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) 1993-2015 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)
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, see <http://www.gnu.org/licenses/>. */
21 - structure the opcode space into opcode+flag.
22 - merge with glibc's regex.[ch].
23 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
24 need to modify the compiled regexp so that re_match can be reentrant.
25 - get rid of on_failure_jump_smart by doing the optimization in re_comp
26 rather than at run-time, so that re_match can be reentrant.
29 /* AIX requires this to be the first thing in the file. */
30 #if defined _AIX && !defined REGEX_MALLOC
34 /* Ignore some GCC warnings for now. This section should go away
35 once the Emacs and Gnulib regex code is merged. */
36 #if 4 < __GNUC__ + (5 <= __GNUC_MINOR__) || defined __clang__
37 # pragma GCC diagnostic ignored "-Wstrict-overflow"
39 # pragma GCC diagnostic ignored "-Wunused-function"
40 # pragma GCC diagnostic ignored "-Wunused-macros"
41 # pragma GCC diagnostic ignored "-Wunused-result"
42 # pragma GCC diagnostic ignored "-Wunused-variable"
46 #if 4 < __GNUC__ + (6 <= __GNUC_MINOR__) && ! defined __clang__
47 # pragma GCC diagnostic ignored "-Wunused-but-set-variable"
55 /* We need this for `regex.h', and perhaps for the Emacs include files. */
56 # include <sys/types.h>
59 /* Whether to use ISO C Amendment 1 wide char functions.
60 Those should not be used for Emacs since it uses its own. */
62 #define WIDE_CHAR_SUPPORT 1
64 #define WIDE_CHAR_SUPPORT \
65 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
68 /* For platform which support the ISO C amendment 1 functionality we
69 support user defined character classes. */
71 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
77 /* We have to keep the namespace clean. */
78 # define regfree(preg) __regfree (preg)
79 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
80 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
81 # define regerror(err_code, preg, errbuf, errbuf_size) \
82 __regerror (err_code, preg, errbuf, errbuf_size)
83 # define re_set_registers(bu, re, nu, st, en) \
84 __re_set_registers (bu, re, nu, st, en)
85 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
86 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
87 # define re_match(bufp, string, size, pos, regs) \
88 __re_match (bufp, string, size, pos, regs)
89 # define re_search(bufp, string, size, startpos, range, regs) \
90 __re_search (bufp, string, size, startpos, range, regs)
91 # define re_compile_pattern(pattern, length, bufp) \
92 __re_compile_pattern (pattern, length, bufp)
93 # define re_set_syntax(syntax) __re_set_syntax (syntax)
94 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
95 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
96 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
98 /* Make sure we call libc's function even if the user overrides them. */
99 # define btowc __btowc
100 # define iswctype __iswctype
101 # define wctype __wctype
103 # define WEAK_ALIAS(a,b) weak_alias (a, b)
105 /* We are also using some library internals. */
106 # include <locale/localeinfo.h>
107 # include <locale/elem-hash.h>
108 # include <langinfo.h>
110 # define WEAK_ALIAS(a,b)
113 /* This is for other GNU distributions with internationalized messages. */
114 #if HAVE_LIBINTL_H || defined _LIBC
115 # include <libintl.h>
117 # define gettext(msgid) (msgid)
121 /* This define is so xgettext can find the internationalizable
123 # define gettext_noop(String) String
126 /* The `emacs' switch turns on certain matching commands
127 that make sense only in Emacs. */
131 # include "character.h"
135 # include "category.h"
137 /* Make syntax table lookup grant data in gl_state. */
138 # define SYNTAX(c) syntax_property (c, 1)
143 # define malloc xmalloc
147 # define realloc xrealloc
153 /* Converts the pointer to the char to BEG-based offset from the start. */
154 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
155 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
157 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
158 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
159 # define RE_STRING_CHAR(p, multibyte) \
160 (multibyte ? (STRING_CHAR (p)) : (*(p)))
161 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
162 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
164 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
166 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
168 /* Set C a (possibly converted to multibyte) character before P. P
169 points into a string which is the virtual concatenation of STR1
170 (which ends at END1) or STR2 (which ends at END2). */
171 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
173 if (target_multibyte) \
175 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
176 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
177 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
178 c = STRING_CHAR (dtemp); \
182 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
183 (c) = RE_CHAR_TO_MULTIBYTE (c); \
187 /* Set C a (possibly converted to multibyte) character at P, and set
188 LEN to the byte length of that character. */
189 # define GET_CHAR_AFTER(c, p, len) \
191 if (target_multibyte) \
192 (c) = STRING_CHAR_AND_LENGTH (p, len); \
197 (c) = RE_CHAR_TO_MULTIBYTE (c); \
201 #else /* not emacs */
203 /* If we are not linking with Emacs proper,
204 we can't use the relocating allocator
205 even if config.h says that we can. */
210 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
213 xmalloc (size_t size
)
215 void *val
= malloc (size
);
218 write (2, "virtual memory exhausted\n", 25);
225 xrealloc (void *block
, size_t size
)
228 /* We must call malloc explicitly when BLOCK is 0, since some
229 reallocs don't do this. */
233 val
= realloc (block
, size
);
236 write (2, "virtual memory exhausted\n", 25);
245 # define malloc xmalloc
249 # define realloc xrealloc
251 # include <stdbool.h>
254 /* Define the syntax stuff for \<, \>, etc. */
256 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
257 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
259 /* Dummy macros for non-Emacs environments. */
260 # define MAX_MULTIBYTE_LENGTH 1
261 # define RE_MULTIBYTE_P(x) 0
262 # define RE_TARGET_MULTIBYTE_P(x) 0
263 # define WORD_BOUNDARY_P(c1, c2) (0)
264 # define BYTES_BY_CHAR_HEAD(p) (1)
265 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
266 # define STRING_CHAR(p) (*(p))
267 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
268 # define CHAR_STRING(c, s) (*(s) = (c), 1)
269 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
270 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
271 # define RE_CHAR_TO_MULTIBYTE(c) (c)
272 # define RE_CHAR_TO_UNIBYTE(c) (c)
273 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
274 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
275 # define GET_CHAR_AFTER(c, p, len) \
277 # define CHAR_BYTE8_P(c) (0)
278 # define CHAR_LEADING_CODE(c) (c)
280 #endif /* not emacs */
283 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
284 # define RE_TRANSLATE_P(TBL) (TBL)
287 /* Get the interface, including the syntax bits. */
290 /* isalpha etc. are used for the character classes. */
295 /* 1 if C is an ASCII character. */
296 # define IS_REAL_ASCII(c) ((c) < 0200)
298 /* 1 if C is a unibyte character. */
299 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
301 /* The Emacs definitions should not be directly affected by locales. */
303 /* In Emacs, these are only used for single-byte characters. */
304 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
305 # define ISCNTRL(c) ((c) < ' ')
306 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
307 || ((c) >= 'a' && (c) <= 'f') \
308 || ((c) >= 'A' && (c) <= 'F'))
310 /* This is only used for single-byte characters. */
311 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
313 /* The rest must handle multibyte characters. */
315 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
316 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
319 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
320 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
323 # define ISALNUM(c) (IS_REAL_ASCII (c) \
324 ? (((c) >= 'a' && (c) <= 'z') \
325 || ((c) >= 'A' && (c) <= 'Z') \
326 || ((c) >= '0' && (c) <= '9')) \
327 : (alphabeticp (c) || decimalnump (c)))
329 # define ISALPHA(c) (IS_REAL_ASCII (c) \
330 ? (((c) >= 'a' && (c) <= 'z') \
331 || ((c) >= 'A' && (c) <= 'Z')) \
334 # define ISLOWER(c) lowercasep (c)
336 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
337 ? ((c) > ' ' && (c) < 0177 \
338 && !(((c) >= 'a' && (c) <= 'z') \
339 || ((c) >= 'A' && (c) <= 'Z') \
340 || ((c) >= '0' && (c) <= '9'))) \
341 : SYNTAX (c) != Sword)
343 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
345 # define ISUPPER(c) uppercasep (c)
347 # define ISWORD(c) (SYNTAX (c) == Sword)
349 #else /* not emacs */
351 /* 1 if C is an ASCII character. */
352 # define IS_REAL_ASCII(c) ((c) < 0200)
354 /* This distinction is not meaningful, except in Emacs. */
355 # define ISUNIBYTE(c) 1
358 # define ISBLANK(c) isblank (c)
360 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
363 # define ISGRAPH(c) isgraph (c)
365 # define ISGRAPH(c) (isprint (c) && !isspace (c))
368 /* Solaris defines ISPRINT so we must undefine it first. */
370 # define ISPRINT(c) isprint (c)
371 # define ISDIGIT(c) isdigit (c)
372 # define ISALNUM(c) isalnum (c)
373 # define ISALPHA(c) isalpha (c)
374 # define ISCNTRL(c) iscntrl (c)
375 # define ISLOWER(c) islower (c)
376 # define ISPUNCT(c) ispunct (c)
377 # define ISSPACE(c) isspace (c)
378 # define ISUPPER(c) isupper (c)
379 # define ISXDIGIT(c) isxdigit (c)
381 # define ISWORD(c) ISALPHA (c)
384 # define TOLOWER(c) _tolower (c)
386 # define TOLOWER(c) tolower (c)
389 /* How many characters in the character set. */
390 # define CHAR_SET_SIZE 256
394 extern char *re_syntax_table
;
396 # else /* not SYNTAX_TABLE */
398 static char re_syntax_table
[CHAR_SET_SIZE
];
401 init_syntax_once (void)
409 memset (re_syntax_table
, 0, sizeof re_syntax_table
);
411 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
413 re_syntax_table
[c
] = Sword
;
415 re_syntax_table
['_'] = Ssymbol
;
420 # endif /* not SYNTAX_TABLE */
422 # define SYNTAX(c) re_syntax_table[(c)]
424 #endif /* not emacs */
426 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
428 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
429 use `alloca' instead of `malloc'. This is because using malloc in
430 re_search* or re_match* could cause memory leaks when C-g is used in
431 Emacs; also, malloc is slower and causes storage fragmentation. On
432 the other hand, malloc is more portable, and easier to debug.
434 Because we sometimes use alloca, some routines have to be macros,
435 not functions -- `alloca'-allocated space disappears at the end of the
436 function it is called in. */
440 # define REGEX_ALLOCATE malloc
441 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
442 # define REGEX_FREE free
444 #else /* not REGEX_MALLOC */
446 /* Emacs already defines alloca, sometimes. */
449 /* Make alloca work the best possible way. */
451 # define alloca __builtin_alloca
452 # else /* not __GNUC__ */
453 # ifdef HAVE_ALLOCA_H
455 # endif /* HAVE_ALLOCA_H */
456 # endif /* not __GNUC__ */
458 # endif /* not alloca */
461 # define REGEX_USE_SAFE_ALLOCA USE_SAFE_ALLOCA
462 # define REGEX_SAFE_FREE() SAFE_FREE ()
463 # define REGEX_ALLOCATE SAFE_ALLOCA
465 # define REGEX_ALLOCATE alloca
468 /* Assumes a `char *destination' variable. */
469 # define REGEX_REALLOCATE(source, osize, nsize) \
470 (destination = REGEX_ALLOCATE (nsize), \
471 memcpy (destination, source, osize))
473 /* No need to do anything to free, after alloca. */
474 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
476 #endif /* not REGEX_MALLOC */
478 #ifndef REGEX_USE_SAFE_ALLOCA
479 # define REGEX_USE_SAFE_ALLOCA ((void) 0)
480 # define REGEX_SAFE_FREE() ((void) 0)
483 /* Define how to allocate the failure stack. */
485 #if defined REL_ALLOC && defined REGEX_MALLOC
487 # define REGEX_ALLOCATE_STACK(size) \
488 r_alloc (&failure_stack_ptr, (size))
489 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
490 r_re_alloc (&failure_stack_ptr, (nsize))
491 # define REGEX_FREE_STACK(ptr) \
492 r_alloc_free (&failure_stack_ptr)
494 #else /* not using relocating allocator */
496 # define REGEX_ALLOCATE_STACK(size) REGEX_ALLOCATE (size)
497 # define REGEX_REALLOCATE_STACK(source, o, n) REGEX_REALLOCATE (source, o, n)
498 # define REGEX_FREE_STACK(ptr) REGEX_FREE (ptr)
500 #endif /* not using relocating allocator */
503 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
504 `string1' or just past its end. This works if PTR is NULL, which is
506 #define FIRST_STRING_P(ptr) \
507 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
509 /* (Re)Allocate N items of type T using malloc, or fail. */
510 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
511 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
512 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
514 #define BYTEWIDTH 8 /* In bits. */
516 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
521 # define max(a, b) ((a) > (b) ? (a) : (b))
522 # define min(a, b) ((a) < (b) ? (a) : (b))
525 /* Type of source-pattern and string chars. */
527 typedef unsigned char re_char
;
528 typedef const re_char const_re_char
;
530 typedef const unsigned char re_char
;
531 typedef re_char const_re_char
;
534 typedef char boolean
;
536 static regoff_t
re_match_2_internal (struct re_pattern_buffer
*bufp
,
537 re_char
*string1
, size_t size1
,
538 re_char
*string2
, size_t size2
,
540 struct re_registers
*regs
,
543 /* These are the command codes that appear in compiled regular
544 expressions. Some opcodes are followed by argument bytes. A
545 command code can specify any interpretation whatsoever for its
546 arguments. Zero bytes may appear in the compiled regular expression. */
552 /* Succeed right away--no more backtracking. */
555 /* Followed by one byte giving n, then by n literal bytes. */
558 /* Matches any (more or less) character. */
561 /* Matches any one char belonging to specified set. First
562 following byte is number of bitmap bytes. Then come bytes
563 for a bitmap saying which chars are in. Bits in each byte
564 are ordered low-bit-first. A character is in the set if its
565 bit is 1. A character too large to have a bit in the map is
566 automatically not in the set.
568 If the length byte has the 0x80 bit set, then that stuff
569 is followed by a range table:
570 2 bytes of flags for character sets (low 8 bits, high 8 bits)
571 See RANGE_TABLE_WORK_BITS below.
572 2 bytes, the number of pairs that follow (upto 32767)
573 pairs, each 2 multibyte characters,
574 each multibyte character represented as 3 bytes. */
577 /* Same parameters as charset, but match any character that is
578 not one of those specified. */
581 /* Start remembering the text that is matched, for storing in a
582 register. Followed by one byte with the register number, in
583 the range 0 to one less than the pattern buffer's re_nsub
587 /* Stop remembering the text that is matched and store it in a
588 memory register. Followed by one byte with the register
589 number, in the range 0 to one less than `re_nsub' in the
593 /* Match a duplicate of something remembered. Followed by one
594 byte containing the register number. */
597 /* Fail unless at beginning of line. */
600 /* Fail unless at end of line. */
603 /* Succeeds if at beginning of buffer (if emacs) or at beginning
604 of string to be matched (if not). */
607 /* Analogously, for end of buffer/string. */
610 /* Followed by two byte relative address to which to jump. */
613 /* Followed by two-byte relative address of place to resume at
614 in case of failure. */
617 /* Like on_failure_jump, but pushes a placeholder instead of the
618 current string position when executed. */
619 on_failure_keep_string_jump
,
621 /* Just like `on_failure_jump', except that it checks that we
622 don't get stuck in an infinite loop (matching an empty string
624 on_failure_jump_loop
,
626 /* Just like `on_failure_jump_loop', except that it checks for
627 a different kind of loop (the kind that shows up with non-greedy
628 operators). This operation has to be immediately preceded
630 on_failure_jump_nastyloop
,
632 /* A smart `on_failure_jump' used for greedy * and + operators.
633 It analyzes the loop before which it is put and if the
634 loop does not require backtracking, it changes itself to
635 `on_failure_keep_string_jump' and short-circuits the loop,
636 else it just defaults to changing itself into `on_failure_jump'.
637 It assumes that it is pointing to just past a `jump'. */
638 on_failure_jump_smart
,
640 /* Followed by two-byte relative address and two-byte number n.
641 After matching N times, jump to the address upon failure.
642 Does not work if N starts at 0: use on_failure_jump_loop
646 /* Followed by two-byte relative address, and two-byte number n.
647 Jump to the address N times, then fail. */
650 /* Set the following two-byte relative address to the
651 subsequent two-byte number. The address *includes* the two
655 wordbeg
, /* Succeeds if at word beginning. */
656 wordend
, /* Succeeds if at word end. */
658 wordbound
, /* Succeeds if at a word boundary. */
659 notwordbound
, /* Succeeds if not at a word boundary. */
661 symbeg
, /* Succeeds if at symbol beginning. */
662 symend
, /* Succeeds if at symbol end. */
664 /* Matches any character whose syntax is specified. Followed by
665 a byte which contains a syntax code, e.g., Sword. */
668 /* Matches any character whose syntax is not that specified. */
672 ,before_dot
, /* Succeeds if before point. */
673 at_dot
, /* Succeeds if at point. */
674 after_dot
, /* Succeeds if after point. */
676 /* Matches any character whose category-set contains the specified
677 category. The operator is followed by a byte which contains a
678 category code (mnemonic ASCII character). */
681 /* Matches any character whose category-set does not contain the
682 specified category. The operator is followed by a byte which
683 contains the category code (mnemonic ASCII character). */
688 /* Common operations on the compiled pattern. */
690 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
692 #define STORE_NUMBER(destination, number) \
694 (destination)[0] = (number) & 0377; \
695 (destination)[1] = (number) >> 8; \
698 /* Same as STORE_NUMBER, except increment DESTINATION to
699 the byte after where the number is stored. Therefore, DESTINATION
700 must be an lvalue. */
702 #define STORE_NUMBER_AND_INCR(destination, number) \
704 STORE_NUMBER (destination, number); \
705 (destination) += 2; \
708 /* Put into DESTINATION a number stored in two contiguous bytes starting
711 #define EXTRACT_NUMBER(destination, source) \
712 ((destination) = extract_number (source))
715 extract_number (re_char
*source
)
717 unsigned leading_byte
= SIGN_EXTEND_CHAR (source
[1]);
718 return (leading_byte
<< 8) + source
[0];
721 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
722 SOURCE must be an lvalue. */
724 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
725 ((destination) = extract_number_and_incr (&source))
728 extract_number_and_incr (re_char
**source
)
730 int num
= extract_number (*source
);
735 /* Store a multibyte character in three contiguous bytes starting
736 DESTINATION, and increment DESTINATION to the byte after where the
737 character is stored. Therefore, DESTINATION must be an lvalue. */
739 #define STORE_CHARACTER_AND_INCR(destination, character) \
741 (destination)[0] = (character) & 0377; \
742 (destination)[1] = ((character) >> 8) & 0377; \
743 (destination)[2] = (character) >> 16; \
744 (destination) += 3; \
747 /* Put into DESTINATION a character stored in three contiguous bytes
748 starting at SOURCE. */
750 #define EXTRACT_CHARACTER(destination, source) \
752 (destination) = ((source)[0] \
753 | ((source)[1] << 8) \
754 | ((source)[2] << 16)); \
758 /* Macros for charset. */
760 /* Size of bitmap of charset P in bytes. P is a start of charset,
761 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
762 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
764 /* Nonzero if charset P has range table. */
765 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
767 /* Return the address of range table of charset P. But not the start
768 of table itself, but the before where the number of ranges is
769 stored. `2 +' means to skip re_opcode_t and size of bitmap,
770 and the 2 bytes of flags at the start of the range table. */
771 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
774 /* Extract the bit flags that start a range table. */
775 #define CHARSET_RANGE_TABLE_BITS(p) \
776 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
777 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
780 /* Return the address of end of RANGE_TABLE. COUNT is number of
781 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
782 is start of range and end of range. `* 3' is size of each start
784 #define CHARSET_RANGE_TABLE_END(range_table, count) \
785 ((range_table) + (count) * 2 * 3)
787 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
788 COUNT is number of ranges in RANGE_TABLE. */
789 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
792 re_wchar_t range_start, range_end; \
794 re_char *range_table_end \
795 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
797 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
799 EXTRACT_CHARACTER (range_start, rtp); \
800 EXTRACT_CHARACTER (range_end, rtp + 3); \
802 if (range_start <= (c) && (c) <= range_end) \
811 /* Test if C is in range table of CHARSET. The flag NOT is negated if
812 C is listed in it. */
813 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
816 /* Number of ranges in range table. */ \
818 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
820 EXTRACT_NUMBER_AND_INCR (count, range_table); \
821 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
825 /* If DEBUG is defined, Regex prints many voluminous messages about what
826 it is doing (if the variable `debug' is nonzero). If linked with the
827 main program in `iregex.c', you can enter patterns and strings
828 interactively. And if linked with the main program in `main.c' and
829 the other test files, you can run the already-written tests. */
833 /* We use standard I/O for debugging. */
836 /* It is useful to test things that ``must'' be true when debugging. */
839 static int debug
= -100000;
841 # define DEBUG_STATEMENT(e) e
842 # define DEBUG_PRINT(...) if (debug > 0) printf (__VA_ARGS__)
843 # define DEBUG_COMPILES_ARGUMENTS
844 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
845 if (debug > 0) print_partial_compiled_pattern (s, e)
846 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
847 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
850 /* Print the fastmap in human-readable form. */
853 print_fastmap (char *fastmap
)
855 unsigned was_a_range
= 0;
858 while (i
< (1 << BYTEWIDTH
))
864 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
880 /* Print a compiled pattern string in human-readable form, starting at
881 the START pointer into it and ending just before the pointer END. */
884 print_partial_compiled_pattern (re_char
*start
, re_char
*end
)
892 fprintf (stderr
, "(null)\n");
896 /* Loop over pattern commands. */
899 fprintf (stderr
, "%td:\t", p
- start
);
901 switch ((re_opcode_t
) *p
++)
904 fprintf (stderr
, "/no_op");
908 fprintf (stderr
, "/succeed");
913 fprintf (stderr
, "/exactn/%d", mcnt
);
916 fprintf (stderr
, "/%c", *p
++);
922 fprintf (stderr
, "/start_memory/%d", *p
++);
926 fprintf (stderr
, "/stop_memory/%d", *p
++);
930 fprintf (stderr
, "/duplicate/%d", *p
++);
934 fprintf (stderr
, "/anychar");
940 register int c
, last
= -100;
941 register int in_range
= 0;
942 int length
= CHARSET_BITMAP_SIZE (p
- 1);
943 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
945 fprintf (stderr
, "/charset [%s",
946 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
949 fprintf (stderr
, " !extends past end of pattern! ");
951 for (c
= 0; c
< 256; c
++)
953 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
955 /* Are we starting a range? */
956 if (last
+ 1 == c
&& ! in_range
)
958 fprintf (stderr
, "-");
961 /* Have we broken a range? */
962 else if (last
+ 1 != c
&& in_range
)
964 fprintf (stderr
, "%c", last
);
969 fprintf (stderr
, "%c", c
);
975 fprintf (stderr
, "%c", last
);
977 fprintf (stderr
, "]");
984 fprintf (stderr
, "has-range-table");
986 /* ??? Should print the range table; for now, just skip it. */
987 p
+= 2; /* skip range table bits */
988 EXTRACT_NUMBER_AND_INCR (count
, p
);
989 p
= CHARSET_RANGE_TABLE_END (p
, count
);
995 fprintf (stderr
, "/begline");
999 fprintf (stderr
, "/endline");
1002 case on_failure_jump
:
1003 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1004 fprintf (stderr
, "/on_failure_jump to %td", p
+ mcnt
- start
);
1007 case on_failure_keep_string_jump
:
1008 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1009 fprintf (stderr
, "/on_failure_keep_string_jump to %td",
1013 case on_failure_jump_nastyloop
:
1014 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1015 fprintf (stderr
, "/on_failure_jump_nastyloop to %td",
1019 case on_failure_jump_loop
:
1020 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1021 fprintf (stderr
, "/on_failure_jump_loop to %td",
1025 case on_failure_jump_smart
:
1026 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1027 fprintf (stderr
, "/on_failure_jump_smart to %td",
1032 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1033 fprintf (stderr
, "/jump to %td", p
+ mcnt
- start
);
1037 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1038 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1039 fprintf (stderr
, "/succeed_n to %td, %d times",
1040 p
- 2 + mcnt
- start
, mcnt2
);
1044 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1045 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1046 fprintf (stderr
, "/jump_n to %td, %d times",
1047 p
- 2 + mcnt
- start
, mcnt2
);
1051 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1052 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1053 fprintf (stderr
, "/set_number_at location %td to %d",
1054 p
- 2 + mcnt
- start
, mcnt2
);
1058 fprintf (stderr
, "/wordbound");
1062 fprintf (stderr
, "/notwordbound");
1066 fprintf (stderr
, "/wordbeg");
1070 fprintf (stderr
, "/wordend");
1074 fprintf (stderr
, "/symbeg");
1078 fprintf (stderr
, "/symend");
1082 fprintf (stderr
, "/syntaxspec");
1084 fprintf (stderr
, "/%d", mcnt
);
1088 fprintf (stderr
, "/notsyntaxspec");
1090 fprintf (stderr
, "/%d", mcnt
);
1095 fprintf (stderr
, "/before_dot");
1099 fprintf (stderr
, "/at_dot");
1103 fprintf (stderr
, "/after_dot");
1107 fprintf (stderr
, "/categoryspec");
1109 fprintf (stderr
, "/%d", mcnt
);
1112 case notcategoryspec
:
1113 fprintf (stderr
, "/notcategoryspec");
1115 fprintf (stderr
, "/%d", mcnt
);
1120 fprintf (stderr
, "/begbuf");
1124 fprintf (stderr
, "/endbuf");
1128 fprintf (stderr
, "?%d", *(p
-1));
1131 fprintf (stderr
, "\n");
1134 fprintf (stderr
, "%td:\tend of pattern.\n", p
- start
);
1139 print_compiled_pattern (struct re_pattern_buffer
*bufp
)
1141 re_char
*buffer
= bufp
->buffer
;
1143 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1144 printf ("%ld bytes used/%ld bytes allocated.\n",
1145 bufp
->used
, bufp
->allocated
);
1147 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1149 printf ("fastmap: ");
1150 print_fastmap (bufp
->fastmap
);
1153 printf ("re_nsub: %zu\t", bufp
->re_nsub
);
1154 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1155 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1156 printf ("no_sub: %d\t", bufp
->no_sub
);
1157 printf ("not_bol: %d\t", bufp
->not_bol
);
1158 printf ("not_eol: %d\t", bufp
->not_eol
);
1159 printf ("syntax: %lx\n", bufp
->syntax
);
1161 /* Perhaps we should print the translate table? */
1166 print_double_string (re_char
*where
, re_char
*string1
, ssize_t size1
,
1167 re_char
*string2
, ssize_t size2
)
1175 if (FIRST_STRING_P (where
))
1177 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1178 putchar (string1
[this_char
]);
1183 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1184 putchar (string2
[this_char
]);
1188 #else /* not DEBUG */
1193 # define DEBUG_STATEMENT(e)
1194 # define DEBUG_PRINT(...)
1195 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1196 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1198 #endif /* not DEBUG */
1200 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1202 # define IF_LINT(Code) Code
1204 # define IF_LINT(Code) /* empty */
1207 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1208 also be assigned to arbitrarily: each pattern buffer stores its own
1209 syntax, so it can be changed between regex compilations. */
1210 /* This has no initializer because initialized variables in Emacs
1211 become read-only after dumping. */
1212 reg_syntax_t re_syntax_options
;
1215 /* Specify the precise syntax of regexps for compilation. This provides
1216 for compatibility for various utilities which historically have
1217 different, incompatible syntaxes.
1219 The argument SYNTAX is a bit mask comprised of the various bits
1220 defined in regex.h. We return the old syntax. */
1223 re_set_syntax (reg_syntax_t syntax
)
1225 reg_syntax_t ret
= re_syntax_options
;
1227 re_syntax_options
= syntax
;
1230 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1232 /* Regexp to use to replace spaces, or NULL meaning don't. */
1233 static const_re_char
*whitespace_regexp
;
1236 re_set_whitespace_regexp (const char *regexp
)
1238 whitespace_regexp
= (const_re_char
*) regexp
;
1240 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1242 /* This table gives an error message for each of the error codes listed
1243 in regex.h. Obviously the order here has to be same as there.
1244 POSIX doesn't require that we do anything for REG_NOERROR,
1245 but why not be nice? */
1247 static const char *re_error_msgid
[] =
1249 gettext_noop ("Success"), /* REG_NOERROR */
1250 gettext_noop ("No match"), /* REG_NOMATCH */
1251 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1252 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1253 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1254 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1255 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1256 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1257 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1258 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1259 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1260 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1261 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1262 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1263 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1264 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1265 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1266 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1269 /* Avoiding alloca during matching, to placate r_alloc. */
1271 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1272 searching and matching functions should not call alloca. On some
1273 systems, alloca is implemented in terms of malloc, and if we're
1274 using the relocating allocator routines, then malloc could cause a
1275 relocation, which might (if the strings being searched are in the
1276 ralloc heap) shift the data out from underneath the regexp
1279 Here's another reason to avoid allocation: Emacs
1280 processes input from X in a signal handler; processing X input may
1281 call malloc; if input arrives while a matching routine is calling
1282 malloc, then we're scrod. But Emacs can't just block input while
1283 calling matching routines; then we don't notice interrupts when
1284 they come in. So, Emacs blocks input around all regexp calls
1285 except the matching calls, which it leaves unprotected, in the
1286 faith that they will not malloc. */
1288 /* Normally, this is fine. */
1289 #define MATCH_MAY_ALLOCATE
1291 /* The match routines may not allocate if (1) they would do it with malloc
1292 and (2) it's not safe for them to use malloc.
1293 Note that if REL_ALLOC is defined, matching would not use malloc for the
1294 failure stack, but we would still use it for the register vectors;
1295 so REL_ALLOC should not affect this. */
1296 #if defined REGEX_MALLOC && defined emacs
1297 # undef MATCH_MAY_ALLOCATE
1301 /* Failure stack declarations and macros; both re_compile_fastmap and
1302 re_match_2 use a failure stack. These have to be macros because of
1303 REGEX_ALLOCATE_STACK. */
1306 /* Approximate number of failure points for which to initially allocate space
1307 when matching. If this number is exceeded, we allocate more
1308 space, so it is not a hard limit. */
1309 #ifndef INIT_FAILURE_ALLOC
1310 # define INIT_FAILURE_ALLOC 20
1313 /* Roughly the maximum number of failure points on the stack. Would be
1314 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1315 This is a variable only so users of regex can assign to it; we never
1316 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1317 before using it, so it should probably be a byte-count instead. */
1318 # if defined MATCH_MAY_ALLOCATE
1319 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1320 whose default stack limit is 2mb. In order for a larger
1321 value to work reliably, you have to try to make it accord
1322 with the process stack limit. */
1323 size_t re_max_failures
= 40000;
1325 size_t re_max_failures
= 4000;
1328 union fail_stack_elt
1331 /* This should be the biggest `int' that's no bigger than a pointer. */
1335 typedef union fail_stack_elt fail_stack_elt_t
;
1339 fail_stack_elt_t
*stack
;
1341 size_t avail
; /* Offset of next open position. */
1342 size_t frame
; /* Offset of the cur constructed frame. */
1345 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1348 /* Define macros to initialize and free the failure stack.
1349 Do `return -2' if the alloc fails. */
1351 #ifdef MATCH_MAY_ALLOCATE
1352 # define INIT_FAIL_STACK() \
1354 fail_stack.stack = \
1355 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1356 * sizeof (fail_stack_elt_t)); \
1358 if (fail_stack.stack == NULL) \
1361 fail_stack.size = INIT_FAILURE_ALLOC; \
1362 fail_stack.avail = 0; \
1363 fail_stack.frame = 0; \
1366 # define INIT_FAIL_STACK() \
1368 fail_stack.avail = 0; \
1369 fail_stack.frame = 0; \
1372 # define RETALLOC_IF(addr, n, t) \
1373 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1377 /* Double the size of FAIL_STACK, up to a limit
1378 which allows approximately `re_max_failures' items.
1380 Return 1 if succeeds, and 0 if either ran out of memory
1381 allocating space for it or it was already too large.
1383 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1385 /* Factor to increase the failure stack size by
1386 when we increase it.
1387 This used to be 2, but 2 was too wasteful
1388 because the old discarded stacks added up to as much space
1389 were as ultimate, maximum-size stack. */
1390 #define FAIL_STACK_GROWTH_FACTOR 4
1392 #define GROW_FAIL_STACK(fail_stack) \
1393 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1394 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1396 : ((fail_stack).stack \
1397 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1398 (fail_stack).size * sizeof (fail_stack_elt_t), \
1399 min (re_max_failures * TYPICAL_FAILURE_SIZE, \
1400 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1401 * FAIL_STACK_GROWTH_FACTOR))), \
1403 (fail_stack).stack == NULL \
1405 : ((fail_stack).size \
1406 = (min (re_max_failures * TYPICAL_FAILURE_SIZE, \
1407 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1408 * FAIL_STACK_GROWTH_FACTOR)) \
1409 / sizeof (fail_stack_elt_t)), \
1413 /* Push a pointer value onto the failure stack.
1414 Assumes the variable `fail_stack'. Probably should only
1415 be called from within `PUSH_FAILURE_POINT'. */
1416 #define PUSH_FAILURE_POINTER(item) \
1417 fail_stack.stack[fail_stack.avail++].pointer = (item)
1419 /* This pushes an integer-valued item onto the failure stack.
1420 Assumes the variable `fail_stack'. Probably should only
1421 be called from within `PUSH_FAILURE_POINT'. */
1422 #define PUSH_FAILURE_INT(item) \
1423 fail_stack.stack[fail_stack.avail++].integer = (item)
1425 /* These POP... operations complement the PUSH... operations.
1426 All assume that `fail_stack' is nonempty. */
1427 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1428 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1430 /* Individual items aside from the registers. */
1431 #define NUM_NONREG_ITEMS 3
1433 /* Used to examine the stack (to detect infinite loops). */
1434 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1435 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1436 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1437 #define TOP_FAILURE_HANDLE() fail_stack.frame
1440 #define ENSURE_FAIL_STACK(space) \
1441 while (REMAINING_AVAIL_SLOTS <= space) { \
1442 if (!GROW_FAIL_STACK (fail_stack)) \
1444 DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\
1445 DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\
1448 /* Push register NUM onto the stack. */
1449 #define PUSH_FAILURE_REG(num) \
1451 char *destination; \
1453 ENSURE_FAIL_STACK(3); \
1454 DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \
1455 n, regstart[n], regend[n]); \
1456 PUSH_FAILURE_POINTER (regstart[n]); \
1457 PUSH_FAILURE_POINTER (regend[n]); \
1458 PUSH_FAILURE_INT (n); \
1461 /* Change the counter's value to VAL, but make sure that it will
1462 be reset when backtracking. */
1463 #define PUSH_NUMBER(ptr,val) \
1465 char *destination; \
1467 ENSURE_FAIL_STACK(3); \
1468 EXTRACT_NUMBER (c, ptr); \
1469 DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \
1470 PUSH_FAILURE_INT (c); \
1471 PUSH_FAILURE_POINTER (ptr); \
1472 PUSH_FAILURE_INT (-1); \
1473 STORE_NUMBER (ptr, val); \
1476 /* Pop a saved register off the stack. */
1477 #define POP_FAILURE_REG_OR_COUNT() \
1479 long pfreg = POP_FAILURE_INT (); \
1482 /* It's a counter. */ \
1483 /* Here, we discard `const', making re_match non-reentrant. */ \
1484 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1485 pfreg = POP_FAILURE_INT (); \
1486 STORE_NUMBER (ptr, pfreg); \
1487 DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \
1491 regend[pfreg] = POP_FAILURE_POINTER (); \
1492 regstart[pfreg] = POP_FAILURE_POINTER (); \
1493 DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \
1494 pfreg, regstart[pfreg], regend[pfreg]); \
1498 /* Check that we are not stuck in an infinite loop. */
1499 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1501 ssize_t failure = TOP_FAILURE_HANDLE (); \
1502 /* Check for infinite matching loops */ \
1503 while (failure > 0 \
1504 && (FAILURE_STR (failure) == string_place \
1505 || FAILURE_STR (failure) == NULL)) \
1507 assert (FAILURE_PAT (failure) >= bufp->buffer \
1508 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1509 if (FAILURE_PAT (failure) == pat_cur) \
1514 DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1515 failure = NEXT_FAILURE_HANDLE(failure); \
1517 DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \
1520 /* Push the information about the state we will need
1521 if we ever fail back to it.
1523 Requires variables fail_stack, regstart, regend and
1524 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1527 Does `return FAILURE_CODE' if runs out of memory. */
1529 #define PUSH_FAILURE_POINT(pattern, string_place) \
1531 char *destination; \
1532 /* Must be int, so when we don't save any registers, the arithmetic \
1533 of 0 + -1 isn't done as unsigned. */ \
1535 DEBUG_STATEMENT (nfailure_points_pushed++); \
1536 DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \
1537 DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \
1538 DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\
1540 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1542 DEBUG_PRINT ("\n"); \
1544 DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \
1545 PUSH_FAILURE_INT (fail_stack.frame); \
1547 DEBUG_PRINT (" Push string %p: `", string_place); \
1548 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1549 DEBUG_PRINT ("'\n"); \
1550 PUSH_FAILURE_POINTER (string_place); \
1552 DEBUG_PRINT (" Push pattern %p: ", pattern); \
1553 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1554 PUSH_FAILURE_POINTER (pattern); \
1556 /* Close the frame by moving the frame pointer past it. */ \
1557 fail_stack.frame = fail_stack.avail; \
1560 /* Estimate the size of data pushed by a typical failure stack entry.
1561 An estimate is all we need, because all we use this for
1562 is to choose a limit for how big to make the failure stack. */
1563 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1564 #define TYPICAL_FAILURE_SIZE 20
1566 /* How many items can still be added to the stack without overflowing it. */
1567 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1570 /* Pops what PUSH_FAIL_STACK pushes.
1572 We restore into the parameters, all of which should be lvalues:
1573 STR -- the saved data position.
1574 PAT -- the saved pattern position.
1575 REGSTART, REGEND -- arrays of string positions.
1577 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1578 `pend', `string1', `size1', `string2', and `size2'. */
1580 #define POP_FAILURE_POINT(str, pat) \
1582 assert (!FAIL_STACK_EMPTY ()); \
1584 /* Remove failure points and point to how many regs pushed. */ \
1585 DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \
1586 DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \
1587 DEBUG_PRINT (" size: %zd\n", fail_stack.size); \
1589 /* Pop the saved registers. */ \
1590 while (fail_stack.frame < fail_stack.avail) \
1591 POP_FAILURE_REG_OR_COUNT (); \
1593 pat = POP_FAILURE_POINTER (); \
1594 DEBUG_PRINT (" Popping pattern %p: ", pat); \
1595 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1597 /* If the saved string location is NULL, it came from an \
1598 on_failure_keep_string_jump opcode, and we want to throw away the \
1599 saved NULL, thus retaining our current position in the string. */ \
1600 str = POP_FAILURE_POINTER (); \
1601 DEBUG_PRINT (" Popping string %p: `", str); \
1602 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1603 DEBUG_PRINT ("'\n"); \
1605 fail_stack.frame = POP_FAILURE_INT (); \
1606 DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \
1608 assert (fail_stack.avail >= 0); \
1609 assert (fail_stack.frame <= fail_stack.avail); \
1611 DEBUG_STATEMENT (nfailure_points_popped++); \
1612 } while (0) /* POP_FAILURE_POINT */
1616 /* Registers are set to a sentinel when they haven't yet matched. */
1617 #define REG_UNSET(e) ((e) == NULL)
1619 /* Subroutine declarations and macros for regex_compile. */
1621 static reg_errcode_t
regex_compile (re_char
*pattern
, size_t size
,
1622 reg_syntax_t syntax
,
1623 struct re_pattern_buffer
*bufp
);
1624 static void store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
);
1625 static void store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
);
1626 static void insert_op1 (re_opcode_t op
, unsigned char *loc
,
1627 int arg
, unsigned char *end
);
1628 static void insert_op2 (re_opcode_t op
, unsigned char *loc
,
1629 int arg1
, int arg2
, unsigned char *end
);
1630 static boolean
at_begline_loc_p (re_char
*pattern
, re_char
*p
,
1631 reg_syntax_t syntax
);
1632 static boolean
at_endline_loc_p (re_char
*p
, re_char
*pend
,
1633 reg_syntax_t syntax
);
1634 static re_char
*skip_one_char (re_char
*p
);
1635 static int analyze_first (re_char
*p
, re_char
*pend
,
1636 char *fastmap
, const int multibyte
);
1638 /* Fetch the next character in the uncompiled pattern, with no
1640 #define PATFETCH(c) \
1643 if (p == pend) return REG_EEND; \
1644 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1649 /* If `translate' is non-null, return translate[D], else just D. We
1650 cast the subscript to translate because some data is declared as
1651 `char *', to avoid warnings when a string constant is passed. But
1652 when we use a character as a subscript we must make it unsigned. */
1654 # define TRANSLATE(d) \
1655 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1659 /* Macros for outputting the compiled pattern into `buffer'. */
1661 /* If the buffer isn't allocated when it comes in, use this. */
1662 #define INIT_BUF_SIZE 32
1664 /* Make sure we have at least N more bytes of space in buffer. */
1665 #define GET_BUFFER_SPACE(n) \
1666 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1669 /* Make sure we have one more byte of buffer space and then add C to it. */
1670 #define BUF_PUSH(c) \
1672 GET_BUFFER_SPACE (1); \
1673 *b++ = (unsigned char) (c); \
1677 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1678 #define BUF_PUSH_2(c1, c2) \
1680 GET_BUFFER_SPACE (2); \
1681 *b++ = (unsigned char) (c1); \
1682 *b++ = (unsigned char) (c2); \
1686 /* Store a jump with opcode OP at LOC to location TO. We store a
1687 relative address offset by the three bytes the jump itself occupies. */
1688 #define STORE_JUMP(op, loc, to) \
1689 store_op1 (op, loc, (to) - (loc) - 3)
1691 /* Likewise, for a two-argument jump. */
1692 #define STORE_JUMP2(op, loc, to, arg) \
1693 store_op2 (op, loc, (to) - (loc) - 3, arg)
1695 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1696 #define INSERT_JUMP(op, loc, to) \
1697 insert_op1 (op, loc, (to) - (loc) - 3, b)
1699 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1700 #define INSERT_JUMP2(op, loc, to, arg) \
1701 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1704 /* This is not an arbitrary limit: the arguments which represent offsets
1705 into the pattern are two bytes long. So if 2^15 bytes turns out to
1706 be too small, many things would have to change. */
1707 # define MAX_BUF_SIZE (1L << 15)
1709 /* Extend the buffer by twice its current size via realloc and
1710 reset the pointers that pointed into the old block to point to the
1711 correct places in the new one. If extending the buffer results in it
1712 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1713 #if __BOUNDED_POINTERS__
1714 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1715 # define MOVE_BUFFER_POINTER(P) \
1716 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1717 SET_HIGH_BOUND (P), \
1718 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1719 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1722 SET_HIGH_BOUND (b); \
1723 SET_HIGH_BOUND (begalt); \
1724 if (fixup_alt_jump) \
1725 SET_HIGH_BOUND (fixup_alt_jump); \
1727 SET_HIGH_BOUND (laststart); \
1728 if (pending_exact) \
1729 SET_HIGH_BOUND (pending_exact); \
1732 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1733 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1735 #define EXTEND_BUFFER() \
1737 unsigned char *old_buffer = bufp->buffer; \
1738 if (bufp->allocated == MAX_BUF_SIZE) \
1740 bufp->allocated <<= 1; \
1741 if (bufp->allocated > MAX_BUF_SIZE) \
1742 bufp->allocated = MAX_BUF_SIZE; \
1743 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1744 if (bufp->buffer == NULL) \
1745 return REG_ESPACE; \
1746 /* If the buffer moved, move all the pointers into it. */ \
1747 if (old_buffer != bufp->buffer) \
1749 unsigned char *new_buffer = bufp->buffer; \
1750 MOVE_BUFFER_POINTER (b); \
1751 MOVE_BUFFER_POINTER (begalt); \
1752 if (fixup_alt_jump) \
1753 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1755 MOVE_BUFFER_POINTER (laststart); \
1756 if (pending_exact) \
1757 MOVE_BUFFER_POINTER (pending_exact); \
1759 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1763 /* Since we have one byte reserved for the register number argument to
1764 {start,stop}_memory, the maximum number of groups we can report
1765 things about is what fits in that byte. */
1766 #define MAX_REGNUM 255
1768 /* But patterns can have more than `MAX_REGNUM' registers. We just
1769 ignore the excess. */
1770 typedef int regnum_t
;
1773 /* Macros for the compile stack. */
1775 /* Since offsets can go either forwards or backwards, this type needs to
1776 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1777 /* int may be not enough when sizeof(int) == 2. */
1778 typedef long pattern_offset_t
;
1782 pattern_offset_t begalt_offset
;
1783 pattern_offset_t fixup_alt_jump
;
1784 pattern_offset_t laststart_offset
;
1786 } compile_stack_elt_t
;
1791 compile_stack_elt_t
*stack
;
1793 size_t avail
; /* Offset of next open position. */
1794 } compile_stack_type
;
1797 #define INIT_COMPILE_STACK_SIZE 32
1799 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1800 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1802 /* The next available element. */
1803 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1805 /* Explicit quit checking is needed for Emacs, which uses polling to
1806 process input events. */
1808 # define IMMEDIATE_QUIT_CHECK \
1810 if (immediate_quit) QUIT; \
1813 # define IMMEDIATE_QUIT_CHECK ((void)0)
1816 /* Structure to manage work area for range table. */
1817 struct range_table_work_area
1819 int *table
; /* actual work area. */
1820 int allocated
; /* allocated size for work area in bytes. */
1821 int used
; /* actually used size in words. */
1822 int bits
; /* flag to record character classes */
1827 /* Make sure that WORK_AREA can hold more N multibyte characters.
1828 This is used only in set_image_of_range and set_image_of_range_1.
1829 It expects WORK_AREA to be a pointer.
1830 If it can't get the space, it returns from the surrounding function. */
1832 #define EXTEND_RANGE_TABLE(work_area, n) \
1834 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1836 extend_range_table_work_area (&work_area); \
1837 if ((work_area).table == 0) \
1838 return (REG_ESPACE); \
1842 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1843 (work_area).bits |= (bit)
1845 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1846 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1848 EXTEND_RANGE_TABLE ((work_area), 2); \
1849 (work_area).table[(work_area).used++] = (range_start); \
1850 (work_area).table[(work_area).used++] = (range_end); \
1855 /* Free allocated memory for WORK_AREA. */
1856 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1858 if ((work_area).table) \
1859 free ((work_area).table); \
1862 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1863 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1864 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1865 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1867 /* Bits used to implement the multibyte-part of the various character classes
1868 such as [:alnum:] in a charset's range table. The code currently assumes
1869 that only the low 16 bits are used. */
1870 #define BIT_WORD 0x1
1871 #define BIT_LOWER 0x2
1872 #define BIT_PUNCT 0x4
1873 #define BIT_SPACE 0x8
1874 #define BIT_UPPER 0x10
1875 #define BIT_MULTIBYTE 0x20
1876 #define BIT_ALPHA 0x40
1877 #define BIT_ALNUM 0x80
1878 #define BIT_PRINT 0x100
1881 /* Set the bit for character C in a list. */
1882 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1887 /* Store characters in the range FROM to TO in the bitmap at B (for
1888 ASCII and unibyte characters) and WORK_AREA (for multibyte
1889 characters) while translating them and paying attention to the
1890 continuity of translated characters.
1892 Implementation note: It is better to implement these fairly big
1893 macros by a function, but it's not that easy because macros called
1894 in this macro assume various local variables already declared. */
1896 /* Both FROM and TO are ASCII characters. */
1898 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1902 for (C0 = (FROM); C0 <= (TO); C0++) \
1904 C1 = TRANSLATE (C0); \
1905 if (! ASCII_CHAR_P (C1)) \
1907 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1908 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1911 SET_LIST_BIT (C1); \
1916 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1918 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1920 int C0, C1, C2, I; \
1921 int USED = RANGE_TABLE_WORK_USED (work_area); \
1923 for (C0 = (FROM); C0 <= (TO); C0++) \
1925 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1926 if (CHAR_BYTE8_P (C1)) \
1927 SET_LIST_BIT (C0); \
1930 C2 = TRANSLATE (C1); \
1932 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1934 SET_LIST_BIT (C1); \
1935 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1937 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1938 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1940 if (C2 >= from - 1 && C2 <= to + 1) \
1942 if (C2 == from - 1) \
1943 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1944 else if (C2 == to + 1) \
1945 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1950 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1956 /* Both FROM and TO are multibyte characters. */
1958 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1960 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1962 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1963 for (C0 = (FROM); C0 <= (TO); C0++) \
1965 C1 = TRANSLATE (C0); \
1966 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1967 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1968 SET_LIST_BIT (C2); \
1969 if (C1 >= (FROM) && C1 <= (TO)) \
1971 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1973 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1974 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1976 if (C1 >= from - 1 && C1 <= to + 1) \
1978 if (C1 == from - 1) \
1979 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1980 else if (C1 == to + 1) \
1981 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1986 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1992 /* Get the next unsigned number in the uncompiled pattern. */
1993 #define GET_INTERVAL_COUNT(num) \
1996 FREE_STACK_RETURN (REG_EBRACE); \
2000 while ('0' <= c && c <= '9') \
2004 if (RE_DUP_MAX / 10 - (RE_DUP_MAX % 10 < c - '0') < num) \
2005 FREE_STACK_RETURN (REG_BADBR); \
2006 num = num * 10 + c - '0'; \
2008 FREE_STACK_RETURN (REG_EBRACE); \
2014 #if ! WIDE_CHAR_SUPPORT
2016 /* Map a string to the char class it names (if any). */
2018 re_wctype (const_re_char
*str
)
2020 const char *string
= (const char *) str
;
2021 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2022 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2023 else if (STREQ (string
, "word")) return RECC_WORD
;
2024 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2025 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2026 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2027 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2028 else if (STREQ (string
, "print")) return RECC_PRINT
;
2029 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2030 else if (STREQ (string
, "space")) return RECC_SPACE
;
2031 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2032 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2033 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2034 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2035 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2036 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2037 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2041 /* True if CH is in the char class CC. */
2043 re_iswctype (int ch
, re_wctype_t cc
)
2047 case RECC_ALNUM
: return ISALNUM (ch
) != 0;
2048 case RECC_ALPHA
: return ISALPHA (ch
) != 0;
2049 case RECC_BLANK
: return ISBLANK (ch
) != 0;
2050 case RECC_CNTRL
: return ISCNTRL (ch
) != 0;
2051 case RECC_DIGIT
: return ISDIGIT (ch
) != 0;
2052 case RECC_GRAPH
: return ISGRAPH (ch
) != 0;
2053 case RECC_LOWER
: return ISLOWER (ch
) != 0;
2054 case RECC_PRINT
: return ISPRINT (ch
) != 0;
2055 case RECC_PUNCT
: return ISPUNCT (ch
) != 0;
2056 case RECC_SPACE
: return ISSPACE (ch
) != 0;
2057 case RECC_UPPER
: return ISUPPER (ch
) != 0;
2058 case RECC_XDIGIT
: return ISXDIGIT (ch
) != 0;
2059 case RECC_ASCII
: return IS_REAL_ASCII (ch
) != 0;
2060 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2061 case RECC_UNIBYTE
: return ISUNIBYTE (ch
) != 0;
2062 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2063 case RECC_WORD
: return ISWORD (ch
) != 0;
2064 case RECC_ERROR
: return false;
2070 /* Return a bit-pattern to use in the range-table bits to match multibyte
2071 chars of class CC. */
2073 re_wctype_to_bit (re_wctype_t cc
)
2077 case RECC_NONASCII
: case RECC_GRAPH
:
2078 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2079 case RECC_ALPHA
: return BIT_ALPHA
;
2080 case RECC_ALNUM
: return BIT_ALNUM
;
2081 case RECC_WORD
: return BIT_WORD
;
2082 case RECC_LOWER
: return BIT_LOWER
;
2083 case RECC_UPPER
: return BIT_UPPER
;
2084 case RECC_PUNCT
: return BIT_PUNCT
;
2085 case RECC_SPACE
: return BIT_SPACE
;
2086 case RECC_PRINT
: return BIT_PRINT
;
2087 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2088 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2095 /* Filling in the work area of a range. */
2097 /* Actually extend the space in WORK_AREA. */
2100 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2102 work_area
->allocated
+= 16 * sizeof (int);
2103 work_area
->table
= realloc (work_area
->table
, work_area
->allocated
);
2109 /* Carefully find the ranges of codes that are equivalent
2110 under case conversion to the range start..end when passed through
2111 TRANSLATE. Handle the case where non-letters can come in between
2112 two upper-case letters (which happens in Latin-1).
2113 Also handle the case of groups of more than 2 case-equivalent chars.
2115 The basic method is to look at consecutive characters and see
2116 if they can form a run that can be handled as one.
2118 Returns -1 if successful, REG_ESPACE if ran out of space. */
2121 set_image_of_range_1 (struct range_table_work_area
*work_area
,
2122 re_wchar_t start
, re_wchar_t end
,
2123 RE_TRANSLATE_TYPE translate
)
2125 /* `one_case' indicates a character, or a run of characters,
2126 each of which is an isolate (no case-equivalents).
2127 This includes all ASCII non-letters.
2129 `two_case' indicates a character, or a run of characters,
2130 each of which has two case-equivalent forms.
2131 This includes all ASCII letters.
2133 `strange' indicates a character that has more than one
2136 enum case_type
{one_case
, two_case
, strange
};
2138 /* Describe the run that is in progress,
2139 which the next character can try to extend.
2140 If run_type is strange, that means there really is no run.
2141 If run_type is one_case, then run_start...run_end is the run.
2142 If run_type is two_case, then the run is run_start...run_end,
2143 and the case-equivalents end at run_eqv_end. */
2145 enum case_type run_type
= strange
;
2146 int run_start
, run_end
, run_eqv_end
;
2148 Lisp_Object eqv_table
;
2150 if (!RE_TRANSLATE_P (translate
))
2152 EXTEND_RANGE_TABLE (work_area
, 2);
2153 work_area
->table
[work_area
->used
++] = (start
);
2154 work_area
->table
[work_area
->used
++] = (end
);
2158 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2160 for (; start
<= end
; start
++)
2162 enum case_type this_type
;
2163 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2164 int minchar
, maxchar
;
2166 /* Classify this character */
2168 this_type
= one_case
;
2169 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2170 this_type
= two_case
;
2172 this_type
= strange
;
2175 minchar
= start
, maxchar
= eqv
;
2177 minchar
= eqv
, maxchar
= start
;
2179 /* Can this character extend the run in progress? */
2180 if (this_type
== strange
|| this_type
!= run_type
2181 || !(minchar
== run_end
+ 1
2182 && (run_type
== two_case
2183 ? maxchar
== run_eqv_end
+ 1 : 1)))
2186 Record each of its equivalent ranges. */
2187 if (run_type
== one_case
)
2189 EXTEND_RANGE_TABLE (work_area
, 2);
2190 work_area
->table
[work_area
->used
++] = run_start
;
2191 work_area
->table
[work_area
->used
++] = run_end
;
2193 else if (run_type
== two_case
)
2195 EXTEND_RANGE_TABLE (work_area
, 4);
2196 work_area
->table
[work_area
->used
++] = run_start
;
2197 work_area
->table
[work_area
->used
++] = run_end
;
2198 work_area
->table
[work_area
->used
++]
2199 = RE_TRANSLATE (eqv_table
, run_start
);
2200 work_area
->table
[work_area
->used
++]
2201 = RE_TRANSLATE (eqv_table
, run_end
);
2206 if (this_type
== strange
)
2208 /* For a strange character, add each of its equivalents, one
2209 by one. Don't start a range. */
2212 EXTEND_RANGE_TABLE (work_area
, 2);
2213 work_area
->table
[work_area
->used
++] = eqv
;
2214 work_area
->table
[work_area
->used
++] = eqv
;
2215 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2217 while (eqv
!= start
);
2220 /* Add this char to the run, or start a new run. */
2221 else if (run_type
== strange
)
2223 /* Initialize a new range. */
2224 run_type
= this_type
;
2227 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2231 /* Extend a running range. */
2233 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2237 /* If a run is still in progress at the end, finish it now
2238 by recording its equivalent ranges. */
2239 if (run_type
== one_case
)
2241 EXTEND_RANGE_TABLE (work_area
, 2);
2242 work_area
->table
[work_area
->used
++] = run_start
;
2243 work_area
->table
[work_area
->used
++] = run_end
;
2245 else if (run_type
== two_case
)
2247 EXTEND_RANGE_TABLE (work_area
, 4);
2248 work_area
->table
[work_area
->used
++] = run_start
;
2249 work_area
->table
[work_area
->used
++] = run_end
;
2250 work_area
->table
[work_area
->used
++]
2251 = RE_TRANSLATE (eqv_table
, run_start
);
2252 work_area
->table
[work_area
->used
++]
2253 = RE_TRANSLATE (eqv_table
, run_end
);
2261 /* Record the image of the range start..end when passed through
2262 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2263 and is not even necessarily contiguous.
2264 Normally we approximate it with the smallest contiguous range that contains
2265 all the chars we need. However, for Latin-1 we go to extra effort
2268 This function is not called for ASCII ranges.
2270 Returns -1 if successful, REG_ESPACE if ran out of space. */
2273 set_image_of_range (struct range_table_work_area
*work_area
,
2274 re_wchar_t start
, re_wchar_t end
,
2275 RE_TRANSLATE_TYPE translate
)
2277 re_wchar_t cmin
, cmax
;
2280 /* For Latin-1 ranges, use set_image_of_range_1
2281 to get proper handling of ranges that include letters and nonletters.
2282 For a range that includes the whole of Latin-1, this is not necessary.
2283 For other character sets, we don't bother to get this right. */
2284 if (RE_TRANSLATE_P (translate
) && start
< 04400
2285 && !(start
< 04200 && end
>= 04377))
2292 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2302 EXTEND_RANGE_TABLE (work_area
, 2);
2303 work_area
->table
[work_area
->used
++] = (start
);
2304 work_area
->table
[work_area
->used
++] = (end
);
2306 cmin
= -1, cmax
= -1;
2308 if (RE_TRANSLATE_P (translate
))
2312 for (ch
= start
; ch
<= end
; ch
++)
2314 re_wchar_t c
= TRANSLATE (ch
);
2315 if (! (start
<= c
&& c
<= end
))
2321 cmin
= min (cmin
, c
);
2322 cmax
= max (cmax
, c
);
2329 EXTEND_RANGE_TABLE (work_area
, 2);
2330 work_area
->table
[work_area
->used
++] = (cmin
);
2331 work_area
->table
[work_area
->used
++] = (cmax
);
2339 #ifndef MATCH_MAY_ALLOCATE
2341 /* If we cannot allocate large objects within re_match_2_internal,
2342 we make the fail stack and register vectors global.
2343 The fail stack, we grow to the maximum size when a regexp
2345 The register vectors, we adjust in size each time we
2346 compile a regexp, according to the number of registers it needs. */
2348 static fail_stack_type fail_stack
;
2350 /* Size with which the following vectors are currently allocated.
2351 That is so we can make them bigger as needed,
2352 but never make them smaller. */
2353 static int regs_allocated_size
;
2355 static re_char
** regstart
, ** regend
;
2356 static re_char
**best_regstart
, **best_regend
;
2358 /* Make the register vectors big enough for NUM_REGS registers,
2359 but don't make them smaller. */
2362 regex_grow_registers (int num_regs
)
2364 if (num_regs
> regs_allocated_size
)
2366 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2367 RETALLOC_IF (regend
, num_regs
, re_char
*);
2368 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2369 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2371 regs_allocated_size
= num_regs
;
2375 #endif /* not MATCH_MAY_ALLOCATE */
2377 static boolean
group_in_compile_stack (compile_stack_type compile_stack
,
2380 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2381 Returns one of error codes defined in `regex.h', or zero for success.
2383 Assumes the `allocated' (and perhaps `buffer') and `translate'
2384 fields are set in BUFP on entry.
2386 If it succeeds, results are put in BUFP (if it returns an error, the
2387 contents of BUFP are undefined):
2388 `buffer' is the compiled pattern;
2389 `syntax' is set to SYNTAX;
2390 `used' is set to the length of the compiled pattern;
2391 `fastmap_accurate' is zero;
2392 `re_nsub' is the number of subexpressions in PATTERN;
2393 `not_bol' and `not_eol' are zero;
2395 The `fastmap' field is neither examined nor set. */
2397 /* Insert the `jump' from the end of last alternative to "here".
2398 The space for the jump has already been allocated. */
2399 #define FIXUP_ALT_JUMP() \
2401 if (fixup_alt_jump) \
2402 STORE_JUMP (jump, fixup_alt_jump, b); \
2406 /* Return, freeing storage we allocated. */
2407 #define FREE_STACK_RETURN(value) \
2409 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2410 free (compile_stack.stack); \
2414 static reg_errcode_t
2415 regex_compile (const_re_char
*pattern
, size_t size
, reg_syntax_t syntax
,
2416 struct re_pattern_buffer
*bufp
)
2418 /* We fetch characters from PATTERN here. */
2419 register re_wchar_t c
, c1
;
2421 /* Points to the end of the buffer, where we should append. */
2422 register unsigned char *b
;
2424 /* Keeps track of unclosed groups. */
2425 compile_stack_type compile_stack
;
2427 /* Points to the current (ending) position in the pattern. */
2429 /* `const' makes AIX compiler fail. */
2430 unsigned char *p
= pattern
;
2432 re_char
*p
= pattern
;
2434 re_char
*pend
= pattern
+ size
;
2436 /* How to translate the characters in the pattern. */
2437 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2439 /* Address of the count-byte of the most recently inserted `exactn'
2440 command. This makes it possible to tell if a new exact-match
2441 character can be added to that command or if the character requires
2442 a new `exactn' command. */
2443 unsigned char *pending_exact
= 0;
2445 /* Address of start of the most recently finished expression.
2446 This tells, e.g., postfix * where to find the start of its
2447 operand. Reset at the beginning of groups and alternatives. */
2448 unsigned char *laststart
= 0;
2450 /* Address of beginning of regexp, or inside of last group. */
2451 unsigned char *begalt
;
2453 /* Place in the uncompiled pattern (i.e., the {) to
2454 which to go back if the interval is invalid. */
2455 re_char
*beg_interval
;
2457 /* Address of the place where a forward jump should go to the end of
2458 the containing expression. Each alternative of an `or' -- except the
2459 last -- ends with a forward jump of this sort. */
2460 unsigned char *fixup_alt_jump
= 0;
2462 /* Work area for range table of charset. */
2463 struct range_table_work_area range_table_work
;
2465 /* If the object matched can contain multibyte characters. */
2466 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2468 /* Nonzero if we have pushed down into a subpattern. */
2469 int in_subpattern
= 0;
2471 /* These hold the values of p, pattern, and pend from the main
2472 pattern when we have pushed into a subpattern. */
2473 re_char
*main_p
IF_LINT (= NULL
);
2474 re_char
*main_pattern
IF_LINT (= NULL
);
2475 re_char
*main_pend
IF_LINT (= NULL
);
2479 DEBUG_PRINT ("\nCompiling pattern: ");
2482 unsigned debug_count
;
2484 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2485 putchar (pattern
[debug_count
]);
2490 /* Initialize the compile stack. */
2491 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2492 if (compile_stack
.stack
== NULL
)
2495 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2496 compile_stack
.avail
= 0;
2498 range_table_work
.table
= 0;
2499 range_table_work
.allocated
= 0;
2501 /* Initialize the pattern buffer. */
2502 bufp
->syntax
= syntax
;
2503 bufp
->fastmap_accurate
= 0;
2504 bufp
->not_bol
= bufp
->not_eol
= 0;
2505 bufp
->used_syntax
= 0;
2507 /* Set `used' to zero, so that if we return an error, the pattern
2508 printer (for debugging) will think there's no pattern. We reset it
2512 /* Always count groups, whether or not bufp->no_sub is set. */
2515 #if !defined emacs && !defined SYNTAX_TABLE
2516 /* Initialize the syntax table. */
2517 init_syntax_once ();
2520 if (bufp
->allocated
== 0)
2523 { /* If zero allocated, but buffer is non-null, try to realloc
2524 enough space. This loses if buffer's address is bogus, but
2525 that is the user's responsibility. */
2526 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2529 { /* Caller did not allocate a buffer. Do it for them. */
2530 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2532 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2534 bufp
->allocated
= INIT_BUF_SIZE
;
2537 begalt
= b
= bufp
->buffer
;
2539 /* Loop through the uncompiled pattern until we're at the end. */
2544 /* If this is the end of an included regexp,
2545 pop back to the main regexp and try again. */
2549 pattern
= main_pattern
;
2554 /* If this is the end of the main regexp, we are done. */
2566 /* If there's no special whitespace regexp, treat
2567 spaces normally. And don't try to do this recursively. */
2568 if (!whitespace_regexp
|| in_subpattern
)
2571 /* Peek past following spaces. */
2578 /* If the spaces are followed by a repetition op,
2579 treat them normally. */
2581 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2582 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2585 /* Replace the spaces with the whitespace regexp. */
2589 main_pattern
= pattern
;
2590 p
= pattern
= whitespace_regexp
;
2591 pend
= p
+ strlen ((const char *) p
);
2597 if ( /* If at start of pattern, it's an operator. */
2599 /* If context independent, it's an operator. */
2600 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2601 /* Otherwise, depends on what's come before. */
2602 || at_begline_loc_p (pattern
, p
, syntax
))
2603 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2612 if ( /* If at end of pattern, it's an operator. */
2614 /* If context independent, it's an operator. */
2615 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2616 /* Otherwise, depends on what's next. */
2617 || at_endline_loc_p (p
, pend
, syntax
))
2618 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2627 if ((syntax
& RE_BK_PLUS_QM
)
2628 || (syntax
& RE_LIMITED_OPS
))
2632 /* If there is no previous pattern... */
2635 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2636 FREE_STACK_RETURN (REG_BADRPT
);
2637 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2642 /* 1 means zero (many) matches is allowed. */
2643 boolean zero_times_ok
= 0, many_times_ok
= 0;
2646 /* If there is a sequence of repetition chars, collapse it
2647 down to just one (the right one). We can't combine
2648 interval operators with these because of, e.g., `a{2}*',
2649 which should only match an even number of `a's. */
2653 if ((syntax
& RE_FRUGAL
)
2654 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2658 zero_times_ok
|= c
!= '+';
2659 many_times_ok
|= c
!= '?';
2665 || (!(syntax
& RE_BK_PLUS_QM
)
2666 && (*p
== '+' || *p
== '?')))
2668 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2671 FREE_STACK_RETURN (REG_EESCAPE
);
2672 if (p
[1] == '+' || p
[1] == '?')
2673 PATFETCH (c
); /* Gobble up the backslash. */
2679 /* If we get here, we found another repeat character. */
2683 /* Star, etc. applied to an empty pattern is equivalent
2684 to an empty pattern. */
2685 if (!laststart
|| laststart
== b
)
2688 /* Now we know whether or not zero matches is allowed
2689 and also whether or not two or more matches is allowed. */
2694 boolean simple
= skip_one_char (laststart
) == b
;
2695 size_t startoffset
= 0;
2697 /* Check if the loop can match the empty string. */
2698 (simple
|| !analyze_first (laststart
, b
, NULL
, 0))
2699 ? on_failure_jump
: on_failure_jump_loop
;
2700 assert (skip_one_char (laststart
) <= b
);
2702 if (!zero_times_ok
&& simple
)
2703 { /* Since simple * loops can be made faster by using
2704 on_failure_keep_string_jump, we turn simple P+
2705 into PP* if P is simple. */
2706 unsigned char *p1
, *p2
;
2707 startoffset
= b
- laststart
;
2708 GET_BUFFER_SPACE (startoffset
);
2709 p1
= b
; p2
= laststart
;
2715 GET_BUFFER_SPACE (6);
2718 STORE_JUMP (ofj
, b
, b
+ 6);
2720 /* Simple * loops can use on_failure_keep_string_jump
2721 depending on what follows. But since we don't know
2722 that yet, we leave the decision up to
2723 on_failure_jump_smart. */
2724 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2725 laststart
+ startoffset
, b
+ 6);
2727 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2732 /* A simple ? pattern. */
2733 assert (zero_times_ok
);
2734 GET_BUFFER_SPACE (3);
2735 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2739 else /* not greedy */
2740 { /* I wish the greedy and non-greedy cases could be merged. */
2742 GET_BUFFER_SPACE (7); /* We might use less. */
2745 boolean emptyp
= analyze_first (laststart
, b
, NULL
, 0);
2747 /* The non-greedy multiple match looks like
2748 a repeat..until: we only need a conditional jump
2749 at the end of the loop. */
2750 if (emptyp
) BUF_PUSH (no_op
);
2751 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2752 : on_failure_jump
, b
, laststart
);
2756 /* The repeat...until naturally matches one or more.
2757 To also match zero times, we need to first jump to
2758 the end of the loop (its conditional jump). */
2759 INSERT_JUMP (jump
, laststart
, b
);
2765 /* non-greedy a?? */
2766 INSERT_JUMP (jump
, laststart
, b
+ 3);
2768 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2787 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2789 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2791 /* Ensure that we have enough space to push a charset: the
2792 opcode, the length count, and the bitset; 34 bytes in all. */
2793 GET_BUFFER_SPACE (34);
2797 /* We test `*p == '^' twice, instead of using an if
2798 statement, so we only need one BUF_PUSH. */
2799 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2803 /* Remember the first position in the bracket expression. */
2806 /* Push the number of bytes in the bitmap. */
2807 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2809 /* Clear the whole map. */
2810 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2812 /* charset_not matches newline according to a syntax bit. */
2813 if ((re_opcode_t
) b
[-2] == charset_not
2814 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2815 SET_LIST_BIT ('\n');
2817 /* Read in characters and ranges, setting map bits. */
2820 boolean escaped_char
= false;
2821 const unsigned char *p2
= p
;
2824 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2826 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2827 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2828 So the translation is done later in a loop. Example:
2829 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2832 /* \ might escape characters inside [...] and [^...]. */
2833 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2835 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2838 escaped_char
= true;
2842 /* Could be the end of the bracket expression. If it's
2843 not (i.e., when the bracket expression is `[]' so
2844 far), the ']' character bit gets set way below. */
2845 if (c
== ']' && p2
!= p1
)
2849 /* See if we're at the beginning of a possible character
2852 if (!escaped_char
&&
2853 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2855 /* Leave room for the null. */
2856 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2857 const unsigned char *class_beg
;
2863 /* If pattern is `[[:'. */
2864 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2869 if ((c
== ':' && *p
== ']') || p
== pend
)
2871 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2874 /* This is in any case an invalid class name. */
2879 /* If isn't a word bracketed by `[:' and `:]':
2880 undo the ending character, the letters, and
2881 leave the leading `:' and `[' (but set bits for
2883 if (c
== ':' && *p
== ']')
2885 re_wctype_t cc
= re_wctype (str
);
2888 FREE_STACK_RETURN (REG_ECTYPE
);
2890 /* Throw away the ] at the end of the character
2894 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2897 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2898 if (re_iswctype (btowc (ch
), cc
))
2901 if (c
< (1 << BYTEWIDTH
))
2905 /* Most character classes in a multibyte match
2906 just set a flag. Exceptions are is_blank,
2907 is_digit, is_cntrl, and is_xdigit, since
2908 they can only match ASCII characters. We
2909 don't need to handle them for multibyte.
2910 They are distinguished by a negative wctype. */
2912 /* Setup the gl_state object to its buffer-defined
2913 value. This hardcodes the buffer-global
2914 syntax-table for ASCII chars, while the other chars
2915 will obey syntax-table properties. It's not ideal,
2916 but it's the way it's been done until now. */
2917 SETUP_BUFFER_SYNTAX_TABLE ();
2919 for (ch
= 0; ch
< 256; ++ch
)
2921 c
= RE_CHAR_TO_MULTIBYTE (ch
);
2922 if (! CHAR_BYTE8_P (c
)
2923 && re_iswctype (c
, cc
))
2929 if (ASCII_CHAR_P (c1
))
2931 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
2935 SET_RANGE_TABLE_WORK_AREA_BIT
2936 (range_table_work
, re_wctype_to_bit (cc
));
2938 /* In most cases the matching rule for char classes
2939 only uses the syntax table for multibyte chars,
2940 so that the content of the syntax-table is not
2941 hardcoded in the range_table. SPACE and WORD are
2942 the two exceptions. */
2943 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2944 bufp
->used_syntax
= 1;
2946 /* Repeat the loop. */
2951 /* Go back to right after the "[:". */
2955 /* Because the `:' may start the range, we
2956 can't simply set bit and repeat the loop.
2957 Instead, just set it to C and handle below. */
2962 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2965 /* Discard the `-'. */
2968 /* Fetch the character which ends the range. */
2971 if (CHAR_BYTE8_P (c1
)
2972 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
2973 /* Treat the range from a multibyte character to
2974 raw-byte character as empty. */
2979 /* Range from C to C. */
2984 if (syntax
& RE_NO_EMPTY_RANGES
)
2985 FREE_STACK_RETURN (REG_ERANGEX
);
2986 /* Else, repeat the loop. */
2991 /* Set the range into bitmap */
2992 for (; c
<= c1
; c
++)
2995 if (ch
< (1 << BYTEWIDTH
))
3002 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
3004 if (CHAR_BYTE8_P (c1
))
3005 c
= BYTE8_TO_CHAR (128);
3009 if (CHAR_BYTE8_P (c
))
3011 c
= CHAR_TO_BYTE8 (c
);
3012 c1
= CHAR_TO_BYTE8 (c1
);
3013 for (; c
<= c1
; c
++)
3018 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
3022 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
3029 /* Discard any (non)matching list bytes that are all 0 at the
3030 end of the map. Decrease the map-length byte too. */
3031 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3035 /* Build real range table from work area. */
3036 if (RANGE_TABLE_WORK_USED (range_table_work
)
3037 || RANGE_TABLE_WORK_BITS (range_table_work
))
3040 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3042 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3043 bytes for flags, two for COUNT, and three bytes for
3045 GET_BUFFER_SPACE (4 + used
* 3);
3047 /* Indicate the existence of range table. */
3048 laststart
[1] |= 0x80;
3050 /* Store the character class flag bits into the range table.
3051 If not in emacs, these flag bits are always 0. */
3052 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3053 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3055 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3056 for (i
= 0; i
< used
; i
++)
3057 STORE_CHARACTER_AND_INCR
3058 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3065 if (syntax
& RE_NO_BK_PARENS
)
3072 if (syntax
& RE_NO_BK_PARENS
)
3079 if (syntax
& RE_NEWLINE_ALT
)
3086 if (syntax
& RE_NO_BK_VBAR
)
3093 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3094 goto handle_interval
;
3100 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3102 /* Do not translate the character after the \, so that we can
3103 distinguish, e.g., \B from \b, even if we normally would
3104 translate, e.g., B to b. */
3110 if (syntax
& RE_NO_BK_PARENS
)
3111 goto normal_backslash
;
3116 regnum_t regnum
= 0;
3119 /* Look for a special (?...) construct */
3120 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3122 PATFETCH (c
); /* Gobble up the '?'. */
3128 case ':': shy
= 1; break;
3130 /* An explicitly specified regnum must start
3133 FREE_STACK_RETURN (REG_BADPAT
);
3134 case '1': case '2': case '3': case '4':
3135 case '5': case '6': case '7': case '8': case '9':
3136 regnum
= 10*regnum
+ (c
- '0'); break;
3138 /* Only (?:...) is supported right now. */
3139 FREE_STACK_RETURN (REG_BADPAT
);
3146 regnum
= ++bufp
->re_nsub
;
3148 { /* It's actually not shy, but explicitly numbered. */
3150 if (regnum
> bufp
->re_nsub
)
3151 bufp
->re_nsub
= regnum
;
3152 else if (regnum
> bufp
->re_nsub
3153 /* Ideally, we'd want to check that the specified
3154 group can't have matched (i.e. all subgroups
3155 using the same regnum are in other branches of
3156 OR patterns), but we don't currently keep track
3157 of enough info to do that easily. */
3158 || group_in_compile_stack (compile_stack
, regnum
))
3159 FREE_STACK_RETURN (REG_BADPAT
);
3162 /* It's really shy. */
3163 regnum
= - bufp
->re_nsub
;
3165 if (COMPILE_STACK_FULL
)
3167 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3168 compile_stack_elt_t
);
3169 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3171 compile_stack
.size
<<= 1;
3174 /* These are the values to restore when we hit end of this
3175 group. They are all relative offsets, so that if the
3176 whole pattern moves because of realloc, they will still
3178 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3179 COMPILE_STACK_TOP
.fixup_alt_jump
3180 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3181 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3182 COMPILE_STACK_TOP
.regnum
= regnum
;
3184 /* Do not push a start_memory for groups beyond the last one
3185 we can represent in the compiled pattern. */
3186 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3187 BUF_PUSH_2 (start_memory
, regnum
);
3189 compile_stack
.avail
++;
3194 /* If we've reached MAX_REGNUM groups, then this open
3195 won't actually generate any code, so we'll have to
3196 clear pending_exact explicitly. */
3202 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3204 if (COMPILE_STACK_EMPTY
)
3206 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3207 goto normal_backslash
;
3209 FREE_STACK_RETURN (REG_ERPAREN
);
3215 /* See similar code for backslashed left paren above. */
3216 if (COMPILE_STACK_EMPTY
)
3218 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3221 FREE_STACK_RETURN (REG_ERPAREN
);
3224 /* Since we just checked for an empty stack above, this
3225 ``can't happen''. */
3226 assert (compile_stack
.avail
!= 0);
3228 /* We don't just want to restore into `regnum', because
3229 later groups should continue to be numbered higher,
3230 as in `(ab)c(de)' -- the second group is #2. */
3233 compile_stack
.avail
--;
3234 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3236 = COMPILE_STACK_TOP
.fixup_alt_jump
3237 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3239 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3240 regnum
= COMPILE_STACK_TOP
.regnum
;
3241 /* If we've reached MAX_REGNUM groups, then this open
3242 won't actually generate any code, so we'll have to
3243 clear pending_exact explicitly. */
3246 /* We're at the end of the group, so now we know how many
3247 groups were inside this one. */
3248 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3249 BUF_PUSH_2 (stop_memory
, regnum
);
3254 case '|': /* `\|'. */
3255 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3256 goto normal_backslash
;
3258 if (syntax
& RE_LIMITED_OPS
)
3261 /* Insert before the previous alternative a jump which
3262 jumps to this alternative if the former fails. */
3263 GET_BUFFER_SPACE (3);
3264 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3268 /* The alternative before this one has a jump after it
3269 which gets executed if it gets matched. Adjust that
3270 jump so it will jump to this alternative's analogous
3271 jump (put in below, which in turn will jump to the next
3272 (if any) alternative's such jump, etc.). The last such
3273 jump jumps to the correct final destination. A picture:
3279 If we are at `b', then fixup_alt_jump right now points to a
3280 three-byte space after `a'. We'll put in the jump, set
3281 fixup_alt_jump to right after `b', and leave behind three
3282 bytes which we'll fill in when we get to after `c'. */
3286 /* Mark and leave space for a jump after this alternative,
3287 to be filled in later either by next alternative or
3288 when know we're at the end of a series of alternatives. */
3290 GET_BUFFER_SPACE (3);
3299 /* If \{ is a literal. */
3300 if (!(syntax
& RE_INTERVALS
)
3301 /* If we're at `\{' and it's not the open-interval
3303 || (syntax
& RE_NO_BK_BRACES
))
3304 goto normal_backslash
;
3308 /* If got here, then the syntax allows intervals. */
3310 /* At least (most) this many matches must be made. */
3311 int lower_bound
= 0, upper_bound
= -1;
3315 GET_INTERVAL_COUNT (lower_bound
);
3318 GET_INTERVAL_COUNT (upper_bound
);
3320 /* Interval such as `{1}' => match exactly once. */
3321 upper_bound
= lower_bound
;
3324 || (0 <= upper_bound
&& upper_bound
< lower_bound
))
3325 FREE_STACK_RETURN (REG_BADBR
);
3327 if (!(syntax
& RE_NO_BK_BRACES
))
3330 FREE_STACK_RETURN (REG_BADBR
);
3332 FREE_STACK_RETURN (REG_EESCAPE
);
3337 FREE_STACK_RETURN (REG_BADBR
);
3339 /* We just parsed a valid interval. */
3341 /* If it's invalid to have no preceding re. */
3344 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3345 FREE_STACK_RETURN (REG_BADRPT
);
3346 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3349 goto unfetch_interval
;
3352 if (upper_bound
== 0)
3353 /* If the upper bound is zero, just drop the sub pattern
3356 else if (lower_bound
== 1 && upper_bound
== 1)
3357 /* Just match it once: nothing to do here. */
3360 /* Otherwise, we have a nontrivial interval. When
3361 we're all done, the pattern will look like:
3362 set_number_at <jump count> <upper bound>
3363 set_number_at <succeed_n count> <lower bound>
3364 succeed_n <after jump addr> <succeed_n count>
3366 jump_n <succeed_n addr> <jump count>
3367 (The upper bound and `jump_n' are omitted if
3368 `upper_bound' is 1, though.) */
3370 { /* If the upper bound is > 1, we need to insert
3371 more at the end of the loop. */
3372 unsigned int nbytes
= (upper_bound
< 0 ? 3
3373 : upper_bound
> 1 ? 5 : 0);
3374 unsigned int startoffset
= 0;
3376 GET_BUFFER_SPACE (20); /* We might use less. */
3378 if (lower_bound
== 0)
3380 /* A succeed_n that starts with 0 is really a
3381 a simple on_failure_jump_loop. */
3382 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3388 /* Initialize lower bound of the `succeed_n', even
3389 though it will be set during matching by its
3390 attendant `set_number_at' (inserted next),
3391 because `re_compile_fastmap' needs to know.
3392 Jump to the `jump_n' we might insert below. */
3393 INSERT_JUMP2 (succeed_n
, laststart
,
3398 /* Code to initialize the lower bound. Insert
3399 before the `succeed_n'. The `5' is the last two
3400 bytes of this `set_number_at', plus 3 bytes of
3401 the following `succeed_n'. */
3402 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3407 if (upper_bound
< 0)
3409 /* A negative upper bound stands for infinity,
3410 in which case it degenerates to a plain jump. */
3411 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3414 else if (upper_bound
> 1)
3415 { /* More than one repetition is allowed, so
3416 append a backward jump to the `succeed_n'
3417 that starts this interval.
3419 When we've reached this during matching,
3420 we'll have matched the interval once, so
3421 jump back only `upper_bound - 1' times. */
3422 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3426 /* The location we want to set is the second
3427 parameter of the `jump_n'; that is `b-2' as
3428 an absolute address. `laststart' will be
3429 the `set_number_at' we're about to insert;
3430 `laststart+3' the number to set, the source
3431 for the relative address. But we are
3432 inserting into the middle of the pattern --
3433 so everything is getting moved up by 5.
3434 Conclusion: (b - 2) - (laststart + 3) + 5,
3435 i.e., b - laststart.
3437 We insert this at the beginning of the loop
3438 so that if we fail during matching, we'll
3439 reinitialize the bounds. */
3440 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3441 upper_bound
- 1, b
);
3446 beg_interval
= NULL
;
3451 /* If an invalid interval, match the characters as literals. */
3452 assert (beg_interval
);
3454 beg_interval
= NULL
;
3456 /* normal_char and normal_backslash need `c'. */
3459 if (!(syntax
& RE_NO_BK_BRACES
))
3461 assert (p
> pattern
&& p
[-1] == '\\');
3462 goto normal_backslash
;
3468 /* There is no way to specify the before_dot and after_dot
3469 operators. rms says this is ok. --karl */
3478 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3484 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3490 BUF_PUSH_2 (categoryspec
, c
);
3496 BUF_PUSH_2 (notcategoryspec
, c
);
3502 if (syntax
& RE_NO_GNU_OPS
)
3505 BUF_PUSH_2 (syntaxspec
, Sword
);
3510 if (syntax
& RE_NO_GNU_OPS
)
3513 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3518 if (syntax
& RE_NO_GNU_OPS
)
3525 if (syntax
& RE_NO_GNU_OPS
)
3532 if (syntax
& RE_NO_GNU_OPS
)
3541 FREE_STACK_RETURN (REG_BADPAT
);
3545 if (syntax
& RE_NO_GNU_OPS
)
3547 BUF_PUSH (wordbound
);
3551 if (syntax
& RE_NO_GNU_OPS
)
3553 BUF_PUSH (notwordbound
);
3557 if (syntax
& RE_NO_GNU_OPS
)
3563 if (syntax
& RE_NO_GNU_OPS
)
3568 case '1': case '2': case '3': case '4': case '5':
3569 case '6': case '7': case '8': case '9':
3573 if (syntax
& RE_NO_BK_REFS
)
3574 goto normal_backslash
;
3578 if (reg
> bufp
->re_nsub
|| reg
< 1
3579 /* Can't back reference to a subexp before its end. */
3580 || group_in_compile_stack (compile_stack
, reg
))
3581 FREE_STACK_RETURN (REG_ESUBREG
);
3584 BUF_PUSH_2 (duplicate
, reg
);
3591 if (syntax
& RE_BK_PLUS_QM
)
3594 goto normal_backslash
;
3598 /* You might think it would be useful for \ to mean
3599 not to translate; but if we don't translate it
3600 it will never match anything. */
3607 /* Expects the character in `c'. */
3609 /* If no exactn currently being built. */
3612 /* If last exactn not at current position. */
3613 || pending_exact
+ *pending_exact
+ 1 != b
3615 /* We have only one byte following the exactn for the count. */
3616 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3618 /* If followed by a repetition operator. */
3619 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3620 || ((syntax
& RE_BK_PLUS_QM
)
3621 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3622 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3623 || ((syntax
& RE_INTERVALS
)
3624 && ((syntax
& RE_NO_BK_BRACES
)
3625 ? p
!= pend
&& *p
== '{'
3626 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3628 /* Start building a new exactn. */
3632 BUF_PUSH_2 (exactn
, 0);
3633 pending_exact
= b
- 1;
3636 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3643 len
= CHAR_STRING (c
, b
);
3648 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3649 if (! CHAR_BYTE8_P (c1
))
3651 re_wchar_t c2
= TRANSLATE (c1
);
3653 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3659 (*pending_exact
) += len
;
3664 } /* while p != pend */
3667 /* Through the pattern now. */
3671 if (!COMPILE_STACK_EMPTY
)
3672 FREE_STACK_RETURN (REG_EPAREN
);
3674 /* If we don't want backtracking, force success
3675 the first time we reach the end of the compiled pattern. */
3676 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3679 /* We have succeeded; set the length of the buffer. */
3680 bufp
->used
= b
- bufp
->buffer
;
3685 re_compile_fastmap (bufp
);
3686 DEBUG_PRINT ("\nCompiled pattern: \n");
3687 print_compiled_pattern (bufp
);
3692 #ifndef MATCH_MAY_ALLOCATE
3693 /* Initialize the failure stack to the largest possible stack. This
3694 isn't necessary unless we're trying to avoid calling alloca in
3695 the search and match routines. */
3697 int num_regs
= bufp
->re_nsub
+ 1;
3699 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3701 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3702 falk_stack
.stack
= realloc (fail_stack
.stack
,
3703 fail_stack
.size
* sizeof *falk_stack
.stack
);
3706 regex_grow_registers (num_regs
);
3708 #endif /* not MATCH_MAY_ALLOCATE */
3710 FREE_STACK_RETURN (REG_NOERROR
);
3711 } /* regex_compile */
3713 /* Subroutines for `regex_compile'. */
3715 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3718 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3720 *loc
= (unsigned char) op
;
3721 STORE_NUMBER (loc
+ 1, arg
);
3725 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3728 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3730 *loc
= (unsigned char) op
;
3731 STORE_NUMBER (loc
+ 1, arg1
);
3732 STORE_NUMBER (loc
+ 3, arg2
);
3736 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3737 for OP followed by two-byte integer parameter ARG. */
3740 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3742 register unsigned char *pfrom
= end
;
3743 register unsigned char *pto
= end
+ 3;
3745 while (pfrom
!= loc
)
3748 store_op1 (op
, loc
, arg
);
3752 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3755 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3757 register unsigned char *pfrom
= end
;
3758 register unsigned char *pto
= end
+ 5;
3760 while (pfrom
!= loc
)
3763 store_op2 (op
, loc
, arg1
, arg2
);
3767 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3768 after an alternative or a begin-subexpression. We assume there is at
3769 least one character before the ^. */
3772 at_begline_loc_p (const_re_char
*pattern
, const_re_char
*p
, reg_syntax_t syntax
)
3774 re_char
*prev
= p
- 2;
3775 boolean odd_backslashes
;
3777 /* After a subexpression? */
3779 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3781 /* After an alternative? */
3782 else if (*prev
== '|')
3783 odd_backslashes
= (syntax
& RE_NO_BK_VBAR
) == 0;
3785 /* After a shy subexpression? */
3786 else if (*prev
== ':' && (syntax
& RE_SHY_GROUPS
))
3788 /* Skip over optional regnum. */
3789 while (prev
- 1 >= pattern
&& prev
[-1] >= '0' && prev
[-1] <= '9')
3792 if (!(prev
- 2 >= pattern
3793 && prev
[-1] == '?' && prev
[-2] == '('))
3796 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3801 /* Count the number of preceding backslashes. */
3803 while (prev
- 1 >= pattern
&& prev
[-1] == '\\')
3805 return (p
- prev
) & odd_backslashes
;
3809 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3810 at least one character after the $, i.e., `P < PEND'. */
3813 at_endline_loc_p (const_re_char
*p
, const_re_char
*pend
, reg_syntax_t syntax
)
3816 boolean next_backslash
= *next
== '\\';
3817 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3820 /* Before a subexpression? */
3821 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3822 : next_backslash
&& next_next
&& *next_next
== ')')
3823 /* Before an alternative? */
3824 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3825 : next_backslash
&& next_next
&& *next_next
== '|');
3829 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3830 false if it's not. */
3833 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3835 ssize_t this_element
;
3837 for (this_element
= compile_stack
.avail
- 1;
3840 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3847 If fastmap is non-NULL, go through the pattern and fill fastmap
3848 with all the possible leading chars. If fastmap is NULL, don't
3849 bother filling it up (obviously) and only return whether the
3850 pattern could potentially match the empty string.
3852 Return 1 if p..pend might match the empty string.
3853 Return 0 if p..pend matches at least one char.
3854 Return -1 if fastmap was not updated accurately. */
3857 analyze_first (const_re_char
*p
, const_re_char
*pend
, char *fastmap
,
3858 const int multibyte
)
3863 /* If all elements for base leading-codes in fastmap is set, this
3864 flag is set true. */
3865 boolean match_any_multibyte_characters
= false;
3869 /* The loop below works as follows:
3870 - It has a working-list kept in the PATTERN_STACK and which basically
3871 starts by only containing a pointer to the first operation.
3872 - If the opcode we're looking at is a match against some set of
3873 chars, then we add those chars to the fastmap and go on to the
3874 next work element from the worklist (done via `break').
3875 - If the opcode is a control operator on the other hand, we either
3876 ignore it (if it's meaningless at this point, such as `start_memory')
3877 or execute it (if it's a jump). If the jump has several destinations
3878 (i.e. `on_failure_jump'), then we push the other destination onto the
3880 We guarantee termination by ignoring backward jumps (more or less),
3881 so that `p' is monotonically increasing. More to the point, we
3882 never set `p' (or push) anything `<= p1'. */
3886 /* `p1' is used as a marker of how far back a `on_failure_jump'
3887 can go without being ignored. It is normally equal to `p'
3888 (which prevents any backward `on_failure_jump') except right
3889 after a plain `jump', to allow patterns such as:
3892 10: on_failure_jump 3
3893 as used for the *? operator. */
3902 /* If the first character has to match a backreference, that means
3903 that the group was empty (since it already matched). Since this
3904 is the only case that interests us here, we can assume that the
3905 backreference must match the empty string. */
3910 /* Following are the cases which match a character. These end
3916 /* If multibyte is nonzero, the first byte of each
3917 character is an ASCII or a leading code. Otherwise,
3918 each byte is a character. Thus, this works in both
3923 /* For the case of matching this unibyte regex
3924 against multibyte, we must set a leading code of
3925 the corresponding multibyte character. */
3926 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
3928 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
3935 /* We could put all the chars except for \n (and maybe \0)
3936 but we don't bother since it is generally not worth it. */
3937 if (!fastmap
) break;
3942 if (!fastmap
) break;
3944 /* Chars beyond end of bitmap are possible matches. */
3945 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3946 j
< (1 << BYTEWIDTH
); j
++)
3952 if (!fastmap
) break;
3953 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3954 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3956 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3960 if (/* Any leading code can possibly start a character
3961 which doesn't match the specified set of characters. */
3964 /* If we can match a character class, we can match any
3965 multibyte characters. */
3966 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3967 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3970 if (match_any_multibyte_characters
== false)
3972 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3973 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3975 match_any_multibyte_characters
= true;
3979 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3980 && match_any_multibyte_characters
== false)
3982 /* Set fastmap[I] to 1 where I is a leading code of each
3983 multibyte character in the range table. */
3985 unsigned char lc1
, lc2
;
3987 /* Make P points the range table. `+ 2' is to skip flag
3988 bits for a character class. */
3989 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3991 /* Extract the number of ranges in range table into COUNT. */
3992 EXTRACT_NUMBER_AND_INCR (count
, p
);
3993 for (; count
> 0; count
--, p
+= 3)
3995 /* Extract the start and end of each range. */
3996 EXTRACT_CHARACTER (c
, p
);
3997 lc1
= CHAR_LEADING_CODE (c
);
3999 EXTRACT_CHARACTER (c
, p
);
4000 lc2
= CHAR_LEADING_CODE (c
);
4001 for (j
= lc1
; j
<= lc2
; j
++)
4010 if (!fastmap
) break;
4012 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4014 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4015 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4019 /* This match depends on text properties. These end with
4020 aborting optimizations. */
4024 case notcategoryspec
:
4025 if (!fastmap
) break;
4026 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4028 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
4029 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4032 /* Any leading code can possibly start a character which
4033 has or doesn't has the specified category. */
4034 if (match_any_multibyte_characters
== false)
4036 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4037 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4039 match_any_multibyte_characters
= true;
4043 /* All cases after this match the empty string. These end with
4065 EXTRACT_NUMBER_AND_INCR (j
, p
);
4067 /* Backward jumps can only go back to code that we've already
4068 visited. `re_compile' should make sure this is true. */
4073 case on_failure_jump
:
4074 case on_failure_keep_string_jump
:
4075 case on_failure_jump_loop
:
4076 case on_failure_jump_nastyloop
:
4077 case on_failure_jump_smart
:
4083 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4084 to jump back to "just after here". */
4087 case on_failure_jump
:
4088 case on_failure_keep_string_jump
:
4089 case on_failure_jump_nastyloop
:
4090 case on_failure_jump_loop
:
4091 case on_failure_jump_smart
:
4092 EXTRACT_NUMBER_AND_INCR (j
, p
);
4094 ; /* Backward jump to be ignored. */
4096 { /* We have to look down both arms.
4097 We first go down the "straight" path so as to minimize
4098 stack usage when going through alternatives. */
4099 int r
= analyze_first (p
, pend
, fastmap
, multibyte
);
4107 /* This code simply does not properly handle forward jump_n. */
4108 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4110 /* jump_n can either jump or fall through. The (backward) jump
4111 case has already been handled, so we only need to look at the
4112 fallthrough case. */
4116 /* If N == 0, it should be an on_failure_jump_loop instead. */
4117 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4119 /* We only care about one iteration of the loop, so we don't
4120 need to consider the case where this behaves like an
4137 abort (); /* We have listed all the cases. */
4140 /* Getting here means we have found the possible starting
4141 characters for one path of the pattern -- and that the empty
4142 string does not match. We need not follow this path further. */
4146 /* We reached the end without matching anything. */
4149 } /* analyze_first */
4151 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4152 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4153 characters can start a string that matches the pattern. This fastmap
4154 is used by re_search to skip quickly over impossible starting points.
4156 Character codes above (1 << BYTEWIDTH) are not represented in the
4157 fastmap, but the leading codes are represented. Thus, the fastmap
4158 indicates which character sets could start a match.
4160 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4161 area as BUFP->fastmap.
4163 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4166 Returns 0 if we succeed, -2 if an internal error. */
4169 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4171 char *fastmap
= bufp
->fastmap
;
4174 assert (fastmap
&& bufp
->buffer
);
4176 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4177 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4179 analysis
= analyze_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4180 fastmap
, RE_MULTIBYTE_P (bufp
));
4181 bufp
->can_be_null
= (analysis
!= 0);
4183 } /* re_compile_fastmap */
4185 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4186 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4187 this memory for recording register information. STARTS and ENDS
4188 must be allocated using the malloc library routine, and must each
4189 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4191 If NUM_REGS == 0, then subsequent matches should allocate their own
4194 Unless this function is called, the first search or match using
4195 PATTERN_BUFFER will allocate its own register data, without
4196 freeing the old data. */
4199 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4203 bufp
->regs_allocated
= REGS_REALLOCATE
;
4204 regs
->num_regs
= num_regs
;
4205 regs
->start
= starts
;
4210 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4212 regs
->start
= regs
->end
= 0;
4215 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4217 /* Searching routines. */
4219 /* Like re_search_2, below, but only one string is specified, and
4220 doesn't let you say where to stop matching. */
4223 re_search (struct re_pattern_buffer
*bufp
, const char *string
, size_t size
,
4224 ssize_t startpos
, ssize_t range
, struct re_registers
*regs
)
4226 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4229 WEAK_ALIAS (__re_search
, re_search
)
4231 /* Head address of virtual concatenation of string. */
4232 #define HEAD_ADDR_VSTRING(P) \
4233 (((P) >= size1 ? string2 : string1))
4235 /* Address of POS in the concatenation of virtual string. */
4236 #define POS_ADDR_VSTRING(POS) \
4237 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4239 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4240 virtual concatenation of STRING1 and STRING2, starting first at index
4241 STARTPOS, then at STARTPOS + 1, and so on.
4243 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4245 RANGE is how far to scan while trying to match. RANGE = 0 means try
4246 only at STARTPOS; in general, the last start tried is STARTPOS +
4249 In REGS, return the indices of the virtual concatenation of STRING1
4250 and STRING2 that matched the entire BUFP->buffer and its contained
4253 Do not consider matching one past the index STOP in the virtual
4254 concatenation of STRING1 and STRING2.
4256 We return either the position in the strings at which the match was
4257 found, -1 if no match, or -2 if error (such as failure
4261 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, size_t size1
,
4262 const char *str2
, size_t size2
, ssize_t startpos
, ssize_t range
,
4263 struct re_registers
*regs
, ssize_t stop
)
4266 re_char
*string1
= (re_char
*) str1
;
4267 re_char
*string2
= (re_char
*) str2
;
4268 register char *fastmap
= bufp
->fastmap
;
4269 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4270 size_t total_size
= size1
+ size2
;
4271 ssize_t endpos
= startpos
+ range
;
4272 boolean anchored_start
;
4273 /* Nonzero if we are searching multibyte string. */
4274 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4276 /* Check for out-of-range STARTPOS. */
4277 if (startpos
< 0 || startpos
> total_size
)
4280 /* Fix up RANGE if it might eventually take us outside
4281 the virtual concatenation of STRING1 and STRING2.
4282 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4284 range
= 0 - startpos
;
4285 else if (endpos
> total_size
)
4286 range
= total_size
- startpos
;
4288 /* If the search isn't to be a backwards one, don't waste time in a
4289 search for a pattern anchored at beginning of buffer. */
4290 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4299 /* In a forward search for something that starts with \=.
4300 don't keep searching past point. */
4301 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4303 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4309 /* Update the fastmap now if not correct already. */
4310 if (fastmap
&& !bufp
->fastmap_accurate
)
4311 re_compile_fastmap (bufp
);
4313 /* See whether the pattern is anchored. */
4314 anchored_start
= (bufp
->buffer
[0] == begline
);
4317 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4319 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4321 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4325 /* Loop through the string, looking for a place to start matching. */
4328 /* If the pattern is anchored,
4329 skip quickly past places we cannot match.
4330 We don't bother to treat startpos == 0 specially
4331 because that case doesn't repeat. */
4332 if (anchored_start
&& startpos
> 0)
4334 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4335 : string2
[startpos
- size1
- 1])
4340 /* If a fastmap is supplied, skip quickly over characters that
4341 cannot be the start of a match. If the pattern can match the
4342 null string, however, we don't need to skip characters; we want
4343 the first null string. */
4344 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4346 register re_char
*d
;
4347 register re_wchar_t buf_ch
;
4349 d
= POS_ADDR_VSTRING (startpos
);
4351 if (range
> 0) /* Searching forwards. */
4353 ssize_t irange
= range
, lim
= 0;
4355 if (startpos
< size1
&& startpos
+ range
>= size1
)
4356 lim
= range
- (size1
- startpos
);
4358 /* Written out as an if-else to avoid testing `translate'
4360 if (RE_TRANSLATE_P (translate
))
4367 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4368 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4369 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4372 range
-= buf_charlen
;
4378 register re_wchar_t ch
, translated
;
4381 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4382 translated
= RE_TRANSLATE (translate
, ch
);
4383 if (translated
!= ch
4384 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4386 if (fastmap
[buf_ch
])
4399 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4400 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4402 range
-= buf_charlen
;
4406 while (range
> lim
&& !fastmap
[*d
])
4412 startpos
+= irange
- range
;
4414 else /* Searching backwards. */
4418 buf_ch
= STRING_CHAR (d
);
4419 buf_ch
= TRANSLATE (buf_ch
);
4420 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4425 register re_wchar_t ch
, translated
;
4428 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4429 translated
= TRANSLATE (ch
);
4430 if (translated
!= ch
4431 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4433 if (! fastmap
[TRANSLATE (buf_ch
)])
4439 /* If can't match the null string, and that's all we have left, fail. */
4440 if (range
>= 0 && startpos
== total_size
&& fastmap
4441 && !bufp
->can_be_null
)
4444 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4445 startpos
, regs
, stop
);
4458 /* Update STARTPOS to the next character boundary. */
4461 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4462 int len
= BYTES_BY_CHAR_HEAD (*p
);
4480 /* Update STARTPOS to the previous character boundary. */
4483 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4485 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4487 /* Find the head of multibyte form. */
4488 PREV_CHAR_BOUNDARY (p
, phead
);
4489 range
+= p0
- 1 - p
;
4493 startpos
-= p0
- 1 - p
;
4499 WEAK_ALIAS (__re_search_2
, re_search_2
)
4501 /* Declarations and macros for re_match_2. */
4503 static int bcmp_translate (re_char
*s1
, re_char
*s2
,
4504 register ssize_t len
,
4505 RE_TRANSLATE_TYPE translate
,
4506 const int multibyte
);
4508 /* This converts PTR, a pointer into one of the search strings `string1'
4509 and `string2' into an offset from the beginning of that string. */
4510 #define POINTER_TO_OFFSET(ptr) \
4511 (FIRST_STRING_P (ptr) \
4513 : (ptr) - string2 + (ptrdiff_t) size1)
4515 /* Call before fetching a character with *d. This switches over to
4516 string2 if necessary.
4517 Check re_match_2_internal for a discussion of why end_match_2 might
4518 not be within string2 (but be equal to end_match_1 instead). */
4519 #define PREFETCH() \
4522 /* End of string2 => fail. */ \
4523 if (dend == end_match_2) \
4525 /* End of string1 => advance to string2. */ \
4527 dend = end_match_2; \
4530 /* Call before fetching a char with *d if you already checked other limits.
4531 This is meant for use in lookahead operations like wordend, etc..
4532 where we might need to look at parts of the string that might be
4533 outside of the LIMITs (i.e past `stop'). */
4534 #define PREFETCH_NOLIMIT() \
4538 dend = end_match_2; \
4541 /* Test if at very beginning or at very end of the virtual concatenation
4542 of `string1' and `string2'. If only one string, it's `string2'. */
4543 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4544 #define AT_STRINGS_END(d) ((d) == end2)
4546 /* Disabled due to a compiler bug -- see comment at case wordbound */
4548 /* The comment at case wordbound is following one, but we don't use
4549 AT_WORD_BOUNDARY anymore to support multibyte form.
4551 The DEC Alpha C compiler 3.x generates incorrect code for the
4552 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4553 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4554 macro and introducing temporary variables works around the bug. */
4557 /* Test if D points to a character which is word-constituent. We have
4558 two special cases to check for: if past the end of string1, look at
4559 the first character in string2; and if before the beginning of
4560 string2, look at the last character in string1. */
4561 #define WORDCHAR_P(d) \
4562 (SYNTAX ((d) == end1 ? *string2 \
4563 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4566 /* Test if the character before D and the one at D differ with respect
4567 to being word-constituent. */
4568 #define AT_WORD_BOUNDARY(d) \
4569 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4570 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4573 /* Free everything we malloc. */
4574 #ifdef MATCH_MAY_ALLOCATE
4575 # define FREE_VAR(var) \
4583 # define FREE_VARIABLES() \
4585 REGEX_FREE_STACK (fail_stack.stack); \
4586 FREE_VAR (regstart); \
4587 FREE_VAR (regend); \
4588 FREE_VAR (best_regstart); \
4589 FREE_VAR (best_regend); \
4590 REGEX_SAFE_FREE (); \
4593 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4594 #endif /* not MATCH_MAY_ALLOCATE */
4597 /* Optimization routines. */
4599 /* If the operation is a match against one or more chars,
4600 return a pointer to the next operation, else return NULL. */
4602 skip_one_char (const_re_char
*p
)
4615 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4618 p
= CHARSET_RANGE_TABLE (p
- 1);
4619 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4620 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4623 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4630 case notcategoryspec
:
4642 /* Jump over non-matching operations. */
4644 skip_noops (const_re_char
*p
, const_re_char
*pend
)
4658 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4669 /* Non-zero if "p1 matches something" implies "p2 fails". */
4671 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, const_re_char
*p1
,
4675 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4676 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4678 assert (p1
>= bufp
->buffer
&& p1
< pend
4679 && p2
>= bufp
->buffer
&& p2
<= pend
);
4681 /* Skip over open/close-group commands.
4682 If what follows this loop is a ...+ construct,
4683 look at what begins its body, since we will have to
4684 match at least one of that. */
4685 p2
= skip_noops (p2
, pend
);
4686 /* The same skip can be done for p1, except that this function
4687 is only used in the case where p1 is a simple match operator. */
4688 /* p1 = skip_noops (p1, pend); */
4690 assert (p1
>= bufp
->buffer
&& p1
< pend
4691 && p2
>= bufp
->buffer
&& p2
<= pend
);
4693 op2
= p2
== pend
? succeed
: *p2
;
4699 /* If we're at the end of the pattern, we can change. */
4700 if (skip_one_char (p1
))
4702 DEBUG_PRINT (" End of pattern: fast loop.\n");
4710 register re_wchar_t c
4711 = (re_opcode_t
) *p2
== endline
? '\n'
4712 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4714 if ((re_opcode_t
) *p1
== exactn
)
4716 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4718 DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4723 else if ((re_opcode_t
) *p1
== charset
4724 || (re_opcode_t
) *p1
== charset_not
)
4726 int not = (re_opcode_t
) *p1
== charset_not
;
4728 /* Test if C is listed in charset (or charset_not)
4730 if (! multibyte
|| IS_REAL_ASCII (c
))
4732 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4733 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4736 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4737 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4739 /* `not' is equal to 1 if c would match, which means
4740 that we can't change to pop_failure_jump. */
4743 DEBUG_PRINT (" No match => fast loop.\n");
4747 else if ((re_opcode_t
) *p1
== anychar
4750 DEBUG_PRINT (" . != \\n => fast loop.\n");
4758 if ((re_opcode_t
) *p1
== exactn
)
4759 /* Reuse the code above. */
4760 return mutually_exclusive_p (bufp
, p2
, p1
);
4762 /* It is hard to list up all the character in charset
4763 P2 if it includes multibyte character. Give up in
4765 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4767 /* Now, we are sure that P2 has no range table.
4768 So, for the size of bitmap in P2, `p2[1]' is
4769 enough. But P1 may have range table, so the
4770 size of bitmap table of P1 is extracted by
4771 using macro `CHARSET_BITMAP_SIZE'.
4773 In a multibyte case, we know that all the character
4774 listed in P2 is ASCII. In a unibyte case, P1 has only a
4775 bitmap table. So, in both cases, it is enough to test
4776 only the bitmap table of P1. */
4778 if ((re_opcode_t
) *p1
== charset
)
4781 /* We win if the charset inside the loop
4782 has no overlap with the one after the loop. */
4785 && idx
< CHARSET_BITMAP_SIZE (p1
));
4787 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4791 || idx
== CHARSET_BITMAP_SIZE (p1
))
4793 DEBUG_PRINT (" No match => fast loop.\n");
4797 else if ((re_opcode_t
) *p1
== charset_not
)
4800 /* We win if the charset_not inside the loop lists
4801 every character listed in the charset after. */
4802 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4803 if (! (p2
[2 + idx
] == 0
4804 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4805 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4810 DEBUG_PRINT (" No match => fast loop.\n");
4823 /* Reuse the code above. */
4824 return mutually_exclusive_p (bufp
, p2
, p1
);
4826 /* When we have two charset_not, it's very unlikely that
4827 they don't overlap. The union of the two sets of excluded
4828 chars should cover all possible chars, which, as a matter of
4829 fact, is virtually impossible in multibyte buffers. */
4835 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4837 return ((re_opcode_t
) *p1
== syntaxspec
4838 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4840 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4843 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4845 return ((re_opcode_t
) *p1
== notsyntaxspec
4846 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4848 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4851 return (((re_opcode_t
) *p1
== notsyntaxspec
4852 || (re_opcode_t
) *p1
== syntaxspec
)
4857 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4858 case notcategoryspec
:
4859 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4871 /* Matching routines. */
4873 #ifndef emacs /* Emacs never uses this. */
4874 /* re_match is like re_match_2 except it takes only a single string. */
4877 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
4878 size_t size
, ssize_t pos
, struct re_registers
*regs
)
4880 regoff_t result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
,
4881 size
, pos
, regs
, size
);
4884 WEAK_ALIAS (__re_match
, re_match
)
4885 #endif /* not emacs */
4888 /* In Emacs, this is the string or buffer in which we
4889 are matching. It is used for looking up syntax properties. */
4890 Lisp_Object re_match_object
;
4893 /* re_match_2 matches the compiled pattern in BUFP against the
4894 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4895 and SIZE2, respectively). We start matching at POS, and stop
4898 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4899 store offsets for the substring each group matched in REGS. See the
4900 documentation for exactly how many groups we fill.
4902 We return -1 if no match, -2 if an internal error (such as the
4903 failure stack overflowing). Otherwise, we return the length of the
4904 matched substring. */
4907 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
,
4908 size_t size1
, const char *string2
, size_t size2
, ssize_t pos
,
4909 struct re_registers
*regs
, ssize_t stop
)
4915 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4916 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4917 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4920 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4921 (re_char
*) string2
, size2
,
4925 WEAK_ALIAS (__re_match_2
, re_match_2
)
4928 /* This is a separate function so that we can force an alloca cleanup
4931 re_match_2_internal (struct re_pattern_buffer
*bufp
, const_re_char
*string1
,
4932 size_t size1
, const_re_char
*string2
, size_t size2
,
4933 ssize_t pos
, struct re_registers
*regs
, ssize_t stop
)
4935 /* General temporaries. */
4939 /* Just past the end of the corresponding string. */
4940 re_char
*end1
, *end2
;
4942 /* Pointers into string1 and string2, just past the last characters in
4943 each to consider matching. */
4944 re_char
*end_match_1
, *end_match_2
;
4946 /* Where we are in the data, and the end of the current string. */
4949 /* Used sometimes to remember where we were before starting matching
4950 an operator so that we can go back in case of failure. This "atomic"
4951 behavior of matching opcodes is indispensable to the correctness
4952 of the on_failure_keep_string_jump optimization. */
4955 /* Where we are in the pattern, and the end of the pattern. */
4956 re_char
*p
= bufp
->buffer
;
4957 re_char
*pend
= p
+ bufp
->used
;
4959 /* We use this to map every character in the string. */
4960 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4962 /* Nonzero if BUFP is setup from a multibyte regex. */
4963 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4965 /* Nonzero if STRING1/STRING2 are multibyte. */
4966 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4968 /* Failure point stack. Each place that can handle a failure further
4969 down the line pushes a failure point on this stack. It consists of
4970 regstart, and regend for all registers corresponding to
4971 the subexpressions we're currently inside, plus the number of such
4972 registers, and, finally, two char *'s. The first char * is where
4973 to resume scanning the pattern; the second one is where to resume
4974 scanning the strings. */
4975 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4976 fail_stack_type fail_stack
;
4978 #ifdef DEBUG_COMPILES_ARGUMENTS
4979 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4982 #if defined REL_ALLOC && defined REGEX_MALLOC
4983 /* This holds the pointer to the failure stack, when
4984 it is allocated relocatably. */
4985 fail_stack_elt_t
*failure_stack_ptr
;
4988 /* We fill all the registers internally, independent of what we
4989 return, for use in backreferences. The number here includes
4990 an element for register zero. */
4991 size_t num_regs
= bufp
->re_nsub
+ 1;
4993 /* Information on the contents of registers. These are pointers into
4994 the input strings; they record just what was matched (on this
4995 attempt) by a subexpression part of the pattern, that is, the
4996 regnum-th regstart pointer points to where in the pattern we began
4997 matching and the regnum-th regend points to right after where we
4998 stopped matching the regnum-th subexpression. (The zeroth register
4999 keeps track of what the whole pattern matches.) */
5000 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5001 re_char
**regstart
, **regend
;
5004 /* The following record the register info as found in the above
5005 variables when we find a match better than any we've seen before.
5006 This happens as we backtrack through the failure points, which in
5007 turn happens only if we have not yet matched the entire string. */
5008 unsigned best_regs_set
= false;
5009 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5010 re_char
**best_regstart
, **best_regend
;
5013 /* Logically, this is `best_regend[0]'. But we don't want to have to
5014 allocate space for that if we're not allocating space for anything
5015 else (see below). Also, we never need info about register 0 for
5016 any of the other register vectors, and it seems rather a kludge to
5017 treat `best_regend' differently than the rest. So we keep track of
5018 the end of the best match so far in a separate variable. We
5019 initialize this to NULL so that when we backtrack the first time
5020 and need to test it, it's not garbage. */
5021 re_char
*match_end
= NULL
;
5023 #ifdef DEBUG_COMPILES_ARGUMENTS
5024 /* Counts the total number of registers pushed. */
5025 unsigned num_regs_pushed
= 0;
5028 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5030 REGEX_USE_SAFE_ALLOCA
;
5034 #ifdef MATCH_MAY_ALLOCATE
5035 /* Do not bother to initialize all the register variables if there are
5036 no groups in the pattern, as it takes a fair amount of time. If
5037 there are groups, we include space for register 0 (the whole
5038 pattern), even though we never use it, since it simplifies the
5039 array indexing. We should fix this. */
5042 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5043 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5044 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5045 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5047 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5055 /* We must initialize all our variables to NULL, so that
5056 `FREE_VARIABLES' doesn't try to free them. */
5057 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5059 #endif /* MATCH_MAY_ALLOCATE */
5061 /* The starting position is bogus. */
5062 if (pos
< 0 || pos
> size1
+ size2
)
5068 /* Initialize subexpression text positions to -1 to mark ones that no
5069 start_memory/stop_memory has been seen for. Also initialize the
5070 register information struct. */
5071 for (reg
= 1; reg
< num_regs
; reg
++)
5072 regstart
[reg
] = regend
[reg
] = NULL
;
5074 /* We move `string1' into `string2' if the latter's empty -- but not if
5075 `string1' is null. */
5076 if (size2
== 0 && string1
!= NULL
)
5083 end1
= string1
+ size1
;
5084 end2
= string2
+ size2
;
5086 /* `p' scans through the pattern as `d' scans through the data.
5087 `dend' is the end of the input string that `d' points within. `d'
5088 is advanced into the following input string whenever necessary, but
5089 this happens before fetching; therefore, at the beginning of the
5090 loop, `d' can be pointing at the end of a string, but it cannot
5094 /* Only match within string2. */
5095 d
= string2
+ pos
- size1
;
5096 dend
= end_match_2
= string2
+ stop
- size1
;
5097 end_match_1
= end1
; /* Just to give it a value. */
5103 /* Only match within string1. */
5104 end_match_1
= string1
+ stop
;
5106 When we reach end_match_1, PREFETCH normally switches to string2.
5107 But in the present case, this means that just doing a PREFETCH
5108 makes us jump from `stop' to `gap' within the string.
5109 What we really want here is for the search to stop as
5110 soon as we hit end_match_1. That's why we set end_match_2
5111 to end_match_1 (since PREFETCH fails as soon as we hit
5113 end_match_2
= end_match_1
;
5116 { /* It's important to use this code when stop == size so that
5117 moving `d' from end1 to string2 will not prevent the d == dend
5118 check from catching the end of string. */
5120 end_match_2
= string2
+ stop
- size1
;
5126 DEBUG_PRINT ("The compiled pattern is: ");
5127 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5128 DEBUG_PRINT ("The string to match is: `");
5129 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5130 DEBUG_PRINT ("'\n");
5132 /* This loops over pattern commands. It exits by returning from the
5133 function if the match is complete, or it drops through if the match
5134 fails at this starting point in the input data. */
5137 DEBUG_PRINT ("\n%p: ", p
);
5143 /* End of pattern means we might have succeeded. */
5144 DEBUG_PRINT ("end of pattern ... ");
5146 /* If we haven't matched the entire string, and we want the
5147 longest match, try backtracking. */
5148 if (d
!= end_match_2
)
5150 /* 1 if this match ends in the same string (string1 or string2)
5151 as the best previous match. */
5152 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5153 == FIRST_STRING_P (d
));
5154 /* 1 if this match is the best seen so far. */
5155 boolean best_match_p
;
5157 /* AIX compiler got confused when this was combined
5158 with the previous declaration. */
5160 best_match_p
= d
> match_end
;
5162 best_match_p
= !FIRST_STRING_P (d
);
5164 DEBUG_PRINT ("backtracking.\n");
5166 if (!FAIL_STACK_EMPTY ())
5167 { /* More failure points to try. */
5169 /* If exceeds best match so far, save it. */
5170 if (!best_regs_set
|| best_match_p
)
5172 best_regs_set
= true;
5175 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5177 for (reg
= 1; reg
< num_regs
; reg
++)
5179 best_regstart
[reg
] = regstart
[reg
];
5180 best_regend
[reg
] = regend
[reg
];
5186 /* If no failure points, don't restore garbage. And if
5187 last match is real best match, don't restore second
5189 else if (best_regs_set
&& !best_match_p
)
5192 /* Restore best match. It may happen that `dend ==
5193 end_match_1' while the restored d is in string2.
5194 For example, the pattern `x.*y.*z' against the
5195 strings `x-' and `y-z-', if the two strings are
5196 not consecutive in memory. */
5197 DEBUG_PRINT ("Restoring best registers.\n");
5200 dend
= ((d
>= string1
&& d
<= end1
)
5201 ? end_match_1
: end_match_2
);
5203 for (reg
= 1; reg
< num_regs
; reg
++)
5205 regstart
[reg
] = best_regstart
[reg
];
5206 regend
[reg
] = best_regend
[reg
];
5209 } /* d != end_match_2 */
5212 DEBUG_PRINT ("Accepting match.\n");
5214 /* If caller wants register contents data back, do it. */
5215 if (regs
&& !bufp
->no_sub
)
5217 /* Have the register data arrays been allocated? */
5218 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5219 { /* No. So allocate them with malloc. We need one
5220 extra element beyond `num_regs' for the `-1' marker
5222 regs
->num_regs
= max (RE_NREGS
, num_regs
+ 1);
5223 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5224 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5225 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5230 bufp
->regs_allocated
= REGS_REALLOCATE
;
5232 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5233 { /* Yes. If we need more elements than were already
5234 allocated, reallocate them. If we need fewer, just
5236 if (regs
->num_regs
< num_regs
+ 1)
5238 regs
->num_regs
= num_regs
+ 1;
5239 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5240 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5241 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5250 /* These braces fend off a "empty body in an else-statement"
5251 warning under GCC when assert expands to nothing. */
5252 assert (bufp
->regs_allocated
== REGS_FIXED
);
5255 /* Convert the pointer data in `regstart' and `regend' to
5256 indices. Register zero has to be set differently,
5257 since we haven't kept track of any info for it. */
5258 if (regs
->num_regs
> 0)
5260 regs
->start
[0] = pos
;
5261 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5264 /* Go through the first `min (num_regs, regs->num_regs)'
5265 registers, since that is all we initialized. */
5266 for (reg
= 1; reg
< min (num_regs
, regs
->num_regs
); reg
++)
5268 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5269 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5272 regs
->start
[reg
] = POINTER_TO_OFFSET (regstart
[reg
]);
5273 regs
->end
[reg
] = POINTER_TO_OFFSET (regend
[reg
]);
5277 /* If the regs structure we return has more elements than
5278 were in the pattern, set the extra elements to -1. If
5279 we (re)allocated the registers, this is the case,
5280 because we always allocate enough to have at least one
5282 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5283 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5284 } /* regs && !bufp->no_sub */
5286 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5287 nfailure_points_pushed
, nfailure_points_popped
,
5288 nfailure_points_pushed
- nfailure_points_popped
);
5289 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed
);
5291 dcnt
= POINTER_TO_OFFSET (d
) - pos
;
5293 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt
);
5299 /* Otherwise match next pattern command. */
5302 /* Ignore these. Used to ignore the n of succeed_n's which
5303 currently have n == 0. */
5305 DEBUG_PRINT ("EXECUTING no_op.\n");
5309 DEBUG_PRINT ("EXECUTING succeed.\n");
5312 /* Match the next n pattern characters exactly. The following
5313 byte in the pattern defines n, and the n bytes after that
5314 are the characters to match. */
5317 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt
);
5319 /* Remember the start point to rollback upon failure. */
5323 /* This is written out as an if-else so we don't waste time
5324 testing `translate' inside the loop. */
5325 if (RE_TRANSLATE_P (translate
))
5329 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5349 /* The cost of testing `translate' is comparatively small. */
5350 if (target_multibyte
)
5353 int pat_charlen
, buf_charlen
;
5358 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5361 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5364 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5366 if (TRANSLATE (buf_ch
) != pat_ch
)
5374 mcnt
-= pat_charlen
;
5386 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5387 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5394 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5395 if (! CHAR_BYTE8_P (buf_ch
))
5397 buf_ch
= TRANSLATE (buf_ch
);
5398 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5404 if (buf_ch
!= pat_ch
)
5417 /* Match any character except possibly a newline or a null. */
5423 DEBUG_PRINT ("EXECUTING anychar.\n");
5426 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5428 buf_ch
= TRANSLATE (buf_ch
);
5430 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5432 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5433 && buf_ch
== '\000'))
5436 DEBUG_PRINT (" Matched `%d'.\n", *d
);
5445 register unsigned int c
;
5446 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5449 /* Start of actual range_table, or end of bitmap if there is no
5451 re_char
*range_table
IF_LINT (= NULL
);
5453 /* Nonzero if there is a range table. */
5454 int range_table_exists
;
5456 /* Number of ranges of range table. This is not included
5457 in the initial byte-length of the command. */
5460 /* Whether matching against a unibyte character. */
5461 boolean unibyte_char
= false;
5463 DEBUG_PRINT ("EXECUTING charset%s.\n", not ? "_not" : "");
5465 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5467 if (range_table_exists
)
5469 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5470 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5474 c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5475 if (target_multibyte
)
5480 c1
= RE_CHAR_TO_UNIBYTE (c
);
5483 unibyte_char
= true;
5489 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5491 if (! CHAR_BYTE8_P (c1
))
5493 c1
= TRANSLATE (c1
);
5494 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5497 unibyte_char
= true;
5502 unibyte_char
= true;
5505 if (unibyte_char
&& c
< (1 << BYTEWIDTH
))
5506 { /* Lookup bitmap. */
5507 /* Cast to `unsigned' instead of `unsigned char' in
5508 case the bit list is a full 32 bytes long. */
5509 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5510 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5514 else if (range_table_exists
)
5516 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5518 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5519 | (class_bits
& BIT_MULTIBYTE
)
5520 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5521 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5522 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5523 | (class_bits
& BIT_WORD
&& ISWORD (c
))
5524 | (class_bits
& BIT_ALPHA
&& ISALPHA (c
))
5525 | (class_bits
& BIT_ALNUM
&& ISALNUM (c
)))
5528 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5532 if (range_table_exists
)
5533 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5535 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5537 if (!not) goto fail
;
5544 /* The beginning of a group is represented by start_memory.
5545 The argument is the register number. The text
5546 matched within the group is recorded (in the internal
5547 registers data structure) under the register number. */
5549 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p
);
5551 /* In case we need to undo this operation (via backtracking). */
5552 PUSH_FAILURE_REG (*p
);
5555 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5556 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart
[*p
]));
5558 /* Move past the register number and inner group count. */
5563 /* The stop_memory opcode represents the end of a group. Its
5564 argument is the same as start_memory's: the register number. */
5566 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p
);
5568 assert (!REG_UNSET (regstart
[*p
]));
5569 /* Strictly speaking, there should be code such as:
5571 assert (REG_UNSET (regend[*p]));
5572 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5574 But the only info to be pushed is regend[*p] and it is known to
5575 be UNSET, so there really isn't anything to push.
5576 Not pushing anything, on the other hand deprives us from the
5577 guarantee that regend[*p] is UNSET since undoing this operation
5578 will not reset its value properly. This is not important since
5579 the value will only be read on the next start_memory or at
5580 the very end and both events can only happen if this stop_memory
5584 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend
[*p
]));
5586 /* Move past the register number and the inner group count. */
5591 /* \<digit> has been turned into a `duplicate' command which is
5592 followed by the numeric value of <digit> as the register number. */
5595 register re_char
*d2
, *dend2
;
5596 int regno
= *p
++; /* Get which register to match against. */
5597 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno
);
5599 /* Can't back reference a group which we've never matched. */
5600 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5603 /* Where in input to try to start matching. */
5604 d2
= regstart
[regno
];
5606 /* Remember the start point to rollback upon failure. */
5609 /* Where to stop matching; if both the place to start and
5610 the place to stop matching are in the same string, then
5611 set to the place to stop, otherwise, for now have to use
5612 the end of the first string. */
5614 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5615 == FIRST_STRING_P (regend
[regno
]))
5616 ? regend
[regno
] : end_match_1
);
5621 /* If necessary, advance to next segment in register
5625 if (dend2
== end_match_2
) break;
5626 if (dend2
== regend
[regno
]) break;
5628 /* End of string1 => advance to string2. */
5630 dend2
= regend
[regno
];
5632 /* At end of register contents => success */
5633 if (d2
== dend2
) break;
5635 /* If necessary, advance to next segment in data. */
5638 /* How many characters left in this segment to match. */
5641 /* Want how many consecutive characters we can match in
5642 one shot, so, if necessary, adjust the count. */
5643 if (dcnt
> dend2
- d2
)
5646 /* Compare that many; failure if mismatch, else move
5648 if (RE_TRANSLATE_P (translate
)
5649 ? bcmp_translate (d
, d2
, dcnt
, translate
, target_multibyte
)
5650 : memcmp (d
, d2
, dcnt
))
5655 d
+= dcnt
, d2
+= dcnt
;
5661 /* begline matches the empty string at the beginning of the string
5662 (unless `not_bol' is set in `bufp'), and after newlines. */
5664 DEBUG_PRINT ("EXECUTING begline.\n");
5666 if (AT_STRINGS_BEG (d
))
5668 if (!bufp
->not_bol
) break;
5673 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5677 /* In all other cases, we fail. */
5681 /* endline is the dual of begline. */
5683 DEBUG_PRINT ("EXECUTING endline.\n");
5685 if (AT_STRINGS_END (d
))
5687 if (!bufp
->not_eol
) break;
5691 PREFETCH_NOLIMIT ();
5698 /* Match at the very beginning of the data. */
5700 DEBUG_PRINT ("EXECUTING begbuf.\n");
5701 if (AT_STRINGS_BEG (d
))
5706 /* Match at the very end of the data. */
5708 DEBUG_PRINT ("EXECUTING endbuf.\n");
5709 if (AT_STRINGS_END (d
))
5714 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5715 pushes NULL as the value for the string on the stack. Then
5716 `POP_FAILURE_POINT' will keep the current value for the
5717 string, instead of restoring it. To see why, consider
5718 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5719 then the . fails against the \n. But the next thing we want
5720 to do is match the \n against the \n; if we restored the
5721 string value, we would be back at the foo.
5723 Because this is used only in specific cases, we don't need to
5724 check all the things that `on_failure_jump' does, to make
5725 sure the right things get saved on the stack. Hence we don't
5726 share its code. The only reason to push anything on the
5727 stack at all is that otherwise we would have to change
5728 `anychar's code to do something besides goto fail in this
5729 case; that seems worse than this. */
5730 case on_failure_keep_string_jump
:
5731 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5732 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5735 PUSH_FAILURE_POINT (p
- 3, NULL
);
5738 /* A nasty loop is introduced by the non-greedy *? and +?.
5739 With such loops, the stack only ever contains one failure point
5740 at a time, so that a plain on_failure_jump_loop kind of
5741 cycle detection cannot work. Worse yet, such a detection
5742 can not only fail to detect a cycle, but it can also wrongly
5743 detect a cycle (between different instantiations of the same
5745 So the method used for those nasty loops is a little different:
5746 We use a special cycle-detection-stack-frame which is pushed
5747 when the on_failure_jump_nastyloop failure-point is *popped*.
5748 This special frame thus marks the beginning of one iteration
5749 through the loop and we can hence easily check right here
5750 whether something matched between the beginning and the end of
5752 case on_failure_jump_nastyloop
:
5753 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5754 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5757 assert ((re_opcode_t
)p
[-4] == no_op
);
5760 CHECK_INFINITE_LOOP (p
- 4, d
);
5762 /* If there's a cycle, just continue without pushing
5763 this failure point. The failure point is the "try again"
5764 option, which shouldn't be tried.
5765 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5766 PUSH_FAILURE_POINT (p
- 3, d
);
5770 /* Simple loop detecting on_failure_jump: just check on the
5771 failure stack if the same spot was already hit earlier. */
5772 case on_failure_jump_loop
:
5774 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5775 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5779 CHECK_INFINITE_LOOP (p
- 3, d
);
5781 /* If there's a cycle, get out of the loop, as if the matching
5782 had failed. We used to just `goto fail' here, but that was
5783 aborting the search a bit too early: we want to keep the
5784 empty-loop-match and keep matching after the loop.
5785 We want (x?)*y\1z to match both xxyz and xxyxz. */
5788 PUSH_FAILURE_POINT (p
- 3, d
);
5793 /* Uses of on_failure_jump:
5795 Each alternative starts with an on_failure_jump that points
5796 to the beginning of the next alternative. Each alternative
5797 except the last ends with a jump that in effect jumps past
5798 the rest of the alternatives. (They really jump to the
5799 ending jump of the following alternative, because tensioning
5800 these jumps is a hassle.)
5802 Repeats start with an on_failure_jump that points past both
5803 the repetition text and either the following jump or
5804 pop_failure_jump back to this on_failure_jump. */
5805 case on_failure_jump
:
5806 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5807 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5810 PUSH_FAILURE_POINT (p
-3, d
);
5813 /* This operation is used for greedy *.
5814 Compare the beginning of the repeat with what in the
5815 pattern follows its end. If we can establish that there
5816 is nothing that they would both match, i.e., that we
5817 would have to backtrack because of (as in, e.g., `a*a')
5818 then we can use a non-backtracking loop based on
5819 on_failure_keep_string_jump instead of on_failure_jump. */
5820 case on_failure_jump_smart
:
5821 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5822 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5825 re_char
*p1
= p
; /* Next operation. */
5826 /* Here, we discard `const', making re_match non-reentrant. */
5827 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5828 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5830 p
-= 3; /* Reset so that we will re-execute the
5831 instruction once it's been changed. */
5833 EXTRACT_NUMBER (mcnt
, p2
- 2);
5835 /* Ensure this is a indeed the trivial kind of loop
5836 we are expecting. */
5837 assert (skip_one_char (p1
) == p2
- 3);
5838 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5839 DEBUG_STATEMENT (debug
+= 2);
5840 if (mutually_exclusive_p (bufp
, p1
, p2
))
5842 /* Use a fast `on_failure_keep_string_jump' loop. */
5843 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5844 *p3
= (unsigned char) on_failure_keep_string_jump
;
5845 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5849 /* Default to a safe `on_failure_jump' loop. */
5850 DEBUG_PRINT (" smart default => slow loop.\n");
5851 *p3
= (unsigned char) on_failure_jump
;
5853 DEBUG_STATEMENT (debug
-= 2);
5857 /* Unconditionally jump (without popping any failure points). */
5860 IMMEDIATE_QUIT_CHECK
;
5861 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5862 DEBUG_PRINT ("EXECUTING jump %d ", mcnt
);
5863 p
+= mcnt
; /* Do the jump. */
5864 DEBUG_PRINT ("(to %p).\n", p
);
5868 /* Have to succeed matching what follows at least n times.
5869 After that, handle like `on_failure_jump'. */
5871 /* Signedness doesn't matter since we only compare MCNT to 0. */
5872 EXTRACT_NUMBER (mcnt
, p
+ 2);
5873 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt
);
5875 /* Originally, mcnt is how many times we HAVE to succeed. */
5878 /* Here, we discard `const', making re_match non-reentrant. */
5879 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5882 PUSH_NUMBER (p2
, mcnt
);
5885 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5890 /* Signedness doesn't matter since we only compare MCNT to 0. */
5891 EXTRACT_NUMBER (mcnt
, p
+ 2);
5892 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt
);
5894 /* Originally, this is how many times we CAN jump. */
5897 /* Here, we discard `const', making re_match non-reentrant. */
5898 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5900 PUSH_NUMBER (p2
, mcnt
);
5901 goto unconditional_jump
;
5903 /* If don't have to jump any more, skip over the rest of command. */
5910 unsigned char *p2
; /* Location of the counter. */
5911 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5913 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5914 /* Here, we discard `const', making re_match non-reentrant. */
5915 p2
= (unsigned char*) p
+ mcnt
;
5916 /* Signedness doesn't matter since we only copy MCNT's bits. */
5917 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5918 DEBUG_PRINT (" Setting %p to %d.\n", p2
, mcnt
);
5919 PUSH_NUMBER (p2
, mcnt
);
5926 boolean
not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5927 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5929 /* We SUCCEED (or FAIL) in one of the following cases: */
5931 /* Case 1: D is at the beginning or the end of string. */
5932 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5936 /* C1 is the character before D, S1 is the syntax of C1, C2
5937 is the character at D, and S2 is the syntax of C2. */
5942 ssize_t offset
= PTR_TO_OFFSET (d
- 1);
5943 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5944 UPDATE_SYNTAX_TABLE (charpos
);
5946 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5949 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5951 PREFETCH_NOLIMIT ();
5952 GET_CHAR_AFTER (c2
, d
, dummy
);
5955 if (/* Case 2: Only one of S1 and S2 is Sword. */
5956 ((s1
== Sword
) != (s2
== Sword
))
5957 /* Case 3: Both of S1 and S2 are Sword, and macro
5958 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5959 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5969 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5971 /* We FAIL in one of the following cases: */
5973 /* Case 1: D is at the end of string. */
5974 if (AT_STRINGS_END (d
))
5978 /* C1 is the character before D, S1 is the syntax of C1, C2
5979 is the character at D, and S2 is the syntax of C2. */
5984 ssize_t offset
= PTR_TO_OFFSET (d
);
5985 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5986 UPDATE_SYNTAX_TABLE (charpos
);
5989 GET_CHAR_AFTER (c2
, d
, dummy
);
5992 /* Case 2: S2 is not Sword. */
5996 /* Case 3: D is not at the beginning of string ... */
5997 if (!AT_STRINGS_BEG (d
))
5999 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6001 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6005 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6007 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6014 DEBUG_PRINT ("EXECUTING wordend.\n");
6016 /* We FAIL in one of the following cases: */
6018 /* Case 1: D is at the beginning of string. */
6019 if (AT_STRINGS_BEG (d
))
6023 /* C1 is the character before D, S1 is the syntax of C1, C2
6024 is the character at D, and S2 is the syntax of C2. */
6029 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6030 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6031 UPDATE_SYNTAX_TABLE (charpos
);
6033 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6036 /* Case 2: S1 is not Sword. */
6040 /* Case 3: D is not at the end of string ... */
6041 if (!AT_STRINGS_END (d
))
6043 PREFETCH_NOLIMIT ();
6044 GET_CHAR_AFTER (c2
, d
, dummy
);
6046 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6050 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6052 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6059 DEBUG_PRINT ("EXECUTING symbeg.\n");
6061 /* We FAIL in one of the following cases: */
6063 /* Case 1: D is at the end of string. */
6064 if (AT_STRINGS_END (d
))
6068 /* C1 is the character before D, S1 is the syntax of C1, C2
6069 is the character at D, and S2 is the syntax of C2. */
6073 ssize_t offset
= PTR_TO_OFFSET (d
);
6074 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6075 UPDATE_SYNTAX_TABLE (charpos
);
6078 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6081 /* Case 2: S2 is neither Sword nor Ssymbol. */
6082 if (s2
!= Sword
&& s2
!= Ssymbol
)
6085 /* Case 3: D is not at the beginning of string ... */
6086 if (!AT_STRINGS_BEG (d
))
6088 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6090 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6094 /* ... and S1 is Sword or Ssymbol. */
6095 if (s1
== Sword
|| s1
== Ssymbol
)
6102 DEBUG_PRINT ("EXECUTING symend.\n");
6104 /* We FAIL in one of the following cases: */
6106 /* Case 1: D is at the beginning of string. */
6107 if (AT_STRINGS_BEG (d
))
6111 /* C1 is the character before D, S1 is the syntax of C1, C2
6112 is the character at D, and S2 is the syntax of C2. */
6116 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6117 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6118 UPDATE_SYNTAX_TABLE (charpos
);
6120 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6123 /* Case 2: S1 is neither Ssymbol nor Sword. */
6124 if (s1
!= Sword
&& s1
!= Ssymbol
)
6127 /* Case 3: D is not at the end of string ... */
6128 if (!AT_STRINGS_END (d
))
6130 PREFETCH_NOLIMIT ();
6131 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6133 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6137 /* ... and S2 is Sword or Ssymbol. */
6138 if (s2
== Sword
|| s2
== Ssymbol
)
6147 boolean
not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6149 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6154 ssize_t offset
= PTR_TO_OFFSET (d
);
6155 ssize_t pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6156 UPDATE_SYNTAX_TABLE (pos1
);
6163 GET_CHAR_AFTER (c
, d
, len
);
6164 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6173 DEBUG_PRINT ("EXECUTING before_dot.\n");
6174 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6179 DEBUG_PRINT ("EXECUTING at_dot.\n");
6180 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6185 DEBUG_PRINT ("EXECUTING after_dot.\n");
6186 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6191 case notcategoryspec
:
6193 boolean
not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6195 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6196 not ? "not" : "", mcnt
);
6202 GET_CHAR_AFTER (c
, d
, len
);
6203 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6215 continue; /* Successfully executed one pattern command; keep going. */
6218 /* We goto here if a matching operation fails. */
6220 IMMEDIATE_QUIT_CHECK
;
6221 if (!FAIL_STACK_EMPTY ())
6224 /* A restart point is known. Restore to that state. */
6225 DEBUG_PRINT ("\nFAIL:\n");
6226 POP_FAILURE_POINT (str
, pat
);
6229 case on_failure_keep_string_jump
:
6230 assert (str
== NULL
);
6231 goto continue_failure_jump
;
6233 case on_failure_jump_nastyloop
:
6234 assert ((re_opcode_t
)pat
[-2] == no_op
);
6235 PUSH_FAILURE_POINT (pat
- 2, str
);
6238 case on_failure_jump_loop
:
6239 case on_failure_jump
:
6242 continue_failure_jump
:
6243 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6248 /* A special frame used for nastyloops. */
6255 assert (p
>= bufp
->buffer
&& p
<= pend
);
6257 if (d
>= string1
&& d
<= end1
)
6261 break; /* Matching at this starting point really fails. */
6265 goto restore_best_regs
;
6269 return -1; /* Failure to match. */
6272 /* Subroutine definitions for re_match_2. */
6274 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6275 bytes; nonzero otherwise. */
6278 bcmp_translate (const_re_char
*s1
, const_re_char
*s2
, register ssize_t len
,
6279 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6281 register re_char
*p1
= s1
, *p2
= s2
;
6282 re_char
*p1_end
= s1
+ len
;
6283 re_char
*p2_end
= s2
+ len
;
6285 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6286 different lengths, but relying on a single `len' would break this. -sm */
6287 while (p1
< p1_end
&& p2
< p2_end
)
6289 int p1_charlen
, p2_charlen
;
6290 re_wchar_t p1_ch
, p2_ch
;
6292 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6293 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6295 if (RE_TRANSLATE (translate
, p1_ch
)
6296 != RE_TRANSLATE (translate
, p2_ch
))
6299 p1
+= p1_charlen
, p2
+= p2_charlen
;
6302 if (p1
!= p1_end
|| p2
!= p2_end
)
6308 /* Entry points for GNU code. */
6310 /* re_compile_pattern is the GNU regular expression compiler: it
6311 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6312 Returns 0 if the pattern was valid, otherwise an error string.
6314 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6315 are set in BUFP on entry.
6317 We call regex_compile to do the actual compilation. */
6320 re_compile_pattern (const char *pattern
, size_t length
,
6321 struct re_pattern_buffer
*bufp
)
6325 /* GNU code is written to assume at least RE_NREGS registers will be set
6326 (and at least one extra will be -1). */
6327 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6329 /* And GNU code determines whether or not to get register information
6330 by passing null for the REGS argument to re_match, etc., not by
6334 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6338 return gettext (re_error_msgid
[(int) ret
]);
6340 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6342 /* Entry points compatible with 4.2 BSD regex library. We don't define
6343 them unless specifically requested. */
6345 #if defined _REGEX_RE_COMP || defined _LIBC
6347 /* BSD has one and only one pattern buffer. */
6348 static struct re_pattern_buffer re_comp_buf
;
6352 /* Make these definitions weak in libc, so POSIX programs can redefine
6353 these names if they don't use our functions, and still use
6354 regcomp/regexec below without link errors. */
6357 re_comp (const char *s
)
6363 if (!re_comp_buf
.buffer
)
6364 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6365 return (char *) gettext ("No previous regular expression");
6369 if (!re_comp_buf
.buffer
)
6371 re_comp_buf
.buffer
= malloc (200);
6372 if (re_comp_buf
.buffer
== NULL
)
6373 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6374 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6375 re_comp_buf
.allocated
= 200;
6377 re_comp_buf
.fastmap
= malloc (1 << BYTEWIDTH
);
6378 if (re_comp_buf
.fastmap
== NULL
)
6379 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6380 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6383 /* Since `re_exec' always passes NULL for the `regs' argument, we
6384 don't need to initialize the pattern buffer fields which affect it. */
6386 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6391 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6392 return (char *) gettext (re_error_msgid
[(int) ret
]);
6400 re_exec (const char *s
)
6402 const size_t len
= strlen (s
);
6403 return re_search (&re_comp_buf
, s
, len
, 0, len
, 0) >= 0;
6405 #endif /* _REGEX_RE_COMP */
6407 /* POSIX.2 functions. Don't define these for Emacs. */
6411 /* regcomp takes a regular expression as a string and compiles it.
6413 PREG is a regex_t *. We do not expect any fields to be initialized,
6414 since POSIX says we shouldn't. Thus, we set
6416 `buffer' to the compiled pattern;
6417 `used' to the length of the compiled pattern;
6418 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6419 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6420 RE_SYNTAX_POSIX_BASIC;
6421 `fastmap' to an allocated space for the fastmap;
6422 `fastmap_accurate' to zero;
6423 `re_nsub' to the number of subexpressions in PATTERN.
6425 PATTERN is the address of the pattern string.
6427 CFLAGS is a series of bits which affect compilation.
6429 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6430 use POSIX basic syntax.
6432 If REG_NEWLINE is set, then . and [^...] don't match newline.
6433 Also, regexec will try a match beginning after every newline.
6435 If REG_ICASE is set, then we considers upper- and lowercase
6436 versions of letters to be equivalent when matching.
6438 If REG_NOSUB is set, then when PREG is passed to regexec, that
6439 routine will report only success or failure, and nothing about the
6442 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6443 the return codes and their meanings.) */
6446 regcomp (regex_t
*_Restrict_ preg
, const char *_Restrict_ pattern
,
6451 = (cflags
& REG_EXTENDED
) ?
6452 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6454 /* regex_compile will allocate the space for the compiled pattern. */
6456 preg
->allocated
= 0;
6459 /* Try to allocate space for the fastmap. */
6460 preg
->fastmap
= malloc (1 << BYTEWIDTH
);
6462 if (cflags
& REG_ICASE
)
6466 preg
->translate
= malloc (CHAR_SET_SIZE
* sizeof *preg
->translate
);
6467 if (preg
->translate
== NULL
)
6468 return (int) REG_ESPACE
;
6470 /* Map uppercase characters to corresponding lowercase ones. */
6471 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6472 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6475 preg
->translate
= NULL
;
6477 /* If REG_NEWLINE is set, newlines are treated differently. */
6478 if (cflags
& REG_NEWLINE
)
6479 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6480 syntax
&= ~RE_DOT_NEWLINE
;
6481 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6484 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6486 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6488 /* POSIX says a null character in the pattern terminates it, so we
6489 can use strlen here in compiling the pattern. */
6490 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6492 /* POSIX doesn't distinguish between an unmatched open-group and an
6493 unmatched close-group: both are REG_EPAREN. */
6494 if (ret
== REG_ERPAREN
)
6497 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6498 { /* Compute the fastmap now, since regexec cannot modify the pattern
6500 re_compile_fastmap (preg
);
6501 if (preg
->can_be_null
)
6502 { /* The fastmap can't be used anyway. */
6503 free (preg
->fastmap
);
6504 preg
->fastmap
= NULL
;
6509 WEAK_ALIAS (__regcomp
, regcomp
)
6512 /* regexec searches for a given pattern, specified by PREG, in the
6515 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6516 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6517 least NMATCH elements, and we set them to the offsets of the
6518 corresponding matched substrings.
6520 EFLAGS specifies `execution flags' which affect matching: if
6521 REG_NOTBOL is set, then ^ does not match at the beginning of the
6522 string; if REG_NOTEOL is set, then $ does not match at the end.
6524 We return 0 if we find a match and REG_NOMATCH if not. */
6527 regexec (const regex_t
*_Restrict_ preg
, const char *_Restrict_ string
,
6528 size_t nmatch
, regmatch_t pmatch
[_Restrict_arr_
], int eflags
)
6531 struct re_registers regs
;
6532 regex_t private_preg
;
6533 size_t len
= strlen (string
);
6534 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6536 private_preg
= *preg
;
6538 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6539 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6541 /* The user has told us exactly how many registers to return
6542 information about, via `nmatch'. We have to pass that on to the
6543 matching routines. */
6544 private_preg
.regs_allocated
= REGS_FIXED
;
6548 regs
.num_regs
= nmatch
;
6549 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6550 if (regs
.start
== NULL
)
6552 regs
.end
= regs
.start
+ nmatch
;
6555 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6556 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6557 was a little bit longer but still only matching the real part.
6558 This works because the `endline' will check for a '\n' and will find a
6559 '\0', correctly deciding that this is not the end of a line.
6560 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6561 a convenient '\0' there. For all we know, the string could be preceded
6562 by '\n' which would throw things off. */
6564 /* Perform the searching operation. */
6565 ret
= re_search (&private_preg
, string
, len
,
6566 /* start: */ 0, /* range: */ len
,
6567 want_reg_info
? ®s
: 0);
6569 /* Copy the register information to the POSIX structure. */
6576 for (r
= 0; r
< nmatch
; r
++)
6578 pmatch
[r
].rm_so
= regs
.start
[r
];
6579 pmatch
[r
].rm_eo
= regs
.end
[r
];
6583 /* If we needed the temporary register info, free the space now. */
6587 /* We want zero return to mean success, unlike `re_search'. */
6588 return ret
>= 0 ? REG_NOERROR
: REG_NOMATCH
;
6590 WEAK_ALIAS (__regexec
, regexec
)
6593 /* Returns a message corresponding to an error code, ERR_CODE, returned
6594 from either regcomp or regexec. We don't use PREG here.
6596 ERR_CODE was previously called ERRCODE, but that name causes an
6597 error with msvc8 compiler. */
6600 regerror (int err_code
, const regex_t
*preg
, char *errbuf
, size_t errbuf_size
)
6606 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6607 /* Only error codes returned by the rest of the code should be passed
6608 to this routine. If we are given anything else, or if other regex
6609 code generates an invalid error code, then the program has a bug.
6610 Dump core so we can fix it. */
6613 msg
= gettext (re_error_msgid
[err_code
]);
6615 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6617 if (errbuf_size
!= 0)
6619 if (msg_size
> errbuf_size
)
6621 memcpy (errbuf
, msg
, errbuf_size
- 1);
6622 errbuf
[errbuf_size
- 1] = 0;
6625 strcpy (errbuf
, msg
);
6630 WEAK_ALIAS (__regerror
, regerror
)
6633 /* Free dynamically allocated space used by PREG. */
6636 regfree (regex_t
*preg
)
6638 free (preg
->buffer
);
6639 preg
->buffer
= NULL
;
6641 preg
->allocated
= 0;
6644 free (preg
->fastmap
);
6645 preg
->fastmap
= NULL
;
6646 preg
->fastmap_accurate
= 0;
6648 free (preg
->translate
);
6649 preg
->translate
= NULL
;
6651 WEAK_ALIAS (__regfree
, regfree
)
6653 #endif /* not emacs */