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"
56 /* We need this for `regex.h', and perhaps for the Emacs include files. */
57 # include <sys/types.h>
60 /* Whether to use ISO C Amendment 1 wide char functions.
61 Those should not be used for Emacs since it uses its own. */
63 #define WIDE_CHAR_SUPPORT 1
65 #define WIDE_CHAR_SUPPORT \
66 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
69 /* For platform which support the ISO C amendment 1 functionality we
70 support user defined character classes. */
72 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
78 /* We have to keep the namespace clean. */
79 # define regfree(preg) __regfree (preg)
80 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
81 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
82 # define regerror(err_code, preg, errbuf, errbuf_size) \
83 __regerror (err_code, preg, errbuf, errbuf_size)
84 # define re_set_registers(bu, re, nu, st, en) \
85 __re_set_registers (bu, re, nu, st, en)
86 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
87 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
88 # define re_match(bufp, string, size, pos, regs) \
89 __re_match (bufp, string, size, pos, regs)
90 # define re_search(bufp, string, size, startpos, range, regs) \
91 __re_search (bufp, string, size, startpos, range, regs)
92 # define re_compile_pattern(pattern, length, bufp) \
93 __re_compile_pattern (pattern, length, bufp)
94 # define re_set_syntax(syntax) __re_set_syntax (syntax)
95 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
96 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
97 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
99 /* Make sure we call libc's function even if the user overrides them. */
100 # define btowc __btowc
101 # define iswctype __iswctype
102 # define wctype __wctype
104 # define WEAK_ALIAS(a,b) weak_alias (a, b)
106 /* We are also using some library internals. */
107 # include <locale/localeinfo.h>
108 # include <locale/elem-hash.h>
109 # include <langinfo.h>
111 # define WEAK_ALIAS(a,b)
114 /* This is for other GNU distributions with internationalized messages. */
115 #if HAVE_LIBINTL_H || defined _LIBC
116 # include <libintl.h>
118 # define gettext(msgid) (msgid)
122 /* This define is so xgettext can find the internationalizable
124 # define gettext_noop(String) String
127 /* The `emacs' switch turns on certain matching commands
128 that make sense only in Emacs. */
132 # include "character.h"
136 # include "category.h"
138 /* Make syntax table lookup grant data in gl_state. */
139 # define SYNTAX(c) syntax_property (c, 1)
144 # define malloc xmalloc
148 # define realloc xrealloc
154 /* Converts the pointer to the char to BEG-based offset from the start. */
155 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
156 /* Strings are 0-indexed, buffers are 1-indexed; we pun on the boolean
157 result to get the right base index. */
158 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
160 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
161 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
162 # define RE_STRING_CHAR(p, multibyte) \
163 (multibyte ? (STRING_CHAR (p)) : (*(p)))
164 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
165 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
167 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
169 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
171 /* Set C a (possibly converted to multibyte) character before P. P
172 points into a string which is the virtual concatenation of STR1
173 (which ends at END1) or STR2 (which ends at END2). */
174 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
176 if (target_multibyte) \
178 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
179 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
180 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
181 c = STRING_CHAR (dtemp); \
185 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
186 (c) = RE_CHAR_TO_MULTIBYTE (c); \
190 /* Set C a (possibly converted to multibyte) character at P, and set
191 LEN to the byte length of that character. */
192 # define GET_CHAR_AFTER(c, p, len) \
194 if (target_multibyte) \
195 (c) = STRING_CHAR_AND_LENGTH (p, len); \
200 (c) = RE_CHAR_TO_MULTIBYTE (c); \
204 #else /* not emacs */
206 /* If we are not linking with Emacs proper,
207 we can't use the relocating allocator
208 even if config.h says that we can. */
213 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
216 xmalloc (size_t size
)
218 void *val
= malloc (size
);
221 write (STDERR_FILENO
, "virtual memory exhausted\n", 25);
228 xrealloc (void *block
, size_t size
)
231 /* We must call malloc explicitly when BLOCK is 0, since some
232 reallocs don't do this. */
236 val
= realloc (block
, size
);
239 write (STDERR_FILENO
, "virtual memory exhausted\n", 25);
248 # define malloc xmalloc
252 # define realloc xrealloc
254 # include <stdbool.h>
257 /* Define the syntax stuff for \<, \>, etc. */
259 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
260 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
262 /* Dummy macros for non-Emacs environments. */
263 # define MAX_MULTIBYTE_LENGTH 1
264 # define RE_MULTIBYTE_P(x) 0
265 # define RE_TARGET_MULTIBYTE_P(x) 0
266 # define WORD_BOUNDARY_P(c1, c2) (0)
267 # define BYTES_BY_CHAR_HEAD(p) (1)
268 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
269 # define STRING_CHAR(p) (*(p))
270 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
271 # define CHAR_STRING(c, s) (*(s) = (c), 1)
272 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
273 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
274 # define RE_CHAR_TO_MULTIBYTE(c) (c)
275 # define RE_CHAR_TO_UNIBYTE(c) (c)
276 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
277 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
278 # define GET_CHAR_AFTER(c, p, len) \
280 # define CHAR_BYTE8_P(c) (0)
281 # define CHAR_LEADING_CODE(c) (c)
283 #endif /* not emacs */
286 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
287 # define RE_TRANSLATE_P(TBL) (TBL)
290 /* Get the interface, including the syntax bits. */
293 /* isalpha etc. are used for the character classes. */
298 /* 1 if C is an ASCII character. */
299 # define IS_REAL_ASCII(c) ((c) < 0200)
301 /* 1 if C is a unibyte character. */
302 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
304 /* The Emacs definitions should not be directly affected by locales. */
306 /* In Emacs, these are only used for single-byte characters. */
307 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
308 # define ISCNTRL(c) ((c) < ' ')
309 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
310 || ((c) >= 'a' && (c) <= 'f') \
311 || ((c) >= 'A' && (c) <= 'F'))
313 /* This is only used for single-byte characters. */
314 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
316 /* The rest must handle multibyte characters. */
318 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
319 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0240) \
322 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
323 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
326 # define ISALNUM(c) (IS_REAL_ASCII (c) \
327 ? (((c) >= 'a' && (c) <= 'z') \
328 || ((c) >= 'A' && (c) <= 'Z') \
329 || ((c) >= '0' && (c) <= '9')) \
332 # define ISALPHA(c) (IS_REAL_ASCII (c) \
333 ? (((c) >= 'a' && (c) <= 'z') \
334 || ((c) >= 'A' && (c) <= 'Z')) \
337 # define ISLOWER(c) lowercasep (c)
339 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
340 ? ((c) > ' ' && (c) < 0177 \
341 && !(((c) >= 'a' && (c) <= 'z') \
342 || ((c) >= 'A' && (c) <= 'Z') \
343 || ((c) >= '0' && (c) <= '9'))) \
344 : SYNTAX (c) != Sword)
346 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
348 # define ISUPPER(c) uppercasep (c)
350 # define ISWORD(c) (SYNTAX (c) == Sword)
352 #else /* not emacs */
354 /* 1 if C is an ASCII character. */
355 # define IS_REAL_ASCII(c) ((c) < 0200)
357 /* This distinction is not meaningful, except in Emacs. */
358 # define ISUNIBYTE(c) 1
361 # define ISBLANK(c) isblank (c)
363 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
366 # define ISGRAPH(c) isgraph (c)
368 # define ISGRAPH(c) (isprint (c) && !isspace (c))
371 /* Solaris defines ISPRINT so we must undefine it first. */
373 # define ISPRINT(c) isprint (c)
374 # define ISDIGIT(c) isdigit (c)
375 # define ISALNUM(c) isalnum (c)
376 # define ISALPHA(c) isalpha (c)
377 # define ISCNTRL(c) iscntrl (c)
378 # define ISLOWER(c) islower (c)
379 # define ISPUNCT(c) ispunct (c)
380 # define ISSPACE(c) isspace (c)
381 # define ISUPPER(c) isupper (c)
382 # define ISXDIGIT(c) isxdigit (c)
384 # define ISWORD(c) ISALPHA (c)
387 # define TOLOWER(c) _tolower (c)
389 # define TOLOWER(c) tolower (c)
392 /* How many characters in the character set. */
393 # define CHAR_SET_SIZE 256
397 extern char *re_syntax_table
;
399 # else /* not SYNTAX_TABLE */
401 static char re_syntax_table
[CHAR_SET_SIZE
];
404 init_syntax_once (void)
412 memset (re_syntax_table
, 0, sizeof re_syntax_table
);
414 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
416 re_syntax_table
[c
] = Sword
;
418 re_syntax_table
['_'] = Ssymbol
;
423 # endif /* not SYNTAX_TABLE */
425 # define SYNTAX(c) re_syntax_table[(c)]
427 #endif /* not emacs */
429 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
431 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
432 use `alloca' instead of `malloc'. This is because using malloc in
433 re_search* or re_match* could cause memory leaks when C-g is used in
434 Emacs; also, malloc is slower and causes storage fragmentation. On
435 the other hand, malloc is more portable, and easier to debug.
437 Because we sometimes use alloca, some routines have to be macros,
438 not functions -- `alloca'-allocated space disappears at the end of the
439 function it is called in. */
443 # define REGEX_ALLOCATE malloc
444 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
445 # define REGEX_FREE free
447 #else /* not REGEX_MALLOC */
450 # define REGEX_USE_SAFE_ALLOCA USE_SAFE_ALLOCA
451 # define REGEX_SAFE_FREE() SAFE_FREE ()
452 # define REGEX_ALLOCATE SAFE_ALLOCA
455 # define REGEX_ALLOCATE alloca
458 /* Assumes a `char *destination' variable. */
459 # define REGEX_REALLOCATE(source, osize, nsize) \
460 (destination = REGEX_ALLOCATE (nsize), \
461 memcpy (destination, source, osize))
463 /* No need to do anything to free, after alloca. */
464 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
466 #endif /* not REGEX_MALLOC */
468 #ifndef REGEX_USE_SAFE_ALLOCA
469 # define REGEX_USE_SAFE_ALLOCA ((void) 0)
470 # define REGEX_SAFE_FREE() ((void) 0)
473 /* Define how to allocate the failure stack. */
475 #if defined REL_ALLOC && defined REGEX_MALLOC
477 # define REGEX_ALLOCATE_STACK(size) \
478 r_alloc (&failure_stack_ptr, (size))
479 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
480 r_re_alloc (&failure_stack_ptr, (nsize))
481 # define REGEX_FREE_STACK(ptr) \
482 r_alloc_free (&failure_stack_ptr)
484 #else /* not using relocating allocator */
486 # define REGEX_ALLOCATE_STACK(size) REGEX_ALLOCATE (size)
487 # define REGEX_REALLOCATE_STACK(source, o, n) REGEX_REALLOCATE (source, o, n)
488 # define REGEX_FREE_STACK(ptr) REGEX_FREE (ptr)
490 #endif /* not using relocating allocator */
493 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
494 `string1' or just past its end. This works if PTR is NULL, which is
496 #define FIRST_STRING_P(ptr) \
497 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
499 /* (Re)Allocate N items of type T using malloc, or fail. */
500 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
501 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
502 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
504 #define BYTEWIDTH 8 /* In bits. */
509 # define max(a, b) ((a) > (b) ? (a) : (b))
510 # define min(a, b) ((a) < (b) ? (a) : (b))
513 /* Type of source-pattern and string chars. */
515 typedef unsigned char re_char
;
516 typedef const re_char const_re_char
;
518 typedef const unsigned char re_char
;
519 typedef re_char const_re_char
;
522 typedef char boolean
;
524 static regoff_t
re_match_2_internal (struct re_pattern_buffer
*bufp
,
525 re_char
*string1
, size_t size1
,
526 re_char
*string2
, size_t size2
,
528 struct re_registers
*regs
,
531 /* These are the command codes that appear in compiled regular
532 expressions. Some opcodes are followed by argument bytes. A
533 command code can specify any interpretation whatsoever for its
534 arguments. Zero bytes may appear in the compiled regular expression. */
540 /* Succeed right away--no more backtracking. */
543 /* Followed by one byte giving n, then by n literal bytes. */
546 /* Matches any (more or less) character. */
549 /* Matches any one char belonging to specified set. First
550 following byte is number of bitmap bytes. Then come bytes
551 for a bitmap saying which chars are in. Bits in each byte
552 are ordered low-bit-first. A character is in the set if its
553 bit is 1. A character too large to have a bit in the map is
554 automatically not in the set.
556 If the length byte has the 0x80 bit set, then that stuff
557 is followed by a range table:
558 2 bytes of flags for character sets (low 8 bits, high 8 bits)
559 See RANGE_TABLE_WORK_BITS below.
560 2 bytes, the number of pairs that follow (upto 32767)
561 pairs, each 2 multibyte characters,
562 each multibyte character represented as 3 bytes. */
565 /* Same parameters as charset, but match any character that is
566 not one of those specified. */
569 /* Start remembering the text that is matched, for storing in a
570 register. Followed by one byte with the register number, in
571 the range 0 to one less than the pattern buffer's re_nsub
575 /* Stop remembering the text that is matched and store it in a
576 memory register. Followed by one byte with the register
577 number, in the range 0 to one less than `re_nsub' in the
581 /* Match a duplicate of something remembered. Followed by one
582 byte containing the register number. */
585 /* Fail unless at beginning of line. */
588 /* Fail unless at end of line. */
591 /* Succeeds if at beginning of buffer (if emacs) or at beginning
592 of string to be matched (if not). */
595 /* Analogously, for end of buffer/string. */
598 /* Followed by two byte relative address to which to jump. */
601 /* Followed by two-byte relative address of place to resume at
602 in case of failure. */
605 /* Like on_failure_jump, but pushes a placeholder instead of the
606 current string position when executed. */
607 on_failure_keep_string_jump
,
609 /* Just like `on_failure_jump', except that it checks that we
610 don't get stuck in an infinite loop (matching an empty string
612 on_failure_jump_loop
,
614 /* Just like `on_failure_jump_loop', except that it checks for
615 a different kind of loop (the kind that shows up with non-greedy
616 operators). This operation has to be immediately preceded
618 on_failure_jump_nastyloop
,
620 /* A smart `on_failure_jump' used for greedy * and + operators.
621 It analyzes the loop before which it is put and if the
622 loop does not require backtracking, it changes itself to
623 `on_failure_keep_string_jump' and short-circuits the loop,
624 else it just defaults to changing itself into `on_failure_jump'.
625 It assumes that it is pointing to just past a `jump'. */
626 on_failure_jump_smart
,
628 /* Followed by two-byte relative address and two-byte number n.
629 After matching N times, jump to the address upon failure.
630 Does not work if N starts at 0: use on_failure_jump_loop
634 /* Followed by two-byte relative address, and two-byte number n.
635 Jump to the address N times, then fail. */
638 /* Set the following two-byte relative address to the
639 subsequent two-byte number. The address *includes* the two
643 wordbeg
, /* Succeeds if at word beginning. */
644 wordend
, /* Succeeds if at word end. */
646 wordbound
, /* Succeeds if at a word boundary. */
647 notwordbound
, /* Succeeds if not at a word boundary. */
649 symbeg
, /* Succeeds if at symbol beginning. */
650 symend
, /* Succeeds if at symbol end. */
652 /* Matches any character whose syntax is specified. Followed by
653 a byte which contains a syntax code, e.g., Sword. */
656 /* Matches any character whose syntax is not that specified. */
660 , at_dot
, /* Succeeds if at point. */
662 /* Matches any character whose category-set contains the specified
663 category. The operator is followed by a byte which contains a
664 category code (mnemonic ASCII character). */
667 /* Matches any character whose category-set does not contain the
668 specified category. The operator is followed by a byte which
669 contains the category code (mnemonic ASCII character). */
674 /* Common operations on the compiled pattern. */
676 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
678 #define STORE_NUMBER(destination, number) \
680 (destination)[0] = (number) & 0377; \
681 (destination)[1] = (number) >> 8; \
684 /* Same as STORE_NUMBER, except increment DESTINATION to
685 the byte after where the number is stored. Therefore, DESTINATION
686 must be an lvalue. */
688 #define STORE_NUMBER_AND_INCR(destination, number) \
690 STORE_NUMBER (destination, number); \
691 (destination) += 2; \
694 /* Put into DESTINATION a number stored in two contiguous bytes starting
697 #define EXTRACT_NUMBER(destination, source) \
698 ((destination) = extract_number (source))
701 extract_number (re_char
*source
)
703 unsigned leading_byte
= SIGN_EXTEND_CHAR (source
[1]);
704 return (leading_byte
<< 8) + source
[0];
707 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
708 SOURCE must be an lvalue. */
710 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
711 ((destination) = extract_number_and_incr (&source))
714 extract_number_and_incr (re_char
**source
)
716 int num
= extract_number (*source
);
721 /* Store a multibyte character in three contiguous bytes starting
722 DESTINATION, and increment DESTINATION to the byte after where the
723 character is stored. Therefore, DESTINATION must be an lvalue. */
725 #define STORE_CHARACTER_AND_INCR(destination, character) \
727 (destination)[0] = (character) & 0377; \
728 (destination)[1] = ((character) >> 8) & 0377; \
729 (destination)[2] = (character) >> 16; \
730 (destination) += 3; \
733 /* Put into DESTINATION a character stored in three contiguous bytes
734 starting at SOURCE. */
736 #define EXTRACT_CHARACTER(destination, source) \
738 (destination) = ((source)[0] \
739 | ((source)[1] << 8) \
740 | ((source)[2] << 16)); \
744 /* Macros for charset. */
746 /* Size of bitmap of charset P in bytes. P is a start of charset,
747 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
748 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
750 /* Nonzero if charset P has range table. */
751 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
753 /* Return the address of range table of charset P. But not the start
754 of table itself, but the before where the number of ranges is
755 stored. `2 +' means to skip re_opcode_t and size of bitmap,
756 and the 2 bytes of flags at the start of the range table. */
757 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
760 /* Extract the bit flags that start a range table. */
761 #define CHARSET_RANGE_TABLE_BITS(p) \
762 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
763 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
766 /* Return the address of end of RANGE_TABLE. COUNT is number of
767 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
768 is start of range and end of range. `* 3' is size of each start
770 #define CHARSET_RANGE_TABLE_END(range_table, count) \
771 ((range_table) + (count) * 2 * 3)
773 /* If DEBUG is defined, Regex prints many voluminous messages about what
774 it is doing (if the variable `debug' is nonzero). If linked with the
775 main program in `iregex.c', you can enter patterns and strings
776 interactively. And if linked with the main program in `main.c' and
777 the other test files, you can run the already-written tests. */
781 /* We use standard I/O for debugging. */
784 /* It is useful to test things that ``must'' be true when debugging. */
787 static int debug
= -100000;
789 # define DEBUG_STATEMENT(e) e
790 # define DEBUG_PRINT(...) if (debug > 0) printf (__VA_ARGS__)
791 # define DEBUG_COMPILES_ARGUMENTS
792 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
793 if (debug > 0) print_partial_compiled_pattern (s, e)
794 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
795 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
798 /* Print the fastmap in human-readable form. */
801 print_fastmap (char *fastmap
)
803 unsigned was_a_range
= 0;
806 while (i
< (1 << BYTEWIDTH
))
812 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
828 /* Print a compiled pattern string in human-readable form, starting at
829 the START pointer into it and ending just before the pointer END. */
832 print_partial_compiled_pattern (re_char
*start
, re_char
*end
)
840 fprintf (stderr
, "(null)\n");
844 /* Loop over pattern commands. */
847 fprintf (stderr
, "%td:\t", p
- start
);
849 switch ((re_opcode_t
) *p
++)
852 fprintf (stderr
, "/no_op");
856 fprintf (stderr
, "/succeed");
861 fprintf (stderr
, "/exactn/%d", mcnt
);
864 fprintf (stderr
, "/%c", *p
++);
870 fprintf (stderr
, "/start_memory/%d", *p
++);
874 fprintf (stderr
, "/stop_memory/%d", *p
++);
878 fprintf (stderr
, "/duplicate/%d", *p
++);
882 fprintf (stderr
, "/anychar");
888 register int c
, last
= -100;
889 register int in_range
= 0;
890 int length
= CHARSET_BITMAP_SIZE (p
- 1);
891 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
893 fprintf (stderr
, "/charset [%s",
894 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
897 fprintf (stderr
, " !extends past end of pattern! ");
899 for (c
= 0; c
< 256; c
++)
901 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
903 /* Are we starting a range? */
904 if (last
+ 1 == c
&& ! in_range
)
906 fprintf (stderr
, "-");
909 /* Have we broken a range? */
910 else if (last
+ 1 != c
&& in_range
)
912 fprintf (stderr
, "%c", last
);
917 fprintf (stderr
, "%c", c
);
923 fprintf (stderr
, "%c", last
);
925 fprintf (stderr
, "]");
932 fprintf (stderr
, "has-range-table");
934 /* ??? Should print the range table; for now, just skip it. */
935 p
+= 2; /* skip range table bits */
936 EXTRACT_NUMBER_AND_INCR (count
, p
);
937 p
= CHARSET_RANGE_TABLE_END (p
, count
);
943 fprintf (stderr
, "/begline");
947 fprintf (stderr
, "/endline");
950 case on_failure_jump
:
951 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
952 fprintf (stderr
, "/on_failure_jump to %td", p
+ mcnt
- start
);
955 case on_failure_keep_string_jump
:
956 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
957 fprintf (stderr
, "/on_failure_keep_string_jump to %td",
961 case on_failure_jump_nastyloop
:
962 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
963 fprintf (stderr
, "/on_failure_jump_nastyloop to %td",
967 case on_failure_jump_loop
:
968 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
969 fprintf (stderr
, "/on_failure_jump_loop to %td",
973 case on_failure_jump_smart
:
974 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
975 fprintf (stderr
, "/on_failure_jump_smart to %td",
980 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
981 fprintf (stderr
, "/jump to %td", p
+ mcnt
- start
);
985 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
986 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
987 fprintf (stderr
, "/succeed_n to %td, %d times",
988 p
- 2 + mcnt
- start
, mcnt2
);
992 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
993 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
994 fprintf (stderr
, "/jump_n to %td, %d times",
995 p
- 2 + mcnt
- start
, mcnt2
);
999 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1000 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1001 fprintf (stderr
, "/set_number_at location %td to %d",
1002 p
- 2 + mcnt
- start
, mcnt2
);
1006 fprintf (stderr
, "/wordbound");
1010 fprintf (stderr
, "/notwordbound");
1014 fprintf (stderr
, "/wordbeg");
1018 fprintf (stderr
, "/wordend");
1022 fprintf (stderr
, "/symbeg");
1026 fprintf (stderr
, "/symend");
1030 fprintf (stderr
, "/syntaxspec");
1032 fprintf (stderr
, "/%d", mcnt
);
1036 fprintf (stderr
, "/notsyntaxspec");
1038 fprintf (stderr
, "/%d", mcnt
);
1043 fprintf (stderr
, "/at_dot");
1047 fprintf (stderr
, "/categoryspec");
1049 fprintf (stderr
, "/%d", mcnt
);
1052 case notcategoryspec
:
1053 fprintf (stderr
, "/notcategoryspec");
1055 fprintf (stderr
, "/%d", mcnt
);
1060 fprintf (stderr
, "/begbuf");
1064 fprintf (stderr
, "/endbuf");
1068 fprintf (stderr
, "?%d", *(p
-1));
1071 fprintf (stderr
, "\n");
1074 fprintf (stderr
, "%td:\tend of pattern.\n", p
- start
);
1079 print_compiled_pattern (struct re_pattern_buffer
*bufp
)
1081 re_char
*buffer
= bufp
->buffer
;
1083 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1084 printf ("%ld bytes used/%ld bytes allocated.\n",
1085 bufp
->used
, bufp
->allocated
);
1087 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1089 printf ("fastmap: ");
1090 print_fastmap (bufp
->fastmap
);
1093 printf ("re_nsub: %zu\t", bufp
->re_nsub
);
1094 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1095 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1096 printf ("no_sub: %d\t", bufp
->no_sub
);
1097 printf ("not_bol: %d\t", bufp
->not_bol
);
1098 printf ("not_eol: %d\t", bufp
->not_eol
);
1100 printf ("syntax: %lx\n", bufp
->syntax
);
1103 /* Perhaps we should print the translate table? */
1108 print_double_string (re_char
*where
, re_char
*string1
, ssize_t size1
,
1109 re_char
*string2
, ssize_t size2
)
1117 if (FIRST_STRING_P (where
))
1119 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1120 putchar (string1
[this_char
]);
1125 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1126 putchar (string2
[this_char
]);
1130 #else /* not DEBUG */
1135 # define DEBUG_STATEMENT(e)
1136 # define DEBUG_PRINT(...)
1137 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1138 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1140 #endif /* not DEBUG */
1143 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1144 also be assigned to arbitrarily: each pattern buffer stores its own
1145 syntax, so it can be changed between regex compilations. */
1146 /* This has no initializer because initialized variables in Emacs
1147 become read-only after dumping. */
1148 reg_syntax_t re_syntax_options
;
1151 /* Specify the precise syntax of regexps for compilation. This provides
1152 for compatibility for various utilities which historically have
1153 different, incompatible syntaxes.
1155 The argument SYNTAX is a bit mask comprised of the various bits
1156 defined in regex.h. We return the old syntax. */
1159 re_set_syntax (reg_syntax_t syntax
)
1161 reg_syntax_t ret
= re_syntax_options
;
1163 re_syntax_options
= syntax
;
1166 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1170 /* This table gives an error message for each of the error codes listed
1171 in regex.h. Obviously the order here has to be same as there.
1172 POSIX doesn't require that we do anything for REG_NOERROR,
1173 but why not be nice? */
1175 static const char *re_error_msgid
[] =
1177 gettext_noop ("Success"), /* REG_NOERROR */
1178 gettext_noop ("No match"), /* REG_NOMATCH */
1179 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1180 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1181 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1182 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1183 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1184 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1185 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1186 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1187 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1188 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1189 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1190 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1191 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1192 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1193 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1194 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1197 /* Avoiding alloca during matching, to placate r_alloc. */
1199 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1200 searching and matching functions should not call alloca. On some
1201 systems, alloca is implemented in terms of malloc, and if we're
1202 using the relocating allocator routines, then malloc could cause a
1203 relocation, which might (if the strings being searched are in the
1204 ralloc heap) shift the data out from underneath the regexp
1207 Here's another reason to avoid allocation: Emacs
1208 processes input from X in a signal handler; processing X input may
1209 call malloc; if input arrives while a matching routine is calling
1210 malloc, then we're scrod. But Emacs can't just block input while
1211 calling matching routines; then we don't notice interrupts when
1212 they come in. So, Emacs blocks input around all regexp calls
1213 except the matching calls, which it leaves unprotected, in the
1214 faith that they will not malloc. */
1216 /* Normally, this is fine. */
1217 #define MATCH_MAY_ALLOCATE
1219 /* The match routines may not allocate if (1) they would do it with malloc
1220 and (2) it's not safe for them to use malloc.
1221 Note that if REL_ALLOC is defined, matching would not use malloc for the
1222 failure stack, but we would still use it for the register vectors;
1223 so REL_ALLOC should not affect this. */
1224 #if defined REGEX_MALLOC && defined emacs
1225 # undef MATCH_MAY_ALLOCATE
1229 /* Failure stack declarations and macros; both re_compile_fastmap and
1230 re_match_2 use a failure stack. These have to be macros because of
1231 REGEX_ALLOCATE_STACK. */
1234 /* Approximate number of failure points for which to initially allocate space
1235 when matching. If this number is exceeded, we allocate more
1236 space, so it is not a hard limit. */
1237 #ifndef INIT_FAILURE_ALLOC
1238 # define INIT_FAILURE_ALLOC 20
1241 /* Roughly the maximum number of failure points on the stack. Would be
1242 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1243 This is a variable only so users of regex can assign to it; we never
1244 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1245 before using it, so it should probably be a byte-count instead. */
1246 # if defined MATCH_MAY_ALLOCATE
1247 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1248 whose default stack limit is 2mb. In order for a larger
1249 value to work reliably, you have to try to make it accord
1250 with the process stack limit. */
1251 size_t re_max_failures
= 40000;
1253 size_t re_max_failures
= 4000;
1256 union fail_stack_elt
1259 /* This should be the biggest `int' that's no bigger than a pointer. */
1263 typedef union fail_stack_elt fail_stack_elt_t
;
1267 fail_stack_elt_t
*stack
;
1269 size_t avail
; /* Offset of next open position. */
1270 size_t frame
; /* Offset of the cur constructed frame. */
1273 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1276 /* Define macros to initialize and free the failure stack.
1277 Do `return -2' if the alloc fails. */
1279 #ifdef MATCH_MAY_ALLOCATE
1280 # define INIT_FAIL_STACK() \
1282 fail_stack.stack = \
1283 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1284 * sizeof (fail_stack_elt_t)); \
1286 if (fail_stack.stack == NULL) \
1289 fail_stack.size = INIT_FAILURE_ALLOC; \
1290 fail_stack.avail = 0; \
1291 fail_stack.frame = 0; \
1294 # define INIT_FAIL_STACK() \
1296 fail_stack.avail = 0; \
1297 fail_stack.frame = 0; \
1300 # define RETALLOC_IF(addr, n, t) \
1301 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1305 /* Double the size of FAIL_STACK, up to a limit
1306 which allows approximately `re_max_failures' items.
1308 Return 1 if succeeds, and 0 if either ran out of memory
1309 allocating space for it or it was already too large.
1311 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1313 /* Factor to increase the failure stack size by
1314 when we increase it.
1315 This used to be 2, but 2 was too wasteful
1316 because the old discarded stacks added up to as much space
1317 were as ultimate, maximum-size stack. */
1318 #define FAIL_STACK_GROWTH_FACTOR 4
1320 #define GROW_FAIL_STACK(fail_stack) \
1321 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1322 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1324 : ((fail_stack).stack \
1325 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1326 (fail_stack).size * sizeof (fail_stack_elt_t), \
1327 min (re_max_failures * TYPICAL_FAILURE_SIZE, \
1328 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1329 * FAIL_STACK_GROWTH_FACTOR))), \
1331 (fail_stack).stack == NULL \
1333 : ((fail_stack).size \
1334 = (min (re_max_failures * TYPICAL_FAILURE_SIZE, \
1335 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1336 * FAIL_STACK_GROWTH_FACTOR)) \
1337 / sizeof (fail_stack_elt_t)), \
1341 /* Push a pointer value onto the failure stack.
1342 Assumes the variable `fail_stack'. Probably should only
1343 be called from within `PUSH_FAILURE_POINT'. */
1344 #define PUSH_FAILURE_POINTER(item) \
1345 fail_stack.stack[fail_stack.avail++].pointer = (item)
1347 /* This pushes an integer-valued item onto the failure stack.
1348 Assumes the variable `fail_stack'. Probably should only
1349 be called from within `PUSH_FAILURE_POINT'. */
1350 #define PUSH_FAILURE_INT(item) \
1351 fail_stack.stack[fail_stack.avail++].integer = (item)
1353 /* These POP... operations complement the PUSH... operations.
1354 All assume that `fail_stack' is nonempty. */
1355 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1356 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1358 /* Individual items aside from the registers. */
1359 #define NUM_NONREG_ITEMS 3
1361 /* Used to examine the stack (to detect infinite loops). */
1362 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1363 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1364 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1365 #define TOP_FAILURE_HANDLE() fail_stack.frame
1368 #define ENSURE_FAIL_STACK(space) \
1369 while (REMAINING_AVAIL_SLOTS <= space) { \
1370 if (!GROW_FAIL_STACK (fail_stack)) \
1372 DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\
1373 DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\
1376 /* Push register NUM onto the stack. */
1377 #define PUSH_FAILURE_REG(num) \
1379 char *destination; \
1381 ENSURE_FAIL_STACK(3); \
1382 DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \
1383 n, regstart[n], regend[n]); \
1384 PUSH_FAILURE_POINTER (regstart[n]); \
1385 PUSH_FAILURE_POINTER (regend[n]); \
1386 PUSH_FAILURE_INT (n); \
1389 /* Change the counter's value to VAL, but make sure that it will
1390 be reset when backtracking. */
1391 #define PUSH_NUMBER(ptr,val) \
1393 char *destination; \
1395 ENSURE_FAIL_STACK(3); \
1396 EXTRACT_NUMBER (c, ptr); \
1397 DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \
1398 PUSH_FAILURE_INT (c); \
1399 PUSH_FAILURE_POINTER (ptr); \
1400 PUSH_FAILURE_INT (-1); \
1401 STORE_NUMBER (ptr, val); \
1404 /* Pop a saved register off the stack. */
1405 #define POP_FAILURE_REG_OR_COUNT() \
1407 long pfreg = POP_FAILURE_INT (); \
1410 /* It's a counter. */ \
1411 /* Here, we discard `const', making re_match non-reentrant. */ \
1412 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1413 pfreg = POP_FAILURE_INT (); \
1414 STORE_NUMBER (ptr, pfreg); \
1415 DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \
1419 regend[pfreg] = POP_FAILURE_POINTER (); \
1420 regstart[pfreg] = POP_FAILURE_POINTER (); \
1421 DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \
1422 pfreg, regstart[pfreg], regend[pfreg]); \
1426 /* Check that we are not stuck in an infinite loop. */
1427 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1429 ssize_t failure = TOP_FAILURE_HANDLE (); \
1430 /* Check for infinite matching loops */ \
1431 while (failure > 0 \
1432 && (FAILURE_STR (failure) == string_place \
1433 || FAILURE_STR (failure) == NULL)) \
1435 assert (FAILURE_PAT (failure) >= bufp->buffer \
1436 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1437 if (FAILURE_PAT (failure) == pat_cur) \
1442 DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1443 failure = NEXT_FAILURE_HANDLE(failure); \
1445 DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \
1448 /* Push the information about the state we will need
1449 if we ever fail back to it.
1451 Requires variables fail_stack, regstart, regend and
1452 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1455 Does `return FAILURE_CODE' if runs out of memory. */
1457 #define PUSH_FAILURE_POINT(pattern, string_place) \
1459 char *destination; \
1460 /* Must be int, so when we don't save any registers, the arithmetic \
1461 of 0 + -1 isn't done as unsigned. */ \
1463 DEBUG_STATEMENT (nfailure_points_pushed++); \
1464 DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \
1465 DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \
1466 DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\
1468 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1470 DEBUG_PRINT ("\n"); \
1472 DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \
1473 PUSH_FAILURE_INT (fail_stack.frame); \
1475 DEBUG_PRINT (" Push string %p: \"", string_place); \
1476 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1477 DEBUG_PRINT ("\"\n"); \
1478 PUSH_FAILURE_POINTER (string_place); \
1480 DEBUG_PRINT (" Push pattern %p: ", pattern); \
1481 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1482 PUSH_FAILURE_POINTER (pattern); \
1484 /* Close the frame by moving the frame pointer past it. */ \
1485 fail_stack.frame = fail_stack.avail; \
1488 /* Estimate the size of data pushed by a typical failure stack entry.
1489 An estimate is all we need, because all we use this for
1490 is to choose a limit for how big to make the failure stack. */
1491 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1492 #define TYPICAL_FAILURE_SIZE 20
1494 /* How many items can still be added to the stack without overflowing it. */
1495 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1498 /* Pops what PUSH_FAIL_STACK pushes.
1500 We restore into the parameters, all of which should be lvalues:
1501 STR -- the saved data position.
1502 PAT -- the saved pattern position.
1503 REGSTART, REGEND -- arrays of string positions.
1505 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1506 `pend', `string1', `size1', `string2', and `size2'. */
1508 #define POP_FAILURE_POINT(str, pat) \
1510 assert (!FAIL_STACK_EMPTY ()); \
1512 /* Remove failure points and point to how many regs pushed. */ \
1513 DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \
1514 DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \
1515 DEBUG_PRINT (" size: %zd\n", fail_stack.size); \
1517 /* Pop the saved registers. */ \
1518 while (fail_stack.frame < fail_stack.avail) \
1519 POP_FAILURE_REG_OR_COUNT (); \
1521 pat = POP_FAILURE_POINTER (); \
1522 DEBUG_PRINT (" Popping pattern %p: ", pat); \
1523 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1525 /* If the saved string location is NULL, it came from an \
1526 on_failure_keep_string_jump opcode, and we want to throw away the \
1527 saved NULL, thus retaining our current position in the string. */ \
1528 str = POP_FAILURE_POINTER (); \
1529 DEBUG_PRINT (" Popping string %p: \"", str); \
1530 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1531 DEBUG_PRINT ("\"\n"); \
1533 fail_stack.frame = POP_FAILURE_INT (); \
1534 DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \
1536 assert (fail_stack.avail >= 0); \
1537 assert (fail_stack.frame <= fail_stack.avail); \
1539 DEBUG_STATEMENT (nfailure_points_popped++); \
1540 } while (0) /* POP_FAILURE_POINT */
1544 /* Registers are set to a sentinel when they haven't yet matched. */
1545 #define REG_UNSET(e) ((e) == NULL)
1547 /* Subroutine declarations and macros for regex_compile. */
1549 static reg_errcode_t
regex_compile (re_char
*pattern
, size_t size
,
1551 bool posix_backtracking
,
1552 const char *whitespace_regexp
,
1554 reg_syntax_t syntax
,
1556 struct re_pattern_buffer
*bufp
);
1557 static void store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
);
1558 static void store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
);
1559 static void insert_op1 (re_opcode_t op
, unsigned char *loc
,
1560 int arg
, unsigned char *end
);
1561 static void insert_op2 (re_opcode_t op
, unsigned char *loc
,
1562 int arg1
, int arg2
, unsigned char *end
);
1563 static boolean
at_begline_loc_p (re_char
*pattern
, re_char
*p
,
1564 reg_syntax_t syntax
);
1565 static boolean
at_endline_loc_p (re_char
*p
, re_char
*pend
,
1566 reg_syntax_t syntax
);
1567 static re_char
*skip_one_char (re_char
*p
);
1568 static int analyze_first (re_char
*p
, re_char
*pend
,
1569 char *fastmap
, const int multibyte
);
1571 /* Fetch the next character in the uncompiled pattern, with no
1573 #define PATFETCH(c) \
1576 if (p == pend) return REG_EEND; \
1577 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1582 /* If `translate' is non-null, return translate[D], else just D. We
1583 cast the subscript to translate because some data is declared as
1584 `char *', to avoid warnings when a string constant is passed. But
1585 when we use a character as a subscript we must make it unsigned. */
1587 # define TRANSLATE(d) \
1588 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1592 /* Macros for outputting the compiled pattern into `buffer'. */
1594 /* If the buffer isn't allocated when it comes in, use this. */
1595 #define INIT_BUF_SIZE 32
1597 /* Make sure we have at least N more bytes of space in buffer. */
1598 #define GET_BUFFER_SPACE(n) \
1599 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1602 /* Make sure we have one more byte of buffer space and then add C to it. */
1603 #define BUF_PUSH(c) \
1605 GET_BUFFER_SPACE (1); \
1606 *b++ = (unsigned char) (c); \
1610 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1611 #define BUF_PUSH_2(c1, c2) \
1613 GET_BUFFER_SPACE (2); \
1614 *b++ = (unsigned char) (c1); \
1615 *b++ = (unsigned char) (c2); \
1619 /* Store a jump with opcode OP at LOC to location TO. We store a
1620 relative address offset by the three bytes the jump itself occupies. */
1621 #define STORE_JUMP(op, loc, to) \
1622 store_op1 (op, loc, (to) - (loc) - 3)
1624 /* Likewise, for a two-argument jump. */
1625 #define STORE_JUMP2(op, loc, to, arg) \
1626 store_op2 (op, loc, (to) - (loc) - 3, arg)
1628 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1629 #define INSERT_JUMP(op, loc, to) \
1630 insert_op1 (op, loc, (to) - (loc) - 3, b)
1632 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1633 #define INSERT_JUMP2(op, loc, to, arg) \
1634 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1637 /* This is not an arbitrary limit: the arguments which represent offsets
1638 into the pattern are two bytes long. So if 2^15 bytes turns out to
1639 be too small, many things would have to change. */
1640 # define MAX_BUF_SIZE (1L << 15)
1642 /* Extend the buffer by twice its current size via realloc and
1643 reset the pointers that pointed into the old block to point to the
1644 correct places in the new one. If extending the buffer results in it
1645 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1646 #define EXTEND_BUFFER() \
1648 unsigned char *old_buffer = bufp->buffer; \
1649 if (bufp->allocated == MAX_BUF_SIZE) \
1651 bufp->allocated <<= 1; \
1652 if (bufp->allocated > MAX_BUF_SIZE) \
1653 bufp->allocated = MAX_BUF_SIZE; \
1654 ptrdiff_t b_off = b - old_buffer; \
1655 ptrdiff_t begalt_off = begalt - old_buffer; \
1656 bool fixup_alt_jump_set = !!fixup_alt_jump; \
1657 bool laststart_set = !!laststart; \
1658 bool pending_exact_set = !!pending_exact; \
1659 ptrdiff_t fixup_alt_jump_off, laststart_off, pending_exact_off; \
1660 if (fixup_alt_jump_set) fixup_alt_jump_off = fixup_alt_jump - old_buffer; \
1661 if (laststart_set) laststart_off = laststart - old_buffer; \
1662 if (pending_exact_set) pending_exact_off = pending_exact - old_buffer; \
1663 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1664 if (bufp->buffer == NULL) \
1665 return REG_ESPACE; \
1666 unsigned char *new_buffer = bufp->buffer; \
1667 b = new_buffer + b_off; \
1668 begalt = new_buffer + begalt_off; \
1669 if (fixup_alt_jump_set) fixup_alt_jump = new_buffer + fixup_alt_jump_off; \
1670 if (laststart_set) laststart = new_buffer + laststart_off; \
1671 if (pending_exact_set) pending_exact = new_buffer + pending_exact_off; \
1675 /* Since we have one byte reserved for the register number argument to
1676 {start,stop}_memory, the maximum number of groups we can report
1677 things about is what fits in that byte. */
1678 #define MAX_REGNUM 255
1680 /* But patterns can have more than `MAX_REGNUM' registers. We just
1681 ignore the excess. */
1682 typedef int regnum_t
;
1685 /* Macros for the compile stack. */
1687 /* Since offsets can go either forwards or backwards, this type needs to
1688 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1689 /* int may be not enough when sizeof(int) == 2. */
1690 typedef long pattern_offset_t
;
1694 pattern_offset_t begalt_offset
;
1695 pattern_offset_t fixup_alt_jump
;
1696 pattern_offset_t laststart_offset
;
1698 } compile_stack_elt_t
;
1703 compile_stack_elt_t
*stack
;
1705 size_t avail
; /* Offset of next open position. */
1706 } compile_stack_type
;
1709 #define INIT_COMPILE_STACK_SIZE 32
1711 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1712 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1714 /* The next available element. */
1715 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1717 /* Explicit quit checking is needed for Emacs, which uses polling to
1718 process input events. */
1720 # define IMMEDIATE_QUIT_CHECK \
1722 if (immediate_quit) QUIT; \
1725 # define IMMEDIATE_QUIT_CHECK ((void)0)
1728 /* Structure to manage work area for range table. */
1729 struct range_table_work_area
1731 int *table
; /* actual work area. */
1732 int allocated
; /* allocated size for work area in bytes. */
1733 int used
; /* actually used size in words. */
1734 int bits
; /* flag to record character classes */
1739 /* Make sure that WORK_AREA can hold more N multibyte characters.
1740 This is used only in set_image_of_range and set_image_of_range_1.
1741 It expects WORK_AREA to be a pointer.
1742 If it can't get the space, it returns from the surrounding function. */
1744 #define EXTEND_RANGE_TABLE(work_area, n) \
1746 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1748 extend_range_table_work_area (&work_area); \
1749 if ((work_area).table == 0) \
1750 return (REG_ESPACE); \
1754 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1755 (work_area).bits |= (bit)
1757 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1758 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1760 EXTEND_RANGE_TABLE ((work_area), 2); \
1761 (work_area).table[(work_area).used++] = (range_start); \
1762 (work_area).table[(work_area).used++] = (range_end); \
1767 /* Free allocated memory for WORK_AREA. */
1768 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1770 if ((work_area).table) \
1771 free ((work_area).table); \
1774 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1775 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1776 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1777 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1779 /* Bits used to implement the multibyte-part of the various character classes
1780 such as [:alnum:] in a charset's range table. The code currently assumes
1781 that only the low 16 bits are used. */
1782 #define BIT_WORD 0x1
1783 #define BIT_LOWER 0x2
1784 #define BIT_PUNCT 0x4
1785 #define BIT_SPACE 0x8
1786 #define BIT_UPPER 0x10
1787 #define BIT_MULTIBYTE 0x20
1788 #define BIT_ALPHA 0x40
1789 #define BIT_ALNUM 0x80
1790 #define BIT_GRAPH 0x100
1791 #define BIT_PRINT 0x200
1794 /* Set the bit for character C in a list. */
1795 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1800 /* Store characters in the range FROM to TO in the bitmap at B (for
1801 ASCII and unibyte characters) and WORK_AREA (for multibyte
1802 characters) while translating them and paying attention to the
1803 continuity of translated characters.
1805 Implementation note: It is better to implement these fairly big
1806 macros by a function, but it's not that easy because macros called
1807 in this macro assume various local variables already declared. */
1809 /* Both FROM and TO are ASCII characters. */
1811 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1815 for (C0 = (FROM); C0 <= (TO); C0++) \
1817 C1 = TRANSLATE (C0); \
1818 if (! ASCII_CHAR_P (C1)) \
1820 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1821 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1824 SET_LIST_BIT (C1); \
1829 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1831 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1833 int C0, C1, C2, I; \
1834 int USED = RANGE_TABLE_WORK_USED (work_area); \
1836 for (C0 = (FROM); C0 <= (TO); C0++) \
1838 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1839 if (CHAR_BYTE8_P (C1)) \
1840 SET_LIST_BIT (C0); \
1843 C2 = TRANSLATE (C1); \
1845 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1847 SET_LIST_BIT (C1); \
1848 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1850 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1851 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1853 if (C2 >= from - 1 && C2 <= to + 1) \
1855 if (C2 == from - 1) \
1856 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1857 else if (C2 == to + 1) \
1858 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1863 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1869 /* Both FROM and TO are multibyte characters. */
1871 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1873 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1875 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1876 for (C0 = (FROM); C0 <= (TO); C0++) \
1878 C1 = TRANSLATE (C0); \
1879 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1880 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1881 SET_LIST_BIT (C2); \
1882 if (C1 >= (FROM) && C1 <= (TO)) \
1884 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1886 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1887 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1889 if (C1 >= from - 1 && C1 <= to + 1) \
1891 if (C1 == from - 1) \
1892 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1893 else if (C1 == to + 1) \
1894 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1899 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1905 /* Get the next unsigned number in the uncompiled pattern. */
1906 #define GET_INTERVAL_COUNT(num) \
1909 FREE_STACK_RETURN (REG_EBRACE); \
1913 while ('0' <= c && c <= '9') \
1917 if (RE_DUP_MAX / 10 - (RE_DUP_MAX % 10 < c - '0') < num) \
1918 FREE_STACK_RETURN (REG_BADBR); \
1919 num = num * 10 + c - '0'; \
1921 FREE_STACK_RETURN (REG_EBRACE); \
1927 #if ! WIDE_CHAR_SUPPORT
1929 /* Parse a character class, i.e. string such as "[:name:]". *strp
1930 points to the string to be parsed and limit is length, in bytes, of
1933 If *strp point to a string that begins with "[:name:]", where name is
1934 a non-empty sequence of lower case letters, *strp will be advanced past the
1935 closing square bracket and RECC_* constant which maps to the name will be
1936 returned. If name is not a valid character class name zero, or RECC_ERROR,
1939 Otherwise, if *strp doesn’t begin with "[:name:]", -1 is returned.
1941 The function can be used on ASCII and multibyte (UTF-8-encoded) strings.
1944 re_wctype_parse (const unsigned char **strp
, unsigned limit
)
1946 const char *beg
= (const char *)*strp
, *it
;
1948 if (limit
< 4 || beg
[0] != '[' || beg
[1] != ':')
1951 beg
+= 2; /* skip opening ‘[:’ */
1952 limit
-= 3; /* opening ‘[:’ and half of closing ‘:]’; --limit handles rest */
1953 for (it
= beg
; it
[0] != ':' || it
[1] != ']'; ++it
)
1957 *strp
= (const unsigned char *)(it
+ 2);
1959 /* Sort tests in the length=five case by frequency the classes to minimize
1960 number of times we fail the comparison. The frequencies of character class
1961 names used in Emacs sources as of 2016-07-27:
1963 $ find \( -name \*.c -o -name \*.el \) -exec grep -h '\[:[a-z]*:]' {} + |
1964 sed 's/]/]\n/g' |grep -o '\[:[a-z]*:]' |sort |uniq -c |sort -nr
1982 If you update this list, consider also updating chain of or’ed conditions
1983 in execute_charset function.
1988 if (!memcmp (beg
, "word", 4)) return RECC_WORD
;
1991 if (!memcmp (beg
, "alnum", 5)) return RECC_ALNUM
;
1992 if (!memcmp (beg
, "alpha", 5)) return RECC_ALPHA
;
1993 if (!memcmp (beg
, "space", 5)) return RECC_SPACE
;
1994 if (!memcmp (beg
, "digit", 5)) return RECC_DIGIT
;
1995 if (!memcmp (beg
, "blank", 5)) return RECC_BLANK
;
1996 if (!memcmp (beg
, "upper", 5)) return RECC_UPPER
;
1997 if (!memcmp (beg
, "lower", 5)) return RECC_LOWER
;
1998 if (!memcmp (beg
, "punct", 5)) return RECC_PUNCT
;
1999 if (!memcmp (beg
, "ascii", 5)) return RECC_ASCII
;
2000 if (!memcmp (beg
, "graph", 5)) return RECC_GRAPH
;
2001 if (!memcmp (beg
, "print", 5)) return RECC_PRINT
;
2002 if (!memcmp (beg
, "cntrl", 5)) return RECC_CNTRL
;
2005 if (!memcmp (beg
, "xdigit", 6)) return RECC_XDIGIT
;
2008 if (!memcmp (beg
, "unibyte", 7)) return RECC_UNIBYTE
;
2011 if (!memcmp (beg
, "nonascii", 8)) return RECC_NONASCII
;
2014 if (!memcmp (beg
, "multibyte", 9)) return RECC_MULTIBYTE
;
2021 /* True if CH is in the char class CC. */
2023 re_iswctype (int ch
, re_wctype_t cc
)
2027 case RECC_ALNUM
: return ISALNUM (ch
) != 0;
2028 case RECC_ALPHA
: return ISALPHA (ch
) != 0;
2029 case RECC_BLANK
: return ISBLANK (ch
) != 0;
2030 case RECC_CNTRL
: return ISCNTRL (ch
) != 0;
2031 case RECC_DIGIT
: return ISDIGIT (ch
) != 0;
2032 case RECC_GRAPH
: return ISGRAPH (ch
) != 0;
2033 case RECC_LOWER
: return ISLOWER (ch
) != 0;
2034 case RECC_PRINT
: return ISPRINT (ch
) != 0;
2035 case RECC_PUNCT
: return ISPUNCT (ch
) != 0;
2036 case RECC_SPACE
: return ISSPACE (ch
) != 0;
2037 case RECC_UPPER
: return ISUPPER (ch
) != 0;
2038 case RECC_XDIGIT
: return ISXDIGIT (ch
) != 0;
2039 case RECC_ASCII
: return IS_REAL_ASCII (ch
) != 0;
2040 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2041 case RECC_UNIBYTE
: return ISUNIBYTE (ch
) != 0;
2042 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2043 case RECC_WORD
: return ISWORD (ch
) != 0;
2044 case RECC_ERROR
: return false;
2050 /* Return a bit-pattern to use in the range-table bits to match multibyte
2051 chars of class CC. */
2053 re_wctype_to_bit (re_wctype_t cc
)
2058 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2059 case RECC_ALPHA
: return BIT_ALPHA
;
2060 case RECC_ALNUM
: return BIT_ALNUM
;
2061 case RECC_WORD
: return BIT_WORD
;
2062 case RECC_LOWER
: return BIT_LOWER
;
2063 case RECC_UPPER
: return BIT_UPPER
;
2064 case RECC_PUNCT
: return BIT_PUNCT
;
2065 case RECC_SPACE
: return BIT_SPACE
;
2066 case RECC_GRAPH
: return BIT_GRAPH
;
2067 case RECC_PRINT
: return BIT_PRINT
;
2068 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2069 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2076 /* Filling in the work area of a range. */
2078 /* Actually extend the space in WORK_AREA. */
2081 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2083 work_area
->allocated
+= 16 * sizeof (int);
2084 work_area
->table
= realloc (work_area
->table
, work_area
->allocated
);
2090 /* Carefully find the ranges of codes that are equivalent
2091 under case conversion to the range start..end when passed through
2092 TRANSLATE. Handle the case where non-letters can come in between
2093 two upper-case letters (which happens in Latin-1).
2094 Also handle the case of groups of more than 2 case-equivalent chars.
2096 The basic method is to look at consecutive characters and see
2097 if they can form a run that can be handled as one.
2099 Returns -1 if successful, REG_ESPACE if ran out of space. */
2102 set_image_of_range_1 (struct range_table_work_area
*work_area
,
2103 re_wchar_t start
, re_wchar_t end
,
2104 RE_TRANSLATE_TYPE translate
)
2106 /* `one_case' indicates a character, or a run of characters,
2107 each of which is an isolate (no case-equivalents).
2108 This includes all ASCII non-letters.
2110 `two_case' indicates a character, or a run of characters,
2111 each of which has two case-equivalent forms.
2112 This includes all ASCII letters.
2114 `strange' indicates a character that has more than one
2117 enum case_type
{one_case
, two_case
, strange
};
2119 /* Describe the run that is in progress,
2120 which the next character can try to extend.
2121 If run_type is strange, that means there really is no run.
2122 If run_type is one_case, then run_start...run_end is the run.
2123 If run_type is two_case, then the run is run_start...run_end,
2124 and the case-equivalents end at run_eqv_end. */
2126 enum case_type run_type
= strange
;
2127 int run_start
, run_end
, run_eqv_end
;
2129 Lisp_Object eqv_table
;
2131 if (!RE_TRANSLATE_P (translate
))
2133 EXTEND_RANGE_TABLE (work_area
, 2);
2134 work_area
->table
[work_area
->used
++] = (start
);
2135 work_area
->table
[work_area
->used
++] = (end
);
2139 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2141 for (; start
<= end
; start
++)
2143 enum case_type this_type
;
2144 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2145 int minchar
, maxchar
;
2147 /* Classify this character */
2149 this_type
= one_case
;
2150 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2151 this_type
= two_case
;
2153 this_type
= strange
;
2156 minchar
= start
, maxchar
= eqv
;
2158 minchar
= eqv
, maxchar
= start
;
2160 /* Can this character extend the run in progress? */
2161 if (this_type
== strange
|| this_type
!= run_type
2162 || !(minchar
== run_end
+ 1
2163 && (run_type
== two_case
2164 ? maxchar
== run_eqv_end
+ 1 : 1)))
2167 Record each of its equivalent ranges. */
2168 if (run_type
== one_case
)
2170 EXTEND_RANGE_TABLE (work_area
, 2);
2171 work_area
->table
[work_area
->used
++] = run_start
;
2172 work_area
->table
[work_area
->used
++] = run_end
;
2174 else if (run_type
== two_case
)
2176 EXTEND_RANGE_TABLE (work_area
, 4);
2177 work_area
->table
[work_area
->used
++] = run_start
;
2178 work_area
->table
[work_area
->used
++] = run_end
;
2179 work_area
->table
[work_area
->used
++]
2180 = RE_TRANSLATE (eqv_table
, run_start
);
2181 work_area
->table
[work_area
->used
++]
2182 = RE_TRANSLATE (eqv_table
, run_end
);
2187 if (this_type
== strange
)
2189 /* For a strange character, add each of its equivalents, one
2190 by one. Don't start a range. */
2193 EXTEND_RANGE_TABLE (work_area
, 2);
2194 work_area
->table
[work_area
->used
++] = eqv
;
2195 work_area
->table
[work_area
->used
++] = eqv
;
2196 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2198 while (eqv
!= start
);
2201 /* Add this char to the run, or start a new run. */
2202 else if (run_type
== strange
)
2204 /* Initialize a new range. */
2205 run_type
= this_type
;
2208 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2212 /* Extend a running range. */
2214 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2218 /* If a run is still in progress at the end, finish it now
2219 by recording its equivalent ranges. */
2220 if (run_type
== one_case
)
2222 EXTEND_RANGE_TABLE (work_area
, 2);
2223 work_area
->table
[work_area
->used
++] = run_start
;
2224 work_area
->table
[work_area
->used
++] = run_end
;
2226 else if (run_type
== two_case
)
2228 EXTEND_RANGE_TABLE (work_area
, 4);
2229 work_area
->table
[work_area
->used
++] = run_start
;
2230 work_area
->table
[work_area
->used
++] = run_end
;
2231 work_area
->table
[work_area
->used
++]
2232 = RE_TRANSLATE (eqv_table
, run_start
);
2233 work_area
->table
[work_area
->used
++]
2234 = RE_TRANSLATE (eqv_table
, run_end
);
2242 /* Record the image of the range start..end when passed through
2243 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2244 and is not even necessarily contiguous.
2245 Normally we approximate it with the smallest contiguous range that contains
2246 all the chars we need. However, for Latin-1 we go to extra effort
2249 This function is not called for ASCII ranges.
2251 Returns -1 if successful, REG_ESPACE if ran out of space. */
2254 set_image_of_range (struct range_table_work_area
*work_area
,
2255 re_wchar_t start
, re_wchar_t end
,
2256 RE_TRANSLATE_TYPE translate
)
2258 re_wchar_t cmin
, cmax
;
2261 /* For Latin-1 ranges, use set_image_of_range_1
2262 to get proper handling of ranges that include letters and nonletters.
2263 For a range that includes the whole of Latin-1, this is not necessary.
2264 For other character sets, we don't bother to get this right. */
2265 if (RE_TRANSLATE_P (translate
) && start
< 04400
2266 && !(start
< 04200 && end
>= 04377))
2273 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2283 EXTEND_RANGE_TABLE (work_area
, 2);
2284 work_area
->table
[work_area
->used
++] = (start
);
2285 work_area
->table
[work_area
->used
++] = (end
);
2287 cmin
= -1, cmax
= -1;
2289 if (RE_TRANSLATE_P (translate
))
2293 for (ch
= start
; ch
<= end
; ch
++)
2295 re_wchar_t c
= TRANSLATE (ch
);
2296 if (! (start
<= c
&& c
<= end
))
2302 cmin
= min (cmin
, c
);
2303 cmax
= max (cmax
, c
);
2310 EXTEND_RANGE_TABLE (work_area
, 2);
2311 work_area
->table
[work_area
->used
++] = (cmin
);
2312 work_area
->table
[work_area
->used
++] = (cmax
);
2320 #ifndef MATCH_MAY_ALLOCATE
2322 /* If we cannot allocate large objects within re_match_2_internal,
2323 we make the fail stack and register vectors global.
2324 The fail stack, we grow to the maximum size when a regexp
2326 The register vectors, we adjust in size each time we
2327 compile a regexp, according to the number of registers it needs. */
2329 static fail_stack_type fail_stack
;
2331 /* Size with which the following vectors are currently allocated.
2332 That is so we can make them bigger as needed,
2333 but never make them smaller. */
2334 static int regs_allocated_size
;
2336 static re_char
** regstart
, ** regend
;
2337 static re_char
**best_regstart
, **best_regend
;
2339 /* Make the register vectors big enough for NUM_REGS registers,
2340 but don't make them smaller. */
2343 regex_grow_registers (int num_regs
)
2345 if (num_regs
> regs_allocated_size
)
2347 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2348 RETALLOC_IF (regend
, num_regs
, re_char
*);
2349 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2350 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2352 regs_allocated_size
= num_regs
;
2356 #endif /* not MATCH_MAY_ALLOCATE */
2358 static boolean
group_in_compile_stack (compile_stack_type compile_stack
,
2361 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2362 Returns one of error codes defined in `regex.h', or zero for success.
2364 If WHITESPACE_REGEXP is given (only #ifdef emacs), it is used instead of
2365 a space character in PATTERN.
2367 Assumes the `allocated' (and perhaps `buffer') and `translate'
2368 fields are set in BUFP on entry.
2370 If it succeeds, results are put in BUFP (if it returns an error, the
2371 contents of BUFP are undefined):
2372 `buffer' is the compiled pattern;
2373 `syntax' is set to SYNTAX;
2374 `used' is set to the length of the compiled pattern;
2375 `fastmap_accurate' is zero;
2376 `re_nsub' is the number of subexpressions in PATTERN;
2377 `not_bol' and `not_eol' are zero;
2379 The `fastmap' field is neither examined nor set. */
2381 /* Insert the `jump' from the end of last alternative to "here".
2382 The space for the jump has already been allocated. */
2383 #define FIXUP_ALT_JUMP() \
2385 if (fixup_alt_jump) \
2386 STORE_JUMP (jump, fixup_alt_jump, b); \
2390 /* Return, freeing storage we allocated. */
2391 #define FREE_STACK_RETURN(value) \
2393 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2394 free (compile_stack.stack); \
2398 static reg_errcode_t
2399 regex_compile (const_re_char
*pattern
, size_t size
,
2401 # define syntax RE_SYNTAX_EMACS
2402 bool posix_backtracking
,
2403 const char *whitespace_regexp
,
2405 reg_syntax_t syntax
,
2406 # define posix_backtracking (!(syntax & RE_NO_POSIX_BACKTRACKING))
2408 struct re_pattern_buffer
*bufp
)
2410 /* We fetch characters from PATTERN here. */
2411 register re_wchar_t c
, c1
;
2413 /* Points to the end of the buffer, where we should append. */
2414 register unsigned char *b
;
2416 /* Keeps track of unclosed groups. */
2417 compile_stack_type compile_stack
;
2419 /* Points to the current (ending) position in the pattern. */
2421 /* `const' makes AIX compiler fail. */
2422 unsigned char *p
= pattern
;
2424 re_char
*p
= pattern
;
2426 re_char
*pend
= pattern
+ size
;
2428 /* How to translate the characters in the pattern. */
2429 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2431 /* Address of the count-byte of the most recently inserted `exactn'
2432 command. This makes it possible to tell if a new exact-match
2433 character can be added to that command or if the character requires
2434 a new `exactn' command. */
2435 unsigned char *pending_exact
= 0;
2437 /* Address of start of the most recently finished expression.
2438 This tells, e.g., postfix * where to find the start of its
2439 operand. Reset at the beginning of groups and alternatives. */
2440 unsigned char *laststart
= 0;
2442 /* Address of beginning of regexp, or inside of last group. */
2443 unsigned char *begalt
;
2445 /* Place in the uncompiled pattern (i.e., the {) to
2446 which to go back if the interval is invalid. */
2447 re_char
*beg_interval
;
2449 /* Address of the place where a forward jump should go to the end of
2450 the containing expression. Each alternative of an `or' -- except the
2451 last -- ends with a forward jump of this sort. */
2452 unsigned char *fixup_alt_jump
= 0;
2454 /* Work area for range table of charset. */
2455 struct range_table_work_area range_table_work
;
2457 /* If the object matched can contain multibyte characters. */
2458 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2461 /* Nonzero if we have pushed down into a subpattern. */
2462 int in_subpattern
= 0;
2464 /* These hold the values of p, pattern, and pend from the main
2465 pattern when we have pushed into a subpattern. */
2467 re_char
*main_pattern
;
2473 DEBUG_PRINT ("\nCompiling pattern: ");
2476 unsigned debug_count
;
2478 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2479 putchar (pattern
[debug_count
]);
2484 /* Initialize the compile stack. */
2485 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2486 if (compile_stack
.stack
== NULL
)
2489 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2490 compile_stack
.avail
= 0;
2492 range_table_work
.table
= 0;
2493 range_table_work
.allocated
= 0;
2495 /* Initialize the pattern buffer. */
2497 bufp
->syntax
= syntax
;
2499 bufp
->fastmap_accurate
= 0;
2500 bufp
->not_bol
= bufp
->not_eol
= 0;
2501 bufp
->used_syntax
= 0;
2503 /* Set `used' to zero, so that if we return an error, the pattern
2504 printer (for debugging) will think there's no pattern. We reset it
2508 /* Always count groups, whether or not bufp->no_sub is set. */
2511 #if !defined emacs && !defined SYNTAX_TABLE
2512 /* Initialize the syntax table. */
2513 init_syntax_once ();
2516 if (bufp
->allocated
== 0)
2519 { /* If zero allocated, but buffer is non-null, try to realloc
2520 enough space. This loses if buffer's address is bogus, but
2521 that is the user's responsibility. */
2522 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2525 { /* Caller did not allocate a buffer. Do it for them. */
2526 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2528 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2530 bufp
->allocated
= INIT_BUF_SIZE
;
2533 begalt
= b
= bufp
->buffer
;
2535 /* Loop through the uncompiled pattern until we're at the end. */
2541 /* If this is the end of an included regexp,
2542 pop back to the main regexp and try again. */
2546 pattern
= main_pattern
;
2552 /* If this is the end of the main regexp, we are done. */
2565 /* If there's no special whitespace regexp, treat
2566 spaces normally. And don't try to do this recursively. */
2567 if (!whitespace_regexp
|| in_subpattern
)
2570 /* Peek past following spaces. */
2577 /* If the spaces are followed by a repetition op,
2578 treat them normally. */
2580 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2581 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2584 /* Replace the spaces with the whitespace regexp. */
2588 main_pattern
= pattern
;
2589 p
= pattern
= (re_char
*) whitespace_regexp
;
2590 pend
= p
+ strlen (whitespace_regexp
);
2597 if ( /* If at start of pattern, it's an operator. */
2599 /* If context independent, it's an operator. */
2600 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2601 /* Otherwise, depends on what's come before. */
2602 || at_begline_loc_p (pattern
, p
, syntax
))
2603 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2612 if ( /* If at end of pattern, it's an operator. */
2614 /* If context independent, it's an operator. */
2615 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2616 /* Otherwise, depends on what's next. */
2617 || at_endline_loc_p (p
, pend
, syntax
))
2618 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2627 if ((syntax
& RE_BK_PLUS_QM
)
2628 || (syntax
& RE_LIMITED_OPS
))
2632 /* If there is no previous pattern... */
2635 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2636 FREE_STACK_RETURN (REG_BADRPT
);
2637 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2642 /* 1 means zero (many) matches is allowed. */
2643 boolean zero_times_ok
= 0, many_times_ok
= 0;
2646 /* If there is a sequence of repetition chars, collapse it
2647 down to just one (the right one). We can't combine
2648 interval operators with these because of, e.g., `a{2}*',
2649 which should only match an even number of `a's. */
2653 if ((syntax
& RE_FRUGAL
)
2654 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2658 zero_times_ok
|= c
!= '+';
2659 many_times_ok
|= c
!= '?';
2665 || (!(syntax
& RE_BK_PLUS_QM
)
2666 && (*p
== '+' || *p
== '?')))
2668 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2671 FREE_STACK_RETURN (REG_EESCAPE
);
2672 if (p
[1] == '+' || p
[1] == '?')
2673 PATFETCH (c
); /* Gobble up the backslash. */
2679 /* If we get here, we found another repeat character. */
2683 /* Star, etc. applied to an empty pattern is equivalent
2684 to an empty pattern. */
2685 if (!laststart
|| laststart
== b
)
2688 /* Now we know whether or not zero matches is allowed
2689 and also whether or not two or more matches is allowed. */
2694 boolean simple
= skip_one_char (laststart
) == b
;
2695 size_t startoffset
= 0;
2697 /* Check if the loop can match the empty string. */
2698 (simple
|| !analyze_first (laststart
, b
, NULL
, 0))
2699 ? on_failure_jump
: on_failure_jump_loop
;
2700 assert (skip_one_char (laststart
) <= b
);
2702 if (!zero_times_ok
&& simple
)
2703 { /* Since simple * loops can be made faster by using
2704 on_failure_keep_string_jump, we turn simple P+
2705 into PP* if P is simple. */
2706 unsigned char *p1
, *p2
;
2707 startoffset
= b
- laststart
;
2708 GET_BUFFER_SPACE (startoffset
);
2709 p1
= b
; p2
= laststart
;
2715 GET_BUFFER_SPACE (6);
2718 STORE_JUMP (ofj
, b
, b
+ 6);
2720 /* Simple * loops can use on_failure_keep_string_jump
2721 depending on what follows. But since we don't know
2722 that yet, we leave the decision up to
2723 on_failure_jump_smart. */
2724 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2725 laststart
+ startoffset
, b
+ 6);
2727 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2732 /* A simple ? pattern. */
2733 assert (zero_times_ok
);
2734 GET_BUFFER_SPACE (3);
2735 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2739 else /* not greedy */
2740 { /* I wish the greedy and non-greedy cases could be merged. */
2742 GET_BUFFER_SPACE (7); /* We might use less. */
2745 boolean emptyp
= analyze_first (laststart
, b
, NULL
, 0);
2747 /* The non-greedy multiple match looks like
2748 a repeat..until: we only need a conditional jump
2749 at the end of the loop. */
2750 if (emptyp
) BUF_PUSH (no_op
);
2751 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2752 : on_failure_jump
, b
, laststart
);
2756 /* The repeat...until naturally matches one or more.
2757 To also match zero times, we need to first jump to
2758 the end of the loop (its conditional jump). */
2759 INSERT_JUMP (jump
, laststart
, b
);
2765 /* non-greedy a?? */
2766 INSERT_JUMP (jump
, laststart
, b
+ 3);
2768 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2787 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2789 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2791 /* Ensure that we have enough space to push a charset: the
2792 opcode, the length count, and the bitset; 34 bytes in all. */
2793 GET_BUFFER_SPACE (34);
2797 /* We test `*p == '^' twice, instead of using an if
2798 statement, so we only need one BUF_PUSH. */
2799 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2803 /* Remember the first position in the bracket expression. */
2806 /* Push the number of bytes in the bitmap. */
2807 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2809 /* Clear the whole map. */
2810 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2812 /* charset_not matches newline according to a syntax bit. */
2813 if ((re_opcode_t
) b
[-2] == charset_not
2814 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2815 SET_LIST_BIT ('\n');
2817 /* Read in characters and ranges, setting map bits. */
2820 boolean escaped_char
= false;
2821 const unsigned char *p2
= p
;
2825 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2827 /* See if we're at the beginning of a possible character
2829 if (syntax
& RE_CHAR_CLASSES
&&
2830 (cc
= re_wctype_parse(&p
, pend
- p
)) != -1)
2833 FREE_STACK_RETURN (REG_ECTYPE
);
2836 FREE_STACK_RETURN (REG_EBRACK
);
2839 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2840 if (re_iswctype (btowc (ch
), cc
))
2843 if (c
< (1 << BYTEWIDTH
))
2847 /* Most character classes in a multibyte match just set
2848 a flag. Exceptions are is_blank, is_digit, is_cntrl, and
2849 is_xdigit, since they can only match ASCII characters.
2850 We don't need to handle them for multibyte. */
2852 /* Setup the gl_state object to its buffer-defined value.
2853 This hardcodes the buffer-global syntax-table for ASCII
2854 chars, while the other chars will obey syntax-table
2855 properties. It's not ideal, but it's the way it's been
2857 SETUP_BUFFER_SYNTAX_TABLE ();
2859 for (c
= 0; c
< 0x80; ++c
)
2860 if (re_iswctype (c
, cc
))
2866 if (ASCII_CHAR_P (c1
))
2868 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
2871 SET_RANGE_TABLE_WORK_AREA_BIT
2872 (range_table_work
, re_wctype_to_bit (cc
));
2874 /* In most cases the matching rule for char classes only
2875 uses the syntax table for multibyte chars, so that the
2876 content of the syntax-table is not hardcoded in the
2877 range_table. SPACE and WORD are the two exceptions. */
2878 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2879 bufp
->used_syntax
= 1;
2881 /* Repeat the loop. */
2885 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2886 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2887 So the translation is done later in a loop. Example:
2888 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2891 /* \ might escape characters inside [...] and [^...]. */
2892 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2894 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2897 escaped_char
= true;
2901 /* Could be the end of the bracket expression. If it's
2902 not (i.e., when the bracket expression is `[]' so
2903 far), the ']' character bit gets set way below. */
2904 if (c
== ']' && p2
!= p1
)
2908 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2911 /* Discard the `-'. */
2914 /* Fetch the character which ends the range. */
2917 if (CHAR_BYTE8_P (c1
)
2918 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
2919 /* Treat the range from a multibyte character to
2920 raw-byte character as empty. */
2925 /* Range from C to C. */
2930 if (syntax
& RE_NO_EMPTY_RANGES
)
2931 FREE_STACK_RETURN (REG_ERANGEX
);
2932 /* Else, repeat the loop. */
2937 /* Set the range into bitmap */
2938 for (; c
<= c1
; c
++)
2941 if (ch
< (1 << BYTEWIDTH
))
2948 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
2950 if (CHAR_BYTE8_P (c1
))
2951 c
= BYTE8_TO_CHAR (128);
2955 if (CHAR_BYTE8_P (c
))
2957 c
= CHAR_TO_BYTE8 (c
);
2958 c1
= CHAR_TO_BYTE8 (c1
);
2959 for (; c
<= c1
; c
++)
2964 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
2968 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
2975 /* Discard any (non)matching list bytes that are all 0 at the
2976 end of the map. Decrease the map-length byte too. */
2977 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2981 /* Build real range table from work area. */
2982 if (RANGE_TABLE_WORK_USED (range_table_work
)
2983 || RANGE_TABLE_WORK_BITS (range_table_work
))
2986 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
2988 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2989 bytes for flags, two for COUNT, and three bytes for
2991 GET_BUFFER_SPACE (4 + used
* 3);
2993 /* Indicate the existence of range table. */
2994 laststart
[1] |= 0x80;
2996 /* Store the character class flag bits into the range table.
2997 If not in emacs, these flag bits are always 0. */
2998 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
2999 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3001 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3002 for (i
= 0; i
< used
; i
++)
3003 STORE_CHARACTER_AND_INCR
3004 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3011 if (syntax
& RE_NO_BK_PARENS
)
3018 if (syntax
& RE_NO_BK_PARENS
)
3025 if (syntax
& RE_NEWLINE_ALT
)
3032 if (syntax
& RE_NO_BK_VBAR
)
3039 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3040 goto handle_interval
;
3046 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3048 /* Do not translate the character after the \, so that we can
3049 distinguish, e.g., \B from \b, even if we normally would
3050 translate, e.g., B to b. */
3056 if (syntax
& RE_NO_BK_PARENS
)
3057 goto normal_backslash
;
3062 regnum_t regnum
= 0;
3065 /* Look for a special (?...) construct */
3066 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3068 PATFETCH (c
); /* Gobble up the '?'. */
3074 case ':': shy
= 1; break;
3076 /* An explicitly specified regnum must start
3079 FREE_STACK_RETURN (REG_BADPAT
);
3080 case '1': case '2': case '3': case '4':
3081 case '5': case '6': case '7': case '8': case '9':
3082 regnum
= 10*regnum
+ (c
- '0'); break;
3084 /* Only (?:...) is supported right now. */
3085 FREE_STACK_RETURN (REG_BADPAT
);
3092 regnum
= ++bufp
->re_nsub
;
3094 { /* It's actually not shy, but explicitly numbered. */
3096 if (regnum
> bufp
->re_nsub
)
3097 bufp
->re_nsub
= regnum
;
3098 else if (regnum
> bufp
->re_nsub
3099 /* Ideally, we'd want to check that the specified
3100 group can't have matched (i.e. all subgroups
3101 using the same regnum are in other branches of
3102 OR patterns), but we don't currently keep track
3103 of enough info to do that easily. */
3104 || group_in_compile_stack (compile_stack
, regnum
))
3105 FREE_STACK_RETURN (REG_BADPAT
);
3108 /* It's really shy. */
3109 regnum
= - bufp
->re_nsub
;
3111 if (COMPILE_STACK_FULL
)
3113 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3114 compile_stack_elt_t
);
3115 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3117 compile_stack
.size
<<= 1;
3120 /* These are the values to restore when we hit end of this
3121 group. They are all relative offsets, so that if the
3122 whole pattern moves because of realloc, they will still
3124 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3125 COMPILE_STACK_TOP
.fixup_alt_jump
3126 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3127 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3128 COMPILE_STACK_TOP
.regnum
= regnum
;
3130 /* Do not push a start_memory for groups beyond the last one
3131 we can represent in the compiled pattern. */
3132 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3133 BUF_PUSH_2 (start_memory
, regnum
);
3135 compile_stack
.avail
++;
3140 /* If we've reached MAX_REGNUM groups, then this open
3141 won't actually generate any code, so we'll have to
3142 clear pending_exact explicitly. */
3148 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3150 if (COMPILE_STACK_EMPTY
)
3152 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3153 goto normal_backslash
;
3155 FREE_STACK_RETURN (REG_ERPAREN
);
3161 /* See similar code for backslashed left paren above. */
3162 if (COMPILE_STACK_EMPTY
)
3164 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3167 FREE_STACK_RETURN (REG_ERPAREN
);
3170 /* Since we just checked for an empty stack above, this
3171 ``can't happen''. */
3172 assert (compile_stack
.avail
!= 0);
3174 /* We don't just want to restore into `regnum', because
3175 later groups should continue to be numbered higher,
3176 as in `(ab)c(de)' -- the second group is #2. */
3179 compile_stack
.avail
--;
3180 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3182 = COMPILE_STACK_TOP
.fixup_alt_jump
3183 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3185 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3186 regnum
= COMPILE_STACK_TOP
.regnum
;
3187 /* If we've reached MAX_REGNUM groups, then this open
3188 won't actually generate any code, so we'll have to
3189 clear pending_exact explicitly. */
3192 /* We're at the end of the group, so now we know how many
3193 groups were inside this one. */
3194 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3195 BUF_PUSH_2 (stop_memory
, regnum
);
3200 case '|': /* `\|'. */
3201 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3202 goto normal_backslash
;
3204 if (syntax
& RE_LIMITED_OPS
)
3207 /* Insert before the previous alternative a jump which
3208 jumps to this alternative if the former fails. */
3209 GET_BUFFER_SPACE (3);
3210 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3214 /* The alternative before this one has a jump after it
3215 which gets executed if it gets matched. Adjust that
3216 jump so it will jump to this alternative's analogous
3217 jump (put in below, which in turn will jump to the next
3218 (if any) alternative's such jump, etc.). The last such
3219 jump jumps to the correct final destination. A picture:
3225 If we are at `b', then fixup_alt_jump right now points to a
3226 three-byte space after `a'. We'll put in the jump, set
3227 fixup_alt_jump to right after `b', and leave behind three
3228 bytes which we'll fill in when we get to after `c'. */
3232 /* Mark and leave space for a jump after this alternative,
3233 to be filled in later either by next alternative or
3234 when know we're at the end of a series of alternatives. */
3236 GET_BUFFER_SPACE (3);
3245 /* If \{ is a literal. */
3246 if (!(syntax
& RE_INTERVALS
)
3247 /* If we're at `\{' and it's not the open-interval
3249 || (syntax
& RE_NO_BK_BRACES
))
3250 goto normal_backslash
;
3254 /* If got here, then the syntax allows intervals. */
3256 /* At least (most) this many matches must be made. */
3257 int lower_bound
= 0, upper_bound
= -1;
3261 GET_INTERVAL_COUNT (lower_bound
);
3264 GET_INTERVAL_COUNT (upper_bound
);
3266 /* Interval such as `{1}' => match exactly once. */
3267 upper_bound
= lower_bound
;
3270 || (0 <= upper_bound
&& upper_bound
< lower_bound
))
3271 FREE_STACK_RETURN (REG_BADBR
);
3273 if (!(syntax
& RE_NO_BK_BRACES
))
3276 FREE_STACK_RETURN (REG_BADBR
);
3278 FREE_STACK_RETURN (REG_EESCAPE
);
3283 FREE_STACK_RETURN (REG_BADBR
);
3285 /* We just parsed a valid interval. */
3287 /* If it's invalid to have no preceding re. */
3290 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3291 FREE_STACK_RETURN (REG_BADRPT
);
3292 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3295 goto unfetch_interval
;
3298 if (upper_bound
== 0)
3299 /* If the upper bound is zero, just drop the sub pattern
3302 else if (lower_bound
== 1 && upper_bound
== 1)
3303 /* Just match it once: nothing to do here. */
3306 /* Otherwise, we have a nontrivial interval. When
3307 we're all done, the pattern will look like:
3308 set_number_at <jump count> <upper bound>
3309 set_number_at <succeed_n count> <lower bound>
3310 succeed_n <after jump addr> <succeed_n count>
3312 jump_n <succeed_n addr> <jump count>
3313 (The upper bound and `jump_n' are omitted if
3314 `upper_bound' is 1, though.) */
3316 { /* If the upper bound is > 1, we need to insert
3317 more at the end of the loop. */
3318 unsigned int nbytes
= (upper_bound
< 0 ? 3
3319 : upper_bound
> 1 ? 5 : 0);
3320 unsigned int startoffset
= 0;
3322 GET_BUFFER_SPACE (20); /* We might use less. */
3324 if (lower_bound
== 0)
3326 /* A succeed_n that starts with 0 is really a
3327 a simple on_failure_jump_loop. */
3328 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3334 /* Initialize lower bound of the `succeed_n', even
3335 though it will be set during matching by its
3336 attendant `set_number_at' (inserted next),
3337 because `re_compile_fastmap' needs to know.
3338 Jump to the `jump_n' we might insert below. */
3339 INSERT_JUMP2 (succeed_n
, laststart
,
3344 /* Code to initialize the lower bound. Insert
3345 before the `succeed_n'. The `5' is the last two
3346 bytes of this `set_number_at', plus 3 bytes of
3347 the following `succeed_n'. */
3348 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3353 if (upper_bound
< 0)
3355 /* A negative upper bound stands for infinity,
3356 in which case it degenerates to a plain jump. */
3357 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3360 else if (upper_bound
> 1)
3361 { /* More than one repetition is allowed, so
3362 append a backward jump to the `succeed_n'
3363 that starts this interval.
3365 When we've reached this during matching,
3366 we'll have matched the interval once, so
3367 jump back only `upper_bound - 1' times. */
3368 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3372 /* The location we want to set is the second
3373 parameter of the `jump_n'; that is `b-2' as
3374 an absolute address. `laststart' will be
3375 the `set_number_at' we're about to insert;
3376 `laststart+3' the number to set, the source
3377 for the relative address. But we are
3378 inserting into the middle of the pattern --
3379 so everything is getting moved up by 5.
3380 Conclusion: (b - 2) - (laststart + 3) + 5,
3381 i.e., b - laststart.
3383 We insert this at the beginning of the loop
3384 so that if we fail during matching, we'll
3385 reinitialize the bounds. */
3386 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3387 upper_bound
- 1, b
);
3392 beg_interval
= NULL
;
3397 /* If an invalid interval, match the characters as literals. */
3398 assert (beg_interval
);
3400 beg_interval
= NULL
;
3402 /* normal_char and normal_backslash need `c'. */
3405 if (!(syntax
& RE_NO_BK_BRACES
))
3407 assert (p
> pattern
&& p
[-1] == '\\');
3408 goto normal_backslash
;
3422 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3428 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3434 BUF_PUSH_2 (categoryspec
, c
);
3440 BUF_PUSH_2 (notcategoryspec
, c
);
3446 if (syntax
& RE_NO_GNU_OPS
)
3449 BUF_PUSH_2 (syntaxspec
, Sword
);
3454 if (syntax
& RE_NO_GNU_OPS
)
3457 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3462 if (syntax
& RE_NO_GNU_OPS
)
3469 if (syntax
& RE_NO_GNU_OPS
)
3476 if (syntax
& RE_NO_GNU_OPS
)
3485 FREE_STACK_RETURN (REG_BADPAT
);
3489 if (syntax
& RE_NO_GNU_OPS
)
3491 BUF_PUSH (wordbound
);
3495 if (syntax
& RE_NO_GNU_OPS
)
3497 BUF_PUSH (notwordbound
);
3501 if (syntax
& RE_NO_GNU_OPS
)
3507 if (syntax
& RE_NO_GNU_OPS
)
3512 case '1': case '2': case '3': case '4': case '5':
3513 case '6': case '7': case '8': case '9':
3517 if (syntax
& RE_NO_BK_REFS
)
3518 goto normal_backslash
;
3522 if (reg
> bufp
->re_nsub
|| reg
< 1
3523 /* Can't back reference to a subexp before its end. */
3524 || group_in_compile_stack (compile_stack
, reg
))
3525 FREE_STACK_RETURN (REG_ESUBREG
);
3528 BUF_PUSH_2 (duplicate
, reg
);
3535 if (syntax
& RE_BK_PLUS_QM
)
3538 goto normal_backslash
;
3542 /* You might think it would be useful for \ to mean
3543 not to translate; but if we don't translate it
3544 it will never match anything. */
3551 /* Expects the character in `c'. */
3553 /* If no exactn currently being built. */
3556 /* If last exactn not at current position. */
3557 || pending_exact
+ *pending_exact
+ 1 != b
3559 /* We have only one byte following the exactn for the count. */
3560 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3562 /* If followed by a repetition operator. */
3563 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3564 || ((syntax
& RE_BK_PLUS_QM
)
3565 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3566 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3567 || ((syntax
& RE_INTERVALS
)
3568 && ((syntax
& RE_NO_BK_BRACES
)
3569 ? p
!= pend
&& *p
== '{'
3570 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3572 /* Start building a new exactn. */
3576 BUF_PUSH_2 (exactn
, 0);
3577 pending_exact
= b
- 1;
3580 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3587 len
= CHAR_STRING (c
, b
);
3592 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3593 if (! CHAR_BYTE8_P (c1
))
3595 re_wchar_t c2
= TRANSLATE (c1
);
3597 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3603 (*pending_exact
) += len
;
3608 } /* while p != pend */
3611 /* Through the pattern now. */
3615 if (!COMPILE_STACK_EMPTY
)
3616 FREE_STACK_RETURN (REG_EPAREN
);
3618 /* If we don't want backtracking, force success
3619 the first time we reach the end of the compiled pattern. */
3620 if (!posix_backtracking
)
3623 /* We have succeeded; set the length of the buffer. */
3624 bufp
->used
= b
- bufp
->buffer
;
3629 re_compile_fastmap (bufp
);
3630 DEBUG_PRINT ("\nCompiled pattern: \n");
3631 print_compiled_pattern (bufp
);
3636 #ifndef MATCH_MAY_ALLOCATE
3637 /* Initialize the failure stack to the largest possible stack. This
3638 isn't necessary unless we're trying to avoid calling alloca in
3639 the search and match routines. */
3641 int num_regs
= bufp
->re_nsub
+ 1;
3643 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3645 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3646 falk_stack
.stack
= realloc (fail_stack
.stack
,
3647 fail_stack
.size
* sizeof *falk_stack
.stack
);
3650 regex_grow_registers (num_regs
);
3652 #endif /* not MATCH_MAY_ALLOCATE */
3654 FREE_STACK_RETURN (REG_NOERROR
);
3659 # undef posix_backtracking
3661 } /* regex_compile */
3663 /* Subroutines for `regex_compile'. */
3665 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3668 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3670 *loc
= (unsigned char) op
;
3671 STORE_NUMBER (loc
+ 1, arg
);
3675 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3678 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3680 *loc
= (unsigned char) op
;
3681 STORE_NUMBER (loc
+ 1, arg1
);
3682 STORE_NUMBER (loc
+ 3, arg2
);
3686 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3687 for OP followed by two-byte integer parameter ARG. */
3690 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3692 register unsigned char *pfrom
= end
;
3693 register unsigned char *pto
= end
+ 3;
3695 while (pfrom
!= loc
)
3698 store_op1 (op
, loc
, arg
);
3702 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3705 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3707 register unsigned char *pfrom
= end
;
3708 register unsigned char *pto
= end
+ 5;
3710 while (pfrom
!= loc
)
3713 store_op2 (op
, loc
, arg1
, arg2
);
3717 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3718 after an alternative or a begin-subexpression. We assume there is at
3719 least one character before the ^. */
3722 at_begline_loc_p (const_re_char
*pattern
, const_re_char
*p
, reg_syntax_t syntax
)
3724 re_char
*prev
= p
- 2;
3725 boolean odd_backslashes
;
3727 /* After a subexpression? */
3729 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3731 /* After an alternative? */
3732 else if (*prev
== '|')
3733 odd_backslashes
= (syntax
& RE_NO_BK_VBAR
) == 0;
3735 /* After a shy subexpression? */
3736 else if (*prev
== ':' && (syntax
& RE_SHY_GROUPS
))
3738 /* Skip over optional regnum. */
3739 while (prev
- 1 >= pattern
&& prev
[-1] >= '0' && prev
[-1] <= '9')
3742 if (!(prev
- 2 >= pattern
3743 && prev
[-1] == '?' && prev
[-2] == '('))
3746 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3751 /* Count the number of preceding backslashes. */
3753 while (prev
- 1 >= pattern
&& prev
[-1] == '\\')
3755 return (p
- prev
) & odd_backslashes
;
3759 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3760 at least one character after the $, i.e., `P < PEND'. */
3763 at_endline_loc_p (const_re_char
*p
, const_re_char
*pend
, reg_syntax_t syntax
)
3766 boolean next_backslash
= *next
== '\\';
3767 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3770 /* Before a subexpression? */
3771 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3772 : next_backslash
&& next_next
&& *next_next
== ')')
3773 /* Before an alternative? */
3774 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3775 : next_backslash
&& next_next
&& *next_next
== '|');
3779 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3780 false if it's not. */
3783 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3785 ssize_t this_element
;
3787 for (this_element
= compile_stack
.avail
- 1;
3790 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3797 If fastmap is non-NULL, go through the pattern and fill fastmap
3798 with all the possible leading chars. If fastmap is NULL, don't
3799 bother filling it up (obviously) and only return whether the
3800 pattern could potentially match the empty string.
3802 Return 1 if p..pend might match the empty string.
3803 Return 0 if p..pend matches at least one char.
3804 Return -1 if fastmap was not updated accurately. */
3807 analyze_first (const_re_char
*p
, const_re_char
*pend
, char *fastmap
,
3808 const int multibyte
)
3813 /* If all elements for base leading-codes in fastmap is set, this
3814 flag is set true. */
3815 boolean match_any_multibyte_characters
= false;
3819 /* The loop below works as follows:
3820 - It has a working-list kept in the PATTERN_STACK and which basically
3821 starts by only containing a pointer to the first operation.
3822 - If the opcode we're looking at is a match against some set of
3823 chars, then we add those chars to the fastmap and go on to the
3824 next work element from the worklist (done via `break').
3825 - If the opcode is a control operator on the other hand, we either
3826 ignore it (if it's meaningless at this point, such as `start_memory')
3827 or execute it (if it's a jump). If the jump has several destinations
3828 (i.e. `on_failure_jump'), then we push the other destination onto the
3830 We guarantee termination by ignoring backward jumps (more or less),
3831 so that `p' is monotonically increasing. More to the point, we
3832 never set `p' (or push) anything `<= p1'. */
3836 /* `p1' is used as a marker of how far back a `on_failure_jump'
3837 can go without being ignored. It is normally equal to `p'
3838 (which prevents any backward `on_failure_jump') except right
3839 after a plain `jump', to allow patterns such as:
3842 10: on_failure_jump 3
3843 as used for the *? operator. */
3852 /* If the first character has to match a backreference, that means
3853 that the group was empty (since it already matched). Since this
3854 is the only case that interests us here, we can assume that the
3855 backreference must match the empty string. */
3860 /* Following are the cases which match a character. These end
3866 /* If multibyte is nonzero, the first byte of each
3867 character is an ASCII or a leading code. Otherwise,
3868 each byte is a character. Thus, this works in both
3873 /* For the case of matching this unibyte regex
3874 against multibyte, we must set a leading code of
3875 the corresponding multibyte character. */
3876 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
3878 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
3885 /* We could put all the chars except for \n (and maybe \0)
3886 but we don't bother since it is generally not worth it. */
3887 if (!fastmap
) break;
3892 if (!fastmap
) break;
3894 /* Chars beyond end of bitmap are possible matches. */
3895 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3896 j
< (1 << BYTEWIDTH
); j
++)
3902 if (!fastmap
) break;
3903 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3904 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3906 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3910 if (/* Any leading code can possibly start a character
3911 which doesn't match the specified set of characters. */
3914 /* If we can match a character class, we can match any
3915 multibyte characters. */
3916 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3917 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3920 if (match_any_multibyte_characters
== false)
3922 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3923 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3925 match_any_multibyte_characters
= true;
3929 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3930 && match_any_multibyte_characters
== false)
3932 /* Set fastmap[I] to 1 where I is a leading code of each
3933 multibyte character in the range table. */
3935 unsigned char lc1
, lc2
;
3937 /* Make P points the range table. `+ 2' is to skip flag
3938 bits for a character class. */
3939 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3941 /* Extract the number of ranges in range table into COUNT. */
3942 EXTRACT_NUMBER_AND_INCR (count
, p
);
3943 for (; count
> 0; count
--, p
+= 3)
3945 /* Extract the start and end of each range. */
3946 EXTRACT_CHARACTER (c
, p
);
3947 lc1
= CHAR_LEADING_CODE (c
);
3949 EXTRACT_CHARACTER (c
, p
);
3950 lc2
= CHAR_LEADING_CODE (c
);
3951 for (j
= lc1
; j
<= lc2
; j
++)
3960 if (!fastmap
) break;
3962 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
3964 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3965 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
3969 /* This match depends on text properties. These end with
3970 aborting optimizations. */
3974 case notcategoryspec
:
3975 if (!fastmap
) break;
3976 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
3978 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
3979 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
3982 /* Any leading code can possibly start a character which
3983 has or doesn't has the specified category. */
3984 if (match_any_multibyte_characters
== false)
3986 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3987 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3989 match_any_multibyte_characters
= true;
3993 /* All cases after this match the empty string. These end with
4013 EXTRACT_NUMBER_AND_INCR (j
, p
);
4015 /* Backward jumps can only go back to code that we've already
4016 visited. `re_compile' should make sure this is true. */
4021 case on_failure_jump
:
4022 case on_failure_keep_string_jump
:
4023 case on_failure_jump_loop
:
4024 case on_failure_jump_nastyloop
:
4025 case on_failure_jump_smart
:
4031 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4032 to jump back to "just after here". */
4035 case on_failure_jump
:
4036 case on_failure_keep_string_jump
:
4037 case on_failure_jump_nastyloop
:
4038 case on_failure_jump_loop
:
4039 case on_failure_jump_smart
:
4040 EXTRACT_NUMBER_AND_INCR (j
, p
);
4042 ; /* Backward jump to be ignored. */
4044 { /* We have to look down both arms.
4045 We first go down the "straight" path so as to minimize
4046 stack usage when going through alternatives. */
4047 int r
= analyze_first (p
, pend
, fastmap
, multibyte
);
4055 /* This code simply does not properly handle forward jump_n. */
4056 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4058 /* jump_n can either jump or fall through. The (backward) jump
4059 case has already been handled, so we only need to look at the
4060 fallthrough case. */
4064 /* If N == 0, it should be an on_failure_jump_loop instead. */
4065 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4067 /* We only care about one iteration of the loop, so we don't
4068 need to consider the case where this behaves like an
4085 abort (); /* We have listed all the cases. */
4088 /* Getting here means we have found the possible starting
4089 characters for one path of the pattern -- and that the empty
4090 string does not match. We need not follow this path further. */
4094 /* We reached the end without matching anything. */
4097 } /* analyze_first */
4099 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4100 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4101 characters can start a string that matches the pattern. This fastmap
4102 is used by re_search to skip quickly over impossible starting points.
4104 Character codes above (1 << BYTEWIDTH) are not represented in the
4105 fastmap, but the leading codes are represented. Thus, the fastmap
4106 indicates which character sets could start a match.
4108 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4109 area as BUFP->fastmap.
4111 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4114 Returns 0 if we succeed, -2 if an internal error. */
4117 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4119 char *fastmap
= bufp
->fastmap
;
4122 assert (fastmap
&& bufp
->buffer
);
4124 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4125 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4127 analysis
= analyze_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4128 fastmap
, RE_MULTIBYTE_P (bufp
));
4129 bufp
->can_be_null
= (analysis
!= 0);
4131 } /* re_compile_fastmap */
4133 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4134 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4135 this memory for recording register information. STARTS and ENDS
4136 must be allocated using the malloc library routine, and must each
4137 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4139 If NUM_REGS == 0, then subsequent matches should allocate their own
4142 Unless this function is called, the first search or match using
4143 PATTERN_BUFFER will allocate its own register data, without
4144 freeing the old data. */
4147 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4151 bufp
->regs_allocated
= REGS_REALLOCATE
;
4152 regs
->num_regs
= num_regs
;
4153 regs
->start
= starts
;
4158 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4160 regs
->start
= regs
->end
= 0;
4163 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4165 /* Searching routines. */
4167 /* Like re_search_2, below, but only one string is specified, and
4168 doesn't let you say where to stop matching. */
4171 re_search (struct re_pattern_buffer
*bufp
, const char *string
, size_t size
,
4172 ssize_t startpos
, ssize_t range
, struct re_registers
*regs
)
4174 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4177 WEAK_ALIAS (__re_search
, re_search
)
4179 /* Head address of virtual concatenation of string. */
4180 #define HEAD_ADDR_VSTRING(P) \
4181 (((P) >= size1 ? string2 : string1))
4183 /* Address of POS in the concatenation of virtual string. */
4184 #define POS_ADDR_VSTRING(POS) \
4185 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4187 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4188 virtual concatenation of STRING1 and STRING2, starting first at index
4189 STARTPOS, then at STARTPOS + 1, and so on.
4191 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4193 RANGE is how far to scan while trying to match. RANGE = 0 means try
4194 only at STARTPOS; in general, the last start tried is STARTPOS +
4197 In REGS, return the indices of the virtual concatenation of STRING1
4198 and STRING2 that matched the entire BUFP->buffer and its contained
4201 Do not consider matching one past the index STOP in the virtual
4202 concatenation of STRING1 and STRING2.
4204 We return either the position in the strings at which the match was
4205 found, -1 if no match, or -2 if error (such as failure
4209 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, size_t size1
,
4210 const char *str2
, size_t size2
, ssize_t startpos
, ssize_t range
,
4211 struct re_registers
*regs
, ssize_t stop
)
4214 re_char
*string1
= (re_char
*) str1
;
4215 re_char
*string2
= (re_char
*) str2
;
4216 register char *fastmap
= bufp
->fastmap
;
4217 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4218 size_t total_size
= size1
+ size2
;
4219 ssize_t endpos
= startpos
+ range
;
4220 boolean anchored_start
;
4221 /* Nonzero if we are searching multibyte string. */
4222 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4224 /* Check for out-of-range STARTPOS. */
4225 if (startpos
< 0 || startpos
> total_size
)
4228 /* Fix up RANGE if it might eventually take us outside
4229 the virtual concatenation of STRING1 and STRING2.
4230 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4232 range
= 0 - startpos
;
4233 else if (endpos
> total_size
)
4234 range
= total_size
- startpos
;
4236 /* If the search isn't to be a backwards one, don't waste time in a
4237 search for a pattern anchored at beginning of buffer. */
4238 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4247 /* In a forward search for something that starts with \=.
4248 don't keep searching past point. */
4249 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4251 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4257 /* Update the fastmap now if not correct already. */
4258 if (fastmap
&& !bufp
->fastmap_accurate
)
4259 re_compile_fastmap (bufp
);
4261 /* See whether the pattern is anchored. */
4262 anchored_start
= (bufp
->buffer
[0] == begline
);
4265 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4267 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4269 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4273 /* Loop through the string, looking for a place to start matching. */
4276 /* If the pattern is anchored,
4277 skip quickly past places we cannot match.
4278 We don't bother to treat startpos == 0 specially
4279 because that case doesn't repeat. */
4280 if (anchored_start
&& startpos
> 0)
4282 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4283 : string2
[startpos
- size1
- 1])
4288 /* If a fastmap is supplied, skip quickly over characters that
4289 cannot be the start of a match. If the pattern can match the
4290 null string, however, we don't need to skip characters; we want
4291 the first null string. */
4292 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4294 register re_char
*d
;
4295 register re_wchar_t buf_ch
;
4297 d
= POS_ADDR_VSTRING (startpos
);
4299 if (range
> 0) /* Searching forwards. */
4301 ssize_t irange
= range
, lim
= 0;
4303 if (startpos
< size1
&& startpos
+ range
>= size1
)
4304 lim
= range
- (size1
- startpos
);
4306 /* Written out as an if-else to avoid testing `translate'
4308 if (RE_TRANSLATE_P (translate
))
4315 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4316 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4317 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4320 range
-= buf_charlen
;
4326 register re_wchar_t ch
, translated
;
4329 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4330 translated
= RE_TRANSLATE (translate
, ch
);
4331 if (translated
!= ch
4332 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4334 if (fastmap
[buf_ch
])
4347 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4348 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4350 range
-= buf_charlen
;
4354 while (range
> lim
&& !fastmap
[*d
])
4360 startpos
+= irange
- range
;
4362 else /* Searching backwards. */
4366 buf_ch
= STRING_CHAR (d
);
4367 buf_ch
= TRANSLATE (buf_ch
);
4368 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4373 register re_wchar_t ch
, translated
;
4376 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4377 translated
= TRANSLATE (ch
);
4378 if (translated
!= ch
4379 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4381 if (! fastmap
[TRANSLATE (buf_ch
)])
4387 /* If can't match the null string, and that's all we have left, fail. */
4388 if (range
>= 0 && startpos
== total_size
&& fastmap
4389 && !bufp
->can_be_null
)
4392 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4393 startpos
, regs
, stop
);
4406 /* Update STARTPOS to the next character boundary. */
4409 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4410 int len
= BYTES_BY_CHAR_HEAD (*p
);
4428 /* Update STARTPOS to the previous character boundary. */
4431 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4433 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4435 /* Find the head of multibyte form. */
4436 PREV_CHAR_BOUNDARY (p
, phead
);
4437 range
+= p0
- 1 - p
;
4441 startpos
-= p0
- 1 - p
;
4447 WEAK_ALIAS (__re_search_2
, re_search_2
)
4449 /* Declarations and macros for re_match_2. */
4451 static int bcmp_translate (re_char
*s1
, re_char
*s2
,
4452 register ssize_t len
,
4453 RE_TRANSLATE_TYPE translate
,
4454 const int multibyte
);
4456 /* This converts PTR, a pointer into one of the search strings `string1'
4457 and `string2' into an offset from the beginning of that string. */
4458 #define POINTER_TO_OFFSET(ptr) \
4459 (FIRST_STRING_P (ptr) \
4461 : (ptr) - string2 + (ptrdiff_t) size1)
4463 /* Call before fetching a character with *d. This switches over to
4464 string2 if necessary.
4465 Check re_match_2_internal for a discussion of why end_match_2 might
4466 not be within string2 (but be equal to end_match_1 instead). */
4467 #define PREFETCH() \
4470 /* End of string2 => fail. */ \
4471 if (dend == end_match_2) \
4473 /* End of string1 => advance to string2. */ \
4475 dend = end_match_2; \
4478 /* Call before fetching a char with *d if you already checked other limits.
4479 This is meant for use in lookahead operations like wordend, etc..
4480 where we might need to look at parts of the string that might be
4481 outside of the LIMITs (i.e past `stop'). */
4482 #define PREFETCH_NOLIMIT() \
4486 dend = end_match_2; \
4489 /* Test if at very beginning or at very end of the virtual concatenation
4490 of `string1' and `string2'. If only one string, it's `string2'. */
4491 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4492 #define AT_STRINGS_END(d) ((d) == end2)
4494 /* Disabled due to a compiler bug -- see comment at case wordbound */
4496 /* The comment at case wordbound is following one, but we don't use
4497 AT_WORD_BOUNDARY anymore to support multibyte form.
4499 The DEC Alpha C compiler 3.x generates incorrect code for the
4500 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4501 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4502 macro and introducing temporary variables works around the bug. */
4505 /* Test if D points to a character which is word-constituent. We have
4506 two special cases to check for: if past the end of string1, look at
4507 the first character in string2; and if before the beginning of
4508 string2, look at the last character in string1. */
4509 #define WORDCHAR_P(d) \
4510 (SYNTAX ((d) == end1 ? *string2 \
4511 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4514 /* Test if the character before D and the one at D differ with respect
4515 to being word-constituent. */
4516 #define AT_WORD_BOUNDARY(d) \
4517 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4518 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4521 /* Free everything we malloc. */
4522 #ifdef MATCH_MAY_ALLOCATE
4523 # define FREE_VAR(var) \
4531 # define FREE_VARIABLES() \
4533 REGEX_FREE_STACK (fail_stack.stack); \
4534 FREE_VAR (regstart); \
4535 FREE_VAR (regend); \
4536 FREE_VAR (best_regstart); \
4537 FREE_VAR (best_regend); \
4538 REGEX_SAFE_FREE (); \
4541 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4542 #endif /* not MATCH_MAY_ALLOCATE */
4545 /* Optimization routines. */
4547 /* If the operation is a match against one or more chars,
4548 return a pointer to the next operation, else return NULL. */
4550 skip_one_char (const_re_char
*p
)
4563 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4566 p
= CHARSET_RANGE_TABLE (p
- 1);
4567 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4568 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4571 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4578 case notcategoryspec
:
4590 /* Jump over non-matching operations. */
4592 skip_noops (const_re_char
*p
, const_re_char
*pend
)
4606 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4617 /* Test if C matches charset op. *PP points to the charset or charset_not
4618 opcode. When the function finishes, *PP will be advanced past that opcode.
4619 C is character to test (possibly after translations) and CORIG is original
4620 character (i.e. without any translations). UNIBYTE denotes whether c is
4621 unibyte or multibyte character. */
4623 execute_charset (const_re_char
**pp
, unsigned c
, unsigned corig
, bool unibyte
)
4625 re_char
*p
= *pp
, *rtp
= NULL
;
4626 bool not = (re_opcode_t
) *p
== charset_not
;
4628 if (CHARSET_RANGE_TABLE_EXISTS_P (p
))
4631 rtp
= CHARSET_RANGE_TABLE (p
);
4632 EXTRACT_NUMBER_AND_INCR (count
, rtp
);
4633 *pp
= CHARSET_RANGE_TABLE_END ((rtp
), (count
));
4636 *pp
+= 2 + CHARSET_BITMAP_SIZE (p
);
4638 if (unibyte
&& c
< (1 << BYTEWIDTH
))
4639 { /* Lookup bitmap. */
4640 /* Cast to `unsigned' instead of `unsigned char' in
4641 case the bit list is a full 32 bytes long. */
4642 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (p
) * BYTEWIDTH
)
4643 && p
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4649 int class_bits
= CHARSET_RANGE_TABLE_BITS (p
);
4650 re_wchar_t range_start
, range_end
;
4652 /* Sort tests by the most commonly used classes with some adjustment to which
4653 tests are easiest to perform. Take a look at comment in re_wctype_parse
4654 for table with frequencies of character class names. */
4656 if ((class_bits
& BIT_MULTIBYTE
) ||
4657 (class_bits
& BIT_ALNUM
&& ISALNUM (c
)) ||
4658 (class_bits
& BIT_ALPHA
&& ISALPHA (c
)) ||
4659 (class_bits
& BIT_SPACE
&& ISSPACE (c
)) ||
4660 (class_bits
& BIT_WORD
&& ISWORD (c
)) ||
4661 ((class_bits
& BIT_UPPER
) &&
4662 (ISUPPER (c
) || (corig
!= c
&&
4663 c
== downcase (corig
) && ISLOWER (c
)))) ||
4664 ((class_bits
& BIT_LOWER
) &&
4665 (ISLOWER (c
) || (corig
!= c
&&
4666 c
== upcase (corig
) && ISUPPER(c
)))) ||
4667 (class_bits
& BIT_PUNCT
&& ISPUNCT (c
)) ||
4668 (class_bits
& BIT_GRAPH
&& ISGRAPH (c
)) ||
4669 (class_bits
& BIT_PRINT
&& ISPRINT (c
)))
4672 for (p
= *pp
; rtp
< p
; rtp
+= 2 * 3)
4674 EXTRACT_CHARACTER (range_start
, rtp
);
4675 EXTRACT_CHARACTER (range_end
, rtp
+ 3);
4676 if (range_start
<= c
&& c
<= range_end
)
4684 /* Non-zero if "p1 matches something" implies "p2 fails". */
4686 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, const_re_char
*p1
,
4690 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4691 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4693 assert (p1
>= bufp
->buffer
&& p1
< pend
4694 && p2
>= bufp
->buffer
&& p2
<= pend
);
4696 /* Skip over open/close-group commands.
4697 If what follows this loop is a ...+ construct,
4698 look at what begins its body, since we will have to
4699 match at least one of that. */
4700 p2
= skip_noops (p2
, pend
);
4701 /* The same skip can be done for p1, except that this function
4702 is only used in the case where p1 is a simple match operator. */
4703 /* p1 = skip_noops (p1, pend); */
4705 assert (p1
>= bufp
->buffer
&& p1
< pend
4706 && p2
>= bufp
->buffer
&& p2
<= pend
);
4708 op2
= p2
== pend
? succeed
: *p2
;
4714 /* If we're at the end of the pattern, we can change. */
4715 if (skip_one_char (p1
))
4717 DEBUG_PRINT (" End of pattern: fast loop.\n");
4725 register re_wchar_t c
4726 = (re_opcode_t
) *p2
== endline
? '\n'
4727 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4729 if ((re_opcode_t
) *p1
== exactn
)
4731 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4733 DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4738 else if ((re_opcode_t
) *p1
== charset
4739 || (re_opcode_t
) *p1
== charset_not
)
4741 if (!execute_charset (&p1
, c
, c
, !multibyte
|| IS_REAL_ASCII (c
)))
4743 DEBUG_PRINT (" No match => fast loop.\n");
4747 else if ((re_opcode_t
) *p1
== anychar
4750 DEBUG_PRINT (" . != \\n => fast loop.\n");
4758 if ((re_opcode_t
) *p1
== exactn
)
4759 /* Reuse the code above. */
4760 return mutually_exclusive_p (bufp
, p2
, p1
);
4762 /* It is hard to list up all the character in charset
4763 P2 if it includes multibyte character. Give up in
4765 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4767 /* Now, we are sure that P2 has no range table.
4768 So, for the size of bitmap in P2, `p2[1]' is
4769 enough. But P1 may have range table, so the
4770 size of bitmap table of P1 is extracted by
4771 using macro `CHARSET_BITMAP_SIZE'.
4773 In a multibyte case, we know that all the character
4774 listed in P2 is ASCII. In a unibyte case, P1 has only a
4775 bitmap table. So, in both cases, it is enough to test
4776 only the bitmap table of P1. */
4778 if ((re_opcode_t
) *p1
== charset
)
4781 /* We win if the charset inside the loop
4782 has no overlap with the one after the loop. */
4785 && idx
< CHARSET_BITMAP_SIZE (p1
));
4787 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4791 || idx
== CHARSET_BITMAP_SIZE (p1
))
4793 DEBUG_PRINT (" No match => fast loop.\n");
4797 else if ((re_opcode_t
) *p1
== charset_not
)
4800 /* We win if the charset_not inside the loop lists
4801 every character listed in the charset after. */
4802 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4803 if (! (p2
[2 + idx
] == 0
4804 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4805 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4810 DEBUG_PRINT (" No match => fast loop.\n");
4823 /* Reuse the code above. */
4824 return mutually_exclusive_p (bufp
, p2
, p1
);
4826 /* When we have two charset_not, it's very unlikely that
4827 they don't overlap. The union of the two sets of excluded
4828 chars should cover all possible chars, which, as a matter of
4829 fact, is virtually impossible in multibyte buffers. */
4835 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4837 return ((re_opcode_t
) *p1
== syntaxspec
4838 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4840 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4843 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4845 return ((re_opcode_t
) *p1
== notsyntaxspec
4846 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4848 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4851 return (((re_opcode_t
) *p1
== notsyntaxspec
4852 || (re_opcode_t
) *p1
== syntaxspec
)
4857 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4858 case notcategoryspec
:
4859 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4871 /* Matching routines. */
4873 #ifndef emacs /* Emacs never uses this. */
4874 /* re_match is like re_match_2 except it takes only a single string. */
4877 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
4878 size_t size
, ssize_t pos
, struct re_registers
*regs
)
4880 regoff_t result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
,
4881 size
, pos
, regs
, size
);
4884 WEAK_ALIAS (__re_match
, re_match
)
4885 #endif /* not emacs */
4887 /* re_match_2 matches the compiled pattern in BUFP against the
4888 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4889 and SIZE2, respectively). We start matching at POS, and stop
4892 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4893 store offsets for the substring each group matched in REGS. See the
4894 documentation for exactly how many groups we fill.
4896 We return -1 if no match, -2 if an internal error (such as the
4897 failure stack overflowing). Otherwise, we return the length of the
4898 matched substring. */
4901 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
,
4902 size_t size1
, const char *string2
, size_t size2
, ssize_t pos
,
4903 struct re_registers
*regs
, ssize_t stop
)
4909 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4910 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4911 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4914 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4915 (re_char
*) string2
, size2
,
4919 WEAK_ALIAS (__re_match_2
, re_match_2
)
4922 /* This is a separate function so that we can force an alloca cleanup
4925 re_match_2_internal (struct re_pattern_buffer
*bufp
, const_re_char
*string1
,
4926 size_t size1
, const_re_char
*string2
, size_t size2
,
4927 ssize_t pos
, struct re_registers
*regs
, ssize_t stop
)
4929 /* General temporaries. */
4933 /* Just past the end of the corresponding string. */
4934 re_char
*end1
, *end2
;
4936 /* Pointers into string1 and string2, just past the last characters in
4937 each to consider matching. */
4938 re_char
*end_match_1
, *end_match_2
;
4940 /* Where we are in the data, and the end of the current string. */
4943 /* Used sometimes to remember where we were before starting matching
4944 an operator so that we can go back in case of failure. This "atomic"
4945 behavior of matching opcodes is indispensable to the correctness
4946 of the on_failure_keep_string_jump optimization. */
4949 /* Where we are in the pattern, and the end of the pattern. */
4950 re_char
*p
= bufp
->buffer
;
4951 re_char
*pend
= p
+ bufp
->used
;
4953 /* We use this to map every character in the string. */
4954 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4956 /* Nonzero if BUFP is setup from a multibyte regex. */
4957 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4959 /* Nonzero if STRING1/STRING2 are multibyte. */
4960 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4962 /* Failure point stack. Each place that can handle a failure further
4963 down the line pushes a failure point on this stack. It consists of
4964 regstart, and regend for all registers corresponding to
4965 the subexpressions we're currently inside, plus the number of such
4966 registers, and, finally, two char *'s. The first char * is where
4967 to resume scanning the pattern; the second one is where to resume
4968 scanning the strings. */
4969 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4970 fail_stack_type fail_stack
;
4972 #ifdef DEBUG_COMPILES_ARGUMENTS
4973 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4976 #if defined REL_ALLOC && defined REGEX_MALLOC
4977 /* This holds the pointer to the failure stack, when
4978 it is allocated relocatably. */
4979 fail_stack_elt_t
*failure_stack_ptr
;
4982 /* We fill all the registers internally, independent of what we
4983 return, for use in backreferences. The number here includes
4984 an element for register zero. */
4985 size_t num_regs
= bufp
->re_nsub
+ 1;
4987 /* Information on the contents of registers. These are pointers into
4988 the input strings; they record just what was matched (on this
4989 attempt) by a subexpression part of the pattern, that is, the
4990 regnum-th regstart pointer points to where in the pattern we began
4991 matching and the regnum-th regend points to right after where we
4992 stopped matching the regnum-th subexpression. (The zeroth register
4993 keeps track of what the whole pattern matches.) */
4994 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4995 re_char
**regstart
, **regend
;
4998 /* The following record the register info as found in the above
4999 variables when we find a match better than any we've seen before.
5000 This happens as we backtrack through the failure points, which in
5001 turn happens only if we have not yet matched the entire string. */
5002 unsigned best_regs_set
= false;
5003 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5004 re_char
**best_regstart
, **best_regend
;
5007 /* Logically, this is `best_regend[0]'. But we don't want to have to
5008 allocate space for that if we're not allocating space for anything
5009 else (see below). Also, we never need info about register 0 for
5010 any of the other register vectors, and it seems rather a kludge to
5011 treat `best_regend' differently than the rest. So we keep track of
5012 the end of the best match so far in a separate variable. We
5013 initialize this to NULL so that when we backtrack the first time
5014 and need to test it, it's not garbage. */
5015 re_char
*match_end
= NULL
;
5017 #ifdef DEBUG_COMPILES_ARGUMENTS
5018 /* Counts the total number of registers pushed. */
5019 unsigned num_regs_pushed
= 0;
5022 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5024 REGEX_USE_SAFE_ALLOCA
;
5028 #ifdef MATCH_MAY_ALLOCATE
5029 /* Do not bother to initialize all the register variables if there are
5030 no groups in the pattern, as it takes a fair amount of time. If
5031 there are groups, we include space for register 0 (the whole
5032 pattern), even though we never use it, since it simplifies the
5033 array indexing. We should fix this. */
5036 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5037 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5038 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5039 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5041 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5049 /* We must initialize all our variables to NULL, so that
5050 `FREE_VARIABLES' doesn't try to free them. */
5051 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5053 #endif /* MATCH_MAY_ALLOCATE */
5055 /* The starting position is bogus. */
5056 if (pos
< 0 || pos
> size1
+ size2
)
5062 /* Initialize subexpression text positions to -1 to mark ones that no
5063 start_memory/stop_memory has been seen for. Also initialize the
5064 register information struct. */
5065 for (reg
= 1; reg
< num_regs
; reg
++)
5066 regstart
[reg
] = regend
[reg
] = NULL
;
5068 /* We move `string1' into `string2' if the latter's empty -- but not if
5069 `string1' is null. */
5070 if (size2
== 0 && string1
!= NULL
)
5077 end1
= string1
+ size1
;
5078 end2
= string2
+ size2
;
5080 /* `p' scans through the pattern as `d' scans through the data.
5081 `dend' is the end of the input string that `d' points within. `d'
5082 is advanced into the following input string whenever necessary, but
5083 this happens before fetching; therefore, at the beginning of the
5084 loop, `d' can be pointing at the end of a string, but it cannot
5088 /* Only match within string2. */
5089 d
= string2
+ pos
- size1
;
5090 dend
= end_match_2
= string2
+ stop
- size1
;
5091 end_match_1
= end1
; /* Just to give it a value. */
5097 /* Only match within string1. */
5098 end_match_1
= string1
+ stop
;
5100 When we reach end_match_1, PREFETCH normally switches to string2.
5101 But in the present case, this means that just doing a PREFETCH
5102 makes us jump from `stop' to `gap' within the string.
5103 What we really want here is for the search to stop as
5104 soon as we hit end_match_1. That's why we set end_match_2
5105 to end_match_1 (since PREFETCH fails as soon as we hit
5107 end_match_2
= end_match_1
;
5110 { /* It's important to use this code when stop == size so that
5111 moving `d' from end1 to string2 will not prevent the d == dend
5112 check from catching the end of string. */
5114 end_match_2
= string2
+ stop
- size1
;
5120 DEBUG_PRINT ("The compiled pattern is: ");
5121 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5122 DEBUG_PRINT ("The string to match is: \"");
5123 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5124 DEBUG_PRINT ("\"\n");
5126 /* This loops over pattern commands. It exits by returning from the
5127 function if the match is complete, or it drops through if the match
5128 fails at this starting point in the input data. */
5131 DEBUG_PRINT ("\n%p: ", p
);
5135 /* End of pattern means we might have succeeded. */
5136 DEBUG_PRINT ("end of pattern ... ");
5138 /* If we haven't matched the entire string, and we want the
5139 longest match, try backtracking. */
5140 if (d
!= end_match_2
)
5142 /* True if this match is the best seen so far. */
5146 /* True if this match ends in the same string (string1
5147 or string2) as the best previous match. */
5148 bool same_str_p
= (FIRST_STRING_P (match_end
)
5149 == FIRST_STRING_P (d
));
5151 /* AIX compiler got confused when this was combined
5152 with the previous declaration. */
5154 best_match_p
= d
> match_end
;
5156 best_match_p
= !FIRST_STRING_P (d
);
5159 DEBUG_PRINT ("backtracking.\n");
5161 if (!FAIL_STACK_EMPTY ())
5162 { /* More failure points to try. */
5164 /* If exceeds best match so far, save it. */
5165 if (!best_regs_set
|| best_match_p
)
5167 best_regs_set
= true;
5170 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5172 for (reg
= 1; reg
< num_regs
; reg
++)
5174 best_regstart
[reg
] = regstart
[reg
];
5175 best_regend
[reg
] = regend
[reg
];
5181 /* If no failure points, don't restore garbage. And if
5182 last match is real best match, don't restore second
5184 else if (best_regs_set
&& !best_match_p
)
5187 /* Restore best match. It may happen that `dend ==
5188 end_match_1' while the restored d is in string2.
5189 For example, the pattern `x.*y.*z' against the
5190 strings `x-' and `y-z-', if the two strings are
5191 not consecutive in memory. */
5192 DEBUG_PRINT ("Restoring best registers.\n");
5195 dend
= ((d
>= string1
&& d
<= end1
)
5196 ? end_match_1
: end_match_2
);
5198 for (reg
= 1; reg
< num_regs
; reg
++)
5200 regstart
[reg
] = best_regstart
[reg
];
5201 regend
[reg
] = best_regend
[reg
];
5204 } /* d != end_match_2 */
5207 DEBUG_PRINT ("Accepting match.\n");
5209 /* If caller wants register contents data back, do it. */
5210 if (regs
&& !bufp
->no_sub
)
5212 /* Have the register data arrays been allocated? */
5213 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5214 { /* No. So allocate them with malloc. We need one
5215 extra element beyond `num_regs' for the `-1' marker
5217 regs
->num_regs
= max (RE_NREGS
, num_regs
+ 1);
5218 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5219 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5220 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5225 bufp
->regs_allocated
= REGS_REALLOCATE
;
5227 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5228 { /* Yes. If we need more elements than were already
5229 allocated, reallocate them. If we need fewer, just
5231 if (regs
->num_regs
< num_regs
+ 1)
5233 regs
->num_regs
= num_regs
+ 1;
5234 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5235 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5236 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5245 /* These braces fend off a "empty body in an else-statement"
5246 warning under GCC when assert expands to nothing. */
5247 assert (bufp
->regs_allocated
== REGS_FIXED
);
5250 /* Convert the pointer data in `regstart' and `regend' to
5251 indices. Register zero has to be set differently,
5252 since we haven't kept track of any info for it. */
5253 if (regs
->num_regs
> 0)
5255 regs
->start
[0] = pos
;
5256 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5259 /* Go through the first `min (num_regs, regs->num_regs)'
5260 registers, since that is all we initialized. */
5261 for (reg
= 1; reg
< min (num_regs
, regs
->num_regs
); reg
++)
5263 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5264 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5267 regs
->start
[reg
] = POINTER_TO_OFFSET (regstart
[reg
]);
5268 regs
->end
[reg
] = POINTER_TO_OFFSET (regend
[reg
]);
5272 /* If the regs structure we return has more elements than
5273 were in the pattern, set the extra elements to -1. If
5274 we (re)allocated the registers, this is the case,
5275 because we always allocate enough to have at least one
5277 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5278 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5279 } /* regs && !bufp->no_sub */
5281 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5282 nfailure_points_pushed
, nfailure_points_popped
,
5283 nfailure_points_pushed
- nfailure_points_popped
);
5284 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed
);
5286 ptrdiff_t dcnt
= POINTER_TO_OFFSET (d
) - pos
;
5288 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt
);
5294 /* Otherwise match next pattern command. */
5297 /* Ignore these. Used to ignore the n of succeed_n's which
5298 currently have n == 0. */
5300 DEBUG_PRINT ("EXECUTING no_op.\n");
5304 DEBUG_PRINT ("EXECUTING succeed.\n");
5307 /* Match the next n pattern characters exactly. The following
5308 byte in the pattern defines n, and the n bytes after that
5309 are the characters to match. */
5312 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt
);
5314 /* Remember the start point to rollback upon failure. */
5318 /* This is written out as an if-else so we don't waste time
5319 testing `translate' inside the loop. */
5320 if (RE_TRANSLATE_P (translate
))
5324 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5344 /* The cost of testing `translate' is comparatively small. */
5345 if (target_multibyte
)
5348 int pat_charlen
, buf_charlen
;
5353 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5356 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5359 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5361 if (TRANSLATE (buf_ch
) != pat_ch
)
5369 mcnt
-= pat_charlen
;
5381 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5382 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5389 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5390 if (! CHAR_BYTE8_P (buf_ch
))
5392 buf_ch
= TRANSLATE (buf_ch
);
5393 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5399 if (buf_ch
!= pat_ch
)
5412 /* Match any character except possibly a newline or a null. */
5417 reg_syntax_t syntax
;
5419 DEBUG_PRINT ("EXECUTING anychar.\n");
5422 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5424 buf_ch
= TRANSLATE (buf_ch
);
5427 syntax
= RE_SYNTAX_EMACS
;
5429 syntax
= bufp
->syntax
;
5432 if ((!(syntax
& RE_DOT_NEWLINE
) && buf_ch
== '\n')
5433 || ((syntax
& RE_DOT_NOT_NULL
) && buf_ch
== '\000'))
5436 DEBUG_PRINT (" Matched \"%d\".\n", *d
);
5445 register unsigned int c
, corig
;
5448 /* Whether matching against a unibyte character. */
5449 boolean unibyte_char
= false;
5451 DEBUG_PRINT ("EXECUTING charset%s.\n",
5452 (re_opcode_t
) *(p
- 1) == charset_not
? "_not" : "");
5455 corig
= c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5456 if (target_multibyte
)
5461 c1
= RE_CHAR_TO_UNIBYTE (c
);
5464 unibyte_char
= true;
5470 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5472 if (! CHAR_BYTE8_P (c1
))
5474 c1
= TRANSLATE (c1
);
5475 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5478 unibyte_char
= true;
5483 unibyte_char
= true;
5487 if (!execute_charset (&p
, c
, corig
, unibyte_char
))
5495 /* The beginning of a group is represented by start_memory.
5496 The argument is the register number. The text
5497 matched within the group is recorded (in the internal
5498 registers data structure) under the register number. */
5500 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p
);
5502 /* In case we need to undo this operation (via backtracking). */
5503 PUSH_FAILURE_REG (*p
);
5506 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5507 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart
[*p
]));
5509 /* Move past the register number and inner group count. */
5514 /* The stop_memory opcode represents the end of a group. Its
5515 argument is the same as start_memory's: the register number. */
5517 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p
);
5519 assert (!REG_UNSET (regstart
[*p
]));
5520 /* Strictly speaking, there should be code such as:
5522 assert (REG_UNSET (regend[*p]));
5523 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5525 But the only info to be pushed is regend[*p] and it is known to
5526 be UNSET, so there really isn't anything to push.
5527 Not pushing anything, on the other hand deprives us from the
5528 guarantee that regend[*p] is UNSET since undoing this operation
5529 will not reset its value properly. This is not important since
5530 the value will only be read on the next start_memory or at
5531 the very end and both events can only happen if this stop_memory
5535 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend
[*p
]));
5537 /* Move past the register number and the inner group count. */
5542 /* \<digit> has been turned into a `duplicate' command which is
5543 followed by the numeric value of <digit> as the register number. */
5546 register re_char
*d2
, *dend2
;
5547 int regno
= *p
++; /* Get which register to match against. */
5548 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno
);
5550 /* Can't back reference a group which we've never matched. */
5551 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5554 /* Where in input to try to start matching. */
5555 d2
= regstart
[regno
];
5557 /* Remember the start point to rollback upon failure. */
5560 /* Where to stop matching; if both the place to start and
5561 the place to stop matching are in the same string, then
5562 set to the place to stop, otherwise, for now have to use
5563 the end of the first string. */
5565 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5566 == FIRST_STRING_P (regend
[regno
]))
5567 ? regend
[regno
] : end_match_1
);
5572 /* If necessary, advance to next segment in register
5576 if (dend2
== end_match_2
) break;
5577 if (dend2
== regend
[regno
]) break;
5579 /* End of string1 => advance to string2. */
5581 dend2
= regend
[regno
];
5583 /* At end of register contents => success */
5584 if (d2
== dend2
) break;
5586 /* If necessary, advance to next segment in data. */
5589 /* How many characters left in this segment to match. */
5592 /* Want how many consecutive characters we can match in
5593 one shot, so, if necessary, adjust the count. */
5594 if (dcnt
> dend2
- d2
)
5597 /* Compare that many; failure if mismatch, else move
5599 if (RE_TRANSLATE_P (translate
)
5600 ? bcmp_translate (d
, d2
, dcnt
, translate
, target_multibyte
)
5601 : memcmp (d
, d2
, dcnt
))
5606 d
+= dcnt
, d2
+= dcnt
;
5612 /* begline matches the empty string at the beginning of the string
5613 (unless `not_bol' is set in `bufp'), and after newlines. */
5615 DEBUG_PRINT ("EXECUTING begline.\n");
5617 if (AT_STRINGS_BEG (d
))
5619 if (!bufp
->not_bol
) break;
5624 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5628 /* In all other cases, we fail. */
5632 /* endline is the dual of begline. */
5634 DEBUG_PRINT ("EXECUTING endline.\n");
5636 if (AT_STRINGS_END (d
))
5638 if (!bufp
->not_eol
) break;
5642 PREFETCH_NOLIMIT ();
5649 /* Match at the very beginning of the data. */
5651 DEBUG_PRINT ("EXECUTING begbuf.\n");
5652 if (AT_STRINGS_BEG (d
))
5657 /* Match at the very end of the data. */
5659 DEBUG_PRINT ("EXECUTING endbuf.\n");
5660 if (AT_STRINGS_END (d
))
5665 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5666 pushes NULL as the value for the string on the stack. Then
5667 `POP_FAILURE_POINT' will keep the current value for the
5668 string, instead of restoring it. To see why, consider
5669 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5670 then the . fails against the \n. But the next thing we want
5671 to do is match the \n against the \n; if we restored the
5672 string value, we would be back at the foo.
5674 Because this is used only in specific cases, we don't need to
5675 check all the things that `on_failure_jump' does, to make
5676 sure the right things get saved on the stack. Hence we don't
5677 share its code. The only reason to push anything on the
5678 stack at all is that otherwise we would have to change
5679 `anychar's code to do something besides goto fail in this
5680 case; that seems worse than this. */
5681 case on_failure_keep_string_jump
:
5682 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5683 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5686 PUSH_FAILURE_POINT (p
- 3, NULL
);
5689 /* A nasty loop is introduced by the non-greedy *? and +?.
5690 With such loops, the stack only ever contains one failure point
5691 at a time, so that a plain on_failure_jump_loop kind of
5692 cycle detection cannot work. Worse yet, such a detection
5693 can not only fail to detect a cycle, but it can also wrongly
5694 detect a cycle (between different instantiations of the same
5696 So the method used for those nasty loops is a little different:
5697 We use a special cycle-detection-stack-frame which is pushed
5698 when the on_failure_jump_nastyloop failure-point is *popped*.
5699 This special frame thus marks the beginning of one iteration
5700 through the loop and we can hence easily check right here
5701 whether something matched between the beginning and the end of
5703 case on_failure_jump_nastyloop
:
5704 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5705 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5708 assert ((re_opcode_t
)p
[-4] == no_op
);
5711 CHECK_INFINITE_LOOP (p
- 4, d
);
5713 /* If there's a cycle, just continue without pushing
5714 this failure point. The failure point is the "try again"
5715 option, which shouldn't be tried.
5716 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5717 PUSH_FAILURE_POINT (p
- 3, d
);
5721 /* Simple loop detecting on_failure_jump: just check on the
5722 failure stack if the same spot was already hit earlier. */
5723 case on_failure_jump_loop
:
5725 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5726 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5730 CHECK_INFINITE_LOOP (p
- 3, d
);
5732 /* If there's a cycle, get out of the loop, as if the matching
5733 had failed. We used to just `goto fail' here, but that was
5734 aborting the search a bit too early: we want to keep the
5735 empty-loop-match and keep matching after the loop.
5736 We want (x?)*y\1z to match both xxyz and xxyxz. */
5739 PUSH_FAILURE_POINT (p
- 3, d
);
5744 /* Uses of on_failure_jump:
5746 Each alternative starts with an on_failure_jump that points
5747 to the beginning of the next alternative. Each alternative
5748 except the last ends with a jump that in effect jumps past
5749 the rest of the alternatives. (They really jump to the
5750 ending jump of the following alternative, because tensioning
5751 these jumps is a hassle.)
5753 Repeats start with an on_failure_jump that points past both
5754 the repetition text and either the following jump or
5755 pop_failure_jump back to this on_failure_jump. */
5756 case on_failure_jump
:
5757 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5758 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5761 PUSH_FAILURE_POINT (p
-3, d
);
5764 /* This operation is used for greedy *.
5765 Compare the beginning of the repeat with what in the
5766 pattern follows its end. If we can establish that there
5767 is nothing that they would both match, i.e., that we
5768 would have to backtrack because of (as in, e.g., `a*a')
5769 then we can use a non-backtracking loop based on
5770 on_failure_keep_string_jump instead of on_failure_jump. */
5771 case on_failure_jump_smart
:
5772 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5773 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5776 re_char
*p1
= p
; /* Next operation. */
5777 /* Here, we discard `const', making re_match non-reentrant. */
5778 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5779 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5781 p
-= 3; /* Reset so that we will re-execute the
5782 instruction once it's been changed. */
5784 EXTRACT_NUMBER (mcnt
, p2
- 2);
5786 /* Ensure this is a indeed the trivial kind of loop
5787 we are expecting. */
5788 assert (skip_one_char (p1
) == p2
- 3);
5789 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5790 DEBUG_STATEMENT (debug
+= 2);
5791 if (mutually_exclusive_p (bufp
, p1
, p2
))
5793 /* Use a fast `on_failure_keep_string_jump' loop. */
5794 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5795 *p3
= (unsigned char) on_failure_keep_string_jump
;
5796 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5800 /* Default to a safe `on_failure_jump' loop. */
5801 DEBUG_PRINT (" smart default => slow loop.\n");
5802 *p3
= (unsigned char) on_failure_jump
;
5804 DEBUG_STATEMENT (debug
-= 2);
5808 /* Unconditionally jump (without popping any failure points). */
5811 IMMEDIATE_QUIT_CHECK
;
5812 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5813 DEBUG_PRINT ("EXECUTING jump %d ", mcnt
);
5814 p
+= mcnt
; /* Do the jump. */
5815 DEBUG_PRINT ("(to %p).\n", p
);
5819 /* Have to succeed matching what follows at least n times.
5820 After that, handle like `on_failure_jump'. */
5822 /* Signedness doesn't matter since we only compare MCNT to 0. */
5823 EXTRACT_NUMBER (mcnt
, p
+ 2);
5824 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt
);
5826 /* Originally, mcnt is how many times we HAVE to succeed. */
5829 /* Here, we discard `const', making re_match non-reentrant. */
5830 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5833 PUSH_NUMBER (p2
, mcnt
);
5836 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5841 /* Signedness doesn't matter since we only compare MCNT to 0. */
5842 EXTRACT_NUMBER (mcnt
, p
+ 2);
5843 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt
);
5845 /* Originally, this is how many times we CAN jump. */
5848 /* Here, we discard `const', making re_match non-reentrant. */
5849 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5851 PUSH_NUMBER (p2
, mcnt
);
5852 goto unconditional_jump
;
5854 /* If don't have to jump any more, skip over the rest of command. */
5861 unsigned char *p2
; /* Location of the counter. */
5862 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5864 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5865 /* Here, we discard `const', making re_match non-reentrant. */
5866 p2
= (unsigned char*) p
+ mcnt
;
5867 /* Signedness doesn't matter since we only copy MCNT's bits. */
5868 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5869 DEBUG_PRINT (" Setting %p to %d.\n", p2
, mcnt
);
5870 PUSH_NUMBER (p2
, mcnt
);
5877 boolean
not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5878 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5880 /* We SUCCEED (or FAIL) in one of the following cases: */
5882 /* Case 1: D is at the beginning or the end of string. */
5883 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5887 /* C1 is the character before D, S1 is the syntax of C1, C2
5888 is the character at D, and S2 is the syntax of C2. */
5893 ssize_t offset
= PTR_TO_OFFSET (d
- 1);
5894 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5895 UPDATE_SYNTAX_TABLE_FAST (charpos
);
5897 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5900 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos
+ 1);
5902 PREFETCH_NOLIMIT ();
5903 GET_CHAR_AFTER (c2
, d
, dummy
);
5906 if (/* Case 2: Only one of S1 and S2 is Sword. */
5907 ((s1
== Sword
) != (s2
== Sword
))
5908 /* Case 3: Both of S1 and S2 are Sword, and macro
5909 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5910 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5920 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5922 /* We FAIL in one of the following cases: */
5924 /* Case 1: D is at the end of string. */
5925 if (AT_STRINGS_END (d
))
5929 /* C1 is the character before D, S1 is the syntax of C1, C2
5930 is the character at D, and S2 is the syntax of C2. */
5935 ssize_t offset
= PTR_TO_OFFSET (d
);
5936 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5937 UPDATE_SYNTAX_TABLE_FAST (charpos
);
5940 GET_CHAR_AFTER (c2
, d
, dummy
);
5943 /* Case 2: S2 is not Sword. */
5947 /* Case 3: D is not at the beginning of string ... */
5948 if (!AT_STRINGS_BEG (d
))
5950 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5952 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5956 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5958 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5965 DEBUG_PRINT ("EXECUTING wordend.\n");
5967 /* We FAIL in one of the following cases: */
5969 /* Case 1: D is at the beginning of string. */
5970 if (AT_STRINGS_BEG (d
))
5974 /* C1 is the character before D, S1 is the syntax of C1, C2
5975 is the character at D, and S2 is the syntax of C2. */
5980 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
5981 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5982 UPDATE_SYNTAX_TABLE_FAST (charpos
);
5984 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5987 /* Case 2: S1 is not Sword. */
5991 /* Case 3: D is not at the end of string ... */
5992 if (!AT_STRINGS_END (d
))
5994 PREFETCH_NOLIMIT ();
5995 GET_CHAR_AFTER (c2
, d
, dummy
);
5997 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos
);
6001 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6003 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6010 DEBUG_PRINT ("EXECUTING symbeg.\n");
6012 /* We FAIL in one of the following cases: */
6014 /* Case 1: D is at the end of string. */
6015 if (AT_STRINGS_END (d
))
6019 /* C1 is the character before D, S1 is the syntax of C1, C2
6020 is the character at D, and S2 is the syntax of C2. */
6024 ssize_t offset
= PTR_TO_OFFSET (d
);
6025 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6026 UPDATE_SYNTAX_TABLE_FAST (charpos
);
6029 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6032 /* Case 2: S2 is neither Sword nor Ssymbol. */
6033 if (s2
!= Sword
&& s2
!= Ssymbol
)
6036 /* Case 3: D is not at the beginning of string ... */
6037 if (!AT_STRINGS_BEG (d
))
6039 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6041 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6045 /* ... and S1 is Sword or Ssymbol. */
6046 if (s1
== Sword
|| s1
== Ssymbol
)
6053 DEBUG_PRINT ("EXECUTING symend.\n");
6055 /* We FAIL in one of the following cases: */
6057 /* Case 1: D is at the beginning of string. */
6058 if (AT_STRINGS_BEG (d
))
6062 /* C1 is the character before D, S1 is the syntax of C1, C2
6063 is the character at D, and S2 is the syntax of C2. */
6067 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6068 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6069 UPDATE_SYNTAX_TABLE_FAST (charpos
);
6071 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6074 /* Case 2: S1 is neither Ssymbol nor Sword. */
6075 if (s1
!= Sword
&& s1
!= Ssymbol
)
6078 /* Case 3: D is not at the end of string ... */
6079 if (!AT_STRINGS_END (d
))
6081 PREFETCH_NOLIMIT ();
6082 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6084 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos
+ 1);
6088 /* ... and S2 is Sword or Ssymbol. */
6089 if (s2
== Sword
|| s2
== Ssymbol
)
6098 boolean
not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6100 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6105 ssize_t offset
= PTR_TO_OFFSET (d
);
6106 ssize_t pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6107 UPDATE_SYNTAX_TABLE_FAST (pos1
);
6114 GET_CHAR_AFTER (c
, d
, len
);
6115 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6124 DEBUG_PRINT ("EXECUTING at_dot.\n");
6125 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6130 case notcategoryspec
:
6132 boolean
not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6134 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6135 not ? "not" : "", mcnt
);
6141 GET_CHAR_AFTER (c
, d
, len
);
6142 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6154 continue; /* Successfully executed one pattern command; keep going. */
6157 /* We goto here if a matching operation fails. */
6159 IMMEDIATE_QUIT_CHECK
;
6160 if (!FAIL_STACK_EMPTY ())
6163 /* A restart point is known. Restore to that state. */
6164 DEBUG_PRINT ("\nFAIL:\n");
6165 POP_FAILURE_POINT (str
, pat
);
6168 case on_failure_keep_string_jump
:
6169 assert (str
== NULL
);
6170 goto continue_failure_jump
;
6172 case on_failure_jump_nastyloop
:
6173 assert ((re_opcode_t
)pat
[-2] == no_op
);
6174 PUSH_FAILURE_POINT (pat
- 2, str
);
6177 case on_failure_jump_loop
:
6178 case on_failure_jump
:
6181 continue_failure_jump
:
6182 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6187 /* A special frame used for nastyloops. */
6194 assert (p
>= bufp
->buffer
&& p
<= pend
);
6196 if (d
>= string1
&& d
<= end1
)
6200 break; /* Matching at this starting point really fails. */
6204 goto restore_best_regs
;
6208 return -1; /* Failure to match. */
6211 /* Subroutine definitions for re_match_2. */
6213 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6214 bytes; nonzero otherwise. */
6217 bcmp_translate (const_re_char
*s1
, const_re_char
*s2
, register ssize_t len
,
6218 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6220 register re_char
*p1
= s1
, *p2
= s2
;
6221 re_char
*p1_end
= s1
+ len
;
6222 re_char
*p2_end
= s2
+ len
;
6224 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6225 different lengths, but relying on a single `len' would break this. -sm */
6226 while (p1
< p1_end
&& p2
< p2_end
)
6228 int p1_charlen
, p2_charlen
;
6229 re_wchar_t p1_ch
, p2_ch
;
6231 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6232 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6234 if (RE_TRANSLATE (translate
, p1_ch
)
6235 != RE_TRANSLATE (translate
, p2_ch
))
6238 p1
+= p1_charlen
, p2
+= p2_charlen
;
6241 if (p1
!= p1_end
|| p2
!= p2_end
)
6247 /* Entry points for GNU code. */
6249 /* re_compile_pattern is the GNU regular expression compiler: it
6250 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6251 Returns 0 if the pattern was valid, otherwise an error string.
6253 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6254 are set in BUFP on entry.
6256 We call regex_compile to do the actual compilation. */
6259 re_compile_pattern (const char *pattern
, size_t length
,
6261 bool posix_backtracking
, const char *whitespace_regexp
,
6263 struct re_pattern_buffer
*bufp
)
6267 /* GNU code is written to assume at least RE_NREGS registers will be set
6268 (and at least one extra will be -1). */
6269 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6271 /* And GNU code determines whether or not to get register information
6272 by passing null for the REGS argument to re_match, etc., not by
6276 ret
= regex_compile ((re_char
*) pattern
, length
,
6287 return gettext (re_error_msgid
[(int) ret
]);
6289 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6291 /* Entry points compatible with 4.2 BSD regex library. We don't define
6292 them unless specifically requested. */
6294 #if defined _REGEX_RE_COMP || defined _LIBC
6296 /* BSD has one and only one pattern buffer. */
6297 static struct re_pattern_buffer re_comp_buf
;
6301 /* Make these definitions weak in libc, so POSIX programs can redefine
6302 these names if they don't use our functions, and still use
6303 regcomp/regexec below without link errors. */
6306 re_comp (const char *s
)
6312 if (!re_comp_buf
.buffer
)
6313 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6314 return (char *) gettext ("No previous regular expression");
6318 if (!re_comp_buf
.buffer
)
6320 re_comp_buf
.buffer
= malloc (200);
6321 if (re_comp_buf
.buffer
== NULL
)
6322 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6323 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6324 re_comp_buf
.allocated
= 200;
6326 re_comp_buf
.fastmap
= malloc (1 << BYTEWIDTH
);
6327 if (re_comp_buf
.fastmap
== NULL
)
6328 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6329 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6332 /* Since `re_exec' always passes NULL for the `regs' argument, we
6333 don't need to initialize the pattern buffer fields which affect it. */
6335 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6340 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6341 return (char *) gettext (re_error_msgid
[(int) ret
]);
6349 re_exec (const char *s
)
6351 const size_t len
= strlen (s
);
6352 return re_search (&re_comp_buf
, s
, len
, 0, len
, 0) >= 0;
6354 #endif /* _REGEX_RE_COMP */
6356 /* POSIX.2 functions. Don't define these for Emacs. */
6360 /* regcomp takes a regular expression as a string and compiles it.
6362 PREG is a regex_t *. We do not expect any fields to be initialized,
6363 since POSIX says we shouldn't. Thus, we set
6365 `buffer' to the compiled pattern;
6366 `used' to the length of the compiled pattern;
6367 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6368 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6369 RE_SYNTAX_POSIX_BASIC;
6370 `fastmap' to an allocated space for the fastmap;
6371 `fastmap_accurate' to zero;
6372 `re_nsub' to the number of subexpressions in PATTERN.
6374 PATTERN is the address of the pattern string.
6376 CFLAGS is a series of bits which affect compilation.
6378 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6379 use POSIX basic syntax.
6381 If REG_NEWLINE is set, then . and [^...] don't match newline.
6382 Also, regexec will try a match beginning after every newline.
6384 If REG_ICASE is set, then we considers upper- and lowercase
6385 versions of letters to be equivalent when matching.
6387 If REG_NOSUB is set, then when PREG is passed to regexec, that
6388 routine will report only success or failure, and nothing about the
6391 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6392 the return codes and their meanings.) */
6395 regcomp (regex_t
*_Restrict_ preg
, const char *_Restrict_ pattern
,
6400 = (cflags
& REG_EXTENDED
) ?
6401 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6403 /* regex_compile will allocate the space for the compiled pattern. */
6405 preg
->allocated
= 0;
6408 /* Try to allocate space for the fastmap. */
6409 preg
->fastmap
= malloc (1 << BYTEWIDTH
);
6411 if (cflags
& REG_ICASE
)
6415 preg
->translate
= malloc (CHAR_SET_SIZE
* sizeof *preg
->translate
);
6416 if (preg
->translate
== NULL
)
6417 return (int) REG_ESPACE
;
6419 /* Map uppercase characters to corresponding lowercase ones. */
6420 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6421 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6424 preg
->translate
= NULL
;
6426 /* If REG_NEWLINE is set, newlines are treated differently. */
6427 if (cflags
& REG_NEWLINE
)
6428 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6429 syntax
&= ~RE_DOT_NEWLINE
;
6430 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6433 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6435 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6437 /* POSIX says a null character in the pattern terminates it, so we
6438 can use strlen here in compiling the pattern. */
6439 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6441 /* POSIX doesn't distinguish between an unmatched open-group and an
6442 unmatched close-group: both are REG_EPAREN. */
6443 if (ret
== REG_ERPAREN
)
6446 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6447 { /* Compute the fastmap now, since regexec cannot modify the pattern
6449 re_compile_fastmap (preg
);
6450 if (preg
->can_be_null
)
6451 { /* The fastmap can't be used anyway. */
6452 free (preg
->fastmap
);
6453 preg
->fastmap
= NULL
;
6458 WEAK_ALIAS (__regcomp
, regcomp
)
6461 /* regexec searches for a given pattern, specified by PREG, in the
6464 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6465 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6466 least NMATCH elements, and we set them to the offsets of the
6467 corresponding matched substrings.
6469 EFLAGS specifies `execution flags' which affect matching: if
6470 REG_NOTBOL is set, then ^ does not match at the beginning of the
6471 string; if REG_NOTEOL is set, then $ does not match at the end.
6473 We return 0 if we find a match and REG_NOMATCH if not. */
6476 regexec (const regex_t
*_Restrict_ preg
, const char *_Restrict_ string
,
6477 size_t nmatch
, regmatch_t pmatch
[_Restrict_arr_
], int eflags
)
6480 struct re_registers regs
;
6481 regex_t private_preg
;
6482 size_t len
= strlen (string
);
6483 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6485 private_preg
= *preg
;
6487 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6488 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6490 /* The user has told us exactly how many registers to return
6491 information about, via `nmatch'. We have to pass that on to the
6492 matching routines. */
6493 private_preg
.regs_allocated
= REGS_FIXED
;
6497 regs
.num_regs
= nmatch
;
6498 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6499 if (regs
.start
== NULL
)
6501 regs
.end
= regs
.start
+ nmatch
;
6504 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6505 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6506 was a little bit longer but still only matching the real part.
6507 This works because the `endline' will check for a '\n' and will find a
6508 '\0', correctly deciding that this is not the end of a line.
6509 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6510 a convenient '\0' there. For all we know, the string could be preceded
6511 by '\n' which would throw things off. */
6513 /* Perform the searching operation. */
6514 ret
= re_search (&private_preg
, string
, len
,
6515 /* start: */ 0, /* range: */ len
,
6516 want_reg_info
? ®s
: 0);
6518 /* Copy the register information to the POSIX structure. */
6525 for (r
= 0; r
< nmatch
; r
++)
6527 pmatch
[r
].rm_so
= regs
.start
[r
];
6528 pmatch
[r
].rm_eo
= regs
.end
[r
];
6532 /* If we needed the temporary register info, free the space now. */
6536 /* We want zero return to mean success, unlike `re_search'. */
6537 return ret
>= 0 ? REG_NOERROR
: REG_NOMATCH
;
6539 WEAK_ALIAS (__regexec
, regexec
)
6542 /* Returns a message corresponding to an error code, ERR_CODE, returned
6543 from either regcomp or regexec. We don't use PREG here.
6545 ERR_CODE was previously called ERRCODE, but that name causes an
6546 error with msvc8 compiler. */
6549 regerror (int err_code
, const regex_t
*preg
, char *errbuf
, size_t errbuf_size
)
6555 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6556 /* Only error codes returned by the rest of the code should be passed
6557 to this routine. If we are given anything else, or if other regex
6558 code generates an invalid error code, then the program has a bug.
6559 Dump core so we can fix it. */
6562 msg
= gettext (re_error_msgid
[err_code
]);
6564 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6566 if (errbuf_size
!= 0)
6568 if (msg_size
> errbuf_size
)
6570 memcpy (errbuf
, msg
, errbuf_size
- 1);
6571 errbuf
[errbuf_size
- 1] = 0;
6574 strcpy (errbuf
, msg
);
6579 WEAK_ALIAS (__regerror
, regerror
)
6582 /* Free dynamically allocated space used by PREG. */
6585 regfree (regex_t
*preg
)
6587 free (preg
->buffer
);
6588 preg
->buffer
= NULL
;
6590 preg
->allocated
= 0;
6593 free (preg
->fastmap
);
6594 preg
->fastmap
= NULL
;
6595 preg
->fastmap_accurate
= 0;
6597 free (preg
->translate
);
6598 preg
->translate
= NULL
;
6600 WEAK_ALIAS (__regfree
, regfree
)
6602 #endif /* not emacs */