
1743 reg_syntax_t syntax));
1746 reg_syntax_t syntax));
1751 /* Fetch the next character in the uncompiled pattern, with no
1752 translation. */
1753 #define PATFETCH(c) \
1754 do { \
1755 int len; \
1756 if (p == pend) return REG_EEND; \
1757 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1758 p += len; \
1759 } while (0)
1762 /* If `translate' is non-null, return translate[D], else just D. We
1763 cast the subscript to translate because some data is declared as
1764 `char *', to avoid warnings when a string constant is passed. But
1765 when we use a character as a subscript we must make it unsigned. */
1766 #ifndef TRANSLATE
1767 # define TRANSLATE(d) \
1768 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1769 #endif
1772 /* Macros for outputting the compiled pattern into `buffer'. */
1774 /* If the buffer isn't allocated when it comes in, use this. */
1775 #define INIT_BUF_SIZE 32
1777 /* Make sure we have at least N more bytes of space in buffer. */
1778 #define GET_BUFFER_SPACE(n) \
1779 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1780 EXTEND_BUFFER ()
1782 /* Make sure we have one more byte of buffer space and then add C to it. */
1783 #define BUF_PUSH(c) \
1784 do { \
1785 GET_BUFFER_SPACE (1); \
1786 *b++ = (unsigned char) (c); \
1787 } while (0)
1790 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1791 #define BUF_PUSH_2(c1, c2) \
1792 do { \
1793 GET_BUFFER_SPACE (2); \
1794 *b++ = (unsigned char) (c1); \
1795 *b++ = (unsigned char) (c2); \
1796 } while (0)
1799 /* As with BUF_PUSH_2, except for three bytes. */
1800 #define BUF_PUSH_3(c1, c2, c3) \
1801 do { \
1802 GET_BUFFER_SPACE (3); \
1803 *b++ = (unsigned char) (c1); \
1804 *b++ = (unsigned char) (c2); \
1805 *b++ = (unsigned char) (c3); \
1806 } while (0)
1809 /* Store a jump with opcode OP at LOC to location TO. We store a
1810 relative address offset by the three bytes the jump itself occupies. */
1811 #define STORE_JUMP(op, loc, to) \
1812 store_op1 (op, loc, (to) - (loc) - 3)
1814 /* Likewise, for a two-argument jump. */
1815 #define STORE_JUMP2(op, loc, to, arg) \
1816 store_op2 (op, loc, (to) - (loc) - 3, arg)
1818 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1819 #define INSERT_JUMP(op, loc, to) \
1820 insert_op1 (op, loc, (to) - (loc) - 3, b)
1822 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1823 #define INSERT_JUMP2(op, loc, to, arg) \
1824 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1827 /* This is not an arbitrary limit: the arguments which represent offsets
1828 into the pattern are two bytes long. So if 2^15 bytes turns out to
1829 be too small, many things would have to change. */
1830 # define MAX_BUF_SIZE (1L << 15)
1833 /* Any other compiler which, like MSC, has allocation limit below 2^16
1834 bytes will have to use approach similar to what was done below for
1835 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1836 reallocating to 0 bytes. Such thing is not going to work too well.
1837 You have been warned!! */
1838 #if defined _MSC_VER && !defined WIN32
1839 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1840 # define MAX_BUF_SIZE 65500L
1841 #else
1842 # define MAX_BUF_SIZE (1L << 16)
1843 #endif
1846 /* Extend the buffer by twice its current size via realloc and
1847 reset the pointers that pointed into the old block to point to the
1848 correct places in the new one. If extending the buffer results in it
1849 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1850 #if __BOUNDED_POINTERS__
1851 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1852 # define MOVE_BUFFER_POINTER(P) \
1853 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1854 SET_HIGH_BOUND (P), \
1855 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1856 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1857 else \
1858 { \
1859 SET_HIGH_BOUND (b); \
1860 SET_HIGH_BOUND (begalt); \
1861 if (fixup_alt_jump) \
1862 SET_HIGH_BOUND (fixup_alt_jump); \
1863 if (laststart) \
1864 SET_HIGH_BOUND (laststart); \
1865 if (pending_exact) \
1866 SET_HIGH_BOUND (pending_exact); \
1867 }
1868 #else
1869 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1870 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1871 #endif
1872 #define EXTEND_BUFFER() \
1873 do { \
1874 unsigned char *old_buffer = bufp->buffer; \
1875 if (bufp->allocated == MAX_BUF_SIZE) \
1876 return REG_ESIZE; \
1877 bufp->allocated <<= 1; \
1878 if (bufp->allocated > MAX_BUF_SIZE) \
1879 bufp->allocated = MAX_BUF_SIZE; \
1880 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1881 if (bufp->buffer == NULL) \
1882 return REG_ESPACE; \
1884 if (old_buffer != bufp->buffer) \
1885 { \
1886 unsigned char *new_buffer = bufp->buffer; \
1887 MOVE_BUFFER_POINTER (b); \
1888 MOVE_BUFFER_POINTER (begalt); \
1889 if (fixup_alt_jump) \
1890 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1891 if (laststart) \
1892 MOVE_BUFFER_POINTER (laststart); \
1893 if (pending_exact) \
1894 MOVE_BUFFER_POINTER (pending_exact); \
1895 } \
1896 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1897 } while (0)
1900 /* Since we have one byte reserved for the register number argument to
1901 {start,stop}_memory, the maximum number of groups we can report
1902 things about is what fits in that byte. */
1903 #define MAX_REGNUM 255
1905 /* But patterns can have more than `MAX_REGNUM' registers. We just
1906 ignore the excess. */
1910 /* Macros for the compile stack. */
1912 /* Since offsets can go either forwards or backwards, this type needs to
1913 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1914 /* int may be not enough when sizeof(int) == 2. */
1918 {
1919 pattern_offset_t begalt_offset;
1920 pattern_offset_t fixup_alt_jump;
1921 pattern_offset_t laststart_offset;
1922 regnum_t regnum;
1927 {
1934 #define INIT_COMPILE_STACK_SIZE 32
1936 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1937 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1939 /* The next available element. */
1940 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1942 /* Explicit quit checking is only used on NTemacs and whenever we
1943 use polling to process input events. */
1944 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1946 # define IMMEDIATE_QUIT_CHECK \
1947 do { \
1948 if (immediate_quit) QUIT; \
1949 } while (0)
1950 #else
1951 # define IMMEDIATE_QUIT_CHECK ((void)0)
1952 #endif
1953 \f
1954 /* Structure to manage work area for range table. */
1955 struct range_table_work_area
1956 {
1961 };
1963 /* Make sure that WORK_AREA can hold more N multibyte characters.
1964 This is used only in set_image_of_range and set_image_of_range_1.
1965 It expects WORK_AREA to be a pointer.
1966 If it can't get the space, it returns from the surrounding function. */
1968 #define EXTEND_RANGE_TABLE(work_area, n) \
1969 do { \
1970 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1971 { \
1972 extend_range_table_work_area (&work_area); \
1973 if ((work_area).table == 0) \
1974 return (REG_ESPACE); \
1975 } \
1976 } while (0)
1978 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1979 (work_area).bits |= (bit)
1981 /* Bits used to implement the multibyte-part of the various character classes
1982 such as [:alnum:] in a charset's range table. */
1983 #define BIT_WORD 0x1
1984 #define BIT_LOWER 0x2
1985 #define BIT_PUNCT 0x4
1986 #define BIT_SPACE 0x8
1987 #define BIT_UPPER 0x10
1988 #define BIT_MULTIBYTE 0x20
1990 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1991 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1992 do { \
1993 EXTEND_RANGE_TABLE ((work_area), 2); \
1994 (work_area).table[(work_area).used++] = (range_start); \
1995 (work_area).table[(work_area).used++] = (range_end); \
1996 } while (0)
1998 /* Free allocated memory for WORK_AREA. */
1999 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2000 do { \
2001 if ((work_area).table) \
2002 free ((work_area).table); \
2003 } while (0)
2005 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2006 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2007 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2008 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2009 \f
2011 /* Set the bit for character C in a list. */
2012 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2015 #ifdef emacs
2017 /* Store characters in the range FROM to TO in the bitmap at B (for
2018 ASCII and unibyte characters) and WORK_AREA (for multibyte
2019 characters) while translating them and paying attention to the
2020 continuity of translated characters.
2022 Implementation note: It is better to implement these fairly big
2023 macros by a function, but it's not that easy because macros called
2024 in this macro assume various local variables already declared. */
2026 /* Both FROM and TO are ASCII characters. */
2028 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2029 do { \
2030 int C0, C1; \
2031 \
2032 for (C0 = (FROM); C0 <= (TO); C0++) \
2033 { \
2034 C1 = TRANSLATE (C0); \
2035 if (! ASCII_CHAR_P (C1)) \
2036 { \
2037 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2038 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2039 C1 = C0; \
2040 } \
2041 SET_LIST_BIT (C1); \
2042 } \
2043 } while (0)
2046 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2048 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2049 do { \
2050 int C0, C1, C2, I; \
2051 int USED = RANGE_TABLE_WORK_USED (work_area); \
2052 \
2053 for (C0 = (FROM); C0 <= (TO); C0++) \
2054 { \
2055 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2056 if (CHAR_BYTE8_P (C1)) \
2057 SET_LIST_BIT (C0); \
2058 else \
2059 { \
2060 C2 = TRANSLATE (C1); \
2061 if (C2 == C1 \
2062 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2063 C1 = C0; \
2064 SET_LIST_BIT (C1); \
2065 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2066 { \
2067 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2068 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2069 \
2070 if (C2 >= from - 1 && C2 <= to + 1) \
2071 { \
2072 if (C2 == from - 1) \
2073 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2074 else if (C2 == to + 1) \
2075 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2076 break; \
2077 } \
2078 } \
2079 if (I < USED) \
2080 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2081 } \
2082 } \
2083 } while (0)
2086 /* Both FROM and TO are multibyte characters. */
2088 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2089 do { \
2090 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2091 \
2092 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2093 for (C0 = (FROM); C0 <= (TO); C0++) \
2094 { \
2095 C1 = TRANSLATE (C0); \
2096 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2097 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2098 SET_LIST_BIT (C2); \
2099 if (C1 >= (FROM) && C1 <= (TO)) \
2100 continue; \
2101 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2102 { \
2103 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2104 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2105 \
2106 if (C1 >= from - 1 && C1 <= to + 1) \
2107 { \
2108 if (C1 == from - 1) \
2109 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2110 else if (C1 == to + 1) \
2111 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2112 break; \
2113 } \
2114 } \
2115 if (I < USED) \
2116 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2117 } \
2118 } while (0)
2122 /* Get the next unsigned number in the uncompiled pattern. */
2123 #define GET_UNSIGNED_NUMBER(num) \
2124 do { \
2125 if (p == pend) \
2126 FREE_STACK_RETURN (REG_EBRACE); \
2127 else \
2128 { \
2129 PATFETCH (c); \
2131 { \
2132 int prev; \
2133 if (num < 0) \
2134 num = 0; \
2135 prev = num; \
2137 if (num / 10 != prev) \
2138 FREE_STACK_RETURN (REG_BADBR); \
2139 if (p == pend) \
2140 FREE_STACK_RETURN (REG_EBRACE); \
2141 PATFETCH (c); \
2142 } \
2143 } \
2144 } while (0)
2145 \f
2146 #if ! WIDE_CHAR_SUPPORT
2148 /* Map a string to the char class it names (if any). */
2149 re_wctype_t
2151 {
2171 }
2173 /* True if CH is in the char class CC. */
2174 boolean
2176 {
2178 {
2199 }
2200 }
2202 /* Return a bit-pattern to use in the range-table bits to match multibyte
2203 chars of class CC. */
2204 static int
2206 {
2208 {
2220 }
2221 }
2222 #endif
2223 \f
2224 /* Filling in the work area of a range. */
2226 /* Actually extend the space in WORK_AREA. */
2228 static void
2230 {
2233 work_area->table
2235 else
2236 work_area->table
2238 }
2240 #if 0
2241 #ifdef emacs
2243 /* Carefully find the ranges of codes that are equivalent
2244 under case conversion to the range start..end when passed through
2245 TRANSLATE. Handle the case where non-letters can come in between
2246 two upper-case letters (which happens in Latin-1).
2247 Also handle the case of groups of more than 2 case-equivalent chars.
2249 The basic method is to look at consecutive characters and see
2250 if they can form a run that can be handled as one.
2252 Returns -1 if successful, REG_ESPACE if ran out of space. */
2254 static int
2256 RE_TRANSLATE_TYPE translate;
2259 {
2260 /* `one_case' indicates a character, or a run of characters,
2261 each of which is an isolate (no case-equivalents).
2262 This includes all ASCII non-letters.
2264 `two_case' indicates a character, or a run of characters,
2265 each of which has two case-equivalent forms.
2266 This includes all ASCII letters.
2268 `strange' indicates a character that has more than one
2269 case-equivalent. */
2273 /* Describe the run that is in progress,
2274 which the next character can try to extend.
2275 If run_type is strange, that means there really is no run.
2276 If run_type is one_case, then run_start...run_end is the run.
2277 If run_type is two_case, then the run is run_start...run_end,
2278 and the case-equivalents end at run_eqv_end. */
2283 Lisp_Object eqv_table;
2286 {
2291 }
2296 {
2301 /* Classify this character */
2306 else
2311 else
2314 /* Can this character extend the run in progress? */
2319 {
2320 /* No, end the run.
2321 Record each of its equivalent ranges. */
2323 {
2327 }
2329 {
2337 }
2339 }
2342 {
2343 /* For a strange character, add each of its equivalents, one
2344 by one. Don't start a range. */
2345 do
2346 {
2351 }
2353 }
2355 /* Add this char to the run, or start a new run. */
2357 {
2358 /* Initialize a new range. */
2363 }
2364 else
2365 {
2366 /* Extend a running range. */
2369 }
2370 }
2372 /* If a run is still in progress at the end, finish it now
2373 by recording its equivalent ranges. */
2375 {
2379 }
2381 {
2389 }
2392 }
2396 /* Record the image of the range start..end when passed through
2397 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2398 and is not even necessarily contiguous.
2399 Normally we approximate it with the smallest contiguous range that contains
2400 all the chars we need. However, for Latin-1 we go to extra effort
2401 to do a better job.
2403 This function is not called for ASCII ranges.
2405 Returns -1 if successful, REG_ESPACE if ran out of space. */
2407 static int
2409 RE_TRANSLATE_TYPE translate;
2412 {
2415 #ifdef emacs
2416 /* For Latin-1 ranges, use set_image_of_range_1
2417 to get proper handling of ranges that include letters and nonletters.
2418 For a range that includes the whole of Latin-1, this is not necessary.
2419 For other character sets, we don't bother to get this right. */
2422 {
2435 }
2436 #endif
2445 {
2449 {
2452 {
2455 else
2456 {
2459 }
2460 }
2461 }
2464 {
2468 }
2469 }
2472 }
2474 \f
2475 #ifndef MATCH_MAY_ALLOCATE
2477 /* If we cannot allocate large objects within re_match_2_internal,
2478 we make the fail stack and register vectors global.
2479 The fail stack, we grow to the maximum size when a regexp
2480 is compiled.
2481 The register vectors, we adjust in size each time we
2482 compile a regexp, according to the number of registers it needs. */
2486 /* Size with which the following vectors are currently allocated.
2487 That is so we can make them bigger as needed,
2488 but never make them smaller. */
2494 /* Make the register vectors big enough for NUM_REGS registers,
2495 but don't make them smaller. */
2497 static
2500 {
2502 {
2509 }
2510 }
2513 \f
2515 compile_stack,
2516 regnum_t regnum));
2518 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2519 Returns one of error codes defined in `regex.h', or zero for success.
2521 Assumes the `allocated' (and perhaps `buffer') and `translate'
2522 fields are set in BUFP on entry.
2524 If it succeeds, results are put in BUFP (if it returns an error, the
2525 contents of BUFP are undefined):
2526 `buffer' is the compiled pattern;
2527 `syntax' is set to SYNTAX;
2528 `used' is set to the length of the compiled pattern;
2529 `fastmap_accurate' is zero;
2530 `re_nsub' is the number of subexpressions in PATTERN;
2531 `not_bol' and `not_eol' are zero;
2533 The `fastmap' field is neither examined nor set. */
2535 /* Insert the `jump' from the end of last alternative to "here".
2536 The space for the jump has already been allocated. */
2537 #define FIXUP_ALT_JUMP() \
2538 do { \
2539 if (fixup_alt_jump) \
2540 STORE_JUMP (jump, fixup_alt_jump, b); \
2541 } while (0)
2544 /* Return, freeing storage we allocated. */
2545 #define FREE_STACK_RETURN(value) \
2546 do { \
2547 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2548 free (compile_stack.stack); \
2549 return value; \
2550 } while (0)
2552 static reg_errcode_t
2553 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2554 {
2555 /* We fetch characters from PATTERN here. */
2558 /* A random temporary spot in PATTERN. */
2561 /* Points to the end of the buffer, where we should append. */
2564 /* Keeps track of unclosed groups. */
2565 compile_stack_type compile_stack;
2567 /* Points to the current (ending) position in the pattern. */
2568 #ifdef AIX
2569 /* `const' makes AIX compiler fail. */
2571 #else
2573 #endif
2576 /* How to translate the characters in the pattern. */
2579 /* Address of the count-byte of the most recently inserted `exactn'
2580 command. This makes it possible to tell if a new exact-match
2581 character can be added to that command or if the character requires
2582 a new `exactn' command. */
2585 /* Address of start of the most recently finished expression.
2586 This tells, e.g., postfix * where to find the start of its
2587 operand. Reset at the beginning of groups and alternatives. */
2590 /* Address of beginning of regexp, or inside of last group. */
2593 /* Place in the uncompiled pattern (i.e., the {) to
2594 which to go back if the interval is invalid. */
2597 /* Address of the place where a forward jump should go to the end of
2598 the containing expression. Each alternative of an `or' -- except the
2599 last -- ends with a forward jump of this sort. */
2602 /* Work area for range table of charset. */
2605 /* If the object matched can contain multibyte characters. */
2608 /* If a target of matching can contain multibyte characters. */
2611 /* Nonzero if we have pushed down into a subpattern. */
2614 /* These hold the values of p, pattern, and pend from the main
2615 pattern when we have pushed into a subpattern. */
2620 #ifdef DEBUG
2621 debug++;
2624 {
2630 }
2633 /* Initialize the compile stack. */
2644 /* Initialize the pattern buffer. */
2650 /* Set `used' to zero, so that if we return an error, the pattern
2651 printer (for debugging) will think there's no pattern. We reset it
2652 at the end. */
2655 /* Always count groups, whether or not bufp->no_sub is set. */
2658 #if !defined emacs && !defined SYNTAX_TABLE
2659 /* Initialize the syntax table. */
2661 #endif
2664 {
2667 enough space. This loses if buffer's address is bogus, but
2668 that is the user's responsibility. */
2670 }
2671 else
2674 }
2678 }
2682 /* Loop through the uncompiled pattern until we're at the end. */
2684 {
2686 {
2687 /* If this is the end of an included regexp,
2688 pop back to the main regexp and try again. */
2690 {
2696 }
2697 /* If this is the end of the main regexp, we are done. */
2699 }
2704 {
2706 {
2709 /* If there's no special whitespace regexp, treat
2710 spaces normally. And don't try to do this recursively. */
2714 /* Peek past following spaces. */
2716 {
2719 p1++;
2720 }
2721 /* If the spaces are followed by a repetition op,
2722 treat them normally. */
2728 /* Replace the spaces with the whitespace regexp. */
2736 }
2739 {
2742 /* If context independent, it's an operator. */
2744 /* Otherwise, depends on what's come before. */
2747 else
2749 }
2754 {
2756 p == pend
2757 /* If context independent, it's an operator. */
2759 /* Otherwise, depends on what's next. */
2762 else
2764 }
2773 handle_plus:
2775 /* If there is no previous pattern... */
2777 {
2782 }
2784 {
2785 /* 1 means zero (many) matches is allowed. */
2789 /* If there is a sequence of repetition chars, collapse it
2790 down to just one (the right one). We can't combine
2791 interval operators with these because of, e.g., `a{2}*',
2792 which should only match an even number of `a's. */
2795 {
2799 else
2800 {
2803 }
2810 ;
2812 {
2817 else
2819 }
2820 else
2822 /* If we get here, we found another repeat character. */
2824 }
2826 /* Star, etc. applied to an empty pattern is equivalent
2827 to an empty pattern. */
2831 /* Now we know whether or not zero matches is allowed
2832 and also whether or not two or more matches is allowed. */
2834 {
2836 {
2839 re_opcode_t ofj =
2840 /* Check if the loop can match the empty string. */
2847 on_failure_keep_string_jump, we turn simple P+
2848 into PP* if P is simple. */
2856 }
2860 /* A + loop. */
2862 else
2863 /* Simple * loops can use on_failure_keep_string_jump
2864 depending on what follows. But since we don't know
2865 that yet, we leave the decision up to
2866 on_failure_jump_smart. */
2872 }
2873 else
2874 {
2875 /* A simple ? pattern. */
2880 }
2881 }
2887 {
2890 /* The non-greedy multiple match looks like
2891 a repeat..until: we only need a conditional jump
2892 at the end of the loop. */
2898 {
2899 /* The repeat...until naturally matches one or more.
2900 To also match zero times, we need to first jump to
2901 the end of the loop (its conditional jump). */
2904 }
2905 }
2906 else
2907 {
2908 /* non-greedy a?? */
2913 }
2914 }
2915 }
2927 {
2932 /* Ensure that we have enough space to push a charset: the
2933 opcode, the length count, and the bitset; 34 bytes in all. */
2938 /* We test `*p == '^' twice, instead of using an if
2939 statement, so we only need one BUF_PUSH. */
2942 p++;
2944 /* Remember the first position in the bracket expression. */
2947 /* Push the number of bytes in the bitmap. */
2950 /* Clear the whole map. */
2953 /* charset_not matches newline according to a syntax bit. */
2958 /* Read in characters and ranges, setting map bits. */
2960 {
2967 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2968 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2969 So the translation is done later in a loop. Example:
2970 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2973 /* \ might escape characters inside [...] and [^...]. */
2975 {
2980 }
2981 else
2982 {
2983 /* Could be the end of the bracket expression. If it's
2984 not (i.e., when the bracket expression is `[]' so
2985 far), the ']' character bit gets set way below. */
2988 }
2990 /* See if we're at the beginning of a possible character
2991 class. */
2995 {
2996 /* Leave room for the null. */
3004 /* If pattern is `[[:'. */
3008 {
3014 else
3015 /* This is in any case an invalid class name. */
3017 }
3020 /* If isn't a word bracketed by `[:' and `:]':
3021 undo the ending character, the letters, and
3022 leave the leading `:' and `[' (but set bits for
3023 them). */
3025 {
3026 re_wctype_t cc;
3034 /* Throw away the ] at the end of the character
3035 class. */
3040 #ifndef emacs
3043 {
3047 }
3049 /* Most character classes in a multibyte match
3050 just set a flag. Exceptions are is_blank,
3051 is_digit, is_cntrl, and is_xdigit, since
3052 they can only match ASCII characters. We
3053 don't need to handle them for multibyte.
3054 They are distinguished by a negative wctype. */
3056 /* Setup the gl_state object to its buffer-defined
3057 value. This hardcodes the buffer-global
3058 syntax-table for ASCII chars, while the other chars
3059 will obey syntax-table properties. It's not ideal,
3060 but it's the way it's been done until now. */
3064 {
3068 {
3077 }
3078 }
3079 SET_RANGE_TABLE_WORK_AREA_BIT
3082 /* In most cases the matching rule for char classes
3083 only uses the syntax table for multibyte chars,
3084 so that the content of the syntax-table it is not
3085 hardcoded in the range_table. SPACE and WORD are
3086 the two exceptions. */
3090 /* Repeat the loop. */
3092 }
3093 else
3094 {
3095 /* Go back to right after the "[:". */
3099 /* Because the `:' may starts the range, we
3100 can't simply set bit and repeat the loop.
3101 Instead, just set it to C and handle below. */
3103 }
3104 }
3107 {
3109 /* Discard the `-'. */
3112 /* Fetch the character which ends the range. */
3114 #ifdef emacs
3117 /* Treat the range from a multibyte character to
3118 raw-byte character as empty. */
3121 }
3122 else
3123 /* Range from C to C. */
3127 {
3130 /* Else, repeat the loop. */
3131 }
3132 else
3133 {
3134 #ifndef emacs
3135 /* Set the range into bitmap */
3137 {
3141 }
3144 {
3150 }
3152 {
3154 {
3159 }
3161 {
3163 }
3164 else
3165 {
3167 }
3168 }
3170 }
3171 }
3173 /* Discard any (non)matching list bytes that are all 0 at the
3174 end of the map. Decrease the map-length byte too. */
3179 /* Build real range table from work area. */
3182 {
3186 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3187 bytes for flags, two for COUNT, and three bytes for
3188 each character. */
3191 /* Indicate the existence of range table. */
3194 /* Store the character class flag bits into the range table.
3195 If not in emacs, these flag bits are always 0. */
3201 STORE_CHARACTER_AND_INCR
3203 }
3204 }
3211 else
3218 else
3225 else
3232 else
3239 else
3246 /* Do not translate the character after the \, so that we can
3247 distinguish, e.g., \B from \b, even if we normally would
3248 translate, e.g., B to b. */
3252 {
3257 handle_open:
3258 {
3262 {
3263 /* Look for a special (?...) construct */
3265 {
3268 {
3271 {
3274 /* An explicitly specified regnum must start
3275 with non-0. */
3282 /* Only (?:...) is supported right now. */
3284 }
3285 }
3286 }
3287 }
3297 /* Ideally, we'd want to check that the specified
3298 group can't have matched (i.e. all subgroups
3299 using the same regnum are in other branches of
3300 OR patterns), but we don't currently keep track
3301 of enough info to do that easily. */
3304 }
3305 else
3306 /* It's really shy. */
3310 {
3312 compile_stack_elt_t);
3316 }
3318 /* These are the values to restore when we hit end of this
3319 group. They are all relative offsets, so that if the
3320 whole pattern moves because of realloc, they will still
3321 be valid. */
3323 COMPILE_STACK_TOP.fixup_alt_jump
3328 /* Do not push a start_memory for groups beyond the last one
3329 we can represent in the compiled pattern. */
3338 /* If we've reached MAX_REGNUM groups, then this open
3339 won't actually generate any code, so we'll have to
3340 clear pending_exact explicitly. */
3343 }
3349 {
3352 else
3354 }
3356 handle_close:
3359 /* See similar code for backslashed left paren above. */
3361 {
3364 else
3366 }
3368 /* Since we just checked for an empty stack above, this
3369 ``can't happen''. */
3371 {
3372 /* We don't just want to restore into `regnum', because
3373 later groups should continue to be numbered higher,
3374 as in `(ab)c(de)' -- the second group is #2. */
3375 regnum_t regnum;
3379 fixup_alt_jump
3385 /* If we've reached MAX_REGNUM groups, then this open
3386 won't actually generate any code, so we'll have to
3387 clear pending_exact explicitly. */
3390 /* We're at the end of the group, so now we know how many
3391 groups were inside this one. */
3394 }
3401 handle_alt:
3405 /* Insert before the previous alternative a jump which
3406 jumps to this alternative if the former fails. */
3412 /* The alternative before this one has a jump after it
3413 which gets executed if it gets matched. Adjust that
3414 jump so it will jump to this alternative's analogous
3415 jump (put in below, which in turn will jump to the next
3416 (if any) alternative's such jump, etc.). The last such
3417 jump jumps to the correct final destination. A picture:
3418 _____ _____
3419 | | | |
3420 | v | v
3421 a | b | c
3423 If we are at `b', then fixup_alt_jump right now points to a
3424 three-byte space after `a'. We'll put in the jump, set
3425 fixup_alt_jump to right after `b', and leave behind three
3426 bytes which we'll fill in when we get to after `c'. */
3430 /* Mark and leave space for a jump after this alternative,
3431 to be filled in later either by next alternative or
3432 when know we're at the end of a series of alternatives. */
3443 /* If \{ is a literal. */
3445 /* If we're at `\{' and it's not the open-interval
3446 operator. */
3450 handle_interval:
3451 {
3452 /* If got here, then the syntax allows intervals. */
3454 /* At least (most) this many matches must be made. */
3463 else
3464 /* Interval such as `{1}' => match exactly once. */
3472 {
3478 }
3483 /* We just parsed a valid interval. */
3485 /* If it's invalid to have no preceding re. */
3487 {
3492 else
3494 }
3497 /* If the upper bound is zero, just drop the sub pattern
3498 altogether. */
3501 /* Just match it once: nothing to do here. */
3502 ;
3504 /* Otherwise, we have a nontrivial interval. When
3505 we're all done, the pattern will look like:
3506 set_number_at <jump count> <upper bound>
3507 set_number_at <succeed_n count> <lower bound>
3508 succeed_n <after jump addr> <succeed_n count>
3509 <body of loop>
3510 jump_n <succeed_n addr> <jump count>
3511 (The upper bound and `jump_n' are omitted if
3512 `upper_bound' is 1, though.) */
3513 else
3515 more at the end of the loop. */
3523 {
3524 /* A succeed_n that starts with 0 is really a
3525 a simple on_failure_jump_loop. */
3529 }
3530 else
3531 {
3532 /* Initialize lower bound of the `succeed_n', even
3533 though it will be set during matching by its
3534 attendant `set_number_at' (inserted next),
3535 because `re_compile_fastmap' needs to know.
3536 Jump to the `jump_n' we might insert below. */
3539 lower_bound);
3542 /* Code to initialize the lower bound. Insert
3543 before the `succeed_n'. The `5' is the last two
3544 bytes of this `set_number_at', plus 3 bytes of
3545 the following `succeed_n'. */
3549 }
3552 {
3553 /* A negative upper bound stands for infinity,
3554 in which case it degenerates to a plain jump. */
3557 }
3560 append a backward jump to the `succeed_n'
3561 that starts this interval.
3563 When we've reached this during matching,
3564 we'll have matched the interval once, so
3565 jump back only `upper_bound - 1' times. */
3570 /* The location we want to set is the second
3571 parameter of the `jump_n'; that is `b-2' as
3572 an absolute address. `laststart' will be
3573 the `set_number_at' we're about to insert;
3574 `laststart+3' the number to set, the source
3575 for the relative address. But we are
3576 inserting into the middle of the pattern --
3577 so everything is getting moved up by 5.
3578 Conclusion: (b - 2) - (laststart + 3) + 5,
3579 i.e., b - laststart.
3581 We insert this at the beginning of the loop
3582 so that if we fail during matching, we'll
3583 reinitialize the bounds. */
3587 }
3588 }
3591 }
3594 unfetch_interval:
3595 /* If an invalid interval, match the characters as literals. */
3600 /* normal_char and normal_backslash need `c'. */
3604 {
3607 }
3608 else
3611 #ifdef emacs
3612 /* There is no way to specify the before_dot and after_dot
3613 operators. rms says this is ok. --karl */
3681 else
3711 {
3712 regnum_t reg;
3720 /* Can't back reference to a subexp before its end. */
3726 }
3734 else
3738 normal_backslash:
3739 /* You might think it would be useful for \ to mean
3740 not to translate; but if we don't translate it
3741 it will never match anything. */
3743 }
3748 /* Expects the character in `c'. */
3749 normal_char:
3750 /* If no exactn currently being built. */
3753 /* If last exactn not at current position. */
3756 /* We have only one byte following the exactn for the count. */
3759 /* If followed by a repetition operator. */
3768 {
3769 /* Start building a new exactn. */
3775 }
3778 {
3782 {
3786 }
3787 else
3788 {
3791 {
3796 }
3799 }
3801 }
3808 /* Through the pattern now. */
3815 /* If we don't want backtracking, force success
3816 the first time we reach the end of the compiled pattern. */
3820 /* We have succeeded; set the length of the buffer. */
3823 #ifdef DEBUG
3825 {
3829 }
3830 debug--;
3833 #ifndef MATCH_MAY_ALLOCATE
3834 /* Initialize the failure stack to the largest possible stack. This
3835 isn't necessary unless we're trying to avoid calling alloca in
3836 the search and match routines. */
3837 {
3841 {
3845 fail_stack.stack
3848 else
3849 fail_stack.stack
3853 }
3856 }
3861 \f
3862 /* Subroutines for `regex_compile'. */
3864 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3866 static void
3868 {
3871 }
3874 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3876 static void
3878 {
3882 }
3885 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3886 for OP followed by two-byte integer parameter ARG. */
3888 static void
3890 {
3898 }
3901 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3903 static void
3905 {
3913 }
3916 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3917 after an alternative or a begin-subexpression. We assume there is at
3918 least one character before the ^. */
3920 static boolean
3922 {
3926 return
3927 /* After a subexpression? */
3929 /* After an alternative? */
3931 /* After a shy subexpression? */
3936 }
3939 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3940 at least one character after the $, i.e., `P < PEND'. */
3942 static boolean
3944 {
3949 return
3950 /* Before a subexpression? */
3953 /* Before an alternative? */
3956 }
3959 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3960 false if it's not. */
3962 static boolean
3964 {
3969 this_element--)
3974 }
3975 \f
3976 /* analyse_first.
3977 If fastmap is non-NULL, go through the pattern and fill fastmap
3978 with all the possible leading chars. If fastmap is NULL, don't
3979 bother filling it up (obviously) and only return whether the
3980 pattern could potentially match the empty string.
3982 Return 1 if p..pend might match the empty string.
3983 Return 0 if p..pend matches at least one char.
3984 Return -1 if fastmap was not updated accurately. */
3986 static int
3988 {
3992 /* If all elements for base leading-codes in fastmap is set, this
3993 flag is set true. */
3998 /* The loop below works as follows:
3999 - It has a working-list kept in the PATTERN_STACK and which basically
4000 starts by only containing a pointer to the first operation.
4001 - If the opcode we're looking at is a match against some set of
4002 chars, then we add those chars to the fastmap and go on to the
4003 next work element from the worklist (done via `break').
4004 - If the opcode is a control operator on the other hand, we either
4005 ignore it (if it's meaningless at this point, such as `start_memory')
4006 or execute it (if it's a jump). If the jump has several destinations
4007 (i.e. `on_failure_jump'), then we push the other destination onto the
4008 worklist.
4009 We guarantee termination by ignoring backward jumps (more or less),
4010 so that `p' is monotonically increasing. More to the point, we
4011 never set `p' (or push) anything `<= p1'. */
4014 {
4015 /* `p1' is used as a marker of how far back a `on_failure_jump'
4016 can go without being ignored. It is normally equal to `p'
4017 (which prevents any backward `on_failure_jump') except right
4018 after a plain `jump', to allow patterns such as:
4019 0: jump 10
4020 3..9: <body>
4021 10: on_failure_jump 3
4022 as used for the *? operator. */
4026 {
4032 /* If the first character has to match a backreference, that means
4033 that the group was empty (since it already matched). Since this
4034 is the only case that interests us here, we can assume that the
4035 backreference must match the empty string. */
4036 p++;
4040 /* Following are the cases which match a character. These end
4041 with `break'. */
4045 {
4046 /* If multibyte is nonzero, the first byte of each
4047 character is an ASCII or a leading code. Otherwise,
4048 each byte is a character. Thus, this works in both
4049 cases. */
4052 {
4053 /* For the case of matching this unibyte regex
4054 against multibyte, we must set a leading code of
4055 the corresponding multibyte character. */
4059 }
4060 }
4065 /* We could put all the chars except for \n (and maybe \0)
4066 but we don't bother since it is generally not worth it. */
4073 {
4074 /* Chars beyond end of bitmap are possible matches. */
4078 }
4080 /* Fallthrough */
4089 #ifdef emacs
4091 which doesn't match the specified set of characters. */
4092 not
4093 ||
4094 /* If we can match a character class, we can match any
4095 multibyte characters. */
4099 {
4101 {
4106 }
4107 }
4111 {
4112 /* Set fastmap[I] to 1 where I is a leading code of each
4113 multibyte characer in the range table. */
4117 /* Make P points the range table. `+ 2' is to skip flag
4118 bits for a character class. */
4121 /* Extract the number of ranges in range table into COUNT. */
4124 {
4125 /* Extract the start and end of each range. */
4133 }
4134 }
4135 #endif
4141 #ifndef emacs
4149 /* This match depends on text properties. These end with
4150 aborting optimizations. */
4162 /* Any leading code can possibly start a character which
4163 has or doesn't has the specified category. */
4165 {
4170 }
4173 /* All cases after this match the empty string. These end with
4174 `continue'. */
4197 /* Backward jumps can only go back to code that we've already
4198 visited. `re_compile' should make sure this is true. */
4202 {
4208 p++;
4212 };
4213 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4214 to jump back to "just after here". */
4215 /* Fallthrough */
4225 else
4227 We first go down the "straight" path so as to minimize
4228 stack usage when going through alternatives. */
4232 }
4237 /* This code simply does not properly handle forward jump_n. */
4240 /* jump_n can either jump or fall through. The (backward) jump
4241 case has already been handled, so we only need to look at the
4242 fallthrough case. */
4246 /* If N == 0, it should be an on_failure_jump_loop instead. */
4249 /* We only care about one iteration of the loop, so we don't
4250 need to consider the case where this behaves like an
4251 on_failure_jump. */
4270 /* Getting here means we have found the possible starting
4271 characters for one path of the pattern -- and that the empty
4272 string does not match. We need not follow this path further. */
4276 /* We reached the end without matching anything. */
4280 \f
4281 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4282 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4283 characters can start a string that matches the pattern. This fastmap
4284 is used by re_search to skip quickly over impossible starting points.
4286 Character codes above (1 << BYTEWIDTH) are not represented in the
4287 fastmap, but the leading codes are represented. Thus, the fastmap
4288 indicates which character sets could start a match.
4290 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4291 area as BUFP->fastmap.
4293 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4294 the pattern buffer.
4296 Returns 0 if we succeed, -2 if an internal error. */
4298 int
4300 {
4314 \f
4315 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4316 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4317 this memory for recording register information. STARTS and ENDS
4318 must be allocated using the malloc library routine, and must each
4319 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4321 If NUM_REGS == 0, then subsequent matches should allocate their own
4322 register data.
4324 Unless this function is called, the first search or match using
4325 PATTERN_BUFFER will allocate its own register data, without
4326 freeing the old data. */
4328 void
4329 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4330 {
4332 {
4337 }
4338 else
4339 {
4343 }
4344 }
4346 \f
4347 /* Searching routines. */
4349 /* Like re_search_2, below, but only one string is specified, and
4350 doesn't let you say where to stop matching. */
4352 int
4353 re_search (struct re_pattern_buffer *bufp, const char *string, int size, int startpos, int range, struct re_registers *regs)
4354 {
4357 }
4360 /* Head address of virtual concatenation of string. */
4361 #define HEAD_ADDR_VSTRING(P) \
4362 (((P) >= size1 ? string2 : string1))
4364 /* End address of virtual concatenation of string. */
4365 #define STOP_ADDR_VSTRING(P) \
4366 (((P) >= size1 ? string2 + size2 : string1 + size1))
4368 /* Address of POS in the concatenation of virtual string. */
4369 #define POS_ADDR_VSTRING(POS) \
4370 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4372 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4373 virtual concatenation of STRING1 and STRING2, starting first at index
4374 STARTPOS, then at STARTPOS + 1, and so on.
4376 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4378 RANGE is how far to scan while trying to match. RANGE = 0 means try
4379 only at STARTPOS; in general, the last start tried is STARTPOS +
4380 RANGE.
4382 In REGS, return the indices of the virtual concatenation of STRING1
4383 and STRING2 that matched the entire BUFP->buffer and its contained
4384 subexpressions.
4386 Do not consider matching one past the index STOP in the virtual
4387 concatenation of STRING1 and STRING2.
4389 We return either the position in the strings at which the match was
4390 found, -1 if no match, or -2 if error (such as failure
4391 stack overflow). */
4393 int
4394 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)
4395 {
4403 boolean anchored_start;
4404 /* Nonzero if we are searching multibyte string. */
4407 /* Check for out-of-range STARTPOS. */
4411 /* Fix up RANGE if it might eventually take us outside
4412 the virtual concatenation of STRING1 and STRING2.
4413 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4419 /* If the search isn't to be a backwards one, don't waste time in a
4420 search for a pattern anchored at beginning of buffer. */
4422 {
4425 else
4427 }
4429 #ifdef emacs
4430 /* In a forward search for something that starts with \=.
4431 don't keep searching past point. */
4433 {
4437 }
4440 /* Update the fastmap now if not correct already. */
4444 /* See whether the pattern is anchored. */
4447 #ifdef emacs
4449 {
4453 }
4454 #endif
4456 /* Loop through the string, looking for a place to start matching. */
4458 {
4459 /* If the pattern is anchored,
4460 skip quickly past places we cannot match.
4461 We don't bother to treat startpos == 0 specially
4462 because that case doesn't repeat. */
4464 {
4469 }
4471 /* If a fastmap is supplied, skip quickly over characters that
4472 cannot be the start of a match. If the pattern can match the
4473 null string, however, we don't need to skip characters; we want
4474 the first null string. */
4476 {
4483 {
4490 /* Written out as an if-else to avoid testing `translate'
4491 inside the loop. */
4493 {
4496 {
4506 }
4507 else
4509 {
4520 d++;
4521 range--;
4522 }
4523 }
4524 else
4525 {
4528 {
4536 }
4537 else
4539 {
4540 d++;
4541 range--;
4542 }
4543 }
4545 }
4547 {
4549 {
4554 }
4555 else
4556 {
4567 }
4568 }
4569 }
4571 /* If can't match the null string, and that's all we have left, fail. */
4585 advance:
4589 {
4590 /* Update STARTPOS to the next character boundary. */
4592 {
4601 }
4602 else
4603 {
4604 range--;
4605 startpos++;
4606 }
4607 }
4608 else
4609 {
4610 range++;
4611 startpos--;
4613 /* Update STARTPOS to the previous character boundary. */
4615 {
4620 /* Find the head of multibyte form. */
4627 }
4628 }
4629 }
4633 \f
4634 /* Declarations and macros for re_match_2. */
4638 RE_TRANSLATE_TYPE translate,
4641 /* This converts PTR, a pointer into one of the search strings `string1'
4642 and `string2' into an offset from the beginning of that string. */
4643 #define POINTER_TO_OFFSET(ptr) \
4644 (FIRST_STRING_P (ptr) \
4645 ? ((regoff_t) ((ptr) - string1)) \
4646 : ((regoff_t) ((ptr) - string2 + size1)))
4648 /* Call before fetching a character with *d. This switches over to
4649 string2 if necessary.
4650 Check re_match_2_internal for a discussion of why end_match_2 might
4651 not be within string2 (but be equal to end_match_1 instead). */
4652 #define PREFETCH() \
4653 while (d == dend) \
4654 { \
4656 if (dend == end_match_2) \
4657 goto fail; \
4659 d = string2; \
4660 dend = end_match_2; \
4661 }
4663 /* Call before fetching a char with *d if you already checked other limits.
4664 This is meant for use in lookahead operations like wordend, etc..
4665 where we might need to look at parts of the string that might be
4666 outside of the LIMITs (i.e past `stop'). */
4667 #define PREFETCH_NOLIMIT() \
4668 if (d == end1) \
4669 { \
4670 d = string2; \
4671 dend = end_match_2; \
4672 } \
4674 /* Test if at very beginning or at very end of the virtual concatenation
4675 of `string1' and `string2'. If only one string, it's `string2'. */
4676 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4677 #define AT_STRINGS_END(d) ((d) == end2)
4680 /* Test if D points to a character which is word-constituent. We have
4681 two special cases to check for: if past the end of string1, look at
4682 the first character in string2; and if before the beginning of
4683 string2, look at the last character in string1. */
4684 #define WORDCHAR_P(d) \
4685 (SYNTAX ((d) == end1 ? *string2 \
4686 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4687 == Sword)
4689 /* Disabled due to a compiler bug -- see comment at case wordbound */
4691 /* The comment at case wordbound is following one, but we don't use
4692 AT_WORD_BOUNDARY anymore to support multibyte form.
4694 The DEC Alpha C compiler 3.x generates incorrect code for the
4695 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4696 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4697 macro and introducing temporary variables works around the bug. */
4699 #if 0
4700 /* Test if the character before D and the one at D differ with respect
4701 to being word-constituent. */
4702 #define AT_WORD_BOUNDARY(d) \
4703 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4704 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4705 #endif
4707 /* Free everything we malloc. */
4708 #ifdef MATCH_MAY_ALLOCATE
4709 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4710 # define FREE_VARIABLES() \
4711 do { \
4712 REGEX_FREE_STACK (fail_stack.stack); \
4713 FREE_VAR (regstart); \
4714 FREE_VAR (regend); \
4715 FREE_VAR (best_regstart); \
4716 FREE_VAR (best_regend); \
4717 } while (0)
4718 #else
4722 \f
4723 /* Optimization routines. */
4725 /* If the operation is a match against one or more chars,
4726 return a pointer to the next operation, else return NULL. */
4729 {
4731 {
4742 {
4747 }
4748 else
4754 #ifdef emacs
4758 p++;
4763 }
4765 }
4768 /* Jump over non-matching operations. */
4771 {
4774 {
4776 {
4789 }
4790 }
4793 }
4795 /* Non-zero if "p1 matches something" implies "p2 fails". */
4796 static int
4798 {
4799 re_opcode_t op2;
4806 /* Skip over open/close-group commands.
4807 If what follows this loop is a ...+ construct,
4808 look at what begins its body, since we will have to
4809 match at least one of that. */
4811 /* The same skip can be done for p1, except that this function
4812 is only used in the case where p1 is a simple match operator. */
4813 /* p1 = skip_noops (p1, pend); */
4821 {
4824 /* If we're at the end of the pattern, we can change. */
4826 {
4829 }
4834 {
4835 register re_wchar_t c
4840 {
4842 {
4845 }
4846 }
4850 {
4853 /* Test if C is listed in charset (or charset_not)
4854 at `p1'. */
4856 {
4860 }
4864 /* `not' is equal to 1 if c would match, which means
4865 that we can't change to pop_failure_jump. */
4867 {
4870 }
4871 }
4874 {
4877 }
4878 }
4882 {
4884 /* Reuse the code above. */
4887 /* It is hard to list up all the character in charset
4888 P2 if it includes multibyte character. Give up in
4889 such case. */
4891 {
4892 /* Now, we are sure that P2 has no range table.
4893 So, for the size of bitmap in P2, `p2[1]' is
4894 enough. But P1 may have range table, so the
4895 size of bitmap table of P1 is extracted by
4896 using macro `CHARSET_BITMAP_SIZE'.
4898 In a multibyte case, we know that all the character
4899 listed in P2 is ASCII. In a unibyte case, P1 has only a
4900 bitmap table. So, in both cases, it is enough to test
4901 only the bitmap table of P1. */
4904 {
4906 /* We win if the charset inside the loop
4907 has no overlap with the one after the loop. */
4911 idx++)
4917 {
4920 }
4921 }
4923 {
4925 /* We win if the charset_not inside the loop lists
4926 every character listed in the charset after. */
4934 {
4937 }
4938 }
4939 }
4940 }
4945 {
4948 /* Reuse the code above. */
4951 /* When we have two charset_not, it's very unlikely that
4952 they don't overlap. The union of the two sets of excluded
4953 chars should cover all possible chars, which, as a matter of
4954 fact, is virtually impossible in multibyte buffers. */
4956 }
4980 #ifdef emacs
4988 ;
4989 }
4991 /* Safe default. */
4993 }
4995 \f
4996 /* Matching routines. */
4999 /* re_match is like re_match_2 except it takes only a single string. */
5001 int
5007 {
5011 }
5015 #ifdef emacs
5016 /* In Emacs, this is the string or buffer in which we
5017 are matching. It is used for looking up syntax properties. */
5018 Lisp_Object re_match_object;
5019 #endif
5021 /* re_match_2 matches the compiled pattern in BUFP against the
5022 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5023 and SIZE2, respectively). We start matching at POS, and stop
5024 matching at STOP.
5026 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5027 store offsets for the substring each group matched in REGS. See the
5028 documentation for exactly how many groups we fill.
5030 We return -1 if no match, -2 if an internal error (such as the
5031 failure stack overflowing). Otherwise, we return the length of the
5032 matched substring. */
5034 int
5035 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)
5036 {
5039 #ifdef emacs
5044 #endif
5050 }
5054 /* This is a separate function so that we can force an alloca cleanup
5055 afterwards. */
5056 static int
5057 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)
5058 {
5059 /* General temporaries. */
5064 /* Just past the end of the corresponding string. */
5067 /* Pointers into string1 and string2, just past the last characters in
5068 each to consider matching. */
5071 /* Where we are in the data, and the end of the current string. */
5074 /* Used sometimes to remember where we were before starting matching
5075 an operator so that we can go back in case of failure. This "atomic"
5076 behavior of matching opcodes is indispensable to the correctness
5077 of the on_failure_keep_string_jump optimization. */
5080 /* Where we are in the pattern, and the end of the pattern. */
5084 /* We use this to map every character in the string. */
5087 /* Nonzero if BUFP is setup from a multibyte regex. */
5090 /* Nonzero if STRING1/STRING2 are multibyte. */
5093 /* Failure point stack. Each place that can handle a failure further
5094 down the line pushes a failure point on this stack. It consists of
5095 regstart, and regend for all registers corresponding to
5096 the subexpressions we're currently inside, plus the number of such
5097 registers, and, finally, two char *'s. The first char * is where
5098 to resume scanning the pattern; the second one is where to resume
5099 scanning the strings. */
5101 fail_stack_type fail_stack;
5102 #endif
5103 #ifdef DEBUG
5105 #endif
5107 #if defined REL_ALLOC && defined REGEX_MALLOC
5108 /* This holds the pointer to the failure stack, when
5109 it is allocated relocatably. */
5111 #endif
5113 /* We fill all the registers internally, independent of what we
5114 return, for use in backreferences. The number here includes
5115 an element for register zero. */
5118 /* Information on the contents of registers. These are pointers into
5119 the input strings; they record just what was matched (on this
5120 attempt) by a subexpression part of the pattern, that is, the
5121 regnum-th regstart pointer points to where in the pattern we began
5122 matching and the regnum-th regend points to right after where we
5123 stopped matching the regnum-th subexpression. (The zeroth register
5124 keeps track of what the whole pattern matches.) */
5127 #endif
5129 /* The following record the register info as found in the above
5130 variables when we find a match better than any we've seen before.
5131 This happens as we backtrack through the failure points, which in
5132 turn happens only if we have not yet matched the entire string. */
5136 #endif
5138 /* Logically, this is `best_regend[0]'. But we don't want to have to
5139 allocate space for that if we're not allocating space for anything
5140 else (see below). Also, we never need info about register 0 for
5141 any of the other register vectors, and it seems rather a kludge to
5142 treat `best_regend' differently than the rest. So we keep track of
5143 the end of the best match so far in a separate variable. We
5144 initialize this to NULL so that when we backtrack the first time
5145 and need to test it, it's not garbage. */
5148 #ifdef DEBUG
5149 /* Counts the total number of registers pushed. */
5151 #endif
5157 #ifdef MATCH_MAY_ALLOCATE
5158 /* Do not bother to initialize all the register variables if there are
5159 no groups in the pattern, as it takes a fair amount of time. If
5160 there are groups, we include space for register 0 (the whole
5161 pattern), even though we never use it, since it simplifies the
5162 array indexing. We should fix this. */
5164 {
5171 {
5174 }
5175 }
5176 else
5177 {
5178 /* We must initialize all our variables to NULL, so that
5179 `FREE_VARIABLES' doesn't try to free them. */
5181 }
5184 /* The starting position is bogus. */
5186 {
5189 }
5191 /* Initialize subexpression text positions to -1 to mark ones that no
5192 start_memory/stop_memory has been seen for. Also initialize the
5193 register information struct. */
5197 /* We move `string1' into `string2' if the latter's empty -- but not if
5198 `string1' is null. */
5200 {
5205 }
5209 /* `p' scans through the pattern as `d' scans through the data.
5210 `dend' is the end of the input string that `d' points within. `d'
5211 is advanced into the following input string whenever necessary, but
5212 this happens before fetching; therefore, at the beginning of the
5213 loop, `d' can be pointing at the end of a string, but it cannot
5214 equal `string2'. */
5216 {
5217 /* Only match within string2. */
5221 }
5222 else
5223 {
5225 {
5226 /* Only match within string1. */
5228 /* BEWARE!
5229 When we reach end_match_1, PREFETCH normally switches to string2.
5230 But in the present case, this means that just doing a PREFETCH
5231 makes us jump from `stop' to `gap' within the string.
5232 What we really want here is for the search to stop as
5233 soon as we hit end_match_1. That's why we set end_match_2
5234 to end_match_1 (since PREFETCH fails as soon as we hit
5235 end_match_2). */
5237 }
5238 else
5240 moving `d' from end1 to string2 will not prevent the d == dend
5241 check from catching the end of string. */
5244 }
5247 }
5255 /* This loops over pattern commands. It exits by returning from the
5256 function if the match is complete, or it drops through if the match
5257 fails at this starting point in the input data. */
5259 {
5266 /* If we haven't matched the entire string, and we want the
5267 longest match, try backtracking. */
5269 {
5270 /* 1 if this match ends in the same string (string1 or string2)
5271 as the best previous match. */
5274 /* 1 if this match is the best seen so far. */
5275 boolean best_match_p;
5277 /* AIX compiler got confused when this was combined
5278 with the previous declaration. */
5281 else
5289 /* If exceeds best match so far, save it. */
5291 {
5298 {
5301 }
5302 }
5304 }
5306 /* If no failure points, don't restore garbage. And if
5307 last match is real best match, don't restore second
5308 best one. */
5310 {
5311 restore_best_regs:
5312 /* Restore best match. It may happen that `dend ==
5313 end_match_1' while the restored d is in string2.
5314 For example, the pattern `x.*y.*z' against the
5315 strings `x-' and `y-z-', if the two strings are
5316 not consecutive in memory. */
5324 {
5327 }
5328 }
5331 succeed_label:
5334 /* If caller wants register contents data back, do it. */
5336 {
5337 /* Have the register data arrays been allocated? */
5340 extra element beyond `num_regs' for the `-1' marker
5341 GNU code uses. */
5346 {
5349 }
5351 }
5354 allocated, reallocate them. If we need fewer, just
5355 leave it alone. */
5357 {
5362 {
5365 }
5366 }
5367 }
5368 else
5369 {
5370 /* These braces fend off a "empty body in an else-statement"
5371 warning under GCC when assert expands to nothing. */
5373 }
5375 /* Convert the pointer data in `regstart' and `regend' to
5376 indices. Register zero has to be set differently,
5377 since we haven't kept track of any info for it. */
5379 {
5382 }
5384 /* Go through the first `min (num_regs, regs->num_regs)'
5385 registers, since that is all we initialized. */
5387 {
5390 else
5391 {
5396 }
5397 }
5399 /* If the regs structure we return has more elements than
5400 were in the pattern, set the extra elements to -1. If
5401 we (re)allocated the registers, this is the case,
5402 because we always allocate enough to have at least one
5403 -1 at the end. */
5419 }
5421 /* Otherwise match next pattern command. */
5423 {
5424 /* Ignore these. Used to ignore the n of succeed_n's which
5425 currently have n == 0. */
5434 /* Match the next n pattern characters exactly. The following
5435 byte in the pattern defines n, and the n bytes after that
5436 are the characters to match. */
5441 /* Remember the start point to rollback upon failure. */
5444 #ifndef emacs
5445 /* This is written out as an if-else so we don't waste time
5446 testing `translate' inside the loop. */
5448 do
5449 {
5452 {
5455 }
5456 d++;
5457 }
5459 else
5460 do
5461 {
5464 {
5467 }
5468 }
5471 /* The cost of testing `translate' is comparatively small. */
5473 do
5474 {
5481 else
5482 {
5485 }
5489 {
5492 }
5497 }
5499 else
5500 do
5501 {
5507 {
5510 }
5511 else
5512 {
5515 }
5518 {
5523 }
5524 else
5527 {
5530 }
5532 d++;
5533 }
5535 #endif
5539 /* Match any character except possibly a newline or a null. */
5541 {
5543 re_wchar_t buf_ch;
5549 target_multibyte);
5560 }
5566 {
5571 /* Start of actual range_table, or end of bitmap if there is no
5572 range table. */
5575 /* Nonzero if there is a range table. */
5578 /* Number of ranges of range table. This is not included
5579 in the initial byte-length of the command. */
5582 /* Whether matching against a unibyte character. */
5590 {
5593 }
5598 {
5604 {
5607 }
5608 }
5609 else
5610 {
5614 {
5618 {
5621 }
5622 }
5623 else
5625 }
5629 /* Cast to `unsigned' instead of `unsigned char' in
5630 case the bit list is a full 32 bytes long. */
5634 }
5635 #ifdef emacs
5637 {
5647 else
5649 }
5654 else
5661 }
5664 /* The beginning of a group is represented by start_memory.
5665 The argument is the register number. The text
5666 matched within the group is recorded (in the internal
5667 registers data structure) under the register number. */
5671 /* In case we need to undo this operation (via backtracking). */
5678 /* Move past the register number and inner group count. */
5683 /* The stop_memory opcode represents the end of a group. Its
5684 argument is the same as start_memory's: the register number. */
5689 /* Strictly speaking, there should be code such as:
5691 assert (REG_UNSET (regend[*p]));
5692 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5694 But the only info to be pushed is regend[*p] and it is known to
5695 be UNSET, so there really isn't anything to push.
5696 Not pushing anything, on the other hand deprives us from the
5697 guarantee that regend[*p] is UNSET since undoing this operation
5698 will not reset its value properly. This is not important since
5699 the value will only be read on the next start_memory or at
5700 the very end and both events can only happen if this stop_memory
5701 is *not* undone. */
5706 /* Move past the register number and the inner group count. */
5711 /* \<digit> has been turned into a `duplicate' command which is
5712 followed by the numeric value of <digit> as the register number. */
5714 {
5719 /* Can't back reference a group which we've never matched. */
5723 /* Where in input to try to start matching. */
5726 /* Remember the start point to rollback upon failure. */
5729 /* Where to stop matching; if both the place to start and
5730 the place to stop matching are in the same string, then
5731 set to the place to stop, otherwise, for now have to use
5732 the end of the first string. */
5738 {
5739 /* If necessary, advance to next segment in register
5740 contents. */
5742 {
5746 /* End of string1 => advance to string2. */
5749 }
5750 /* At end of register contents => success */
5753 /* If necessary, advance to next segment in data. */
5756 /* How many characters left in this segment to match. */
5759 /* Want how many consecutive characters we can match in
5760 one shot, so, if necessary, adjust the count. */
5764 /* Compare that many; failure if mismatch, else move
5765 past them. */
5769 {
5772 }
5774 }
5775 }
5779 /* begline matches the empty string at the beginning of the string
5780 (unless `not_bol' is set in `bufp'), and after newlines. */
5785 {
5787 }
5788 else
5789 {
5794 }
5795 /* In all other cases, we fail. */
5799 /* endline is the dual of begline. */
5804 {
5806 }
5807 else
5808 {
5812 }
5816 /* Match at the very beginning of the data. */
5824 /* Match at the very end of the data. */
5832 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5833 pushes NULL as the value for the string on the stack. Then
5834 `POP_FAILURE_POINT' will keep the current value for the
5835 string, instead of restoring it. To see why, consider
5836 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5837 then the . fails against the \n. But the next thing we want
5838 to do is match the \n against the \n; if we restored the
5839 string value, we would be back at the foo.
5841 Because this is used only in specific cases, we don't need to
5842 check all the things that `on_failure_jump' does, to make
5843 sure the right things get saved on the stack. Hence we don't
5844 share its code. The only reason to push anything on the
5845 stack at all is that otherwise we would have to change
5846 `anychar's code to do something besides goto fail in this
5847 case; that seems worse than this. */
5856 /* A nasty loop is introduced by the non-greedy *? and +?.
5857 With such loops, the stack only ever contains one failure point
5858 at a time, so that a plain on_failure_jump_loop kind of
5859 cycle detection cannot work. Worse yet, such a detection
5860 can not only fail to detect a cycle, but it can also wrongly
5861 detect a cycle (between different instantiations of the same
5862 loop).
5863 So the method used for those nasty loops is a little different:
5864 We use a special cycle-detection-stack-frame which is pushed
5865 when the on_failure_jump_nastyloop failure-point is *popped*.
5866 This special frame thus marks the beginning of one iteration
5867 through the loop and we can hence easily check right here
5868 whether something matched between the beginning and the end of
5869 the loop. */
5876 {
5880 /* If there's a cycle, just continue without pushing
5881 this failure point. The failure point is the "try again"
5882 option, which shouldn't be tried.
5883 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5885 }
5888 /* Simple loop detecting on_failure_jump: just check on the
5889 failure stack if the same spot was already hit earlier. */
5891 on_failure:
5895 {
5899 /* If there's a cycle, get out of the loop, as if the matching
5900 had failed. We used to just `goto fail' here, but that was
5901 aborting the search a bit too early: we want to keep the
5902 empty-loop-match and keep matching after the loop.
5903 We want (x?)*y\1z to match both xxyz and xxyxz. */
5905 else
5907 }
5911 /* Uses of on_failure_jump:
5913 Each alternative starts with an on_failure_jump that points
5914 to the beginning of the next alternative. Each alternative
5915 except the last ends with a jump that in effect jumps past
5916 the rest of the alternatives. (They really jump to the
5917 ending jump of the following alternative, because tensioning
5918 these jumps is a hassle.)
5920 Repeats start with an on_failure_jump that points past both
5921 the repetition text and either the following jump or
5922 pop_failure_jump back to this on_failure_jump. */
5931 /* This operation is used for greedy *.
5932 Compare the beginning of the repeat with what in the
5933 pattern follows its end. If we can establish that there
5934 is nothing that they would both match, i.e., that we
5935 would have to backtrack because of (as in, e.g., `a*a')
5936 then we can use a non-backtracking loop based on
5937 on_failure_keep_string_jump instead of on_failure_jump. */
5942 {
5944 /* Here, we discard `const', making re_match non-reentrant. */
5949 instruction once it's been changed. */
5953 /* Ensure this is a indeed the trivial kind of loop
5954 we are expecting. */
5959 {
5960 /* Use a fast `on_failure_keep_string_jump' loop. */
5964 }
5965 else
5966 {
5967 /* Default to a safe `on_failure_jump' loop. */
5970 }
5972 }
5975 /* Unconditionally jump (without popping any failure points). */
5977 unconditional_jump:
5978 IMMEDIATE_QUIT_CHECK;
5986 /* Have to succeed matching what follows at least n times.
5987 After that, handle like `on_failure_jump'. */
5989 /* Signedness doesn't matter since we only compare MCNT to 0. */
5993 /* Originally, mcnt is how many times we HAVE to succeed. */
5995 {
5996 /* Here, we discard `const', making re_match non-reentrant. */
5998 mcnt--;
6001 }
6002 else
6003 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
6008 /* Signedness doesn't matter since we only compare MCNT to 0. */
6012 /* Originally, this is how many times we CAN jump. */
6014 {
6015 /* Here, we discard `const', making re_match non-reentrant. */
6017 mcnt--;
6020 }
6021 /* If don't have to jump any more, skip over the rest of command. */
6022 else
6027 {
6032 /* Here, we discard `const', making re_match non-reentrant. */
6034 /* Signedness doesn't matter since we only copy MCNT's bits . */
6039 }
6046 /* We SUCCEED (or FAIL) in one of the following cases: */
6048 /* Case 1: D is at the beginning or the end of string. */
6051 else
6052 {
6053 /* C1 is the character before D, S1 is the syntax of C1, C2
6054 is the character at D, and S2 is the syntax of C2. */
6058 #ifdef emacs
6062 #endif
6065 #ifdef emacs
6067 #endif
6074 /* Case 3: Both of S1 and S2 are Sword, and macro
6075 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6078 }
6081 else
6087 /* We FAIL in one of the following cases: */
6089 /* Case 1: D is at the end of string. */
6092 else
6093 {
6094 /* C1 is the character before D, S1 is the syntax of C1, C2
6095 is the character at D, and S2 is the syntax of C2. */
6099 #ifdef emacs
6103 #endif
6108 /* Case 2: S2 is not Sword. */
6112 /* Case 3: D is not at the beginning of string ... */
6114 {
6116 #ifdef emacs
6118 #endif
6121 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6122 returns 0. */
6125 }
6126 }
6132 /* We FAIL in one of the following cases: */
6134 /* Case 1: D is at the beginning of string. */
6137 else
6138 {
6139 /* C1 is the character before D, S1 is the syntax of C1, C2
6140 is the character at D, and S2 is the syntax of C2. */
6144 #ifdef emacs
6148 #endif
6152 /* Case 2: S1 is not Sword. */
6156 /* Case 3: D is not at the end of string ... */
6158 {
6161 #ifdef emacs
6163 #endif
6166 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6167 returns 0. */
6170 }
6171 }
6177 /* We FAIL in one of the following cases: */
6179 /* Case 1: D is at the end of string. */
6182 else
6183 {
6184 /* C1 is the character before D, S1 is the syntax of C1, C2
6185 is the character at D, and S2 is the syntax of C2. */
6188 #ifdef emacs
6192 #endif
6197 /* Case 2: S2 is neither Sword nor Ssymbol. */
6201 /* Case 3: D is not at the beginning of string ... */
6203 {
6205 #ifdef emacs
6207 #endif
6210 /* ... and S1 is Sword or Ssymbol. */
6213 }
6214 }
6220 /* We FAIL in one of the following cases: */
6222 /* Case 1: D is at the beginning of string. */
6225 else
6226 {
6227 /* C1 is the character before D, S1 is the syntax of C1, C2
6228 is the character at D, and S2 is the syntax of C2. */
6231 #ifdef emacs
6235 #endif
6239 /* Case 2: S1 is neither Ssymbol nor Sword. */
6243 /* Case 3: D is not at the end of string ... */
6245 {
6248 #ifdef emacs
6250 #endif
6253 /* ... and S2 is Sword or Ssymbol. */
6256 }
6257 }
6266 #ifdef emacs
6267 {
6271 }
6272 #endif
6273 {
6275 re_wchar_t c;
6281 }
6284 #ifdef emacs
6309 {
6311 re_wchar_t c;
6317 }
6324 }
6328 /* We goto here if a matching operation fails. */
6329 fail:
6330 IMMEDIATE_QUIT_CHECK;
6332 {
6334 /* A restart point is known. Restore to that state. */
6338 {
6346 /* Fallthrough */
6352 continue_failure_jump:
6358 /* A special frame used for nastyloops. */
6363 }
6369 }
6370 else
6381 \f
6382 /* Subroutine definitions for re_match_2. */
6384 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6385 bytes; nonzero otherwise. */
6387 static int
6390 {
6395 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6396 different lengths, but relying on a single `len' would break this. -sm */
6398 {
6410 }
6416 }
6417 \f
6418 /* Entry points for GNU code. */
6420 /* re_compile_pattern is the GNU regular expression compiler: it
6421 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6422 Returns 0 if the pattern was valid, otherwise an error string.
6424 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6425 are set in BUFP on entry.
6427 We call regex_compile to do the actual compilation. */
6431 {
6432 reg_errcode_t ret;
6434 /* GNU code is written to assume at least RE_NREGS registers will be set
6435 (and at least one extra will be -1). */
6438 /* And GNU code determines whether or not to get register information
6439 by passing null for the REGS argument to re_match, etc., not by
6440 setting no_sub. */
6448 }
6450 \f
6451 /* Entry points compatible with 4.2 BSD regex library. We don't define
6452 them unless specifically requested. */
6454 #if defined _REGEX_RE_COMP || defined _LIBC
6456 /* BSD has one and only one pattern buffer. */
6460 # ifdef _LIBC
6461 /* Make these definitions weak in libc, so POSIX programs can redefine
6462 these names if they don't use our functions, and still use
6463 regcomp/regexec below without link errors. */
6464 weak_function
6465 # endif
6468 {
6469 reg_errcode_t ret;
6472 {
6474 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6477 }
6480 {
6483 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6489 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6491 }
6493 /* Since `re_exec' always passes NULL for the `regs' argument, we
6494 don't need to initialize the pattern buffer fields which affect it. */
6501 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6503 }
6506 int
6507 # ifdef _LIBC
6508 weak_function
6509 # endif
6512 {
6514 return
6516 }
6518 \f
6519 /* POSIX.2 functions. Don't define these for Emacs. */
6521 #ifndef emacs
6523 /* regcomp takes a regular expression as a string and compiles it.
6525 PREG is a regex_t *. We do not expect any fields to be initialized,
6526 since POSIX says we shouldn't. Thus, we set
6528 `buffer' to the compiled pattern;
6529 `used' to the length of the compiled pattern;
6530 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6531 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6532 RE_SYNTAX_POSIX_BASIC;
6533 `fastmap' to an allocated space for the fastmap;
6534 `fastmap_accurate' to zero;
6535 `re_nsub' to the number of subexpressions in PATTERN.
6537 PATTERN is the address of the pattern string.
6539 CFLAGS is a series of bits which affect compilation.
6541 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6542 use POSIX basic syntax.
6544 If REG_NEWLINE is set, then . and [^...] don't match newline.
6545 Also, regexec will try a match beginning after every newline.
6547 If REG_ICASE is set, then we considers upper- and lowercase
6548 versions of letters to be equivalent when matching.
6550 If REG_NOSUB is set, then when PREG is passed to regexec, that
6551 routine will report only success or failure, and nothing about the
6552 registers.
6554 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6555 the return codes and their meanings.) */
6557 int
6562 {
6563 reg_errcode_t ret;
6564 reg_syntax_t syntax
6568 /* regex_compile will allocate the space for the compiled pattern. */
6573 /* Try to allocate space for the fastmap. */
6577 {
6580 preg->translate
6586 /* Map uppercase characters to corresponding lowercase ones. */
6589 }
6590 else
6593 /* If REG_NEWLINE is set, newlines are treated differently. */
6598 }
6599 else
6604 /* POSIX says a null character in the pattern terminates it, so we
6605 can use strlen here in compiling the pattern. */
6608 /* POSIX doesn't distinguish between an unmatched open-group and an
6609 unmatched close-group: both are REG_EPAREN. */
6615 buffer. */
6621 }
6622 }
6624 }
6628 /* regexec searches for a given pattern, specified by PREG, in the
6629 string STRING.
6631 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6632 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6633 least NMATCH elements, and we set them to the offsets of the
6634 corresponding matched substrings.
6636 EFLAGS specifies `execution flags' which affect matching: if
6637 REG_NOTBOL is set, then ^ does not match at the beginning of the
6638 string; if REG_NOTEOL is set, then $ does not match at the end.
6640 We return 0 if we find a match and REG_NOMATCH if not. */
6642 int
6649 {
6652 regex_t private_preg;
6661 /* The user has told us exactly how many registers to return
6662 information about, via `nmatch'. We have to pass that on to the
6663 matching routines. */
6667 {
6673 }
6675 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6676 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6677 was a little bit longer but still only matching the real part.
6678 This works because the `endline' will check for a '\n' and will find a
6679 '\0', correctly deciding that this is not the end of a line.
6680 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6681 a convenient '\0' there. For all we know, the string could be preceded
6682 by '\n' which would throw things off. */
6684 /* Perform the searching operation. */
6689 /* Copy the register information to the POSIX structure. */
6691 {
6693 {
6697 {
6700 }
6701 }
6703 /* If we needed the temporary register info, free the space now. */
6705 }
6707 /* We want zero return to mean success, unlike `re_search'. */
6709 }
6713 /* Returns a message corresponding to an error code, ERR_CODE, returned
6714 from either regcomp or regexec. We don't use PREG here.
6716 ERR_CODE was previously called ERRCODE, but that name causes an
6717 error with msvc8 compiler. */
6719 size_t
6725 {
6731 /* Only error codes returned by the rest of the code should be passed
6732 to this routine. If we are given anything else, or if other regex
6733 code generates an invalid error code, then the program has a bug.
6734 Dump core so we can fix it. */
6742 {
6744 {
6747 }
6748 else
6750 }
6753 }
6757 /* Free dynamically allocated space used by PREG. */
6759 void
6762 {
6775 }
6780 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6781 (do not change this comment) */