| blob | 9284be95ffbb83056a21285e5539b711725b5909 |
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) 1993-2011 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3, or (at your option)
10 any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
20 USA. */
22 /* TODO:
23 - structure the opcode space into opcode+flag.
24 - merge with glibc's regex.[ch].
25 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
26 need to modify the compiled regexp so that re_match can be reentrant.
27 - get rid of on_failure_jump_smart by doing the optimization in re_comp
28 rather than at run-time, so that re_match can be reentrant.
29 */
31 /* AIX requires this to be the first thing in the file. */
32 #if defined _AIX && !defined REGEX_MALLOC
33 #pragma alloca
34 #endif
36 #ifdef HAVE_CONFIG_H
37 # include <config.h>
38 #endif
40 #if defined STDC_HEADERS && !defined emacs
41 # include <stddef.h>
42 #else
43 /* We need this for `regex.h', and perhaps for the Emacs include files. */
44 # include <sys/types.h>
45 #endif
47 /* Whether to use ISO C Amendment 1 wide char functions.
48 Those should not be used for Emacs since it uses its own. */
49 #if defined _LIBC
50 #define WIDE_CHAR_SUPPORT 1
51 #else
52 #define WIDE_CHAR_SUPPORT \
53 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
54 #endif
56 /* For platform which support the ISO C amendement 1 functionality we
57 support user defined character classes. */
58 #if WIDE_CHAR_SUPPORT
59 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
60 # include <wchar.h>
61 # include <wctype.h>
62 #endif
64 #ifdef _LIBC
65 /* We have to keep the namespace clean. */
66 # define regfree(preg) __regfree (preg)
67 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
68 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
69 # define regerror(err_code, preg, errbuf, errbuf_size) \
70 __regerror(err_code, preg, errbuf, errbuf_size)
71 # define re_set_registers(bu, re, nu, st, en) \
72 __re_set_registers (bu, re, nu, st, en)
73 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
74 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
75 # define re_match(bufp, string, size, pos, regs) \
76 __re_match (bufp, string, size, pos, regs)
77 # define re_search(bufp, string, size, startpos, range, regs) \
78 __re_search (bufp, string, size, startpos, range, regs)
79 # define re_compile_pattern(pattern, length, bufp) \
80 __re_compile_pattern (pattern, length, bufp)
81 # define re_set_syntax(syntax) __re_set_syntax (syntax)
82 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
83 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
84 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
86 /* Make sure we call libc's function even if the user overrides them. */
87 # define btowc __btowc
88 # define iswctype __iswctype
89 # define wctype __wctype
91 # define WEAK_ALIAS(a,b) weak_alias (a, b)
93 /* We are also using some library internals. */
94 # include <locale/localeinfo.h>
95 # include <locale/elem-hash.h>
96 # include <langinfo.h>
97 #else
98 # define WEAK_ALIAS(a,b)
99 #endif
101 /* This is for other GNU distributions with internationalized messages. */
102 #if HAVE_LIBINTL_H || defined _LIBC
103 # include <libintl.h>
104 #else
105 # define gettext(msgid) (msgid)
106 #endif
108 #ifndef gettext_noop
109 /* This define is so xgettext can find the internationalizable
110 strings. */
111 # define gettext_noop(String) String
112 #endif
114 /* The `emacs' switch turns on certain matching commands
115 that make sense only in Emacs. */
116 #ifdef emacs
118 # include <setjmp.h>
122 /* Make syntax table lookup grant data in gl_state. */
123 # define SYNTAX_ENTRY_VIA_PROPERTY
129 # ifdef malloc
130 # undef malloc
131 # endif
132 # define malloc xmalloc
133 # ifdef realloc
134 # undef realloc
135 # endif
136 # define realloc xrealloc
137 # ifdef free
138 # undef free
139 # endif
140 # define free xfree
142 /* Converts the pointer to the char to BEG-based offset from the start. */
143 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
144 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
146 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
147 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
148 # define RE_STRING_CHAR(p, multibyte) \
149 (multibyte ? (STRING_CHAR (p)) : (*(p)))
150 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
151 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
153 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
155 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
157 /* Set C a (possibly converted to multibyte) character before P. P
158 points into a string which is the virtual concatenation of STR1
159 (which ends at END1) or STR2 (which ends at END2). */
160 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
161 do { \
162 if (target_multibyte) \
163 { \
164 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
165 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
166 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
167 c = STRING_CHAR (dtemp); \
168 } \
169 else \
170 { \
171 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
172 (c) = RE_CHAR_TO_MULTIBYTE (c); \
173 } \
174 } while (0)
176 /* Set C a (possibly converted to multibyte) character at P, and set
177 LEN to the byte length of that character. */
178 # define GET_CHAR_AFTER(c, p, len) \
179 do { \
180 if (target_multibyte) \
181 (c) = STRING_CHAR_AND_LENGTH (p, len); \
182 else \
183 { \
184 (c) = *p; \
185 len = 1; \
186 (c) = RE_CHAR_TO_MULTIBYTE (c); \
187 } \
188 } while (0)
192 /* If we are not linking with Emacs proper,
193 we can't use the relocating allocator
194 even if config.h says that we can. */
195 # undef REL_ALLOC
197 # include <unistd.h>
199 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
203 {
207 {
210 }
212 }
216 {
218 /* We must call malloc explicitly when BLOCK is 0, since some
219 reallocs don't do this. */
222 else
225 {
228 }
230 }
232 # ifdef malloc
233 # undef malloc
234 # endif
235 # define malloc xmalloc
236 # ifdef realloc
237 # undef realloc
238 # endif
239 # define realloc xrealloc
241 /* This is the normal way of making sure we have memcpy, memcmp and memset. */
242 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
243 # include <string.h>
244 # else
245 # include <strings.h>
246 # ifndef memcmp
247 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
248 # endif
249 # ifndef memcpy
250 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
251 # endif
252 # endif
254 /* Define the syntax stuff for \<, \>, etc. */
256 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
259 # define SWITCH_ENUM_CAST(x) (x)
261 /* Dummy macros for non-Emacs environments. */
262 # define CHAR_CHARSET(c) 0
263 # define CHARSET_LEADING_CODE_BASE(c) 0
264 # define MAX_MULTIBYTE_LENGTH 1
265 # define RE_MULTIBYTE_P(x) 0
266 # define RE_TARGET_MULTIBYTE_P(x) 0
267 # define WORD_BOUNDARY_P(c1, c2) (0)
268 # define CHAR_HEAD_P(p) (1)
269 # define SINGLE_BYTE_CHAR_P(c) (1)
270 # define SAME_CHARSET_P(c1, c2) (1)
271 # define BYTES_BY_CHAR_HEAD(p) (1)
272 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
273 # define STRING_CHAR(p) (*(p))
274 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
275 # define CHAR_STRING(c, s) (*(s) = (c), 1)
276 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
277 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
278 # define RE_CHAR_TO_MULTIBYTE(c) (c)
279 # define RE_CHAR_TO_UNIBYTE(c) (c)
280 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
281 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
282 # define GET_CHAR_AFTER(c, p, len) \
283 (c = *p, len = 1)
284 # define MAKE_CHAR(charset, c1, c2) (c1)
285 # define BYTE8_TO_CHAR(c) (c)
286 # define CHAR_BYTE8_P(c) (0)
287 # define CHAR_LEADING_CODE(c) (c)
291 #ifndef RE_TRANSLATE
292 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
293 # define RE_TRANSLATE_P(TBL) (TBL)
294 #endif
295 \f
296 /* Get the interface, including the syntax bits. */
299 /* isalpha etc. are used for the character classes. */
300 #include <ctype.h>
302 #ifdef emacs
304 /* 1 if C is an ASCII character. */
305 # define IS_REAL_ASCII(c) ((c) < 0200)
307 /* 1 if C is a unibyte character. */
308 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
310 /* The Emacs definitions should not be directly affected by locales. */
312 /* In Emacs, these are only used for single-byte characters. */
319 /* This is only used for single-byte characters. */
322 /* The rest must handle multibyte characters. */
324 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
326 : 1)
328 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
330 : 1)
332 # define ISALNUM(c) (IS_REAL_ASCII (c) \
336 : SYNTAX (c) == Sword)
338 # define ISALPHA(c) (IS_REAL_ASCII (c) \
341 : SYNTAX (c) == Sword)
343 # define ISLOWER(c) (LOWERCASEP (c))
345 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
350 : SYNTAX (c) != Sword)
352 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
354 # define ISUPPER(c) (UPPERCASEP (c))
356 # define ISWORD(c) (SYNTAX (c) == Sword)
360 /* Jim Meyering writes:
362 "... Some ctype macros are valid only for character codes that
363 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
364 using /bin/cc or gcc but without giving an ansi option). So, all
365 ctype uses should be through macros like ISPRINT... If
366 STDC_HEADERS is defined, then autoconf has verified that the ctype
367 macros don't need to be guarded with references to isascii. ...
368 Defining isascii to 1 should let any compiler worth its salt
369 eliminate the && through constant folding."
370 Solaris defines some of these symbols so we must undefine them first. */
372 # undef ISASCII
373 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
374 # define ISASCII(c) 1
375 # else
376 # define ISASCII(c) isascii(c)
377 # endif
379 /* 1 if C is an ASCII character. */
380 # define IS_REAL_ASCII(c) ((c) < 0200)
382 /* This distinction is not meaningful, except in Emacs. */
383 # define ISUNIBYTE(c) 1
385 # ifdef isblank
386 # define ISBLANK(c) (ISASCII (c) && isblank (c))
387 # else
389 # endif
390 # ifdef isgraph
391 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
392 # else
393 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
394 # endif
396 # undef ISPRINT
397 # define ISPRINT(c) (ISASCII (c) && isprint (c))
398 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
399 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
400 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
401 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
402 # define ISLOWER(c) (ISASCII (c) && islower (c))
403 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
404 # define ISSPACE(c) (ISASCII (c) && isspace (c))
405 # define ISUPPER(c) (ISASCII (c) && isupper (c))
406 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
408 # define ISWORD(c) ISALPHA(c)
410 # ifdef _tolower
411 # define TOLOWER(c) _tolower(c)
412 # else
413 # define TOLOWER(c) tolower(c)
414 # endif
416 /* How many characters in the character set. */
417 # define CHAR_SET_SIZE 256
419 # ifdef SYNTAX_TABLE
427 static void
429 {
445 }
449 # define SYNTAX(c) re_syntax_table[(c)]
452 \f
453 #ifndef NULL
454 # define NULL (void *)0
455 #endif
457 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
458 since ours (we hope) works properly with all combinations of
459 machines, compilers, `char' and `unsigned char' argument types.
460 (Per Bothner suggested the basic approach.) */
461 #undef SIGN_EXTEND_CHAR
462 #if __STDC__
463 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
465 /* As in Harbison and Steele. */
466 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
467 #endif
468 \f
469 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
470 use `alloca' instead of `malloc'. This is because using malloc in
471 re_search* or re_match* could cause memory leaks when C-g is used in
472 Emacs; also, malloc is slower and causes storage fragmentation. On
473 the other hand, malloc is more portable, and easier to debug.
475 Because we sometimes use alloca, some routines have to be macros,
476 not functions -- `alloca'-allocated space disappears at the end of the
477 function it is called in. */
479 #ifdef REGEX_MALLOC
481 # define REGEX_ALLOCATE malloc
482 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
483 # define REGEX_FREE free
487 /* Emacs already defines alloca, sometimes. */
488 # ifndef alloca
490 /* Make alloca work the best possible way. */
491 # ifdef __GNUC__
492 # define alloca __builtin_alloca
494 # ifdef HAVE_ALLOCA_H
495 # include <alloca.h>
501 # define REGEX_ALLOCATE alloca
503 /* Assumes a `char *destination' variable. */
504 # define REGEX_REALLOCATE(source, osize, nsize) \
505 (destination = (char *) alloca (nsize), \
506 memcpy (destination, source, osize))
508 /* No need to do anything to free, after alloca. */
513 /* Define how to allocate the failure stack. */
515 #if defined REL_ALLOC && defined REGEX_MALLOC
517 # define REGEX_ALLOCATE_STACK(size) \
518 r_alloc (&failure_stack_ptr, (size))
519 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
520 r_re_alloc (&failure_stack_ptr, (nsize))
521 # define REGEX_FREE_STACK(ptr) \
522 r_alloc_free (&failure_stack_ptr)
526 # ifdef REGEX_MALLOC
528 # define REGEX_ALLOCATE_STACK malloc
529 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
530 # define REGEX_FREE_STACK free
534 # define REGEX_ALLOCATE_STACK alloca
536 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
537 REGEX_REALLOCATE (source, osize, nsize)
538 /* No need to explicitly free anything. */
539 # define REGEX_FREE_STACK(arg) ((void)0)
545 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
546 `string1' or just past its end. This works if PTR is NULL, which is
547 a good thing. */
548 #define FIRST_STRING_P(ptr) \
549 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
551 /* (Re)Allocate N items of type T using malloc, or fail. */
552 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
553 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
554 #define RETALLOC_IF(addr, n, t) \
555 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
556 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
560 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
562 #undef MAX
563 #undef MIN
564 #define MAX(a, b) ((a) > (b) ? (a) : (b))
565 #define MIN(a, b) ((a) < (b) ? (a) : (b))
567 /* Type of source-pattern and string chars. */
571 #define false 0
572 #define true 1
580 \f
581 /* These are the command codes that appear in compiled regular
582 expressions. Some opcodes are followed by argument bytes. A
583 command code can specify any interpretation whatsoever for its
584 arguments. Zero bytes may appear in the compiled regular expression. */
587 {
590 /* Succeed right away--no more backtracking. */
591 succeed,
593 /* Followed by one byte giving n, then by n literal bytes. */
594 exactn,
596 /* Matches any (more or less) character. */
597 anychar,
599 /* Matches any one char belonging to specified set. First
600 following byte is number of bitmap bytes. Then come bytes
601 for a bitmap saying which chars are in. Bits in each byte
602 are ordered low-bit-first. A character is in the set if its
603 bit is 1. A character too large to have a bit in the map is
604 automatically not in the set.
606 If the length byte has the 0x80 bit set, then that stuff
607 is followed by a range table:
608 2 bytes of flags for character sets (low 8 bits, high 8 bits)
609 See RANGE_TABLE_WORK_BITS below.
610 2 bytes, the number of pairs that follow (upto 32767)
611 pairs, each 2 multibyte characters,
612 each multibyte character represented as 3 bytes. */
613 charset,
615 /* Same parameters as charset, but match any character that is
616 not one of those specified. */
617 charset_not,
619 /* Start remembering the text that is matched, for storing in a
620 register. Followed by one byte with the register number, in
621 the range 0 to one less than the pattern buffer's re_nsub
622 field. */
623 start_memory,
625 /* Stop remembering the text that is matched and store it in a
626 memory register. Followed by one byte with the register
627 number, in the range 0 to one less than `re_nsub' in the
628 pattern buffer. */
629 stop_memory,
631 /* Match a duplicate of something remembered. Followed by one
632 byte containing the register number. */
633 duplicate,
635 /* Fail unless at beginning of line. */
636 begline,
638 /* Fail unless at end of line. */
639 endline,
641 /* Succeeds if at beginning of buffer (if emacs) or at beginning
642 of string to be matched (if not). */
643 begbuf,
645 /* Analogously, for end of buffer/string. */
646 endbuf,
648 /* Followed by two byte relative address to which to jump. */
649 jump,
651 /* Followed by two-byte relative address of place to resume at
652 in case of failure. */
653 on_failure_jump,
655 /* Like on_failure_jump, but pushes a placeholder instead of the
656 current string position when executed. */
657 on_failure_keep_string_jump,
659 /* Just like `on_failure_jump', except that it checks that we
660 don't get stuck in an infinite loop (matching an empty string
661 indefinitely). */
662 on_failure_jump_loop,
664 /* Just like `on_failure_jump_loop', except that it checks for
665 a different kind of loop (the kind that shows up with non-greedy
666 operators). This operation has to be immediately preceded
667 by a `no_op'. */
668 on_failure_jump_nastyloop,
670 /* A smart `on_failure_jump' used for greedy * and + operators.
671 It analyses the loop before which it is put and if the
672 loop does not require backtracking, it changes itself to
673 `on_failure_keep_string_jump' and short-circuits the loop,
674 else it just defaults to changing itself into `on_failure_jump'.
675 It assumes that it is pointing to just past a `jump'. */
676 on_failure_jump_smart,
678 /* Followed by two-byte relative address and two-byte number n.
679 After matching N times, jump to the address upon failure.
680 Does not work if N starts at 0: use on_failure_jump_loop
681 instead. */
682 succeed_n,
684 /* Followed by two-byte relative address, and two-byte number n.
685 Jump to the address N times, then fail. */
686 jump_n,
688 /* Set the following two-byte relative address to the
689 subsequent two-byte number. The address *includes* the two
690 bytes of number. */
691 set_number_at,
702 /* Matches any character whose syntax is specified. Followed by
703 a byte which contains a syntax code, e.g., Sword. */
704 syntaxspec,
706 /* Matches any character whose syntax is not that specified. */
707 notsyntaxspec
709 #ifdef emacs
714 /* Matches any character whose category-set contains the specified
715 category. The operator is followed by a byte which contains a
716 category code (mnemonic ASCII character). */
717 categoryspec,
719 /* Matches any character whose category-set does not contain the
720 specified category. The operator is followed by a byte which
721 contains the category code (mnemonic ASCII character). */
722 notcategoryspec
725 \f
726 /* Common operations on the compiled pattern. */
728 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
730 #define STORE_NUMBER(destination, number) \
731 do { \
732 (destination)[0] = (number) & 0377; \
733 (destination)[1] = (number) >> 8; \
734 } while (0)
736 /* Same as STORE_NUMBER, except increment DESTINATION to
737 the byte after where the number is stored. Therefore, DESTINATION
738 must be an lvalue. */
740 #define STORE_NUMBER_AND_INCR(destination, number) \
741 do { \
742 STORE_NUMBER (destination, number); \
743 (destination) += 2; \
744 } while (0)
746 /* Put into DESTINATION a number stored in two contiguous bytes starting
747 at SOURCE. */
749 #define EXTRACT_NUMBER(destination, source) \
750 do { \
751 (destination) = *(source) & 0377; \
752 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
753 } while (0)
755 #ifdef DEBUG
757 static void
761 {
765 }
768 # undef EXTRACT_NUMBER
769 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
774 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
775 SOURCE must be an lvalue. */
777 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
778 do { \
779 EXTRACT_NUMBER (destination, source); \
780 (source) += 2; \
781 } while (0)
783 #ifdef DEBUG
786 static void
790 {
793 }
795 # ifndef EXTRACT_MACROS
796 # undef EXTRACT_NUMBER_AND_INCR
797 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
798 extract_number_and_incr (&dest, &src)
802 \f
803 /* Store a multibyte character in three contiguous bytes starting
804 DESTINATION, and increment DESTINATION to the byte after where the
805 character is stored. Therefore, DESTINATION must be an lvalue. */
807 #define STORE_CHARACTER_AND_INCR(destination, character) \
808 do { \
809 (destination)[0] = (character) & 0377; \
810 (destination)[1] = ((character) >> 8) & 0377; \
811 (destination)[2] = (character) >> 16; \
812 (destination) += 3; \
813 } while (0)
815 /* Put into DESTINATION a character stored in three contiguous bytes
816 starting at SOURCE. */
818 #define EXTRACT_CHARACTER(destination, source) \
819 do { \
820 (destination) = ((source)[0] \
821 | ((source)[1] << 8) \
822 | ((source)[2] << 16)); \
823 } while (0)
826 /* Macros for charset. */
828 /* Size of bitmap of charset P in bytes. P is a start of charset,
829 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
830 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
832 /* Nonzero if charset P has range table. */
833 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
835 /* Return the address of range table of charset P. But not the start
836 of table itself, but the before where the number of ranges is
837 stored. `2 +' means to skip re_opcode_t and size of bitmap,
838 and the 2 bytes of flags at the start of the range table. */
839 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
841 /* Extract the bit flags that start a range table. */
842 #define CHARSET_RANGE_TABLE_BITS(p) \
843 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
844 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
846 /* Test if C is listed in the bitmap of charset P. */
847 #define CHARSET_LOOKUP_BITMAP(p, c) \
848 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
849 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
851 /* Return the address of end of RANGE_TABLE. COUNT is number of
852 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
853 is start of range and end of range. `* 3' is size of each start
854 and end. */
855 #define CHARSET_RANGE_TABLE_END(range_table, count) \
856 ((range_table) + (count) * 2 * 3)
858 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
859 COUNT is number of ranges in RANGE_TABLE. */
860 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
861 do \
862 { \
863 re_wchar_t range_start, range_end; \
864 re_char *p; \
865 re_char *range_table_end \
866 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
867 \
868 for (p = (range_table); p < range_table_end; p += 2 * 3) \
869 { \
870 EXTRACT_CHARACTER (range_start, p); \
871 EXTRACT_CHARACTER (range_end, p + 3); \
872 \
873 if (range_start <= (c) && (c) <= range_end) \
874 { \
875 (not) = !(not); \
876 break; \
877 } \
878 } \
879 } \
880 while (0)
882 /* Test if C is in range table of CHARSET. The flag NOT is negated if
883 C is listed in it. */
884 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
885 do \
886 { \
888 int count; \
889 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
890 \
891 EXTRACT_NUMBER_AND_INCR (count, range_table); \
892 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
893 } \
894 while (0)
895 \f
896 /* If DEBUG is defined, Regex prints many voluminous messages about what
897 it is doing (if the variable `debug' is nonzero). If linked with the
898 main program in `iregex.c', you can enter patterns and strings
899 interactively. And if linked with the main program in `main.c' and
900 the other test files, you can run the already-written tests. */
902 #ifdef DEBUG
904 /* We use standard I/O for debugging. */
905 # include <stdio.h>
907 /* It is useful to test things that ``must'' be true when debugging. */
908 # include <assert.h>
912 # define DEBUG_STATEMENT(e) e
913 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
914 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
915 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
916 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
917 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
918 if (debug > 0) print_partial_compiled_pattern (s, e)
919 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
920 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
923 /* Print the fastmap in human-readable form. */
925 void
928 {
933 {
935 {
939 {
941 i++;
942 }
944 {
947 }
948 }
949 }
951 }
954 /* Print a compiled pattern string in human-readable form, starting at
955 the START pointer into it and ending just before the pointer END. */
957 void
961 {
967 {
970 }
972 /* Loop over pattern commands. */
974 {
978 {
990 do
991 {
993 }
1015 {
1030 {
1031 /* Are we starting a range? */
1033 {
1036 }
1037 /* Have we broken a range? */
1039 {
1042 }
1048 }
1058 {
1062 /* ??? Should print the range table; for now, just skip it. */
1066 }
1067 }
1162 # ifdef emacs
1198 }
1201 }
1204 }
1207 void
1210 {
1218 {
1221 }
1231 /* Perhaps we should print the translate table? */
1232 }
1235 void
1242 {
1247 else
1248 {
1250 {
1255 }
1259 }
1260 }
1264 # undef assert
1265 # define assert(e)
1267 # define DEBUG_STATEMENT(e)
1268 # define DEBUG_PRINT1(x)
1269 # define DEBUG_PRINT2(x1, x2)
1270 # define DEBUG_PRINT3(x1, x2, x3)
1271 # define DEBUG_PRINT4(x1, x2, x3, x4)
1272 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1273 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1276 \f
1277 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1278 also be assigned to arbitrarily: each pattern buffer stores its own
1279 syntax, so it can be changed between regex compilations. */
1280 /* This has no initializer because initialized variables in Emacs
1281 become read-only after dumping. */
1282 reg_syntax_t re_syntax_options;
1285 /* Specify the precise syntax of regexps for compilation. This provides
1286 for compatibility for various utilities which historically have
1287 different, incompatible syntaxes.
1289 The argument SYNTAX is a bit mask comprised of the various bits
1290 defined in regex.h. We return the old syntax. */
1292 reg_syntax_t
1294 {
1299 }
1302 /* Regexp to use to replace spaces, or NULL meaning don't. */
1305 void
1307 {
1309 }
1311 \f
1312 /* This table gives an error message for each of the error codes listed
1313 in regex.h. Obviously the order here has to be same as there.
1314 POSIX doesn't require that we do anything for REG_NOERROR,
1315 but why not be nice? */
1318 {
1337 };
1338 \f
1339 /* Avoiding alloca during matching, to placate r_alloc. */
1341 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1342 searching and matching functions should not call alloca. On some
1343 systems, alloca is implemented in terms of malloc, and if we're
1344 using the relocating allocator routines, then malloc could cause a
1345 relocation, which might (if the strings being searched are in the
1346 ralloc heap) shift the data out from underneath the regexp
1347 routines.
1349 Here's another reason to avoid allocation: Emacs
1350 processes input from X in a signal handler; processing X input may
1351 call malloc; if input arrives while a matching routine is calling
1352 malloc, then we're scrod. But Emacs can't just block input while
1353 calling matching routines; then we don't notice interrupts when
1354 they come in. So, Emacs blocks input around all regexp calls
1355 except the matching calls, which it leaves unprotected, in the
1356 faith that they will not malloc. */
1358 /* Normally, this is fine. */
1359 #define MATCH_MAY_ALLOCATE
1361 /* The match routines may not allocate if (1) they would do it with malloc
1362 and (2) it's not safe for them to use malloc.
1363 Note that if REL_ALLOC is defined, matching would not use malloc for the
1364 failure stack, but we would still use it for the register vectors;
1365 so REL_ALLOC should not affect this. */
1366 #if defined REGEX_MALLOC && defined emacs
1367 # undef MATCH_MAY_ALLOCATE
1368 #endif
1370 \f
1371 /* Failure stack declarations and macros; both re_compile_fastmap and
1372 re_match_2 use a failure stack. These have to be macros because of
1373 REGEX_ALLOCATE_STACK. */
1376 /* Approximate number of failure points for which to initially allocate space
1377 when matching. If this number is exceeded, we allocate more
1378 space, so it is not a hard limit. */
1379 #ifndef INIT_FAILURE_ALLOC
1380 # define INIT_FAILURE_ALLOC 20
1381 #endif
1383 /* Roughly the maximum number of failure points on the stack. Would be
1384 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1385 This is a variable only so users of regex can assign to it; we never
1386 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1387 before using it, so it should probably be a byte-count instead. */
1388 # if defined MATCH_MAY_ALLOCATE
1389 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1390 whose default stack limit is 2mb. In order for a larger
1391 value to work reliably, you have to try to make it accord
1392 with the process stack limit. */
1394 # else
1396 # endif
1398 union fail_stack_elt
1399 {
1401 /* This should be the biggest `int' that's no bigger than a pointer. */
1403 };
1408 {
1415 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1416 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1419 /* Define macros to initialize and free the failure stack.
1420 Do `return -2' if the alloc fails. */
1422 #ifdef MATCH_MAY_ALLOCATE
1423 # define INIT_FAIL_STACK() \
1424 do { \
1425 fail_stack.stack = (fail_stack_elt_t *) \
1426 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1427 * sizeof (fail_stack_elt_t)); \
1428 \
1429 if (fail_stack.stack == NULL) \
1430 return -2; \
1431 \
1432 fail_stack.size = INIT_FAILURE_ALLOC; \
1433 fail_stack.avail = 0; \
1434 fail_stack.frame = 0; \
1435 } while (0)
1437 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1438 #else
1439 # define INIT_FAIL_STACK() \
1440 do { \
1441 fail_stack.avail = 0; \
1442 fail_stack.frame = 0; \
1443 } while (0)
1445 # define RESET_FAIL_STACK() ((void)0)
1446 #endif
1449 /* Double the size of FAIL_STACK, up to a limit
1450 which allows approximately `re_max_failures' items.
1452 Return 1 if succeeds, and 0 if either ran out of memory
1453 allocating space for it or it was already too large.
1455 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1457 /* Factor to increase the failure stack size by
1458 when we increase it.
1459 This used to be 2, but 2 was too wasteful
1460 because the old discarded stacks added up to as much space
1461 were as ultimate, maximum-size stack. */
1462 #define FAIL_STACK_GROWTH_FACTOR 4
1464 #define GROW_FAIL_STACK(fail_stack) \
1465 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1466 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1467 ? 0 \
1468 : ((fail_stack).stack \
1469 = (fail_stack_elt_t *) \
1470 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1471 (fail_stack).size * sizeof (fail_stack_elt_t), \
1472 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1473 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1474 * FAIL_STACK_GROWTH_FACTOR))), \
1475 \
1476 (fail_stack).stack == NULL \
1477 ? 0 \
1478 : ((fail_stack).size \
1479 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1480 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1481 * FAIL_STACK_GROWTH_FACTOR)) \
1482 / sizeof (fail_stack_elt_t)), \
1483 1)))
1486 /* Push a pointer value onto the failure stack.
1487 Assumes the variable `fail_stack'. Probably should only
1488 be called from within `PUSH_FAILURE_POINT'. */
1489 #define PUSH_FAILURE_POINTER(item) \
1490 fail_stack.stack[fail_stack.avail++].pointer = (item)
1492 /* This pushes an integer-valued item onto the failure stack.
1493 Assumes the variable `fail_stack'. Probably should only
1494 be called from within `PUSH_FAILURE_POINT'. */
1495 #define PUSH_FAILURE_INT(item) \
1496 fail_stack.stack[fail_stack.avail++].integer = (item)
1498 /* Push a fail_stack_elt_t value onto the failure stack.
1499 Assumes the variable `fail_stack'. Probably should only
1500 be called from within `PUSH_FAILURE_POINT'. */
1501 #define PUSH_FAILURE_ELT(item) \
1502 fail_stack.stack[fail_stack.avail++] = (item)
1504 /* These three POP... operations complement the three PUSH... operations.
1505 All assume that `fail_stack' is nonempty. */
1506 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1507 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1508 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1510 /* Individual items aside from the registers. */
1511 #define NUM_NONREG_ITEMS 3
1513 /* Used to examine the stack (to detect infinite loops). */
1514 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1515 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1516 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1517 #define TOP_FAILURE_HANDLE() fail_stack.frame
1520 #define ENSURE_FAIL_STACK(space) \
1521 while (REMAINING_AVAIL_SLOTS <= space) { \
1522 if (!GROW_FAIL_STACK (fail_stack)) \
1523 return -2; \
1526 }
1528 /* Push register NUM onto the stack. */
1529 #define PUSH_FAILURE_REG(num) \
1530 do { \
1531 char *destination; \
1532 ENSURE_FAIL_STACK(3); \
1534 num, regstart[num], regend[num]); \
1535 PUSH_FAILURE_POINTER (regstart[num]); \
1536 PUSH_FAILURE_POINTER (regend[num]); \
1537 PUSH_FAILURE_INT (num); \
1538 } while (0)
1540 /* Change the counter's value to VAL, but make sure that it will
1541 be reset when backtracking. */
1542 #define PUSH_NUMBER(ptr,val) \
1543 do { \
1544 char *destination; \
1545 int c; \
1546 ENSURE_FAIL_STACK(3); \
1547 EXTRACT_NUMBER (c, ptr); \
1549 PUSH_FAILURE_INT (c); \
1550 PUSH_FAILURE_POINTER (ptr); \
1551 PUSH_FAILURE_INT (-1); \
1552 STORE_NUMBER (ptr, val); \
1553 } while (0)
1555 /* Pop a saved register off the stack. */
1556 #define POP_FAILURE_REG_OR_COUNT() \
1557 do { \
1558 int reg = POP_FAILURE_INT (); \
1559 if (reg == -1) \
1560 { \
1563 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1564 reg = POP_FAILURE_INT (); \
1565 STORE_NUMBER (ptr, reg); \
1567 } \
1568 else \
1569 { \
1570 regend[reg] = POP_FAILURE_POINTER (); \
1571 regstart[reg] = POP_FAILURE_POINTER (); \
1573 reg, regstart[reg], regend[reg]); \
1574 } \
1575 } while (0)
1577 /* Check that we are not stuck in an infinite loop. */
1578 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1579 do { \
1580 int failure = TOP_FAILURE_HANDLE (); \
1582 while (failure > 0 \
1583 && (FAILURE_STR (failure) == string_place \
1584 || FAILURE_STR (failure) == NULL)) \
1585 { \
1586 assert (FAILURE_PAT (failure) >= bufp->buffer \
1587 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1588 if (FAILURE_PAT (failure) == pat_cur) \
1589 { \
1590 cycle = 1; \
1591 break; \
1592 } \
1594 failure = NEXT_FAILURE_HANDLE(failure); \
1595 } \
1597 } while (0)
1599 /* Push the information about the state we will need
1600 if we ever fail back to it.
1602 Requires variables fail_stack, regstart, regend and
1603 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1604 declared.
1606 Does `return FAILURE_CODE' if runs out of memory. */
1608 #define PUSH_FAILURE_POINT(pattern, string_place) \
1609 do { \
1610 char *destination; \
1611 /* Must be int, so when we don't save any registers, the arithmetic \
1613 \
1614 DEBUG_STATEMENT (nfailure_points_pushed++); \
1618 \
1619 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1620 \
1622 \
1624 PUSH_FAILURE_INT (fail_stack.frame); \
1625 \
1627 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1629 PUSH_FAILURE_POINTER (string_place); \
1630 \
1632 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1633 PUSH_FAILURE_POINTER (pattern); \
1634 \
1636 fail_stack.frame = fail_stack.avail; \
1637 } while (0)
1639 /* Estimate the size of data pushed by a typical failure stack entry.
1640 An estimate is all we need, because all we use this for
1641 is to choose a limit for how big to make the failure stack. */
1642 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1643 #define TYPICAL_FAILURE_SIZE 20
1645 /* How many items can still be added to the stack without overflowing it. */
1646 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1649 /* Pops what PUSH_FAIL_STACK pushes.
1651 We restore into the parameters, all of which should be lvalues:
1652 STR -- the saved data position.
1653 PAT -- the saved pattern position.
1654 REGSTART, REGEND -- arrays of string positions.
1656 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1657 `pend', `string1', `size1', `string2', and `size2'. */
1659 #define POP_FAILURE_POINT(str, pat) \
1660 do { \
1661 assert (!FAIL_STACK_EMPTY ()); \
1662 \
1667 \
1669 while (fail_stack.frame < fail_stack.avail) \
1670 POP_FAILURE_REG_OR_COUNT (); \
1671 \
1672 pat = POP_FAILURE_POINTER (); \
1674 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1675 \
1676 /* If the saved string location is NULL, it came from an \
1677 on_failure_keep_string_jump opcode, and we want to throw away the \
1679 str = POP_FAILURE_POINTER (); \
1681 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1683 \
1684 fail_stack.frame = POP_FAILURE_INT (); \
1686 \
1687 assert (fail_stack.avail >= 0); \
1688 assert (fail_stack.frame <= fail_stack.avail); \
1689 \
1690 DEBUG_STATEMENT (nfailure_points_popped++); \
1694 \f
1695 /* Registers are set to a sentinel when they haven't yet matched. */
1696 #define REG_UNSET(e) ((e) == NULL)
1697 \f
1698 /* Subroutine declarations and macros for regex_compile. */
1701 reg_syntax_t syntax,
1712 reg_syntax_t syntax));
1715 reg_syntax_t syntax));
1720 /* Fetch the next character in the uncompiled pattern, with no
1721 translation. */
1722 #define PATFETCH(c) \
1723 do { \
1724 int len; \
1725 if (p == pend) return REG_EEND; \
1726 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1727 p += len; \
1728 } while (0)
1731 /* If `translate' is non-null, return translate[D], else just D. We
1732 cast the subscript to translate because some data is declared as
1733 `char *', to avoid warnings when a string constant is passed. But
1734 when we use a character as a subscript we must make it unsigned. */
1735 #ifndef TRANSLATE
1736 # define TRANSLATE(d) \
1737 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1738 #endif
1741 /* Macros for outputting the compiled pattern into `buffer'. */
1743 /* If the buffer isn't allocated when it comes in, use this. */
1744 #define INIT_BUF_SIZE 32
1746 /* Make sure we have at least N more bytes of space in buffer. */
1747 #define GET_BUFFER_SPACE(n) \
1748 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1749 EXTEND_BUFFER ()
1751 /* Make sure we have one more byte of buffer space and then add C to it. */
1752 #define BUF_PUSH(c) \
1753 do { \
1754 GET_BUFFER_SPACE (1); \
1755 *b++ = (unsigned char) (c); \
1756 } while (0)
1759 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1760 #define BUF_PUSH_2(c1, c2) \
1761 do { \
1762 GET_BUFFER_SPACE (2); \
1763 *b++ = (unsigned char) (c1); \
1764 *b++ = (unsigned char) (c2); \
1765 } while (0)
1768 /* As with BUF_PUSH_2, except for three bytes. */
1769 #define BUF_PUSH_3(c1, c2, c3) \
1770 do { \
1771 GET_BUFFER_SPACE (3); \
1772 *b++ = (unsigned char) (c1); \
1773 *b++ = (unsigned char) (c2); \
1774 *b++ = (unsigned char) (c3); \
1775 } while (0)
1778 /* Store a jump with opcode OP at LOC to location TO. We store a
1779 relative address offset by the three bytes the jump itself occupies. */
1780 #define STORE_JUMP(op, loc, to) \
1781 store_op1 (op, loc, (to) - (loc) - 3)
1783 /* Likewise, for a two-argument jump. */
1784 #define STORE_JUMP2(op, loc, to, arg) \
1785 store_op2 (op, loc, (to) - (loc) - 3, arg)
1787 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1788 #define INSERT_JUMP(op, loc, to) \
1789 insert_op1 (op, loc, (to) - (loc) - 3, b)
1791 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1792 #define INSERT_JUMP2(op, loc, to, arg) \
1793 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1796 /* This is not an arbitrary limit: the arguments which represent offsets
1797 into the pattern are two bytes long. So if 2^15 bytes turns out to
1798 be too small, many things would have to change. */
1799 # define MAX_BUF_SIZE (1L << 15)
1802 /* Any other compiler which, like MSC, has allocation limit below 2^16
1803 bytes will have to use approach similar to what was done below for
1804 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1805 reallocating to 0 bytes. Such thing is not going to work too well.
1806 You have been warned!! */
1807 #if defined _MSC_VER && !defined WIN32
1808 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1809 # define MAX_BUF_SIZE 65500L
1810 #else
1811 # define MAX_BUF_SIZE (1L << 16)
1812 #endif
1815 /* Extend the buffer by twice its current size via realloc and
1816 reset the pointers that pointed into the old block to point to the
1817 correct places in the new one. If extending the buffer results in it
1818 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1819 #if __BOUNDED_POINTERS__
1820 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1821 # define MOVE_BUFFER_POINTER(P) \
1822 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1823 SET_HIGH_BOUND (P), \
1824 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1825 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1826 else \
1827 { \
1828 SET_HIGH_BOUND (b); \
1829 SET_HIGH_BOUND (begalt); \
1830 if (fixup_alt_jump) \
1831 SET_HIGH_BOUND (fixup_alt_jump); \
1832 if (laststart) \
1833 SET_HIGH_BOUND (laststart); \
1834 if (pending_exact) \
1835 SET_HIGH_BOUND (pending_exact); \
1836 }
1837 #else
1838 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1839 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1840 #endif
1841 #define EXTEND_BUFFER() \
1842 do { \
1843 unsigned char *old_buffer = bufp->buffer; \
1844 if (bufp->allocated == MAX_BUF_SIZE) \
1845 return REG_ESIZE; \
1846 bufp->allocated <<= 1; \
1847 if (bufp->allocated > MAX_BUF_SIZE) \
1848 bufp->allocated = MAX_BUF_SIZE; \
1849 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1850 if (bufp->buffer == NULL) \
1851 return REG_ESPACE; \
1853 if (old_buffer != bufp->buffer) \
1854 { \
1855 unsigned char *new_buffer = bufp->buffer; \
1856 MOVE_BUFFER_POINTER (b); \
1857 MOVE_BUFFER_POINTER (begalt); \
1858 if (fixup_alt_jump) \
1859 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1860 if (laststart) \
1861 MOVE_BUFFER_POINTER (laststart); \
1862 if (pending_exact) \
1863 MOVE_BUFFER_POINTER (pending_exact); \
1864 } \
1865 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1866 } while (0)
1869 /* Since we have one byte reserved for the register number argument to
1870 {start,stop}_memory, the maximum number of groups we can report
1871 things about is what fits in that byte. */
1872 #define MAX_REGNUM 255
1874 /* But patterns can have more than `MAX_REGNUM' registers. We just
1875 ignore the excess. */
1879 /* Macros for the compile stack. */
1881 /* Since offsets can go either forwards or backwards, this type needs to
1882 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1883 /* int may be not enough when sizeof(int) == 2. */
1887 {
1888 pattern_offset_t begalt_offset;
1889 pattern_offset_t fixup_alt_jump;
1890 pattern_offset_t laststart_offset;
1891 regnum_t regnum;
1896 {
1903 #define INIT_COMPILE_STACK_SIZE 32
1905 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1906 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1908 /* The next available element. */
1909 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1911 /* Explicit quit checking is only used on NTemacs and whenever we
1912 use polling to process input events. */
1913 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1915 # define IMMEDIATE_QUIT_CHECK \
1916 do { \
1917 if (immediate_quit) QUIT; \
1918 } while (0)
1919 #else
1920 # define IMMEDIATE_QUIT_CHECK ((void)0)
1921 #endif
1922 \f
1923 /* Structure to manage work area for range table. */
1924 struct range_table_work_area
1925 {
1930 };
1932 /* Make sure that WORK_AREA can hold more N multibyte characters.
1933 This is used only in set_image_of_range and set_image_of_range_1.
1934 It expects WORK_AREA to be a pointer.
1935 If it can't get the space, it returns from the surrounding function. */
1937 #define EXTEND_RANGE_TABLE(work_area, n) \
1938 do { \
1939 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1940 { \
1941 extend_range_table_work_area (&work_area); \
1942 if ((work_area).table == 0) \
1943 return (REG_ESPACE); \
1944 } \
1945 } while (0)
1947 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1948 (work_area).bits |= (bit)
1950 /* Bits used to implement the multibyte-part of the various character classes
1951 such as [:alnum:] in a charset's range table. */
1952 #define BIT_WORD 0x1
1953 #define BIT_LOWER 0x2
1954 #define BIT_PUNCT 0x4
1955 #define BIT_SPACE 0x8
1956 #define BIT_UPPER 0x10
1957 #define BIT_MULTIBYTE 0x20
1959 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1960 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1961 do { \
1962 EXTEND_RANGE_TABLE ((work_area), 2); \
1963 (work_area).table[(work_area).used++] = (range_start); \
1964 (work_area).table[(work_area).used++] = (range_end); \
1965 } while (0)
1967 /* Free allocated memory for WORK_AREA. */
1968 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1969 do { \
1970 if ((work_area).table) \
1971 free ((work_area).table); \
1972 } while (0)
1974 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1975 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1976 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1977 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1978 \f
1980 /* Set the bit for character C in a list. */
1981 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1984 #ifdef emacs
1986 /* Store characters in the range FROM to TO in the bitmap at B (for
1987 ASCII and unibyte characters) and WORK_AREA (for multibyte
1988 characters) while translating them and paying attention to the
1989 continuity of translated characters.
1991 Implementation note: It is better to implement these fairly big
1992 macros by a function, but it's not that easy because macros called
1993 in this macro assume various local variables already declared. */
1995 /* Both FROM and TO are ASCII characters. */
1997 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1998 do { \
1999 int C0, C1; \
2000 \
2001 for (C0 = (FROM); C0 <= (TO); C0++) \
2002 { \
2003 C1 = TRANSLATE (C0); \
2004 if (! ASCII_CHAR_P (C1)) \
2005 { \
2006 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2007 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2008 C1 = C0; \
2009 } \
2010 SET_LIST_BIT (C1); \
2011 } \
2012 } while (0)
2015 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2017 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2018 do { \
2019 int C0, C1, C2, I; \
2020 int USED = RANGE_TABLE_WORK_USED (work_area); \
2021 \
2022 for (C0 = (FROM); C0 <= (TO); C0++) \
2023 { \
2024 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2025 if (CHAR_BYTE8_P (C1)) \
2026 SET_LIST_BIT (C0); \
2027 else \
2028 { \
2029 C2 = TRANSLATE (C1); \
2030 if (C2 == C1 \
2031 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2032 C1 = C0; \
2033 SET_LIST_BIT (C1); \
2034 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2035 { \
2036 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2037 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2038 \
2039 if (C2 >= from - 1 && C2 <= to + 1) \
2040 { \
2041 if (C2 == from - 1) \
2042 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2043 else if (C2 == to + 1) \
2044 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2045 break; \
2046 } \
2047 } \
2048 if (I < USED) \
2049 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2050 } \
2051 } \
2052 } while (0)
2055 /* Both FROM and TO are multibyte characters. */
2057 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2058 do { \
2059 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2060 \
2061 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2062 for (C0 = (FROM); C0 <= (TO); C0++) \
2063 { \
2064 C1 = TRANSLATE (C0); \
2065 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2066 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2067 SET_LIST_BIT (C2); \
2068 if (C1 >= (FROM) && C1 <= (TO)) \
2069 continue; \
2070 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2071 { \
2072 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2073 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2074 \
2075 if (C1 >= from - 1 && C1 <= to + 1) \
2076 { \
2077 if (C1 == from - 1) \
2078 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2079 else if (C1 == to + 1) \
2080 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2081 break; \
2082 } \
2083 } \
2084 if (I < USED) \
2085 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2086 } \
2087 } while (0)
2091 /* Get the next unsigned number in the uncompiled pattern. */
2092 #define GET_UNSIGNED_NUMBER(num) \
2093 do { \
2094 if (p == pend) \
2095 FREE_STACK_RETURN (REG_EBRACE); \
2096 else \
2097 { \
2098 PATFETCH (c); \
2100 { \
2101 int prev; \
2102 if (num < 0) \
2103 num = 0; \
2104 prev = num; \
2106 if (num / 10 != prev) \
2107 FREE_STACK_RETURN (REG_BADBR); \
2108 if (p == pend) \
2109 FREE_STACK_RETURN (REG_EBRACE); \
2110 PATFETCH (c); \
2111 } \
2112 } \
2113 } while (0)
2114 \f
2115 #if ! WIDE_CHAR_SUPPORT
2117 /* Map a string to the char class it names (if any). */
2118 re_wctype_t
2120 {
2140 }
2142 /* True if CH is in the char class CC. */
2143 boolean
2145 {
2147 {
2168 }
2169 }
2171 /* Return a bit-pattern to use in the range-table bits to match multibyte
2172 chars of class CC. */
2173 static int
2175 {
2177 {
2189 }
2190 }
2191 #endif
2192 \f
2193 /* Filling in the work area of a range. */
2195 /* Actually extend the space in WORK_AREA. */
2197 static void
2199 {
2202 work_area->table
2204 else
2205 work_area->table
2207 }
2209 #if 0
2210 #ifdef emacs
2212 /* Carefully find the ranges of codes that are equivalent
2213 under case conversion to the range start..end when passed through
2214 TRANSLATE. Handle the case where non-letters can come in between
2215 two upper-case letters (which happens in Latin-1).
2216 Also handle the case of groups of more than 2 case-equivalent chars.
2218 The basic method is to look at consecutive characters and see
2219 if they can form a run that can be handled as one.
2221 Returns -1 if successful, REG_ESPACE if ran out of space. */
2223 static int
2225 RE_TRANSLATE_TYPE translate;
2228 {
2229 /* `one_case' indicates a character, or a run of characters,
2230 each of which is an isolate (no case-equivalents).
2231 This includes all ASCII non-letters.
2233 `two_case' indicates a character, or a run of characters,
2234 each of which has two case-equivalent forms.
2235 This includes all ASCII letters.
2237 `strange' indicates a character that has more than one
2238 case-equivalent. */
2242 /* Describe the run that is in progress,
2243 which the next character can try to extend.
2244 If run_type is strange, that means there really is no run.
2245 If run_type is one_case, then run_start...run_end is the run.
2246 If run_type is two_case, then the run is run_start...run_end,
2247 and the case-equivalents end at run_eqv_end. */
2252 Lisp_Object eqv_table;
2255 {
2260 }
2265 {
2270 /* Classify this character */
2275 else
2280 else
2283 /* Can this character extend the run in progress? */
2288 {
2289 /* No, end the run.
2290 Record each of its equivalent ranges. */
2292 {
2296 }
2298 {
2306 }
2308 }
2311 {
2312 /* For a strange character, add each of its equivalents, one
2313 by one. Don't start a range. */
2314 do
2315 {
2320 }
2322 }
2324 /* Add this char to the run, or start a new run. */
2326 {
2327 /* Initialize a new range. */
2332 }
2333 else
2334 {
2335 /* Extend a running range. */
2338 }
2339 }
2341 /* If a run is still in progress at the end, finish it now
2342 by recording its equivalent ranges. */
2344 {
2348 }
2350 {
2358 }
2361 }
2365 /* Record the image of the range start..end when passed through
2366 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2367 and is not even necessarily contiguous.
2368 Normally we approximate it with the smallest contiguous range that contains
2369 all the chars we need. However, for Latin-1 we go to extra effort
2370 to do a better job.
2372 This function is not called for ASCII ranges.
2374 Returns -1 if successful, REG_ESPACE if ran out of space. */
2376 static int
2378 RE_TRANSLATE_TYPE translate;
2381 {
2384 #ifdef emacs
2385 /* For Latin-1 ranges, use set_image_of_range_1
2386 to get proper handling of ranges that include letters and nonletters.
2387 For a range that includes the whole of Latin-1, this is not necessary.
2388 For other character sets, we don't bother to get this right. */
2391 {
2404 }
2405 #endif
2414 {
2418 {
2421 {
2424 else
2425 {
2428 }
2429 }
2430 }
2433 {
2437 }
2438 }
2441 }
2443 \f
2444 #ifndef MATCH_MAY_ALLOCATE
2446 /* If we cannot allocate large objects within re_match_2_internal,
2447 we make the fail stack and register vectors global.
2448 The fail stack, we grow to the maximum size when a regexp
2449 is compiled.
2450 The register vectors, we adjust in size each time we
2451 compile a regexp, according to the number of registers it needs. */
2455 /* Size with which the following vectors are currently allocated.
2456 That is so we can make them bigger as needed,
2457 but never make them smaller. */
2463 /* Make the register vectors big enough for NUM_REGS registers,
2464 but don't make them smaller. */
2466 static
2469 {
2471 {
2478 }
2479 }
2482 \f
2484 compile_stack,
2485 regnum_t regnum));
2487 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2488 Returns one of error codes defined in `regex.h', or zero for success.
2490 Assumes the `allocated' (and perhaps `buffer') and `translate'
2491 fields are set in BUFP on entry.
2493 If it succeeds, results are put in BUFP (if it returns an error, the
2494 contents of BUFP are undefined):
2495 `buffer' is the compiled pattern;
2496 `syntax' is set to SYNTAX;
2497 `used' is set to the length of the compiled pattern;
2498 `fastmap_accurate' is zero;
2499 `re_nsub' is the number of subexpressions in PATTERN;
2500 `not_bol' and `not_eol' are zero;
2502 The `fastmap' field is neither examined nor set. */
2504 /* Insert the `jump' from the end of last alternative to "here".
2505 The space for the jump has already been allocated. */
2506 #define FIXUP_ALT_JUMP() \
2507 do { \
2508 if (fixup_alt_jump) \
2509 STORE_JUMP (jump, fixup_alt_jump, b); \
2510 } while (0)
2513 /* Return, freeing storage we allocated. */
2514 #define FREE_STACK_RETURN(value) \
2515 do { \
2516 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2517 free (compile_stack.stack); \
2518 return value; \
2519 } while (0)
2521 static reg_errcode_t
2522 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2523 {
2524 /* We fetch characters from PATTERN here. */
2527 /* A random temporary spot in PATTERN. */
2530 /* Points to the end of the buffer, where we should append. */
2533 /* Keeps track of unclosed groups. */
2534 compile_stack_type compile_stack;
2536 /* Points to the current (ending) position in the pattern. */
2537 #ifdef AIX
2538 /* `const' makes AIX compiler fail. */
2540 #else
2542 #endif
2545 /* How to translate the characters in the pattern. */
2548 /* Address of the count-byte of the most recently inserted `exactn'
2549 command. This makes it possible to tell if a new exact-match
2550 character can be added to that command or if the character requires
2551 a new `exactn' command. */
2554 /* Address of start of the most recently finished expression.
2555 This tells, e.g., postfix * where to find the start of its
2556 operand. Reset at the beginning of groups and alternatives. */
2559 /* Address of beginning of regexp, or inside of last group. */
2562 /* Place in the uncompiled pattern (i.e., the {) to
2563 which to go back if the interval is invalid. */
2566 /* Address of the place where a forward jump should go to the end of
2567 the containing expression. Each alternative of an `or' -- except the
2568 last -- ends with a forward jump of this sort. */
2571 /* Work area for range table of charset. */
2574 /* If the object matched can contain multibyte characters. */
2577 /* If a target of matching can contain multibyte characters. */
2580 /* Nonzero if we have pushed down into a subpattern. */
2583 /* These hold the values of p, pattern, and pend from the main
2584 pattern when we have pushed into a subpattern. */
2589 #ifdef DEBUG
2590 debug++;
2593 {
2599 }
2602 /* Initialize the compile stack. */
2613 /* Initialize the pattern buffer. */
2619 /* Set `used' to zero, so that if we return an error, the pattern
2620 printer (for debugging) will think there's no pattern. We reset it
2621 at the end. */
2624 /* Always count groups, whether or not bufp->no_sub is set. */
2627 #if !defined emacs && !defined SYNTAX_TABLE
2628 /* Initialize the syntax table. */
2630 #endif
2633 {
2636 enough space. This loses if buffer's address is bogus, but
2637 that is the user's responsibility. */
2639 }
2640 else
2643 }
2647 }
2651 /* Loop through the uncompiled pattern until we're at the end. */
2653 {
2655 {
2656 /* If this is the end of an included regexp,
2657 pop back to the main regexp and try again. */
2659 {
2665 }
2666 /* If this is the end of the main regexp, we are done. */
2668 }
2673 {
2675 {
2678 /* If there's no special whitespace regexp, treat
2679 spaces normally. And don't try to do this recursively. */
2683 /* Peek past following spaces. */
2685 {
2688 p1++;
2689 }
2690 /* If the spaces are followed by a repetition op,
2691 treat them normally. */
2697 /* Replace the spaces with the whitespace regexp. */
2705 }
2708 {
2711 /* If context independent, it's an operator. */
2713 /* Otherwise, depends on what's come before. */
2716 else
2718 }
2723 {
2725 p == pend
2726 /* If context independent, it's an operator. */
2728 /* Otherwise, depends on what's next. */
2731 else
2733 }
2742 handle_plus:
2744 /* If there is no previous pattern... */
2746 {
2751 }
2753 {
2754 /* 1 means zero (many) matches is allowed. */
2758 /* If there is a sequence of repetition chars, collapse it
2759 down to just one (the right one). We can't combine
2760 interval operators with these because of, e.g., `a{2}*',
2761 which should only match an even number of `a's. */
2764 {
2768 else
2769 {
2772 }
2779 ;
2781 {
2786 else
2788 }
2789 else
2791 /* If we get here, we found another repeat character. */
2793 }
2795 /* Star, etc. applied to an empty pattern is equivalent
2796 to an empty pattern. */
2800 /* Now we know whether or not zero matches is allowed
2801 and also whether or not two or more matches is allowed. */
2803 {
2805 {
2808 re_opcode_t ofj =
2809 /* Check if the loop can match the empty string. */
2816 on_failure_keep_string_jump, we turn simple P+
2817 into PP* if P is simple. */
2825 }
2829 /* A + loop. */
2831 else
2832 /* Simple * loops can use on_failure_keep_string_jump
2833 depending on what follows. But since we don't know
2834 that yet, we leave the decision up to
2835 on_failure_jump_smart. */
2841 }
2842 else
2843 {
2844 /* A simple ? pattern. */
2849 }
2850 }
2856 {
2859 /* The non-greedy multiple match looks like
2860 a repeat..until: we only need a conditional jump
2861 at the end of the loop. */
2867 {
2868 /* The repeat...until naturally matches one or more.
2869 To also match zero times, we need to first jump to
2870 the end of the loop (its conditional jump). */
2873 }
2874 }
2875 else
2876 {
2877 /* non-greedy a?? */
2882 }
2883 }
2884 }
2896 {
2901 /* Ensure that we have enough space to push a charset: the
2902 opcode, the length count, and the bitset; 34 bytes in all. */
2907 /* We test `*p == '^' twice, instead of using an if
2908 statement, so we only need one BUF_PUSH. */
2911 p++;
2913 /* Remember the first position in the bracket expression. */
2916 /* Push the number of bytes in the bitmap. */
2919 /* Clear the whole map. */
2922 /* charset_not matches newline according to a syntax bit. */
2927 /* Read in characters and ranges, setting map bits. */
2929 {
2936 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2937 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2938 So the translation is done later in a loop. Example:
2939 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2942 /* \ might escape characters inside [...] and [^...]. */
2944 {
2949 }
2950 else
2951 {
2952 /* Could be the end of the bracket expression. If it's
2953 not (i.e., when the bracket expression is `[]' so
2954 far), the ']' character bit gets set way below. */
2957 }
2959 /* See if we're at the beginning of a possible character
2960 class. */
2964 {
2965 /* Leave room for the null. */
2973 /* If pattern is `[[:'. */
2977 {
2983 else
2984 /* This is in any case an invalid class name. */
2986 }
2989 /* If isn't a word bracketed by `[:' and `:]':
2990 undo the ending character, the letters, and
2991 leave the leading `:' and `[' (but set bits for
2992 them). */
2994 {
2995 re_wctype_t cc;
3003 /* Throw away the ] at the end of the character
3004 class. */
3009 #ifndef emacs
3012 {
3016 }
3018 /* Most character classes in a multibyte match
3019 just set a flag. Exceptions are is_blank,
3020 is_digit, is_cntrl, and is_xdigit, since
3021 they can only match ASCII characters. We
3022 don't need to handle them for multibyte.
3023 They are distinguished by a negative wctype. */
3025 /* Setup the gl_state object to its buffer-defined
3026 value. This hardcodes the buffer-global
3027 syntax-table for ASCII chars, while the other chars
3028 will obey syntax-table properties. It's not ideal,
3029 but it's the way it's been done until now. */
3033 {
3037 {
3046 }
3047 }
3048 SET_RANGE_TABLE_WORK_AREA_BIT
3051 /* In most cases the matching rule for char classes
3052 only uses the syntax table for multibyte chars,
3053 so that the content of the syntax-table it is not
3054 hardcoded in the range_table. SPACE and WORD are
3055 the two exceptions. */
3059 /* Repeat the loop. */
3061 }
3062 else
3063 {
3064 /* Go back to right after the "[:". */
3068 /* Because the `:' may starts the range, we
3069 can't simply set bit and repeat the loop.
3070 Instead, just set it to C and handle below. */
3072 }
3073 }
3076 {
3078 /* Discard the `-'. */
3081 /* Fetch the character which ends the range. */
3083 #ifdef emacs
3086 /* Treat the range from a multibyte character to
3087 raw-byte character as empty. */
3090 }
3091 else
3092 /* Range from C to C. */
3096 {
3099 /* Else, repeat the loop. */
3100 }
3101 else
3102 {
3103 #ifndef emacs
3104 /* Set the range into bitmap */
3106 {
3110 }
3113 {
3119 }
3121 {
3123 {
3128 }
3130 {
3132 }
3133 else
3134 {
3136 }
3137 }
3139 }
3140 }
3142 /* Discard any (non)matching list bytes that are all 0 at the
3143 end of the map. Decrease the map-length byte too. */
3148 /* Build real range table from work area. */
3151 {
3155 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3156 bytes for flags, two for COUNT, and three bytes for
3157 each character. */
3160 /* Indicate the existence of range table. */
3163 /* Store the character class flag bits into the range table.
3164 If not in emacs, these flag bits are always 0. */
3170 STORE_CHARACTER_AND_INCR
3172 }
3173 }
3180 else
3187 else
3194 else
3201 else
3208 else
3215 /* Do not translate the character after the \, so that we can
3216 distinguish, e.g., \B from \b, even if we normally would
3217 translate, e.g., B to b. */
3221 {
3226 handle_open:
3227 {
3231 {
3232 /* Look for a special (?...) construct */
3234 {
3237 {
3240 {
3243 /* An explicitly specified regnum must start
3244 with non-0. */
3251 /* Only (?:...) is supported right now. */
3253 }
3254 }
3255 }
3256 }
3266 /* Ideally, we'd want to check that the specified
3267 group can't have matched (i.e. all subgroups
3268 using the same regnum are in other branches of
3269 OR patterns), but we don't currently keep track
3270 of enough info to do that easily. */
3273 }
3274 else
3275 /* It's really shy. */
3279 {
3281 compile_stack_elt_t);
3285 }
3287 /* These are the values to restore when we hit end of this
3288 group. They are all relative offsets, so that if the
3289 whole pattern moves because of realloc, they will still
3290 be valid. */
3292 COMPILE_STACK_TOP.fixup_alt_jump
3297 /* Do not push a start_memory for groups beyond the last one
3298 we can represent in the compiled pattern. */
3307 /* If we've reached MAX_REGNUM groups, then this open
3308 won't actually generate any code, so we'll have to
3309 clear pending_exact explicitly. */
3312 }
3318 {
3321 else
3323 }
3325 handle_close:
3328 /* See similar code for backslashed left paren above. */
3330 {
3333 else
3335 }
3337 /* Since we just checked for an empty stack above, this
3338 ``can't happen''. */
3340 {
3341 /* We don't just want to restore into `regnum', because
3342 later groups should continue to be numbered higher,
3343 as in `(ab)c(de)' -- the second group is #2. */
3344 regnum_t regnum;
3348 fixup_alt_jump
3354 /* If we've reached MAX_REGNUM groups, then this open
3355 won't actually generate any code, so we'll have to
3356 clear pending_exact explicitly. */
3359 /* We're at the end of the group, so now we know how many
3360 groups were inside this one. */
3363 }
3370 handle_alt:
3374 /* Insert before the previous alternative a jump which
3375 jumps to this alternative if the former fails. */
3381 /* The alternative before this one has a jump after it
3382 which gets executed if it gets matched. Adjust that
3383 jump so it will jump to this alternative's analogous
3384 jump (put in below, which in turn will jump to the next
3385 (if any) alternative's such jump, etc.). The last such
3386 jump jumps to the correct final destination. A picture:
3387 _____ _____
3388 | | | |
3389 | v | v
3390 a | b | c
3392 If we are at `b', then fixup_alt_jump right now points to a
3393 three-byte space after `a'. We'll put in the jump, set
3394 fixup_alt_jump to right after `b', and leave behind three
3395 bytes which we'll fill in when we get to after `c'. */
3399 /* Mark and leave space for a jump after this alternative,
3400 to be filled in later either by next alternative or
3401 when know we're at the end of a series of alternatives. */
3412 /* If \{ is a literal. */
3414 /* If we're at `\{' and it's not the open-interval
3415 operator. */
3419 handle_interval:
3420 {
3421 /* If got here, then the syntax allows intervals. */
3423 /* At least (most) this many matches must be made. */
3432 else
3433 /* Interval such as `{1}' => match exactly once. */
3441 {
3447 }
3452 /* We just parsed a valid interval. */
3454 /* If it's invalid to have no preceding re. */
3456 {
3461 else
3463 }
3466 /* If the upper bound is zero, just drop the sub pattern
3467 altogether. */
3470 /* Just match it once: nothing to do here. */
3471 ;
3473 /* Otherwise, we have a nontrivial interval. When
3474 we're all done, the pattern will look like:
3475 set_number_at <jump count> <upper bound>
3476 set_number_at <succeed_n count> <lower bound>
3477 succeed_n <after jump addr> <succeed_n count>
3478 <body of loop>
3479 jump_n <succeed_n addr> <jump count>
3480 (The upper bound and `jump_n' are omitted if
3481 `upper_bound' is 1, though.) */
3482 else
3484 more at the end of the loop. */
3492 {
3493 /* A succeed_n that starts with 0 is really a
3494 a simple on_failure_jump_loop. */
3498 }
3499 else
3500 {
3501 /* Initialize lower bound of the `succeed_n', even
3502 though it will be set during matching by its
3503 attendant `set_number_at' (inserted next),
3504 because `re_compile_fastmap' needs to know.
3505 Jump to the `jump_n' we might insert below. */
3508 lower_bound);
3511 /* Code to initialize the lower bound. Insert
3512 before the `succeed_n'. The `5' is the last two
3513 bytes of this `set_number_at', plus 3 bytes of
3514 the following `succeed_n'. */
3518 }
3521 {
3522 /* A negative upper bound stands for infinity,
3523 in which case it degenerates to a plain jump. */
3526 }
3529 append a backward jump to the `succeed_n'
3530 that starts this interval.
3532 When we've reached this during matching,
3533 we'll have matched the interval once, so
3534 jump back only `upper_bound - 1' times. */
3539 /* The location we want to set is the second
3540 parameter of the `jump_n'; that is `b-2' as
3541 an absolute address. `laststart' will be
3542 the `set_number_at' we're about to insert;
3543 `laststart+3' the number to set, the source
3544 for the relative address. But we are
3545 inserting into the middle of the pattern --
3546 so everything is getting moved up by 5.
3547 Conclusion: (b - 2) - (laststart + 3) + 5,
3548 i.e., b - laststart.
3550 We insert this at the beginning of the loop
3551 so that if we fail during matching, we'll
3552 reinitialize the bounds. */
3556 }
3557 }
3560 }
3563 unfetch_interval:
3564 /* If an invalid interval, match the characters as literals. */
3569 /* normal_char and normal_backslash need `c'. */
3573 {
3576 }
3577 else
3580 #ifdef emacs
3581 /* There is no way to specify the before_dot and after_dot
3582 operators. rms says this is ok. --karl */
3650 else
3680 {
3681 regnum_t reg;
3689 /* Can't back reference to a subexp before its end. */
3695 }
3703 else
3707 normal_backslash:
3708 /* You might think it would be useful for \ to mean
3709 not to translate; but if we don't translate it
3710 it will never match anything. */
3712 }
3717 /* Expects the character in `c'. */
3718 normal_char:
3719 /* If no exactn currently being built. */
3722 /* If last exactn not at current position. */
3725 /* We have only one byte following the exactn for the count. */
3728 /* If followed by a repetition operator. */
3737 {
3738 /* Start building a new exactn. */
3744 }
3747 {
3751 {
3755 }
3756 else
3757 {
3760 {
3765 }
3768 }
3770 }
3777 /* Through the pattern now. */
3784 /* If we don't want backtracking, force success
3785 the first time we reach the end of the compiled pattern. */
3789 /* We have succeeded; set the length of the buffer. */
3792 #ifdef DEBUG
3794 {
3798 }
3799 debug--;
3802 #ifndef MATCH_MAY_ALLOCATE
3803 /* Initialize the failure stack to the largest possible stack. This
3804 isn't necessary unless we're trying to avoid calling alloca in
3805 the search and match routines. */
3806 {
3810 {
3814 fail_stack.stack
3817 else
3818 fail_stack.stack
3822 }
3825 }
3830 \f
3831 /* Subroutines for `regex_compile'. */
3833 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3835 static void
3837 {
3840 }
3843 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3845 static void
3847 {
3851 }
3854 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3855 for OP followed by two-byte integer parameter ARG. */
3857 static void
3859 {
3867 }
3870 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3872 static void
3874 {
3882 }
3885 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3886 after an alternative or a begin-subexpression. We assume there is at
3887 least one character before the ^. */
3889 static boolean
3891 {
3895 return
3896 /* After a subexpression? */
3898 /* After an alternative? */
3900 /* After a shy subexpression? */
3905 }
3908 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3909 at least one character after the $, i.e., `P < PEND'. */
3911 static boolean
3913 {
3918 return
3919 /* Before a subexpression? */
3922 /* Before an alternative? */
3925 }
3928 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3929 false if it's not. */
3931 static boolean
3933 {
3938 this_element--)
3943 }
3944 \f
3945 /* analyse_first.
3946 If fastmap is non-NULL, go through the pattern and fill fastmap
3947 with all the possible leading chars. If fastmap is NULL, don't
3948 bother filling it up (obviously) and only return whether the
3949 pattern could potentially match the empty string.
3951 Return 1 if p..pend might match the empty string.
3952 Return 0 if p..pend matches at least one char.
3953 Return -1 if fastmap was not updated accurately. */
3955 static int
3957 {
3961 /* If all elements for base leading-codes in fastmap is set, this
3962 flag is set true. */
3967 /* The loop below works as follows:
3968 - It has a working-list kept in the PATTERN_STACK and which basically
3969 starts by only containing a pointer to the first operation.
3970 - If the opcode we're looking at is a match against some set of
3971 chars, then we add those chars to the fastmap and go on to the
3972 next work element from the worklist (done via `break').
3973 - If the opcode is a control operator on the other hand, we either
3974 ignore it (if it's meaningless at this point, such as `start_memory')
3975 or execute it (if it's a jump). If the jump has several destinations
3976 (i.e. `on_failure_jump'), then we push the other destination onto the
3977 worklist.
3978 We guarantee termination by ignoring backward jumps (more or less),
3979 so that `p' is monotonically increasing. More to the point, we
3980 never set `p' (or push) anything `<= p1'. */
3983 {
3984 /* `p1' is used as a marker of how far back a `on_failure_jump'
3985 can go without being ignored. It is normally equal to `p'
3986 (which prevents any backward `on_failure_jump') except right
3987 after a plain `jump', to allow patterns such as:
3988 0: jump 10
3989 3..9: <body>
3990 10: on_failure_jump 3
3991 as used for the *? operator. */
3995 {
4000 /* If the first character has to match a backreference, that means
4001 that the group was empty (since it already matched). Since this
4002 is the only case that interests us here, we can assume that the
4003 backreference must match the empty string. */
4004 p++;
4008 /* Following are the cases which match a character. These end
4009 with `break'. */
4013 {
4014 /* If multibyte is nonzero, the first byte of each
4015 character is an ASCII or a leading code. Otherwise,
4016 each byte is a character. Thus, this works in both
4017 cases. */
4020 {
4021 /* For the case of matching this unibyte regex
4022 against multibyte, we must set a leading code of
4023 the corresponding multibyte character. */
4027 }
4028 }
4033 /* We could put all the chars except for \n (and maybe \0)
4034 but we don't bother since it is generally not worth it. */
4041 {
4042 /* Chars beyond end of bitmap are possible matches. */
4046 }
4048 /* Fallthrough */
4057 #ifdef emacs
4059 which doesn't match the specified set of characters. */
4060 not
4061 ||
4062 /* If we can match a character class, we can match any
4063 multibyte characters. */
4067 {
4069 {
4074 }
4075 }
4079 {
4080 /* Set fastmap[I] to 1 where I is a leading code of each
4081 multibyte character in the range table. */
4085 /* Make P points the range table. `+ 2' is to skip flag
4086 bits for a character class. */
4089 /* Extract the number of ranges in range table into COUNT. */
4092 {
4093 /* Extract the start and end of each range. */
4101 }
4102 }
4103 #endif
4109 #ifndef emacs
4117 /* This match depends on text properties. These end with
4118 aborting optimizations. */
4130 /* Any leading code can possibly start a character which
4131 has or doesn't has the specified category. */
4133 {
4138 }
4141 /* All cases after this match the empty string. These end with
4142 `continue'. */
4165 /* Backward jumps can only go back to code that we've already
4166 visited. `re_compile' should make sure this is true. */
4170 {
4176 p++;
4180 };
4181 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4182 to jump back to "just after here". */
4183 /* Fallthrough */
4193 else
4195 We first go down the "straight" path so as to minimize
4196 stack usage when going through alternatives. */
4200 }
4205 /* This code simply does not properly handle forward jump_n. */
4208 /* jump_n can either jump or fall through. The (backward) jump
4209 case has already been handled, so we only need to look at the
4210 fallthrough case. */
4214 /* If N == 0, it should be an on_failure_jump_loop instead. */
4217 /* We only care about one iteration of the loop, so we don't
4218 need to consider the case where this behaves like an
4219 on_failure_jump. */
4238 /* Getting here means we have found the possible starting
4239 characters for one path of the pattern -- and that the empty
4240 string does not match. We need not follow this path further. */
4244 /* We reached the end without matching anything. */
4248 \f
4249 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4250 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4251 characters can start a string that matches the pattern. This fastmap
4252 is used by re_search to skip quickly over impossible starting points.
4254 Character codes above (1 << BYTEWIDTH) are not represented in the
4255 fastmap, but the leading codes are represented. Thus, the fastmap
4256 indicates which character sets could start a match.
4258 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4259 area as BUFP->fastmap.
4261 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4262 the pattern buffer.
4264 Returns 0 if we succeed, -2 if an internal error. */
4266 int
4268 {
4282 \f
4283 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4284 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4285 this memory for recording register information. STARTS and ENDS
4286 must be allocated using the malloc library routine, and must each
4287 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4289 If NUM_REGS == 0, then subsequent matches should allocate their own
4290 register data.
4292 Unless this function is called, the first search or match using
4293 PATTERN_BUFFER will allocate its own register data, without
4294 freeing the old data. */
4296 void
4297 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4298 {
4300 {
4305 }
4306 else
4307 {
4311 }
4312 }
4314 \f
4315 /* Searching routines. */
4317 /* Like re_search_2, below, but only one string is specified, and
4318 doesn't let you say where to stop matching. */
4320 int
4321 re_search (struct re_pattern_buffer *bufp, const char *string, int size, int startpos, int range, struct re_registers *regs)
4322 {
4325 }
4328 /* Head address of virtual concatenation of string. */
4329 #define HEAD_ADDR_VSTRING(P) \
4330 (((P) >= size1 ? string2 : string1))
4332 /* End address of virtual concatenation of string. */
4333 #define STOP_ADDR_VSTRING(P) \
4334 (((P) >= size1 ? string2 + size2 : string1 + size1))
4336 /* Address of POS in the concatenation of virtual string. */
4337 #define POS_ADDR_VSTRING(POS) \
4338 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4340 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4341 virtual concatenation of STRING1 and STRING2, starting first at index
4342 STARTPOS, then at STARTPOS + 1, and so on.
4344 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4346 RANGE is how far to scan while trying to match. RANGE = 0 means try
4347 only at STARTPOS; in general, the last start tried is STARTPOS +
4348 RANGE.
4350 In REGS, return the indices of the virtual concatenation of STRING1
4351 and STRING2 that matched the entire BUFP->buffer and its contained
4352 subexpressions.
4354 Do not consider matching one past the index STOP in the virtual
4355 concatenation of STRING1 and STRING2.
4357 We return either the position in the strings at which the match was
4358 found, -1 if no match, or -2 if error (such as failure
4359 stack overflow). */
4361 int
4362 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, int size1, const char *str2, int size2, int startpos, int range, struct re_registers *regs, int stop)
4363 {
4371 boolean anchored_start;
4372 /* Nonzero if we are searching multibyte string. */
4375 /* Check for out-of-range STARTPOS. */
4379 /* Fix up RANGE if it might eventually take us outside
4380 the virtual concatenation of STRING1 and STRING2.
4381 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4387 /* If the search isn't to be a backwards one, don't waste time in a
4388 search for a pattern anchored at beginning of buffer. */
4390 {
4393 else
4395 }
4397 #ifdef emacs
4398 /* In a forward search for something that starts with \=.
4399 don't keep searching past point. */
4401 {
4405 }
4408 /* Update the fastmap now if not correct already. */
4412 /* See whether the pattern is anchored. */
4415 #ifdef emacs
4417 {
4421 }
4422 #endif
4424 /* Loop through the string, looking for a place to start matching. */
4426 {
4427 /* If the pattern is anchored,
4428 skip quickly past places we cannot match.
4429 We don't bother to treat startpos == 0 specially
4430 because that case doesn't repeat. */
4432 {
4437 }
4439 /* If a fastmap is supplied, skip quickly over characters that
4440 cannot be the start of a match. If the pattern can match the
4441 null string, however, we don't need to skip characters; we want
4442 the first null string. */
4444 {
4451 {
4458 /* Written out as an if-else to avoid testing `translate'
4459 inside the loop. */
4461 {
4464 {
4474 }
4475 else
4477 {
4488 d++;
4489 range--;
4490 }
4491 }
4492 else
4493 {
4496 {
4504 }
4505 else
4507 {
4508 d++;
4509 range--;
4510 }
4511 }
4513 }
4515 {
4517 {
4522 }
4523 else
4524 {
4535 }
4536 }
4537 }
4539 /* If can't match the null string, and that's all we have left, fail. */
4553 advance:
4557 {
4558 /* Update STARTPOS to the next character boundary. */
4560 {
4569 }
4570 else
4571 {
4572 range--;
4573 startpos++;
4574 }
4575 }
4576 else
4577 {
4578 range++;
4579 startpos--;
4581 /* Update STARTPOS to the previous character boundary. */
4583 {
4588 /* Find the head of multibyte form. */
4595 }
4596 }
4597 }
4601 \f
4602 /* Declarations and macros for re_match_2. */
4606 RE_TRANSLATE_TYPE translate,
4609 /* This converts PTR, a pointer into one of the search strings `string1'
4610 and `string2' into an offset from the beginning of that string. */
4611 #define POINTER_TO_OFFSET(ptr) \
4612 (FIRST_STRING_P (ptr) \
4613 ? ((regoff_t) ((ptr) - string1)) \
4614 : ((regoff_t) ((ptr) - string2 + size1)))
4616 /* Call before fetching a character with *d. This switches over to
4617 string2 if necessary.
4618 Check re_match_2_internal for a discussion of why end_match_2 might
4619 not be within string2 (but be equal to end_match_1 instead). */
4620 #define PREFETCH() \
4621 while (d == dend) \
4622 { \
4624 if (dend == end_match_2) \
4625 goto fail; \
4627 d = string2; \
4628 dend = end_match_2; \
4629 }
4631 /* Call before fetching a char with *d if you already checked other limits.
4632 This is meant for use in lookahead operations like wordend, etc..
4633 where we might need to look at parts of the string that might be
4634 outside of the LIMITs (i.e past `stop'). */
4635 #define PREFETCH_NOLIMIT() \
4636 if (d == end1) \
4637 { \
4638 d = string2; \
4639 dend = end_match_2; \
4640 } \
4642 /* Test if at very beginning or at very end of the virtual concatenation
4643 of `string1' and `string2'. If only one string, it's `string2'. */
4644 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4645 #define AT_STRINGS_END(d) ((d) == end2)
4648 /* Test if D points to a character which is word-constituent. We have
4649 two special cases to check for: if past the end of string1, look at
4650 the first character in string2; and if before the beginning of
4651 string2, look at the last character in string1. */
4652 #define WORDCHAR_P(d) \
4653 (SYNTAX ((d) == end1 ? *string2 \
4654 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4655 == Sword)
4657 /* Disabled due to a compiler bug -- see comment at case wordbound */
4659 /* The comment at case wordbound is following one, but we don't use
4660 AT_WORD_BOUNDARY anymore to support multibyte form.
4662 The DEC Alpha C compiler 3.x generates incorrect code for the
4663 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4664 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4665 macro and introducing temporary variables works around the bug. */
4667 #if 0
4668 /* Test if the character before D and the one at D differ with respect
4669 to being word-constituent. */
4670 #define AT_WORD_BOUNDARY(d) \
4671 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4672 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4673 #endif
4675 /* Free everything we malloc. */
4676 #ifdef MATCH_MAY_ALLOCATE
4677 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4678 # define FREE_VARIABLES() \
4679 do { \
4680 REGEX_FREE_STACK (fail_stack.stack); \
4681 FREE_VAR (regstart); \
4682 FREE_VAR (regend); \
4683 FREE_VAR (best_regstart); \
4684 FREE_VAR (best_regend); \
4685 } while (0)
4686 #else
4690 \f
4691 /* Optimization routines. */
4693 /* If the operation is a match against one or more chars,
4694 return a pointer to the next operation, else return NULL. */
4697 {
4699 {
4710 {
4715 }
4716 else
4722 #ifdef emacs
4726 p++;
4731 }
4733 }
4736 /* Jump over non-matching operations. */
4739 {
4742 {
4744 {
4757 }
4758 }
4761 }
4763 /* Non-zero if "p1 matches something" implies "p2 fails". */
4764 static int
4766 {
4767 re_opcode_t op2;
4774 /* Skip over open/close-group commands.
4775 If what follows this loop is a ...+ construct,
4776 look at what begins its body, since we will have to
4777 match at least one of that. */
4779 /* The same skip can be done for p1, except that this function
4780 is only used in the case where p1 is a simple match operator. */
4781 /* p1 = skip_noops (p1, pend); */
4789 {
4792 /* If we're at the end of the pattern, we can change. */
4794 {
4797 }
4802 {
4803 register re_wchar_t c
4808 {
4810 {
4813 }
4814 }
4818 {
4821 /* Test if C is listed in charset (or charset_not)
4822 at `p1'. */
4824 {
4828 }
4832 /* `not' is equal to 1 if c would match, which means
4833 that we can't change to pop_failure_jump. */
4835 {
4838 }
4839 }
4842 {
4845 }
4846 }
4850 {
4852 /* Reuse the code above. */
4855 /* It is hard to list up all the character in charset
4856 P2 if it includes multibyte character. Give up in
4857 such case. */
4859 {
4860 /* Now, we are sure that P2 has no range table.
4861 So, for the size of bitmap in P2, `p2[1]' is
4862 enough. But P1 may have range table, so the
4863 size of bitmap table of P1 is extracted by
4864 using macro `CHARSET_BITMAP_SIZE'.
4866 In a multibyte case, we know that all the character
4867 listed in P2 is ASCII. In a unibyte case, P1 has only a
4868 bitmap table. So, in both cases, it is enough to test
4869 only the bitmap table of P1. */
4872 {
4874 /* We win if the charset inside the loop
4875 has no overlap with the one after the loop. */
4879 idx++)
4885 {
4888 }
4889 }
4891 {
4893 /* We win if the charset_not inside the loop lists
4894 every character listed in the charset after. */
4902 {
4905 }
4906 }
4907 }
4908 }
4913 {
4916 /* Reuse the code above. */
4919 /* When we have two charset_not, it's very unlikely that
4920 they don't overlap. The union of the two sets of excluded
4921 chars should cover all possible chars, which, as a matter of
4922 fact, is virtually impossible in multibyte buffers. */
4924 }
4948 #ifdef emacs
4956 ;
4957 }
4959 /* Safe default. */
4961 }
4963 \f
4964 /* Matching routines. */
4967 /* re_match is like re_match_2 except it takes only a single string. */
4969 int
4972 {
4976 }
4980 #ifdef emacs
4981 /* In Emacs, this is the string or buffer in which we
4982 are matching. It is used for looking up syntax properties. */
4983 Lisp_Object re_match_object;
4984 #endif
4986 /* re_match_2 matches the compiled pattern in BUFP against the
4987 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4988 and SIZE2, respectively). We start matching at POS, and stop
4989 matching at STOP.
4991 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4992 store offsets for the substring each group matched in REGS. See the
4993 documentation for exactly how many groups we fill.
4995 We return -1 if no match, -2 if an internal error (such as the
4996 failure stack overflowing). Otherwise, we return the length of the
4997 matched substring. */
4999 int
5000 re_match_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, const char *string2, int size2, int pos, struct re_registers *regs, int stop)
5001 {
5004 #ifdef emacs
5009 #endif
5015 }
5019 /* This is a separate function so that we can force an alloca cleanup
5020 afterwards. */
5021 static int
5022 re_match_2_internal (struct re_pattern_buffer *bufp, const re_char *string1, int size1, const re_char *string2, int size2, int pos, struct re_registers *regs, int stop)
5023 {
5024 /* General temporaries. */
5029 /* Just past the end of the corresponding string. */
5032 /* Pointers into string1 and string2, just past the last characters in
5033 each to consider matching. */
5036 /* Where we are in the data, and the end of the current string. */
5039 /* Used sometimes to remember where we were before starting matching
5040 an operator so that we can go back in case of failure. This "atomic"
5041 behavior of matching opcodes is indispensable to the correctness
5042 of the on_failure_keep_string_jump optimization. */
5045 /* Where we are in the pattern, and the end of the pattern. */
5049 /* We use this to map every character in the string. */
5052 /* Nonzero if BUFP is setup from a multibyte regex. */
5055 /* Nonzero if STRING1/STRING2 are multibyte. */
5058 /* Failure point stack. Each place that can handle a failure further
5059 down the line pushes a failure point on this stack. It consists of
5060 regstart, and regend for all registers corresponding to
5061 the subexpressions we're currently inside, plus the number of such
5062 registers, and, finally, two char *'s. The first char * is where
5063 to resume scanning the pattern; the second one is where to resume
5064 scanning the strings. */
5066 fail_stack_type fail_stack;
5067 #endif
5068 #ifdef DEBUG
5070 #endif
5072 #if defined REL_ALLOC && defined REGEX_MALLOC
5073 /* This holds the pointer to the failure stack, when
5074 it is allocated relocatably. */
5076 #endif
5078 /* We fill all the registers internally, independent of what we
5079 return, for use in backreferences. The number here includes
5080 an element for register zero. */
5083 /* Information on the contents of registers. These are pointers into
5084 the input strings; they record just what was matched (on this
5085 attempt) by a subexpression part of the pattern, that is, the
5086 regnum-th regstart pointer points to where in the pattern we began
5087 matching and the regnum-th regend points to right after where we
5088 stopped matching the regnum-th subexpression. (The zeroth register
5089 keeps track of what the whole pattern matches.) */
5092 #endif
5094 /* The following record the register info as found in the above
5095 variables when we find a match better than any we've seen before.
5096 This happens as we backtrack through the failure points, which in
5097 turn happens only if we have not yet matched the entire string. */
5101 #endif
5103 /* Logically, this is `best_regend[0]'. But we don't want to have to
5104 allocate space for that if we're not allocating space for anything
5105 else (see below). Also, we never need info about register 0 for
5106 any of the other register vectors, and it seems rather a kludge to
5107 treat `best_regend' differently than the rest. So we keep track of
5108 the end of the best match so far in a separate variable. We
5109 initialize this to NULL so that when we backtrack the first time
5110 and need to test it, it's not garbage. */
5113 #ifdef DEBUG
5114 /* Counts the total number of registers pushed. */
5116 #endif
5122 #ifdef MATCH_MAY_ALLOCATE
5123 /* Do not bother to initialize all the register variables if there are
5124 no groups in the pattern, as it takes a fair amount of time. If
5125 there are groups, we include space for register 0 (the whole
5126 pattern), even though we never use it, since it simplifies the
5127 array indexing. We should fix this. */
5129 {
5136 {
5139 }
5140 }
5141 else
5142 {
5143 /* We must initialize all our variables to NULL, so that
5144 `FREE_VARIABLES' doesn't try to free them. */
5146 }
5149 /* The starting position is bogus. */
5151 {
5154 }
5156 /* Initialize subexpression text positions to -1 to mark ones that no
5157 start_memory/stop_memory has been seen for. Also initialize the
5158 register information struct. */
5162 /* We move `string1' into `string2' if the latter's empty -- but not if
5163 `string1' is null. */
5165 {
5170 }
5174 /* `p' scans through the pattern as `d' scans through the data.
5175 `dend' is the end of the input string that `d' points within. `d'
5176 is advanced into the following input string whenever necessary, but
5177 this happens before fetching; therefore, at the beginning of the
5178 loop, `d' can be pointing at the end of a string, but it cannot
5179 equal `string2'. */
5181 {
5182 /* Only match within string2. */
5186 }
5187 else
5188 {
5190 {
5191 /* Only match within string1. */
5193 /* BEWARE!
5194 When we reach end_match_1, PREFETCH normally switches to string2.
5195 But in the present case, this means that just doing a PREFETCH
5196 makes us jump from `stop' to `gap' within the string.
5197 What we really want here is for the search to stop as
5198 soon as we hit end_match_1. That's why we set end_match_2
5199 to end_match_1 (since PREFETCH fails as soon as we hit
5200 end_match_2). */
5202 }
5203 else
5205 moving `d' from end1 to string2 will not prevent the d == dend
5206 check from catching the end of string. */
5209 }
5212 }
5220 /* This loops over pattern commands. It exits by returning from the
5221 function if the match is complete, or it drops through if the match
5222 fails at this starting point in the input data. */
5224 {
5231 /* If we haven't matched the entire string, and we want the
5232 longest match, try backtracking. */
5234 {
5235 /* 1 if this match ends in the same string (string1 or string2)
5236 as the best previous match. */
5239 /* 1 if this match is the best seen so far. */
5240 boolean best_match_p;
5242 /* AIX compiler got confused when this was combined
5243 with the previous declaration. */
5246 else
5254 /* If exceeds best match so far, save it. */
5256 {
5263 {
5266 }
5267 }
5269 }
5271 /* If no failure points, don't restore garbage. And if
5272 last match is real best match, don't restore second
5273 best one. */
5275 {
5276 restore_best_regs:
5277 /* Restore best match. It may happen that `dend ==
5278 end_match_1' while the restored d is in string2.
5279 For example, the pattern `x.*y.*z' against the
5280 strings `x-' and `y-z-', if the two strings are
5281 not consecutive in memory. */
5289 {
5292 }
5293 }
5296 succeed_label:
5299 /* If caller wants register contents data back, do it. */
5301 {
5302 /* Have the register data arrays been allocated? */
5305 extra element beyond `num_regs' for the `-1' marker
5306 GNU code uses. */
5311 {
5314 }
5316 }
5319 allocated, reallocate them. If we need fewer, just
5320 leave it alone. */
5322 {
5327 {
5330 }
5331 }
5332 }
5333 else
5334 {
5335 /* These braces fend off a "empty body in an else-statement"
5336 warning under GCC when assert expands to nothing. */
5338 }
5340 /* Convert the pointer data in `regstart' and `regend' to
5341 indices. Register zero has to be set differently,
5342 since we haven't kept track of any info for it. */
5344 {
5347 }
5349 /* Go through the first `min (num_regs, regs->num_regs)'
5350 registers, since that is all we initialized. */
5352 {
5355 else
5356 {
5361 }
5362 }
5364 /* If the regs structure we return has more elements than
5365 were in the pattern, set the extra elements to -1. If
5366 we (re)allocated the registers, this is the case,
5367 because we always allocate enough to have at least one
5368 -1 at the end. */
5384 }
5386 /* Otherwise match next pattern command. */
5388 {
5389 /* Ignore these. Used to ignore the n of succeed_n's which
5390 currently have n == 0. */
5399 /* Match the next n pattern characters exactly. The following
5400 byte in the pattern defines n, and the n bytes after that
5401 are the characters to match. */
5406 /* Remember the start point to rollback upon failure. */
5409 #ifndef emacs
5410 /* This is written out as an if-else so we don't waste time
5411 testing `translate' inside the loop. */
5413 do
5414 {
5417 {
5420 }
5421 d++;
5422 }
5424 else
5425 do
5426 {
5429 {
5432 }
5433 }
5436 /* The cost of testing `translate' is comparatively small. */
5438 do
5439 {
5446 else
5447 {
5450 }
5454 {
5457 }
5462 }
5464 else
5465 do
5466 {
5472 {
5475 }
5476 else
5477 {
5480 }
5483 {
5488 }
5489 else
5492 {
5495 }
5497 d++;
5498 }
5500 #endif
5504 /* Match any character except possibly a newline or a null. */
5506 {
5508 re_wchar_t buf_ch;
5514 target_multibyte);
5525 }
5531 {
5536 /* Start of actual range_table, or end of bitmap if there is no
5537 range table. */
5540 /* Nonzero if there is a range table. */
5543 /* Number of ranges of range table. This is not included
5544 in the initial byte-length of the command. */
5547 /* Whether matching against a unibyte character. */
5555 {
5558 }
5563 {
5569 {
5572 }
5573 }
5574 else
5575 {
5579 {
5583 {
5586 }
5587 }
5588 else
5590 }
5594 /* Cast to `unsigned' instead of `unsigned char' in
5595 case the bit list is a full 32 bytes long. */
5599 }
5600 #ifdef emacs
5602 {
5612 else
5614 }
5619 else
5626 }
5629 /* The beginning of a group is represented by start_memory.
5630 The argument is the register number. The text
5631 matched within the group is recorded (in the internal
5632 registers data structure) under the register number. */
5636 /* In case we need to undo this operation (via backtracking). */
5643 /* Move past the register number and inner group count. */
5648 /* The stop_memory opcode represents the end of a group. Its
5649 argument is the same as start_memory's: the register number. */
5654 /* Strictly speaking, there should be code such as:
5656 assert (REG_UNSET (regend[*p]));
5657 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5659 But the only info to be pushed is regend[*p] and it is known to
5660 be UNSET, so there really isn't anything to push.
5661 Not pushing anything, on the other hand deprives us from the
5662 guarantee that regend[*p] is UNSET since undoing this operation
5663 will not reset its value properly. This is not important since
5664 the value will only be read on the next start_memory or at
5665 the very end and both events can only happen if this stop_memory
5666 is *not* undone. */
5671 /* Move past the register number and the inner group count. */
5676 /* \<digit> has been turned into a `duplicate' command which is
5677 followed by the numeric value of <digit> as the register number. */
5679 {
5684 /* Can't back reference a group which we've never matched. */
5688 /* Where in input to try to start matching. */
5691 /* Remember the start point to rollback upon failure. */
5694 /* Where to stop matching; if both the place to start and
5695 the place to stop matching are in the same string, then
5696 set to the place to stop, otherwise, for now have to use
5697 the end of the first string. */
5703 {
5704 /* If necessary, advance to next segment in register
5705 contents. */
5707 {
5711 /* End of string1 => advance to string2. */
5714 }
5715 /* At end of register contents => success */
5718 /* If necessary, advance to next segment in data. */
5721 /* How many characters left in this segment to match. */
5724 /* Want how many consecutive characters we can match in
5725 one shot, so, if necessary, adjust the count. */
5729 /* Compare that many; failure if mismatch, else move
5730 past them. */
5734 {
5737 }
5739 }
5740 }
5744 /* begline matches the empty string at the beginning of the string
5745 (unless `not_bol' is set in `bufp'), and after newlines. */
5750 {
5752 }
5753 else
5754 {
5759 }
5760 /* In all other cases, we fail. */
5764 /* endline is the dual of begline. */
5769 {
5771 }
5772 else
5773 {
5777 }
5781 /* Match at the very beginning of the data. */
5789 /* Match at the very end of the data. */
5797 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5798 pushes NULL as the value for the string on the stack. Then
5799 `POP_FAILURE_POINT' will keep the current value for the
5800 string, instead of restoring it. To see why, consider
5801 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5802 then the . fails against the \n. But the next thing we want
5803 to do is match the \n against the \n; if we restored the
5804 string value, we would be back at the foo.
5806 Because this is used only in specific cases, we don't need to
5807 check all the things that `on_failure_jump' does, to make
5808 sure the right things get saved on the stack. Hence we don't
5809 share its code. The only reason to push anything on the
5810 stack at all is that otherwise we would have to change
5811 `anychar's code to do something besides goto fail in this
5812 case; that seems worse than this. */
5821 /* A nasty loop is introduced by the non-greedy *? and +?.
5822 With such loops, the stack only ever contains one failure point
5823 at a time, so that a plain on_failure_jump_loop kind of
5824 cycle detection cannot work. Worse yet, such a detection
5825 can not only fail to detect a cycle, but it can also wrongly
5826 detect a cycle (between different instantiations of the same
5827 loop).
5828 So the method used for those nasty loops is a little different:
5829 We use a special cycle-detection-stack-frame which is pushed
5830 when the on_failure_jump_nastyloop failure-point is *popped*.
5831 This special frame thus marks the beginning of one iteration
5832 through the loop and we can hence easily check right here
5833 whether something matched between the beginning and the end of
5834 the loop. */
5841 {
5845 /* If there's a cycle, just continue without pushing
5846 this failure point. The failure point is the "try again"
5847 option, which shouldn't be tried.
5848 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5850 }
5853 /* Simple loop detecting on_failure_jump: just check on the
5854 failure stack if the same spot was already hit earlier. */
5856 on_failure:
5860 {
5864 /* If there's a cycle, get out of the loop, as if the matching
5865 had failed. We used to just `goto fail' here, but that was
5866 aborting the search a bit too early: we want to keep the
5867 empty-loop-match and keep matching after the loop.
5868 We want (x?)*y\1z to match both xxyz and xxyxz. */
5870 else
5872 }
5876 /* Uses of on_failure_jump:
5878 Each alternative starts with an on_failure_jump that points
5879 to the beginning of the next alternative. Each alternative
5880 except the last ends with a jump that in effect jumps past
5881 the rest of the alternatives. (They really jump to the
5882 ending jump of the following alternative, because tensioning
5883 these jumps is a hassle.)
5885 Repeats start with an on_failure_jump that points past both
5886 the repetition text and either the following jump or
5887 pop_failure_jump back to this on_failure_jump. */
5896 /* This operation is used for greedy *.
5897 Compare the beginning of the repeat with what in the
5898 pattern follows its end. If we can establish that there
5899 is nothing that they would both match, i.e., that we
5900 would have to backtrack because of (as in, e.g., `a*a')
5901 then we can use a non-backtracking loop based on
5902 on_failure_keep_string_jump instead of on_failure_jump. */
5907 {
5909 /* Here, we discard `const', making re_match non-reentrant. */
5914 instruction once it's been changed. */
5918 /* Ensure this is a indeed the trivial kind of loop
5919 we are expecting. */
5924 {
5925 /* Use a fast `on_failure_keep_string_jump' loop. */
5929 }
5930 else
5931 {
5932 /* Default to a safe `on_failure_jump' loop. */
5935 }
5937 }
5940 /* Unconditionally jump (without popping any failure points). */
5942 unconditional_jump:
5943 IMMEDIATE_QUIT_CHECK;
5951 /* Have to succeed matching what follows at least n times.
5952 After that, handle like `on_failure_jump'. */
5954 /* Signedness doesn't matter since we only compare MCNT to 0. */
5958 /* Originally, mcnt is how many times we HAVE to succeed. */
5960 {
5961 /* Here, we discard `const', making re_match non-reentrant. */
5963 mcnt--;
5966 }
5967 else
5968 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5973 /* Signedness doesn't matter since we only compare MCNT to 0. */
5977 /* Originally, this is how many times we CAN jump. */
5979 {
5980 /* Here, we discard `const', making re_match non-reentrant. */
5982 mcnt--;
5985 }
5986 /* If don't have to jump any more, skip over the rest of command. */
5987 else
5992 {
5997 /* Here, we discard `const', making re_match non-reentrant. */
5999 /* Signedness doesn't matter since we only copy MCNT's bits . */
6004 }
6011 /* We SUCCEED (or FAIL) in one of the following cases: */
6013 /* Case 1: D is at the beginning or the end of string. */
6016 else
6017 {
6018 /* C1 is the character before D, S1 is the syntax of C1, C2
6019 is the character at D, and S2 is the syntax of C2. */
6023 #ifdef emacs
6027 #endif
6030 #ifdef emacs
6032 #endif
6039 /* Case 3: Both of S1 and S2 are Sword, and macro
6040 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6043 }
6046 else
6052 /* We FAIL in one of the following cases: */
6054 /* Case 1: D is at the end of string. */
6057 else
6058 {
6059 /* C1 is the character before D, S1 is the syntax of C1, C2
6060 is the character at D, and S2 is the syntax of C2. */
6064 #ifdef emacs
6068 #endif
6073 /* Case 2: S2 is not Sword. */
6077 /* Case 3: D is not at the beginning of string ... */
6079 {
6081 #ifdef emacs
6083 #endif
6086 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6087 returns 0. */
6090 }
6091 }
6097 /* We FAIL in one of the following cases: */
6099 /* Case 1: D is at the beginning of string. */
6102 else
6103 {
6104 /* C1 is the character before D, S1 is the syntax of C1, C2
6105 is the character at D, and S2 is the syntax of C2. */
6109 #ifdef emacs
6113 #endif
6117 /* Case 2: S1 is not Sword. */
6121 /* Case 3: D is not at the end of string ... */
6123 {
6126 #ifdef emacs
6128 #endif
6131 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6132 returns 0. */
6135 }
6136 }
6142 /* We FAIL in one of the following cases: */
6144 /* Case 1: D is at the end of string. */
6147 else
6148 {
6149 /* C1 is the character before D, S1 is the syntax of C1, C2
6150 is the character at D, and S2 is the syntax of C2. */
6153 #ifdef emacs
6157 #endif
6162 /* Case 2: S2 is neither Sword nor Ssymbol. */
6166 /* Case 3: D is not at the beginning of string ... */
6168 {
6170 #ifdef emacs
6172 #endif
6175 /* ... and S1 is Sword or Ssymbol. */
6178 }
6179 }
6185 /* We FAIL in one of the following cases: */
6187 /* Case 1: D is at the beginning of string. */
6190 else
6191 {
6192 /* C1 is the character before D, S1 is the syntax of C1, C2
6193 is the character at D, and S2 is the syntax of C2. */
6196 #ifdef emacs
6200 #endif
6204 /* Case 2: S1 is neither Ssymbol nor Sword. */
6208 /* Case 3: D is not at the end of string ... */
6210 {
6213 #ifdef emacs
6215 #endif
6218 /* ... and S2 is Sword or Ssymbol. */
6221 }
6222 }
6231 #ifdef emacs
6232 {
6236 }
6237 #endif
6238 {
6240 re_wchar_t c;
6246 }
6249 #ifdef emacs
6274 {
6276 re_wchar_t c;
6282 }
6289 }
6293 /* We goto here if a matching operation fails. */
6294 fail:
6295 IMMEDIATE_QUIT_CHECK;
6297 {
6299 /* A restart point is known. Restore to that state. */
6303 {
6311 /* Fallthrough */
6317 continue_failure_jump:
6323 /* A special frame used for nastyloops. */
6328 }
6334 }
6335 else
6346 \f
6347 /* Subroutine definitions for re_match_2. */
6349 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6350 bytes; nonzero otherwise. */
6352 static int
6355 {
6360 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6361 different lengths, but relying on a single `len' would break this. -sm */
6363 {
6375 }
6381 }
6382 \f
6383 /* Entry points for GNU code. */
6385 /* re_compile_pattern is the GNU regular expression compiler: it
6386 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6387 Returns 0 if the pattern was valid, otherwise an error string.
6389 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6390 are set in BUFP on entry.
6392 We call regex_compile to do the actual compilation. */
6396 {
6397 reg_errcode_t ret;
6399 /* GNU code is written to assume at least RE_NREGS registers will be set
6400 (and at least one extra will be -1). */
6403 /* And GNU code determines whether or not to get register information
6404 by passing null for the REGS argument to re_match, etc., not by
6405 setting no_sub. */
6413 }
6415 \f
6416 /* Entry points compatible with 4.2 BSD regex library. We don't define
6417 them unless specifically requested. */
6419 #if defined _REGEX_RE_COMP || defined _LIBC
6421 /* BSD has one and only one pattern buffer. */
6425 # ifdef _LIBC
6426 /* Make these definitions weak in libc, so POSIX programs can redefine
6427 these names if they don't use our functions, and still use
6428 regcomp/regexec below without link errors. */
6429 weak_function
6430 # endif
6433 {
6434 reg_errcode_t ret;
6437 {
6439 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6442 }
6445 {
6448 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6454 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6456 }
6458 /* Since `re_exec' always passes NULL for the `regs' argument, we
6459 don't need to initialize the pattern buffer fields which affect it. */
6466 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6468 }
6471 int
6472 # ifdef _LIBC
6473 weak_function
6474 # endif
6477 {
6479 return
6481 }
6483 \f
6484 /* POSIX.2 functions. Don't define these for Emacs. */
6486 #ifndef emacs
6488 /* regcomp takes a regular expression as a string and compiles it.
6490 PREG is a regex_t *. We do not expect any fields to be initialized,
6491 since POSIX says we shouldn't. Thus, we set
6493 `buffer' to the compiled pattern;
6494 `used' to the length of the compiled pattern;
6495 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6496 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6497 RE_SYNTAX_POSIX_BASIC;
6498 `fastmap' to an allocated space for the fastmap;
6499 `fastmap_accurate' to zero;
6500 `re_nsub' to the number of subexpressions in PATTERN.
6502 PATTERN is the address of the pattern string.
6504 CFLAGS is a series of bits which affect compilation.
6506 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6507 use POSIX basic syntax.
6509 If REG_NEWLINE is set, then . and [^...] don't match newline.
6510 Also, regexec will try a match beginning after every newline.
6512 If REG_ICASE is set, then we considers upper- and lowercase
6513 versions of letters to be equivalent when matching.
6515 If REG_NOSUB is set, then when PREG is passed to regexec, that
6516 routine will report only success or failure, and nothing about the
6517 registers.
6519 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6520 the return codes and their meanings.) */
6522 int
6525 {
6526 reg_errcode_t ret;
6527 reg_syntax_t syntax
6531 /* regex_compile will allocate the space for the compiled pattern. */
6536 /* Try to allocate space for the fastmap. */
6540 {
6543 preg->translate
6549 /* Map uppercase characters to corresponding lowercase ones. */
6552 }
6553 else
6556 /* If REG_NEWLINE is set, newlines are treated differently. */
6561 }
6562 else
6567 /* POSIX says a null character in the pattern terminates it, so we
6568 can use strlen here in compiling the pattern. */
6571 /* POSIX doesn't distinguish between an unmatched open-group and an
6572 unmatched close-group: both are REG_EPAREN. */
6578 buffer. */
6584 }
6585 }
6587 }
6591 /* regexec searches for a given pattern, specified by PREG, in the
6592 string STRING.
6594 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6595 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6596 least NMATCH elements, and we set them to the offsets of the
6597 corresponding matched substrings.
6599 EFLAGS specifies `execution flags' which affect matching: if
6600 REG_NOTBOL is set, then ^ does not match at the beginning of the
6601 string; if REG_NOTEOL is set, then $ does not match at the end.
6603 We return 0 if we find a match and REG_NOMATCH if not. */
6605 int
6608 {
6611 regex_t private_preg;
6620 /* The user has told us exactly how many registers to return
6621 information about, via `nmatch'. We have to pass that on to the
6622 matching routines. */
6626 {
6632 }
6634 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6635 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6636 was a little bit longer but still only matching the real part.
6637 This works because the `endline' will check for a '\n' and will find a
6638 '\0', correctly deciding that this is not the end of a line.
6639 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6640 a convenient '\0' there. For all we know, the string could be preceded
6641 by '\n' which would throw things off. */
6643 /* Perform the searching operation. */
6648 /* Copy the register information to the POSIX structure. */
6650 {
6652 {
6656 {
6659 }
6660 }
6662 /* If we needed the temporary register info, free the space now. */
6664 }
6666 /* We want zero return to mean success, unlike `re_search'. */
6668 }
6672 /* Returns a message corresponding to an error code, ERR_CODE, returned
6673 from either regcomp or regexec. We don't use PREG here.
6675 ERR_CODE was previously called ERRCODE, but that name causes an
6676 error with msvc8 compiler. */
6678 size_t
6680 {
6686 /* Only error codes returned by the rest of the code should be passed
6687 to this routine. If we are given anything else, or if other regex
6688 code generates an invalid error code, then the program has a bug.
6689 Dump core so we can fix it. */
6697 {
6699 {
6702 }
6703 else
6705 }
6708 }
6712 /* Free dynamically allocated space used by PREG. */
6714 void
6716 {
6729 }