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. */
1869 #define BIT_WORD 0x1
1870 #define BIT_LOWER 0x2
1871 #define BIT_PUNCT 0x4
1872 #define BIT_SPACE 0x8
1873 #define BIT_UPPER 0x10
1874 #define BIT_MULTIBYTE 0x20
1875 #define BIT_ALPHA 0x40
1876 #define BIT_ALNUM 0x80
1879 /* Set the bit for character C in a list. */
1880 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1885 /* Store characters in the range FROM to TO in the bitmap at B (for
1886 ASCII and unibyte characters) and WORK_AREA (for multibyte
1887 characters) while translating them and paying attention to the
1888 continuity of translated characters.
1890 Implementation note: It is better to implement these fairly big
1891 macros by a function, but it's not that easy because macros called
1892 in this macro assume various local variables already declared. */
1894 /* Both FROM and TO are ASCII characters. */
1896 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1900 for (C0 = (FROM); C0 <= (TO); C0++) \
1902 C1 = TRANSLATE (C0); \
1903 if (! ASCII_CHAR_P (C1)) \
1905 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1906 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1909 SET_LIST_BIT (C1); \
1914 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1916 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1918 int C0, C1, C2, I; \
1919 int USED = RANGE_TABLE_WORK_USED (work_area); \
1921 for (C0 = (FROM); C0 <= (TO); C0++) \
1923 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1924 if (CHAR_BYTE8_P (C1)) \
1925 SET_LIST_BIT (C0); \
1928 C2 = TRANSLATE (C1); \
1930 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1932 SET_LIST_BIT (C1); \
1933 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1935 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1936 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1938 if (C2 >= from - 1 && C2 <= to + 1) \
1940 if (C2 == from - 1) \
1941 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1942 else if (C2 == to + 1) \
1943 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1948 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1954 /* Both FROM and TO are multibyte characters. */
1956 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1958 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1960 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1961 for (C0 = (FROM); C0 <= (TO); C0++) \
1963 C1 = TRANSLATE (C0); \
1964 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1965 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1966 SET_LIST_BIT (C2); \
1967 if (C1 >= (FROM) && C1 <= (TO)) \
1969 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1971 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1972 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1974 if (C1 >= from - 1 && C1 <= to + 1) \
1976 if (C1 == from - 1) \
1977 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1978 else if (C1 == to + 1) \
1979 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1984 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1990 /* Get the next unsigned number in the uncompiled pattern. */
1991 #define GET_INTERVAL_COUNT(num) \
1994 FREE_STACK_RETURN (REG_EBRACE); \
1998 while ('0' <= c && c <= '9') \
2002 if (RE_DUP_MAX / 10 - (RE_DUP_MAX % 10 < c - '0') < num) \
2003 FREE_STACK_RETURN (REG_BADBR); \
2004 num = num * 10 + c - '0'; \
2006 FREE_STACK_RETURN (REG_EBRACE); \
2012 #if ! WIDE_CHAR_SUPPORT
2014 /* Map a string to the char class it names (if any). */
2016 re_wctype (const_re_char
*str
)
2018 const char *string
= (const char *) str
;
2019 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2020 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2021 else if (STREQ (string
, "word")) return RECC_WORD
;
2022 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2023 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2024 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2025 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2026 else if (STREQ (string
, "print")) return RECC_PRINT
;
2027 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2028 else if (STREQ (string
, "space")) return RECC_SPACE
;
2029 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2030 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2031 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2032 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2033 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2034 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2035 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2039 /* True if CH is in the char class CC. */
2041 re_iswctype (int ch
, re_wctype_t cc
)
2045 case RECC_ALNUM
: return ISALNUM (ch
) != 0;
2046 case RECC_ALPHA
: return ISALPHA (ch
) != 0;
2047 case RECC_BLANK
: return ISBLANK (ch
) != 0;
2048 case RECC_CNTRL
: return ISCNTRL (ch
) != 0;
2049 case RECC_DIGIT
: return ISDIGIT (ch
) != 0;
2050 case RECC_GRAPH
: return ISGRAPH (ch
) != 0;
2051 case RECC_LOWER
: return ISLOWER (ch
) != 0;
2052 case RECC_PRINT
: return ISPRINT (ch
) != 0;
2053 case RECC_PUNCT
: return ISPUNCT (ch
) != 0;
2054 case RECC_SPACE
: return ISSPACE (ch
) != 0;
2055 case RECC_UPPER
: return ISUPPER (ch
) != 0;
2056 case RECC_XDIGIT
: return ISXDIGIT (ch
) != 0;
2057 case RECC_ASCII
: return IS_REAL_ASCII (ch
) != 0;
2058 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2059 case RECC_UNIBYTE
: return ISUNIBYTE (ch
) != 0;
2060 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2061 case RECC_WORD
: return ISWORD (ch
) != 0;
2062 case RECC_ERROR
: return false;
2068 /* Return a bit-pattern to use in the range-table bits to match multibyte
2069 chars of class CC. */
2071 re_wctype_to_bit (re_wctype_t cc
)
2075 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2076 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2077 case RECC_ALPHA
: return BIT_ALPHA
;
2078 case RECC_ALNUM
: return BIT_ALNUM
;
2079 case RECC_WORD
: return BIT_WORD
;
2080 case RECC_LOWER
: return BIT_LOWER
;
2081 case RECC_UPPER
: return BIT_UPPER
;
2082 case RECC_PUNCT
: return BIT_PUNCT
;
2083 case RECC_SPACE
: return BIT_SPACE
;
2084 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2085 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2092 /* Filling in the work area of a range. */
2094 /* Actually extend the space in WORK_AREA. */
2097 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2099 work_area
->allocated
+= 16 * sizeof (int);
2100 work_area
->table
= realloc (work_area
->table
, work_area
->allocated
);
2106 /* Carefully find the ranges of codes that are equivalent
2107 under case conversion to the range start..end when passed through
2108 TRANSLATE. Handle the case where non-letters can come in between
2109 two upper-case letters (which happens in Latin-1).
2110 Also handle the case of groups of more than 2 case-equivalent chars.
2112 The basic method is to look at consecutive characters and see
2113 if they can form a run that can be handled as one.
2115 Returns -1 if successful, REG_ESPACE if ran out of space. */
2118 set_image_of_range_1 (struct range_table_work_area
*work_area
,
2119 re_wchar_t start
, re_wchar_t end
,
2120 RE_TRANSLATE_TYPE translate
)
2122 /* `one_case' indicates a character, or a run of characters,
2123 each of which is an isolate (no case-equivalents).
2124 This includes all ASCII non-letters.
2126 `two_case' indicates a character, or a run of characters,
2127 each of which has two case-equivalent forms.
2128 This includes all ASCII letters.
2130 `strange' indicates a character that has more than one
2133 enum case_type
{one_case
, two_case
, strange
};
2135 /* Describe the run that is in progress,
2136 which the next character can try to extend.
2137 If run_type is strange, that means there really is no run.
2138 If run_type is one_case, then run_start...run_end is the run.
2139 If run_type is two_case, then the run is run_start...run_end,
2140 and the case-equivalents end at run_eqv_end. */
2142 enum case_type run_type
= strange
;
2143 int run_start
, run_end
, run_eqv_end
;
2145 Lisp_Object eqv_table
;
2147 if (!RE_TRANSLATE_P (translate
))
2149 EXTEND_RANGE_TABLE (work_area
, 2);
2150 work_area
->table
[work_area
->used
++] = (start
);
2151 work_area
->table
[work_area
->used
++] = (end
);
2155 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2157 for (; start
<= end
; start
++)
2159 enum case_type this_type
;
2160 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2161 int minchar
, maxchar
;
2163 /* Classify this character */
2165 this_type
= one_case
;
2166 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2167 this_type
= two_case
;
2169 this_type
= strange
;
2172 minchar
= start
, maxchar
= eqv
;
2174 minchar
= eqv
, maxchar
= start
;
2176 /* Can this character extend the run in progress? */
2177 if (this_type
== strange
|| this_type
!= run_type
2178 || !(minchar
== run_end
+ 1
2179 && (run_type
== two_case
2180 ? maxchar
== run_eqv_end
+ 1 : 1)))
2183 Record each of its equivalent ranges. */
2184 if (run_type
== one_case
)
2186 EXTEND_RANGE_TABLE (work_area
, 2);
2187 work_area
->table
[work_area
->used
++] = run_start
;
2188 work_area
->table
[work_area
->used
++] = run_end
;
2190 else if (run_type
== two_case
)
2192 EXTEND_RANGE_TABLE (work_area
, 4);
2193 work_area
->table
[work_area
->used
++] = run_start
;
2194 work_area
->table
[work_area
->used
++] = run_end
;
2195 work_area
->table
[work_area
->used
++]
2196 = RE_TRANSLATE (eqv_table
, run_start
);
2197 work_area
->table
[work_area
->used
++]
2198 = RE_TRANSLATE (eqv_table
, run_end
);
2203 if (this_type
== strange
)
2205 /* For a strange character, add each of its equivalents, one
2206 by one. Don't start a range. */
2209 EXTEND_RANGE_TABLE (work_area
, 2);
2210 work_area
->table
[work_area
->used
++] = eqv
;
2211 work_area
->table
[work_area
->used
++] = eqv
;
2212 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2214 while (eqv
!= start
);
2217 /* Add this char to the run, or start a new run. */
2218 else if (run_type
== strange
)
2220 /* Initialize a new range. */
2221 run_type
= this_type
;
2224 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2228 /* Extend a running range. */
2230 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2234 /* If a run is still in progress at the end, finish it now
2235 by recording its equivalent ranges. */
2236 if (run_type
== one_case
)
2238 EXTEND_RANGE_TABLE (work_area
, 2);
2239 work_area
->table
[work_area
->used
++] = run_start
;
2240 work_area
->table
[work_area
->used
++] = run_end
;
2242 else if (run_type
== two_case
)
2244 EXTEND_RANGE_TABLE (work_area
, 4);
2245 work_area
->table
[work_area
->used
++] = run_start
;
2246 work_area
->table
[work_area
->used
++] = run_end
;
2247 work_area
->table
[work_area
->used
++]
2248 = RE_TRANSLATE (eqv_table
, run_start
);
2249 work_area
->table
[work_area
->used
++]
2250 = RE_TRANSLATE (eqv_table
, run_end
);
2258 /* Record the image of the range start..end when passed through
2259 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2260 and is not even necessarily contiguous.
2261 Normally we approximate it with the smallest contiguous range that contains
2262 all the chars we need. However, for Latin-1 we go to extra effort
2265 This function is not called for ASCII ranges.
2267 Returns -1 if successful, REG_ESPACE if ran out of space. */
2270 set_image_of_range (struct range_table_work_area
*work_area
,
2271 re_wchar_t start
, re_wchar_t end
,
2272 RE_TRANSLATE_TYPE translate
)
2274 re_wchar_t cmin
, cmax
;
2277 /* For Latin-1 ranges, use set_image_of_range_1
2278 to get proper handling of ranges that include letters and nonletters.
2279 For a range that includes the whole of Latin-1, this is not necessary.
2280 For other character sets, we don't bother to get this right. */
2281 if (RE_TRANSLATE_P (translate
) && start
< 04400
2282 && !(start
< 04200 && end
>= 04377))
2289 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2299 EXTEND_RANGE_TABLE (work_area
, 2);
2300 work_area
->table
[work_area
->used
++] = (start
);
2301 work_area
->table
[work_area
->used
++] = (end
);
2303 cmin
= -1, cmax
= -1;
2305 if (RE_TRANSLATE_P (translate
))
2309 for (ch
= start
; ch
<= end
; ch
++)
2311 re_wchar_t c
= TRANSLATE (ch
);
2312 if (! (start
<= c
&& c
<= end
))
2318 cmin
= min (cmin
, c
);
2319 cmax
= max (cmax
, c
);
2326 EXTEND_RANGE_TABLE (work_area
, 2);
2327 work_area
->table
[work_area
->used
++] = (cmin
);
2328 work_area
->table
[work_area
->used
++] = (cmax
);
2336 #ifndef MATCH_MAY_ALLOCATE
2338 /* If we cannot allocate large objects within re_match_2_internal,
2339 we make the fail stack and register vectors global.
2340 The fail stack, we grow to the maximum size when a regexp
2342 The register vectors, we adjust in size each time we
2343 compile a regexp, according to the number of registers it needs. */
2345 static fail_stack_type fail_stack
;
2347 /* Size with which the following vectors are currently allocated.
2348 That is so we can make them bigger as needed,
2349 but never make them smaller. */
2350 static int regs_allocated_size
;
2352 static re_char
** regstart
, ** regend
;
2353 static re_char
**best_regstart
, **best_regend
;
2355 /* Make the register vectors big enough for NUM_REGS registers,
2356 but don't make them smaller. */
2359 regex_grow_registers (int num_regs
)
2361 if (num_regs
> regs_allocated_size
)
2363 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2364 RETALLOC_IF (regend
, num_regs
, re_char
*);
2365 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2366 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2368 regs_allocated_size
= num_regs
;
2372 #endif /* not MATCH_MAY_ALLOCATE */
2374 static boolean
group_in_compile_stack (compile_stack_type compile_stack
,
2377 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2378 Returns one of error codes defined in `regex.h', or zero for success.
2380 Assumes the `allocated' (and perhaps `buffer') and `translate'
2381 fields are set in BUFP on entry.
2383 If it succeeds, results are put in BUFP (if it returns an error, the
2384 contents of BUFP are undefined):
2385 `buffer' is the compiled pattern;
2386 `syntax' is set to SYNTAX;
2387 `used' is set to the length of the compiled pattern;
2388 `fastmap_accurate' is zero;
2389 `re_nsub' is the number of subexpressions in PATTERN;
2390 `not_bol' and `not_eol' are zero;
2392 The `fastmap' field is neither examined nor set. */
2394 /* Insert the `jump' from the end of last alternative to "here".
2395 The space for the jump has already been allocated. */
2396 #define FIXUP_ALT_JUMP() \
2398 if (fixup_alt_jump) \
2399 STORE_JUMP (jump, fixup_alt_jump, b); \
2403 /* Return, freeing storage we allocated. */
2404 #define FREE_STACK_RETURN(value) \
2406 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2407 free (compile_stack.stack); \
2411 static reg_errcode_t
2412 regex_compile (const_re_char
*pattern
, size_t size
, reg_syntax_t syntax
,
2413 struct re_pattern_buffer
*bufp
)
2415 /* We fetch characters from PATTERN here. */
2416 register re_wchar_t c
, c1
;
2418 /* Points to the end of the buffer, where we should append. */
2419 register unsigned char *b
;
2421 /* Keeps track of unclosed groups. */
2422 compile_stack_type compile_stack
;
2424 /* Points to the current (ending) position in the pattern. */
2426 /* `const' makes AIX compiler fail. */
2427 unsigned char *p
= pattern
;
2429 re_char
*p
= pattern
;
2431 re_char
*pend
= pattern
+ size
;
2433 /* How to translate the characters in the pattern. */
2434 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2436 /* Address of the count-byte of the most recently inserted `exactn'
2437 command. This makes it possible to tell if a new exact-match
2438 character can be added to that command or if the character requires
2439 a new `exactn' command. */
2440 unsigned char *pending_exact
= 0;
2442 /* Address of start of the most recently finished expression.
2443 This tells, e.g., postfix * where to find the start of its
2444 operand. Reset at the beginning of groups and alternatives. */
2445 unsigned char *laststart
= 0;
2447 /* Address of beginning of regexp, or inside of last group. */
2448 unsigned char *begalt
;
2450 /* Place in the uncompiled pattern (i.e., the {) to
2451 which to go back if the interval is invalid. */
2452 re_char
*beg_interval
;
2454 /* Address of the place where a forward jump should go to the end of
2455 the containing expression. Each alternative of an `or' -- except the
2456 last -- ends with a forward jump of this sort. */
2457 unsigned char *fixup_alt_jump
= 0;
2459 /* Work area for range table of charset. */
2460 struct range_table_work_area range_table_work
;
2462 /* If the object matched can contain multibyte characters. */
2463 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2465 /* Nonzero if we have pushed down into a subpattern. */
2466 int in_subpattern
= 0;
2468 /* These hold the values of p, pattern, and pend from the main
2469 pattern when we have pushed into a subpattern. */
2470 re_char
*main_p
IF_LINT (= NULL
);
2471 re_char
*main_pattern
IF_LINT (= NULL
);
2472 re_char
*main_pend
IF_LINT (= NULL
);
2476 DEBUG_PRINT ("\nCompiling pattern: ");
2479 unsigned debug_count
;
2481 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2482 putchar (pattern
[debug_count
]);
2487 /* Initialize the compile stack. */
2488 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2489 if (compile_stack
.stack
== NULL
)
2492 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2493 compile_stack
.avail
= 0;
2495 range_table_work
.table
= 0;
2496 range_table_work
.allocated
= 0;
2498 /* Initialize the pattern buffer. */
2499 bufp
->syntax
= syntax
;
2500 bufp
->fastmap_accurate
= 0;
2501 bufp
->not_bol
= bufp
->not_eol
= 0;
2502 bufp
->used_syntax
= 0;
2504 /* Set `used' to zero, so that if we return an error, the pattern
2505 printer (for debugging) will think there's no pattern. We reset it
2509 /* Always count groups, whether or not bufp->no_sub is set. */
2512 #if !defined emacs && !defined SYNTAX_TABLE
2513 /* Initialize the syntax table. */
2514 init_syntax_once ();
2517 if (bufp
->allocated
== 0)
2520 { /* If zero allocated, but buffer is non-null, try to realloc
2521 enough space. This loses if buffer's address is bogus, but
2522 that is the user's responsibility. */
2523 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2526 { /* Caller did not allocate a buffer. Do it for them. */
2527 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2529 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2531 bufp
->allocated
= INIT_BUF_SIZE
;
2534 begalt
= b
= bufp
->buffer
;
2536 /* Loop through the uncompiled pattern until we're at the end. */
2541 /* If this is the end of an included regexp,
2542 pop back to the main regexp and try again. */
2546 pattern
= main_pattern
;
2551 /* If this is the end of the main regexp, we are done. */
2563 /* If there's no special whitespace regexp, treat
2564 spaces normally. And don't try to do this recursively. */
2565 if (!whitespace_regexp
|| in_subpattern
)
2568 /* Peek past following spaces. */
2575 /* If the spaces are followed by a repetition op,
2576 treat them normally. */
2578 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2579 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2582 /* Replace the spaces with the whitespace regexp. */
2586 main_pattern
= pattern
;
2587 p
= pattern
= whitespace_regexp
;
2588 pend
= p
+ strlen ((const char *) p
);
2594 if ( /* If at start of pattern, it's an operator. */
2596 /* If context independent, it's an operator. */
2597 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2598 /* Otherwise, depends on what's come before. */
2599 || at_begline_loc_p (pattern
, p
, syntax
))
2600 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2609 if ( /* If at end of pattern, it's an operator. */
2611 /* If context independent, it's an operator. */
2612 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2613 /* Otherwise, depends on what's next. */
2614 || at_endline_loc_p (p
, pend
, syntax
))
2615 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2624 if ((syntax
& RE_BK_PLUS_QM
)
2625 || (syntax
& RE_LIMITED_OPS
))
2629 /* If there is no previous pattern... */
2632 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2633 FREE_STACK_RETURN (REG_BADRPT
);
2634 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2639 /* 1 means zero (many) matches is allowed. */
2640 boolean zero_times_ok
= 0, many_times_ok
= 0;
2643 /* If there is a sequence of repetition chars, collapse it
2644 down to just one (the right one). We can't combine
2645 interval operators with these because of, e.g., `a{2}*',
2646 which should only match an even number of `a's. */
2650 if ((syntax
& RE_FRUGAL
)
2651 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2655 zero_times_ok
|= c
!= '+';
2656 many_times_ok
|= c
!= '?';
2662 || (!(syntax
& RE_BK_PLUS_QM
)
2663 && (*p
== '+' || *p
== '?')))
2665 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2668 FREE_STACK_RETURN (REG_EESCAPE
);
2669 if (p
[1] == '+' || p
[1] == '?')
2670 PATFETCH (c
); /* Gobble up the backslash. */
2676 /* If we get here, we found another repeat character. */
2680 /* Star, etc. applied to an empty pattern is equivalent
2681 to an empty pattern. */
2682 if (!laststart
|| laststart
== b
)
2685 /* Now we know whether or not zero matches is allowed
2686 and also whether or not two or more matches is allowed. */
2691 boolean simple
= skip_one_char (laststart
) == b
;
2692 size_t startoffset
= 0;
2694 /* Check if the loop can match the empty string. */
2695 (simple
|| !analyze_first (laststart
, b
, NULL
, 0))
2696 ? on_failure_jump
: on_failure_jump_loop
;
2697 assert (skip_one_char (laststart
) <= b
);
2699 if (!zero_times_ok
&& simple
)
2700 { /* Since simple * loops can be made faster by using
2701 on_failure_keep_string_jump, we turn simple P+
2702 into PP* if P is simple. */
2703 unsigned char *p1
, *p2
;
2704 startoffset
= b
- laststart
;
2705 GET_BUFFER_SPACE (startoffset
);
2706 p1
= b
; p2
= laststart
;
2712 GET_BUFFER_SPACE (6);
2715 STORE_JUMP (ofj
, b
, b
+ 6);
2717 /* Simple * loops can use on_failure_keep_string_jump
2718 depending on what follows. But since we don't know
2719 that yet, we leave the decision up to
2720 on_failure_jump_smart. */
2721 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2722 laststart
+ startoffset
, b
+ 6);
2724 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2729 /* A simple ? pattern. */
2730 assert (zero_times_ok
);
2731 GET_BUFFER_SPACE (3);
2732 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2736 else /* not greedy */
2737 { /* I wish the greedy and non-greedy cases could be merged. */
2739 GET_BUFFER_SPACE (7); /* We might use less. */
2742 boolean emptyp
= analyze_first (laststart
, b
, NULL
, 0);
2744 /* The non-greedy multiple match looks like
2745 a repeat..until: we only need a conditional jump
2746 at the end of the loop. */
2747 if (emptyp
) BUF_PUSH (no_op
);
2748 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2749 : on_failure_jump
, b
, laststart
);
2753 /* The repeat...until naturally matches one or more.
2754 To also match zero times, we need to first jump to
2755 the end of the loop (its conditional jump). */
2756 INSERT_JUMP (jump
, laststart
, b
);
2762 /* non-greedy a?? */
2763 INSERT_JUMP (jump
, laststart
, b
+ 3);
2765 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2784 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2786 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2788 /* Ensure that we have enough space to push a charset: the
2789 opcode, the length count, and the bitset; 34 bytes in all. */
2790 GET_BUFFER_SPACE (34);
2794 /* We test `*p == '^' twice, instead of using an if
2795 statement, so we only need one BUF_PUSH. */
2796 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2800 /* Remember the first position in the bracket expression. */
2803 /* Push the number of bytes in the bitmap. */
2804 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2806 /* Clear the whole map. */
2807 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2809 /* charset_not matches newline according to a syntax bit. */
2810 if ((re_opcode_t
) b
[-2] == charset_not
2811 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2812 SET_LIST_BIT ('\n');
2814 /* Read in characters and ranges, setting map bits. */
2817 boolean escaped_char
= false;
2818 const unsigned char *p2
= p
;
2821 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2823 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2824 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2825 So the translation is done later in a loop. Example:
2826 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2829 /* \ might escape characters inside [...] and [^...]. */
2830 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2832 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2835 escaped_char
= true;
2839 /* Could be the end of the bracket expression. If it's
2840 not (i.e., when the bracket expression is `[]' so
2841 far), the ']' character bit gets set way below. */
2842 if (c
== ']' && p2
!= p1
)
2846 /* See if we're at the beginning of a possible character
2849 if (!escaped_char
&&
2850 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2852 /* Leave room for the null. */
2853 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2854 const unsigned char *class_beg
;
2860 /* If pattern is `[[:'. */
2861 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2866 if ((c
== ':' && *p
== ']') || p
== pend
)
2868 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2871 /* This is in any case an invalid class name. */
2876 /* If isn't a word bracketed by `[:' and `:]':
2877 undo the ending character, the letters, and
2878 leave the leading `:' and `[' (but set bits for
2880 if (c
== ':' && *p
== ']')
2882 re_wctype_t cc
= re_wctype (str
);
2885 FREE_STACK_RETURN (REG_ECTYPE
);
2887 /* Throw away the ] at the end of the character
2891 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2894 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2895 if (re_iswctype (btowc (ch
), cc
))
2898 if (c
< (1 << BYTEWIDTH
))
2902 /* Most character classes in a multibyte match
2903 just set a flag. Exceptions are is_blank,
2904 is_digit, is_cntrl, and is_xdigit, since
2905 they can only match ASCII characters. We
2906 don't need to handle them for multibyte.
2907 They are distinguished by a negative wctype. */
2909 /* Setup the gl_state object to its buffer-defined
2910 value. This hardcodes the buffer-global
2911 syntax-table for ASCII chars, while the other chars
2912 will obey syntax-table properties. It's not ideal,
2913 but it's the way it's been done until now. */
2914 SETUP_BUFFER_SYNTAX_TABLE ();
2916 for (ch
= 0; ch
< 256; ++ch
)
2918 c
= RE_CHAR_TO_MULTIBYTE (ch
);
2919 if (! CHAR_BYTE8_P (c
)
2920 && re_iswctype (c
, cc
))
2926 if (ASCII_CHAR_P (c1
))
2928 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
2932 SET_RANGE_TABLE_WORK_AREA_BIT
2933 (range_table_work
, re_wctype_to_bit (cc
));
2935 /* In most cases the matching rule for char classes
2936 only uses the syntax table for multibyte chars,
2937 so that the content of the syntax-table is not
2938 hardcoded in the range_table. SPACE and WORD are
2939 the two exceptions. */
2940 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2941 bufp
->used_syntax
= 1;
2943 /* Repeat the loop. */
2948 /* Go back to right after the "[:". */
2952 /* Because the `:' may start the range, we
2953 can't simply set bit and repeat the loop.
2954 Instead, just set it to C and handle below. */
2959 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2962 /* Discard the `-'. */
2965 /* Fetch the character which ends the range. */
2968 if (CHAR_BYTE8_P (c1
)
2969 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
2970 /* Treat the range from a multibyte character to
2971 raw-byte character as empty. */
2976 /* Range from C to C. */
2981 if (syntax
& RE_NO_EMPTY_RANGES
)
2982 FREE_STACK_RETURN (REG_ERANGEX
);
2983 /* Else, repeat the loop. */
2988 /* Set the range into bitmap */
2989 for (; c
<= c1
; c
++)
2992 if (ch
< (1 << BYTEWIDTH
))
2999 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
3001 if (CHAR_BYTE8_P (c1
))
3002 c
= BYTE8_TO_CHAR (128);
3006 if (CHAR_BYTE8_P (c
))
3008 c
= CHAR_TO_BYTE8 (c
);
3009 c1
= CHAR_TO_BYTE8 (c1
);
3010 for (; c
<= c1
; c
++)
3015 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
3019 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
3026 /* Discard any (non)matching list bytes that are all 0 at the
3027 end of the map. Decrease the map-length byte too. */
3028 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3032 /* Build real range table from work area. */
3033 if (RANGE_TABLE_WORK_USED (range_table_work
)
3034 || RANGE_TABLE_WORK_BITS (range_table_work
))
3037 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3039 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3040 bytes for flags, two for COUNT, and three bytes for
3042 GET_BUFFER_SPACE (4 + used
* 3);
3044 /* Indicate the existence of range table. */
3045 laststart
[1] |= 0x80;
3047 /* Store the character class flag bits into the range table.
3048 If not in emacs, these flag bits are always 0. */
3049 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3050 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3052 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3053 for (i
= 0; i
< used
; i
++)
3054 STORE_CHARACTER_AND_INCR
3055 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3062 if (syntax
& RE_NO_BK_PARENS
)
3069 if (syntax
& RE_NO_BK_PARENS
)
3076 if (syntax
& RE_NEWLINE_ALT
)
3083 if (syntax
& RE_NO_BK_VBAR
)
3090 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3091 goto handle_interval
;
3097 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3099 /* Do not translate the character after the \, so that we can
3100 distinguish, e.g., \B from \b, even if we normally would
3101 translate, e.g., B to b. */
3107 if (syntax
& RE_NO_BK_PARENS
)
3108 goto normal_backslash
;
3113 regnum_t regnum
= 0;
3116 /* Look for a special (?...) construct */
3117 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3119 PATFETCH (c
); /* Gobble up the '?'. */
3125 case ':': shy
= 1; break;
3127 /* An explicitly specified regnum must start
3130 FREE_STACK_RETURN (REG_BADPAT
);
3131 case '1': case '2': case '3': case '4':
3132 case '5': case '6': case '7': case '8': case '9':
3133 regnum
= 10*regnum
+ (c
- '0'); break;
3135 /* Only (?:...) is supported right now. */
3136 FREE_STACK_RETURN (REG_BADPAT
);
3143 regnum
= ++bufp
->re_nsub
;
3145 { /* It's actually not shy, but explicitly numbered. */
3147 if (regnum
> bufp
->re_nsub
)
3148 bufp
->re_nsub
= regnum
;
3149 else if (regnum
> bufp
->re_nsub
3150 /* Ideally, we'd want to check that the specified
3151 group can't have matched (i.e. all subgroups
3152 using the same regnum are in other branches of
3153 OR patterns), but we don't currently keep track
3154 of enough info to do that easily. */
3155 || group_in_compile_stack (compile_stack
, regnum
))
3156 FREE_STACK_RETURN (REG_BADPAT
);
3159 /* It's really shy. */
3160 regnum
= - bufp
->re_nsub
;
3162 if (COMPILE_STACK_FULL
)
3164 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3165 compile_stack_elt_t
);
3166 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3168 compile_stack
.size
<<= 1;
3171 /* These are the values to restore when we hit end of this
3172 group. They are all relative offsets, so that if the
3173 whole pattern moves because of realloc, they will still
3175 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3176 COMPILE_STACK_TOP
.fixup_alt_jump
3177 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3178 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3179 COMPILE_STACK_TOP
.regnum
= regnum
;
3181 /* Do not push a start_memory for groups beyond the last one
3182 we can represent in the compiled pattern. */
3183 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3184 BUF_PUSH_2 (start_memory
, regnum
);
3186 compile_stack
.avail
++;
3191 /* If we've reached MAX_REGNUM groups, then this open
3192 won't actually generate any code, so we'll have to
3193 clear pending_exact explicitly. */
3199 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3201 if (COMPILE_STACK_EMPTY
)
3203 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3204 goto normal_backslash
;
3206 FREE_STACK_RETURN (REG_ERPAREN
);
3212 /* See similar code for backslashed left paren above. */
3213 if (COMPILE_STACK_EMPTY
)
3215 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3218 FREE_STACK_RETURN (REG_ERPAREN
);
3221 /* Since we just checked for an empty stack above, this
3222 ``can't happen''. */
3223 assert (compile_stack
.avail
!= 0);
3225 /* We don't just want to restore into `regnum', because
3226 later groups should continue to be numbered higher,
3227 as in `(ab)c(de)' -- the second group is #2. */
3230 compile_stack
.avail
--;
3231 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3233 = COMPILE_STACK_TOP
.fixup_alt_jump
3234 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3236 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3237 regnum
= COMPILE_STACK_TOP
.regnum
;
3238 /* If we've reached MAX_REGNUM groups, then this open
3239 won't actually generate any code, so we'll have to
3240 clear pending_exact explicitly. */
3243 /* We're at the end of the group, so now we know how many
3244 groups were inside this one. */
3245 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3246 BUF_PUSH_2 (stop_memory
, regnum
);
3251 case '|': /* `\|'. */
3252 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3253 goto normal_backslash
;
3255 if (syntax
& RE_LIMITED_OPS
)
3258 /* Insert before the previous alternative a jump which
3259 jumps to this alternative if the former fails. */
3260 GET_BUFFER_SPACE (3);
3261 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3265 /* The alternative before this one has a jump after it
3266 which gets executed if it gets matched. Adjust that
3267 jump so it will jump to this alternative's analogous
3268 jump (put in below, which in turn will jump to the next
3269 (if any) alternative's such jump, etc.). The last such
3270 jump jumps to the correct final destination. A picture:
3276 If we are at `b', then fixup_alt_jump right now points to a
3277 three-byte space after `a'. We'll put in the jump, set
3278 fixup_alt_jump to right after `b', and leave behind three
3279 bytes which we'll fill in when we get to after `c'. */
3283 /* Mark and leave space for a jump after this alternative,
3284 to be filled in later either by next alternative or
3285 when know we're at the end of a series of alternatives. */
3287 GET_BUFFER_SPACE (3);
3296 /* If \{ is a literal. */
3297 if (!(syntax
& RE_INTERVALS
)
3298 /* If we're at `\{' and it's not the open-interval
3300 || (syntax
& RE_NO_BK_BRACES
))
3301 goto normal_backslash
;
3305 /* If got here, then the syntax allows intervals. */
3307 /* At least (most) this many matches must be made. */
3308 int lower_bound
= 0, upper_bound
= -1;
3312 GET_INTERVAL_COUNT (lower_bound
);
3315 GET_INTERVAL_COUNT (upper_bound
);
3317 /* Interval such as `{1}' => match exactly once. */
3318 upper_bound
= lower_bound
;
3321 || (0 <= upper_bound
&& upper_bound
< lower_bound
))
3322 FREE_STACK_RETURN (REG_BADBR
);
3324 if (!(syntax
& RE_NO_BK_BRACES
))
3327 FREE_STACK_RETURN (REG_BADBR
);
3329 FREE_STACK_RETURN (REG_EESCAPE
);
3334 FREE_STACK_RETURN (REG_BADBR
);
3336 /* We just parsed a valid interval. */
3338 /* If it's invalid to have no preceding re. */
3341 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3342 FREE_STACK_RETURN (REG_BADRPT
);
3343 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3346 goto unfetch_interval
;
3349 if (upper_bound
== 0)
3350 /* If the upper bound is zero, just drop the sub pattern
3353 else if (lower_bound
== 1 && upper_bound
== 1)
3354 /* Just match it once: nothing to do here. */
3357 /* Otherwise, we have a nontrivial interval. When
3358 we're all done, the pattern will look like:
3359 set_number_at <jump count> <upper bound>
3360 set_number_at <succeed_n count> <lower bound>
3361 succeed_n <after jump addr> <succeed_n count>
3363 jump_n <succeed_n addr> <jump count>
3364 (The upper bound and `jump_n' are omitted if
3365 `upper_bound' is 1, though.) */
3367 { /* If the upper bound is > 1, we need to insert
3368 more at the end of the loop. */
3369 unsigned int nbytes
= (upper_bound
< 0 ? 3
3370 : upper_bound
> 1 ? 5 : 0);
3371 unsigned int startoffset
= 0;
3373 GET_BUFFER_SPACE (20); /* We might use less. */
3375 if (lower_bound
== 0)
3377 /* A succeed_n that starts with 0 is really a
3378 a simple on_failure_jump_loop. */
3379 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3385 /* Initialize lower bound of the `succeed_n', even
3386 though it will be set during matching by its
3387 attendant `set_number_at' (inserted next),
3388 because `re_compile_fastmap' needs to know.
3389 Jump to the `jump_n' we might insert below. */
3390 INSERT_JUMP2 (succeed_n
, laststart
,
3395 /* Code to initialize the lower bound. Insert
3396 before the `succeed_n'. The `5' is the last two
3397 bytes of this `set_number_at', plus 3 bytes of
3398 the following `succeed_n'. */
3399 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3404 if (upper_bound
< 0)
3406 /* A negative upper bound stands for infinity,
3407 in which case it degenerates to a plain jump. */
3408 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3411 else if (upper_bound
> 1)
3412 { /* More than one repetition is allowed, so
3413 append a backward jump to the `succeed_n'
3414 that starts this interval.
3416 When we've reached this during matching,
3417 we'll have matched the interval once, so
3418 jump back only `upper_bound - 1' times. */
3419 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3423 /* The location we want to set is the second
3424 parameter of the `jump_n'; that is `b-2' as
3425 an absolute address. `laststart' will be
3426 the `set_number_at' we're about to insert;
3427 `laststart+3' the number to set, the source
3428 for the relative address. But we are
3429 inserting into the middle of the pattern --
3430 so everything is getting moved up by 5.
3431 Conclusion: (b - 2) - (laststart + 3) + 5,
3432 i.e., b - laststart.
3434 We insert this at the beginning of the loop
3435 so that if we fail during matching, we'll
3436 reinitialize the bounds. */
3437 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3438 upper_bound
- 1, b
);
3443 beg_interval
= NULL
;
3448 /* If an invalid interval, match the characters as literals. */
3449 assert (beg_interval
);
3451 beg_interval
= NULL
;
3453 /* normal_char and normal_backslash need `c'. */
3456 if (!(syntax
& RE_NO_BK_BRACES
))
3458 assert (p
> pattern
&& p
[-1] == '\\');
3459 goto normal_backslash
;
3465 /* There is no way to specify the before_dot and after_dot
3466 operators. rms says this is ok. --karl */
3475 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3481 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3487 BUF_PUSH_2 (categoryspec
, c
);
3493 BUF_PUSH_2 (notcategoryspec
, c
);
3499 if (syntax
& RE_NO_GNU_OPS
)
3502 BUF_PUSH_2 (syntaxspec
, Sword
);
3507 if (syntax
& RE_NO_GNU_OPS
)
3510 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3515 if (syntax
& RE_NO_GNU_OPS
)
3522 if (syntax
& RE_NO_GNU_OPS
)
3529 if (syntax
& RE_NO_GNU_OPS
)
3538 FREE_STACK_RETURN (REG_BADPAT
);
3542 if (syntax
& RE_NO_GNU_OPS
)
3544 BUF_PUSH (wordbound
);
3548 if (syntax
& RE_NO_GNU_OPS
)
3550 BUF_PUSH (notwordbound
);
3554 if (syntax
& RE_NO_GNU_OPS
)
3560 if (syntax
& RE_NO_GNU_OPS
)
3565 case '1': case '2': case '3': case '4': case '5':
3566 case '6': case '7': case '8': case '9':
3570 if (syntax
& RE_NO_BK_REFS
)
3571 goto normal_backslash
;
3575 if (reg
> bufp
->re_nsub
|| reg
< 1
3576 /* Can't back reference to a subexp before its end. */
3577 || group_in_compile_stack (compile_stack
, reg
))
3578 FREE_STACK_RETURN (REG_ESUBREG
);
3581 BUF_PUSH_2 (duplicate
, reg
);
3588 if (syntax
& RE_BK_PLUS_QM
)
3591 goto normal_backslash
;
3595 /* You might think it would be useful for \ to mean
3596 not to translate; but if we don't translate it
3597 it will never match anything. */
3604 /* Expects the character in `c'. */
3606 /* If no exactn currently being built. */
3609 /* If last exactn not at current position. */
3610 || pending_exact
+ *pending_exact
+ 1 != b
3612 /* We have only one byte following the exactn for the count. */
3613 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3615 /* If followed by a repetition operator. */
3616 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3617 || ((syntax
& RE_BK_PLUS_QM
)
3618 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3619 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3620 || ((syntax
& RE_INTERVALS
)
3621 && ((syntax
& RE_NO_BK_BRACES
)
3622 ? p
!= pend
&& *p
== '{'
3623 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3625 /* Start building a new exactn. */
3629 BUF_PUSH_2 (exactn
, 0);
3630 pending_exact
= b
- 1;
3633 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3640 len
= CHAR_STRING (c
, b
);
3645 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3646 if (! CHAR_BYTE8_P (c1
))
3648 re_wchar_t c2
= TRANSLATE (c1
);
3650 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3656 (*pending_exact
) += len
;
3661 } /* while p != pend */
3664 /* Through the pattern now. */
3668 if (!COMPILE_STACK_EMPTY
)
3669 FREE_STACK_RETURN (REG_EPAREN
);
3671 /* If we don't want backtracking, force success
3672 the first time we reach the end of the compiled pattern. */
3673 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3676 /* We have succeeded; set the length of the buffer. */
3677 bufp
->used
= b
- bufp
->buffer
;
3682 re_compile_fastmap (bufp
);
3683 DEBUG_PRINT ("\nCompiled pattern: \n");
3684 print_compiled_pattern (bufp
);
3689 #ifndef MATCH_MAY_ALLOCATE
3690 /* Initialize the failure stack to the largest possible stack. This
3691 isn't necessary unless we're trying to avoid calling alloca in
3692 the search and match routines. */
3694 int num_regs
= bufp
->re_nsub
+ 1;
3696 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3698 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3699 falk_stack
.stack
= realloc (fail_stack
.stack
,
3700 fail_stack
.size
* sizeof *falk_stack
.stack
);
3703 regex_grow_registers (num_regs
);
3705 #endif /* not MATCH_MAY_ALLOCATE */
3707 FREE_STACK_RETURN (REG_NOERROR
);
3708 } /* regex_compile */
3710 /* Subroutines for `regex_compile'. */
3712 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3715 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3717 *loc
= (unsigned char) op
;
3718 STORE_NUMBER (loc
+ 1, arg
);
3722 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3725 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3727 *loc
= (unsigned char) op
;
3728 STORE_NUMBER (loc
+ 1, arg1
);
3729 STORE_NUMBER (loc
+ 3, arg2
);
3733 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3734 for OP followed by two-byte integer parameter ARG. */
3737 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3739 register unsigned char *pfrom
= end
;
3740 register unsigned char *pto
= end
+ 3;
3742 while (pfrom
!= loc
)
3745 store_op1 (op
, loc
, arg
);
3749 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3752 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3754 register unsigned char *pfrom
= end
;
3755 register unsigned char *pto
= end
+ 5;
3757 while (pfrom
!= loc
)
3760 store_op2 (op
, loc
, arg1
, arg2
);
3764 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3765 after an alternative or a begin-subexpression. We assume there is at
3766 least one character before the ^. */
3769 at_begline_loc_p (const_re_char
*pattern
, const_re_char
*p
, reg_syntax_t syntax
)
3771 re_char
*prev
= p
- 2;
3772 boolean odd_backslashes
;
3774 /* After a subexpression? */
3776 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3778 /* After an alternative? */
3779 else if (*prev
== '|')
3780 odd_backslashes
= (syntax
& RE_NO_BK_VBAR
) == 0;
3782 /* After a shy subexpression? */
3783 else if (*prev
== ':' && (syntax
& RE_SHY_GROUPS
))
3785 /* Skip over optional regnum. */
3786 while (prev
- 1 >= pattern
&& prev
[-1] >= '0' && prev
[-1] <= '9')
3789 if (!(prev
- 2 >= pattern
3790 && prev
[-1] == '?' && prev
[-2] == '('))
3793 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3798 /* Count the number of preceding backslashes. */
3800 while (prev
- 1 >= pattern
&& prev
[-1] == '\\')
3802 return (p
- prev
) & odd_backslashes
;
3806 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3807 at least one character after the $, i.e., `P < PEND'. */
3810 at_endline_loc_p (const_re_char
*p
, const_re_char
*pend
, reg_syntax_t syntax
)
3813 boolean next_backslash
= *next
== '\\';
3814 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3817 /* Before a subexpression? */
3818 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3819 : next_backslash
&& next_next
&& *next_next
== ')')
3820 /* Before an alternative? */
3821 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3822 : next_backslash
&& next_next
&& *next_next
== '|');
3826 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3827 false if it's not. */
3830 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3832 ssize_t this_element
;
3834 for (this_element
= compile_stack
.avail
- 1;
3837 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3844 If fastmap is non-NULL, go through the pattern and fill fastmap
3845 with all the possible leading chars. If fastmap is NULL, don't
3846 bother filling it up (obviously) and only return whether the
3847 pattern could potentially match the empty string.
3849 Return 1 if p..pend might match the empty string.
3850 Return 0 if p..pend matches at least one char.
3851 Return -1 if fastmap was not updated accurately. */
3854 analyze_first (const_re_char
*p
, const_re_char
*pend
, char *fastmap
,
3855 const int multibyte
)
3860 /* If all elements for base leading-codes in fastmap is set, this
3861 flag is set true. */
3862 boolean match_any_multibyte_characters
= false;
3866 /* The loop below works as follows:
3867 - It has a working-list kept in the PATTERN_STACK and which basically
3868 starts by only containing a pointer to the first operation.
3869 - If the opcode we're looking at is a match against some set of
3870 chars, then we add those chars to the fastmap and go on to the
3871 next work element from the worklist (done via `break').
3872 - If the opcode is a control operator on the other hand, we either
3873 ignore it (if it's meaningless at this point, such as `start_memory')
3874 or execute it (if it's a jump). If the jump has several destinations
3875 (i.e. `on_failure_jump'), then we push the other destination onto the
3877 We guarantee termination by ignoring backward jumps (more or less),
3878 so that `p' is monotonically increasing. More to the point, we
3879 never set `p' (or push) anything `<= p1'. */
3883 /* `p1' is used as a marker of how far back a `on_failure_jump'
3884 can go without being ignored. It is normally equal to `p'
3885 (which prevents any backward `on_failure_jump') except right
3886 after a plain `jump', to allow patterns such as:
3889 10: on_failure_jump 3
3890 as used for the *? operator. */
3899 /* If the first character has to match a backreference, that means
3900 that the group was empty (since it already matched). Since this
3901 is the only case that interests us here, we can assume that the
3902 backreference must match the empty string. */
3907 /* Following are the cases which match a character. These end
3913 /* If multibyte is nonzero, the first byte of each
3914 character is an ASCII or a leading code. Otherwise,
3915 each byte is a character. Thus, this works in both
3920 /* For the case of matching this unibyte regex
3921 against multibyte, we must set a leading code of
3922 the corresponding multibyte character. */
3923 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
3925 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
3932 /* We could put all the chars except for \n (and maybe \0)
3933 but we don't bother since it is generally not worth it. */
3934 if (!fastmap
) break;
3939 if (!fastmap
) break;
3941 /* Chars beyond end of bitmap are possible matches. */
3942 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3943 j
< (1 << BYTEWIDTH
); j
++)
3949 if (!fastmap
) break;
3950 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3951 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3953 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3957 if (/* Any leading code can possibly start a character
3958 which doesn't match the specified set of characters. */
3961 /* If we can match a character class, we can match any
3962 multibyte characters. */
3963 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3964 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3967 if (match_any_multibyte_characters
== false)
3969 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3970 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3972 match_any_multibyte_characters
= true;
3976 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3977 && match_any_multibyte_characters
== false)
3979 /* Set fastmap[I] to 1 where I is a leading code of each
3980 multibyte character in the range table. */
3982 unsigned char lc1
, lc2
;
3984 /* Make P points the range table. `+ 2' is to skip flag
3985 bits for a character class. */
3986 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3988 /* Extract the number of ranges in range table into COUNT. */
3989 EXTRACT_NUMBER_AND_INCR (count
, p
);
3990 for (; count
> 0; count
--, p
+= 3)
3992 /* Extract the start and end of each range. */
3993 EXTRACT_CHARACTER (c
, p
);
3994 lc1
= CHAR_LEADING_CODE (c
);
3996 EXTRACT_CHARACTER (c
, p
);
3997 lc2
= CHAR_LEADING_CODE (c
);
3998 for (j
= lc1
; j
<= lc2
; j
++)
4007 if (!fastmap
) break;
4009 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4011 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4012 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4016 /* This match depends on text properties. These end with
4017 aborting optimizations. */
4021 case notcategoryspec
:
4022 if (!fastmap
) break;
4023 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4025 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
4026 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4029 /* Any leading code can possibly start a character which
4030 has or doesn't has the specified category. */
4031 if (match_any_multibyte_characters
== false)
4033 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4034 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4036 match_any_multibyte_characters
= true;
4040 /* All cases after this match the empty string. These end with
4062 EXTRACT_NUMBER_AND_INCR (j
, p
);
4064 /* Backward jumps can only go back to code that we've already
4065 visited. `re_compile' should make sure this is true. */
4070 case on_failure_jump
:
4071 case on_failure_keep_string_jump
:
4072 case on_failure_jump_loop
:
4073 case on_failure_jump_nastyloop
:
4074 case on_failure_jump_smart
:
4080 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4081 to jump back to "just after here". */
4084 case on_failure_jump
:
4085 case on_failure_keep_string_jump
:
4086 case on_failure_jump_nastyloop
:
4087 case on_failure_jump_loop
:
4088 case on_failure_jump_smart
:
4089 EXTRACT_NUMBER_AND_INCR (j
, p
);
4091 ; /* Backward jump to be ignored. */
4093 { /* We have to look down both arms.
4094 We first go down the "straight" path so as to minimize
4095 stack usage when going through alternatives. */
4096 int r
= analyze_first (p
, pend
, fastmap
, multibyte
);
4104 /* This code simply does not properly handle forward jump_n. */
4105 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4107 /* jump_n can either jump or fall through. The (backward) jump
4108 case has already been handled, so we only need to look at the
4109 fallthrough case. */
4113 /* If N == 0, it should be an on_failure_jump_loop instead. */
4114 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4116 /* We only care about one iteration of the loop, so we don't
4117 need to consider the case where this behaves like an
4134 abort (); /* We have listed all the cases. */
4137 /* Getting here means we have found the possible starting
4138 characters for one path of the pattern -- and that the empty
4139 string does not match. We need not follow this path further. */
4143 /* We reached the end without matching anything. */
4146 } /* analyze_first */
4148 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4149 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4150 characters can start a string that matches the pattern. This fastmap
4151 is used by re_search to skip quickly over impossible starting points.
4153 Character codes above (1 << BYTEWIDTH) are not represented in the
4154 fastmap, but the leading codes are represented. Thus, the fastmap
4155 indicates which character sets could start a match.
4157 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4158 area as BUFP->fastmap.
4160 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4163 Returns 0 if we succeed, -2 if an internal error. */
4166 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4168 char *fastmap
= bufp
->fastmap
;
4171 assert (fastmap
&& bufp
->buffer
);
4173 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4174 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4176 analysis
= analyze_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4177 fastmap
, RE_MULTIBYTE_P (bufp
));
4178 bufp
->can_be_null
= (analysis
!= 0);
4180 } /* re_compile_fastmap */
4182 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4183 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4184 this memory for recording register information. STARTS and ENDS
4185 must be allocated using the malloc library routine, and must each
4186 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4188 If NUM_REGS == 0, then subsequent matches should allocate their own
4191 Unless this function is called, the first search or match using
4192 PATTERN_BUFFER will allocate its own register data, without
4193 freeing the old data. */
4196 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4200 bufp
->regs_allocated
= REGS_REALLOCATE
;
4201 regs
->num_regs
= num_regs
;
4202 regs
->start
= starts
;
4207 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4209 regs
->start
= regs
->end
= 0;
4212 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4214 /* Searching routines. */
4216 /* Like re_search_2, below, but only one string is specified, and
4217 doesn't let you say where to stop matching. */
4220 re_search (struct re_pattern_buffer
*bufp
, const char *string
, size_t size
,
4221 ssize_t startpos
, ssize_t range
, struct re_registers
*regs
)
4223 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4226 WEAK_ALIAS (__re_search
, re_search
)
4228 /* Head address of virtual concatenation of string. */
4229 #define HEAD_ADDR_VSTRING(P) \
4230 (((P) >= size1 ? string2 : string1))
4232 /* Address of POS in the concatenation of virtual string. */
4233 #define POS_ADDR_VSTRING(POS) \
4234 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4236 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4237 virtual concatenation of STRING1 and STRING2, starting first at index
4238 STARTPOS, then at STARTPOS + 1, and so on.
4240 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4242 RANGE is how far to scan while trying to match. RANGE = 0 means try
4243 only at STARTPOS; in general, the last start tried is STARTPOS +
4246 In REGS, return the indices of the virtual concatenation of STRING1
4247 and STRING2 that matched the entire BUFP->buffer and its contained
4250 Do not consider matching one past the index STOP in the virtual
4251 concatenation of STRING1 and STRING2.
4253 We return either the position in the strings at which the match was
4254 found, -1 if no match, or -2 if error (such as failure
4258 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, size_t size1
,
4259 const char *str2
, size_t size2
, ssize_t startpos
, ssize_t range
,
4260 struct re_registers
*regs
, ssize_t stop
)
4263 re_char
*string1
= (re_char
*) str1
;
4264 re_char
*string2
= (re_char
*) str2
;
4265 register char *fastmap
= bufp
->fastmap
;
4266 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4267 size_t total_size
= size1
+ size2
;
4268 ssize_t endpos
= startpos
+ range
;
4269 boolean anchored_start
;
4270 /* Nonzero if we are searching multibyte string. */
4271 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4273 /* Check for out-of-range STARTPOS. */
4274 if (startpos
< 0 || startpos
> total_size
)
4277 /* Fix up RANGE if it might eventually take us outside
4278 the virtual concatenation of STRING1 and STRING2.
4279 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4281 range
= 0 - startpos
;
4282 else if (endpos
> total_size
)
4283 range
= total_size
- startpos
;
4285 /* If the search isn't to be a backwards one, don't waste time in a
4286 search for a pattern anchored at beginning of buffer. */
4287 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4296 /* In a forward search for something that starts with \=.
4297 don't keep searching past point. */
4298 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4300 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4306 /* Update the fastmap now if not correct already. */
4307 if (fastmap
&& !bufp
->fastmap_accurate
)
4308 re_compile_fastmap (bufp
);
4310 /* See whether the pattern is anchored. */
4311 anchored_start
= (bufp
->buffer
[0] == begline
);
4314 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4316 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4318 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4322 /* Loop through the string, looking for a place to start matching. */
4325 /* If the pattern is anchored,
4326 skip quickly past places we cannot match.
4327 We don't bother to treat startpos == 0 specially
4328 because that case doesn't repeat. */
4329 if (anchored_start
&& startpos
> 0)
4331 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4332 : string2
[startpos
- size1
- 1])
4337 /* If a fastmap is supplied, skip quickly over characters that
4338 cannot be the start of a match. If the pattern can match the
4339 null string, however, we don't need to skip characters; we want
4340 the first null string. */
4341 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4343 register re_char
*d
;
4344 register re_wchar_t buf_ch
;
4346 d
= POS_ADDR_VSTRING (startpos
);
4348 if (range
> 0) /* Searching forwards. */
4350 ssize_t irange
= range
, lim
= 0;
4352 if (startpos
< size1
&& startpos
+ range
>= size1
)
4353 lim
= range
- (size1
- startpos
);
4355 /* Written out as an if-else to avoid testing `translate'
4357 if (RE_TRANSLATE_P (translate
))
4364 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4365 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4366 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4369 range
-= buf_charlen
;
4375 register re_wchar_t ch
, translated
;
4378 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4379 translated
= RE_TRANSLATE (translate
, ch
);
4380 if (translated
!= ch
4381 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4383 if (fastmap
[buf_ch
])
4396 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4397 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4399 range
-= buf_charlen
;
4403 while (range
> lim
&& !fastmap
[*d
])
4409 startpos
+= irange
- range
;
4411 else /* Searching backwards. */
4415 buf_ch
= STRING_CHAR (d
);
4416 buf_ch
= TRANSLATE (buf_ch
);
4417 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4422 register re_wchar_t ch
, translated
;
4425 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4426 translated
= TRANSLATE (ch
);
4427 if (translated
!= ch
4428 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4430 if (! fastmap
[TRANSLATE (buf_ch
)])
4436 /* If can't match the null string, and that's all we have left, fail. */
4437 if (range
>= 0 && startpos
== total_size
&& fastmap
4438 && !bufp
->can_be_null
)
4441 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4442 startpos
, regs
, stop
);
4455 /* Update STARTPOS to the next character boundary. */
4458 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4459 int len
= BYTES_BY_CHAR_HEAD (*p
);
4477 /* Update STARTPOS to the previous character boundary. */
4480 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4482 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4484 /* Find the head of multibyte form. */
4485 PREV_CHAR_BOUNDARY (p
, phead
);
4486 range
+= p0
- 1 - p
;
4490 startpos
-= p0
- 1 - p
;
4496 WEAK_ALIAS (__re_search_2
, re_search_2
)
4498 /* Declarations and macros for re_match_2. */
4500 static int bcmp_translate (re_char
*s1
, re_char
*s2
,
4501 register ssize_t len
,
4502 RE_TRANSLATE_TYPE translate
,
4503 const int multibyte
);
4505 /* This converts PTR, a pointer into one of the search strings `string1'
4506 and `string2' into an offset from the beginning of that string. */
4507 #define POINTER_TO_OFFSET(ptr) \
4508 (FIRST_STRING_P (ptr) \
4510 : (ptr) - string2 + (ptrdiff_t) size1)
4512 /* Call before fetching a character with *d. This switches over to
4513 string2 if necessary.
4514 Check re_match_2_internal for a discussion of why end_match_2 might
4515 not be within string2 (but be equal to end_match_1 instead). */
4516 #define PREFETCH() \
4519 /* End of string2 => fail. */ \
4520 if (dend == end_match_2) \
4522 /* End of string1 => advance to string2. */ \
4524 dend = end_match_2; \
4527 /* Call before fetching a char with *d if you already checked other limits.
4528 This is meant for use in lookahead operations like wordend, etc..
4529 where we might need to look at parts of the string that might be
4530 outside of the LIMITs (i.e past `stop'). */
4531 #define PREFETCH_NOLIMIT() \
4535 dend = end_match_2; \
4538 /* Test if at very beginning or at very end of the virtual concatenation
4539 of `string1' and `string2'. If only one string, it's `string2'. */
4540 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4541 #define AT_STRINGS_END(d) ((d) == end2)
4543 /* Disabled due to a compiler bug -- see comment at case wordbound */
4545 /* The comment at case wordbound is following one, but we don't use
4546 AT_WORD_BOUNDARY anymore to support multibyte form.
4548 The DEC Alpha C compiler 3.x generates incorrect code for the
4549 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4550 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4551 macro and introducing temporary variables works around the bug. */
4554 /* Test if D points to a character which is word-constituent. We have
4555 two special cases to check for: if past the end of string1, look at
4556 the first character in string2; and if before the beginning of
4557 string2, look at the last character in string1. */
4558 #define WORDCHAR_P(d) \
4559 (SYNTAX ((d) == end1 ? *string2 \
4560 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4563 /* Test if the character before D and the one at D differ with respect
4564 to being word-constituent. */
4565 #define AT_WORD_BOUNDARY(d) \
4566 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4567 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4570 /* Free everything we malloc. */
4571 #ifdef MATCH_MAY_ALLOCATE
4572 # define FREE_VAR(var) \
4580 # define FREE_VARIABLES() \
4582 REGEX_FREE_STACK (fail_stack.stack); \
4583 FREE_VAR (regstart); \
4584 FREE_VAR (regend); \
4585 FREE_VAR (best_regstart); \
4586 FREE_VAR (best_regend); \
4587 REGEX_SAFE_FREE (); \
4590 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4591 #endif /* not MATCH_MAY_ALLOCATE */
4594 /* Optimization routines. */
4596 /* If the operation is a match against one or more chars,
4597 return a pointer to the next operation, else return NULL. */
4599 skip_one_char (const_re_char
*p
)
4612 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4615 p
= CHARSET_RANGE_TABLE (p
- 1);
4616 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4617 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4620 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4627 case notcategoryspec
:
4639 /* Jump over non-matching operations. */
4641 skip_noops (const_re_char
*p
, const_re_char
*pend
)
4655 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4666 /* Non-zero if "p1 matches something" implies "p2 fails". */
4668 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, const_re_char
*p1
,
4672 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4673 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4675 assert (p1
>= bufp
->buffer
&& p1
< pend
4676 && p2
>= bufp
->buffer
&& p2
<= pend
);
4678 /* Skip over open/close-group commands.
4679 If what follows this loop is a ...+ construct,
4680 look at what begins its body, since we will have to
4681 match at least one of that. */
4682 p2
= skip_noops (p2
, pend
);
4683 /* The same skip can be done for p1, except that this function
4684 is only used in the case where p1 is a simple match operator. */
4685 /* p1 = skip_noops (p1, pend); */
4687 assert (p1
>= bufp
->buffer
&& p1
< pend
4688 && p2
>= bufp
->buffer
&& p2
<= pend
);
4690 op2
= p2
== pend
? succeed
: *p2
;
4696 /* If we're at the end of the pattern, we can change. */
4697 if (skip_one_char (p1
))
4699 DEBUG_PRINT (" End of pattern: fast loop.\n");
4707 register re_wchar_t c
4708 = (re_opcode_t
) *p2
== endline
? '\n'
4709 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4711 if ((re_opcode_t
) *p1
== exactn
)
4713 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4715 DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4720 else if ((re_opcode_t
) *p1
== charset
4721 || (re_opcode_t
) *p1
== charset_not
)
4723 int not = (re_opcode_t
) *p1
== charset_not
;
4725 /* Test if C is listed in charset (or charset_not)
4727 if (! multibyte
|| IS_REAL_ASCII (c
))
4729 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4730 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4733 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4734 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4736 /* `not' is equal to 1 if c would match, which means
4737 that we can't change to pop_failure_jump. */
4740 DEBUG_PRINT (" No match => fast loop.\n");
4744 else if ((re_opcode_t
) *p1
== anychar
4747 DEBUG_PRINT (" . != \\n => fast loop.\n");
4755 if ((re_opcode_t
) *p1
== exactn
)
4756 /* Reuse the code above. */
4757 return mutually_exclusive_p (bufp
, p2
, p1
);
4759 /* It is hard to list up all the character in charset
4760 P2 if it includes multibyte character. Give up in
4762 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4764 /* Now, we are sure that P2 has no range table.
4765 So, for the size of bitmap in P2, `p2[1]' is
4766 enough. But P1 may have range table, so the
4767 size of bitmap table of P1 is extracted by
4768 using macro `CHARSET_BITMAP_SIZE'.
4770 In a multibyte case, we know that all the character
4771 listed in P2 is ASCII. In a unibyte case, P1 has only a
4772 bitmap table. So, in both cases, it is enough to test
4773 only the bitmap table of P1. */
4775 if ((re_opcode_t
) *p1
== charset
)
4778 /* We win if the charset inside the loop
4779 has no overlap with the one after the loop. */
4782 && idx
< CHARSET_BITMAP_SIZE (p1
));
4784 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4788 || idx
== CHARSET_BITMAP_SIZE (p1
))
4790 DEBUG_PRINT (" No match => fast loop.\n");
4794 else if ((re_opcode_t
) *p1
== charset_not
)
4797 /* We win if the charset_not inside the loop lists
4798 every character listed in the charset after. */
4799 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4800 if (! (p2
[2 + idx
] == 0
4801 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4802 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4807 DEBUG_PRINT (" No match => fast loop.\n");
4820 /* Reuse the code above. */
4821 return mutually_exclusive_p (bufp
, p2
, p1
);
4823 /* When we have two charset_not, it's very unlikely that
4824 they don't overlap. The union of the two sets of excluded
4825 chars should cover all possible chars, which, as a matter of
4826 fact, is virtually impossible in multibyte buffers. */
4832 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4834 return ((re_opcode_t
) *p1
== syntaxspec
4835 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4837 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4840 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4842 return ((re_opcode_t
) *p1
== notsyntaxspec
4843 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4845 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4848 return (((re_opcode_t
) *p1
== notsyntaxspec
4849 || (re_opcode_t
) *p1
== syntaxspec
)
4854 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4855 case notcategoryspec
:
4856 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4868 /* Matching routines. */
4870 #ifndef emacs /* Emacs never uses this. */
4871 /* re_match is like re_match_2 except it takes only a single string. */
4874 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
4875 size_t size
, ssize_t pos
, struct re_registers
*regs
)
4877 regoff_t result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
,
4878 size
, pos
, regs
, size
);
4881 WEAK_ALIAS (__re_match
, re_match
)
4882 #endif /* not emacs */
4885 /* In Emacs, this is the string or buffer in which we
4886 are matching. It is used for looking up syntax properties. */
4887 Lisp_Object re_match_object
;
4890 /* re_match_2 matches the compiled pattern in BUFP against the
4891 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4892 and SIZE2, respectively). We start matching at POS, and stop
4895 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4896 store offsets for the substring each group matched in REGS. See the
4897 documentation for exactly how many groups we fill.
4899 We return -1 if no match, -2 if an internal error (such as the
4900 failure stack overflowing). Otherwise, we return the length of the
4901 matched substring. */
4904 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
,
4905 size_t size1
, const char *string2
, size_t size2
, ssize_t pos
,
4906 struct re_registers
*regs
, ssize_t stop
)
4912 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4913 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4914 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4917 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4918 (re_char
*) string2
, size2
,
4922 WEAK_ALIAS (__re_match_2
, re_match_2
)
4925 /* This is a separate function so that we can force an alloca cleanup
4928 re_match_2_internal (struct re_pattern_buffer
*bufp
, const_re_char
*string1
,
4929 size_t size1
, const_re_char
*string2
, size_t size2
,
4930 ssize_t pos
, struct re_registers
*regs
, ssize_t stop
)
4932 /* General temporaries. */
4936 /* Just past the end of the corresponding string. */
4937 re_char
*end1
, *end2
;
4939 /* Pointers into string1 and string2, just past the last characters in
4940 each to consider matching. */
4941 re_char
*end_match_1
, *end_match_2
;
4943 /* Where we are in the data, and the end of the current string. */
4946 /* Used sometimes to remember where we were before starting matching
4947 an operator so that we can go back in case of failure. This "atomic"
4948 behavior of matching opcodes is indispensable to the correctness
4949 of the on_failure_keep_string_jump optimization. */
4952 /* Where we are in the pattern, and the end of the pattern. */
4953 re_char
*p
= bufp
->buffer
;
4954 re_char
*pend
= p
+ bufp
->used
;
4956 /* We use this to map every character in the string. */
4957 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4959 /* Nonzero if BUFP is setup from a multibyte regex. */
4960 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4962 /* Nonzero if STRING1/STRING2 are multibyte. */
4963 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4965 /* Failure point stack. Each place that can handle a failure further
4966 down the line pushes a failure point on this stack. It consists of
4967 regstart, and regend for all registers corresponding to
4968 the subexpressions we're currently inside, plus the number of such
4969 registers, and, finally, two char *'s. The first char * is where
4970 to resume scanning the pattern; the second one is where to resume
4971 scanning the strings. */
4972 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4973 fail_stack_type fail_stack
;
4975 #ifdef DEBUG_COMPILES_ARGUMENTS
4976 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4979 #if defined REL_ALLOC && defined REGEX_MALLOC
4980 /* This holds the pointer to the failure stack, when
4981 it is allocated relocatably. */
4982 fail_stack_elt_t
*failure_stack_ptr
;
4985 /* We fill all the registers internally, independent of what we
4986 return, for use in backreferences. The number here includes
4987 an element for register zero. */
4988 size_t num_regs
= bufp
->re_nsub
+ 1;
4990 /* Information on the contents of registers. These are pointers into
4991 the input strings; they record just what was matched (on this
4992 attempt) by a subexpression part of the pattern, that is, the
4993 regnum-th regstart pointer points to where in the pattern we began
4994 matching and the regnum-th regend points to right after where we
4995 stopped matching the regnum-th subexpression. (The zeroth register
4996 keeps track of what the whole pattern matches.) */
4997 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4998 re_char
**regstart
, **regend
;
5001 /* The following record the register info as found in the above
5002 variables when we find a match better than any we've seen before.
5003 This happens as we backtrack through the failure points, which in
5004 turn happens only if we have not yet matched the entire string. */
5005 unsigned best_regs_set
= false;
5006 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5007 re_char
**best_regstart
, **best_regend
;
5010 /* Logically, this is `best_regend[0]'. But we don't want to have to
5011 allocate space for that if we're not allocating space for anything
5012 else (see below). Also, we never need info about register 0 for
5013 any of the other register vectors, and it seems rather a kludge to
5014 treat `best_regend' differently than the rest. So we keep track of
5015 the end of the best match so far in a separate variable. We
5016 initialize this to NULL so that when we backtrack the first time
5017 and need to test it, it's not garbage. */
5018 re_char
*match_end
= NULL
;
5020 #ifdef DEBUG_COMPILES_ARGUMENTS
5021 /* Counts the total number of registers pushed. */
5022 unsigned num_regs_pushed
= 0;
5025 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5027 REGEX_USE_SAFE_ALLOCA
;
5031 #ifdef MATCH_MAY_ALLOCATE
5032 /* Do not bother to initialize all the register variables if there are
5033 no groups in the pattern, as it takes a fair amount of time. If
5034 there are groups, we include space for register 0 (the whole
5035 pattern), even though we never use it, since it simplifies the
5036 array indexing. We should fix this. */
5039 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5040 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5041 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5042 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5044 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5052 /* We must initialize all our variables to NULL, so that
5053 `FREE_VARIABLES' doesn't try to free them. */
5054 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5056 #endif /* MATCH_MAY_ALLOCATE */
5058 /* The starting position is bogus. */
5059 if (pos
< 0 || pos
> size1
+ size2
)
5065 /* Initialize subexpression text positions to -1 to mark ones that no
5066 start_memory/stop_memory has been seen for. Also initialize the
5067 register information struct. */
5068 for (reg
= 1; reg
< num_regs
; reg
++)
5069 regstart
[reg
] = regend
[reg
] = NULL
;
5071 /* We move `string1' into `string2' if the latter's empty -- but not if
5072 `string1' is null. */
5073 if (size2
== 0 && string1
!= NULL
)
5080 end1
= string1
+ size1
;
5081 end2
= string2
+ size2
;
5083 /* `p' scans through the pattern as `d' scans through the data.
5084 `dend' is the end of the input string that `d' points within. `d'
5085 is advanced into the following input string whenever necessary, but
5086 this happens before fetching; therefore, at the beginning of the
5087 loop, `d' can be pointing at the end of a string, but it cannot
5091 /* Only match within string2. */
5092 d
= string2
+ pos
- size1
;
5093 dend
= end_match_2
= string2
+ stop
- size1
;
5094 end_match_1
= end1
; /* Just to give it a value. */
5100 /* Only match within string1. */
5101 end_match_1
= string1
+ stop
;
5103 When we reach end_match_1, PREFETCH normally switches to string2.
5104 But in the present case, this means that just doing a PREFETCH
5105 makes us jump from `stop' to `gap' within the string.
5106 What we really want here is for the search to stop as
5107 soon as we hit end_match_1. That's why we set end_match_2
5108 to end_match_1 (since PREFETCH fails as soon as we hit
5110 end_match_2
= end_match_1
;
5113 { /* It's important to use this code when stop == size so that
5114 moving `d' from end1 to string2 will not prevent the d == dend
5115 check from catching the end of string. */
5117 end_match_2
= string2
+ stop
- size1
;
5123 DEBUG_PRINT ("The compiled pattern is: ");
5124 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5125 DEBUG_PRINT ("The string to match is: `");
5126 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5127 DEBUG_PRINT ("'\n");
5129 /* This loops over pattern commands. It exits by returning from the
5130 function if the match is complete, or it drops through if the match
5131 fails at this starting point in the input data. */
5134 DEBUG_PRINT ("\n%p: ", p
);
5140 /* End of pattern means we might have succeeded. */
5141 DEBUG_PRINT ("end of pattern ... ");
5143 /* If we haven't matched the entire string, and we want the
5144 longest match, try backtracking. */
5145 if (d
!= end_match_2
)
5147 /* 1 if this match ends in the same string (string1 or string2)
5148 as the best previous match. */
5149 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5150 == FIRST_STRING_P (d
));
5151 /* 1 if this match is the best seen so far. */
5152 boolean best_match_p
;
5154 /* AIX compiler got confused when this was combined
5155 with the previous declaration. */
5157 best_match_p
= d
> match_end
;
5159 best_match_p
= !FIRST_STRING_P (d
);
5161 DEBUG_PRINT ("backtracking.\n");
5163 if (!FAIL_STACK_EMPTY ())
5164 { /* More failure points to try. */
5166 /* If exceeds best match so far, save it. */
5167 if (!best_regs_set
|| best_match_p
)
5169 best_regs_set
= true;
5172 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5174 for (reg
= 1; reg
< num_regs
; reg
++)
5176 best_regstart
[reg
] = regstart
[reg
];
5177 best_regend
[reg
] = regend
[reg
];
5183 /* If no failure points, don't restore garbage. And if
5184 last match is real best match, don't restore second
5186 else if (best_regs_set
&& !best_match_p
)
5189 /* Restore best match. It may happen that `dend ==
5190 end_match_1' while the restored d is in string2.
5191 For example, the pattern `x.*y.*z' against the
5192 strings `x-' and `y-z-', if the two strings are
5193 not consecutive in memory. */
5194 DEBUG_PRINT ("Restoring best registers.\n");
5197 dend
= ((d
>= string1
&& d
<= end1
)
5198 ? end_match_1
: end_match_2
);
5200 for (reg
= 1; reg
< num_regs
; reg
++)
5202 regstart
[reg
] = best_regstart
[reg
];
5203 regend
[reg
] = best_regend
[reg
];
5206 } /* d != end_match_2 */
5209 DEBUG_PRINT ("Accepting match.\n");
5211 /* If caller wants register contents data back, do it. */
5212 if (regs
&& !bufp
->no_sub
)
5214 /* Have the register data arrays been allocated? */
5215 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5216 { /* No. So allocate them with malloc. We need one
5217 extra element beyond `num_regs' for the `-1' marker
5219 regs
->num_regs
= max (RE_NREGS
, num_regs
+ 1);
5220 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5221 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5222 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5227 bufp
->regs_allocated
= REGS_REALLOCATE
;
5229 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5230 { /* Yes. If we need more elements than were already
5231 allocated, reallocate them. If we need fewer, just
5233 if (regs
->num_regs
< num_regs
+ 1)
5235 regs
->num_regs
= num_regs
+ 1;
5236 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5237 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5238 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5247 /* These braces fend off a "empty body in an else-statement"
5248 warning under GCC when assert expands to nothing. */
5249 assert (bufp
->regs_allocated
== REGS_FIXED
);
5252 /* Convert the pointer data in `regstart' and `regend' to
5253 indices. Register zero has to be set differently,
5254 since we haven't kept track of any info for it. */
5255 if (regs
->num_regs
> 0)
5257 regs
->start
[0] = pos
;
5258 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5261 /* Go through the first `min (num_regs, regs->num_regs)'
5262 registers, since that is all we initialized. */
5263 for (reg
= 1; reg
< min (num_regs
, regs
->num_regs
); reg
++)
5265 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5266 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5269 regs
->start
[reg
] = POINTER_TO_OFFSET (regstart
[reg
]);
5270 regs
->end
[reg
] = POINTER_TO_OFFSET (regend
[reg
]);
5274 /* If the regs structure we return has more elements than
5275 were in the pattern, set the extra elements to -1. If
5276 we (re)allocated the registers, this is the case,
5277 because we always allocate enough to have at least one
5279 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5280 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5281 } /* regs && !bufp->no_sub */
5283 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5284 nfailure_points_pushed
, nfailure_points_popped
,
5285 nfailure_points_pushed
- nfailure_points_popped
);
5286 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed
);
5288 dcnt
= POINTER_TO_OFFSET (d
) - pos
;
5290 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt
);
5296 /* Otherwise match next pattern command. */
5299 /* Ignore these. Used to ignore the n of succeed_n's which
5300 currently have n == 0. */
5302 DEBUG_PRINT ("EXECUTING no_op.\n");
5306 DEBUG_PRINT ("EXECUTING succeed.\n");
5309 /* Match the next n pattern characters exactly. The following
5310 byte in the pattern defines n, and the n bytes after that
5311 are the characters to match. */
5314 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt
);
5316 /* Remember the start point to rollback upon failure. */
5320 /* This is written out as an if-else so we don't waste time
5321 testing `translate' inside the loop. */
5322 if (RE_TRANSLATE_P (translate
))
5326 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5346 /* The cost of testing `translate' is comparatively small. */
5347 if (target_multibyte
)
5350 int pat_charlen
, buf_charlen
;
5355 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5358 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5361 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5363 if (TRANSLATE (buf_ch
) != pat_ch
)
5371 mcnt
-= pat_charlen
;
5383 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5384 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5391 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5392 if (! CHAR_BYTE8_P (buf_ch
))
5394 buf_ch
= TRANSLATE (buf_ch
);
5395 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5401 if (buf_ch
!= pat_ch
)
5414 /* Match any character except possibly a newline or a null. */
5420 DEBUG_PRINT ("EXECUTING anychar.\n");
5423 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5425 buf_ch
= TRANSLATE (buf_ch
);
5427 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5429 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5430 && buf_ch
== '\000'))
5433 DEBUG_PRINT (" Matched `%d'.\n", *d
);
5442 register unsigned int c
;
5443 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5446 /* Start of actual range_table, or end of bitmap if there is no
5448 re_char
*range_table
IF_LINT (= NULL
);
5450 /* Nonzero if there is a range table. */
5451 int range_table_exists
;
5453 /* Number of ranges of range table. This is not included
5454 in the initial byte-length of the command. */
5457 /* Whether matching against a unibyte character. */
5458 boolean unibyte_char
= false;
5460 DEBUG_PRINT ("EXECUTING charset%s.\n", not ? "_not" : "");
5462 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5464 if (range_table_exists
)
5466 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5467 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5471 c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5472 if (target_multibyte
)
5477 c1
= RE_CHAR_TO_UNIBYTE (c
);
5480 unibyte_char
= true;
5486 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5488 if (! CHAR_BYTE8_P (c1
))
5490 c1
= TRANSLATE (c1
);
5491 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5494 unibyte_char
= true;
5499 unibyte_char
= true;
5502 if (unibyte_char
&& c
< (1 << BYTEWIDTH
))
5503 { /* Lookup bitmap. */
5504 /* Cast to `unsigned' instead of `unsigned char' in
5505 case the bit list is a full 32 bytes long. */
5506 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5507 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5511 else if (range_table_exists
)
5513 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5515 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5516 | (class_bits
& BIT_MULTIBYTE
)
5517 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5518 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5519 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5520 | (class_bits
& BIT_WORD
&& ISWORD (c
))
5521 | (class_bits
& BIT_ALPHA
&& ISALPHA (c
))
5522 | (class_bits
& BIT_ALNUM
&& ISALNUM (c
)))
5525 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5529 if (range_table_exists
)
5530 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5532 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5534 if (!not) goto fail
;
5541 /* The beginning of a group is represented by start_memory.
5542 The argument is the register number. The text
5543 matched within the group is recorded (in the internal
5544 registers data structure) under the register number. */
5546 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p
);
5548 /* In case we need to undo this operation (via backtracking). */
5549 PUSH_FAILURE_REG (*p
);
5552 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5553 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart
[*p
]));
5555 /* Move past the register number and inner group count. */
5560 /* The stop_memory opcode represents the end of a group. Its
5561 argument is the same as start_memory's: the register number. */
5563 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p
);
5565 assert (!REG_UNSET (regstart
[*p
]));
5566 /* Strictly speaking, there should be code such as:
5568 assert (REG_UNSET (regend[*p]));
5569 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5571 But the only info to be pushed is regend[*p] and it is known to
5572 be UNSET, so there really isn't anything to push.
5573 Not pushing anything, on the other hand deprives us from the
5574 guarantee that regend[*p] is UNSET since undoing this operation
5575 will not reset its value properly. This is not important since
5576 the value will only be read on the next start_memory or at
5577 the very end and both events can only happen if this stop_memory
5581 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend
[*p
]));
5583 /* Move past the register number and the inner group count. */
5588 /* \<digit> has been turned into a `duplicate' command which is
5589 followed by the numeric value of <digit> as the register number. */
5592 register re_char
*d2
, *dend2
;
5593 int regno
= *p
++; /* Get which register to match against. */
5594 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno
);
5596 /* Can't back reference a group which we've never matched. */
5597 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5600 /* Where in input to try to start matching. */
5601 d2
= regstart
[regno
];
5603 /* Remember the start point to rollback upon failure. */
5606 /* Where to stop matching; if both the place to start and
5607 the place to stop matching are in the same string, then
5608 set to the place to stop, otherwise, for now have to use
5609 the end of the first string. */
5611 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5612 == FIRST_STRING_P (regend
[regno
]))
5613 ? regend
[regno
] : end_match_1
);
5618 /* If necessary, advance to next segment in register
5622 if (dend2
== end_match_2
) break;
5623 if (dend2
== regend
[regno
]) break;
5625 /* End of string1 => advance to string2. */
5627 dend2
= regend
[regno
];
5629 /* At end of register contents => success */
5630 if (d2
== dend2
) break;
5632 /* If necessary, advance to next segment in data. */
5635 /* How many characters left in this segment to match. */
5638 /* Want how many consecutive characters we can match in
5639 one shot, so, if necessary, adjust the count. */
5640 if (dcnt
> dend2
- d2
)
5643 /* Compare that many; failure if mismatch, else move
5645 if (RE_TRANSLATE_P (translate
)
5646 ? bcmp_translate (d
, d2
, dcnt
, translate
, target_multibyte
)
5647 : memcmp (d
, d2
, dcnt
))
5652 d
+= dcnt
, d2
+= dcnt
;
5658 /* begline matches the empty string at the beginning of the string
5659 (unless `not_bol' is set in `bufp'), and after newlines. */
5661 DEBUG_PRINT ("EXECUTING begline.\n");
5663 if (AT_STRINGS_BEG (d
))
5665 if (!bufp
->not_bol
) break;
5670 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5674 /* In all other cases, we fail. */
5678 /* endline is the dual of begline. */
5680 DEBUG_PRINT ("EXECUTING endline.\n");
5682 if (AT_STRINGS_END (d
))
5684 if (!bufp
->not_eol
) break;
5688 PREFETCH_NOLIMIT ();
5695 /* Match at the very beginning of the data. */
5697 DEBUG_PRINT ("EXECUTING begbuf.\n");
5698 if (AT_STRINGS_BEG (d
))
5703 /* Match at the very end of the data. */
5705 DEBUG_PRINT ("EXECUTING endbuf.\n");
5706 if (AT_STRINGS_END (d
))
5711 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5712 pushes NULL as the value for the string on the stack. Then
5713 `POP_FAILURE_POINT' will keep the current value for the
5714 string, instead of restoring it. To see why, consider
5715 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5716 then the . fails against the \n. But the next thing we want
5717 to do is match the \n against the \n; if we restored the
5718 string value, we would be back at the foo.
5720 Because this is used only in specific cases, we don't need to
5721 check all the things that `on_failure_jump' does, to make
5722 sure the right things get saved on the stack. Hence we don't
5723 share its code. The only reason to push anything on the
5724 stack at all is that otherwise we would have to change
5725 `anychar's code to do something besides goto fail in this
5726 case; that seems worse than this. */
5727 case on_failure_keep_string_jump
:
5728 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5729 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5732 PUSH_FAILURE_POINT (p
- 3, NULL
);
5735 /* A nasty loop is introduced by the non-greedy *? and +?.
5736 With such loops, the stack only ever contains one failure point
5737 at a time, so that a plain on_failure_jump_loop kind of
5738 cycle detection cannot work. Worse yet, such a detection
5739 can not only fail to detect a cycle, but it can also wrongly
5740 detect a cycle (between different instantiations of the same
5742 So the method used for those nasty loops is a little different:
5743 We use a special cycle-detection-stack-frame which is pushed
5744 when the on_failure_jump_nastyloop failure-point is *popped*.
5745 This special frame thus marks the beginning of one iteration
5746 through the loop and we can hence easily check right here
5747 whether something matched between the beginning and the end of
5749 case on_failure_jump_nastyloop
:
5750 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5751 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5754 assert ((re_opcode_t
)p
[-4] == no_op
);
5757 CHECK_INFINITE_LOOP (p
- 4, d
);
5759 /* If there's a cycle, just continue without pushing
5760 this failure point. The failure point is the "try again"
5761 option, which shouldn't be tried.
5762 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5763 PUSH_FAILURE_POINT (p
- 3, d
);
5767 /* Simple loop detecting on_failure_jump: just check on the
5768 failure stack if the same spot was already hit earlier. */
5769 case on_failure_jump_loop
:
5771 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5772 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5776 CHECK_INFINITE_LOOP (p
- 3, d
);
5778 /* If there's a cycle, get out of the loop, as if the matching
5779 had failed. We used to just `goto fail' here, but that was
5780 aborting the search a bit too early: we want to keep the
5781 empty-loop-match and keep matching after the loop.
5782 We want (x?)*y\1z to match both xxyz and xxyxz. */
5785 PUSH_FAILURE_POINT (p
- 3, d
);
5790 /* Uses of on_failure_jump:
5792 Each alternative starts with an on_failure_jump that points
5793 to the beginning of the next alternative. Each alternative
5794 except the last ends with a jump that in effect jumps past
5795 the rest of the alternatives. (They really jump to the
5796 ending jump of the following alternative, because tensioning
5797 these jumps is a hassle.)
5799 Repeats start with an on_failure_jump that points past both
5800 the repetition text and either the following jump or
5801 pop_failure_jump back to this on_failure_jump. */
5802 case on_failure_jump
:
5803 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5804 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5807 PUSH_FAILURE_POINT (p
-3, d
);
5810 /* This operation is used for greedy *.
5811 Compare the beginning of the repeat with what in the
5812 pattern follows its end. If we can establish that there
5813 is nothing that they would both match, i.e., that we
5814 would have to backtrack because of (as in, e.g., `a*a')
5815 then we can use a non-backtracking loop based on
5816 on_failure_keep_string_jump instead of on_failure_jump. */
5817 case on_failure_jump_smart
:
5818 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5819 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5822 re_char
*p1
= p
; /* Next operation. */
5823 /* Here, we discard `const', making re_match non-reentrant. */
5824 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5825 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5827 p
-= 3; /* Reset so that we will re-execute the
5828 instruction once it's been changed. */
5830 EXTRACT_NUMBER (mcnt
, p2
- 2);
5832 /* Ensure this is a indeed the trivial kind of loop
5833 we are expecting. */
5834 assert (skip_one_char (p1
) == p2
- 3);
5835 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5836 DEBUG_STATEMENT (debug
+= 2);
5837 if (mutually_exclusive_p (bufp
, p1
, p2
))
5839 /* Use a fast `on_failure_keep_string_jump' loop. */
5840 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5841 *p3
= (unsigned char) on_failure_keep_string_jump
;
5842 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5846 /* Default to a safe `on_failure_jump' loop. */
5847 DEBUG_PRINT (" smart default => slow loop.\n");
5848 *p3
= (unsigned char) on_failure_jump
;
5850 DEBUG_STATEMENT (debug
-= 2);
5854 /* Unconditionally jump (without popping any failure points). */
5857 IMMEDIATE_QUIT_CHECK
;
5858 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5859 DEBUG_PRINT ("EXECUTING jump %d ", mcnt
);
5860 p
+= mcnt
; /* Do the jump. */
5861 DEBUG_PRINT ("(to %p).\n", p
);
5865 /* Have to succeed matching what follows at least n times.
5866 After that, handle like `on_failure_jump'. */
5868 /* Signedness doesn't matter since we only compare MCNT to 0. */
5869 EXTRACT_NUMBER (mcnt
, p
+ 2);
5870 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt
);
5872 /* Originally, mcnt is how many times we HAVE to succeed. */
5875 /* Here, we discard `const', making re_match non-reentrant. */
5876 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5879 PUSH_NUMBER (p2
, mcnt
);
5882 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5887 /* Signedness doesn't matter since we only compare MCNT to 0. */
5888 EXTRACT_NUMBER (mcnt
, p
+ 2);
5889 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt
);
5891 /* Originally, this is how many times we CAN jump. */
5894 /* Here, we discard `const', making re_match non-reentrant. */
5895 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5897 PUSH_NUMBER (p2
, mcnt
);
5898 goto unconditional_jump
;
5900 /* If don't have to jump any more, skip over the rest of command. */
5907 unsigned char *p2
; /* Location of the counter. */
5908 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5910 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5911 /* Here, we discard `const', making re_match non-reentrant. */
5912 p2
= (unsigned char*) p
+ mcnt
;
5913 /* Signedness doesn't matter since we only copy MCNT's bits. */
5914 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5915 DEBUG_PRINT (" Setting %p to %d.\n", p2
, mcnt
);
5916 PUSH_NUMBER (p2
, mcnt
);
5923 boolean
not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5924 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5926 /* We SUCCEED (or FAIL) in one of the following cases: */
5928 /* Case 1: D is at the beginning or the end of string. */
5929 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5933 /* C1 is the character before D, S1 is the syntax of C1, C2
5934 is the character at D, and S2 is the syntax of C2. */
5939 ssize_t offset
= PTR_TO_OFFSET (d
- 1);
5940 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5941 UPDATE_SYNTAX_TABLE (charpos
);
5943 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5946 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5948 PREFETCH_NOLIMIT ();
5949 GET_CHAR_AFTER (c2
, d
, dummy
);
5952 if (/* Case 2: Only one of S1 and S2 is Sword. */
5953 ((s1
== Sword
) != (s2
== Sword
))
5954 /* Case 3: Both of S1 and S2 are Sword, and macro
5955 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5956 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5966 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5968 /* We FAIL in one of the following cases: */
5970 /* Case 1: D is at the end of string. */
5971 if (AT_STRINGS_END (d
))
5975 /* C1 is the character before D, S1 is the syntax of C1, C2
5976 is the character at D, and S2 is the syntax of C2. */
5981 ssize_t offset
= PTR_TO_OFFSET (d
);
5982 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5983 UPDATE_SYNTAX_TABLE (charpos
);
5986 GET_CHAR_AFTER (c2
, d
, dummy
);
5989 /* Case 2: S2 is not Sword. */
5993 /* Case 3: D is not at the beginning of string ... */
5994 if (!AT_STRINGS_BEG (d
))
5996 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5998 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6002 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6004 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6011 DEBUG_PRINT ("EXECUTING wordend.\n");
6013 /* We FAIL in one of the following cases: */
6015 /* Case 1: D is at the beginning of string. */
6016 if (AT_STRINGS_BEG (d
))
6020 /* C1 is the character before D, S1 is the syntax of C1, C2
6021 is the character at D, and S2 is the syntax of C2. */
6026 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6027 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6028 UPDATE_SYNTAX_TABLE (charpos
);
6030 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6033 /* Case 2: S1 is not Sword. */
6037 /* Case 3: D is not at the end of string ... */
6038 if (!AT_STRINGS_END (d
))
6040 PREFETCH_NOLIMIT ();
6041 GET_CHAR_AFTER (c2
, d
, dummy
);
6043 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6047 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6049 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6056 DEBUG_PRINT ("EXECUTING symbeg.\n");
6058 /* We FAIL in one of the following cases: */
6060 /* Case 1: D is at the end of string. */
6061 if (AT_STRINGS_END (d
))
6065 /* C1 is the character before D, S1 is the syntax of C1, C2
6066 is the character at D, and S2 is the syntax of C2. */
6070 ssize_t offset
= PTR_TO_OFFSET (d
);
6071 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6072 UPDATE_SYNTAX_TABLE (charpos
);
6075 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6078 /* Case 2: S2 is neither Sword nor Ssymbol. */
6079 if (s2
!= Sword
&& s2
!= Ssymbol
)
6082 /* Case 3: D is not at the beginning of string ... */
6083 if (!AT_STRINGS_BEG (d
))
6085 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6087 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6091 /* ... and S1 is Sword or Ssymbol. */
6092 if (s1
== Sword
|| s1
== Ssymbol
)
6099 DEBUG_PRINT ("EXECUTING symend.\n");
6101 /* We FAIL in one of the following cases: */
6103 /* Case 1: D is at the beginning of string. */
6104 if (AT_STRINGS_BEG (d
))
6108 /* C1 is the character before D, S1 is the syntax of C1, C2
6109 is the character at D, and S2 is the syntax of C2. */
6113 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6114 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6115 UPDATE_SYNTAX_TABLE (charpos
);
6117 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6120 /* Case 2: S1 is neither Ssymbol nor Sword. */
6121 if (s1
!= Sword
&& s1
!= Ssymbol
)
6124 /* Case 3: D is not at the end of string ... */
6125 if (!AT_STRINGS_END (d
))
6127 PREFETCH_NOLIMIT ();
6128 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6130 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6134 /* ... and S2 is Sword or Ssymbol. */
6135 if (s2
== Sword
|| s2
== Ssymbol
)
6144 boolean
not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6146 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6151 ssize_t offset
= PTR_TO_OFFSET (d
);
6152 ssize_t pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6153 UPDATE_SYNTAX_TABLE (pos1
);
6160 GET_CHAR_AFTER (c
, d
, len
);
6161 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6170 DEBUG_PRINT ("EXECUTING before_dot.\n");
6171 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6176 DEBUG_PRINT ("EXECUTING at_dot.\n");
6177 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6182 DEBUG_PRINT ("EXECUTING after_dot.\n");
6183 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6188 case notcategoryspec
:
6190 boolean
not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6192 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6193 not ? "not" : "", mcnt
);
6199 GET_CHAR_AFTER (c
, d
, len
);
6200 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6212 continue; /* Successfully executed one pattern command; keep going. */
6215 /* We goto here if a matching operation fails. */
6217 IMMEDIATE_QUIT_CHECK
;
6218 if (!FAIL_STACK_EMPTY ())
6221 /* A restart point is known. Restore to that state. */
6222 DEBUG_PRINT ("\nFAIL:\n");
6223 POP_FAILURE_POINT (str
, pat
);
6226 case on_failure_keep_string_jump
:
6227 assert (str
== NULL
);
6228 goto continue_failure_jump
;
6230 case on_failure_jump_nastyloop
:
6231 assert ((re_opcode_t
)pat
[-2] == no_op
);
6232 PUSH_FAILURE_POINT (pat
- 2, str
);
6235 case on_failure_jump_loop
:
6236 case on_failure_jump
:
6239 continue_failure_jump
:
6240 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6245 /* A special frame used for nastyloops. */
6252 assert (p
>= bufp
->buffer
&& p
<= pend
);
6254 if (d
>= string1
&& d
<= end1
)
6258 break; /* Matching at this starting point really fails. */
6262 goto restore_best_regs
;
6266 return -1; /* Failure to match. */
6269 /* Subroutine definitions for re_match_2. */
6271 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6272 bytes; nonzero otherwise. */
6275 bcmp_translate (const_re_char
*s1
, const_re_char
*s2
, register ssize_t len
,
6276 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6278 register re_char
*p1
= s1
, *p2
= s2
;
6279 re_char
*p1_end
= s1
+ len
;
6280 re_char
*p2_end
= s2
+ len
;
6282 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6283 different lengths, but relying on a single `len' would break this. -sm */
6284 while (p1
< p1_end
&& p2
< p2_end
)
6286 int p1_charlen
, p2_charlen
;
6287 re_wchar_t p1_ch
, p2_ch
;
6289 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6290 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6292 if (RE_TRANSLATE (translate
, p1_ch
)
6293 != RE_TRANSLATE (translate
, p2_ch
))
6296 p1
+= p1_charlen
, p2
+= p2_charlen
;
6299 if (p1
!= p1_end
|| p2
!= p2_end
)
6305 /* Entry points for GNU code. */
6307 /* re_compile_pattern is the GNU regular expression compiler: it
6308 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6309 Returns 0 if the pattern was valid, otherwise an error string.
6311 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6312 are set in BUFP on entry.
6314 We call regex_compile to do the actual compilation. */
6317 re_compile_pattern (const char *pattern
, size_t length
,
6318 struct re_pattern_buffer
*bufp
)
6322 /* GNU code is written to assume at least RE_NREGS registers will be set
6323 (and at least one extra will be -1). */
6324 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6326 /* And GNU code determines whether or not to get register information
6327 by passing null for the REGS argument to re_match, etc., not by
6331 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6335 return gettext (re_error_msgid
[(int) ret
]);
6337 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6339 /* Entry points compatible with 4.2 BSD regex library. We don't define
6340 them unless specifically requested. */
6342 #if defined _REGEX_RE_COMP || defined _LIBC
6344 /* BSD has one and only one pattern buffer. */
6345 static struct re_pattern_buffer re_comp_buf
;
6349 /* Make these definitions weak in libc, so POSIX programs can redefine
6350 these names if they don't use our functions, and still use
6351 regcomp/regexec below without link errors. */
6354 re_comp (const char *s
)
6360 if (!re_comp_buf
.buffer
)
6361 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6362 return (char *) gettext ("No previous regular expression");
6366 if (!re_comp_buf
.buffer
)
6368 re_comp_buf
.buffer
= malloc (200);
6369 if (re_comp_buf
.buffer
== NULL
)
6370 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6371 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6372 re_comp_buf
.allocated
= 200;
6374 re_comp_buf
.fastmap
= malloc (1 << BYTEWIDTH
);
6375 if (re_comp_buf
.fastmap
== NULL
)
6376 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6377 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6380 /* Since `re_exec' always passes NULL for the `regs' argument, we
6381 don't need to initialize the pattern buffer fields which affect it. */
6383 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6388 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6389 return (char *) gettext (re_error_msgid
[(int) ret
]);
6397 re_exec (const char *s
)
6399 const size_t len
= strlen (s
);
6400 return re_search (&re_comp_buf
, s
, len
, 0, len
, 0) >= 0;
6402 #endif /* _REGEX_RE_COMP */
6404 /* POSIX.2 functions. Don't define these for Emacs. */
6408 /* regcomp takes a regular expression as a string and compiles it.
6410 PREG is a regex_t *. We do not expect any fields to be initialized,
6411 since POSIX says we shouldn't. Thus, we set
6413 `buffer' to the compiled pattern;
6414 `used' to the length of the compiled pattern;
6415 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6416 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6417 RE_SYNTAX_POSIX_BASIC;
6418 `fastmap' to an allocated space for the fastmap;
6419 `fastmap_accurate' to zero;
6420 `re_nsub' to the number of subexpressions in PATTERN.
6422 PATTERN is the address of the pattern string.
6424 CFLAGS is a series of bits which affect compilation.
6426 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6427 use POSIX basic syntax.
6429 If REG_NEWLINE is set, then . and [^...] don't match newline.
6430 Also, regexec will try a match beginning after every newline.
6432 If REG_ICASE is set, then we considers upper- and lowercase
6433 versions of letters to be equivalent when matching.
6435 If REG_NOSUB is set, then when PREG is passed to regexec, that
6436 routine will report only success or failure, and nothing about the
6439 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6440 the return codes and their meanings.) */
6443 regcomp (regex_t
*_Restrict_ preg
, const char *_Restrict_ pattern
,
6448 = (cflags
& REG_EXTENDED
) ?
6449 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6451 /* regex_compile will allocate the space for the compiled pattern. */
6453 preg
->allocated
= 0;
6456 /* Try to allocate space for the fastmap. */
6457 preg
->fastmap
= malloc (1 << BYTEWIDTH
);
6459 if (cflags
& REG_ICASE
)
6463 preg
->translate
= malloc (CHAR_SET_SIZE
* sizeof *preg
->translate
);
6464 if (preg
->translate
== NULL
)
6465 return (int) REG_ESPACE
;
6467 /* Map uppercase characters to corresponding lowercase ones. */
6468 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6469 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6472 preg
->translate
= NULL
;
6474 /* If REG_NEWLINE is set, newlines are treated differently. */
6475 if (cflags
& REG_NEWLINE
)
6476 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6477 syntax
&= ~RE_DOT_NEWLINE
;
6478 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6481 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6483 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6485 /* POSIX says a null character in the pattern terminates it, so we
6486 can use strlen here in compiling the pattern. */
6487 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6489 /* POSIX doesn't distinguish between an unmatched open-group and an
6490 unmatched close-group: both are REG_EPAREN. */
6491 if (ret
== REG_ERPAREN
)
6494 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6495 { /* Compute the fastmap now, since regexec cannot modify the pattern
6497 re_compile_fastmap (preg
);
6498 if (preg
->can_be_null
)
6499 { /* The fastmap can't be used anyway. */
6500 free (preg
->fastmap
);
6501 preg
->fastmap
= NULL
;
6506 WEAK_ALIAS (__regcomp
, regcomp
)
6509 /* regexec searches for a given pattern, specified by PREG, in the
6512 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6513 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6514 least NMATCH elements, and we set them to the offsets of the
6515 corresponding matched substrings.
6517 EFLAGS specifies `execution flags' which affect matching: if
6518 REG_NOTBOL is set, then ^ does not match at the beginning of the
6519 string; if REG_NOTEOL is set, then $ does not match at the end.
6521 We return 0 if we find a match and REG_NOMATCH if not. */
6524 regexec (const regex_t
*_Restrict_ preg
, const char *_Restrict_ string
,
6525 size_t nmatch
, regmatch_t pmatch
[_Restrict_arr_
], int eflags
)
6528 struct re_registers regs
;
6529 regex_t private_preg
;
6530 size_t len
= strlen (string
);
6531 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6533 private_preg
= *preg
;
6535 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6536 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6538 /* The user has told us exactly how many registers to return
6539 information about, via `nmatch'. We have to pass that on to the
6540 matching routines. */
6541 private_preg
.regs_allocated
= REGS_FIXED
;
6545 regs
.num_regs
= nmatch
;
6546 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6547 if (regs
.start
== NULL
)
6549 regs
.end
= regs
.start
+ nmatch
;
6552 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6553 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6554 was a little bit longer but still only matching the real part.
6555 This works because the `endline' will check for a '\n' and will find a
6556 '\0', correctly deciding that this is not the end of a line.
6557 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6558 a convenient '\0' there. For all we know, the string could be preceded
6559 by '\n' which would throw things off. */
6561 /* Perform the searching operation. */
6562 ret
= re_search (&private_preg
, string
, len
,
6563 /* start: */ 0, /* range: */ len
,
6564 want_reg_info
? ®s
: 0);
6566 /* Copy the register information to the POSIX structure. */
6573 for (r
= 0; r
< nmatch
; r
++)
6575 pmatch
[r
].rm_so
= regs
.start
[r
];
6576 pmatch
[r
].rm_eo
= regs
.end
[r
];
6580 /* If we needed the temporary register info, free the space now. */
6584 /* We want zero return to mean success, unlike `re_search'. */
6585 return ret
>= 0 ? REG_NOERROR
: REG_NOMATCH
;
6587 WEAK_ALIAS (__regexec
, regexec
)
6590 /* Returns a message corresponding to an error code, ERR_CODE, returned
6591 from either regcomp or regexec. We don't use PREG here.
6593 ERR_CODE was previously called ERRCODE, but that name causes an
6594 error with msvc8 compiler. */
6597 regerror (int err_code
, const regex_t
*preg
, char *errbuf
, size_t errbuf_size
)
6603 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6604 /* Only error codes returned by the rest of the code should be passed
6605 to this routine. If we are given anything else, or if other regex
6606 code generates an invalid error code, then the program has a bug.
6607 Dump core so we can fix it. */
6610 msg
= gettext (re_error_msgid
[err_code
]);
6612 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6614 if (errbuf_size
!= 0)
6616 if (msg_size
> errbuf_size
)
6618 memcpy (errbuf
, msg
, errbuf_size
- 1);
6619 errbuf
[errbuf_size
- 1] = 0;
6622 strcpy (errbuf
, msg
);
6627 WEAK_ALIAS (__regerror
, regerror
)
6630 /* Free dynamically allocated space used by PREG. */
6633 regfree (regex_t
*preg
)
6635 free (preg
->buffer
);
6636 preg
->buffer
= NULL
;
6638 preg
->allocated
= 0;
6641 free (preg
->fastmap
);
6642 preg
->fastmap
= NULL
;
6643 preg
->fastmap_accurate
= 0;
6645 free (preg
->translate
);
6646 preg
->translate
= NULL
;
6648 WEAK_ALIAS (__regfree
, regfree
)
6650 #endif /* not emacs */