| blob | e19183fcf526c7c8135e436dab0668ce33a7350a |
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
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
123 /* Make syntax table lookup grant data in gl_state. */
124 # define SYNTAX_ENTRY_VIA_PROPERTY
130 # ifdef malloc
131 # undef malloc
132 # endif
133 # define malloc xmalloc
134 # ifdef realloc
135 # undef realloc
136 # endif
137 # define realloc xrealloc
138 # ifdef free
139 # undef free
140 # endif
141 # define free xfree
143 /* Converts the pointer to the char to BEG-based offset from the start. */
144 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
145 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
147 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
148 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
149 # define RE_STRING_CHAR(p, s, multibyte) \
150 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
151 # define RE_STRING_CHAR_AND_LENGTH(p, s, len, multibyte) \
152 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
154 # define RE_CHAR_TO_MULTIBYTE(c) unibyte_to_multibyte_table[(c)]
156 # define RE_CHAR_TO_UNIBYTE(c) \
157 (ASCII_CHAR_P (c) ? (c) \
158 : CHAR_BYTE8_P (c) ? CHAR_TO_BYTE8 (c) \
159 : multibyte_char_to_unibyte_safe (c))
161 /* Set C a (possibly converted to multibyte) character before P. P
162 points into a string which is the virtual concatenation of STR1
163 (which ends at END1) or STR2 (which ends at END2). */
164 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
165 do { \
166 if (target_multibyte) \
167 { \
168 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
169 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
170 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
171 c = STRING_CHAR (dtemp, (p) - dtemp); \
172 } \
173 else \
174 { \
175 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
176 (c) = RE_CHAR_TO_MULTIBYTE (c); \
177 } \
178 } while (0)
180 /* Set C a (possibly converted to multibyte) character at P, and set
181 LEN to the byte length of that character. */
182 # define GET_CHAR_AFTER(c, p, len) \
183 do { \
184 if (target_multibyte) \
185 (c) = STRING_CHAR_AND_LENGTH (p, 0, len); \
186 else \
187 { \
188 (c) = *p; \
189 len = 1; \
190 (c) = RE_CHAR_TO_MULTIBYTE (c); \
191 } \
192 } while (0)
196 /* If we are not linking with Emacs proper,
197 we can't use the relocating allocator
198 even if config.h says that we can. */
199 # undef REL_ALLOC
201 # if defined STDC_HEADERS || defined _LIBC
202 # include <stdlib.h>
203 # else
206 # endif
208 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
213 {
217 {
220 }
222 }
228 {
230 /* We must call malloc explicitly when BLOCK is 0, since some
231 reallocs don't do this. */
234 else
237 {
240 }
242 }
244 # ifdef malloc
245 # undef malloc
246 # endif
247 # define malloc xmalloc
248 # ifdef realloc
249 # undef realloc
250 # endif
251 # define realloc xrealloc
253 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
254 If nothing else has been done, use the method below. */
255 # ifdef INHIBIT_STRING_HEADER
256 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
257 # if !defined bzero && !defined bcopy
258 # undef INHIBIT_STRING_HEADER
259 # endif
260 # endif
261 # endif
263 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
264 This is used in most programs--a few other programs avoid this
265 by defining INHIBIT_STRING_HEADER. */
266 # ifndef INHIBIT_STRING_HEADER
267 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
268 # include <string.h>
269 # ifndef bzero
270 # ifndef _LIBC
272 # else
273 # define bzero(s, n) __bzero (s, n)
274 # endif
275 # endif
276 # else
277 # include <strings.h>
278 # ifndef memcmp
279 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
280 # endif
281 # ifndef memcpy
282 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
283 # endif
284 # endif
285 # endif
287 /* Define the syntax stuff for \<, \>, etc. */
289 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
292 # define SWITCH_ENUM_CAST(x) (x)
294 /* Dummy macros for non-Emacs environments. */
295 # define BASE_LEADING_CODE_P(c) (0)
296 # define CHAR_CHARSET(c) 0
297 # define CHARSET_LEADING_CODE_BASE(c) 0
298 # define MAX_MULTIBYTE_LENGTH 1
299 # define RE_MULTIBYTE_P(x) 0
300 # define RE_TARGET_MULTIBYTE_P(x) 0
301 # define WORD_BOUNDARY_P(c1, c2) (0)
302 # define CHAR_HEAD_P(p) (1)
303 # define SINGLE_BYTE_CHAR_P(c) (1)
304 # define SAME_CHARSET_P(c1, c2) (1)
305 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
306 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
307 # define STRING_CHAR(p, s) (*(p))
308 # define RE_STRING_CHAR(p, s, multibyte) STRING_CHAR ((p), (s))
309 # define CHAR_STRING(c, s) (*(s) = (c), 1)
310 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
311 # define RE_STRING_CHAR_AND_LENGTH(p, s, len, multibyte) STRING_CHAR_AND_LENGTH ((p), (s), (len))
312 # define RE_CHAR_TO_MULTIBYTE(c) (c)
313 # define RE_CHAR_TO_UNIBYTE(c) (c)
314 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
315 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
316 # define GET_CHAR_AFTER(c, p, len) \
317 (c = *p, len = 1)
318 # define MAKE_CHAR(charset, c1, c2) (c1)
319 # define BYTE8_TO_CHAR(c) (c)
320 # define CHAR_BYTE8_P(c) (0)
321 # define CHAR_LEADING_CODE(c) (c)
325 #ifndef RE_TRANSLATE
326 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
327 # define RE_TRANSLATE_P(TBL) (TBL)
328 #endif
329 \f
330 /* Get the interface, including the syntax bits. */
333 /* isalpha etc. are used for the character classes. */
334 #include <ctype.h>
336 #ifdef emacs
338 /* 1 if C is an ASCII character. */
339 # define IS_REAL_ASCII(c) ((c) < 0200)
341 /* 1 if C is a unibyte character. */
342 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
344 /* The Emacs definitions should not be directly affected by locales. */
346 /* In Emacs, these are only used for single-byte characters. */
353 /* This is only used for single-byte characters. */
356 /* The rest must handle multibyte characters. */
358 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
360 : 1)
362 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
364 : 1)
366 # define ISALNUM(c) (IS_REAL_ASCII (c) \
370 : SYNTAX (c) == Sword)
372 # define ISALPHA(c) (IS_REAL_ASCII (c) \
375 : SYNTAX (c) == Sword)
377 # define ISLOWER(c) (LOWERCASEP (c))
379 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
384 : SYNTAX (c) != Sword)
386 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
388 # define ISUPPER(c) (UPPERCASEP (c))
390 # define ISWORD(c) (SYNTAX (c) == Sword)
394 /* Jim Meyering writes:
396 "... Some ctype macros are valid only for character codes that
397 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
398 using /bin/cc or gcc but without giving an ansi option). So, all
399 ctype uses should be through macros like ISPRINT... If
400 STDC_HEADERS is defined, then autoconf has verified that the ctype
401 macros don't need to be guarded with references to isascii. ...
402 Defining isascii to 1 should let any compiler worth its salt
403 eliminate the && through constant folding."
404 Solaris defines some of these symbols so we must undefine them first. */
406 # undef ISASCII
407 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
408 # define ISASCII(c) 1
409 # else
410 # define ISASCII(c) isascii(c)
411 # endif
413 /* 1 if C is an ASCII character. */
414 # define IS_REAL_ASCII(c) ((c) < 0200)
416 /* This distinction is not meaningful, except in Emacs. */
417 # define ISUNIBYTE(c) 1
419 # ifdef isblank
420 # define ISBLANK(c) (ISASCII (c) && isblank (c))
421 # else
423 # endif
424 # ifdef isgraph
425 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
426 # else
427 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
428 # endif
430 # undef ISPRINT
431 # define ISPRINT(c) (ISASCII (c) && isprint (c))
432 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
433 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
434 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
435 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
436 # define ISLOWER(c) (ISASCII (c) && islower (c))
437 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
438 # define ISSPACE(c) (ISASCII (c) && isspace (c))
439 # define ISUPPER(c) (ISASCII (c) && isupper (c))
440 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
442 # define ISWORD(c) ISALPHA(c)
444 # ifdef _tolower
445 # define TOLOWER(c) _tolower(c)
446 # else
447 # define TOLOWER(c) tolower(c)
448 # endif
450 /* How many characters in the character set. */
451 # define CHAR_SET_SIZE 256
453 # ifdef SYNTAX_TABLE
461 static void
463 {
479 }
483 # define SYNTAX(c) re_syntax_table[(c)]
486 \f
487 #ifndef NULL
488 # define NULL (void *)0
489 #endif
491 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
492 since ours (we hope) works properly with all combinations of
493 machines, compilers, `char' and `unsigned char' argument types.
494 (Per Bothner suggested the basic approach.) */
495 #undef SIGN_EXTEND_CHAR
496 #if __STDC__
497 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
499 /* As in Harbison and Steele. */
500 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
501 #endif
502 \f
503 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
504 use `alloca' instead of `malloc'. This is because using malloc in
505 re_search* or re_match* could cause memory leaks when C-g is used in
506 Emacs; also, malloc is slower and causes storage fragmentation. On
507 the other hand, malloc is more portable, and easier to debug.
509 Because we sometimes use alloca, some routines have to be macros,
510 not functions -- `alloca'-allocated space disappears at the end of the
511 function it is called in. */
513 #ifdef REGEX_MALLOC
515 # define REGEX_ALLOCATE malloc
516 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
517 # define REGEX_FREE free
521 /* Emacs already defines alloca, sometimes. */
522 # ifndef alloca
524 /* Make alloca work the best possible way. */
525 # ifdef __GNUC__
526 # define alloca __builtin_alloca
528 # ifdef HAVE_ALLOCA_H
529 # include <alloca.h>
535 # define REGEX_ALLOCATE alloca
537 /* Assumes a `char *destination' variable. */
538 # define REGEX_REALLOCATE(source, osize, nsize) \
539 (destination = (char *) alloca (nsize), \
540 memcpy (destination, source, osize))
542 /* No need to do anything to free, after alloca. */
547 /* Define how to allocate the failure stack. */
549 #if defined REL_ALLOC && defined REGEX_MALLOC
551 # define REGEX_ALLOCATE_STACK(size) \
552 r_alloc (&failure_stack_ptr, (size))
553 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
554 r_re_alloc (&failure_stack_ptr, (nsize))
555 # define REGEX_FREE_STACK(ptr) \
556 r_alloc_free (&failure_stack_ptr)
560 # ifdef REGEX_MALLOC
562 # define REGEX_ALLOCATE_STACK malloc
563 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
564 # define REGEX_FREE_STACK free
568 # define REGEX_ALLOCATE_STACK alloca
570 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
571 REGEX_REALLOCATE (source, osize, nsize)
572 /* No need to explicitly free anything. */
573 # define REGEX_FREE_STACK(arg) ((void)0)
579 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
580 `string1' or just past its end. This works if PTR is NULL, which is
581 a good thing. */
582 #define FIRST_STRING_P(ptr) \
583 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
585 /* (Re)Allocate N items of type T using malloc, or fail. */
586 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
587 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
588 #define RETALLOC_IF(addr, n, t) \
589 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
590 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
594 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
596 #undef MAX
597 #undef MIN
598 #define MAX(a, b) ((a) > (b) ? (a) : (b))
599 #define MIN(a, b) ((a) < (b) ? (a) : (b))
601 /* Type of source-pattern and string chars. */
605 #define false 0
606 #define true 1
614 \f
615 /* These are the command codes that appear in compiled regular
616 expressions. Some opcodes are followed by argument bytes. A
617 command code can specify any interpretation whatsoever for its
618 arguments. Zero bytes may appear in the compiled regular expression. */
621 {
624 /* Succeed right away--no more backtracking. */
625 succeed,
627 /* Followed by one byte giving n, then by n literal bytes. */
628 exactn,
630 /* Matches any (more or less) character. */
631 anychar,
633 /* Matches any one char belonging to specified set. First
634 following byte is number of bitmap bytes. Then come bytes
635 for a bitmap saying which chars are in. Bits in each byte
636 are ordered low-bit-first. A character is in the set if its
637 bit is 1. A character too large to have a bit in the map is
638 automatically not in the set.
640 If the length byte has the 0x80 bit set, then that stuff
641 is followed by a range table:
642 2 bytes of flags for character sets (low 8 bits, high 8 bits)
643 See RANGE_TABLE_WORK_BITS below.
644 2 bytes, the number of pairs that follow (upto 32767)
645 pairs, each 2 multibyte characters,
646 each multibyte character represented as 3 bytes. */
647 charset,
649 /* Same parameters as charset, but match any character that is
650 not one of those specified. */
651 charset_not,
653 /* Start remembering the text that is matched, for storing in a
654 register. Followed by one byte with the register number, in
655 the range 0 to one less than the pattern buffer's re_nsub
656 field. */
657 start_memory,
659 /* Stop remembering the text that is matched and store it in a
660 memory register. Followed by one byte with the register
661 number, in the range 0 to one less than `re_nsub' in the
662 pattern buffer. */
663 stop_memory,
665 /* Match a duplicate of something remembered. Followed by one
666 byte containing the register number. */
667 duplicate,
669 /* Fail unless at beginning of line. */
670 begline,
672 /* Fail unless at end of line. */
673 endline,
675 /* Succeeds if at beginning of buffer (if emacs) or at beginning
676 of string to be matched (if not). */
677 begbuf,
679 /* Analogously, for end of buffer/string. */
680 endbuf,
682 /* Followed by two byte relative address to which to jump. */
683 jump,
685 /* Followed by two-byte relative address of place to resume at
686 in case of failure. */
687 on_failure_jump,
689 /* Like on_failure_jump, but pushes a placeholder instead of the
690 current string position when executed. */
691 on_failure_keep_string_jump,
693 /* Just like `on_failure_jump', except that it checks that we
694 don't get stuck in an infinite loop (matching an empty string
695 indefinitely). */
696 on_failure_jump_loop,
698 /* Just like `on_failure_jump_loop', except that it checks for
699 a different kind of loop (the kind that shows up with non-greedy
700 operators). This operation has to be immediately preceded
701 by a `no_op'. */
702 on_failure_jump_nastyloop,
704 /* A smart `on_failure_jump' used for greedy * and + operators.
705 It analyses the loop before which it is put and if the
706 loop does not require backtracking, it changes itself to
707 `on_failure_keep_string_jump' and short-circuits the loop,
708 else it just defaults to changing itself into `on_failure_jump'.
709 It assumes that it is pointing to just past a `jump'. */
710 on_failure_jump_smart,
712 /* Followed by two-byte relative address and two-byte number n.
713 After matching N times, jump to the address upon failure.
714 Does not work if N starts at 0: use on_failure_jump_loop
715 instead. */
716 succeed_n,
718 /* Followed by two-byte relative address, and two-byte number n.
719 Jump to the address N times, then fail. */
720 jump_n,
722 /* Set the following two-byte relative address to the
723 subsequent two-byte number. The address *includes* the two
724 bytes of number. */
725 set_number_at,
736 /* Matches any character whose syntax is specified. Followed by
737 a byte which contains a syntax code, e.g., Sword. */
738 syntaxspec,
740 /* Matches any character whose syntax is not that specified. */
741 notsyntaxspec
743 #ifdef emacs
748 /* Matches any character whose category-set contains the specified
749 category. The operator is followed by a byte which contains a
750 category code (mnemonic ASCII character). */
751 categoryspec,
753 /* Matches any character whose category-set does not contain the
754 specified category. The operator is followed by a byte which
755 contains the category code (mnemonic ASCII character). */
756 notcategoryspec
759 \f
760 /* Common operations on the compiled pattern. */
762 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
764 #define STORE_NUMBER(destination, number) \
765 do { \
766 (destination)[0] = (number) & 0377; \
767 (destination)[1] = (number) >> 8; \
768 } while (0)
770 /* Same as STORE_NUMBER, except increment DESTINATION to
771 the byte after where the number is stored. Therefore, DESTINATION
772 must be an lvalue. */
774 #define STORE_NUMBER_AND_INCR(destination, number) \
775 do { \
776 STORE_NUMBER (destination, number); \
777 (destination) += 2; \
778 } while (0)
780 /* Put into DESTINATION a number stored in two contiguous bytes starting
781 at SOURCE. */
783 #define EXTRACT_NUMBER(destination, source) \
784 do { \
785 (destination) = *(source) & 0377; \
786 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
787 } while (0)
789 #ifdef DEBUG
791 static void
795 {
799 }
802 # undef EXTRACT_NUMBER
803 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
808 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
809 SOURCE must be an lvalue. */
811 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
812 do { \
813 EXTRACT_NUMBER (destination, source); \
814 (source) += 2; \
815 } while (0)
817 #ifdef DEBUG
820 static void
824 {
827 }
829 # ifndef EXTRACT_MACROS
830 # undef EXTRACT_NUMBER_AND_INCR
831 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
832 extract_number_and_incr (&dest, &src)
836 \f
837 /* Store a multibyte character in three contiguous bytes starting
838 DESTINATION, and increment DESTINATION to the byte after where the
839 character is stored. Therefore, DESTINATION must be an lvalue. */
841 #define STORE_CHARACTER_AND_INCR(destination, character) \
842 do { \
843 (destination)[0] = (character) & 0377; \
844 (destination)[1] = ((character) >> 8) & 0377; \
845 (destination)[2] = (character) >> 16; \
846 (destination) += 3; \
847 } while (0)
849 /* Put into DESTINATION a character stored in three contiguous bytes
850 starting at SOURCE. */
852 #define EXTRACT_CHARACTER(destination, source) \
853 do { \
854 (destination) = ((source)[0] \
855 | ((source)[1] << 8) \
856 | ((source)[2] << 16)); \
857 } while (0)
860 /* Macros for charset. */
862 /* Size of bitmap of charset P in bytes. P is a start of charset,
863 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
864 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
866 /* Nonzero if charset P has range table. */
867 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
869 /* Return the address of range table of charset P. But not the start
870 of table itself, but the before where the number of ranges is
871 stored. `2 +' means to skip re_opcode_t and size of bitmap,
872 and the 2 bytes of flags at the start of the range table. */
873 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
875 /* Extract the bit flags that start a range table. */
876 #define CHARSET_RANGE_TABLE_BITS(p) \
877 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
878 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
880 /* Test if C is listed in the bitmap of charset P. */
881 #define CHARSET_LOOKUP_BITMAP(p, c) \
882 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
883 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
885 /* Return the address of end of RANGE_TABLE. COUNT is number of
886 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
887 is start of range and end of range. `* 3' is size of each start
888 and end. */
889 #define CHARSET_RANGE_TABLE_END(range_table, count) \
890 ((range_table) + (count) * 2 * 3)
892 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
893 COUNT is number of ranges in RANGE_TABLE. */
894 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
895 do \
896 { \
897 re_wchar_t range_start, range_end; \
898 re_char *p; \
899 re_char *range_table_end \
900 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
901 \
902 for (p = (range_table); p < range_table_end; p += 2 * 3) \
903 { \
904 EXTRACT_CHARACTER (range_start, p); \
905 EXTRACT_CHARACTER (range_end, p + 3); \
906 \
907 if (range_start <= (c) && (c) <= range_end) \
908 { \
909 (not) = !(not); \
910 break; \
911 } \
912 } \
913 } \
914 while (0)
916 /* Test if C is in range table of CHARSET. The flag NOT is negated if
917 C is listed in it. */
918 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
919 do \
920 { \
922 int count; \
923 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
924 \
925 EXTRACT_NUMBER_AND_INCR (count, range_table); \
926 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
927 } \
928 while (0)
929 \f
930 /* If DEBUG is defined, Regex prints many voluminous messages about what
931 it is doing (if the variable `debug' is nonzero). If linked with the
932 main program in `iregex.c', you can enter patterns and strings
933 interactively. And if linked with the main program in `main.c' and
934 the other test files, you can run the already-written tests. */
936 #ifdef DEBUG
938 /* We use standard I/O for debugging. */
939 # include <stdio.h>
941 /* It is useful to test things that ``must'' be true when debugging. */
942 # include <assert.h>
946 # define DEBUG_STATEMENT(e) e
947 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
948 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
949 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
950 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
951 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
952 if (debug > 0) print_partial_compiled_pattern (s, e)
953 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
954 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
957 /* Print the fastmap in human-readable form. */
959 void
962 {
967 {
969 {
973 {
975 i++;
976 }
978 {
981 }
982 }
983 }
985 }
988 /* Print a compiled pattern string in human-readable form, starting at
989 the START pointer into it and ending just before the pointer END. */
991 void
995 {
1001 {
1004 }
1006 /* Loop over pattern commands. */
1008 {
1012 {
1024 do
1025 {
1027 }
1049 {
1064 {
1065 /* Are we starting a range? */
1067 {
1070 }
1071 /* Have we broken a range? */
1073 {
1076 }
1082 }
1092 {
1096 /* ??? Should print the range table; for now, just skip it. */
1100 }
1101 }
1196 # ifdef emacs
1232 }
1235 }
1238 }
1241 void
1244 {
1252 {
1255 }
1265 /* Perhaps we should print the translate table? */
1266 }
1269 void
1276 {
1281 else
1282 {
1284 {
1289 }
1293 }
1294 }
1298 # undef assert
1299 # define assert(e)
1301 # define DEBUG_STATEMENT(e)
1302 # define DEBUG_PRINT1(x)
1303 # define DEBUG_PRINT2(x1, x2)
1304 # define DEBUG_PRINT3(x1, x2, x3)
1305 # define DEBUG_PRINT4(x1, x2, x3, x4)
1306 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1307 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1310 \f
1311 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1312 also be assigned to arbitrarily: each pattern buffer stores its own
1313 syntax, so it can be changed between regex compilations. */
1314 /* This has no initializer because initialized variables in Emacs
1315 become read-only after dumping. */
1316 reg_syntax_t re_syntax_options;
1319 /* Specify the precise syntax of regexps for compilation. This provides
1320 for compatibility for various utilities which historically have
1321 different, incompatible syntaxes.
1323 The argument SYNTAX is a bit mask comprised of the various bits
1324 defined in regex.h. We return the old syntax. */
1326 reg_syntax_t
1328 reg_syntax_t syntax;
1329 {
1334 }
1337 /* Regexp to use to replace spaces, or NULL meaning don't. */
1340 void
1343 {
1345 }
1347 \f
1348 /* This table gives an error message for each of the error codes listed
1349 in regex.h. Obviously the order here has to be same as there.
1350 POSIX doesn't require that we do anything for REG_NOERROR,
1351 but why not be nice? */
1354 {
1373 };
1374 \f
1375 /* Avoiding alloca during matching, to placate r_alloc. */
1377 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1378 searching and matching functions should not call alloca. On some
1379 systems, alloca is implemented in terms of malloc, and if we're
1380 using the relocating allocator routines, then malloc could cause a
1381 relocation, which might (if the strings being searched are in the
1382 ralloc heap) shift the data out from underneath the regexp
1383 routines.
1385 Here's another reason to avoid allocation: Emacs
1386 processes input from X in a signal handler; processing X input may
1387 call malloc; if input arrives while a matching routine is calling
1388 malloc, then we're scrod. But Emacs can't just block input while
1389 calling matching routines; then we don't notice interrupts when
1390 they come in. So, Emacs blocks input around all regexp calls
1391 except the matching calls, which it leaves unprotected, in the
1392 faith that they will not malloc. */
1394 /* Normally, this is fine. */
1395 #define MATCH_MAY_ALLOCATE
1397 /* The match routines may not allocate if (1) they would do it with malloc
1398 and (2) it's not safe for them to use malloc.
1399 Note that if REL_ALLOC is defined, matching would not use malloc for the
1400 failure stack, but we would still use it for the register vectors;
1401 so REL_ALLOC should not affect this. */
1402 #if defined REGEX_MALLOC && defined emacs
1403 # undef MATCH_MAY_ALLOCATE
1404 #endif
1406 \f
1407 /* Failure stack declarations and macros; both re_compile_fastmap and
1408 re_match_2 use a failure stack. These have to be macros because of
1409 REGEX_ALLOCATE_STACK. */
1412 /* Approximate number of failure points for which to initially allocate space
1413 when matching. If this number is exceeded, we allocate more
1414 space, so it is not a hard limit. */
1415 #ifndef INIT_FAILURE_ALLOC
1416 # define INIT_FAILURE_ALLOC 20
1417 #endif
1419 /* Roughly the maximum number of failure points on the stack. Would be
1420 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1421 This is a variable only so users of regex can assign to it; we never
1422 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1423 before using it, so it should probably be a byte-count instead. */
1424 # if defined MATCH_MAY_ALLOCATE
1425 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1426 whose default stack limit is 2mb. In order for a larger
1427 value to work reliably, you have to try to make it accord
1428 with the process stack limit. */
1430 # else
1432 # endif
1434 union fail_stack_elt
1435 {
1437 /* This should be the biggest `int' that's no bigger than a pointer. */
1439 };
1444 {
1451 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1452 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1455 /* Define macros to initialize and free the failure stack.
1456 Do `return -2' if the alloc fails. */
1458 #ifdef MATCH_MAY_ALLOCATE
1459 # define INIT_FAIL_STACK() \
1460 do { \
1461 fail_stack.stack = (fail_stack_elt_t *) \
1462 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1463 * sizeof (fail_stack_elt_t)); \
1464 \
1465 if (fail_stack.stack == NULL) \
1466 return -2; \
1467 \
1468 fail_stack.size = INIT_FAILURE_ALLOC; \
1469 fail_stack.avail = 0; \
1470 fail_stack.frame = 0; \
1471 } while (0)
1473 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1474 #else
1475 # define INIT_FAIL_STACK() \
1476 do { \
1477 fail_stack.avail = 0; \
1478 fail_stack.frame = 0; \
1479 } while (0)
1481 # define RESET_FAIL_STACK() ((void)0)
1482 #endif
1485 /* Double the size of FAIL_STACK, up to a limit
1486 which allows approximately `re_max_failures' items.
1488 Return 1 if succeeds, and 0 if either ran out of memory
1489 allocating space for it or it was already too large.
1491 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1493 /* Factor to increase the failure stack size by
1494 when we increase it.
1495 This used to be 2, but 2 was too wasteful
1496 because the old discarded stacks added up to as much space
1497 were as ultimate, maximum-size stack. */
1498 #define FAIL_STACK_GROWTH_FACTOR 4
1500 #define GROW_FAIL_STACK(fail_stack) \
1501 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1502 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1503 ? 0 \
1504 : ((fail_stack).stack \
1505 = (fail_stack_elt_t *) \
1506 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1507 (fail_stack).size * sizeof (fail_stack_elt_t), \
1508 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1509 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1510 * FAIL_STACK_GROWTH_FACTOR))), \
1511 \
1512 (fail_stack).stack == NULL \
1513 ? 0 \
1514 : ((fail_stack).size \
1515 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1516 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1517 * FAIL_STACK_GROWTH_FACTOR)) \
1518 / sizeof (fail_stack_elt_t)), \
1519 1)))
1522 /* Push a pointer value onto the failure stack.
1523 Assumes the variable `fail_stack'. Probably should only
1524 be called from within `PUSH_FAILURE_POINT'. */
1525 #define PUSH_FAILURE_POINTER(item) \
1526 fail_stack.stack[fail_stack.avail++].pointer = (item)
1528 /* This pushes an integer-valued item onto the failure stack.
1529 Assumes the variable `fail_stack'. Probably should only
1530 be called from within `PUSH_FAILURE_POINT'. */
1531 #define PUSH_FAILURE_INT(item) \
1532 fail_stack.stack[fail_stack.avail++].integer = (item)
1534 /* Push a fail_stack_elt_t value onto the failure stack.
1535 Assumes the variable `fail_stack'. Probably should only
1536 be called from within `PUSH_FAILURE_POINT'. */
1537 #define PUSH_FAILURE_ELT(item) \
1538 fail_stack.stack[fail_stack.avail++] = (item)
1540 /* These three POP... operations complement the three PUSH... operations.
1541 All assume that `fail_stack' is nonempty. */
1542 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1543 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1544 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1546 /* Individual items aside from the registers. */
1547 #define NUM_NONREG_ITEMS 3
1549 /* Used to examine the stack (to detect infinite loops). */
1550 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1551 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1552 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1553 #define TOP_FAILURE_HANDLE() fail_stack.frame
1556 #define ENSURE_FAIL_STACK(space) \
1557 while (REMAINING_AVAIL_SLOTS <= space) { \
1558 if (!GROW_FAIL_STACK (fail_stack)) \
1559 return -2; \
1562 }
1564 /* Push register NUM onto the stack. */
1565 #define PUSH_FAILURE_REG(num) \
1566 do { \
1567 char *destination; \
1568 ENSURE_FAIL_STACK(3); \
1570 num, regstart[num], regend[num]); \
1571 PUSH_FAILURE_POINTER (regstart[num]); \
1572 PUSH_FAILURE_POINTER (regend[num]); \
1573 PUSH_FAILURE_INT (num); \
1574 } while (0)
1576 /* Change the counter's value to VAL, but make sure that it will
1577 be reset when backtracking. */
1578 #define PUSH_NUMBER(ptr,val) \
1579 do { \
1580 char *destination; \
1581 int c; \
1582 ENSURE_FAIL_STACK(3); \
1583 EXTRACT_NUMBER (c, ptr); \
1585 PUSH_FAILURE_INT (c); \
1586 PUSH_FAILURE_POINTER (ptr); \
1587 PUSH_FAILURE_INT (-1); \
1588 STORE_NUMBER (ptr, val); \
1589 } while (0)
1591 /* Pop a saved register off the stack. */
1592 #define POP_FAILURE_REG_OR_COUNT() \
1593 do { \
1594 int reg = POP_FAILURE_INT (); \
1595 if (reg == -1) \
1596 { \
1599 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1600 reg = POP_FAILURE_INT (); \
1601 STORE_NUMBER (ptr, reg); \
1603 } \
1604 else \
1605 { \
1606 regend[reg] = POP_FAILURE_POINTER (); \
1607 regstart[reg] = POP_FAILURE_POINTER (); \
1609 reg, regstart[reg], regend[reg]); \
1610 } \
1611 } while (0)
1613 /* Check that we are not stuck in an infinite loop. */
1614 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1615 do { \
1616 int failure = TOP_FAILURE_HANDLE (); \
1618 while (failure > 0 \
1619 && (FAILURE_STR (failure) == string_place \
1620 || FAILURE_STR (failure) == NULL)) \
1621 { \
1622 assert (FAILURE_PAT (failure) >= bufp->buffer \
1623 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1624 if (FAILURE_PAT (failure) == pat_cur) \
1625 { \
1626 cycle = 1; \
1627 break; \
1628 } \
1630 failure = NEXT_FAILURE_HANDLE(failure); \
1631 } \
1633 } while (0)
1635 /* Push the information about the state we will need
1636 if we ever fail back to it.
1638 Requires variables fail_stack, regstart, regend and
1639 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1640 declared.
1642 Does `return FAILURE_CODE' if runs out of memory. */
1644 #define PUSH_FAILURE_POINT(pattern, string_place) \
1645 do { \
1646 char *destination; \
1647 /* Must be int, so when we don't save any registers, the arithmetic \
1649 \
1650 DEBUG_STATEMENT (nfailure_points_pushed++); \
1654 \
1655 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1656 \
1658 \
1660 PUSH_FAILURE_INT (fail_stack.frame); \
1661 \
1663 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1665 PUSH_FAILURE_POINTER (string_place); \
1666 \
1668 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1669 PUSH_FAILURE_POINTER (pattern); \
1670 \
1672 fail_stack.frame = fail_stack.avail; \
1673 } while (0)
1675 /* Estimate the size of data pushed by a typical failure stack entry.
1676 An estimate is all we need, because all we use this for
1677 is to choose a limit for how big to make the failure stack. */
1678 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1679 #define TYPICAL_FAILURE_SIZE 20
1681 /* How many items can still be added to the stack without overflowing it. */
1682 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1685 /* Pops what PUSH_FAIL_STACK pushes.
1687 We restore into the parameters, all of which should be lvalues:
1688 STR -- the saved data position.
1689 PAT -- the saved pattern position.
1690 REGSTART, REGEND -- arrays of string positions.
1692 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1693 `pend', `string1', `size1', `string2', and `size2'. */
1695 #define POP_FAILURE_POINT(str, pat) \
1696 do { \
1697 assert (!FAIL_STACK_EMPTY ()); \
1698 \
1703 \
1705 while (fail_stack.frame < fail_stack.avail) \
1706 POP_FAILURE_REG_OR_COUNT (); \
1707 \
1708 pat = POP_FAILURE_POINTER (); \
1710 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1711 \
1712 /* If the saved string location is NULL, it came from an \
1713 on_failure_keep_string_jump opcode, and we want to throw away the \
1715 str = POP_FAILURE_POINTER (); \
1717 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1719 \
1720 fail_stack.frame = POP_FAILURE_INT (); \
1722 \
1723 assert (fail_stack.avail >= 0); \
1724 assert (fail_stack.frame <= fail_stack.avail); \
1725 \
1726 DEBUG_STATEMENT (nfailure_points_popped++); \
1730 \f
1731 /* Registers are set to a sentinel when they haven't yet matched. */
1732 #define REG_UNSET(e) ((e) == NULL)
1733 \f
1734 /* Subroutine declarations and macros for regex_compile. */
1737 reg_syntax_t syntax,
1748 reg_syntax_t syntax));
1751 reg_syntax_t syntax));
1756 /* Fetch the next character in the uncompiled pattern, with no
1757 translation. */
1758 #define PATFETCH(c) \
1759 do { \
1760 int len; \
1761 if (p == pend) return REG_EEND; \
1762 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len, multibyte); \
1763 p += len; \
1764 } while (0)
1767 /* If `translate' is non-null, return translate[D], else just D. We
1768 cast the subscript to translate because some data is declared as
1769 `char *', to avoid warnings when a string constant is passed. But
1770 when we use a character as a subscript we must make it unsigned. */
1771 #ifndef TRANSLATE
1772 # define TRANSLATE(d) \
1773 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1774 #endif
1777 /* Macros for outputting the compiled pattern into `buffer'. */
1779 /* If the buffer isn't allocated when it comes in, use this. */
1780 #define INIT_BUF_SIZE 32
1782 /* Make sure we have at least N more bytes of space in buffer. */
1783 #define GET_BUFFER_SPACE(n) \
1784 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1785 EXTEND_BUFFER ()
1787 /* Make sure we have one more byte of buffer space and then add C to it. */
1788 #define BUF_PUSH(c) \
1789 do { \
1790 GET_BUFFER_SPACE (1); \
1791 *b++ = (unsigned char) (c); \
1792 } while (0)
1795 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1796 #define BUF_PUSH_2(c1, c2) \
1797 do { \
1798 GET_BUFFER_SPACE (2); \
1799 *b++ = (unsigned char) (c1); \
1800 *b++ = (unsigned char) (c2); \
1801 } while (0)
1804 /* As with BUF_PUSH_2, except for three bytes. */
1805 #define BUF_PUSH_3(c1, c2, c3) \
1806 do { \
1807 GET_BUFFER_SPACE (3); \
1808 *b++ = (unsigned char) (c1); \
1809 *b++ = (unsigned char) (c2); \
1810 *b++ = (unsigned char) (c3); \
1811 } while (0)
1814 /* Store a jump with opcode OP at LOC to location TO. We store a
1815 relative address offset by the three bytes the jump itself occupies. */
1816 #define STORE_JUMP(op, loc, to) \
1817 store_op1 (op, loc, (to) - (loc) - 3)
1819 /* Likewise, for a two-argument jump. */
1820 #define STORE_JUMP2(op, loc, to, arg) \
1821 store_op2 (op, loc, (to) - (loc) - 3, arg)
1823 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1824 #define INSERT_JUMP(op, loc, to) \
1825 insert_op1 (op, loc, (to) - (loc) - 3, b)
1827 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1828 #define INSERT_JUMP2(op, loc, to, arg) \
1829 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1832 /* This is not an arbitrary limit: the arguments which represent offsets
1833 into the pattern are two bytes long. So if 2^15 bytes turns out to
1834 be too small, many things would have to change. */
1835 # define MAX_BUF_SIZE (1L << 15)
1838 /* Any other compiler which, like MSC, has allocation limit below 2^16
1839 bytes will have to use approach similar to what was done below for
1840 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1841 reallocating to 0 bytes. Such thing is not going to work too well.
1842 You have been warned!! */
1843 #if defined _MSC_VER && !defined WIN32
1844 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1845 # define MAX_BUF_SIZE 65500L
1846 #else
1847 # define MAX_BUF_SIZE (1L << 16)
1848 #endif
1851 /* Extend the buffer by twice its current size via realloc and
1852 reset the pointers that pointed into the old block to point to the
1853 correct places in the new one. If extending the buffer results in it
1854 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1855 #if __BOUNDED_POINTERS__
1856 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1857 # define MOVE_BUFFER_POINTER(P) \
1858 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1859 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1860 else \
1861 { \
1862 SET_HIGH_BOUND (b); \
1863 SET_HIGH_BOUND (begalt); \
1864 if (fixup_alt_jump) \
1865 SET_HIGH_BOUND (fixup_alt_jump); \
1866 if (laststart) \
1867 SET_HIGH_BOUND (laststart); \
1868 if (pending_exact) \
1869 SET_HIGH_BOUND (pending_exact); \
1870 }
1871 #else
1872 # define MOVE_BUFFER_POINTER(P) (P) += incr
1873 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1874 #endif
1875 #define EXTEND_BUFFER() \
1876 do { \
1877 re_char *old_buffer = bufp->buffer; \
1878 if (bufp->allocated == MAX_BUF_SIZE) \
1879 return REG_ESIZE; \
1880 bufp->allocated <<= 1; \
1881 if (bufp->allocated > MAX_BUF_SIZE) \
1882 bufp->allocated = MAX_BUF_SIZE; \
1883 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1884 if (bufp->buffer == NULL) \
1885 return REG_ESPACE; \
1887 if (old_buffer != bufp->buffer) \
1888 { \
1889 int incr = bufp->buffer - old_buffer; \
1890 MOVE_BUFFER_POINTER (b); \
1891 MOVE_BUFFER_POINTER (begalt); \
1892 if (fixup_alt_jump) \
1893 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1894 if (laststart) \
1895 MOVE_BUFFER_POINTER (laststart); \
1896 if (pending_exact) \
1897 MOVE_BUFFER_POINTER (pending_exact); \
1898 } \
1899 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1900 } while (0)
1903 /* Since we have one byte reserved for the register number argument to
1904 {start,stop}_memory, the maximum number of groups we can report
1905 things about is what fits in that byte. */
1906 #define MAX_REGNUM 255
1908 /* But patterns can have more than `MAX_REGNUM' registers. We just
1909 ignore the excess. */
1913 /* Macros for the compile stack. */
1915 /* Since offsets can go either forwards or backwards, this type needs to
1916 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1917 /* int may be not enough when sizeof(int) == 2. */
1921 {
1922 pattern_offset_t begalt_offset;
1923 pattern_offset_t fixup_alt_jump;
1924 pattern_offset_t laststart_offset;
1925 regnum_t regnum;
1930 {
1937 #define INIT_COMPILE_STACK_SIZE 32
1939 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1940 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1942 /* The next available element. */
1943 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1945 /* Explicit quit checking is only used on NTemacs and whenever we
1946 use polling to process input events. */
1947 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1949 # define IMMEDIATE_QUIT_CHECK \
1950 do { \
1951 if (immediate_quit) QUIT; \
1952 } while (0)
1953 #else
1954 # define IMMEDIATE_QUIT_CHECK ((void)0)
1955 #endif
1956 \f
1957 /* Structure to manage work area for range table. */
1958 struct range_table_work_area
1959 {
1964 };
1966 /* Make sure that WORK_AREA can hold more N multibyte characters.
1967 This is used only in set_image_of_range and set_image_of_range_1.
1968 It expects WORK_AREA to be a pointer.
1969 If it can't get the space, it returns from the surrounding function. */
1971 #define EXTEND_RANGE_TABLE(work_area, n) \
1972 do { \
1973 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1974 { \
1975 extend_range_table_work_area (&work_area); \
1976 if ((work_area).table == 0) \
1977 return (REG_ESPACE); \
1978 } \
1979 } while (0)
1981 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1982 (work_area).bits |= (bit)
1984 /* Bits used to implement the multibyte-part of the various character classes
1985 such as [:alnum:] in a charset's range table. */
1986 #define BIT_WORD 0x1
1987 #define BIT_LOWER 0x2
1988 #define BIT_PUNCT 0x4
1989 #define BIT_SPACE 0x8
1990 #define BIT_UPPER 0x10
1991 #define BIT_MULTIBYTE 0x20
1993 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1994 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1995 do { \
1996 EXTEND_RANGE_TABLE ((work_area), 2); \
1997 (work_area).table[(work_area).used++] = (range_start); \
1998 (work_area).table[(work_area).used++] = (range_end); \
1999 } while (0)
2001 /* Free allocated memory for WORK_AREA. */
2002 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2003 do { \
2004 if ((work_area).table) \
2005 free ((work_area).table); \
2006 } while (0)
2008 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2009 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2010 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2011 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2012 \f
2014 /* Set the bit for character C in a list. */
2015 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2018 #ifdef emacs
2020 /* Store characters in the range FROM to TO in the bitmap at B (for
2021 ASCII and unibyte characters) and WORK_AREA (for multibyte
2022 characters) while translating them and paying attention to the
2023 continuity of translated characters.
2025 Implementation note: It is better to implement these fairly big
2026 macros by a function, but it's not that easy because macros called
2027 in this macro assume various local variables already declared. */
2029 /* Both FROM and TO are ASCII characters. */
2031 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2032 do { \
2033 int C0, C1; \
2034 \
2035 for (C0 = (FROM); C0 <= (TO); C0++) \
2036 { \
2037 C1 = TRANSLATE (C0); \
2038 if (! ASCII_CHAR_P (C1)) \
2039 { \
2040 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2041 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2042 C1 = C0; \
2043 } \
2044 SET_LIST_BIT (C1); \
2045 } \
2046 } while (0)
2049 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2051 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2052 do { \
2053 int C0, C1, C2, I; \
2054 int USED = RANGE_TABLE_WORK_USED (work_area); \
2055 \
2056 for (C0 = (FROM); C0 <= (TO); C0++) \
2057 { \
2058 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2059 if (CHAR_BYTE8_P (C1)) \
2060 SET_LIST_BIT (C0); \
2061 else \
2062 { \
2063 C2 = TRANSLATE (C1); \
2064 if (C2 == C1 \
2065 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2066 C1 = C0; \
2067 SET_LIST_BIT (C1); \
2068 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2069 { \
2070 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2071 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2072 \
2073 if (C2 >= from - 1 && C2 <= to + 1) \
2074 { \
2075 if (C2 == from - 1) \
2076 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2077 else if (C2 == to + 1) \
2078 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2079 break; \
2080 } \
2081 } \
2082 if (I < USED) \
2083 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2084 } \
2085 } \
2086 } while (0)
2089 /* Both FROM and TO are mulitbyte characters. */
2091 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2092 do { \
2093 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2094 \
2095 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2096 for (C0 = (FROM); C0 <= (TO); C0++) \
2097 { \
2098 C1 = TRANSLATE (C0); \
2099 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2100 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2101 SET_LIST_BIT (C2); \
2102 if (C1 >= (FROM) && C1 <= (TO)) \
2103 continue; \
2104 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2105 { \
2106 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2107 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2108 \
2109 if (C1 >= from - 1 && C1 <= to + 1) \
2110 { \
2111 if (C1 == from - 1) \
2112 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2113 else if (C1 == to + 1) \
2114 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2115 break; \
2116 } \
2117 } \
2118 if (I < USED) \
2119 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2120 } \
2121 } while (0)
2125 /* Get the next unsigned number in the uncompiled pattern. */
2126 #define GET_UNSIGNED_NUMBER(num) \
2127 do { \
2128 if (p == pend) \
2129 FREE_STACK_RETURN (REG_EBRACE); \
2130 else \
2131 { \
2132 PATFETCH (c); \
2134 { \
2135 int prev; \
2136 if (num < 0) \
2137 num = 0; \
2138 prev = num; \
2140 if (num / 10 != prev) \
2141 FREE_STACK_RETURN (REG_BADBR); \
2142 if (p == pend) \
2143 FREE_STACK_RETURN (REG_EBRACE); \
2144 PATFETCH (c); \
2145 } \
2146 } \
2147 } while (0)
2148 \f
2149 #if ! WIDE_CHAR_SUPPORT
2151 /* Map a string to the char class it names (if any). */
2152 re_wctype_t
2155 {
2175 }
2177 /* True if CH is in the char class CC. */
2178 boolean
2181 re_wctype_t cc;
2182 {
2184 {
2205 }
2206 }
2208 /* Return a bit-pattern to use in the range-table bits to match multibyte
2209 chars of class CC. */
2210 static int
2212 re_wctype_t cc;
2213 {
2215 {
2227 }
2228 }
2229 #endif
2230 \f
2231 /* Filling in the work area of a range. */
2233 /* Actually extend the space in WORK_AREA. */
2235 static void
2238 {
2241 work_area->table
2243 else
2244 work_area->table
2246 }
2248 #if 0
2249 #ifdef emacs
2251 /* Carefully find the ranges of codes that are equivalent
2252 under case conversion to the range start..end when passed through
2253 TRANSLATE. Handle the case where non-letters can come in between
2254 two upper-case letters (which happens in Latin-1).
2255 Also handle the case of groups of more than 2 case-equivalent chars.
2257 The basic method is to look at consecutive characters and see
2258 if they can form a run that can be handled as one.
2260 Returns -1 if successful, REG_ESPACE if ran out of space. */
2262 static int
2264 RE_TRANSLATE_TYPE translate;
2267 {
2268 /* `one_case' indicates a character, or a run of characters,
2269 each of which is an isolate (no case-equivalents).
2270 This includes all ASCII non-letters.
2272 `two_case' indicates a character, or a run of characters,
2273 each of which has two case-equivalent forms.
2274 This includes all ASCII letters.
2276 `strange' indicates a character that has more than one
2277 case-equivalent. */
2281 /* Describe the run that is in progress,
2282 which the next character can try to extend.
2283 If run_type is strange, that means there really is no run.
2284 If run_type is one_case, then run_start...run_end is the run.
2285 If run_type is two_case, then the run is run_start...run_end,
2286 and the case-equivalents end at run_eqv_end. */
2291 Lisp_Object eqv_table;
2294 {
2299 }
2304 {
2309 /* Classify this character */
2314 else
2319 else
2322 /* Can this character extend the run in progress? */
2327 {
2328 /* No, end the run.
2329 Record each of its equivalent ranges. */
2331 {
2335 }
2337 {
2345 }
2347 }
2350 {
2351 /* For a strange character, add each of its equivalents, one
2352 by one. Don't start a range. */
2353 do
2354 {
2359 }
2361 }
2363 /* Add this char to the run, or start a new run. */
2365 {
2366 /* Initialize a new range. */
2371 }
2372 else
2373 {
2374 /* Extend a running range. */
2377 }
2378 }
2380 /* If a run is still in progress at the end, finish it now
2381 by recording its equivalent ranges. */
2383 {
2387 }
2389 {
2397 }
2400 }
2404 /* Record the the image of the range start..end when passed through
2405 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2406 and is not even necessarily contiguous.
2407 Normally we approximate it with the smallest contiguous range that contains
2408 all the chars we need. However, for Latin-1 we go to extra effort
2409 to do a better job.
2411 This function is not called for ASCII ranges.
2413 Returns -1 if successful, REG_ESPACE if ran out of space. */
2415 static int
2417 RE_TRANSLATE_TYPE translate;
2420 {
2423 #ifdef emacs
2424 /* For Latin-1 ranges, use set_image_of_range_1
2425 to get proper handling of ranges that include letters and nonletters.
2426 For a range that includes the whole of Latin-1, this is not necessary.
2427 For other character sets, we don't bother to get this right. */
2430 {
2443 }
2444 #endif
2453 {
2457 {
2460 {
2463 else
2464 {
2467 }
2468 }
2469 }
2472 {
2476 }
2477 }
2480 }
2482 \f
2483 #ifndef MATCH_MAY_ALLOCATE
2485 /* If we cannot allocate large objects within re_match_2_internal,
2486 we make the fail stack and register vectors global.
2487 The fail stack, we grow to the maximum size when a regexp
2488 is compiled.
2489 The register vectors, we adjust in size each time we
2490 compile a regexp, according to the number of registers it needs. */
2494 /* Size with which the following vectors are currently allocated.
2495 That is so we can make them bigger as needed,
2496 but never make them smaller. */
2502 /* Make the register vectors big enough for NUM_REGS registers,
2503 but don't make them smaller. */
2505 static
2508 {
2510 {
2517 }
2518 }
2521 \f
2523 compile_stack,
2524 regnum_t regnum));
2526 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2527 Returns one of error codes defined in `regex.h', or zero for success.
2529 Assumes the `allocated' (and perhaps `buffer') and `translate'
2530 fields are set in BUFP on entry.
2532 If it succeeds, results are put in BUFP (if it returns an error, the
2533 contents of BUFP are undefined):
2534 `buffer' is the compiled pattern;
2535 `syntax' is set to SYNTAX;
2536 `used' is set to the length of the compiled pattern;
2537 `fastmap_accurate' is zero;
2538 `re_nsub' is the number of subexpressions in PATTERN;
2539 `not_bol' and `not_eol' are zero;
2541 The `fastmap' field is neither examined nor set. */
2543 /* Insert the `jump' from the end of last alternative to "here".
2544 The space for the jump has already been allocated. */
2545 #define FIXUP_ALT_JUMP() \
2546 do { \
2547 if (fixup_alt_jump) \
2548 STORE_JUMP (jump, fixup_alt_jump, b); \
2549 } while (0)
2552 /* Return, freeing storage we allocated. */
2553 #define FREE_STACK_RETURN(value) \
2554 do { \
2555 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2556 free (compile_stack.stack); \
2557 return value; \
2558 } while (0)
2560 static reg_errcode_t
2564 reg_syntax_t syntax;
2566 {
2567 /* We fetch characters from PATTERN here. */
2570 /* A random temporary spot in PATTERN. */
2573 /* Points to the end of the buffer, where we should append. */
2576 /* Keeps track of unclosed groups. */
2577 compile_stack_type compile_stack;
2579 /* Points to the current (ending) position in the pattern. */
2580 #ifdef AIX
2581 /* `const' makes AIX compiler fail. */
2583 #else
2585 #endif
2588 /* How to translate the characters in the pattern. */
2591 /* Address of the count-byte of the most recently inserted `exactn'
2592 command. This makes it possible to tell if a new exact-match
2593 character can be added to that command or if the character requires
2594 a new `exactn' command. */
2597 /* Address of start of the most recently finished expression.
2598 This tells, e.g., postfix * where to find the start of its
2599 operand. Reset at the beginning of groups and alternatives. */
2602 /* Address of beginning of regexp, or inside of last group. */
2605 /* Place in the uncompiled pattern (i.e., the {) to
2606 which to go back if the interval is invalid. */
2609 /* Address of the place where a forward jump should go to the end of
2610 the containing expression. Each alternative of an `or' -- except the
2611 last -- ends with a forward jump of this sort. */
2614 /* Work area for range table of charset. */
2617 /* If the object matched can contain multibyte characters. */
2620 /* If a target of matching can contain multibyte characters. */
2623 /* Nonzero if we have pushed down into a subpattern. */
2626 /* These hold the values of p, pattern, and pend from the main
2627 pattern when we have pushed into a subpattern. */
2632 #ifdef DEBUG
2633 debug++;
2636 {
2642 }
2645 /* Initialize the compile stack. */
2656 /* Initialize the pattern buffer. */
2662 /* Set `used' to zero, so that if we return an error, the pattern
2663 printer (for debugging) will think there's no pattern. We reset it
2664 at the end. */
2667 /* Always count groups, whether or not bufp->no_sub is set. */
2670 #if !defined emacs && !defined SYNTAX_TABLE
2671 /* Initialize the syntax table. */
2673 #endif
2676 {
2679 enough space. This loses if buffer's address is bogus, but
2680 that is the user's responsibility. */
2682 }
2683 else
2686 }
2690 }
2694 /* Loop through the uncompiled pattern until we're at the end. */
2696 {
2698 {
2699 /* If this is the end of an included regexp,
2700 pop back to the main regexp and try again. */
2702 {
2708 }
2709 /* If this is the end of the main regexp, we are done. */
2711 }
2716 {
2718 {
2721 /* If there's no special whitespace regexp, treat
2722 spaces normally. And don't try to do this recursively. */
2726 /* Peek past following spaces. */
2728 {
2731 p1++;
2732 }
2733 /* If the spaces are followed by a repetition op,
2734 treat them normally. */
2740 /* Replace the spaces with the whitespace regexp. */
2748 }
2751 {
2754 /* If context independent, it's an operator. */
2756 /* Otherwise, depends on what's come before. */
2759 else
2761 }
2766 {
2768 p == pend
2769 /* If context independent, it's an operator. */
2771 /* Otherwise, depends on what's next. */
2774 else
2776 }
2785 handle_plus:
2787 /* If there is no previous pattern... */
2789 {
2794 }
2796 {
2797 /* 1 means zero (many) matches is allowed. */
2801 /* If there is a sequence of repetition chars, collapse it
2802 down to just one (the right one). We can't combine
2803 interval operators with these because of, e.g., `a{2}*',
2804 which should only match an even number of `a's. */
2807 {
2811 else
2812 {
2815 }
2822 ;
2824 {
2829 else
2831 }
2832 else
2834 /* If we get here, we found another repeat character. */
2836 }
2838 /* Star, etc. applied to an empty pattern is equivalent
2839 to an empty pattern. */
2843 /* Now we know whether or not zero matches is allowed
2844 and also whether or not two or more matches is allowed. */
2846 {
2848 {
2851 re_opcode_t ofj =
2852 /* Check if the loop can match the empty string. */
2859 on_failure_keep_string_jump, we turn simple P+
2860 into PP* if P is simple. */
2868 }
2872 /* A + loop. */
2874 else
2875 /* Simple * loops can use on_failure_keep_string_jump
2876 depending on what follows. But since we don't know
2877 that yet, we leave the decision up to
2878 on_failure_jump_smart. */
2884 }
2885 else
2886 {
2887 /* A simple ? pattern. */
2892 }
2893 }
2899 {
2902 /* The non-greedy multiple match looks like
2903 a repeat..until: we only need a conditional jump
2904 at the end of the loop. */
2910 {
2911 /* The repeat...until naturally matches one or more.
2912 To also match zero times, we need to first jump to
2913 the end of the loop (its conditional jump). */
2916 }
2917 }
2918 else
2919 {
2920 /* non-greedy a?? */
2925 }
2926 }
2927 }
2939 {
2944 /* Ensure that we have enough space to push a charset: the
2945 opcode, the length count, and the bitset; 34 bytes in all. */
2950 /* We test `*p == '^' twice, instead of using an if
2951 statement, so we only need one BUF_PUSH. */
2954 p++;
2956 /* Remember the first position in the bracket expression. */
2959 /* Push the number of bytes in the bitmap. */
2962 /* Clear the whole map. */
2965 /* charset_not matches newline according to a syntax bit. */
2970 /* Read in characters and ranges, setting map bits. */
2972 {
2979 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2980 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2981 So the translation is done later in a loop. Example:
2982 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2985 /* \ might escape characters inside [...] and [^...]. */
2987 {
2992 }
2993 else
2994 {
2995 /* Could be the end of the bracket expression. If it's
2996 not (i.e., when the bracket expression is `[]' so
2997 far), the ']' character bit gets set way below. */
3000 }
3002 /* See if we're at the beginning of a possible character
3003 class. */
3007 {
3008 /* Leave room for the null. */
3016 /* If pattern is `[[:'. */
3020 {
3026 else
3027 /* This is in any case an invalid class name. */
3029 }
3032 /* If isn't a word bracketed by `[:' and `:]':
3033 undo the ending character, the letters, and
3034 leave the leading `:' and `[' (but set bits for
3035 them). */
3037 {
3038 re_wctype_t cc;
3046 /* Throw away the ] at the end of the character
3047 class. */
3052 #ifndef emacs
3055 {
3059 }
3061 /* Most character classes in a multibyte match
3062 just set a flag. Exceptions are is_blank,
3063 is_digit, is_cntrl, and is_xdigit, since
3064 they can only match ASCII characters. We
3065 don't need to handle them for multibyte.
3066 They are distinguished by a negative wctype. */
3069 {
3073 {
3082 }
3083 }
3084 SET_RANGE_TABLE_WORK_AREA_BIT
3087 /* In most cases the matching rule for char classes
3088 only uses the syntax table for multibyte chars,
3089 so that the content of the syntax-table it is not
3090 hardcoded in the range_table. SPACE and WORD are
3091 the two exceptions. */
3095 /* Repeat the loop. */
3097 }
3098 else
3099 {
3100 /* Go back to right after the "[:". */
3104 /* Because the `:' may starts the range, we
3105 can't simply set bit and repeat the loop.
3106 Instead, just set it to C and handle below. */
3108 }
3109 }
3112 {
3114 /* Discard the `-'. */
3117 /* Fetch the character which ends the range. */
3119 #ifdef emacs
3122 /* Treat the range from a multibyte character to
3123 raw-byte character as empty. */
3126 }
3127 else
3128 /* Range from C to C. */
3132 {
3135 /* Else, repeat the loop. */
3136 }
3137 else
3138 {
3139 #ifndef emacs
3140 /* Set the range into bitmap */
3142 {
3146 }
3149 {
3155 }
3157 {
3159 {
3164 }
3166 {
3168 }
3169 else
3170 {
3172 }
3173 }
3175 }
3176 }
3178 /* Discard any (non)matching list bytes that are all 0 at the
3179 end of the map. Decrease the map-length byte too. */
3184 /* Build real range table from work area. */
3187 {
3191 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3192 bytes for flags, two for COUNT, and three bytes for
3193 each character. */
3196 /* Indicate the existence of range table. */
3199 /* Store the character class flag bits into the range table.
3200 If not in emacs, these flag bits are always 0. */
3206 STORE_CHARACTER_AND_INCR
3208 }
3209 }
3216 else
3223 else
3230 else
3237 else
3244 else
3251 /* Do not translate the character after the \, so that we can
3252 distinguish, e.g., \B from \b, even if we normally would
3253 translate, e.g., B to b. */
3257 {
3262 handle_open:
3263 {
3267 {
3268 /* Look for a special (?...) construct */
3270 {
3273 {
3276 {
3279 /* An explicitly specified regnum must start
3280 with non-0. */
3287 /* Only (?:...) is supported right now. */
3289 }
3290 }
3291 }
3292 }
3302 /* Ideally, we'd want to check that the specified
3303 group can't have matched (i.e. all subgroups
3304 using the same regnum are in other branches of
3305 OR patterns), but we don't currently keep track
3306 of enough info to do that easily. */
3309 }
3310 else
3311 /* It's really shy. */
3315 {
3317 compile_stack_elt_t);
3321 }
3323 /* These are the values to restore when we hit end of this
3324 group. They are all relative offsets, so that if the
3325 whole pattern moves because of realloc, they will still
3326 be valid. */
3328 COMPILE_STACK_TOP.fixup_alt_jump
3333 /* Do not push a start_memory for groups beyond the last one
3334 we can represent in the compiled pattern. */
3343 /* If we've reached MAX_REGNUM groups, then this open
3344 won't actually generate any code, so we'll have to
3345 clear pending_exact explicitly. */
3348 }
3354 {
3357 else
3359 }
3361 handle_close:
3364 /* See similar code for backslashed left paren above. */
3366 {
3369 else
3371 }
3373 /* Since we just checked for an empty stack above, this
3374 ``can't happen''. */
3376 {
3377 /* We don't just want to restore into `regnum', because
3378 later groups should continue to be numbered higher,
3379 as in `(ab)c(de)' -- the second group is #2. */
3380 regnum_t regnum;
3384 fixup_alt_jump
3390 /* If we've reached MAX_REGNUM groups, then this open
3391 won't actually generate any code, so we'll have to
3392 clear pending_exact explicitly. */
3395 /* We're at the end of the group, so now we know how many
3396 groups were inside this one. */
3399 }
3406 handle_alt:
3410 /* Insert before the previous alternative a jump which
3411 jumps to this alternative if the former fails. */
3417 /* The alternative before this one has a jump after it
3418 which gets executed if it gets matched. Adjust that
3419 jump so it will jump to this alternative's analogous
3420 jump (put in below, which in turn will jump to the next
3421 (if any) alternative's such jump, etc.). The last such
3422 jump jumps to the correct final destination. A picture:
3423 _____ _____
3424 | | | |
3425 | v | v
3426 a | b | c
3428 If we are at `b', then fixup_alt_jump right now points to a
3429 three-byte space after `a'. We'll put in the jump, set
3430 fixup_alt_jump to right after `b', and leave behind three
3431 bytes which we'll fill in when we get to after `c'. */
3435 /* Mark and leave space for a jump after this alternative,
3436 to be filled in later either by next alternative or
3437 when know we're at the end of a series of alternatives. */
3448 /* If \{ is a literal. */
3450 /* If we're at `\{' and it's not the open-interval
3451 operator. */
3455 handle_interval:
3456 {
3457 /* If got here, then the syntax allows intervals. */
3459 /* At least (most) this many matches must be made. */
3468 else
3469 /* Interval such as `{1}' => match exactly once. */
3477 {
3483 }
3488 /* We just parsed a valid interval. */
3490 /* If it's invalid to have no preceding re. */
3492 {
3497 else
3499 }
3502 /* If the upper bound is zero, just drop the sub pattern
3503 altogether. */
3506 /* Just match it once: nothing to do here. */
3507 ;
3509 /* Otherwise, we have a nontrivial interval. When
3510 we're all done, the pattern will look like:
3511 set_number_at <jump count> <upper bound>
3512 set_number_at <succeed_n count> <lower bound>
3513 succeed_n <after jump addr> <succeed_n count>
3514 <body of loop>
3515 jump_n <succeed_n addr> <jump count>
3516 (The upper bound and `jump_n' are omitted if
3517 `upper_bound' is 1, though.) */
3518 else
3520 more at the end of the loop. */
3528 {
3529 /* A succeed_n that starts with 0 is really a
3530 a simple on_failure_jump_loop. */
3534 }
3535 else
3536 {
3537 /* Initialize lower bound of the `succeed_n', even
3538 though it will be set during matching by its
3539 attendant `set_number_at' (inserted next),
3540 because `re_compile_fastmap' needs to know.
3541 Jump to the `jump_n' we might insert below. */
3544 lower_bound);
3547 /* Code to initialize the lower bound. Insert
3548 before the `succeed_n'. The `5' is the last two
3549 bytes of this `set_number_at', plus 3 bytes of
3550 the following `succeed_n'. */
3554 }
3557 {
3558 /* A negative upper bound stands for infinity,
3559 in which case it degenerates to a plain jump. */
3562 }
3565 append a backward jump to the `succeed_n'
3566 that starts this interval.
3568 When we've reached this during matching,
3569 we'll have matched the interval once, so
3570 jump back only `upper_bound - 1' times. */
3575 /* The location we want to set is the second
3576 parameter of the `jump_n'; that is `b-2' as
3577 an absolute address. `laststart' will be
3578 the `set_number_at' we're about to insert;
3579 `laststart+3' the number to set, the source
3580 for the relative address. But we are
3581 inserting into the middle of the pattern --
3582 so everything is getting moved up by 5.
3583 Conclusion: (b - 2) - (laststart + 3) + 5,
3584 i.e., b - laststart.
3586 We insert this at the beginning of the loop
3587 so that if we fail during matching, we'll
3588 reinitialize the bounds. */
3592 }
3593 }
3596 }
3599 unfetch_interval:
3600 /* If an invalid interval, match the characters as literals. */
3605 /* normal_char and normal_backslash need `c'. */
3609 {
3612 }
3613 else
3616 #ifdef emacs
3617 /* There is no way to specify the before_dot and after_dot
3618 operators. rms says this is ok. --karl */
3686 else
3716 {
3717 regnum_t reg;
3725 /* Can't back reference to a subexp before its end. */
3731 }
3739 else
3743 normal_backslash:
3744 /* You might think it would be useful for \ to mean
3745 not to translate; but if we don't translate it
3746 it will never match anything. */
3748 }
3753 /* Expects the character in `c'. */
3754 normal_char:
3755 /* If no exactn currently being built. */
3758 /* If last exactn not at current position. */
3761 /* We have only one byte following the exactn for the count. */
3764 /* If followed by a repetition operator. */
3773 {
3774 /* Start building a new exactn. */
3780 }
3783 {
3787 {
3791 }
3792 else
3793 {
3796 {
3801 }
3804 }
3806 }
3813 /* Through the pattern now. */
3820 /* If we don't want backtracking, force success
3821 the first time we reach the end of the compiled pattern. */
3825 /* We have succeeded; set the length of the buffer. */
3828 #ifdef DEBUG
3830 {
3834 }
3835 debug--;
3838 #ifndef MATCH_MAY_ALLOCATE
3839 /* Initialize the failure stack to the largest possible stack. This
3840 isn't necessary unless we're trying to avoid calling alloca in
3841 the search and match routines. */
3842 {
3846 {
3850 fail_stack.stack
3853 else
3854 fail_stack.stack
3858 }
3861 }
3866 \f
3867 /* Subroutines for `regex_compile'. */
3869 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3871 static void
3873 re_opcode_t op;
3876 {
3879 }
3882 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3884 static void
3886 re_opcode_t op;
3889 {
3893 }
3896 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3897 for OP followed by two-byte integer parameter ARG. */
3899 static void
3901 re_opcode_t op;
3905 {
3913 }
3916 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3918 static void
3920 re_opcode_t op;
3924 {
3932 }
3935 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3936 after an alternative or a begin-subexpression. We assume there is at
3937 least one character before the ^. */
3939 static boolean
3942 reg_syntax_t syntax;
3943 {
3947 return
3948 /* After a subexpression? */
3950 /* After an alternative? */
3952 /* After a shy subexpression? */
3957 }
3960 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3961 at least one character after the $, i.e., `P < PEND'. */
3963 static boolean
3966 reg_syntax_t syntax;
3967 {
3972 return
3973 /* Before a subexpression? */
3976 /* Before an alternative? */
3979 }
3982 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3983 false if it's not. */
3985 static boolean
3987 compile_stack_type compile_stack;
3988 regnum_t regnum;
3989 {
3994 this_element--)
3999 }
4000 \f
4001 /* analyse_first.
4002 If fastmap is non-NULL, go through the pattern and fill fastmap
4003 with all the possible leading chars. If fastmap is NULL, don't
4004 bother filling it up (obviously) and only return whether the
4005 pattern could potentially match the empty string.
4007 Return 1 if p..pend might match the empty string.
4008 Return 0 if p..pend matches at least one char.
4009 Return -1 if fastmap was not updated accurately. */
4011 static int
4016 {
4020 /* If all elements for base leading-codes in fastmap is set, this
4021 flag is set true. */
4026 /* The loop below works as follows:
4027 - It has a working-list kept in the PATTERN_STACK and which basically
4028 starts by only containing a pointer to the first operation.
4029 - If the opcode we're looking at is a match against some set of
4030 chars, then we add those chars to the fastmap and go on to the
4031 next work element from the worklist (done via `break').
4032 - If the opcode is a control operator on the other hand, we either
4033 ignore it (if it's meaningless at this point, such as `start_memory')
4034 or execute it (if it's a jump). If the jump has several destinations
4035 (i.e. `on_failure_jump'), then we push the other destination onto the
4036 worklist.
4037 We guarantee termination by ignoring backward jumps (more or less),
4038 so that `p' is monotonically increasing. More to the point, we
4039 never set `p' (or push) anything `<= p1'. */
4042 {
4043 /* `p1' is used as a marker of how far back a `on_failure_jump'
4044 can go without being ignored. It is normally equal to `p'
4045 (which prevents any backward `on_failure_jump') except right
4046 after a plain `jump', to allow patterns such as:
4047 0: jump 10
4048 3..9: <body>
4049 10: on_failure_jump 3
4050 as used for the *? operator. */
4054 {
4060 /* If the first character has to match a backreference, that means
4061 that the group was empty (since it already matched). Since this
4062 is the only case that interests us here, we can assume that the
4063 backreference must match the empty string. */
4064 p++;
4068 /* Following are the cases which match a character. These end
4069 with `break'. */
4073 {
4074 /* If multibyte is nonzero, the first byte of each
4075 character is an ASCII or a leading code. Otherwise,
4076 each byte is a character. Thus, this works in both
4077 cases. */
4080 {
4081 /* For the case of matching this unibyte regex
4082 against multibyte, we must set a leading code of
4083 the corresponding multibyte character. */
4088 }
4089 }
4094 /* We could put all the chars except for \n (and maybe \0)
4095 but we don't bother since it is generally not worth it. */
4102 {
4103 /* Chars beyond end of bitmap are possible matches. */
4107 }
4109 /* Fallthrough */
4118 #ifdef emacs
4120 which doesn't match the specified set of characters. */
4121 not
4122 ||
4123 /* If we can match a character class, we can match any
4124 multibyte characters. */
4128 {
4130 {
4135 }
4136 }
4140 {
4141 /* Set fastmap[I] to 1 where I is a leading code of each
4142 multibyte characer in the range table. */
4146 /* Make P points the range table. `+ 2' is to skip flag
4147 bits for a character class. */
4150 /* Extract the number of ranges in range table into COUNT. */
4153 {
4154 /* Extract the start and end of each range. */
4162 }
4163 }
4164 #endif
4170 #ifndef emacs
4178 /* This match depends on text properties. These end with
4179 aborting optimizations. */
4191 /* Any leading code can possibly start a character which
4192 has or doesn't has the specified category. */
4194 {
4199 }
4202 /* All cases after this match the empty string. These end with
4203 `continue'. */
4226 /* Backward jumps can only go back to code that we've already
4227 visited. `re_compile' should make sure this is true. */
4231 {
4237 p++;
4241 };
4242 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4243 to jump back to "just after here". */
4244 /* Fallthrough */
4254 else
4256 We first go down the "straight" path so as to minimize
4257 stack usage when going through alternatives. */
4261 }
4266 /* This code simply does not properly handle forward jump_n. */
4269 /* jump_n can either jump or fall through. The (backward) jump
4270 case has already been handled, so we only need to look at the
4271 fallthrough case. */
4275 /* If N == 0, it should be an on_failure_jump_loop instead. */
4278 /* We only care about one iteration of the loop, so we don't
4279 need to consider the case where this behaves like an
4280 on_failure_jump. */
4299 /* Getting here means we have found the possible starting
4300 characters for one path of the pattern -- and that the empty
4301 string does not match. We need not follow this path further. */
4305 /* We reached the end without matching anything. */
4309 \f
4310 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4311 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4312 characters can start a string that matches the pattern. This fastmap
4313 is used by re_search to skip quickly over impossible starting points.
4315 Character codes above (1 << BYTEWIDTH) are not represented in the
4316 fastmap, but the leading codes are represented. Thus, the fastmap
4317 indicates which character sets could start a match.
4319 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4320 area as BUFP->fastmap.
4322 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4323 the pattern buffer.
4325 Returns 0 if we succeed, -2 if an internal error. */
4327 int
4330 {
4344 \f
4345 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4346 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4347 this memory for recording register information. STARTS and ENDS
4348 must be allocated using the malloc library routine, and must each
4349 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4351 If NUM_REGS == 0, then subsequent matches should allocate their own
4352 register data.
4354 Unless this function is called, the first search or match using
4355 PATTERN_BUFFER will allocate its own register data, without
4356 freeing the old data. */
4358 void
4364 {
4366 {
4371 }
4372 else
4373 {
4377 }
4378 }
4380 \f
4381 /* Searching routines. */
4383 /* Like re_search_2, below, but only one string is specified, and
4384 doesn't let you say where to stop matching. */
4386 int
4392 {
4395 }
4398 /* Head address of virtual concatenation of string. */
4399 #define HEAD_ADDR_VSTRING(P) \
4400 (((P) >= size1 ? string2 : string1))
4402 /* End address of virtual concatenation of string. */
4403 #define STOP_ADDR_VSTRING(P) \
4404 (((P) >= size1 ? string2 + size2 : string1 + size1))
4406 /* Address of POS in the concatenation of virtual string. */
4407 #define POS_ADDR_VSTRING(POS) \
4408 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4410 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4411 virtual concatenation of STRING1 and STRING2, starting first at index
4412 STARTPOS, then at STARTPOS + 1, and so on.
4414 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4416 RANGE is how far to scan while trying to match. RANGE = 0 means try
4417 only at STARTPOS; in general, the last start tried is STARTPOS +
4418 RANGE.
4420 In REGS, return the indices of the virtual concatenation of STRING1
4421 and STRING2 that matched the entire BUFP->buffer and its contained
4422 subexpressions.
4424 Do not consider matching one past the index STOP in the virtual
4425 concatenation of STRING1 and STRING2.
4427 We return either the position in the strings at which the match was
4428 found, -1 if no match, or -2 if error (such as failure
4429 stack overflow). */
4431 int
4440 {
4448 boolean anchored_start;
4449 /* Nonzero if we are searching multibyte string. */
4452 /* Check for out-of-range STARTPOS. */
4456 /* Fix up RANGE if it might eventually take us outside
4457 the virtual concatenation of STRING1 and STRING2.
4458 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4464 /* If the search isn't to be a backwards one, don't waste time in a
4465 search for a pattern anchored at beginning of buffer. */
4467 {
4470 else
4472 }
4474 #ifdef emacs
4475 /* In a forward search for something that starts with \=.
4476 don't keep searching past point. */
4478 {
4482 }
4485 /* Update the fastmap now if not correct already. */
4489 /* See whether the pattern is anchored. */
4492 #ifdef emacs
4494 {
4498 }
4499 #endif
4501 /* Loop through the string, looking for a place to start matching. */
4503 {
4504 /* If the pattern is anchored,
4505 skip quickly past places we cannot match.
4506 We don't bother to treat startpos == 0 specially
4507 because that case doesn't repeat. */
4509 {
4514 }
4516 /* If a fastmap is supplied, skip quickly over characters that
4517 cannot be the start of a match. If the pattern can match the
4518 null string, however, we don't need to skip characters; we want
4519 the first null string. */
4521 {
4528 {
4535 /* Written out as an if-else to avoid testing `translate'
4536 inside the loop. */
4538 {
4541 {
4545 buf_charlen);
4552 }
4553 else
4555 {
4566 d++;
4567 range--;
4568 }
4569 }
4570 else
4571 {
4574 {
4578 buf_charlen);
4583 }
4584 else
4586 {
4587 d++;
4588 range--;
4589 }
4590 }
4592 }
4594 {
4599 {
4604 }
4605 else
4606 {
4617 }
4618 }
4619 }
4621 /* If can't match the null string, and that's all we have left, fail. */
4635 advance:
4639 {
4640 /* Update STARTPOS to the next character boundary. */
4642 {
4651 }
4652 else
4653 {
4654 range--;
4655 startpos++;
4656 }
4657 }
4658 else
4659 {
4660 range++;
4661 startpos--;
4663 /* Update STARTPOS to the previous character boundary. */
4665 {
4670 /* Find the head of multibyte form. */
4677 }
4678 }
4679 }
4683 \f
4684 /* Declarations and macros for re_match_2. */
4688 RE_TRANSLATE_TYPE translate,
4691 /* This converts PTR, a pointer into one of the search strings `string1'
4692 and `string2' into an offset from the beginning of that string. */
4693 #define POINTER_TO_OFFSET(ptr) \
4694 (FIRST_STRING_P (ptr) \
4695 ? ((regoff_t) ((ptr) - string1)) \
4696 : ((regoff_t) ((ptr) - string2 + size1)))
4698 /* Call before fetching a character with *d. This switches over to
4699 string2 if necessary.
4700 Check re_match_2_internal for a discussion of why end_match_2 might
4701 not be within string2 (but be equal to end_match_1 instead). */
4702 #define PREFETCH() \
4703 while (d == dend) \
4704 { \
4706 if (dend == end_match_2) \
4707 goto fail; \
4709 d = string2; \
4710 dend = end_match_2; \
4711 }
4713 /* Call before fetching a char with *d if you already checked other limits.
4714 This is meant for use in lookahead operations like wordend, etc..
4715 where we might need to look at parts of the string that might be
4716 outside of the LIMITs (i.e past `stop'). */
4717 #define PREFETCH_NOLIMIT() \
4718 if (d == end1) \
4719 { \
4720 d = string2; \
4721 dend = end_match_2; \
4722 } \
4724 /* Test if at very beginning or at very end of the virtual concatenation
4725 of `string1' and `string2'. If only one string, it's `string2'. */
4726 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4727 #define AT_STRINGS_END(d) ((d) == end2)
4730 /* Test if D points to a character which is word-constituent. We have
4731 two special cases to check for: if past the end of string1, look at
4732 the first character in string2; and if before the beginning of
4733 string2, look at the last character in string1. */
4734 #define WORDCHAR_P(d) \
4735 (SYNTAX ((d) == end1 ? *string2 \
4736 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4737 == Sword)
4739 /* Disabled due to a compiler bug -- see comment at case wordbound */
4741 /* The comment at case wordbound is following one, but we don't use
4742 AT_WORD_BOUNDARY anymore to support multibyte form.
4744 The DEC Alpha C compiler 3.x generates incorrect code for the
4745 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4746 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4747 macro and introducing temporary variables works around the bug. */
4749 #if 0
4750 /* Test if the character before D and the one at D differ with respect
4751 to being word-constituent. */
4752 #define AT_WORD_BOUNDARY(d) \
4753 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4754 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4755 #endif
4757 /* Free everything we malloc. */
4758 #ifdef MATCH_MAY_ALLOCATE
4759 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4760 # define FREE_VARIABLES() \
4761 do { \
4762 REGEX_FREE_STACK (fail_stack.stack); \
4763 FREE_VAR (regstart); \
4764 FREE_VAR (regend); \
4765 FREE_VAR (best_regstart); \
4766 FREE_VAR (best_regend); \
4767 } while (0)
4768 #else
4772 \f
4773 /* Optimization routines. */
4775 /* If the operation is a match against one or more chars,
4776 return a pointer to the next operation, else return NULL. */
4780 {
4782 {
4793 {
4798 }
4799 else
4805 #ifdef emacs
4809 p++;
4814 }
4816 }
4819 /* Jump over non-matching operations. */
4823 {
4826 {
4828 {
4841 }
4842 }
4845 }
4847 /* Non-zero if "p1 matches something" implies "p2 fails". */
4848 static int
4852 {
4853 re_opcode_t op2;
4860 /* Skip over open/close-group commands.
4861 If what follows this loop is a ...+ construct,
4862 look at what begins its body, since we will have to
4863 match at least one of that. */
4865 /* The same skip can be done for p1, except that this function
4866 is only used in the case where p1 is a simple match operator. */
4867 /* p1 = skip_noops (p1, pend); */
4875 {
4878 /* If we're at the end of the pattern, we can change. */
4880 {
4883 }
4888 {
4889 register re_wchar_t c
4894 {
4896 {
4899 }
4900 }
4904 {
4907 /* Test if C is listed in charset (or charset_not)
4908 at `p1'. */
4910 {
4914 }
4918 /* `not' is equal to 1 if c would match, which means
4919 that we can't change to pop_failure_jump. */
4921 {
4924 }
4925 }
4928 {
4931 }
4932 }
4936 {
4938 /* Reuse the code above. */
4941 /* It is hard to list up all the character in charset
4942 P2 if it includes multibyte character. Give up in
4943 such case. */
4945 {
4946 /* Now, we are sure that P2 has no range table.
4947 So, for the size of bitmap in P2, `p2[1]' is
4948 enough. But P1 may have range table, so the
4949 size of bitmap table of P1 is extracted by
4950 using macro `CHARSET_BITMAP_SIZE'.
4952 In a multibyte case, we know that all the character
4953 listed in P2 is ASCII. In a unibyte case, P1 has only a
4954 bitmap table. So, in both cases, it is enough to test
4955 only the bitmap table of P1. */
4958 {
4960 /* We win if the charset inside the loop
4961 has no overlap with the one after the loop. */
4965 idx++)
4971 {
4974 }
4975 }
4977 {
4979 /* We win if the charset_not inside the loop lists
4980 every character listed in the charset after. */
4988 {
4991 }
4992 }
4993 }
4994 }
4999 {
5002 /* Reuse the code above. */
5005 /* When we have two charset_not, it's very unlikely that
5006 they don't overlap. The union of the two sets of excluded
5007 chars should cover all possible chars, which, as a matter of
5008 fact, is virtually impossible in multibyte buffers. */
5010 }
5034 #ifdef emacs
5042 ;
5043 }
5045 /* Safe default. */
5047 }
5049 \f
5050 /* Matching routines. */
5053 /* re_match is like re_match_2 except it takes only a single string. */
5055 int
5061 {
5065 }
5069 #ifdef emacs
5070 /* In Emacs, this is the string or buffer in which we
5071 are matching. It is used for looking up syntax properties. */
5072 Lisp_Object re_match_object;
5073 #endif
5075 /* re_match_2 matches the compiled pattern in BUFP against the
5076 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5077 and SIZE2, respectively). We start matching at POS, and stop
5078 matching at STOP.
5080 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5081 store offsets for the substring each group matched in REGS. See the
5082 documentation for exactly how many groups we fill.
5084 We return -1 if no match, -2 if an internal error (such as the
5085 failure stack overflowing). Otherwise, we return the length of the
5086 matched substring. */
5088 int
5096 {
5099 #ifdef emacs
5104 #endif
5110 }
5114 /* This is a separate function so that we can force an alloca cleanup
5115 afterwards. */
5116 static int
5124 {
5125 /* General temporaries. */
5130 /* Just past the end of the corresponding string. */
5133 /* Pointers into string1 and string2, just past the last characters in
5134 each to consider matching. */
5137 /* Where we are in the data, and the end of the current string. */
5140 /* Used sometimes to remember where we were before starting matching
5141 an operator so that we can go back in case of failure. This "atomic"
5142 behavior of matching opcodes is indispensable to the correctness
5143 of the on_failure_keep_string_jump optimization. */
5146 /* Where we are in the pattern, and the end of the pattern. */
5150 /* We use this to map every character in the string. */
5153 /* Nonzero if BUFP is setup from a multibyte regex. */
5156 /* Nonzero if STRING1/STRING2 are multibyte. */
5159 /* Failure point stack. Each place that can handle a failure further
5160 down the line pushes a failure point on this stack. It consists of
5161 regstart, and regend for all registers corresponding to
5162 the subexpressions we're currently inside, plus the number of such
5163 registers, and, finally, two char *'s. The first char * is where
5164 to resume scanning the pattern; the second one is where to resume
5165 scanning the strings. */
5167 fail_stack_type fail_stack;
5168 #endif
5169 #ifdef DEBUG
5171 #endif
5173 #if defined REL_ALLOC && defined REGEX_MALLOC
5174 /* This holds the pointer to the failure stack, when
5175 it is allocated relocatably. */
5177 #endif
5179 /* We fill all the registers internally, independent of what we
5180 return, for use in backreferences. The number here includes
5181 an element for register zero. */
5184 /* Information on the contents of registers. These are pointers into
5185 the input strings; they record just what was matched (on this
5186 attempt) by a subexpression part of the pattern, that is, the
5187 regnum-th regstart pointer points to where in the pattern we began
5188 matching and the regnum-th regend points to right after where we
5189 stopped matching the regnum-th subexpression. (The zeroth register
5190 keeps track of what the whole pattern matches.) */
5193 #endif
5195 /* The following record the register info as found in the above
5196 variables when we find a match better than any we've seen before.
5197 This happens as we backtrack through the failure points, which in
5198 turn happens only if we have not yet matched the entire string. */
5202 #endif
5204 /* Logically, this is `best_regend[0]'. But we don't want to have to
5205 allocate space for that if we're not allocating space for anything
5206 else (see below). Also, we never need info about register 0 for
5207 any of the other register vectors, and it seems rather a kludge to
5208 treat `best_regend' differently than the rest. So we keep track of
5209 the end of the best match so far in a separate variable. We
5210 initialize this to NULL so that when we backtrack the first time
5211 and need to test it, it's not garbage. */
5214 #ifdef DEBUG
5215 /* Counts the total number of registers pushed. */
5217 #endif
5223 #ifdef MATCH_MAY_ALLOCATE
5224 /* Do not bother to initialize all the register variables if there are
5225 no groups in the pattern, as it takes a fair amount of time. If
5226 there are groups, we include space for register 0 (the whole
5227 pattern), even though we never use it, since it simplifies the
5228 array indexing. We should fix this. */
5230 {
5237 {
5240 }
5241 }
5242 else
5243 {
5244 /* We must initialize all our variables to NULL, so that
5245 `FREE_VARIABLES' doesn't try to free them. */
5247 }
5250 /* The starting position is bogus. */
5252 {
5255 }
5257 /* Initialize subexpression text positions to -1 to mark ones that no
5258 start_memory/stop_memory has been seen for. Also initialize the
5259 register information struct. */
5263 /* We move `string1' into `string2' if the latter's empty -- but not if
5264 `string1' is null. */
5266 {
5271 }
5275 /* `p' scans through the pattern as `d' scans through the data.
5276 `dend' is the end of the input string that `d' points within. `d'
5277 is advanced into the following input string whenever necessary, but
5278 this happens before fetching; therefore, at the beginning of the
5279 loop, `d' can be pointing at the end of a string, but it cannot
5280 equal `string2'. */
5282 {
5283 /* Only match within string2. */
5287 }
5288 else
5289 {
5291 {
5292 /* Only match within string1. */
5294 /* BEWARE!
5295 When we reach end_match_1, PREFETCH normally switches to string2.
5296 But in the present case, this means that just doing a PREFETCH
5297 makes us jump from `stop' to `gap' within the string.
5298 What we really want here is for the search to stop as
5299 soon as we hit end_match_1. That's why we set end_match_2
5300 to end_match_1 (since PREFETCH fails as soon as we hit
5301 end_match_2). */
5303 }
5304 else
5306 moving `d' from end1 to string2 will not prevent the d == dend
5307 check from catching the end of string. */
5310 }
5313 }
5321 /* This loops over pattern commands. It exits by returning from the
5322 function if the match is complete, or it drops through if the match
5323 fails at this starting point in the input data. */
5325 {
5332 /* If we haven't matched the entire string, and we want the
5333 longest match, try backtracking. */
5335 {
5336 /* 1 if this match ends in the same string (string1 or string2)
5337 as the best previous match. */
5340 /* 1 if this match is the best seen so far. */
5341 boolean best_match_p;
5343 /* AIX compiler got confused when this was combined
5344 with the previous declaration. */
5347 else
5355 /* If exceeds best match so far, save it. */
5357 {
5364 {
5367 }
5368 }
5370 }
5372 /* If no failure points, don't restore garbage. And if
5373 last match is real best match, don't restore second
5374 best one. */
5376 {
5377 restore_best_regs:
5378 /* Restore best match. It may happen that `dend ==
5379 end_match_1' while the restored d is in string2.
5380 For example, the pattern `x.*y.*z' against the
5381 strings `x-' and `y-z-', if the two strings are
5382 not consecutive in memory. */
5390 {
5393 }
5394 }
5397 succeed_label:
5400 /* If caller wants register contents data back, do it. */
5402 {
5403 /* Have the register data arrays been allocated? */
5406 extra element beyond `num_regs' for the `-1' marker
5407 GNU code uses. */
5412 {
5415 }
5417 }
5420 allocated, reallocate them. If we need fewer, just
5421 leave it alone. */
5423 {
5428 {
5431 }
5432 }
5433 }
5434 else
5435 {
5436 /* These braces fend off a "empty body in an else-statement"
5437 warning under GCC when assert expands to nothing. */
5439 }
5441 /* Convert the pointer data in `regstart' and `regend' to
5442 indices. Register zero has to be set differently,
5443 since we haven't kept track of any info for it. */
5445 {
5448 }
5450 /* Go through the first `min (num_regs, regs->num_regs)'
5451 registers, since that is all we initialized. */
5453 {
5456 else
5457 {
5462 }
5463 }
5465 /* If the regs structure we return has more elements than
5466 were in the pattern, set the extra elements to -1. If
5467 we (re)allocated the registers, this is the case,
5468 because we always allocate enough to have at least one
5469 -1 at the end. */
5485 }
5487 /* Otherwise match next pattern command. */
5489 {
5490 /* Ignore these. Used to ignore the n of succeed_n's which
5491 currently have n == 0. */
5500 /* Match the next n pattern characters exactly. The following
5501 byte in the pattern defines n, and the n bytes after that
5502 are the characters to match. */
5507 /* Remember the start point to rollback upon failure. */
5510 #ifndef emacs
5511 /* This is written out as an if-else so we don't waste time
5512 testing `translate' inside the loop. */
5514 do
5515 {
5518 {
5521 }
5522 d++;
5523 }
5525 else
5526 do
5527 {
5530 {
5533 }
5534 }
5537 /* The cost of testing `translate' is comparatively small. */
5539 do
5540 {
5547 else
5548 {
5551 }
5555 {
5558 }
5563 }
5565 else
5566 do
5567 {
5573 {
5577 else
5579 }
5580 else
5581 {
5584 }
5587 {
5592 }
5594 {
5597 }
5599 d++;
5600 }
5602 #endif
5606 /* Match any character except possibly a newline or a null. */
5608 {
5610 re_wchar_t buf_ch;
5616 target_multibyte);
5627 }
5633 {
5638 /* Start of actual range_table, or end of bitmap if there is no
5639 range table. */
5642 /* Nonzero if there is a range table. */
5645 /* Number of ranges of range table. This is not included
5646 in the initial byte-length of the command. */
5654 {
5657 }
5662 {
5669 }
5670 else
5671 {
5675 {
5680 }
5681 }
5685 /* Cast to `unsigned' instead of `unsigned char' in
5686 case the bit list is a full 32 bytes long. */
5690 }
5691 #ifdef emacs
5693 {
5703 else
5705 }
5710 else
5717 }
5720 /* The beginning of a group is represented by start_memory.
5721 The argument is the register number. The text
5722 matched within the group is recorded (in the internal
5723 registers data structure) under the register number. */
5727 /* In case we need to undo this operation (via backtracking). */
5734 /* Move past the register number and inner group count. */
5739 /* The stop_memory opcode represents the end of a group. Its
5740 argument is the same as start_memory's: the register number. */
5745 /* Strictly speaking, there should be code such as:
5747 assert (REG_UNSET (regend[*p]));
5748 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5750 But the only info to be pushed is regend[*p] and it is known to
5751 be UNSET, so there really isn't anything to push.
5752 Not pushing anything, on the other hand deprives us from the
5753 guarantee that regend[*p] is UNSET since undoing this operation
5754 will not reset its value properly. This is not important since
5755 the value will only be read on the next start_memory or at
5756 the very end and both events can only happen if this stop_memory
5757 is *not* undone. */
5762 /* Move past the register number and the inner group count. */
5767 /* \<digit> has been turned into a `duplicate' command which is
5768 followed by the numeric value of <digit> as the register number. */
5770 {
5775 /* Can't back reference a group which we've never matched. */
5779 /* Where in input to try to start matching. */
5782 /* Remember the start point to rollback upon failure. */
5785 /* Where to stop matching; if both the place to start and
5786 the place to stop matching are in the same string, then
5787 set to the place to stop, otherwise, for now have to use
5788 the end of the first string. */
5794 {
5795 /* If necessary, advance to next segment in register
5796 contents. */
5798 {
5802 /* End of string1 => advance to string2. */
5805 }
5806 /* At end of register contents => success */
5809 /* If necessary, advance to next segment in data. */
5812 /* How many characters left in this segment to match. */
5815 /* Want how many consecutive characters we can match in
5816 one shot, so, if necessary, adjust the count. */
5820 /* Compare that many; failure if mismatch, else move
5821 past them. */
5825 {
5828 }
5830 }
5831 }
5835 /* begline matches the empty string at the beginning of the string
5836 (unless `not_bol' is set in `bufp'), and after newlines. */
5841 {
5843 }
5844 else
5845 {
5850 }
5851 /* In all other cases, we fail. */
5855 /* endline is the dual of begline. */
5860 {
5862 }
5863 else
5864 {
5868 }
5872 /* Match at the very beginning of the data. */
5880 /* Match at the very end of the data. */
5888 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5889 pushes NULL as the value for the string on the stack. Then
5890 `POP_FAILURE_POINT' will keep the current value for the
5891 string, instead of restoring it. To see why, consider
5892 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5893 then the . fails against the \n. But the next thing we want
5894 to do is match the \n against the \n; if we restored the
5895 string value, we would be back at the foo.
5897 Because this is used only in specific cases, we don't need to
5898 check all the things that `on_failure_jump' does, to make
5899 sure the right things get saved on the stack. Hence we don't
5900 share its code. The only reason to push anything on the
5901 stack at all is that otherwise we would have to change
5902 `anychar's code to do something besides goto fail in this
5903 case; that seems worse than this. */
5912 /* A nasty loop is introduced by the non-greedy *? and +?.
5913 With such loops, the stack only ever contains one failure point
5914 at a time, so that a plain on_failure_jump_loop kind of
5915 cycle detection cannot work. Worse yet, such a detection
5916 can not only fail to detect a cycle, but it can also wrongly
5917 detect a cycle (between different instantiations of the same
5918 loop).
5919 So the method used for those nasty loops is a little different:
5920 We use a special cycle-detection-stack-frame which is pushed
5921 when the on_failure_jump_nastyloop failure-point is *popped*.
5922 This special frame thus marks the beginning of one iteration
5923 through the loop and we can hence easily check right here
5924 whether something matched between the beginning and the end of
5925 the loop. */
5932 {
5936 /* If there's a cycle, just continue without pushing
5937 this failure point. The failure point is the "try again"
5938 option, which shouldn't be tried.
5939 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5941 }
5944 /* Simple loop detecting on_failure_jump: just check on the
5945 failure stack if the same spot was already hit earlier. */
5947 on_failure:
5951 {
5955 /* If there's a cycle, get out of the loop, as if the matching
5956 had failed. We used to just `goto fail' here, but that was
5957 aborting the search a bit too early: we want to keep the
5958 empty-loop-match and keep matching after the loop.
5959 We want (x?)*y\1z to match both xxyz and xxyxz. */
5961 else
5963 }
5967 /* Uses of on_failure_jump:
5969 Each alternative starts with an on_failure_jump that points
5970 to the beginning of the next alternative. Each alternative
5971 except the last ends with a jump that in effect jumps past
5972 the rest of the alternatives. (They really jump to the
5973 ending jump of the following alternative, because tensioning
5974 these jumps is a hassle.)
5976 Repeats start with an on_failure_jump that points past both
5977 the repetition text and either the following jump or
5978 pop_failure_jump back to this on_failure_jump. */
5987 /* This operation is used for greedy *.
5988 Compare the beginning of the repeat with what in the
5989 pattern follows its end. If we can establish that there
5990 is nothing that they would both match, i.e., that we
5991 would have to backtrack because of (as in, e.g., `a*a')
5992 then we can use a non-backtracking loop based on
5993 on_failure_keep_string_jump instead of on_failure_jump. */
5998 {
6000 /* Here, we discard `const', making re_match non-reentrant. */
6005 instruction once it's been changed. */
6009 /* Ensure this is a indeed the trivial kind of loop
6010 we are expecting. */
6015 {
6016 /* Use a fast `on_failure_keep_string_jump' loop. */
6020 }
6021 else
6022 {
6023 /* Default to a safe `on_failure_jump' loop. */
6026 }
6028 }
6031 /* Unconditionally jump (without popping any failure points). */
6033 unconditional_jump:
6034 IMMEDIATE_QUIT_CHECK;
6042 /* Have to succeed matching what follows at least n times.
6043 After that, handle like `on_failure_jump'. */
6045 /* Signedness doesn't matter since we only compare MCNT to 0. */
6049 /* Originally, mcnt is how many times we HAVE to succeed. */
6051 {
6052 /* Here, we discard `const', making re_match non-reentrant. */
6054 mcnt--;
6057 }
6058 else
6059 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
6064 /* Signedness doesn't matter since we only compare MCNT to 0. */
6068 /* Originally, this is how many times we CAN jump. */
6070 {
6071 /* Here, we discard `const', making re_match non-reentrant. */
6073 mcnt--;
6076 }
6077 /* If don't have to jump any more, skip over the rest of command. */
6078 else
6083 {
6088 /* Here, we discard `const', making re_match non-reentrant. */
6090 /* Signedness doesn't matter since we only copy MCNT's bits . */
6095 }
6102 /* We SUCCEED (or FAIL) in one of the following cases: */
6104 /* Case 1: D is at the beginning or the end of string. */
6107 else
6108 {
6109 /* C1 is the character before D, S1 is the syntax of C1, C2
6110 is the character at D, and S2 is the syntax of C2. */
6114 #ifdef emacs
6118 #endif
6121 #ifdef emacs
6123 #endif
6130 /* Case 3: Both of S1 and S2 are Sword, and macro
6131 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6134 }
6137 else
6143 /* We FAIL in one of the following cases: */
6145 /* Case 1: D is at the end of string. */
6148 else
6149 {
6150 /* C1 is the character before D, S1 is the syntax of C1, C2
6151 is the character at D, and S2 is the syntax of C2. */
6155 #ifdef emacs
6159 #endif
6164 /* Case 2: S2 is not Sword. */
6168 /* Case 3: D is not at the beginning of string ... */
6170 {
6172 #ifdef emacs
6174 #endif
6177 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6178 returns 0. */
6181 }
6182 }
6188 /* We FAIL in one of the following cases: */
6190 /* Case 1: D is at the beginning of string. */
6193 else
6194 {
6195 /* C1 is the character before D, S1 is the syntax of C1, C2
6196 is the character at D, and S2 is the syntax of C2. */
6200 #ifdef emacs
6204 #endif
6208 /* Case 2: S1 is not Sword. */
6212 /* Case 3: D is not at the end of string ... */
6214 {
6217 #ifdef emacs
6219 #endif
6222 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6223 returns 0. */
6226 }
6227 }
6233 /* We FAIL in one of the following cases: */
6235 /* Case 1: D is at the end of string. */
6238 else
6239 {
6240 /* C1 is the character before D, S1 is the syntax of C1, C2
6241 is the character at D, and S2 is the syntax of C2. */
6244 #ifdef emacs
6248 #endif
6253 /* Case 2: S2 is neither Sword nor Ssymbol. */
6257 /* Case 3: D is not at the beginning of string ... */
6259 {
6261 #ifdef emacs
6263 #endif
6266 /* ... and S1 is Sword or Ssymbol. */
6269 }
6270 }
6276 /* We FAIL in one of the following cases: */
6278 /* Case 1: D is at the beginning of string. */
6281 else
6282 {
6283 /* C1 is the character before D, S1 is the syntax of C1, C2
6284 is the character at D, and S2 is the syntax of C2. */
6287 #ifdef emacs
6291 #endif
6295 /* Case 2: S1 is neither Ssymbol nor Sword. */
6299 /* Case 3: D is not at the end of string ... */
6301 {
6304 #ifdef emacs
6306 #endif
6309 /* ... and S2 is Sword or Ssymbol. */
6312 }
6313 }
6322 #ifdef emacs
6323 {
6327 }
6328 #endif
6329 {
6331 re_wchar_t c;
6337 }
6340 #ifdef emacs
6365 {
6367 re_wchar_t c;
6373 }
6380 }
6384 /* We goto here if a matching operation fails. */
6385 fail:
6386 IMMEDIATE_QUIT_CHECK;
6388 {
6390 /* A restart point is known. Restore to that state. */
6394 {
6402 /* Fallthrough */
6408 continue_failure_jump:
6414 /* A special frame used for nastyloops. */
6419 }
6425 }
6426 else
6437 \f
6438 /* Subroutine definitions for re_match_2. */
6440 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6441 bytes; nonzero otherwise. */
6443 static int
6447 RE_TRANSLATE_TYPE translate;
6449 {
6454 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6455 different lengths, but relying on a single `len' would break this. -sm */
6457 {
6469 }
6475 }
6476 \f
6477 /* Entry points for GNU code. */
6479 /* re_compile_pattern is the GNU regular expression compiler: it
6480 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6481 Returns 0 if the pattern was valid, otherwise an error string.
6483 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6484 are set in BUFP on entry.
6486 We call regex_compile to do the actual compilation. */
6493 {
6494 reg_errcode_t ret;
6496 #ifdef emacs
6498 #endif
6500 /* GNU code is written to assume at least RE_NREGS registers will be set
6501 (and at least one extra will be -1). */
6504 /* And GNU code determines whether or not to get register information
6505 by passing null for the REGS argument to re_match, etc., not by
6506 setting no_sub. */
6514 }
6516 \f
6517 /* Entry points compatible with 4.2 BSD regex library. We don't define
6518 them unless specifically requested. */
6520 #if defined _REGEX_RE_COMP || defined _LIBC
6522 /* BSD has one and only one pattern buffer. */
6526 # ifdef _LIBC
6527 /* Make these definitions weak in libc, so POSIX programs can redefine
6528 these names if they don't use our functions, and still use
6529 regcomp/regexec below without link errors. */
6530 weak_function
6531 # endif
6534 {
6535 reg_errcode_t ret;
6538 {
6540 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6543 }
6546 {
6549 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6555 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6557 }
6559 /* Since `re_exec' always passes NULL for the `regs' argument, we
6560 don't need to initialize the pattern buffer fields which affect it. */
6567 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6569 }
6572 int
6573 # ifdef _LIBC
6574 weak_function
6575 # endif
6578 {
6580 return
6582 }
6584 \f
6585 /* POSIX.2 functions. Don't define these for Emacs. */
6587 #ifndef emacs
6589 /* regcomp takes a regular expression as a string and compiles it.
6591 PREG is a regex_t *. We do not expect any fields to be initialized,
6592 since POSIX says we shouldn't. Thus, we set
6594 `buffer' to the compiled pattern;
6595 `used' to the length of the compiled pattern;
6596 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6597 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6598 RE_SYNTAX_POSIX_BASIC;
6599 `fastmap' to an allocated space for the fastmap;
6600 `fastmap_accurate' to zero;
6601 `re_nsub' to the number of subexpressions in PATTERN.
6603 PATTERN is the address of the pattern string.
6605 CFLAGS is a series of bits which affect compilation.
6607 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6608 use POSIX basic syntax.
6610 If REG_NEWLINE is set, then . and [^...] don't match newline.
6611 Also, regexec will try a match beginning after every newline.
6613 If REG_ICASE is set, then we considers upper- and lowercase
6614 versions of letters to be equivalent when matching.
6616 If REG_NOSUB is set, then when PREG is passed to regexec, that
6617 routine will report only success or failure, and nothing about the
6618 registers.
6620 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6621 the return codes and their meanings.) */
6623 int
6628 {
6629 reg_errcode_t ret;
6630 reg_syntax_t syntax
6634 /* regex_compile will allocate the space for the compiled pattern. */
6639 /* Try to allocate space for the fastmap. */
6643 {
6646 preg->translate
6652 /* Map uppercase characters to corresponding lowercase ones. */
6655 }
6656 else
6659 /* If REG_NEWLINE is set, newlines are treated differently. */
6664 }
6665 else
6670 /* POSIX says a null character in the pattern terminates it, so we
6671 can use strlen here in compiling the pattern. */
6674 /* POSIX doesn't distinguish between an unmatched open-group and an
6675 unmatched close-group: both are REG_EPAREN. */
6681 buffer. */
6687 }
6688 }
6690 }
6694 /* regexec searches for a given pattern, specified by PREG, in the
6695 string STRING.
6697 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6698 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6699 least NMATCH elements, and we set them to the offsets of the
6700 corresponding matched substrings.
6702 EFLAGS specifies `execution flags' which affect matching: if
6703 REG_NOTBOL is set, then ^ does not match at the beginning of the
6704 string; if REG_NOTEOL is set, then $ does not match at the end.
6706 We return 0 if we find a match and REG_NOMATCH if not. */
6708 int
6715 {
6718 regex_t private_preg;
6727 /* The user has told us exactly how many registers to return
6728 information about, via `nmatch'. We have to pass that on to the
6729 matching routines. */
6733 {
6739 }
6741 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6742 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6743 was a little bit longer but still only matching the real part.
6744 This works because the `endline' will check for a '\n' and will find a
6745 '\0', correctly deciding that this is not the end of a line.
6746 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6747 a convenient '\0' there. For all we know, the string could be preceded
6748 by '\n' which would throw things off. */
6750 /* Perform the searching operation. */
6755 /* Copy the register information to the POSIX structure. */
6757 {
6759 {
6763 {
6766 }
6767 }
6769 /* If we needed the temporary register info, free the space now. */
6771 }
6773 /* We want zero return to mean success, unlike `re_search'. */
6775 }
6779 /* Returns a message corresponding to an error code, ERR_CODE, returned
6780 from either regcomp or regexec. We don't use PREG here.
6782 ERR_CODE was previously called ERRCODE, but that name causes an
6783 error with msvc8 compiler. */
6785 size_t
6791 {
6797 /* Only error codes returned by the rest of the code should be passed
6798 to this routine. If we are given anything else, or if other regex
6799 code generates an invalid error code, then the program has a bug.
6800 Dump core so we can fix it. */
6808 {
6810 {
6813 }
6814 else
6816 }
6819 }
6823 /* Free dynamically allocated space used by PREG. */
6825 void
6828 {
6844 }
6849 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6850 (do not change this comment) */