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) 1993-2016 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 (STDERR_FILENO
, "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 (STDERR_FILENO
, "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) <= 0240) \
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 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1201 also be assigned to arbitrarily: each pattern buffer stores its own
1202 syntax, so it can be changed between regex compilations. */
1203 /* This has no initializer because initialized variables in Emacs
1204 become read-only after dumping. */
1205 reg_syntax_t re_syntax_options
;
1208 /* Specify the precise syntax of regexps for compilation. This provides
1209 for compatibility for various utilities which historically have
1210 different, incompatible syntaxes.
1212 The argument SYNTAX is a bit mask comprised of the various bits
1213 defined in regex.h. We return the old syntax. */
1216 re_set_syntax (reg_syntax_t syntax
)
1218 reg_syntax_t ret
= re_syntax_options
;
1220 re_syntax_options
= syntax
;
1223 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1225 /* Regexp to use to replace spaces, or NULL meaning don't. */
1226 static const_re_char
*whitespace_regexp
;
1229 re_set_whitespace_regexp (const char *regexp
)
1231 whitespace_regexp
= (const_re_char
*) regexp
;
1233 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1235 /* This table gives an error message for each of the error codes listed
1236 in regex.h. Obviously the order here has to be same as there.
1237 POSIX doesn't require that we do anything for REG_NOERROR,
1238 but why not be nice? */
1240 static const char *re_error_msgid
[] =
1242 gettext_noop ("Success"), /* REG_NOERROR */
1243 gettext_noop ("No match"), /* REG_NOMATCH */
1244 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1245 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1246 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1247 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1248 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1249 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1250 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1251 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1252 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1253 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1254 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1255 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1256 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1257 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1258 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1259 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1262 /* Avoiding alloca during matching, to placate r_alloc. */
1264 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1265 searching and matching functions should not call alloca. On some
1266 systems, alloca is implemented in terms of malloc, and if we're
1267 using the relocating allocator routines, then malloc could cause a
1268 relocation, which might (if the strings being searched are in the
1269 ralloc heap) shift the data out from underneath the regexp
1272 Here's another reason to avoid allocation: Emacs
1273 processes input from X in a signal handler; processing X input may
1274 call malloc; if input arrives while a matching routine is calling
1275 malloc, then we're scrod. But Emacs can't just block input while
1276 calling matching routines; then we don't notice interrupts when
1277 they come in. So, Emacs blocks input around all regexp calls
1278 except the matching calls, which it leaves unprotected, in the
1279 faith that they will not malloc. */
1281 /* Normally, this is fine. */
1282 #define MATCH_MAY_ALLOCATE
1284 /* The match routines may not allocate if (1) they would do it with malloc
1285 and (2) it's not safe for them to use malloc.
1286 Note that if REL_ALLOC is defined, matching would not use malloc for the
1287 failure stack, but we would still use it for the register vectors;
1288 so REL_ALLOC should not affect this. */
1289 #if defined REGEX_MALLOC && defined emacs
1290 # undef MATCH_MAY_ALLOCATE
1294 /* Failure stack declarations and macros; both re_compile_fastmap and
1295 re_match_2 use a failure stack. These have to be macros because of
1296 REGEX_ALLOCATE_STACK. */
1299 /* Approximate number of failure points for which to initially allocate space
1300 when matching. If this number is exceeded, we allocate more
1301 space, so it is not a hard limit. */
1302 #ifndef INIT_FAILURE_ALLOC
1303 # define INIT_FAILURE_ALLOC 20
1306 /* Roughly the maximum number of failure points on the stack. Would be
1307 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1308 This is a variable only so users of regex can assign to it; we never
1309 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1310 before using it, so it should probably be a byte-count instead. */
1311 # if defined MATCH_MAY_ALLOCATE
1312 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1313 whose default stack limit is 2mb. In order for a larger
1314 value to work reliably, you have to try to make it accord
1315 with the process stack limit. */
1316 size_t re_max_failures
= 40000;
1318 size_t re_max_failures
= 4000;
1321 union fail_stack_elt
1324 /* This should be the biggest `int' that's no bigger than a pointer. */
1328 typedef union fail_stack_elt fail_stack_elt_t
;
1332 fail_stack_elt_t
*stack
;
1334 size_t avail
; /* Offset of next open position. */
1335 size_t frame
; /* Offset of the cur constructed frame. */
1338 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1341 /* Define macros to initialize and free the failure stack.
1342 Do `return -2' if the alloc fails. */
1344 #ifdef MATCH_MAY_ALLOCATE
1345 # define INIT_FAIL_STACK() \
1347 fail_stack.stack = \
1348 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1349 * sizeof (fail_stack_elt_t)); \
1351 if (fail_stack.stack == NULL) \
1354 fail_stack.size = INIT_FAILURE_ALLOC; \
1355 fail_stack.avail = 0; \
1356 fail_stack.frame = 0; \
1359 # define INIT_FAIL_STACK() \
1361 fail_stack.avail = 0; \
1362 fail_stack.frame = 0; \
1365 # define RETALLOC_IF(addr, n, t) \
1366 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1370 /* Double the size of FAIL_STACK, up to a limit
1371 which allows approximately `re_max_failures' items.
1373 Return 1 if succeeds, and 0 if either ran out of memory
1374 allocating space for it or it was already too large.
1376 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1378 /* Factor to increase the failure stack size by
1379 when we increase it.
1380 This used to be 2, but 2 was too wasteful
1381 because the old discarded stacks added up to as much space
1382 were as ultimate, maximum-size stack. */
1383 #define FAIL_STACK_GROWTH_FACTOR 4
1385 #define GROW_FAIL_STACK(fail_stack) \
1386 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1387 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1389 : ((fail_stack).stack \
1390 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1391 (fail_stack).size * sizeof (fail_stack_elt_t), \
1392 min (re_max_failures * TYPICAL_FAILURE_SIZE, \
1393 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1394 * FAIL_STACK_GROWTH_FACTOR))), \
1396 (fail_stack).stack == NULL \
1398 : ((fail_stack).size \
1399 = (min (re_max_failures * TYPICAL_FAILURE_SIZE, \
1400 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1401 * FAIL_STACK_GROWTH_FACTOR)) \
1402 / sizeof (fail_stack_elt_t)), \
1406 /* Push a pointer value onto the failure stack.
1407 Assumes the variable `fail_stack'. Probably should only
1408 be called from within `PUSH_FAILURE_POINT'. */
1409 #define PUSH_FAILURE_POINTER(item) \
1410 fail_stack.stack[fail_stack.avail++].pointer = (item)
1412 /* This pushes an integer-valued item onto the failure stack.
1413 Assumes the variable `fail_stack'. Probably should only
1414 be called from within `PUSH_FAILURE_POINT'. */
1415 #define PUSH_FAILURE_INT(item) \
1416 fail_stack.stack[fail_stack.avail++].integer = (item)
1418 /* These POP... operations complement the PUSH... operations.
1419 All assume that `fail_stack' is nonempty. */
1420 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1421 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1423 /* Individual items aside from the registers. */
1424 #define NUM_NONREG_ITEMS 3
1426 /* Used to examine the stack (to detect infinite loops). */
1427 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1428 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1429 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1430 #define TOP_FAILURE_HANDLE() fail_stack.frame
1433 #define ENSURE_FAIL_STACK(space) \
1434 while (REMAINING_AVAIL_SLOTS <= space) { \
1435 if (!GROW_FAIL_STACK (fail_stack)) \
1437 DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\
1438 DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\
1441 /* Push register NUM onto the stack. */
1442 #define PUSH_FAILURE_REG(num) \
1444 char *destination; \
1446 ENSURE_FAIL_STACK(3); \
1447 DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \
1448 n, regstart[n], regend[n]); \
1449 PUSH_FAILURE_POINTER (regstart[n]); \
1450 PUSH_FAILURE_POINTER (regend[n]); \
1451 PUSH_FAILURE_INT (n); \
1454 /* Change the counter's value to VAL, but make sure that it will
1455 be reset when backtracking. */
1456 #define PUSH_NUMBER(ptr,val) \
1458 char *destination; \
1460 ENSURE_FAIL_STACK(3); \
1461 EXTRACT_NUMBER (c, ptr); \
1462 DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \
1463 PUSH_FAILURE_INT (c); \
1464 PUSH_FAILURE_POINTER (ptr); \
1465 PUSH_FAILURE_INT (-1); \
1466 STORE_NUMBER (ptr, val); \
1469 /* Pop a saved register off the stack. */
1470 #define POP_FAILURE_REG_OR_COUNT() \
1472 long pfreg = POP_FAILURE_INT (); \
1475 /* It's a counter. */ \
1476 /* Here, we discard `const', making re_match non-reentrant. */ \
1477 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1478 pfreg = POP_FAILURE_INT (); \
1479 STORE_NUMBER (ptr, pfreg); \
1480 DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \
1484 regend[pfreg] = POP_FAILURE_POINTER (); \
1485 regstart[pfreg] = POP_FAILURE_POINTER (); \
1486 DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \
1487 pfreg, regstart[pfreg], regend[pfreg]); \
1491 /* Check that we are not stuck in an infinite loop. */
1492 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1494 ssize_t failure = TOP_FAILURE_HANDLE (); \
1495 /* Check for infinite matching loops */ \
1496 while (failure > 0 \
1497 && (FAILURE_STR (failure) == string_place \
1498 || FAILURE_STR (failure) == NULL)) \
1500 assert (FAILURE_PAT (failure) >= bufp->buffer \
1501 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1502 if (FAILURE_PAT (failure) == pat_cur) \
1507 DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1508 failure = NEXT_FAILURE_HANDLE(failure); \
1510 DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \
1513 /* Push the information about the state we will need
1514 if we ever fail back to it.
1516 Requires variables fail_stack, regstart, regend and
1517 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1520 Does `return FAILURE_CODE' if runs out of memory. */
1522 #define PUSH_FAILURE_POINT(pattern, string_place) \
1524 char *destination; \
1525 /* Must be int, so when we don't save any registers, the arithmetic \
1526 of 0 + -1 isn't done as unsigned. */ \
1528 DEBUG_STATEMENT (nfailure_points_pushed++); \
1529 DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \
1530 DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \
1531 DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\
1533 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1535 DEBUG_PRINT ("\n"); \
1537 DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \
1538 PUSH_FAILURE_INT (fail_stack.frame); \
1540 DEBUG_PRINT (" Push string %p: \"", string_place); \
1541 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1542 DEBUG_PRINT ("\"\n"); \
1543 PUSH_FAILURE_POINTER (string_place); \
1545 DEBUG_PRINT (" Push pattern %p: ", pattern); \
1546 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1547 PUSH_FAILURE_POINTER (pattern); \
1549 /* Close the frame by moving the frame pointer past it. */ \
1550 fail_stack.frame = fail_stack.avail; \
1553 /* Estimate the size of data pushed by a typical failure stack entry.
1554 An estimate is all we need, because all we use this for
1555 is to choose a limit for how big to make the failure stack. */
1556 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1557 #define TYPICAL_FAILURE_SIZE 20
1559 /* How many items can still be added to the stack without overflowing it. */
1560 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1563 /* Pops what PUSH_FAIL_STACK pushes.
1565 We restore into the parameters, all of which should be lvalues:
1566 STR -- the saved data position.
1567 PAT -- the saved pattern position.
1568 REGSTART, REGEND -- arrays of string positions.
1570 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1571 `pend', `string1', `size1', `string2', and `size2'. */
1573 #define POP_FAILURE_POINT(str, pat) \
1575 assert (!FAIL_STACK_EMPTY ()); \
1577 /* Remove failure points and point to how many regs pushed. */ \
1578 DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \
1579 DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \
1580 DEBUG_PRINT (" size: %zd\n", fail_stack.size); \
1582 /* Pop the saved registers. */ \
1583 while (fail_stack.frame < fail_stack.avail) \
1584 POP_FAILURE_REG_OR_COUNT (); \
1586 pat = POP_FAILURE_POINTER (); \
1587 DEBUG_PRINT (" Popping pattern %p: ", pat); \
1588 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1590 /* If the saved string location is NULL, it came from an \
1591 on_failure_keep_string_jump opcode, and we want to throw away the \
1592 saved NULL, thus retaining our current position in the string. */ \
1593 str = POP_FAILURE_POINTER (); \
1594 DEBUG_PRINT (" Popping string %p: \"", str); \
1595 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1596 DEBUG_PRINT ("\"\n"); \
1598 fail_stack.frame = POP_FAILURE_INT (); \
1599 DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \
1601 assert (fail_stack.avail >= 0); \
1602 assert (fail_stack.frame <= fail_stack.avail); \
1604 DEBUG_STATEMENT (nfailure_points_popped++); \
1605 } while (0) /* POP_FAILURE_POINT */
1609 /* Registers are set to a sentinel when they haven't yet matched. */
1610 #define REG_UNSET(e) ((e) == NULL)
1612 /* Subroutine declarations and macros for regex_compile. */
1614 static reg_errcode_t
regex_compile (re_char
*pattern
, size_t size
,
1615 reg_syntax_t syntax
,
1616 struct re_pattern_buffer
*bufp
);
1617 static void store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
);
1618 static void store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
);
1619 static void insert_op1 (re_opcode_t op
, unsigned char *loc
,
1620 int arg
, unsigned char *end
);
1621 static void insert_op2 (re_opcode_t op
, unsigned char *loc
,
1622 int arg1
, int arg2
, unsigned char *end
);
1623 static boolean
at_begline_loc_p (re_char
*pattern
, re_char
*p
,
1624 reg_syntax_t syntax
);
1625 static boolean
at_endline_loc_p (re_char
*p
, re_char
*pend
,
1626 reg_syntax_t syntax
);
1627 static re_char
*skip_one_char (re_char
*p
);
1628 static int analyze_first (re_char
*p
, re_char
*pend
,
1629 char *fastmap
, const int multibyte
);
1631 /* Fetch the next character in the uncompiled pattern, with no
1633 #define PATFETCH(c) \
1636 if (p == pend) return REG_EEND; \
1637 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1642 /* If `translate' is non-null, return translate[D], else just D. We
1643 cast the subscript to translate because some data is declared as
1644 `char *', to avoid warnings when a string constant is passed. But
1645 when we use a character as a subscript we must make it unsigned. */
1647 # define TRANSLATE(d) \
1648 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1652 /* Macros for outputting the compiled pattern into `buffer'. */
1654 /* If the buffer isn't allocated when it comes in, use this. */
1655 #define INIT_BUF_SIZE 32
1657 /* Make sure we have at least N more bytes of space in buffer. */
1658 #define GET_BUFFER_SPACE(n) \
1659 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1662 /* Make sure we have one more byte of buffer space and then add C to it. */
1663 #define BUF_PUSH(c) \
1665 GET_BUFFER_SPACE (1); \
1666 *b++ = (unsigned char) (c); \
1670 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1671 #define BUF_PUSH_2(c1, c2) \
1673 GET_BUFFER_SPACE (2); \
1674 *b++ = (unsigned char) (c1); \
1675 *b++ = (unsigned char) (c2); \
1679 /* Store a jump with opcode OP at LOC to location TO. We store a
1680 relative address offset by the three bytes the jump itself occupies. */
1681 #define STORE_JUMP(op, loc, to) \
1682 store_op1 (op, loc, (to) - (loc) - 3)
1684 /* Likewise, for a two-argument jump. */
1685 #define STORE_JUMP2(op, loc, to, arg) \
1686 store_op2 (op, loc, (to) - (loc) - 3, arg)
1688 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1689 #define INSERT_JUMP(op, loc, to) \
1690 insert_op1 (op, loc, (to) - (loc) - 3, b)
1692 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1693 #define INSERT_JUMP2(op, loc, to, arg) \
1694 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1697 /* This is not an arbitrary limit: the arguments which represent offsets
1698 into the pattern are two bytes long. So if 2^15 bytes turns out to
1699 be too small, many things would have to change. */
1700 # define MAX_BUF_SIZE (1L << 15)
1702 /* Extend the buffer by twice its current size via realloc and
1703 reset the pointers that pointed into the old block to point to the
1704 correct places in the new one. If extending the buffer results in it
1705 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1706 #if __BOUNDED_POINTERS__
1707 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1708 # define MOVE_BUFFER_POINTER(P) \
1709 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1710 SET_HIGH_BOUND (P), \
1711 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1712 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1715 SET_HIGH_BOUND (b); \
1716 SET_HIGH_BOUND (begalt); \
1717 if (fixup_alt_jump) \
1718 SET_HIGH_BOUND (fixup_alt_jump); \
1720 SET_HIGH_BOUND (laststart); \
1721 if (pending_exact) \
1722 SET_HIGH_BOUND (pending_exact); \
1725 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1726 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1728 #define EXTEND_BUFFER() \
1730 unsigned char *old_buffer = bufp->buffer; \
1731 if (bufp->allocated == MAX_BUF_SIZE) \
1733 bufp->allocated <<= 1; \
1734 if (bufp->allocated > MAX_BUF_SIZE) \
1735 bufp->allocated = MAX_BUF_SIZE; \
1736 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1737 if (bufp->buffer == NULL) \
1738 return REG_ESPACE; \
1739 /* If the buffer moved, move all the pointers into it. */ \
1740 if (old_buffer != bufp->buffer) \
1742 unsigned char *new_buffer = bufp->buffer; \
1743 MOVE_BUFFER_POINTER (b); \
1744 MOVE_BUFFER_POINTER (begalt); \
1745 if (fixup_alt_jump) \
1746 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1748 MOVE_BUFFER_POINTER (laststart); \
1749 if (pending_exact) \
1750 MOVE_BUFFER_POINTER (pending_exact); \
1752 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1756 /* Since we have one byte reserved for the register number argument to
1757 {start,stop}_memory, the maximum number of groups we can report
1758 things about is what fits in that byte. */
1759 #define MAX_REGNUM 255
1761 /* But patterns can have more than `MAX_REGNUM' registers. We just
1762 ignore the excess. */
1763 typedef int regnum_t
;
1766 /* Macros for the compile stack. */
1768 /* Since offsets can go either forwards or backwards, this type needs to
1769 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1770 /* int may be not enough when sizeof(int) == 2. */
1771 typedef long pattern_offset_t
;
1775 pattern_offset_t begalt_offset
;
1776 pattern_offset_t fixup_alt_jump
;
1777 pattern_offset_t laststart_offset
;
1779 } compile_stack_elt_t
;
1784 compile_stack_elt_t
*stack
;
1786 size_t avail
; /* Offset of next open position. */
1787 } compile_stack_type
;
1790 #define INIT_COMPILE_STACK_SIZE 32
1792 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1793 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1795 /* The next available element. */
1796 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1798 /* Explicit quit checking is needed for Emacs, which uses polling to
1799 process input events. */
1801 # define IMMEDIATE_QUIT_CHECK \
1803 if (immediate_quit) QUIT; \
1806 # define IMMEDIATE_QUIT_CHECK ((void)0)
1809 /* Structure to manage work area for range table. */
1810 struct range_table_work_area
1812 int *table
; /* actual work area. */
1813 int allocated
; /* allocated size for work area in bytes. */
1814 int used
; /* actually used size in words. */
1815 int bits
; /* flag to record character classes */
1820 /* Make sure that WORK_AREA can hold more N multibyte characters.
1821 This is used only in set_image_of_range and set_image_of_range_1.
1822 It expects WORK_AREA to be a pointer.
1823 If it can't get the space, it returns from the surrounding function. */
1825 #define EXTEND_RANGE_TABLE(work_area, n) \
1827 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1829 extend_range_table_work_area (&work_area); \
1830 if ((work_area).table == 0) \
1831 return (REG_ESPACE); \
1835 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1836 (work_area).bits |= (bit)
1838 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1839 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1841 EXTEND_RANGE_TABLE ((work_area), 2); \
1842 (work_area).table[(work_area).used++] = (range_start); \
1843 (work_area).table[(work_area).used++] = (range_end); \
1848 /* Free allocated memory for WORK_AREA. */
1849 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1851 if ((work_area).table) \
1852 free ((work_area).table); \
1855 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1856 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1857 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1858 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1860 /* Bits used to implement the multibyte-part of the various character classes
1861 such as [:alnum:] in a charset's range table. The code currently assumes
1862 that only the low 16 bits are used. */
1863 #define BIT_WORD 0x1
1864 #define BIT_LOWER 0x2
1865 #define BIT_PUNCT 0x4
1866 #define BIT_SPACE 0x8
1867 #define BIT_UPPER 0x10
1868 #define BIT_MULTIBYTE 0x20
1869 #define BIT_ALPHA 0x40
1870 #define BIT_ALNUM 0x80
1871 #define BIT_GRAPH 0x100
1872 #define BIT_PRINT 0x200
1875 /* Set the bit for character C in a list. */
1876 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1881 /* Store characters in the range FROM to TO in the bitmap at B (for
1882 ASCII and unibyte characters) and WORK_AREA (for multibyte
1883 characters) while translating them and paying attention to the
1884 continuity of translated characters.
1886 Implementation note: It is better to implement these fairly big
1887 macros by a function, but it's not that easy because macros called
1888 in this macro assume various local variables already declared. */
1890 /* Both FROM and TO are ASCII characters. */
1892 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1896 for (C0 = (FROM); C0 <= (TO); C0++) \
1898 C1 = TRANSLATE (C0); \
1899 if (! ASCII_CHAR_P (C1)) \
1901 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1902 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1905 SET_LIST_BIT (C1); \
1910 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1912 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1914 int C0, C1, C2, I; \
1915 int USED = RANGE_TABLE_WORK_USED (work_area); \
1917 for (C0 = (FROM); C0 <= (TO); C0++) \
1919 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1920 if (CHAR_BYTE8_P (C1)) \
1921 SET_LIST_BIT (C0); \
1924 C2 = TRANSLATE (C1); \
1926 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1928 SET_LIST_BIT (C1); \
1929 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1931 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1932 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1934 if (C2 >= from - 1 && C2 <= to + 1) \
1936 if (C2 == from - 1) \
1937 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1938 else if (C2 == to + 1) \
1939 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1944 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1950 /* Both FROM and TO are multibyte characters. */
1952 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1954 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1956 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1957 for (C0 = (FROM); C0 <= (TO); C0++) \
1959 C1 = TRANSLATE (C0); \
1960 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1961 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1962 SET_LIST_BIT (C2); \
1963 if (C1 >= (FROM) && C1 <= (TO)) \
1965 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1967 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1968 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1970 if (C1 >= from - 1 && C1 <= to + 1) \
1972 if (C1 == from - 1) \
1973 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1974 else if (C1 == to + 1) \
1975 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1980 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1986 /* Get the next unsigned number in the uncompiled pattern. */
1987 #define GET_INTERVAL_COUNT(num) \
1990 FREE_STACK_RETURN (REG_EBRACE); \
1994 while ('0' <= c && c <= '9') \
1998 if (RE_DUP_MAX / 10 - (RE_DUP_MAX % 10 < c - '0') < num) \
1999 FREE_STACK_RETURN (REG_BADBR); \
2000 num = num * 10 + c - '0'; \
2002 FREE_STACK_RETURN (REG_EBRACE); \
2008 #if ! WIDE_CHAR_SUPPORT
2010 /* Map a string to the char class it names (if any). */
2012 re_wctype (const_re_char
*str
)
2014 const char *string
= (const char *) str
;
2015 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2016 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2017 else if (STREQ (string
, "word")) return RECC_WORD
;
2018 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2019 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2020 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2021 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2022 else if (STREQ (string
, "print")) return RECC_PRINT
;
2023 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2024 else if (STREQ (string
, "space")) return RECC_SPACE
;
2025 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2026 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2027 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2028 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2029 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2030 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2031 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2035 /* True if CH is in the char class CC. */
2037 re_iswctype (int ch
, re_wctype_t cc
)
2041 case RECC_ALNUM
: return ISALNUM (ch
) != 0;
2042 case RECC_ALPHA
: return ISALPHA (ch
) != 0;
2043 case RECC_BLANK
: return ISBLANK (ch
) != 0;
2044 case RECC_CNTRL
: return ISCNTRL (ch
) != 0;
2045 case RECC_DIGIT
: return ISDIGIT (ch
) != 0;
2046 case RECC_GRAPH
: return ISGRAPH (ch
) != 0;
2047 case RECC_LOWER
: return ISLOWER (ch
) != 0;
2048 case RECC_PRINT
: return ISPRINT (ch
) != 0;
2049 case RECC_PUNCT
: return ISPUNCT (ch
) != 0;
2050 case RECC_SPACE
: return ISSPACE (ch
) != 0;
2051 case RECC_UPPER
: return ISUPPER (ch
) != 0;
2052 case RECC_XDIGIT
: return ISXDIGIT (ch
) != 0;
2053 case RECC_ASCII
: return IS_REAL_ASCII (ch
) != 0;
2054 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2055 case RECC_UNIBYTE
: return ISUNIBYTE (ch
) != 0;
2056 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2057 case RECC_WORD
: return ISWORD (ch
) != 0;
2058 case RECC_ERROR
: return false;
2064 /* Return a bit-pattern to use in the range-table bits to match multibyte
2065 chars of class CC. */
2067 re_wctype_to_bit (re_wctype_t cc
)
2072 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2073 case RECC_ALPHA
: return BIT_ALPHA
;
2074 case RECC_ALNUM
: return BIT_ALNUM
;
2075 case RECC_WORD
: return BIT_WORD
;
2076 case RECC_LOWER
: return BIT_LOWER
;
2077 case RECC_UPPER
: return BIT_UPPER
;
2078 case RECC_PUNCT
: return BIT_PUNCT
;
2079 case RECC_SPACE
: return BIT_SPACE
;
2080 case RECC_GRAPH
: return BIT_GRAPH
;
2081 case RECC_PRINT
: return BIT_PRINT
;
2082 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2083 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2090 /* Filling in the work area of a range. */
2092 /* Actually extend the space in WORK_AREA. */
2095 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2097 work_area
->allocated
+= 16 * sizeof (int);
2098 work_area
->table
= realloc (work_area
->table
, work_area
->allocated
);
2104 /* Carefully find the ranges of codes that are equivalent
2105 under case conversion to the range start..end when passed through
2106 TRANSLATE. Handle the case where non-letters can come in between
2107 two upper-case letters (which happens in Latin-1).
2108 Also handle the case of groups of more than 2 case-equivalent chars.
2110 The basic method is to look at consecutive characters and see
2111 if they can form a run that can be handled as one.
2113 Returns -1 if successful, REG_ESPACE if ran out of space. */
2116 set_image_of_range_1 (struct range_table_work_area
*work_area
,
2117 re_wchar_t start
, re_wchar_t end
,
2118 RE_TRANSLATE_TYPE translate
)
2120 /* `one_case' indicates a character, or a run of characters,
2121 each of which is an isolate (no case-equivalents).
2122 This includes all ASCII non-letters.
2124 `two_case' indicates a character, or a run of characters,
2125 each of which has two case-equivalent forms.
2126 This includes all ASCII letters.
2128 `strange' indicates a character that has more than one
2131 enum case_type
{one_case
, two_case
, strange
};
2133 /* Describe the run that is in progress,
2134 which the next character can try to extend.
2135 If run_type is strange, that means there really is no run.
2136 If run_type is one_case, then run_start...run_end is the run.
2137 If run_type is two_case, then the run is run_start...run_end,
2138 and the case-equivalents end at run_eqv_end. */
2140 enum case_type run_type
= strange
;
2141 int run_start
, run_end
, run_eqv_end
;
2143 Lisp_Object eqv_table
;
2145 if (!RE_TRANSLATE_P (translate
))
2147 EXTEND_RANGE_TABLE (work_area
, 2);
2148 work_area
->table
[work_area
->used
++] = (start
);
2149 work_area
->table
[work_area
->used
++] = (end
);
2153 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2155 for (; start
<= end
; start
++)
2157 enum case_type this_type
;
2158 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2159 int minchar
, maxchar
;
2161 /* Classify this character */
2163 this_type
= one_case
;
2164 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2165 this_type
= two_case
;
2167 this_type
= strange
;
2170 minchar
= start
, maxchar
= eqv
;
2172 minchar
= eqv
, maxchar
= start
;
2174 /* Can this character extend the run in progress? */
2175 if (this_type
== strange
|| this_type
!= run_type
2176 || !(minchar
== run_end
+ 1
2177 && (run_type
== two_case
2178 ? maxchar
== run_eqv_end
+ 1 : 1)))
2181 Record each of its equivalent ranges. */
2182 if (run_type
== one_case
)
2184 EXTEND_RANGE_TABLE (work_area
, 2);
2185 work_area
->table
[work_area
->used
++] = run_start
;
2186 work_area
->table
[work_area
->used
++] = run_end
;
2188 else if (run_type
== two_case
)
2190 EXTEND_RANGE_TABLE (work_area
, 4);
2191 work_area
->table
[work_area
->used
++] = run_start
;
2192 work_area
->table
[work_area
->used
++] = run_end
;
2193 work_area
->table
[work_area
->used
++]
2194 = RE_TRANSLATE (eqv_table
, run_start
);
2195 work_area
->table
[work_area
->used
++]
2196 = RE_TRANSLATE (eqv_table
, run_end
);
2201 if (this_type
== strange
)
2203 /* For a strange character, add each of its equivalents, one
2204 by one. Don't start a range. */
2207 EXTEND_RANGE_TABLE (work_area
, 2);
2208 work_area
->table
[work_area
->used
++] = eqv
;
2209 work_area
->table
[work_area
->used
++] = eqv
;
2210 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2212 while (eqv
!= start
);
2215 /* Add this char to the run, or start a new run. */
2216 else if (run_type
== strange
)
2218 /* Initialize a new range. */
2219 run_type
= this_type
;
2222 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2226 /* Extend a running range. */
2228 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2232 /* If a run is still in progress at the end, finish it now
2233 by recording its equivalent ranges. */
2234 if (run_type
== one_case
)
2236 EXTEND_RANGE_TABLE (work_area
, 2);
2237 work_area
->table
[work_area
->used
++] = run_start
;
2238 work_area
->table
[work_area
->used
++] = run_end
;
2240 else if (run_type
== two_case
)
2242 EXTEND_RANGE_TABLE (work_area
, 4);
2243 work_area
->table
[work_area
->used
++] = run_start
;
2244 work_area
->table
[work_area
->used
++] = run_end
;
2245 work_area
->table
[work_area
->used
++]
2246 = RE_TRANSLATE (eqv_table
, run_start
);
2247 work_area
->table
[work_area
->used
++]
2248 = RE_TRANSLATE (eqv_table
, run_end
);
2256 /* Record the image of the range start..end when passed through
2257 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2258 and is not even necessarily contiguous.
2259 Normally we approximate it with the smallest contiguous range that contains
2260 all the chars we need. However, for Latin-1 we go to extra effort
2263 This function is not called for ASCII ranges.
2265 Returns -1 if successful, REG_ESPACE if ran out of space. */
2268 set_image_of_range (struct range_table_work_area
*work_area
,
2269 re_wchar_t start
, re_wchar_t end
,
2270 RE_TRANSLATE_TYPE translate
)
2272 re_wchar_t cmin
, cmax
;
2275 /* For Latin-1 ranges, use set_image_of_range_1
2276 to get proper handling of ranges that include letters and nonletters.
2277 For a range that includes the whole of Latin-1, this is not necessary.
2278 For other character sets, we don't bother to get this right. */
2279 if (RE_TRANSLATE_P (translate
) && start
< 04400
2280 && !(start
< 04200 && end
>= 04377))
2287 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2297 EXTEND_RANGE_TABLE (work_area
, 2);
2298 work_area
->table
[work_area
->used
++] = (start
);
2299 work_area
->table
[work_area
->used
++] = (end
);
2301 cmin
= -1, cmax
= -1;
2303 if (RE_TRANSLATE_P (translate
))
2307 for (ch
= start
; ch
<= end
; ch
++)
2309 re_wchar_t c
= TRANSLATE (ch
);
2310 if (! (start
<= c
&& c
<= end
))
2316 cmin
= min (cmin
, c
);
2317 cmax
= max (cmax
, c
);
2324 EXTEND_RANGE_TABLE (work_area
, 2);
2325 work_area
->table
[work_area
->used
++] = (cmin
);
2326 work_area
->table
[work_area
->used
++] = (cmax
);
2334 #ifndef MATCH_MAY_ALLOCATE
2336 /* If we cannot allocate large objects within re_match_2_internal,
2337 we make the fail stack and register vectors global.
2338 The fail stack, we grow to the maximum size when a regexp
2340 The register vectors, we adjust in size each time we
2341 compile a regexp, according to the number of registers it needs. */
2343 static fail_stack_type fail_stack
;
2345 /* Size with which the following vectors are currently allocated.
2346 That is so we can make them bigger as needed,
2347 but never make them smaller. */
2348 static int regs_allocated_size
;
2350 static re_char
** regstart
, ** regend
;
2351 static re_char
**best_regstart
, **best_regend
;
2353 /* Make the register vectors big enough for NUM_REGS registers,
2354 but don't make them smaller. */
2357 regex_grow_registers (int num_regs
)
2359 if (num_regs
> regs_allocated_size
)
2361 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2362 RETALLOC_IF (regend
, num_regs
, re_char
*);
2363 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2364 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2366 regs_allocated_size
= num_regs
;
2370 #endif /* not MATCH_MAY_ALLOCATE */
2372 static boolean
group_in_compile_stack (compile_stack_type compile_stack
,
2375 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2376 Returns one of error codes defined in `regex.h', or zero for success.
2378 Assumes the `allocated' (and perhaps `buffer') and `translate'
2379 fields are set in BUFP on entry.
2381 If it succeeds, results are put in BUFP (if it returns an error, the
2382 contents of BUFP are undefined):
2383 `buffer' is the compiled pattern;
2384 `syntax' is set to SYNTAX;
2385 `used' is set to the length of the compiled pattern;
2386 `fastmap_accurate' is zero;
2387 `re_nsub' is the number of subexpressions in PATTERN;
2388 `not_bol' and `not_eol' are zero;
2390 The `fastmap' field is neither examined nor set. */
2392 /* Insert the `jump' from the end of last alternative to "here".
2393 The space for the jump has already been allocated. */
2394 #define FIXUP_ALT_JUMP() \
2396 if (fixup_alt_jump) \
2397 STORE_JUMP (jump, fixup_alt_jump, b); \
2401 /* Return, freeing storage we allocated. */
2402 #define FREE_STACK_RETURN(value) \
2404 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2405 free (compile_stack.stack); \
2409 static reg_errcode_t
2410 regex_compile (const_re_char
*pattern
, size_t size
, reg_syntax_t syntax
,
2411 struct re_pattern_buffer
*bufp
)
2413 /* We fetch characters from PATTERN here. */
2414 register re_wchar_t c
, c1
;
2416 /* Points to the end of the buffer, where we should append. */
2417 register unsigned char *b
;
2419 /* Keeps track of unclosed groups. */
2420 compile_stack_type compile_stack
;
2422 /* Points to the current (ending) position in the pattern. */
2424 /* `const' makes AIX compiler fail. */
2425 unsigned char *p
= pattern
;
2427 re_char
*p
= pattern
;
2429 re_char
*pend
= pattern
+ size
;
2431 /* How to translate the characters in the pattern. */
2432 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2434 /* Address of the count-byte of the most recently inserted `exactn'
2435 command. This makes it possible to tell if a new exact-match
2436 character can be added to that command or if the character requires
2437 a new `exactn' command. */
2438 unsigned char *pending_exact
= 0;
2440 /* Address of start of the most recently finished expression.
2441 This tells, e.g., postfix * where to find the start of its
2442 operand. Reset at the beginning of groups and alternatives. */
2443 unsigned char *laststart
= 0;
2445 /* Address of beginning of regexp, or inside of last group. */
2446 unsigned char *begalt
;
2448 /* Place in the uncompiled pattern (i.e., the {) to
2449 which to go back if the interval is invalid. */
2450 re_char
*beg_interval
;
2452 /* Address of the place where a forward jump should go to the end of
2453 the containing expression. Each alternative of an `or' -- except the
2454 last -- ends with a forward jump of this sort. */
2455 unsigned char *fixup_alt_jump
= 0;
2457 /* Work area for range table of charset. */
2458 struct range_table_work_area range_table_work
;
2460 /* If the object matched can contain multibyte characters. */
2461 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2463 /* Nonzero if we have pushed down into a subpattern. */
2464 int in_subpattern
= 0;
2466 /* These hold the values of p, pattern, and pend from the main
2467 pattern when we have pushed into a subpattern. */
2469 re_char
*main_pattern
;
2474 DEBUG_PRINT ("\nCompiling pattern: ");
2477 unsigned debug_count
;
2479 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2480 putchar (pattern
[debug_count
]);
2485 /* Initialize the compile stack. */
2486 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2487 if (compile_stack
.stack
== NULL
)
2490 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2491 compile_stack
.avail
= 0;
2493 range_table_work
.table
= 0;
2494 range_table_work
.allocated
= 0;
2496 /* Initialize the pattern buffer. */
2497 bufp
->syntax
= syntax
;
2498 bufp
->fastmap_accurate
= 0;
2499 bufp
->not_bol
= bufp
->not_eol
= 0;
2500 bufp
->used_syntax
= 0;
2502 /* Set `used' to zero, so that if we return an error, the pattern
2503 printer (for debugging) will think there's no pattern. We reset it
2507 /* Always count groups, whether or not bufp->no_sub is set. */
2510 #if !defined emacs && !defined SYNTAX_TABLE
2511 /* Initialize the syntax table. */
2512 init_syntax_once ();
2515 if (bufp
->allocated
== 0)
2518 { /* If zero allocated, but buffer is non-null, try to realloc
2519 enough space. This loses if buffer's address is bogus, but
2520 that is the user's responsibility. */
2521 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2524 { /* Caller did not allocate a buffer. Do it for them. */
2525 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2527 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2529 bufp
->allocated
= INIT_BUF_SIZE
;
2532 begalt
= b
= bufp
->buffer
;
2534 /* Loop through the uncompiled pattern until we're at the end. */
2539 /* If this is the end of an included regexp,
2540 pop back to the main regexp and try again. */
2544 pattern
= main_pattern
;
2549 /* If this is the end of the main regexp, we are done. */
2561 /* If there's no special whitespace regexp, treat
2562 spaces normally. And don't try to do this recursively. */
2563 if (!whitespace_regexp
|| in_subpattern
)
2566 /* Peek past following spaces. */
2573 /* If the spaces are followed by a repetition op,
2574 treat them normally. */
2576 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2577 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2580 /* Replace the spaces with the whitespace regexp. */
2584 main_pattern
= pattern
;
2585 p
= pattern
= whitespace_regexp
;
2586 pend
= p
+ strlen ((const char *) p
);
2592 if ( /* If at start of pattern, it's an operator. */
2594 /* If context independent, it's an operator. */
2595 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2596 /* Otherwise, depends on what's come before. */
2597 || at_begline_loc_p (pattern
, p
, syntax
))
2598 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2607 if ( /* If at end of pattern, it's an operator. */
2609 /* If context independent, it's an operator. */
2610 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2611 /* Otherwise, depends on what's next. */
2612 || at_endline_loc_p (p
, pend
, syntax
))
2613 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2622 if ((syntax
& RE_BK_PLUS_QM
)
2623 || (syntax
& RE_LIMITED_OPS
))
2627 /* If there is no previous pattern... */
2630 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2631 FREE_STACK_RETURN (REG_BADRPT
);
2632 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2637 /* 1 means zero (many) matches is allowed. */
2638 boolean zero_times_ok
= 0, many_times_ok
= 0;
2641 /* If there is a sequence of repetition chars, collapse it
2642 down to just one (the right one). We can't combine
2643 interval operators with these because of, e.g., `a{2}*',
2644 which should only match an even number of `a's. */
2648 if ((syntax
& RE_FRUGAL
)
2649 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2653 zero_times_ok
|= c
!= '+';
2654 many_times_ok
|= c
!= '?';
2660 || (!(syntax
& RE_BK_PLUS_QM
)
2661 && (*p
== '+' || *p
== '?')))
2663 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2666 FREE_STACK_RETURN (REG_EESCAPE
);
2667 if (p
[1] == '+' || p
[1] == '?')
2668 PATFETCH (c
); /* Gobble up the backslash. */
2674 /* If we get here, we found another repeat character. */
2678 /* Star, etc. applied to an empty pattern is equivalent
2679 to an empty pattern. */
2680 if (!laststart
|| laststart
== b
)
2683 /* Now we know whether or not zero matches is allowed
2684 and also whether or not two or more matches is allowed. */
2689 boolean simple
= skip_one_char (laststart
) == b
;
2690 size_t startoffset
= 0;
2692 /* Check if the loop can match the empty string. */
2693 (simple
|| !analyze_first (laststart
, b
, NULL
, 0))
2694 ? on_failure_jump
: on_failure_jump_loop
;
2695 assert (skip_one_char (laststart
) <= b
);
2697 if (!zero_times_ok
&& simple
)
2698 { /* Since simple * loops can be made faster by using
2699 on_failure_keep_string_jump, we turn simple P+
2700 into PP* if P is simple. */
2701 unsigned char *p1
, *p2
;
2702 startoffset
= b
- laststart
;
2703 GET_BUFFER_SPACE (startoffset
);
2704 p1
= b
; p2
= laststart
;
2710 GET_BUFFER_SPACE (6);
2713 STORE_JUMP (ofj
, b
, b
+ 6);
2715 /* Simple * loops can use on_failure_keep_string_jump
2716 depending on what follows. But since we don't know
2717 that yet, we leave the decision up to
2718 on_failure_jump_smart. */
2719 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2720 laststart
+ startoffset
, b
+ 6);
2722 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2727 /* A simple ? pattern. */
2728 assert (zero_times_ok
);
2729 GET_BUFFER_SPACE (3);
2730 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2734 else /* not greedy */
2735 { /* I wish the greedy and non-greedy cases could be merged. */
2737 GET_BUFFER_SPACE (7); /* We might use less. */
2740 boolean emptyp
= analyze_first (laststart
, b
, NULL
, 0);
2742 /* The non-greedy multiple match looks like
2743 a repeat..until: we only need a conditional jump
2744 at the end of the loop. */
2745 if (emptyp
) BUF_PUSH (no_op
);
2746 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2747 : on_failure_jump
, b
, laststart
);
2751 /* The repeat...until naturally matches one or more.
2752 To also match zero times, we need to first jump to
2753 the end of the loop (its conditional jump). */
2754 INSERT_JUMP (jump
, laststart
, b
);
2760 /* non-greedy a?? */
2761 INSERT_JUMP (jump
, laststart
, b
+ 3);
2763 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2782 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2784 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2786 /* Ensure that we have enough space to push a charset: the
2787 opcode, the length count, and the bitset; 34 bytes in all. */
2788 GET_BUFFER_SPACE (34);
2792 /* We test `*p == '^' twice, instead of using an if
2793 statement, so we only need one BUF_PUSH. */
2794 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2798 /* Remember the first position in the bracket expression. */
2801 /* Push the number of bytes in the bitmap. */
2802 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2804 /* Clear the whole map. */
2805 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2807 /* charset_not matches newline according to a syntax bit. */
2808 if ((re_opcode_t
) b
[-2] == charset_not
2809 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2810 SET_LIST_BIT ('\n');
2812 /* Read in characters and ranges, setting map bits. */
2815 boolean escaped_char
= false;
2816 const unsigned char *p2
= p
;
2819 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2821 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2822 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2823 So the translation is done later in a loop. Example:
2824 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2827 /* \ might escape characters inside [...] and [^...]. */
2828 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2830 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2833 escaped_char
= true;
2837 /* Could be the end of the bracket expression. If it's
2838 not (i.e., when the bracket expression is `[]' so
2839 far), the ']' character bit gets set way below. */
2840 if (c
== ']' && p2
!= p1
)
2844 /* See if we're at the beginning of a possible character
2847 if (!escaped_char
&&
2848 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2850 /* Leave room for the null. */
2851 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2852 const unsigned char *class_beg
;
2858 /* If pattern is `[[:'. */
2859 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2864 if ((c
== ':' && *p
== ']') || p
== pend
)
2866 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2869 /* This is in any case an invalid class name. */
2874 /* If isn't a word bracketed by `[:' and `:]':
2875 undo the ending character, the letters, and
2876 leave the leading `:' and `[' (but set bits for
2878 if (c
== ':' && *p
== ']')
2880 re_wctype_t cc
= re_wctype (str
);
2883 FREE_STACK_RETURN (REG_ECTYPE
);
2885 /* Throw away the ] at the end of the character
2889 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2892 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2893 if (re_iswctype (btowc (ch
), cc
))
2896 if (c
< (1 << BYTEWIDTH
))
2900 /* Most character classes in a multibyte match
2901 just set a flag. Exceptions are is_blank,
2902 is_digit, is_cntrl, and is_xdigit, since
2903 they can only match ASCII characters. We
2904 don't need to handle them for multibyte.
2905 They are distinguished by a negative wctype. */
2907 /* Setup the gl_state object to its buffer-defined
2908 value. This hardcodes the buffer-global
2909 syntax-table for ASCII chars, while the other chars
2910 will obey syntax-table properties. It's not ideal,
2911 but it's the way it's been done until now. */
2912 SETUP_BUFFER_SYNTAX_TABLE ();
2914 for (ch
= 0; ch
< 256; ++ch
)
2916 c
= RE_CHAR_TO_MULTIBYTE (ch
);
2917 if (! CHAR_BYTE8_P (c
)
2918 && re_iswctype (c
, cc
))
2924 if (ASCII_CHAR_P (c1
))
2926 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
2930 SET_RANGE_TABLE_WORK_AREA_BIT
2931 (range_table_work
, re_wctype_to_bit (cc
));
2933 /* In most cases the matching rule for char classes
2934 only uses the syntax table for multibyte chars,
2935 so that the content of the syntax-table is not
2936 hardcoded in the range_table. SPACE and WORD are
2937 the two exceptions. */
2938 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2939 bufp
->used_syntax
= 1;
2941 /* Repeat the loop. */
2946 /* Go back to right after the "[:". */
2950 /* Because the `:' may start the range, we
2951 can't simply set bit and repeat the loop.
2952 Instead, just set it to C and handle below. */
2957 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2960 /* Discard the `-'. */
2963 /* Fetch the character which ends the range. */
2966 if (CHAR_BYTE8_P (c1
)
2967 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
2968 /* Treat the range from a multibyte character to
2969 raw-byte character as empty. */
2974 /* Range from C to C. */
2979 if (syntax
& RE_NO_EMPTY_RANGES
)
2980 FREE_STACK_RETURN (REG_ERANGEX
);
2981 /* Else, repeat the loop. */
2986 /* Set the range into bitmap */
2987 for (; c
<= c1
; c
++)
2990 if (ch
< (1 << BYTEWIDTH
))
2997 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
2999 if (CHAR_BYTE8_P (c1
))
3000 c
= BYTE8_TO_CHAR (128);
3004 if (CHAR_BYTE8_P (c
))
3006 c
= CHAR_TO_BYTE8 (c
);
3007 c1
= CHAR_TO_BYTE8 (c1
);
3008 for (; c
<= c1
; c
++)
3013 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
3017 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
3024 /* Discard any (non)matching list bytes that are all 0 at the
3025 end of the map. Decrease the map-length byte too. */
3026 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3030 /* Build real range table from work area. */
3031 if (RANGE_TABLE_WORK_USED (range_table_work
)
3032 || RANGE_TABLE_WORK_BITS (range_table_work
))
3035 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3037 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3038 bytes for flags, two for COUNT, and three bytes for
3040 GET_BUFFER_SPACE (4 + used
* 3);
3042 /* Indicate the existence of range table. */
3043 laststart
[1] |= 0x80;
3045 /* Store the character class flag bits into the range table.
3046 If not in emacs, these flag bits are always 0. */
3047 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3048 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3050 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3051 for (i
= 0; i
< used
; i
++)
3052 STORE_CHARACTER_AND_INCR
3053 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3060 if (syntax
& RE_NO_BK_PARENS
)
3067 if (syntax
& RE_NO_BK_PARENS
)
3074 if (syntax
& RE_NEWLINE_ALT
)
3081 if (syntax
& RE_NO_BK_VBAR
)
3088 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3089 goto handle_interval
;
3095 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3097 /* Do not translate the character after the \, so that we can
3098 distinguish, e.g., \B from \b, even if we normally would
3099 translate, e.g., B to b. */
3105 if (syntax
& RE_NO_BK_PARENS
)
3106 goto normal_backslash
;
3111 regnum_t regnum
= 0;
3114 /* Look for a special (?...) construct */
3115 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3117 PATFETCH (c
); /* Gobble up the '?'. */
3123 case ':': shy
= 1; break;
3125 /* An explicitly specified regnum must start
3128 FREE_STACK_RETURN (REG_BADPAT
);
3129 case '1': case '2': case '3': case '4':
3130 case '5': case '6': case '7': case '8': case '9':
3131 regnum
= 10*regnum
+ (c
- '0'); break;
3133 /* Only (?:...) is supported right now. */
3134 FREE_STACK_RETURN (REG_BADPAT
);
3141 regnum
= ++bufp
->re_nsub
;
3143 { /* It's actually not shy, but explicitly numbered. */
3145 if (regnum
> bufp
->re_nsub
)
3146 bufp
->re_nsub
= regnum
;
3147 else if (regnum
> bufp
->re_nsub
3148 /* Ideally, we'd want to check that the specified
3149 group can't have matched (i.e. all subgroups
3150 using the same regnum are in other branches of
3151 OR patterns), but we don't currently keep track
3152 of enough info to do that easily. */
3153 || group_in_compile_stack (compile_stack
, regnum
))
3154 FREE_STACK_RETURN (REG_BADPAT
);
3157 /* It's really shy. */
3158 regnum
= - bufp
->re_nsub
;
3160 if (COMPILE_STACK_FULL
)
3162 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3163 compile_stack_elt_t
);
3164 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3166 compile_stack
.size
<<= 1;
3169 /* These are the values to restore when we hit end of this
3170 group. They are all relative offsets, so that if the
3171 whole pattern moves because of realloc, they will still
3173 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3174 COMPILE_STACK_TOP
.fixup_alt_jump
3175 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3176 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3177 COMPILE_STACK_TOP
.regnum
= regnum
;
3179 /* Do not push a start_memory for groups beyond the last one
3180 we can represent in the compiled pattern. */
3181 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3182 BUF_PUSH_2 (start_memory
, regnum
);
3184 compile_stack
.avail
++;
3189 /* If we've reached MAX_REGNUM groups, then this open
3190 won't actually generate any code, so we'll have to
3191 clear pending_exact explicitly. */
3197 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3199 if (COMPILE_STACK_EMPTY
)
3201 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3202 goto normal_backslash
;
3204 FREE_STACK_RETURN (REG_ERPAREN
);
3210 /* See similar code for backslashed left paren above. */
3211 if (COMPILE_STACK_EMPTY
)
3213 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3216 FREE_STACK_RETURN (REG_ERPAREN
);
3219 /* Since we just checked for an empty stack above, this
3220 ``can't happen''. */
3221 assert (compile_stack
.avail
!= 0);
3223 /* We don't just want to restore into `regnum', because
3224 later groups should continue to be numbered higher,
3225 as in `(ab)c(de)' -- the second group is #2. */
3228 compile_stack
.avail
--;
3229 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3231 = COMPILE_STACK_TOP
.fixup_alt_jump
3232 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3234 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3235 regnum
= COMPILE_STACK_TOP
.regnum
;
3236 /* If we've reached MAX_REGNUM groups, then this open
3237 won't actually generate any code, so we'll have to
3238 clear pending_exact explicitly. */
3241 /* We're at the end of the group, so now we know how many
3242 groups were inside this one. */
3243 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3244 BUF_PUSH_2 (stop_memory
, regnum
);
3249 case '|': /* `\|'. */
3250 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3251 goto normal_backslash
;
3253 if (syntax
& RE_LIMITED_OPS
)
3256 /* Insert before the previous alternative a jump which
3257 jumps to this alternative if the former fails. */
3258 GET_BUFFER_SPACE (3);
3259 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3263 /* The alternative before this one has a jump after it
3264 which gets executed if it gets matched. Adjust that
3265 jump so it will jump to this alternative's analogous
3266 jump (put in below, which in turn will jump to the next
3267 (if any) alternative's such jump, etc.). The last such
3268 jump jumps to the correct final destination. A picture:
3274 If we are at `b', then fixup_alt_jump right now points to a
3275 three-byte space after `a'. We'll put in the jump, set
3276 fixup_alt_jump to right after `b', and leave behind three
3277 bytes which we'll fill in when we get to after `c'. */
3281 /* Mark and leave space for a jump after this alternative,
3282 to be filled in later either by next alternative or
3283 when know we're at the end of a series of alternatives. */
3285 GET_BUFFER_SPACE (3);
3294 /* If \{ is a literal. */
3295 if (!(syntax
& RE_INTERVALS
)
3296 /* If we're at `\{' and it's not the open-interval
3298 || (syntax
& RE_NO_BK_BRACES
))
3299 goto normal_backslash
;
3303 /* If got here, then the syntax allows intervals. */
3305 /* At least (most) this many matches must be made. */
3306 int lower_bound
= 0, upper_bound
= -1;
3310 GET_INTERVAL_COUNT (lower_bound
);
3313 GET_INTERVAL_COUNT (upper_bound
);
3315 /* Interval such as `{1}' => match exactly once. */
3316 upper_bound
= lower_bound
;
3319 || (0 <= upper_bound
&& upper_bound
< lower_bound
))
3320 FREE_STACK_RETURN (REG_BADBR
);
3322 if (!(syntax
& RE_NO_BK_BRACES
))
3325 FREE_STACK_RETURN (REG_BADBR
);
3327 FREE_STACK_RETURN (REG_EESCAPE
);
3332 FREE_STACK_RETURN (REG_BADBR
);
3334 /* We just parsed a valid interval. */
3336 /* If it's invalid to have no preceding re. */
3339 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3340 FREE_STACK_RETURN (REG_BADRPT
);
3341 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3344 goto unfetch_interval
;
3347 if (upper_bound
== 0)
3348 /* If the upper bound is zero, just drop the sub pattern
3351 else if (lower_bound
== 1 && upper_bound
== 1)
3352 /* Just match it once: nothing to do here. */
3355 /* Otherwise, we have a nontrivial interval. When
3356 we're all done, the pattern will look like:
3357 set_number_at <jump count> <upper bound>
3358 set_number_at <succeed_n count> <lower bound>
3359 succeed_n <after jump addr> <succeed_n count>
3361 jump_n <succeed_n addr> <jump count>
3362 (The upper bound and `jump_n' are omitted if
3363 `upper_bound' is 1, though.) */
3365 { /* If the upper bound is > 1, we need to insert
3366 more at the end of the loop. */
3367 unsigned int nbytes
= (upper_bound
< 0 ? 3
3368 : upper_bound
> 1 ? 5 : 0);
3369 unsigned int startoffset
= 0;
3371 GET_BUFFER_SPACE (20); /* We might use less. */
3373 if (lower_bound
== 0)
3375 /* A succeed_n that starts with 0 is really a
3376 a simple on_failure_jump_loop. */
3377 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3383 /* Initialize lower bound of the `succeed_n', even
3384 though it will be set during matching by its
3385 attendant `set_number_at' (inserted next),
3386 because `re_compile_fastmap' needs to know.
3387 Jump to the `jump_n' we might insert below. */
3388 INSERT_JUMP2 (succeed_n
, laststart
,
3393 /* Code to initialize the lower bound. Insert
3394 before the `succeed_n'. The `5' is the last two
3395 bytes of this `set_number_at', plus 3 bytes of
3396 the following `succeed_n'. */
3397 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3402 if (upper_bound
< 0)
3404 /* A negative upper bound stands for infinity,
3405 in which case it degenerates to a plain jump. */
3406 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3409 else if (upper_bound
> 1)
3410 { /* More than one repetition is allowed, so
3411 append a backward jump to the `succeed_n'
3412 that starts this interval.
3414 When we've reached this during matching,
3415 we'll have matched the interval once, so
3416 jump back only `upper_bound - 1' times. */
3417 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3421 /* The location we want to set is the second
3422 parameter of the `jump_n'; that is `b-2' as
3423 an absolute address. `laststart' will be
3424 the `set_number_at' we're about to insert;
3425 `laststart+3' the number to set, the source
3426 for the relative address. But we are
3427 inserting into the middle of the pattern --
3428 so everything is getting moved up by 5.
3429 Conclusion: (b - 2) - (laststart + 3) + 5,
3430 i.e., b - laststart.
3432 We insert this at the beginning of the loop
3433 so that if we fail during matching, we'll
3434 reinitialize the bounds. */
3435 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3436 upper_bound
- 1, b
);
3441 beg_interval
= NULL
;
3446 /* If an invalid interval, match the characters as literals. */
3447 assert (beg_interval
);
3449 beg_interval
= NULL
;
3451 /* normal_char and normal_backslash need `c'. */
3454 if (!(syntax
& RE_NO_BK_BRACES
))
3456 assert (p
> pattern
&& p
[-1] == '\\');
3457 goto normal_backslash
;
3463 /* There is no way to specify the before_dot and after_dot
3464 operators. rms says this is ok. --karl */
3473 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3479 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3485 BUF_PUSH_2 (categoryspec
, c
);
3491 BUF_PUSH_2 (notcategoryspec
, c
);
3497 if (syntax
& RE_NO_GNU_OPS
)
3500 BUF_PUSH_2 (syntaxspec
, Sword
);
3505 if (syntax
& RE_NO_GNU_OPS
)
3508 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3513 if (syntax
& RE_NO_GNU_OPS
)
3520 if (syntax
& RE_NO_GNU_OPS
)
3527 if (syntax
& RE_NO_GNU_OPS
)
3536 FREE_STACK_RETURN (REG_BADPAT
);
3540 if (syntax
& RE_NO_GNU_OPS
)
3542 BUF_PUSH (wordbound
);
3546 if (syntax
& RE_NO_GNU_OPS
)
3548 BUF_PUSH (notwordbound
);
3552 if (syntax
& RE_NO_GNU_OPS
)
3558 if (syntax
& RE_NO_GNU_OPS
)
3563 case '1': case '2': case '3': case '4': case '5':
3564 case '6': case '7': case '8': case '9':
3568 if (syntax
& RE_NO_BK_REFS
)
3569 goto normal_backslash
;
3573 if (reg
> bufp
->re_nsub
|| reg
< 1
3574 /* Can't back reference to a subexp before its end. */
3575 || group_in_compile_stack (compile_stack
, reg
))
3576 FREE_STACK_RETURN (REG_ESUBREG
);
3579 BUF_PUSH_2 (duplicate
, reg
);
3586 if (syntax
& RE_BK_PLUS_QM
)
3589 goto normal_backslash
;
3593 /* You might think it would be useful for \ to mean
3594 not to translate; but if we don't translate it
3595 it will never match anything. */
3602 /* Expects the character in `c'. */
3604 /* If no exactn currently being built. */
3607 /* If last exactn not at current position. */
3608 || pending_exact
+ *pending_exact
+ 1 != b
3610 /* We have only one byte following the exactn for the count. */
3611 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3613 /* If followed by a repetition operator. */
3614 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3615 || ((syntax
& RE_BK_PLUS_QM
)
3616 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3617 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3618 || ((syntax
& RE_INTERVALS
)
3619 && ((syntax
& RE_NO_BK_BRACES
)
3620 ? p
!= pend
&& *p
== '{'
3621 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3623 /* Start building a new exactn. */
3627 BUF_PUSH_2 (exactn
, 0);
3628 pending_exact
= b
- 1;
3631 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3638 len
= CHAR_STRING (c
, b
);
3643 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3644 if (! CHAR_BYTE8_P (c1
))
3646 re_wchar_t c2
= TRANSLATE (c1
);
3648 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3654 (*pending_exact
) += len
;
3659 } /* while p != pend */
3662 /* Through the pattern now. */
3666 if (!COMPILE_STACK_EMPTY
)
3667 FREE_STACK_RETURN (REG_EPAREN
);
3669 /* If we don't want backtracking, force success
3670 the first time we reach the end of the compiled pattern. */
3671 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3674 /* We have succeeded; set the length of the buffer. */
3675 bufp
->used
= b
- bufp
->buffer
;
3680 re_compile_fastmap (bufp
);
3681 DEBUG_PRINT ("\nCompiled pattern: \n");
3682 print_compiled_pattern (bufp
);
3687 #ifndef MATCH_MAY_ALLOCATE
3688 /* Initialize the failure stack to the largest possible stack. This
3689 isn't necessary unless we're trying to avoid calling alloca in
3690 the search and match routines. */
3692 int num_regs
= bufp
->re_nsub
+ 1;
3694 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3696 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3697 falk_stack
.stack
= realloc (fail_stack
.stack
,
3698 fail_stack
.size
* sizeof *falk_stack
.stack
);
3701 regex_grow_registers (num_regs
);
3703 #endif /* not MATCH_MAY_ALLOCATE */
3705 FREE_STACK_RETURN (REG_NOERROR
);
3706 } /* regex_compile */
3708 /* Subroutines for `regex_compile'. */
3710 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3713 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3715 *loc
= (unsigned char) op
;
3716 STORE_NUMBER (loc
+ 1, arg
);
3720 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3723 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3725 *loc
= (unsigned char) op
;
3726 STORE_NUMBER (loc
+ 1, arg1
);
3727 STORE_NUMBER (loc
+ 3, arg2
);
3731 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3732 for OP followed by two-byte integer parameter ARG. */
3735 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3737 register unsigned char *pfrom
= end
;
3738 register unsigned char *pto
= end
+ 3;
3740 while (pfrom
!= loc
)
3743 store_op1 (op
, loc
, arg
);
3747 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3750 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3752 register unsigned char *pfrom
= end
;
3753 register unsigned char *pto
= end
+ 5;
3755 while (pfrom
!= loc
)
3758 store_op2 (op
, loc
, arg1
, arg2
);
3762 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3763 after an alternative or a begin-subexpression. We assume there is at
3764 least one character before the ^. */
3767 at_begline_loc_p (const_re_char
*pattern
, const_re_char
*p
, reg_syntax_t syntax
)
3769 re_char
*prev
= p
- 2;
3770 boolean odd_backslashes
;
3772 /* After a subexpression? */
3774 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3776 /* After an alternative? */
3777 else if (*prev
== '|')
3778 odd_backslashes
= (syntax
& RE_NO_BK_VBAR
) == 0;
3780 /* After a shy subexpression? */
3781 else if (*prev
== ':' && (syntax
& RE_SHY_GROUPS
))
3783 /* Skip over optional regnum. */
3784 while (prev
- 1 >= pattern
&& prev
[-1] >= '0' && prev
[-1] <= '9')
3787 if (!(prev
- 2 >= pattern
3788 && prev
[-1] == '?' && prev
[-2] == '('))
3791 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3796 /* Count the number of preceding backslashes. */
3798 while (prev
- 1 >= pattern
&& prev
[-1] == '\\')
3800 return (p
- prev
) & odd_backslashes
;
3804 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3805 at least one character after the $, i.e., `P < PEND'. */
3808 at_endline_loc_p (const_re_char
*p
, const_re_char
*pend
, reg_syntax_t syntax
)
3811 boolean next_backslash
= *next
== '\\';
3812 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3815 /* Before a subexpression? */
3816 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3817 : next_backslash
&& next_next
&& *next_next
== ')')
3818 /* Before an alternative? */
3819 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3820 : next_backslash
&& next_next
&& *next_next
== '|');
3824 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3825 false if it's not. */
3828 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3830 ssize_t this_element
;
3832 for (this_element
= compile_stack
.avail
- 1;
3835 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3842 If fastmap is non-NULL, go through the pattern and fill fastmap
3843 with all the possible leading chars. If fastmap is NULL, don't
3844 bother filling it up (obviously) and only return whether the
3845 pattern could potentially match the empty string.
3847 Return 1 if p..pend might match the empty string.
3848 Return 0 if p..pend matches at least one char.
3849 Return -1 if fastmap was not updated accurately. */
3852 analyze_first (const_re_char
*p
, const_re_char
*pend
, char *fastmap
,
3853 const int multibyte
)
3858 /* If all elements for base leading-codes in fastmap is set, this
3859 flag is set true. */
3860 boolean match_any_multibyte_characters
= false;
3864 /* The loop below works as follows:
3865 - It has a working-list kept in the PATTERN_STACK and which basically
3866 starts by only containing a pointer to the first operation.
3867 - If the opcode we're looking at is a match against some set of
3868 chars, then we add those chars to the fastmap and go on to the
3869 next work element from the worklist (done via `break').
3870 - If the opcode is a control operator on the other hand, we either
3871 ignore it (if it's meaningless at this point, such as `start_memory')
3872 or execute it (if it's a jump). If the jump has several destinations
3873 (i.e. `on_failure_jump'), then we push the other destination onto the
3875 We guarantee termination by ignoring backward jumps (more or less),
3876 so that `p' is monotonically increasing. More to the point, we
3877 never set `p' (or push) anything `<= p1'. */
3881 /* `p1' is used as a marker of how far back a `on_failure_jump'
3882 can go without being ignored. It is normally equal to `p'
3883 (which prevents any backward `on_failure_jump') except right
3884 after a plain `jump', to allow patterns such as:
3887 10: on_failure_jump 3
3888 as used for the *? operator. */
3897 /* If the first character has to match a backreference, that means
3898 that the group was empty (since it already matched). Since this
3899 is the only case that interests us here, we can assume that the
3900 backreference must match the empty string. */
3905 /* Following are the cases which match a character. These end
3911 /* If multibyte is nonzero, the first byte of each
3912 character is an ASCII or a leading code. Otherwise,
3913 each byte is a character. Thus, this works in both
3918 /* For the case of matching this unibyte regex
3919 against multibyte, we must set a leading code of
3920 the corresponding multibyte character. */
3921 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
3923 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
3930 /* We could put all the chars except for \n (and maybe \0)
3931 but we don't bother since it is generally not worth it. */
3932 if (!fastmap
) break;
3937 if (!fastmap
) break;
3939 /* Chars beyond end of bitmap are possible matches. */
3940 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3941 j
< (1 << BYTEWIDTH
); j
++)
3947 if (!fastmap
) break;
3948 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3949 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3951 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3955 if (/* Any leading code can possibly start a character
3956 which doesn't match the specified set of characters. */
3959 /* If we can match a character class, we can match any
3960 multibyte characters. */
3961 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3962 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3965 if (match_any_multibyte_characters
== false)
3967 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3968 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3970 match_any_multibyte_characters
= true;
3974 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3975 && match_any_multibyte_characters
== false)
3977 /* Set fastmap[I] to 1 where I is a leading code of each
3978 multibyte character in the range table. */
3980 unsigned char lc1
, lc2
;
3982 /* Make P points the range table. `+ 2' is to skip flag
3983 bits for a character class. */
3984 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3986 /* Extract the number of ranges in range table into COUNT. */
3987 EXTRACT_NUMBER_AND_INCR (count
, p
);
3988 for (; count
> 0; count
--, p
+= 3)
3990 /* Extract the start and end of each range. */
3991 EXTRACT_CHARACTER (c
, p
);
3992 lc1
= CHAR_LEADING_CODE (c
);
3994 EXTRACT_CHARACTER (c
, p
);
3995 lc2
= CHAR_LEADING_CODE (c
);
3996 for (j
= lc1
; j
<= lc2
; j
++)
4005 if (!fastmap
) break;
4007 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4009 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4010 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4014 /* This match depends on text properties. These end with
4015 aborting optimizations. */
4019 case notcategoryspec
:
4020 if (!fastmap
) break;
4021 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4023 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
4024 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4027 /* Any leading code can possibly start a character which
4028 has or doesn't has the specified category. */
4029 if (match_any_multibyte_characters
== false)
4031 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4032 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4034 match_any_multibyte_characters
= true;
4038 /* All cases after this match the empty string. These end with
4060 EXTRACT_NUMBER_AND_INCR (j
, p
);
4062 /* Backward jumps can only go back to code that we've already
4063 visited. `re_compile' should make sure this is true. */
4068 case on_failure_jump
:
4069 case on_failure_keep_string_jump
:
4070 case on_failure_jump_loop
:
4071 case on_failure_jump_nastyloop
:
4072 case on_failure_jump_smart
:
4078 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4079 to jump back to "just after here". */
4082 case on_failure_jump
:
4083 case on_failure_keep_string_jump
:
4084 case on_failure_jump_nastyloop
:
4085 case on_failure_jump_loop
:
4086 case on_failure_jump_smart
:
4087 EXTRACT_NUMBER_AND_INCR (j
, p
);
4089 ; /* Backward jump to be ignored. */
4091 { /* We have to look down both arms.
4092 We first go down the "straight" path so as to minimize
4093 stack usage when going through alternatives. */
4094 int r
= analyze_first (p
, pend
, fastmap
, multibyte
);
4102 /* This code simply does not properly handle forward jump_n. */
4103 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4105 /* jump_n can either jump or fall through. The (backward) jump
4106 case has already been handled, so we only need to look at the
4107 fallthrough case. */
4111 /* If N == 0, it should be an on_failure_jump_loop instead. */
4112 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4114 /* We only care about one iteration of the loop, so we don't
4115 need to consider the case where this behaves like an
4132 abort (); /* We have listed all the cases. */
4135 /* Getting here means we have found the possible starting
4136 characters for one path of the pattern -- and that the empty
4137 string does not match. We need not follow this path further. */
4141 /* We reached the end without matching anything. */
4144 } /* analyze_first */
4146 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4147 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4148 characters can start a string that matches the pattern. This fastmap
4149 is used by re_search to skip quickly over impossible starting points.
4151 Character codes above (1 << BYTEWIDTH) are not represented in the
4152 fastmap, but the leading codes are represented. Thus, the fastmap
4153 indicates which character sets could start a match.
4155 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4156 area as BUFP->fastmap.
4158 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4161 Returns 0 if we succeed, -2 if an internal error. */
4164 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4166 char *fastmap
= bufp
->fastmap
;
4169 assert (fastmap
&& bufp
->buffer
);
4171 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4172 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4174 analysis
= analyze_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4175 fastmap
, RE_MULTIBYTE_P (bufp
));
4176 bufp
->can_be_null
= (analysis
!= 0);
4178 } /* re_compile_fastmap */
4180 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4181 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4182 this memory for recording register information. STARTS and ENDS
4183 must be allocated using the malloc library routine, and must each
4184 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4186 If NUM_REGS == 0, then subsequent matches should allocate their own
4189 Unless this function is called, the first search or match using
4190 PATTERN_BUFFER will allocate its own register data, without
4191 freeing the old data. */
4194 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4198 bufp
->regs_allocated
= REGS_REALLOCATE
;
4199 regs
->num_regs
= num_regs
;
4200 regs
->start
= starts
;
4205 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4207 regs
->start
= regs
->end
= 0;
4210 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4212 /* Searching routines. */
4214 /* Like re_search_2, below, but only one string is specified, and
4215 doesn't let you say where to stop matching. */
4218 re_search (struct re_pattern_buffer
*bufp
, const char *string
, size_t size
,
4219 ssize_t startpos
, ssize_t range
, struct re_registers
*regs
)
4221 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4224 WEAK_ALIAS (__re_search
, re_search
)
4226 /* Head address of virtual concatenation of string. */
4227 #define HEAD_ADDR_VSTRING(P) \
4228 (((P) >= size1 ? string2 : string1))
4230 /* Address of POS in the concatenation of virtual string. */
4231 #define POS_ADDR_VSTRING(POS) \
4232 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4234 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4235 virtual concatenation of STRING1 and STRING2, starting first at index
4236 STARTPOS, then at STARTPOS + 1, and so on.
4238 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4240 RANGE is how far to scan while trying to match. RANGE = 0 means try
4241 only at STARTPOS; in general, the last start tried is STARTPOS +
4244 In REGS, return the indices of the virtual concatenation of STRING1
4245 and STRING2 that matched the entire BUFP->buffer and its contained
4248 Do not consider matching one past the index STOP in the virtual
4249 concatenation of STRING1 and STRING2.
4251 We return either the position in the strings at which the match was
4252 found, -1 if no match, or -2 if error (such as failure
4256 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, size_t size1
,
4257 const char *str2
, size_t size2
, ssize_t startpos
, ssize_t range
,
4258 struct re_registers
*regs
, ssize_t stop
)
4261 re_char
*string1
= (re_char
*) str1
;
4262 re_char
*string2
= (re_char
*) str2
;
4263 register char *fastmap
= bufp
->fastmap
;
4264 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4265 size_t total_size
= size1
+ size2
;
4266 ssize_t endpos
= startpos
+ range
;
4267 boolean anchored_start
;
4268 /* Nonzero if we are searching multibyte string. */
4269 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4271 /* Check for out-of-range STARTPOS. */
4272 if (startpos
< 0 || startpos
> total_size
)
4275 /* Fix up RANGE if it might eventually take us outside
4276 the virtual concatenation of STRING1 and STRING2.
4277 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4279 range
= 0 - startpos
;
4280 else if (endpos
> total_size
)
4281 range
= total_size
- startpos
;
4283 /* If the search isn't to be a backwards one, don't waste time in a
4284 search for a pattern anchored at beginning of buffer. */
4285 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4294 /* In a forward search for something that starts with \=.
4295 don't keep searching past point. */
4296 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4298 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4304 /* Update the fastmap now if not correct already. */
4305 if (fastmap
&& !bufp
->fastmap_accurate
)
4306 re_compile_fastmap (bufp
);
4308 /* See whether the pattern is anchored. */
4309 anchored_start
= (bufp
->buffer
[0] == begline
);
4312 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4314 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4316 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4320 /* Loop through the string, looking for a place to start matching. */
4323 /* If the pattern is anchored,
4324 skip quickly past places we cannot match.
4325 We don't bother to treat startpos == 0 specially
4326 because that case doesn't repeat. */
4327 if (anchored_start
&& startpos
> 0)
4329 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4330 : string2
[startpos
- size1
- 1])
4335 /* If a fastmap is supplied, skip quickly over characters that
4336 cannot be the start of a match. If the pattern can match the
4337 null string, however, we don't need to skip characters; we want
4338 the first null string. */
4339 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4341 register re_char
*d
;
4342 register re_wchar_t buf_ch
;
4344 d
= POS_ADDR_VSTRING (startpos
);
4346 if (range
> 0) /* Searching forwards. */
4348 ssize_t irange
= range
, lim
= 0;
4350 if (startpos
< size1
&& startpos
+ range
>= size1
)
4351 lim
= range
- (size1
- startpos
);
4353 /* Written out as an if-else to avoid testing `translate'
4355 if (RE_TRANSLATE_P (translate
))
4362 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4363 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4364 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4367 range
-= buf_charlen
;
4373 register re_wchar_t ch
, translated
;
4376 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4377 translated
= RE_TRANSLATE (translate
, ch
);
4378 if (translated
!= ch
4379 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4381 if (fastmap
[buf_ch
])
4394 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4395 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4397 range
-= buf_charlen
;
4401 while (range
> lim
&& !fastmap
[*d
])
4407 startpos
+= irange
- range
;
4409 else /* Searching backwards. */
4413 buf_ch
= STRING_CHAR (d
);
4414 buf_ch
= TRANSLATE (buf_ch
);
4415 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4420 register re_wchar_t ch
, translated
;
4423 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4424 translated
= TRANSLATE (ch
);
4425 if (translated
!= ch
4426 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4428 if (! fastmap
[TRANSLATE (buf_ch
)])
4434 /* If can't match the null string, and that's all we have left, fail. */
4435 if (range
>= 0 && startpos
== total_size
&& fastmap
4436 && !bufp
->can_be_null
)
4439 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4440 startpos
, regs
, stop
);
4453 /* Update STARTPOS to the next character boundary. */
4456 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4457 int len
= BYTES_BY_CHAR_HEAD (*p
);
4475 /* Update STARTPOS to the previous character boundary. */
4478 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4480 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4482 /* Find the head of multibyte form. */
4483 PREV_CHAR_BOUNDARY (p
, phead
);
4484 range
+= p0
- 1 - p
;
4488 startpos
-= p0
- 1 - p
;
4494 WEAK_ALIAS (__re_search_2
, re_search_2
)
4496 /* Declarations and macros for re_match_2. */
4498 static int bcmp_translate (re_char
*s1
, re_char
*s2
,
4499 register ssize_t len
,
4500 RE_TRANSLATE_TYPE translate
,
4501 const int multibyte
);
4503 /* This converts PTR, a pointer into one of the search strings `string1'
4504 and `string2' into an offset from the beginning of that string. */
4505 #define POINTER_TO_OFFSET(ptr) \
4506 (FIRST_STRING_P (ptr) \
4508 : (ptr) - string2 + (ptrdiff_t) size1)
4510 /* Call before fetching a character with *d. This switches over to
4511 string2 if necessary.
4512 Check re_match_2_internal for a discussion of why end_match_2 might
4513 not be within string2 (but be equal to end_match_1 instead). */
4514 #define PREFETCH() \
4517 /* End of string2 => fail. */ \
4518 if (dend == end_match_2) \
4520 /* End of string1 => advance to string2. */ \
4522 dend = end_match_2; \
4525 /* Call before fetching a char with *d if you already checked other limits.
4526 This is meant for use in lookahead operations like wordend, etc..
4527 where we might need to look at parts of the string that might be
4528 outside of the LIMITs (i.e past `stop'). */
4529 #define PREFETCH_NOLIMIT() \
4533 dend = end_match_2; \
4536 /* Test if at very beginning or at very end of the virtual concatenation
4537 of `string1' and `string2'. If only one string, it's `string2'. */
4538 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4539 #define AT_STRINGS_END(d) ((d) == end2)
4541 /* Disabled due to a compiler bug -- see comment at case wordbound */
4543 /* The comment at case wordbound is following one, but we don't use
4544 AT_WORD_BOUNDARY anymore to support multibyte form.
4546 The DEC Alpha C compiler 3.x generates incorrect code for the
4547 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4548 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4549 macro and introducing temporary variables works around the bug. */
4552 /* Test if D points to a character which is word-constituent. We have
4553 two special cases to check for: if past the end of string1, look at
4554 the first character in string2; and if before the beginning of
4555 string2, look at the last character in string1. */
4556 #define WORDCHAR_P(d) \
4557 (SYNTAX ((d) == end1 ? *string2 \
4558 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4561 /* Test if the character before D and the one at D differ with respect
4562 to being word-constituent. */
4563 #define AT_WORD_BOUNDARY(d) \
4564 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4565 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4568 /* Free everything we malloc. */
4569 #ifdef MATCH_MAY_ALLOCATE
4570 # define FREE_VAR(var) \
4578 # define FREE_VARIABLES() \
4580 REGEX_FREE_STACK (fail_stack.stack); \
4581 FREE_VAR (regstart); \
4582 FREE_VAR (regend); \
4583 FREE_VAR (best_regstart); \
4584 FREE_VAR (best_regend); \
4585 REGEX_SAFE_FREE (); \
4588 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4589 #endif /* not MATCH_MAY_ALLOCATE */
4592 /* Optimization routines. */
4594 /* If the operation is a match against one or more chars,
4595 return a pointer to the next operation, else return NULL. */
4597 skip_one_char (const_re_char
*p
)
4610 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4613 p
= CHARSET_RANGE_TABLE (p
- 1);
4614 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4615 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4618 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4625 case notcategoryspec
:
4637 /* Jump over non-matching operations. */
4639 skip_noops (const_re_char
*p
, const_re_char
*pend
)
4653 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4664 /* Non-zero if "p1 matches something" implies "p2 fails". */
4666 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, const_re_char
*p1
,
4670 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4671 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4673 assert (p1
>= bufp
->buffer
&& p1
< pend
4674 && p2
>= bufp
->buffer
&& p2
<= pend
);
4676 /* Skip over open/close-group commands.
4677 If what follows this loop is a ...+ construct,
4678 look at what begins its body, since we will have to
4679 match at least one of that. */
4680 p2
= skip_noops (p2
, pend
);
4681 /* The same skip can be done for p1, except that this function
4682 is only used in the case where p1 is a simple match operator. */
4683 /* p1 = skip_noops (p1, pend); */
4685 assert (p1
>= bufp
->buffer
&& p1
< pend
4686 && p2
>= bufp
->buffer
&& p2
<= pend
);
4688 op2
= p2
== pend
? succeed
: *p2
;
4694 /* If we're at the end of the pattern, we can change. */
4695 if (skip_one_char (p1
))
4697 DEBUG_PRINT (" End of pattern: fast loop.\n");
4705 register re_wchar_t c
4706 = (re_opcode_t
) *p2
== endline
? '\n'
4707 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4709 if ((re_opcode_t
) *p1
== exactn
)
4711 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4713 DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4718 else if ((re_opcode_t
) *p1
== charset
4719 || (re_opcode_t
) *p1
== charset_not
)
4721 int not = (re_opcode_t
) *p1
== charset_not
;
4723 /* Test if C is listed in charset (or charset_not)
4725 if (! multibyte
|| IS_REAL_ASCII (c
))
4727 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4728 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4731 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4732 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4734 /* `not' is equal to 1 if c would match, which means
4735 that we can't change to pop_failure_jump. */
4738 DEBUG_PRINT (" No match => fast loop.\n");
4742 else if ((re_opcode_t
) *p1
== anychar
4745 DEBUG_PRINT (" . != \\n => fast loop.\n");
4753 if ((re_opcode_t
) *p1
== exactn
)
4754 /* Reuse the code above. */
4755 return mutually_exclusive_p (bufp
, p2
, p1
);
4757 /* It is hard to list up all the character in charset
4758 P2 if it includes multibyte character. Give up in
4760 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4762 /* Now, we are sure that P2 has no range table.
4763 So, for the size of bitmap in P2, `p2[1]' is
4764 enough. But P1 may have range table, so the
4765 size of bitmap table of P1 is extracted by
4766 using macro `CHARSET_BITMAP_SIZE'.
4768 In a multibyte case, we know that all the character
4769 listed in P2 is ASCII. In a unibyte case, P1 has only a
4770 bitmap table. So, in both cases, it is enough to test
4771 only the bitmap table of P1. */
4773 if ((re_opcode_t
) *p1
== charset
)
4776 /* We win if the charset inside the loop
4777 has no overlap with the one after the loop. */
4780 && idx
< CHARSET_BITMAP_SIZE (p1
));
4782 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4786 || idx
== CHARSET_BITMAP_SIZE (p1
))
4788 DEBUG_PRINT (" No match => fast loop.\n");
4792 else if ((re_opcode_t
) *p1
== charset_not
)
4795 /* We win if the charset_not inside the loop lists
4796 every character listed in the charset after. */
4797 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4798 if (! (p2
[2 + idx
] == 0
4799 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4800 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4805 DEBUG_PRINT (" No match => fast loop.\n");
4818 /* Reuse the code above. */
4819 return mutually_exclusive_p (bufp
, p2
, p1
);
4821 /* When we have two charset_not, it's very unlikely that
4822 they don't overlap. The union of the two sets of excluded
4823 chars should cover all possible chars, which, as a matter of
4824 fact, is virtually impossible in multibyte buffers. */
4830 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4832 return ((re_opcode_t
) *p1
== syntaxspec
4833 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4835 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4838 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4840 return ((re_opcode_t
) *p1
== notsyntaxspec
4841 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4843 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4846 return (((re_opcode_t
) *p1
== notsyntaxspec
4847 || (re_opcode_t
) *p1
== syntaxspec
)
4852 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4853 case notcategoryspec
:
4854 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4866 /* Matching routines. */
4868 #ifndef emacs /* Emacs never uses this. */
4869 /* re_match is like re_match_2 except it takes only a single string. */
4872 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
4873 size_t size
, ssize_t pos
, struct re_registers
*regs
)
4875 regoff_t result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
,
4876 size
, pos
, regs
, size
);
4879 WEAK_ALIAS (__re_match
, re_match
)
4880 #endif /* not emacs */
4883 /* In Emacs, this is the string or buffer in which we
4884 are matching. It is used for looking up syntax properties. */
4885 Lisp_Object re_match_object
;
4888 /* re_match_2 matches the compiled pattern in BUFP against the
4889 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4890 and SIZE2, respectively). We start matching at POS, and stop
4893 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4894 store offsets for the substring each group matched in REGS. See the
4895 documentation for exactly how many groups we fill.
4897 We return -1 if no match, -2 if an internal error (such as the
4898 failure stack overflowing). Otherwise, we return the length of the
4899 matched substring. */
4902 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
,
4903 size_t size1
, const char *string2
, size_t size2
, ssize_t pos
,
4904 struct re_registers
*regs
, ssize_t stop
)
4910 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4911 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4912 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4915 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4916 (re_char
*) string2
, size2
,
4920 WEAK_ALIAS (__re_match_2
, re_match_2
)
4923 /* This is a separate function so that we can force an alloca cleanup
4926 re_match_2_internal (struct re_pattern_buffer
*bufp
, const_re_char
*string1
,
4927 size_t size1
, const_re_char
*string2
, size_t size2
,
4928 ssize_t pos
, struct re_registers
*regs
, ssize_t stop
)
4930 /* General temporaries. */
4934 /* Just past the end of the corresponding string. */
4935 re_char
*end1
, *end2
;
4937 /* Pointers into string1 and string2, just past the last characters in
4938 each to consider matching. */
4939 re_char
*end_match_1
, *end_match_2
;
4941 /* Where we are in the data, and the end of the current string. */
4944 /* Used sometimes to remember where we were before starting matching
4945 an operator so that we can go back in case of failure. This "atomic"
4946 behavior of matching opcodes is indispensable to the correctness
4947 of the on_failure_keep_string_jump optimization. */
4950 /* Where we are in the pattern, and the end of the pattern. */
4951 re_char
*p
= bufp
->buffer
;
4952 re_char
*pend
= p
+ bufp
->used
;
4954 /* We use this to map every character in the string. */
4955 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4957 /* Nonzero if BUFP is setup from a multibyte regex. */
4958 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4960 /* Nonzero if STRING1/STRING2 are multibyte. */
4961 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4963 /* Failure point stack. Each place that can handle a failure further
4964 down the line pushes a failure point on this stack. It consists of
4965 regstart, and regend for all registers corresponding to
4966 the subexpressions we're currently inside, plus the number of such
4967 registers, and, finally, two char *'s. The first char * is where
4968 to resume scanning the pattern; the second one is where to resume
4969 scanning the strings. */
4970 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4971 fail_stack_type fail_stack
;
4973 #ifdef DEBUG_COMPILES_ARGUMENTS
4974 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4977 #if defined REL_ALLOC && defined REGEX_MALLOC
4978 /* This holds the pointer to the failure stack, when
4979 it is allocated relocatably. */
4980 fail_stack_elt_t
*failure_stack_ptr
;
4983 /* We fill all the registers internally, independent of what we
4984 return, for use in backreferences. The number here includes
4985 an element for register zero. */
4986 size_t num_regs
= bufp
->re_nsub
+ 1;
4988 /* Information on the contents of registers. These are pointers into
4989 the input strings; they record just what was matched (on this
4990 attempt) by a subexpression part of the pattern, that is, the
4991 regnum-th regstart pointer points to where in the pattern we began
4992 matching and the regnum-th regend points to right after where we
4993 stopped matching the regnum-th subexpression. (The zeroth register
4994 keeps track of what the whole pattern matches.) */
4995 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4996 re_char
**regstart
, **regend
;
4999 /* The following record the register info as found in the above
5000 variables when we find a match better than any we've seen before.
5001 This happens as we backtrack through the failure points, which in
5002 turn happens only if we have not yet matched the entire string. */
5003 unsigned best_regs_set
= false;
5004 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5005 re_char
**best_regstart
, **best_regend
;
5008 /* Logically, this is `best_regend[0]'. But we don't want to have to
5009 allocate space for that if we're not allocating space for anything
5010 else (see below). Also, we never need info about register 0 for
5011 any of the other register vectors, and it seems rather a kludge to
5012 treat `best_regend' differently than the rest. So we keep track of
5013 the end of the best match so far in a separate variable. We
5014 initialize this to NULL so that when we backtrack the first time
5015 and need to test it, it's not garbage. */
5016 re_char
*match_end
= NULL
;
5018 #ifdef DEBUG_COMPILES_ARGUMENTS
5019 /* Counts the total number of registers pushed. */
5020 unsigned num_regs_pushed
= 0;
5023 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5025 REGEX_USE_SAFE_ALLOCA
;
5029 #ifdef MATCH_MAY_ALLOCATE
5030 /* Do not bother to initialize all the register variables if there are
5031 no groups in the pattern, as it takes a fair amount of time. If
5032 there are groups, we include space for register 0 (the whole
5033 pattern), even though we never use it, since it simplifies the
5034 array indexing. We should fix this. */
5037 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5038 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5039 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5040 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5042 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5050 /* We must initialize all our variables to NULL, so that
5051 `FREE_VARIABLES' doesn't try to free them. */
5052 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5054 #endif /* MATCH_MAY_ALLOCATE */
5056 /* The starting position is bogus. */
5057 if (pos
< 0 || pos
> size1
+ size2
)
5063 /* Initialize subexpression text positions to -1 to mark ones that no
5064 start_memory/stop_memory has been seen for. Also initialize the
5065 register information struct. */
5066 for (reg
= 1; reg
< num_regs
; reg
++)
5067 regstart
[reg
] = regend
[reg
] = NULL
;
5069 /* We move `string1' into `string2' if the latter's empty -- but not if
5070 `string1' is null. */
5071 if (size2
== 0 && string1
!= NULL
)
5078 end1
= string1
+ size1
;
5079 end2
= string2
+ size2
;
5081 /* `p' scans through the pattern as `d' scans through the data.
5082 `dend' is the end of the input string that `d' points within. `d'
5083 is advanced into the following input string whenever necessary, but
5084 this happens before fetching; therefore, at the beginning of the
5085 loop, `d' can be pointing at the end of a string, but it cannot
5089 /* Only match within string2. */
5090 d
= string2
+ pos
- size1
;
5091 dend
= end_match_2
= string2
+ stop
- size1
;
5092 end_match_1
= end1
; /* Just to give it a value. */
5098 /* Only match within string1. */
5099 end_match_1
= string1
+ stop
;
5101 When we reach end_match_1, PREFETCH normally switches to string2.
5102 But in the present case, this means that just doing a PREFETCH
5103 makes us jump from `stop' to `gap' within the string.
5104 What we really want here is for the search to stop as
5105 soon as we hit end_match_1. That's why we set end_match_2
5106 to end_match_1 (since PREFETCH fails as soon as we hit
5108 end_match_2
= end_match_1
;
5111 { /* It's important to use this code when stop == size so that
5112 moving `d' from end1 to string2 will not prevent the d == dend
5113 check from catching the end of string. */
5115 end_match_2
= string2
+ stop
- size1
;
5121 DEBUG_PRINT ("The compiled pattern is: ");
5122 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5123 DEBUG_PRINT ("The string to match is: \"");
5124 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5125 DEBUG_PRINT ("\"\n");
5127 /* This loops over pattern commands. It exits by returning from the
5128 function if the match is complete, or it drops through if the match
5129 fails at this starting point in the input data. */
5132 DEBUG_PRINT ("\n%p: ", p
);
5136 /* End of pattern means we might have succeeded. */
5137 DEBUG_PRINT ("end of pattern ... ");
5139 /* If we haven't matched the entire string, and we want the
5140 longest match, try backtracking. */
5141 if (d
!= end_match_2
)
5143 /* True if this match is the best seen so far. */
5147 /* True if this match ends in the same string (string1
5148 or string2) as the best previous match. */
5149 bool same_str_p
= (FIRST_STRING_P (match_end
)
5150 == FIRST_STRING_P (d
));
5152 /* AIX compiler got confused when this was combined
5153 with the previous declaration. */
5155 best_match_p
= d
> match_end
;
5157 best_match_p
= !FIRST_STRING_P (d
);
5160 DEBUG_PRINT ("backtracking.\n");
5162 if (!FAIL_STACK_EMPTY ())
5163 { /* More failure points to try. */
5165 /* If exceeds best match so far, save it. */
5166 if (!best_regs_set
|| best_match_p
)
5168 best_regs_set
= true;
5171 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5173 for (reg
= 1; reg
< num_regs
; reg
++)
5175 best_regstart
[reg
] = regstart
[reg
];
5176 best_regend
[reg
] = regend
[reg
];
5182 /* If no failure points, don't restore garbage. And if
5183 last match is real best match, don't restore second
5185 else if (best_regs_set
&& !best_match_p
)
5188 /* Restore best match. It may happen that `dend ==
5189 end_match_1' while the restored d is in string2.
5190 For example, the pattern `x.*y.*z' against the
5191 strings `x-' and `y-z-', if the two strings are
5192 not consecutive in memory. */
5193 DEBUG_PRINT ("Restoring best registers.\n");
5196 dend
= ((d
>= string1
&& d
<= end1
)
5197 ? end_match_1
: end_match_2
);
5199 for (reg
= 1; reg
< num_regs
; reg
++)
5201 regstart
[reg
] = best_regstart
[reg
];
5202 regend
[reg
] = best_regend
[reg
];
5205 } /* d != end_match_2 */
5208 DEBUG_PRINT ("Accepting match.\n");
5210 /* If caller wants register contents data back, do it. */
5211 if (regs
&& !bufp
->no_sub
)
5213 /* Have the register data arrays been allocated? */
5214 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5215 { /* No. So allocate them with malloc. We need one
5216 extra element beyond `num_regs' for the `-1' marker
5218 regs
->num_regs
= max (RE_NREGS
, num_regs
+ 1);
5219 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5220 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5221 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5226 bufp
->regs_allocated
= REGS_REALLOCATE
;
5228 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5229 { /* Yes. If we need more elements than were already
5230 allocated, reallocate them. If we need fewer, just
5232 if (regs
->num_regs
< num_regs
+ 1)
5234 regs
->num_regs
= num_regs
+ 1;
5235 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5236 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5237 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5246 /* These braces fend off a "empty body in an else-statement"
5247 warning under GCC when assert expands to nothing. */
5248 assert (bufp
->regs_allocated
== REGS_FIXED
);
5251 /* Convert the pointer data in `regstart' and `regend' to
5252 indices. Register zero has to be set differently,
5253 since we haven't kept track of any info for it. */
5254 if (regs
->num_regs
> 0)
5256 regs
->start
[0] = pos
;
5257 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5260 /* Go through the first `min (num_regs, regs->num_regs)'
5261 registers, since that is all we initialized. */
5262 for (reg
= 1; reg
< min (num_regs
, regs
->num_regs
); reg
++)
5264 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5265 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5268 regs
->start
[reg
] = POINTER_TO_OFFSET (regstart
[reg
]);
5269 regs
->end
[reg
] = POINTER_TO_OFFSET (regend
[reg
]);
5273 /* If the regs structure we return has more elements than
5274 were in the pattern, set the extra elements to -1. If
5275 we (re)allocated the registers, this is the case,
5276 because we always allocate enough to have at least one
5278 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5279 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5280 } /* regs && !bufp->no_sub */
5282 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5283 nfailure_points_pushed
, nfailure_points_popped
,
5284 nfailure_points_pushed
- nfailure_points_popped
);
5285 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed
);
5287 ptrdiff_t dcnt
= POINTER_TO_OFFSET (d
) - pos
;
5289 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt
);
5295 /* Otherwise match next pattern command. */
5298 /* Ignore these. Used to ignore the n of succeed_n's which
5299 currently have n == 0. */
5301 DEBUG_PRINT ("EXECUTING no_op.\n");
5305 DEBUG_PRINT ("EXECUTING succeed.\n");
5308 /* Match the next n pattern characters exactly. The following
5309 byte in the pattern defines n, and the n bytes after that
5310 are the characters to match. */
5313 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt
);
5315 /* Remember the start point to rollback upon failure. */
5319 /* This is written out as an if-else so we don't waste time
5320 testing `translate' inside the loop. */
5321 if (RE_TRANSLATE_P (translate
))
5325 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5345 /* The cost of testing `translate' is comparatively small. */
5346 if (target_multibyte
)
5349 int pat_charlen
, buf_charlen
;
5354 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5357 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5360 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5362 if (TRANSLATE (buf_ch
) != pat_ch
)
5370 mcnt
-= pat_charlen
;
5382 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5383 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5390 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5391 if (! CHAR_BYTE8_P (buf_ch
))
5393 buf_ch
= TRANSLATE (buf_ch
);
5394 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5400 if (buf_ch
!= pat_ch
)
5413 /* Match any character except possibly a newline or a null. */
5419 DEBUG_PRINT ("EXECUTING anychar.\n");
5422 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5424 buf_ch
= TRANSLATE (buf_ch
);
5426 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5428 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5429 && buf_ch
== '\000'))
5432 DEBUG_PRINT (" Matched \"%d\".\n", *d
);
5441 register unsigned int c
, corig
;
5442 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5445 /* Start of actual range_table, or end of bitmap if there is no
5447 re_char
*range_table UNINIT
;
5449 /* Nonzero if there is a range table. */
5450 int range_table_exists
;
5452 /* Number of ranges of range table. This is not included
5453 in the initial byte-length of the command. */
5456 /* Whether matching against a unibyte character. */
5457 boolean unibyte_char
= false;
5459 DEBUG_PRINT ("EXECUTING charset%s.\n", not ? "_not" : "");
5461 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5463 if (range_table_exists
)
5465 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5466 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5470 corig
= c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5471 if (target_multibyte
)
5476 c1
= RE_CHAR_TO_UNIBYTE (c
);
5479 unibyte_char
= true;
5485 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5487 if (! CHAR_BYTE8_P (c1
))
5489 c1
= TRANSLATE (c1
);
5490 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5493 unibyte_char
= true;
5498 unibyte_char
= true;
5501 if (unibyte_char
&& c
< (1 << BYTEWIDTH
))
5502 { /* Lookup bitmap. */
5503 /* Cast to `unsigned' instead of `unsigned char' in
5504 case the bit list is a full 32 bytes long. */
5505 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5506 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5510 else if (range_table_exists
)
5512 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5514 if ( (class_bits
& BIT_LOWER
5517 && c
== upcase (corig
) && ISUPPER(c
))))
5518 | (class_bits
& BIT_MULTIBYTE
)
5519 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5520 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5521 | (class_bits
& BIT_UPPER
5524 && c
== downcase (corig
) && ISLOWER (c
))))
5525 | (class_bits
& BIT_WORD
&& ISWORD (c
))
5526 | (class_bits
& BIT_ALPHA
&& ISALPHA (c
))
5527 | (class_bits
& BIT_ALNUM
&& ISALNUM (c
))
5528 | (class_bits
& BIT_GRAPH
&& ISGRAPH (c
))
5529 | (class_bits
& BIT_PRINT
&& ISPRINT (c
)))
5532 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5536 if (range_table_exists
)
5537 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5539 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5541 if (!not) goto fail
;
5548 /* The beginning of a group is represented by start_memory.
5549 The argument is the register number. The text
5550 matched within the group is recorded (in the internal
5551 registers data structure) under the register number. */
5553 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p
);
5555 /* In case we need to undo this operation (via backtracking). */
5556 PUSH_FAILURE_REG (*p
);
5559 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5560 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart
[*p
]));
5562 /* Move past the register number and inner group count. */
5567 /* The stop_memory opcode represents the end of a group. Its
5568 argument is the same as start_memory's: the register number. */
5570 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p
);
5572 assert (!REG_UNSET (regstart
[*p
]));
5573 /* Strictly speaking, there should be code such as:
5575 assert (REG_UNSET (regend[*p]));
5576 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5578 But the only info to be pushed is regend[*p] and it is known to
5579 be UNSET, so there really isn't anything to push.
5580 Not pushing anything, on the other hand deprives us from the
5581 guarantee that regend[*p] is UNSET since undoing this operation
5582 will not reset its value properly. This is not important since
5583 the value will only be read on the next start_memory or at
5584 the very end and both events can only happen if this stop_memory
5588 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend
[*p
]));
5590 /* Move past the register number and the inner group count. */
5595 /* \<digit> has been turned into a `duplicate' command which is
5596 followed by the numeric value of <digit> as the register number. */
5599 register re_char
*d2
, *dend2
;
5600 int regno
= *p
++; /* Get which register to match against. */
5601 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno
);
5603 /* Can't back reference a group which we've never matched. */
5604 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5607 /* Where in input to try to start matching. */
5608 d2
= regstart
[regno
];
5610 /* Remember the start point to rollback upon failure. */
5613 /* Where to stop matching; if both the place to start and
5614 the place to stop matching are in the same string, then
5615 set to the place to stop, otherwise, for now have to use
5616 the end of the first string. */
5618 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5619 == FIRST_STRING_P (regend
[regno
]))
5620 ? regend
[regno
] : end_match_1
);
5625 /* If necessary, advance to next segment in register
5629 if (dend2
== end_match_2
) break;
5630 if (dend2
== regend
[regno
]) break;
5632 /* End of string1 => advance to string2. */
5634 dend2
= regend
[regno
];
5636 /* At end of register contents => success */
5637 if (d2
== dend2
) break;
5639 /* If necessary, advance to next segment in data. */
5642 /* How many characters left in this segment to match. */
5645 /* Want how many consecutive characters we can match in
5646 one shot, so, if necessary, adjust the count. */
5647 if (dcnt
> dend2
- d2
)
5650 /* Compare that many; failure if mismatch, else move
5652 if (RE_TRANSLATE_P (translate
)
5653 ? bcmp_translate (d
, d2
, dcnt
, translate
, target_multibyte
)
5654 : memcmp (d
, d2
, dcnt
))
5659 d
+= dcnt
, d2
+= dcnt
;
5665 /* begline matches the empty string at the beginning of the string
5666 (unless `not_bol' is set in `bufp'), and after newlines. */
5668 DEBUG_PRINT ("EXECUTING begline.\n");
5670 if (AT_STRINGS_BEG (d
))
5672 if (!bufp
->not_bol
) break;
5677 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5681 /* In all other cases, we fail. */
5685 /* endline is the dual of begline. */
5687 DEBUG_PRINT ("EXECUTING endline.\n");
5689 if (AT_STRINGS_END (d
))
5691 if (!bufp
->not_eol
) break;
5695 PREFETCH_NOLIMIT ();
5702 /* Match at the very beginning of the data. */
5704 DEBUG_PRINT ("EXECUTING begbuf.\n");
5705 if (AT_STRINGS_BEG (d
))
5710 /* Match at the very end of the data. */
5712 DEBUG_PRINT ("EXECUTING endbuf.\n");
5713 if (AT_STRINGS_END (d
))
5718 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5719 pushes NULL as the value for the string on the stack. Then
5720 `POP_FAILURE_POINT' will keep the current value for the
5721 string, instead of restoring it. To see why, consider
5722 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5723 then the . fails against the \n. But the next thing we want
5724 to do is match the \n against the \n; if we restored the
5725 string value, we would be back at the foo.
5727 Because this is used only in specific cases, we don't need to
5728 check all the things that `on_failure_jump' does, to make
5729 sure the right things get saved on the stack. Hence we don't
5730 share its code. The only reason to push anything on the
5731 stack at all is that otherwise we would have to change
5732 `anychar's code to do something besides goto fail in this
5733 case; that seems worse than this. */
5734 case on_failure_keep_string_jump
:
5735 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5736 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5739 PUSH_FAILURE_POINT (p
- 3, NULL
);
5742 /* A nasty loop is introduced by the non-greedy *? and +?.
5743 With such loops, the stack only ever contains one failure point
5744 at a time, so that a plain on_failure_jump_loop kind of
5745 cycle detection cannot work. Worse yet, such a detection
5746 can not only fail to detect a cycle, but it can also wrongly
5747 detect a cycle (between different instantiations of the same
5749 So the method used for those nasty loops is a little different:
5750 We use a special cycle-detection-stack-frame which is pushed
5751 when the on_failure_jump_nastyloop failure-point is *popped*.
5752 This special frame thus marks the beginning of one iteration
5753 through the loop and we can hence easily check right here
5754 whether something matched between the beginning and the end of
5756 case on_failure_jump_nastyloop
:
5757 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5758 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5761 assert ((re_opcode_t
)p
[-4] == no_op
);
5764 CHECK_INFINITE_LOOP (p
- 4, d
);
5766 /* If there's a cycle, just continue without pushing
5767 this failure point. The failure point is the "try again"
5768 option, which shouldn't be tried.
5769 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5770 PUSH_FAILURE_POINT (p
- 3, d
);
5774 /* Simple loop detecting on_failure_jump: just check on the
5775 failure stack if the same spot was already hit earlier. */
5776 case on_failure_jump_loop
:
5778 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5779 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5783 CHECK_INFINITE_LOOP (p
- 3, d
);
5785 /* If there's a cycle, get out of the loop, as if the matching
5786 had failed. We used to just `goto fail' here, but that was
5787 aborting the search a bit too early: we want to keep the
5788 empty-loop-match and keep matching after the loop.
5789 We want (x?)*y\1z to match both xxyz and xxyxz. */
5792 PUSH_FAILURE_POINT (p
- 3, d
);
5797 /* Uses of on_failure_jump:
5799 Each alternative starts with an on_failure_jump that points
5800 to the beginning of the next alternative. Each alternative
5801 except the last ends with a jump that in effect jumps past
5802 the rest of the alternatives. (They really jump to the
5803 ending jump of the following alternative, because tensioning
5804 these jumps is a hassle.)
5806 Repeats start with an on_failure_jump that points past both
5807 the repetition text and either the following jump or
5808 pop_failure_jump back to this on_failure_jump. */
5809 case on_failure_jump
:
5810 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5811 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5814 PUSH_FAILURE_POINT (p
-3, d
);
5817 /* This operation is used for greedy *.
5818 Compare the beginning of the repeat with what in the
5819 pattern follows its end. If we can establish that there
5820 is nothing that they would both match, i.e., that we
5821 would have to backtrack because of (as in, e.g., `a*a')
5822 then we can use a non-backtracking loop based on
5823 on_failure_keep_string_jump instead of on_failure_jump. */
5824 case on_failure_jump_smart
:
5825 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5826 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5829 re_char
*p1
= p
; /* Next operation. */
5830 /* Here, we discard `const', making re_match non-reentrant. */
5831 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5832 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5834 p
-= 3; /* Reset so that we will re-execute the
5835 instruction once it's been changed. */
5837 EXTRACT_NUMBER (mcnt
, p2
- 2);
5839 /* Ensure this is a indeed the trivial kind of loop
5840 we are expecting. */
5841 assert (skip_one_char (p1
) == p2
- 3);
5842 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5843 DEBUG_STATEMENT (debug
+= 2);
5844 if (mutually_exclusive_p (bufp
, p1
, p2
))
5846 /* Use a fast `on_failure_keep_string_jump' loop. */
5847 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5848 *p3
= (unsigned char) on_failure_keep_string_jump
;
5849 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5853 /* Default to a safe `on_failure_jump' loop. */
5854 DEBUG_PRINT (" smart default => slow loop.\n");
5855 *p3
= (unsigned char) on_failure_jump
;
5857 DEBUG_STATEMENT (debug
-= 2);
5861 /* Unconditionally jump (without popping any failure points). */
5864 IMMEDIATE_QUIT_CHECK
;
5865 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5866 DEBUG_PRINT ("EXECUTING jump %d ", mcnt
);
5867 p
+= mcnt
; /* Do the jump. */
5868 DEBUG_PRINT ("(to %p).\n", p
);
5872 /* Have to succeed matching what follows at least n times.
5873 After that, handle like `on_failure_jump'. */
5875 /* Signedness doesn't matter since we only compare MCNT to 0. */
5876 EXTRACT_NUMBER (mcnt
, p
+ 2);
5877 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt
);
5879 /* Originally, mcnt is how many times we HAVE to succeed. */
5882 /* Here, we discard `const', making re_match non-reentrant. */
5883 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5886 PUSH_NUMBER (p2
, mcnt
);
5889 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5894 /* Signedness doesn't matter since we only compare MCNT to 0. */
5895 EXTRACT_NUMBER (mcnt
, p
+ 2);
5896 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt
);
5898 /* Originally, this is how many times we CAN jump. */
5901 /* Here, we discard `const', making re_match non-reentrant. */
5902 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5904 PUSH_NUMBER (p2
, mcnt
);
5905 goto unconditional_jump
;
5907 /* If don't have to jump any more, skip over the rest of command. */
5914 unsigned char *p2
; /* Location of the counter. */
5915 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5917 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5918 /* Here, we discard `const', making re_match non-reentrant. */
5919 p2
= (unsigned char*) p
+ mcnt
;
5920 /* Signedness doesn't matter since we only copy MCNT's bits. */
5921 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5922 DEBUG_PRINT (" Setting %p to %d.\n", p2
, mcnt
);
5923 PUSH_NUMBER (p2
, mcnt
);
5930 boolean
not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5931 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5933 /* We SUCCEED (or FAIL) in one of the following cases: */
5935 /* Case 1: D is at the beginning or the end of string. */
5936 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5940 /* C1 is the character before D, S1 is the syntax of C1, C2
5941 is the character at D, and S2 is the syntax of C2. */
5946 ssize_t offset
= PTR_TO_OFFSET (d
- 1);
5947 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5948 UPDATE_SYNTAX_TABLE_FAST (charpos
);
5950 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5953 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos
+ 1);
5955 PREFETCH_NOLIMIT ();
5956 GET_CHAR_AFTER (c2
, d
, dummy
);
5959 if (/* Case 2: Only one of S1 and S2 is Sword. */
5960 ((s1
== Sword
) != (s2
== Sword
))
5961 /* Case 3: Both of S1 and S2 are Sword, and macro
5962 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5963 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5973 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5975 /* We FAIL in one of the following cases: */
5977 /* Case 1: D is at the end of string. */
5978 if (AT_STRINGS_END (d
))
5982 /* C1 is the character before D, S1 is the syntax of C1, C2
5983 is the character at D, and S2 is the syntax of C2. */
5988 ssize_t offset
= PTR_TO_OFFSET (d
);
5989 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5990 UPDATE_SYNTAX_TABLE_FAST (charpos
);
5993 GET_CHAR_AFTER (c2
, d
, dummy
);
5996 /* Case 2: S2 is not Sword. */
6000 /* Case 3: D is not at the beginning of string ... */
6001 if (!AT_STRINGS_BEG (d
))
6003 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6005 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6009 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6011 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6018 DEBUG_PRINT ("EXECUTING wordend.\n");
6020 /* We FAIL in one of the following cases: */
6022 /* Case 1: D is at the beginning of string. */
6023 if (AT_STRINGS_BEG (d
))
6027 /* C1 is the character before D, S1 is the syntax of C1, C2
6028 is the character at D, and S2 is the syntax of C2. */
6033 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6034 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6035 UPDATE_SYNTAX_TABLE_FAST (charpos
);
6037 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6040 /* Case 2: S1 is not Sword. */
6044 /* Case 3: D is not at the end of string ... */
6045 if (!AT_STRINGS_END (d
))
6047 PREFETCH_NOLIMIT ();
6048 GET_CHAR_AFTER (c2
, d
, dummy
);
6050 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos
);
6054 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6056 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6063 DEBUG_PRINT ("EXECUTING symbeg.\n");
6065 /* We FAIL in one of the following cases: */
6067 /* Case 1: D is at the end of string. */
6068 if (AT_STRINGS_END (d
))
6072 /* C1 is the character before D, S1 is the syntax of C1, C2
6073 is the character at D, and S2 is the syntax of C2. */
6077 ssize_t offset
= PTR_TO_OFFSET (d
);
6078 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6079 UPDATE_SYNTAX_TABLE_FAST (charpos
);
6082 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6085 /* Case 2: S2 is neither Sword nor Ssymbol. */
6086 if (s2
!= Sword
&& s2
!= Ssymbol
)
6089 /* Case 3: D is not at the beginning of string ... */
6090 if (!AT_STRINGS_BEG (d
))
6092 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6094 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6098 /* ... and S1 is Sword or Ssymbol. */
6099 if (s1
== Sword
|| s1
== Ssymbol
)
6106 DEBUG_PRINT ("EXECUTING symend.\n");
6108 /* We FAIL in one of the following cases: */
6110 /* Case 1: D is at the beginning of string. */
6111 if (AT_STRINGS_BEG (d
))
6115 /* C1 is the character before D, S1 is the syntax of C1, C2
6116 is the character at D, and S2 is the syntax of C2. */
6120 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6121 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6122 UPDATE_SYNTAX_TABLE_FAST (charpos
);
6124 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6127 /* Case 2: S1 is neither Ssymbol nor Sword. */
6128 if (s1
!= Sword
&& s1
!= Ssymbol
)
6131 /* Case 3: D is not at the end of string ... */
6132 if (!AT_STRINGS_END (d
))
6134 PREFETCH_NOLIMIT ();
6135 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6137 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos
+ 1);
6141 /* ... and S2 is Sword or Ssymbol. */
6142 if (s2
== Sword
|| s2
== Ssymbol
)
6151 boolean
not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6153 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6158 ssize_t offset
= PTR_TO_OFFSET (d
);
6159 ssize_t pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6160 UPDATE_SYNTAX_TABLE_FAST (pos1
);
6167 GET_CHAR_AFTER (c
, d
, len
);
6168 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6177 DEBUG_PRINT ("EXECUTING before_dot.\n");
6178 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6183 DEBUG_PRINT ("EXECUTING at_dot.\n");
6184 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6189 DEBUG_PRINT ("EXECUTING after_dot.\n");
6190 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6195 case notcategoryspec
:
6197 boolean
not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6199 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6200 not ? "not" : "", mcnt
);
6206 GET_CHAR_AFTER (c
, d
, len
);
6207 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6219 continue; /* Successfully executed one pattern command; keep going. */
6222 /* We goto here if a matching operation fails. */
6224 IMMEDIATE_QUIT_CHECK
;
6225 if (!FAIL_STACK_EMPTY ())
6228 /* A restart point is known. Restore to that state. */
6229 DEBUG_PRINT ("\nFAIL:\n");
6230 POP_FAILURE_POINT (str
, pat
);
6233 case on_failure_keep_string_jump
:
6234 assert (str
== NULL
);
6235 goto continue_failure_jump
;
6237 case on_failure_jump_nastyloop
:
6238 assert ((re_opcode_t
)pat
[-2] == no_op
);
6239 PUSH_FAILURE_POINT (pat
- 2, str
);
6242 case on_failure_jump_loop
:
6243 case on_failure_jump
:
6246 continue_failure_jump
:
6247 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6252 /* A special frame used for nastyloops. */
6259 assert (p
>= bufp
->buffer
&& p
<= pend
);
6261 if (d
>= string1
&& d
<= end1
)
6265 break; /* Matching at this starting point really fails. */
6269 goto restore_best_regs
;
6273 return -1; /* Failure to match. */
6276 /* Subroutine definitions for re_match_2. */
6278 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6279 bytes; nonzero otherwise. */
6282 bcmp_translate (const_re_char
*s1
, const_re_char
*s2
, register ssize_t len
,
6283 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6285 register re_char
*p1
= s1
, *p2
= s2
;
6286 re_char
*p1_end
= s1
+ len
;
6287 re_char
*p2_end
= s2
+ len
;
6289 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6290 different lengths, but relying on a single `len' would break this. -sm */
6291 while (p1
< p1_end
&& p2
< p2_end
)
6293 int p1_charlen
, p2_charlen
;
6294 re_wchar_t p1_ch
, p2_ch
;
6296 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6297 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6299 if (RE_TRANSLATE (translate
, p1_ch
)
6300 != RE_TRANSLATE (translate
, p2_ch
))
6303 p1
+= p1_charlen
, p2
+= p2_charlen
;
6306 if (p1
!= p1_end
|| p2
!= p2_end
)
6312 /* Entry points for GNU code. */
6314 /* re_compile_pattern is the GNU regular expression compiler: it
6315 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6316 Returns 0 if the pattern was valid, otherwise an error string.
6318 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6319 are set in BUFP on entry.
6321 We call regex_compile to do the actual compilation. */
6324 re_compile_pattern (const char *pattern
, size_t length
,
6325 struct re_pattern_buffer
*bufp
)
6329 /* GNU code is written to assume at least RE_NREGS registers will be set
6330 (and at least one extra will be -1). */
6331 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6333 /* And GNU code determines whether or not to get register information
6334 by passing null for the REGS argument to re_match, etc., not by
6338 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6342 return gettext (re_error_msgid
[(int) ret
]);
6344 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6346 /* Entry points compatible with 4.2 BSD regex library. We don't define
6347 them unless specifically requested. */
6349 #if defined _REGEX_RE_COMP || defined _LIBC
6351 /* BSD has one and only one pattern buffer. */
6352 static struct re_pattern_buffer re_comp_buf
;
6356 /* Make these definitions weak in libc, so POSIX programs can redefine
6357 these names if they don't use our functions, and still use
6358 regcomp/regexec below without link errors. */
6361 re_comp (const char *s
)
6367 if (!re_comp_buf
.buffer
)
6368 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6369 return (char *) gettext ("No previous regular expression");
6373 if (!re_comp_buf
.buffer
)
6375 re_comp_buf
.buffer
= malloc (200);
6376 if (re_comp_buf
.buffer
== NULL
)
6377 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6378 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6379 re_comp_buf
.allocated
= 200;
6381 re_comp_buf
.fastmap
= malloc (1 << BYTEWIDTH
);
6382 if (re_comp_buf
.fastmap
== NULL
)
6383 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6384 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6387 /* Since `re_exec' always passes NULL for the `regs' argument, we
6388 don't need to initialize the pattern buffer fields which affect it. */
6390 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6395 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6396 return (char *) gettext (re_error_msgid
[(int) ret
]);
6404 re_exec (const char *s
)
6406 const size_t len
= strlen (s
);
6407 return re_search (&re_comp_buf
, s
, len
, 0, len
, 0) >= 0;
6409 #endif /* _REGEX_RE_COMP */
6411 /* POSIX.2 functions. Don't define these for Emacs. */
6415 /* regcomp takes a regular expression as a string and compiles it.
6417 PREG is a regex_t *. We do not expect any fields to be initialized,
6418 since POSIX says we shouldn't. Thus, we set
6420 `buffer' to the compiled pattern;
6421 `used' to the length of the compiled pattern;
6422 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6423 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6424 RE_SYNTAX_POSIX_BASIC;
6425 `fastmap' to an allocated space for the fastmap;
6426 `fastmap_accurate' to zero;
6427 `re_nsub' to the number of subexpressions in PATTERN.
6429 PATTERN is the address of the pattern string.
6431 CFLAGS is a series of bits which affect compilation.
6433 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6434 use POSIX basic syntax.
6436 If REG_NEWLINE is set, then . and [^...] don't match newline.
6437 Also, regexec will try a match beginning after every newline.
6439 If REG_ICASE is set, then we considers upper- and lowercase
6440 versions of letters to be equivalent when matching.
6442 If REG_NOSUB is set, then when PREG is passed to regexec, that
6443 routine will report only success or failure, and nothing about the
6446 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6447 the return codes and their meanings.) */
6450 regcomp (regex_t
*_Restrict_ preg
, const char *_Restrict_ pattern
,
6455 = (cflags
& REG_EXTENDED
) ?
6456 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6458 /* regex_compile will allocate the space for the compiled pattern. */
6460 preg
->allocated
= 0;
6463 /* Try to allocate space for the fastmap. */
6464 preg
->fastmap
= malloc (1 << BYTEWIDTH
);
6466 if (cflags
& REG_ICASE
)
6470 preg
->translate
= malloc (CHAR_SET_SIZE
* sizeof *preg
->translate
);
6471 if (preg
->translate
== NULL
)
6472 return (int) REG_ESPACE
;
6474 /* Map uppercase characters to corresponding lowercase ones. */
6475 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6476 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6479 preg
->translate
= NULL
;
6481 /* If REG_NEWLINE is set, newlines are treated differently. */
6482 if (cflags
& REG_NEWLINE
)
6483 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6484 syntax
&= ~RE_DOT_NEWLINE
;
6485 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6488 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6490 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6492 /* POSIX says a null character in the pattern terminates it, so we
6493 can use strlen here in compiling the pattern. */
6494 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6496 /* POSIX doesn't distinguish between an unmatched open-group and an
6497 unmatched close-group: both are REG_EPAREN. */
6498 if (ret
== REG_ERPAREN
)
6501 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6502 { /* Compute the fastmap now, since regexec cannot modify the pattern
6504 re_compile_fastmap (preg
);
6505 if (preg
->can_be_null
)
6506 { /* The fastmap can't be used anyway. */
6507 free (preg
->fastmap
);
6508 preg
->fastmap
= NULL
;
6513 WEAK_ALIAS (__regcomp
, regcomp
)
6516 /* regexec searches for a given pattern, specified by PREG, in the
6519 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6520 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6521 least NMATCH elements, and we set them to the offsets of the
6522 corresponding matched substrings.
6524 EFLAGS specifies `execution flags' which affect matching: if
6525 REG_NOTBOL is set, then ^ does not match at the beginning of the
6526 string; if REG_NOTEOL is set, then $ does not match at the end.
6528 We return 0 if we find a match and REG_NOMATCH if not. */
6531 regexec (const regex_t
*_Restrict_ preg
, const char *_Restrict_ string
,
6532 size_t nmatch
, regmatch_t pmatch
[_Restrict_arr_
], int eflags
)
6535 struct re_registers regs
;
6536 regex_t private_preg
;
6537 size_t len
= strlen (string
);
6538 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6540 private_preg
= *preg
;
6542 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6543 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6545 /* The user has told us exactly how many registers to return
6546 information about, via `nmatch'. We have to pass that on to the
6547 matching routines. */
6548 private_preg
.regs_allocated
= REGS_FIXED
;
6552 regs
.num_regs
= nmatch
;
6553 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6554 if (regs
.start
== NULL
)
6556 regs
.end
= regs
.start
+ nmatch
;
6559 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6560 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6561 was a little bit longer but still only matching the real part.
6562 This works because the `endline' will check for a '\n' and will find a
6563 '\0', correctly deciding that this is not the end of a line.
6564 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6565 a convenient '\0' there. For all we know, the string could be preceded
6566 by '\n' which would throw things off. */
6568 /* Perform the searching operation. */
6569 ret
= re_search (&private_preg
, string
, len
,
6570 /* start: */ 0, /* range: */ len
,
6571 want_reg_info
? ®s
: 0);
6573 /* Copy the register information to the POSIX structure. */
6580 for (r
= 0; r
< nmatch
; r
++)
6582 pmatch
[r
].rm_so
= regs
.start
[r
];
6583 pmatch
[r
].rm_eo
= regs
.end
[r
];
6587 /* If we needed the temporary register info, free the space now. */
6591 /* We want zero return to mean success, unlike `re_search'. */
6592 return ret
>= 0 ? REG_NOERROR
: REG_NOMATCH
;
6594 WEAK_ALIAS (__regexec
, regexec
)
6597 /* Returns a message corresponding to an error code, ERR_CODE, returned
6598 from either regcomp or regexec. We don't use PREG here.
6600 ERR_CODE was previously called ERRCODE, but that name causes an
6601 error with msvc8 compiler. */
6604 regerror (int err_code
, const regex_t
*preg
, char *errbuf
, size_t errbuf_size
)
6610 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6611 /* Only error codes returned by the rest of the code should be passed
6612 to this routine. If we are given anything else, or if other regex
6613 code generates an invalid error code, then the program has a bug.
6614 Dump core so we can fix it. */
6617 msg
= gettext (re_error_msgid
[err_code
]);
6619 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6621 if (errbuf_size
!= 0)
6623 if (msg_size
> errbuf_size
)
6625 memcpy (errbuf
, msg
, errbuf_size
- 1);
6626 errbuf
[errbuf_size
- 1] = 0;
6629 strcpy (errbuf
, msg
);
6634 WEAK_ALIAS (__regerror
, regerror
)
6637 /* Free dynamically allocated space used by PREG. */
6640 regfree (regex_t
*preg
)
6642 free (preg
->buffer
);
6643 preg
->buffer
= NULL
;
6645 preg
->allocated
= 0;
6648 free (preg
->fastmap
);
6649 preg
->fastmap
= NULL
;
6650 preg
->fastmap_accurate
= 0;
6652 free (preg
->translate
);
6653 preg
->translate
= NULL
;
6655 WEAK_ALIAS (__regfree
, regfree
)
6657 #endif /* not emacs */