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) 1993-2018 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 <https://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) \
159 ((p) + (NILP (gl_state.object) || BUFFERP (gl_state.object)))
161 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
162 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
163 # define RE_STRING_CHAR(p, multibyte) \
164 (multibyte ? (STRING_CHAR (p)) : (*(p)))
165 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
166 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
168 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
170 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
172 /* Set C a (possibly converted to multibyte) character before P. P
173 points into a string which is the virtual concatenation of STR1
174 (which ends at END1) or STR2 (which ends at END2). */
175 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
177 if (target_multibyte) \
179 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
180 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
181 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
182 c = STRING_CHAR (dtemp); \
186 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
187 (c) = RE_CHAR_TO_MULTIBYTE (c); \
191 /* Set C a (possibly converted to multibyte) character at P, and set
192 LEN to the byte length of that character. */
193 # define GET_CHAR_AFTER(c, p, len) \
195 if (target_multibyte) \
196 (c) = STRING_CHAR_AND_LENGTH (p, len); \
201 (c) = RE_CHAR_TO_MULTIBYTE (c); \
205 #else /* not emacs */
207 /* If we are not linking with Emacs proper,
208 we can't use the relocating allocator
209 even if config.h says that we can. */
214 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
216 static ATTRIBUTE_MALLOC
void *
217 xmalloc (size_t size
)
219 void *val
= malloc (size
);
222 write (STDERR_FILENO
, "virtual memory exhausted\n", 25);
229 xrealloc (void *block
, size_t size
)
232 /* We must call malloc explicitly when BLOCK is 0, since some
233 reallocs don't do this. */
237 val
= realloc (block
, size
);
240 write (STDERR_FILENO
, "virtual memory exhausted\n", 25);
249 # define malloc xmalloc
253 # define realloc xrealloc
255 # include <stdbool.h>
258 /* Define the syntax stuff for \<, \>, etc. */
260 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
261 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
263 /* Dummy macros for non-Emacs environments. */
264 # define MAX_MULTIBYTE_LENGTH 1
265 # define RE_MULTIBYTE_P(x) 0
266 # define RE_TARGET_MULTIBYTE_P(x) 0
267 # define WORD_BOUNDARY_P(c1, c2) (0)
268 # define BYTES_BY_CHAR_HEAD(p) (1)
269 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
270 # define STRING_CHAR(p) (*(p))
271 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
272 # define CHAR_STRING(c, s) (*(s) = (c), 1)
273 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
274 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
275 # define RE_CHAR_TO_MULTIBYTE(c) (c)
276 # define RE_CHAR_TO_UNIBYTE(c) (c)
277 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
278 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
279 # define GET_CHAR_AFTER(c, p, len) \
281 # define CHAR_BYTE8_P(c) (0)
282 # define CHAR_LEADING_CODE(c) (c)
284 #endif /* not emacs */
287 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
288 # define RE_TRANSLATE_P(TBL) (TBL)
291 /* Get the interface, including the syntax bits. */
294 /* isalpha etc. are used for the character classes. */
299 /* 1 if C is an ASCII character. */
300 # define IS_REAL_ASCII(c) ((c) < 0200)
302 /* 1 if C is a unibyte character. */
303 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
305 /* The Emacs definitions should not be directly affected by locales. */
307 /* In Emacs, these are only used for single-byte characters. */
308 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
309 # define ISCNTRL(c) ((c) < ' ')
310 # define ISXDIGIT(c) (0 <= char_hexdigit (c))
312 /* The rest must handle multibyte characters. */
314 # define ISBLANK(c) (IS_REAL_ASCII (c) \
315 ? ((c) == ' ' || (c) == '\t') \
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
434 in Emacs (note that SAFE_ALLOCA could also call malloc, but does so
435 via `record_xmalloc' which uses `unwind_protect' to ensure the
436 memory is freed even in case of non-local exits); also, malloc is
437 slower and causes storage fragmentation. On the other hand, malloc
438 is more portable, and easier to debug.
440 Because we sometimes use alloca, some routines have to be macros,
441 not functions -- `alloca'-allocated space disappears at the end of the
442 function it is called in. */
446 # define REGEX_ALLOCATE malloc
447 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
448 # define REGEX_FREE free
450 #else /* not REGEX_MALLOC */
453 /* This may be adjusted in main(), if the stack is successfully grown. */
454 ptrdiff_t emacs_re_safe_alloca
= MAX_ALLOCA
;
455 /* Like USE_SAFE_ALLOCA, but use emacs_re_safe_alloca. */
456 # define REGEX_USE_SAFE_ALLOCA \
457 ptrdiff_t sa_avail = emacs_re_safe_alloca; \
458 ptrdiff_t sa_count = SPECPDL_INDEX (); bool sa_must_free = false
460 # define REGEX_SAFE_FREE() SAFE_FREE ()
461 # define REGEX_ALLOCATE SAFE_ALLOCA
464 # define REGEX_ALLOCATE alloca
467 /* Assumes a `char *destination' variable. */
468 # define REGEX_REALLOCATE(source, osize, nsize) \
469 (destination = REGEX_ALLOCATE (nsize), \
470 memcpy (destination, source, osize))
472 /* No need to do anything to free, after alloca. */
473 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
475 #endif /* not REGEX_MALLOC */
477 #ifndef REGEX_USE_SAFE_ALLOCA
478 # define REGEX_USE_SAFE_ALLOCA ((void) 0)
479 # define REGEX_SAFE_FREE() ((void) 0)
482 /* Define how to allocate the failure stack. */
484 #if defined REL_ALLOC && defined REGEX_MALLOC
486 # define REGEX_ALLOCATE_STACK(size) \
487 r_alloc (&failure_stack_ptr, (size))
488 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
489 r_re_alloc (&failure_stack_ptr, (nsize))
490 # define REGEX_FREE_STACK(ptr) \
491 r_alloc_free (&failure_stack_ptr)
493 #else /* not using relocating allocator */
495 # define REGEX_ALLOCATE_STACK(size) REGEX_ALLOCATE (size)
496 # define REGEX_REALLOCATE_STACK(source, o, n) REGEX_REALLOCATE (source, o, n)
497 # define REGEX_FREE_STACK(ptr) REGEX_FREE (ptr)
499 #endif /* not using relocating allocator */
502 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
503 `string1' or just past its end. This works if PTR is NULL, which is
505 #define FIRST_STRING_P(ptr) \
506 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
508 /* (Re)Allocate N items of type T using malloc, or fail. */
509 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
510 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
511 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
513 #define BYTEWIDTH 8 /* In bits. */
518 # define max(a, b) ((a) > (b) ? (a) : (b))
519 # define min(a, b) ((a) < (b) ? (a) : (b))
522 /* Type of source-pattern and string chars. */
523 typedef const unsigned char re_char
;
525 typedef char boolean
;
527 static regoff_t
re_match_2_internal (struct re_pattern_buffer
*bufp
,
528 re_char
*string1
, size_t size1
,
529 re_char
*string2
, size_t size2
,
531 struct re_registers
*regs
,
534 /* These are the command codes that appear in compiled regular
535 expressions. Some opcodes are followed by argument bytes. A
536 command code can specify any interpretation whatsoever for its
537 arguments. Zero bytes may appear in the compiled regular expression. */
543 /* Succeed right away--no more backtracking. */
546 /* Followed by one byte giving n, then by n literal bytes. */
549 /* Matches any (more or less) character. */
552 /* Matches any one char belonging to specified set. First
553 following byte is number of bitmap bytes. Then come bytes
554 for a bitmap saying which chars are in. Bits in each byte
555 are ordered low-bit-first. A character is in the set if its
556 bit is 1. A character too large to have a bit in the map is
557 automatically not in the set.
559 If the length byte has the 0x80 bit set, then that stuff
560 is followed by a range table:
561 2 bytes of flags for character sets (low 8 bits, high 8 bits)
562 See RANGE_TABLE_WORK_BITS below.
563 2 bytes, the number of pairs that follow (upto 32767)
564 pairs, each 2 multibyte characters,
565 each multibyte character represented as 3 bytes. */
568 /* Same parameters as charset, but match any character that is
569 not one of those specified. */
572 /* Start remembering the text that is matched, for storing in a
573 register. Followed by one byte with the register number, in
574 the range 0 to one less than the pattern buffer's re_nsub
578 /* Stop remembering the text that is matched and store it in a
579 memory register. Followed by one byte with the register
580 number, in the range 0 to one less than `re_nsub' in the
584 /* Match a duplicate of something remembered. Followed by one
585 byte containing the register number. */
588 /* Fail unless at beginning of line. */
591 /* Fail unless at end of line. */
594 /* Succeeds if at beginning of buffer (if emacs) or at beginning
595 of string to be matched (if not). */
598 /* Analogously, for end of buffer/string. */
601 /* Followed by two byte relative address to which to jump. */
604 /* Followed by two-byte relative address of place to resume at
605 in case of failure. */
608 /* Like on_failure_jump, but pushes a placeholder instead of the
609 current string position when executed. */
610 on_failure_keep_string_jump
,
612 /* Just like `on_failure_jump', except that it checks that we
613 don't get stuck in an infinite loop (matching an empty string
615 on_failure_jump_loop
,
617 /* Just like `on_failure_jump_loop', except that it checks for
618 a different kind of loop (the kind that shows up with non-greedy
619 operators). This operation has to be immediately preceded
621 on_failure_jump_nastyloop
,
623 /* A smart `on_failure_jump' used for greedy * and + operators.
624 It analyzes the loop before which it is put and if the
625 loop does not require backtracking, it changes itself to
626 `on_failure_keep_string_jump' and short-circuits the loop,
627 else it just defaults to changing itself into `on_failure_jump'.
628 It assumes that it is pointing to just past a `jump'. */
629 on_failure_jump_smart
,
631 /* Followed by two-byte relative address and two-byte number n.
632 After matching N times, jump to the address upon failure.
633 Does not work if N starts at 0: use on_failure_jump_loop
637 /* Followed by two-byte relative address, and two-byte number n.
638 Jump to the address N times, then fail. */
641 /* Set the following two-byte relative address to the
642 subsequent two-byte number. The address *includes* the two
646 wordbeg
, /* Succeeds if at word beginning. */
647 wordend
, /* Succeeds if at word end. */
649 wordbound
, /* Succeeds if at a word boundary. */
650 notwordbound
, /* Succeeds if not at a word boundary. */
652 symbeg
, /* Succeeds if at symbol beginning. */
653 symend
, /* Succeeds if at symbol end. */
655 /* Matches any character whose syntax is specified. Followed by
656 a byte which contains a syntax code, e.g., Sword. */
659 /* Matches any character whose syntax is not that specified. */
663 , at_dot
, /* Succeeds if at point. */
665 /* Matches any character whose category-set contains the specified
666 category. The operator is followed by a byte which contains a
667 category code (mnemonic ASCII character). */
670 /* Matches any character whose category-set does not contain the
671 specified category. The operator is followed by a byte which
672 contains the category code (mnemonic ASCII character). */
677 /* Common operations on the compiled pattern. */
679 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
681 #define STORE_NUMBER(destination, number) \
683 (destination)[0] = (number) & 0377; \
684 (destination)[1] = (number) >> 8; \
687 /* Same as STORE_NUMBER, except increment DESTINATION to
688 the byte after where the number is stored. Therefore, DESTINATION
689 must be an lvalue. */
691 #define STORE_NUMBER_AND_INCR(destination, number) \
693 STORE_NUMBER (destination, number); \
694 (destination) += 2; \
697 /* Put into DESTINATION a number stored in two contiguous bytes starting
700 #define EXTRACT_NUMBER(destination, source) \
701 ((destination) = extract_number (source))
704 extract_number (re_char
*source
)
706 unsigned leading_byte
= SIGN_EXTEND_CHAR (source
[1]);
707 return (leading_byte
<< 8) + source
[0];
710 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
711 SOURCE must be an lvalue. */
713 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
714 ((destination) = extract_number_and_incr (&source))
717 extract_number_and_incr (re_char
**source
)
719 int num
= extract_number (*source
);
724 /* Store a multibyte character in three contiguous bytes starting
725 DESTINATION, and increment DESTINATION to the byte after where the
726 character is stored. Therefore, DESTINATION must be an lvalue. */
728 #define STORE_CHARACTER_AND_INCR(destination, character) \
730 (destination)[0] = (character) & 0377; \
731 (destination)[1] = ((character) >> 8) & 0377; \
732 (destination)[2] = (character) >> 16; \
733 (destination) += 3; \
736 /* Put into DESTINATION a character stored in three contiguous bytes
737 starting at SOURCE. */
739 #define EXTRACT_CHARACTER(destination, source) \
741 (destination) = ((source)[0] \
742 | ((source)[1] << 8) \
743 | ((source)[2] << 16)); \
747 /* Macros for charset. */
749 /* Size of bitmap of charset P in bytes. P is a start of charset,
750 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
751 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
753 /* Nonzero if charset P has range table. */
754 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
756 /* Return the address of range table of charset P. But not the start
757 of table itself, but the before where the number of ranges is
758 stored. `2 +' means to skip re_opcode_t and size of bitmap,
759 and the 2 bytes of flags at the start of the range table. */
760 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
763 /* Extract the bit flags that start a range table. */
764 #define CHARSET_RANGE_TABLE_BITS(p) \
765 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
766 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
769 /* Return the address of end of RANGE_TABLE. COUNT is number of
770 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
771 is start of range and end of range. `* 3' is size of each start
773 #define CHARSET_RANGE_TABLE_END(range_table, count) \
774 ((range_table) + (count) * 2 * 3)
776 /* If DEBUG is defined, Regex prints many voluminous messages about what
777 it is doing (if the variable `debug' is nonzero). If linked with the
778 main program in `iregex.c', you can enter patterns and strings
779 interactively. And if linked with the main program in `main.c' and
780 the other test files, you can run the already-written tests. */
784 /* We use standard I/O for debugging. */
787 /* It is useful to test things that ``must'' be true when debugging. */
790 static int debug
= -100000;
792 # define DEBUG_STATEMENT(e) e
793 # define DEBUG_PRINT(...) if (debug > 0) printf (__VA_ARGS__)
794 # define DEBUG_COMPILES_ARGUMENTS
795 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
796 if (debug > 0) print_partial_compiled_pattern (s, e)
797 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
798 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
801 /* Print the fastmap in human-readable form. */
804 print_fastmap (char *fastmap
)
806 unsigned was_a_range
= 0;
809 while (i
< (1 << BYTEWIDTH
))
815 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
831 /* Print a compiled pattern string in human-readable form, starting at
832 the START pointer into it and ending just before the pointer END. */
835 print_partial_compiled_pattern (re_char
*start
, re_char
*end
)
843 fprintf (stderr
, "(null)\n");
847 /* Loop over pattern commands. */
850 fprintf (stderr
, "%td:\t", p
- start
);
852 switch ((re_opcode_t
) *p
++)
855 fprintf (stderr
, "/no_op");
859 fprintf (stderr
, "/succeed");
864 fprintf (stderr
, "/exactn/%d", mcnt
);
867 fprintf (stderr
, "/%c", *p
++);
873 fprintf (stderr
, "/start_memory/%d", *p
++);
877 fprintf (stderr
, "/stop_memory/%d", *p
++);
881 fprintf (stderr
, "/duplicate/%d", *p
++);
885 fprintf (stderr
, "/anychar");
891 register int c
, last
= -100;
892 register int in_range
= 0;
893 int length
= CHARSET_BITMAP_SIZE (p
- 1);
894 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
896 fprintf (stderr
, "/charset [%s",
897 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
900 fprintf (stderr
, " !extends past end of pattern! ");
902 for (c
= 0; c
< 256; c
++)
904 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
906 /* Are we starting a range? */
907 if (last
+ 1 == c
&& ! in_range
)
909 fprintf (stderr
, "-");
912 /* Have we broken a range? */
913 else if (last
+ 1 != c
&& in_range
)
915 fprintf (stderr
, "%c", last
);
920 fprintf (stderr
, "%c", c
);
926 fprintf (stderr
, "%c", last
);
928 fprintf (stderr
, "]");
935 fprintf (stderr
, "has-range-table");
937 /* ??? Should print the range table; for now, just skip it. */
938 p
+= 2; /* skip range table bits */
939 EXTRACT_NUMBER_AND_INCR (count
, p
);
940 p
= CHARSET_RANGE_TABLE_END (p
, count
);
946 fprintf (stderr
, "/begline");
950 fprintf (stderr
, "/endline");
953 case on_failure_jump
:
954 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
955 fprintf (stderr
, "/on_failure_jump to %td", p
+ mcnt
- start
);
958 case on_failure_keep_string_jump
:
959 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
960 fprintf (stderr
, "/on_failure_keep_string_jump to %td",
964 case on_failure_jump_nastyloop
:
965 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
966 fprintf (stderr
, "/on_failure_jump_nastyloop to %td",
970 case on_failure_jump_loop
:
971 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
972 fprintf (stderr
, "/on_failure_jump_loop to %td",
976 case on_failure_jump_smart
:
977 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
978 fprintf (stderr
, "/on_failure_jump_smart to %td",
983 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
984 fprintf (stderr
, "/jump to %td", p
+ mcnt
- start
);
988 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
989 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
990 fprintf (stderr
, "/succeed_n to %td, %d times",
991 p
- 2 + mcnt
- start
, mcnt2
);
995 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
996 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
997 fprintf (stderr
, "/jump_n to %td, %d times",
998 p
- 2 + mcnt
- start
, mcnt2
);
1002 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1003 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1004 fprintf (stderr
, "/set_number_at location %td to %d",
1005 p
- 2 + mcnt
- start
, mcnt2
);
1009 fprintf (stderr
, "/wordbound");
1013 fprintf (stderr
, "/notwordbound");
1017 fprintf (stderr
, "/wordbeg");
1021 fprintf (stderr
, "/wordend");
1025 fprintf (stderr
, "/symbeg");
1029 fprintf (stderr
, "/symend");
1033 fprintf (stderr
, "/syntaxspec");
1035 fprintf (stderr
, "/%d", mcnt
);
1039 fprintf (stderr
, "/notsyntaxspec");
1041 fprintf (stderr
, "/%d", mcnt
);
1046 fprintf (stderr
, "/at_dot");
1050 fprintf (stderr
, "/categoryspec");
1052 fprintf (stderr
, "/%d", mcnt
);
1055 case notcategoryspec
:
1056 fprintf (stderr
, "/notcategoryspec");
1058 fprintf (stderr
, "/%d", mcnt
);
1063 fprintf (stderr
, "/begbuf");
1067 fprintf (stderr
, "/endbuf");
1071 fprintf (stderr
, "?%d", *(p
-1));
1074 fprintf (stderr
, "\n");
1077 fprintf (stderr
, "%td:\tend of pattern.\n", p
- start
);
1082 print_compiled_pattern (struct re_pattern_buffer
*bufp
)
1084 re_char
*buffer
= bufp
->buffer
;
1086 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1087 printf ("%ld bytes used/%ld bytes allocated.\n",
1088 bufp
->used
, bufp
->allocated
);
1090 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1092 printf ("fastmap: ");
1093 print_fastmap (bufp
->fastmap
);
1096 printf ("re_nsub: %zu\t", bufp
->re_nsub
);
1097 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1098 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1099 printf ("no_sub: %d\t", bufp
->no_sub
);
1100 printf ("not_bol: %d\t", bufp
->not_bol
);
1101 printf ("not_eol: %d\t", bufp
->not_eol
);
1103 printf ("syntax: %lx\n", bufp
->syntax
);
1106 /* Perhaps we should print the translate table? */
1111 print_double_string (re_char
*where
, re_char
*string1
, ssize_t size1
,
1112 re_char
*string2
, ssize_t size2
)
1120 if (FIRST_STRING_P (where
))
1122 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1123 putchar (string1
[this_char
]);
1128 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1129 putchar (string2
[this_char
]);
1133 #else /* not DEBUG */
1138 # define DEBUG_STATEMENT(e)
1139 # define DEBUG_PRINT(...)
1140 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1141 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1143 #endif /* not DEBUG */
1147 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1148 also be assigned to arbitrarily: each pattern buffer stores its own
1149 syntax, so it can be changed between regex compilations. */
1150 /* This has no initializer because initialized variables in Emacs
1151 become read-only after dumping. */
1152 reg_syntax_t re_syntax_options
;
1155 /* Specify the precise syntax of regexps for compilation. This provides
1156 for compatibility for various utilities which historically have
1157 different, incompatible syntaxes.
1159 The argument SYNTAX is a bit mask comprised of the various bits
1160 defined in regex.h. We return the old syntax. */
1163 re_set_syntax (reg_syntax_t syntax
)
1165 reg_syntax_t ret
= re_syntax_options
;
1167 re_syntax_options
= syntax
;
1170 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1174 /* This table gives an error message for each of the error codes listed
1175 in regex.h. Obviously the order here has to be same as there.
1176 POSIX doesn't require that we do anything for REG_NOERROR,
1177 but why not be nice? */
1179 static const char *re_error_msgid
[] =
1181 gettext_noop ("Success"), /* REG_NOERROR */
1182 gettext_noop ("No match"), /* REG_NOMATCH */
1183 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1184 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1185 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1186 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1187 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1188 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1189 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1190 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1191 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1192 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1193 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1194 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1195 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1196 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1197 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1198 gettext_noop ("Range striding over charsets"), /* REG_ERANGEX */
1199 gettext_noop ("Invalid content of \\{\\}, repetitions too big") /* REG_ESIZEBR */
1202 /* Whether to allocate memory during matching. */
1204 /* Define MATCH_MAY_ALLOCATE to allow the searching and matching
1205 functions allocate memory for the failure stack and registers.
1206 Normally should be defined, because otherwise searching and
1207 matching routines will have much smaller memory resources at their
1208 disposal, and therefore might fail to handle complex regexps.
1209 Therefore undefine MATCH_MAY_ALLOCATE only in the following
1210 exceptional situations:
1212 . When running on a system where memory is at premium.
1213 . When alloca cannot be used at all, perhaps due to bugs in
1214 its implementation, or its being unavailable, or due to a
1215 very small stack size. This requires to define REGEX_MALLOC
1216 to use malloc instead, which in turn could lead to memory
1217 leaks if search is interrupted by a signal. (For these
1218 reasons, defining REGEX_MALLOC when building Emacs
1219 automatically undefines MATCH_MAY_ALLOCATE, but outside
1220 Emacs you may not care about memory leaks.) If you want to
1221 prevent the memory leaks, undefine MATCH_MAY_ALLOCATE.
1222 . When code that calls the searching and matching functions
1223 cannot allow memory allocation, for whatever reasons. */
1225 /* Normally, this is fine. */
1226 #define MATCH_MAY_ALLOCATE
1228 /* The match routines may not allocate if (1) they would do it with malloc
1229 and (2) it's not safe for them to use malloc.
1230 Note that if REL_ALLOC is defined, matching would not use malloc for the
1231 failure stack, but we would still use it for the register vectors;
1232 so REL_ALLOC should not affect this. */
1233 #if defined REGEX_MALLOC && defined emacs
1234 # undef MATCH_MAY_ALLOCATE
1237 /* While regex matching of a single compiled pattern isn't reentrant
1238 (because we compile regexes to bytecode programs, and the bytecode
1239 programs are self-modifying), the regex machinery must nevertheless
1240 be reentrant with respect to _different_ patterns, and we do that
1241 by avoiding global variables and using MATCH_MAY_ALLOCATE. */
1242 #if !defined MATCH_MAY_ALLOCATE && defined emacs
1243 # error "Emacs requires MATCH_MAY_ALLOCATE"
1247 /* Failure stack declarations and macros; both re_compile_fastmap and
1248 re_match_2 use a failure stack. These have to be macros because of
1249 REGEX_ALLOCATE_STACK. */
1252 /* Approximate number of failure points for which to initially allocate space
1253 when matching. If this number is exceeded, we allocate more
1254 space, so it is not a hard limit. */
1255 #ifndef INIT_FAILURE_ALLOC
1256 # define INIT_FAILURE_ALLOC 20
1259 /* Roughly the maximum number of failure points on the stack. Would be
1260 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1261 This is a variable only so users of regex can assign to it; we never
1262 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1263 before using it, so it should probably be a byte-count instead. */
1264 # if defined MATCH_MAY_ALLOCATE
1265 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1266 whose default stack limit is 2mb. In order for a larger
1267 value to work reliably, you have to try to make it accord
1268 with the process stack limit. */
1269 size_t emacs_re_max_failures
= 40000;
1271 size_t emacs_re_max_failures
= 4000;
1274 union fail_stack_elt
1277 /* This should be the biggest `int' that's no bigger than a pointer. */
1281 typedef union fail_stack_elt fail_stack_elt_t
;
1285 fail_stack_elt_t
*stack
;
1287 size_t avail
; /* Offset of next open position. */
1288 size_t frame
; /* Offset of the cur constructed frame. */
1291 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1294 /* Define macros to initialize and free the failure stack.
1295 Do `return -2' if the alloc fails. */
1297 #ifdef MATCH_MAY_ALLOCATE
1298 # define INIT_FAIL_STACK() \
1300 fail_stack.stack = \
1301 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1302 * sizeof (fail_stack_elt_t)); \
1304 if (fail_stack.stack == NULL) \
1307 fail_stack.size = INIT_FAILURE_ALLOC; \
1308 fail_stack.avail = 0; \
1309 fail_stack.frame = 0; \
1312 # define INIT_FAIL_STACK() \
1314 fail_stack.avail = 0; \
1315 fail_stack.frame = 0; \
1318 # define RETALLOC_IF(addr, n, t) \
1319 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1323 /* Double the size of FAIL_STACK, up to a limit
1324 which allows approximately `emacs_re_max_failures' items.
1326 Return 1 if succeeds, and 0 if either ran out of memory
1327 allocating space for it or it was already too large.
1329 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1331 /* Factor to increase the failure stack size by
1332 when we increase it.
1333 This used to be 2, but 2 was too wasteful
1334 because the old discarded stacks added up to as much space
1335 were as ultimate, maximum-size stack. */
1336 #define FAIL_STACK_GROWTH_FACTOR 4
1338 #define GROW_FAIL_STACK(fail_stack) \
1339 (((fail_stack).size >= emacs_re_max_failures * TYPICAL_FAILURE_SIZE) \
1341 : ((fail_stack).stack \
1342 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1343 (fail_stack).size * sizeof (fail_stack_elt_t), \
1344 min (emacs_re_max_failures * TYPICAL_FAILURE_SIZE, \
1345 ((fail_stack).size * FAIL_STACK_GROWTH_FACTOR)) \
1346 * sizeof (fail_stack_elt_t)), \
1348 (fail_stack).stack == NULL \
1350 : ((fail_stack).size \
1351 = (min (emacs_re_max_failures * TYPICAL_FAILURE_SIZE, \
1352 ((fail_stack).size * FAIL_STACK_GROWTH_FACTOR))), \
1356 /* Push a pointer value onto the failure stack.
1357 Assumes the variable `fail_stack'. Probably should only
1358 be called from within `PUSH_FAILURE_POINT'. */
1359 #define PUSH_FAILURE_POINTER(item) \
1360 fail_stack.stack[fail_stack.avail++].pointer = (item)
1362 /* This pushes an integer-valued item onto the failure stack.
1363 Assumes the variable `fail_stack'. Probably should only
1364 be called from within `PUSH_FAILURE_POINT'. */
1365 #define PUSH_FAILURE_INT(item) \
1366 fail_stack.stack[fail_stack.avail++].integer = (item)
1368 /* These POP... operations complement the PUSH... operations.
1369 All assume that `fail_stack' is nonempty. */
1370 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1371 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1373 /* Individual items aside from the registers. */
1374 #define NUM_NONREG_ITEMS 3
1376 /* Used to examine the stack (to detect infinite loops). */
1377 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1378 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1379 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1380 #define TOP_FAILURE_HANDLE() fail_stack.frame
1383 #define ENSURE_FAIL_STACK(space) \
1384 while (REMAINING_AVAIL_SLOTS <= space) { \
1385 if (!GROW_FAIL_STACK (fail_stack)) \
1387 DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\
1388 DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\
1391 /* Push register NUM onto the stack. */
1392 #define PUSH_FAILURE_REG(num) \
1394 char *destination; \
1396 ENSURE_FAIL_STACK(3); \
1397 DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \
1398 n, regstart[n], regend[n]); \
1399 PUSH_FAILURE_POINTER (regstart[n]); \
1400 PUSH_FAILURE_POINTER (regend[n]); \
1401 PUSH_FAILURE_INT (n); \
1404 /* Change the counter's value to VAL, but make sure that it will
1405 be reset when backtracking. */
1406 #define PUSH_NUMBER(ptr,val) \
1408 char *destination; \
1410 ENSURE_FAIL_STACK(3); \
1411 EXTRACT_NUMBER (c, ptr); \
1412 DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \
1413 PUSH_FAILURE_INT (c); \
1414 PUSH_FAILURE_POINTER (ptr); \
1415 PUSH_FAILURE_INT (-1); \
1416 STORE_NUMBER (ptr, val); \
1419 /* Pop a saved register off the stack. */
1420 #define POP_FAILURE_REG_OR_COUNT() \
1422 long pfreg = POP_FAILURE_INT (); \
1425 /* It's a counter. */ \
1426 /* Here, we discard `const', making re_match non-reentrant. */ \
1427 unsigned char *ptr = (unsigned char *) POP_FAILURE_POINTER (); \
1428 pfreg = POP_FAILURE_INT (); \
1429 STORE_NUMBER (ptr, pfreg); \
1430 DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \
1434 regend[pfreg] = POP_FAILURE_POINTER (); \
1435 regstart[pfreg] = POP_FAILURE_POINTER (); \
1436 DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \
1437 pfreg, regstart[pfreg], regend[pfreg]); \
1441 /* Check that we are not stuck in an infinite loop. */
1442 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1444 ssize_t failure = TOP_FAILURE_HANDLE (); \
1445 /* Check for infinite matching loops */ \
1446 while (failure > 0 \
1447 && (FAILURE_STR (failure) == string_place \
1448 || FAILURE_STR (failure) == NULL)) \
1450 assert (FAILURE_PAT (failure) >= bufp->buffer \
1451 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1452 if (FAILURE_PAT (failure) == pat_cur) \
1457 DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1458 failure = NEXT_FAILURE_HANDLE(failure); \
1460 DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \
1463 /* Push the information about the state we will need
1464 if we ever fail back to it.
1466 Requires variables fail_stack, regstart, regend and
1467 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1470 Does `return FAILURE_CODE' if runs out of memory. */
1472 #define PUSH_FAILURE_POINT(pattern, string_place) \
1474 char *destination; \
1475 /* Must be int, so when we don't save any registers, the arithmetic \
1476 of 0 + -1 isn't done as unsigned. */ \
1478 DEBUG_STATEMENT (nfailure_points_pushed++); \
1479 DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \
1480 DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \
1481 DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\
1483 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1485 DEBUG_PRINT ("\n"); \
1487 DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \
1488 PUSH_FAILURE_INT (fail_stack.frame); \
1490 DEBUG_PRINT (" Push string %p: \"", string_place); \
1491 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1492 DEBUG_PRINT ("\"\n"); \
1493 PUSH_FAILURE_POINTER (string_place); \
1495 DEBUG_PRINT (" Push pattern %p: ", pattern); \
1496 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1497 PUSH_FAILURE_POINTER (pattern); \
1499 /* Close the frame by moving the frame pointer past it. */ \
1500 fail_stack.frame = fail_stack.avail; \
1503 /* Estimate the size of data pushed by a typical failure stack entry.
1504 An estimate is all we need, because all we use this for
1505 is to choose a limit for how big to make the failure stack. */
1506 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1507 #define TYPICAL_FAILURE_SIZE 20
1509 /* How many items can still be added to the stack without overflowing it. */
1510 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1513 /* Pops what PUSH_FAIL_STACK pushes.
1515 We restore into the parameters, all of which should be lvalues:
1516 STR -- the saved data position.
1517 PAT -- the saved pattern position.
1518 REGSTART, REGEND -- arrays of string positions.
1520 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1521 `pend', `string1', `size1', `string2', and `size2'. */
1523 #define POP_FAILURE_POINT(str, pat) \
1525 assert (!FAIL_STACK_EMPTY ()); \
1527 /* Remove failure points and point to how many regs pushed. */ \
1528 DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \
1529 DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \
1530 DEBUG_PRINT (" size: %zd\n", fail_stack.size); \
1532 /* Pop the saved registers. */ \
1533 while (fail_stack.frame < fail_stack.avail) \
1534 POP_FAILURE_REG_OR_COUNT (); \
1536 pat = POP_FAILURE_POINTER (); \
1537 DEBUG_PRINT (" Popping pattern %p: ", pat); \
1538 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1540 /* If the saved string location is NULL, it came from an \
1541 on_failure_keep_string_jump opcode, and we want to throw away the \
1542 saved NULL, thus retaining our current position in the string. */ \
1543 str = POP_FAILURE_POINTER (); \
1544 DEBUG_PRINT (" Popping string %p: \"", str); \
1545 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1546 DEBUG_PRINT ("\"\n"); \
1548 fail_stack.frame = POP_FAILURE_INT (); \
1549 DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \
1551 assert (fail_stack.avail >= 0); \
1552 assert (fail_stack.frame <= fail_stack.avail); \
1554 DEBUG_STATEMENT (nfailure_points_popped++); \
1555 } while (0) /* POP_FAILURE_POINT */
1559 /* Registers are set to a sentinel when they haven't yet matched. */
1560 #define REG_UNSET(e) ((e) == NULL)
1562 /* Subroutine declarations and macros for regex_compile. */
1564 static reg_errcode_t
regex_compile (re_char
*pattern
, size_t size
,
1566 bool posix_backtracking
,
1567 const char *whitespace_regexp
,
1569 reg_syntax_t syntax
,
1571 struct re_pattern_buffer
*bufp
);
1572 static void store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
);
1573 static void store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
);
1574 static void insert_op1 (re_opcode_t op
, unsigned char *loc
,
1575 int arg
, unsigned char *end
);
1576 static void insert_op2 (re_opcode_t op
, unsigned char *loc
,
1577 int arg1
, int arg2
, unsigned char *end
);
1578 static boolean
at_begline_loc_p (re_char
*pattern
, re_char
*p
,
1579 reg_syntax_t syntax
);
1580 static boolean
at_endline_loc_p (re_char
*p
, re_char
*pend
,
1581 reg_syntax_t syntax
);
1582 static re_char
*skip_one_char (re_char
*p
);
1583 static int analyze_first (re_char
*p
, re_char
*pend
,
1584 char *fastmap
, const int multibyte
);
1586 /* Fetch the next character in the uncompiled pattern, with no
1588 #define PATFETCH(c) \
1591 if (p == pend) return REG_EEND; \
1592 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1597 /* If `translate' is non-null, return translate[D], else just D. We
1598 cast the subscript to translate because some data is declared as
1599 `char *', to avoid warnings when a string constant is passed. But
1600 when we use a character as a subscript we must make it unsigned. */
1602 # define TRANSLATE(d) \
1603 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1607 /* Macros for outputting the compiled pattern into `buffer'. */
1609 /* If the buffer isn't allocated when it comes in, use this. */
1610 #define INIT_BUF_SIZE 32
1612 /* Make sure we have at least N more bytes of space in buffer. */
1613 #define GET_BUFFER_SPACE(n) \
1614 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1617 /* Make sure we have one more byte of buffer space and then add C to it. */
1618 #define BUF_PUSH(c) \
1620 GET_BUFFER_SPACE (1); \
1621 *b++ = (unsigned char) (c); \
1625 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1626 #define BUF_PUSH_2(c1, c2) \
1628 GET_BUFFER_SPACE (2); \
1629 *b++ = (unsigned char) (c1); \
1630 *b++ = (unsigned char) (c2); \
1634 /* Store a jump with opcode OP at LOC to location TO. We store a
1635 relative address offset by the three bytes the jump itself occupies. */
1636 #define STORE_JUMP(op, loc, to) \
1637 store_op1 (op, loc, (to) - (loc) - 3)
1639 /* Likewise, for a two-argument jump. */
1640 #define STORE_JUMP2(op, loc, to, arg) \
1641 store_op2 (op, loc, (to) - (loc) - 3, arg)
1643 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1644 #define INSERT_JUMP(op, loc, to) \
1645 insert_op1 (op, loc, (to) - (loc) - 3, b)
1647 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1648 #define INSERT_JUMP2(op, loc, to, arg) \
1649 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1652 /* This is not an arbitrary limit: the arguments which represent offsets
1653 into the pattern are two bytes long. So if 2^15 bytes turns out to
1654 be too small, many things would have to change. */
1655 # define MAX_BUF_SIZE (1L << 15)
1657 /* Extend the buffer by twice its current size via realloc and
1658 reset the pointers that pointed into the old block to point to the
1659 correct places in the new one. If extending the buffer results in it
1660 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1661 #define EXTEND_BUFFER() \
1663 unsigned char *old_buffer = bufp->buffer; \
1664 if (bufp->allocated == MAX_BUF_SIZE) \
1666 bufp->allocated <<= 1; \
1667 if (bufp->allocated > MAX_BUF_SIZE) \
1668 bufp->allocated = MAX_BUF_SIZE; \
1669 ptrdiff_t b_off = b - old_buffer; \
1670 ptrdiff_t begalt_off = begalt - old_buffer; \
1671 bool fixup_alt_jump_set = !!fixup_alt_jump; \
1672 bool laststart_set = !!laststart; \
1673 bool pending_exact_set = !!pending_exact; \
1674 ptrdiff_t fixup_alt_jump_off, laststart_off, pending_exact_off; \
1675 if (fixup_alt_jump_set) fixup_alt_jump_off = fixup_alt_jump - old_buffer; \
1676 if (laststart_set) laststart_off = laststart - old_buffer; \
1677 if (pending_exact_set) pending_exact_off = pending_exact - old_buffer; \
1678 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1679 if (bufp->buffer == NULL) \
1680 return REG_ESPACE; \
1681 unsigned char *new_buffer = bufp->buffer; \
1682 b = new_buffer + b_off; \
1683 begalt = new_buffer + begalt_off; \
1684 if (fixup_alt_jump_set) fixup_alt_jump = new_buffer + fixup_alt_jump_off; \
1685 if (laststart_set) laststart = new_buffer + laststart_off; \
1686 if (pending_exact_set) pending_exact = new_buffer + pending_exact_off; \
1690 /* Since we have one byte reserved for the register number argument to
1691 {start,stop}_memory, the maximum number of groups we can report
1692 things about is what fits in that byte. */
1693 #define MAX_REGNUM 255
1695 /* But patterns can have more than `MAX_REGNUM' registers. We just
1696 ignore the excess. */
1697 typedef int regnum_t
;
1700 /* Macros for the compile stack. */
1702 /* Since offsets can go either forwards or backwards, this type needs to
1703 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1704 /* int may be not enough when sizeof(int) == 2. */
1705 typedef long pattern_offset_t
;
1709 pattern_offset_t begalt_offset
;
1710 pattern_offset_t fixup_alt_jump
;
1711 pattern_offset_t laststart_offset
;
1713 } compile_stack_elt_t
;
1718 compile_stack_elt_t
*stack
;
1720 size_t avail
; /* Offset of next open position. */
1721 } compile_stack_type
;
1724 #define INIT_COMPILE_STACK_SIZE 32
1726 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1727 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1729 /* The next available element. */
1730 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1732 /* Explicit quit checking is needed for Emacs, which uses polling to
1733 process input events. */
1735 static void maybe_quit (void) {}
1738 /* Structure to manage work area for range table. */
1739 struct range_table_work_area
1741 int *table
; /* actual work area. */
1742 int allocated
; /* allocated size for work area in bytes. */
1743 int used
; /* actually used size in words. */
1744 int bits
; /* flag to record character classes */
1749 /* Make sure that WORK_AREA can hold more N multibyte characters.
1750 This is used only in set_image_of_range and set_image_of_range_1.
1751 It expects WORK_AREA to be a pointer.
1752 If it can't get the space, it returns from the surrounding function. */
1754 #define EXTEND_RANGE_TABLE(work_area, n) \
1756 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1758 extend_range_table_work_area (&work_area); \
1759 if ((work_area).table == 0) \
1760 return (REG_ESPACE); \
1764 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1765 (work_area).bits |= (bit)
1767 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1768 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1770 EXTEND_RANGE_TABLE ((work_area), 2); \
1771 (work_area).table[(work_area).used++] = (range_start); \
1772 (work_area).table[(work_area).used++] = (range_end); \
1777 /* Free allocated memory for WORK_AREA. */
1778 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1780 if ((work_area).table) \
1781 free ((work_area).table); \
1784 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1785 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1786 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1787 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1789 /* Bits used to implement the multibyte-part of the various character classes
1790 such as [:alnum:] in a charset's range table. The code currently assumes
1791 that only the low 16 bits are used. */
1792 #define BIT_WORD 0x1
1793 #define BIT_LOWER 0x2
1794 #define BIT_PUNCT 0x4
1795 #define BIT_SPACE 0x8
1796 #define BIT_UPPER 0x10
1797 #define BIT_MULTIBYTE 0x20
1798 #define BIT_ALPHA 0x40
1799 #define BIT_ALNUM 0x80
1800 #define BIT_GRAPH 0x100
1801 #define BIT_PRINT 0x200
1802 #define BIT_BLANK 0x400
1805 /* Set the bit for character C in a list. */
1806 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1811 /* Store characters in the range FROM to TO in the bitmap at B (for
1812 ASCII and unibyte characters) and WORK_AREA (for multibyte
1813 characters) while translating them and paying attention to the
1814 continuity of translated characters.
1816 Implementation note: It is better to implement these fairly big
1817 macros by a function, but it's not that easy because macros called
1818 in this macro assume various local variables already declared. */
1820 /* Both FROM and TO are ASCII characters. */
1822 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1826 for (C0 = (FROM); C0 <= (TO); C0++) \
1828 C1 = TRANSLATE (C0); \
1829 if (! ASCII_CHAR_P (C1)) \
1831 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1832 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1835 SET_LIST_BIT (C1); \
1840 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1842 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1844 int C0, C1, C2, I; \
1845 int USED = RANGE_TABLE_WORK_USED (work_area); \
1847 for (C0 = (FROM); C0 <= (TO); C0++) \
1849 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1850 if (CHAR_BYTE8_P (C1)) \
1851 SET_LIST_BIT (C0); \
1854 C2 = TRANSLATE (C1); \
1856 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1858 SET_LIST_BIT (C1); \
1859 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1861 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1862 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1864 if (C2 >= from - 1 && C2 <= to + 1) \
1866 if (C2 == from - 1) \
1867 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1868 else if (C2 == to + 1) \
1869 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1874 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1880 /* Both FROM and TO are multibyte characters. */
1882 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1884 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1886 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1887 for (C0 = (FROM); C0 <= (TO); C0++) \
1889 C1 = TRANSLATE (C0); \
1890 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1891 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1892 SET_LIST_BIT (C2); \
1893 if (C1 >= (FROM) && C1 <= (TO)) \
1895 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1897 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1898 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1900 if (C1 >= from - 1 && C1 <= to + 1) \
1902 if (C1 == from - 1) \
1903 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1904 else if (C1 == to + 1) \
1905 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1910 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1916 /* Get the next unsigned number in the uncompiled pattern. */
1917 #define GET_INTERVAL_COUNT(num) \
1920 FREE_STACK_RETURN (REG_EBRACE); \
1924 while ('0' <= c && c <= '9') \
1928 if (RE_DUP_MAX / 10 - (RE_DUP_MAX % 10 < c - '0') < num) \
1929 FREE_STACK_RETURN (REG_ESIZEBR); \
1930 num = num * 10 + c - '0'; \
1932 FREE_STACK_RETURN (REG_EBRACE); \
1938 #if ! WIDE_CHAR_SUPPORT
1940 /* Parse a character class, i.e. string such as "[:name:]". *strp
1941 points to the string to be parsed and limit is length, in bytes, of
1944 If *strp point to a string that begins with "[:name:]", where name is
1945 a non-empty sequence of lower case letters, *strp will be advanced past the
1946 closing square bracket and RECC_* constant which maps to the name will be
1947 returned. If name is not a valid character class name zero, or RECC_ERROR,
1950 Otherwise, if *strp doesn't begin with "[:name:]", -1 is returned.
1952 The function can be used on ASCII and multibyte (UTF-8-encoded) strings.
1955 re_wctype_parse (const unsigned char **strp
, unsigned limit
)
1957 const char *beg
= (const char *)*strp
, *it
;
1959 if (limit
< 4 || beg
[0] != '[' || beg
[1] != ':')
1962 beg
+= 2; /* skip opening "[:" */
1963 limit
-= 3; /* opening "[:" and half of closing ":]"; --limit handles rest */
1964 for (it
= beg
; it
[0] != ':' || it
[1] != ']'; ++it
)
1968 *strp
= (const unsigned char *)(it
+ 2);
1970 /* Sort tests in the length=five case by frequency the classes to minimize
1971 number of times we fail the comparison. The frequencies of character class
1972 names used in Emacs sources as of 2016-07-27:
1974 $ find \( -name \*.c -o -name \*.el \) -exec grep -h '\[:[a-z]*:]' {} + |
1975 sed 's/]/]\n/g' |grep -o '\[:[a-z]*:]' |sort |uniq -c |sort -nr
1993 If you update this list, consider also updating chain of or'ed conditions
1994 in execute_charset function.
1999 if (!memcmp (beg
, "word", 4)) return RECC_WORD
;
2002 if (!memcmp (beg
, "alnum", 5)) return RECC_ALNUM
;
2003 if (!memcmp (beg
, "alpha", 5)) return RECC_ALPHA
;
2004 if (!memcmp (beg
, "space", 5)) return RECC_SPACE
;
2005 if (!memcmp (beg
, "digit", 5)) return RECC_DIGIT
;
2006 if (!memcmp (beg
, "blank", 5)) return RECC_BLANK
;
2007 if (!memcmp (beg
, "upper", 5)) return RECC_UPPER
;
2008 if (!memcmp (beg
, "lower", 5)) return RECC_LOWER
;
2009 if (!memcmp (beg
, "punct", 5)) return RECC_PUNCT
;
2010 if (!memcmp (beg
, "ascii", 5)) return RECC_ASCII
;
2011 if (!memcmp (beg
, "graph", 5)) return RECC_GRAPH
;
2012 if (!memcmp (beg
, "print", 5)) return RECC_PRINT
;
2013 if (!memcmp (beg
, "cntrl", 5)) return RECC_CNTRL
;
2016 if (!memcmp (beg
, "xdigit", 6)) return RECC_XDIGIT
;
2019 if (!memcmp (beg
, "unibyte", 7)) return RECC_UNIBYTE
;
2022 if (!memcmp (beg
, "nonascii", 8)) return RECC_NONASCII
;
2025 if (!memcmp (beg
, "multibyte", 9)) return RECC_MULTIBYTE
;
2032 /* True if CH is in the char class CC. */
2034 re_iswctype (int ch
, re_wctype_t cc
)
2038 case RECC_ALNUM
: return ISALNUM (ch
) != 0;
2039 case RECC_ALPHA
: return ISALPHA (ch
) != 0;
2040 case RECC_BLANK
: return ISBLANK (ch
) != 0;
2041 case RECC_CNTRL
: return ISCNTRL (ch
) != 0;
2042 case RECC_DIGIT
: return ISDIGIT (ch
) != 0;
2043 case RECC_GRAPH
: return ISGRAPH (ch
) != 0;
2044 case RECC_LOWER
: return ISLOWER (ch
) != 0;
2045 case RECC_PRINT
: return ISPRINT (ch
) != 0;
2046 case RECC_PUNCT
: return ISPUNCT (ch
) != 0;
2047 case RECC_SPACE
: return ISSPACE (ch
) != 0;
2048 case RECC_UPPER
: return ISUPPER (ch
) != 0;
2049 case RECC_XDIGIT
: return ISXDIGIT (ch
) != 0;
2050 case RECC_ASCII
: return IS_REAL_ASCII (ch
) != 0;
2051 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2052 case RECC_UNIBYTE
: return ISUNIBYTE (ch
) != 0;
2053 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2054 case RECC_WORD
: return ISWORD (ch
) != 0;
2055 case RECC_ERROR
: return false;
2061 /* Return a bit-pattern to use in the range-table bits to match multibyte
2062 chars of class CC. */
2064 re_wctype_to_bit (re_wctype_t cc
)
2069 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2070 case RECC_ALPHA
: return BIT_ALPHA
;
2071 case RECC_ALNUM
: return BIT_ALNUM
;
2072 case RECC_WORD
: return BIT_WORD
;
2073 case RECC_LOWER
: return BIT_LOWER
;
2074 case RECC_UPPER
: return BIT_UPPER
;
2075 case RECC_PUNCT
: return BIT_PUNCT
;
2076 case RECC_SPACE
: return BIT_SPACE
;
2077 case RECC_GRAPH
: return BIT_GRAPH
;
2078 case RECC_PRINT
: return BIT_PRINT
;
2079 case RECC_BLANK
: return BIT_BLANK
;
2080 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2081 case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2088 /* Filling in the work area of a range. */
2090 /* Actually extend the space in WORK_AREA. */
2093 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2095 work_area
->allocated
+= 16 * sizeof (int);
2096 work_area
->table
= realloc (work_area
->table
, work_area
->allocated
);
2102 /* Carefully find the ranges of codes that are equivalent
2103 under case conversion to the range start..end when passed through
2104 TRANSLATE. Handle the case where non-letters can come in between
2105 two upper-case letters (which happens in Latin-1).
2106 Also handle the case of groups of more than 2 case-equivalent chars.
2108 The basic method is to look at consecutive characters and see
2109 if they can form a run that can be handled as one.
2111 Returns -1 if successful, REG_ESPACE if ran out of space. */
2114 set_image_of_range_1 (struct range_table_work_area
*work_area
,
2115 re_wchar_t start
, re_wchar_t end
,
2116 RE_TRANSLATE_TYPE translate
)
2118 /* `one_case' indicates a character, or a run of characters,
2119 each of which is an isolate (no case-equivalents).
2120 This includes all ASCII non-letters.
2122 `two_case' indicates a character, or a run of characters,
2123 each of which has two case-equivalent forms.
2124 This includes all ASCII letters.
2126 `strange' indicates a character that has more than one
2129 enum case_type
{one_case
, two_case
, strange
};
2131 /* Describe the run that is in progress,
2132 which the next character can try to extend.
2133 If run_type is strange, that means there really is no run.
2134 If run_type is one_case, then run_start...run_end is the run.
2135 If run_type is two_case, then the run is run_start...run_end,
2136 and the case-equivalents end at run_eqv_end. */
2138 enum case_type run_type
= strange
;
2139 int run_start
, run_end
, run_eqv_end
;
2141 Lisp_Object eqv_table
;
2143 if (!RE_TRANSLATE_P (translate
))
2145 EXTEND_RANGE_TABLE (work_area
, 2);
2146 work_area
->table
[work_area
->used
++] = (start
);
2147 work_area
->table
[work_area
->used
++] = (end
);
2151 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2153 for (; start
<= end
; start
++)
2155 enum case_type this_type
;
2156 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2157 int minchar
, maxchar
;
2159 /* Classify this character */
2161 this_type
= one_case
;
2162 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2163 this_type
= two_case
;
2165 this_type
= strange
;
2168 minchar
= start
, maxchar
= eqv
;
2170 minchar
= eqv
, maxchar
= start
;
2172 /* Can this character extend the run in progress? */
2173 if (this_type
== strange
|| this_type
!= run_type
2174 || !(minchar
== run_end
+ 1
2175 && (run_type
== two_case
2176 ? maxchar
== run_eqv_end
+ 1 : 1)))
2179 Record each of its equivalent ranges. */
2180 if (run_type
== one_case
)
2182 EXTEND_RANGE_TABLE (work_area
, 2);
2183 work_area
->table
[work_area
->used
++] = run_start
;
2184 work_area
->table
[work_area
->used
++] = run_end
;
2186 else if (run_type
== two_case
)
2188 EXTEND_RANGE_TABLE (work_area
, 4);
2189 work_area
->table
[work_area
->used
++] = run_start
;
2190 work_area
->table
[work_area
->used
++] = run_end
;
2191 work_area
->table
[work_area
->used
++]
2192 = RE_TRANSLATE (eqv_table
, run_start
);
2193 work_area
->table
[work_area
->used
++]
2194 = RE_TRANSLATE (eqv_table
, run_end
);
2199 if (this_type
== strange
)
2201 /* For a strange character, add each of its equivalents, one
2202 by one. Don't start a range. */
2205 EXTEND_RANGE_TABLE (work_area
, 2);
2206 work_area
->table
[work_area
->used
++] = eqv
;
2207 work_area
->table
[work_area
->used
++] = eqv
;
2208 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2210 while (eqv
!= start
);
2213 /* Add this char to the run, or start a new run. */
2214 else if (run_type
== strange
)
2216 /* Initialize a new range. */
2217 run_type
= this_type
;
2220 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2224 /* Extend a running range. */
2226 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2230 /* If a run is still in progress at the end, finish it now
2231 by recording its equivalent ranges. */
2232 if (run_type
== one_case
)
2234 EXTEND_RANGE_TABLE (work_area
, 2);
2235 work_area
->table
[work_area
->used
++] = run_start
;
2236 work_area
->table
[work_area
->used
++] = run_end
;
2238 else if (run_type
== two_case
)
2240 EXTEND_RANGE_TABLE (work_area
, 4);
2241 work_area
->table
[work_area
->used
++] = run_start
;
2242 work_area
->table
[work_area
->used
++] = run_end
;
2243 work_area
->table
[work_area
->used
++]
2244 = RE_TRANSLATE (eqv_table
, run_start
);
2245 work_area
->table
[work_area
->used
++]
2246 = RE_TRANSLATE (eqv_table
, run_end
);
2254 /* Record the image of the range start..end when passed through
2255 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2256 and is not even necessarily contiguous.
2257 Normally we approximate it with the smallest contiguous range that contains
2258 all the chars we need. However, for Latin-1 we go to extra effort
2261 This function is not called for ASCII ranges.
2263 Returns -1 if successful, REG_ESPACE if ran out of space. */
2266 set_image_of_range (struct range_table_work_area
*work_area
,
2267 re_wchar_t start
, re_wchar_t end
,
2268 RE_TRANSLATE_TYPE translate
)
2270 re_wchar_t cmin
, cmax
;
2273 /* For Latin-1 ranges, use set_image_of_range_1
2274 to get proper handling of ranges that include letters and nonletters.
2275 For a range that includes the whole of Latin-1, this is not necessary.
2276 For other character sets, we don't bother to get this right. */
2277 if (RE_TRANSLATE_P (translate
) && start
< 04400
2278 && !(start
< 04200 && end
>= 04377))
2285 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2295 EXTEND_RANGE_TABLE (work_area
, 2);
2296 work_area
->table
[work_area
->used
++] = (start
);
2297 work_area
->table
[work_area
->used
++] = (end
);
2299 cmin
= -1, cmax
= -1;
2301 if (RE_TRANSLATE_P (translate
))
2305 for (ch
= start
; ch
<= end
; ch
++)
2307 re_wchar_t c
= TRANSLATE (ch
);
2308 if (! (start
<= c
&& c
<= end
))
2314 cmin
= min (cmin
, c
);
2315 cmax
= max (cmax
, c
);
2322 EXTEND_RANGE_TABLE (work_area
, 2);
2323 work_area
->table
[work_area
->used
++] = (cmin
);
2324 work_area
->table
[work_area
->used
++] = (cmax
);
2332 #ifndef MATCH_MAY_ALLOCATE
2334 /* If we cannot allocate large objects within re_match_2_internal,
2335 we make the fail stack and register vectors global.
2336 The fail stack, we grow to the maximum size when a regexp
2338 The register vectors, we adjust in size each time we
2339 compile a regexp, according to the number of registers it needs. */
2341 static fail_stack_type fail_stack
;
2343 /* Size with which the following vectors are currently allocated.
2344 That is so we can make them bigger as needed,
2345 but never make them smaller. */
2346 static int regs_allocated_size
;
2348 static re_char
** regstart
, ** regend
;
2349 static re_char
**best_regstart
, **best_regend
;
2351 /* Make the register vectors big enough for NUM_REGS registers,
2352 but don't make them smaller. */
2355 regex_grow_registers (int num_regs
)
2357 if (num_regs
> regs_allocated_size
)
2359 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2360 RETALLOC_IF (regend
, num_regs
, re_char
*);
2361 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2362 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2364 regs_allocated_size
= num_regs
;
2368 #endif /* not MATCH_MAY_ALLOCATE */
2370 static boolean
group_in_compile_stack (compile_stack_type compile_stack
,
2373 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2374 Returns one of error codes defined in `regex.h', or zero for success.
2376 If WHITESPACE_REGEXP is given (only #ifdef emacs), it is used instead of
2377 a space character in PATTERN.
2379 Assumes the `allocated' (and perhaps `buffer') and `translate'
2380 fields are set in BUFP on entry.
2382 If it succeeds, results are put in BUFP (if it returns an error, the
2383 contents of BUFP are undefined):
2384 `buffer' is the compiled pattern;
2385 `syntax' is set to SYNTAX;
2386 `used' is set to the length of the compiled pattern;
2387 `fastmap_accurate' is zero;
2388 `re_nsub' is the number of subexpressions in PATTERN;
2389 `not_bol' and `not_eol' are zero;
2391 The `fastmap' field is neither examined nor set. */
2393 /* Insert the `jump' from the end of last alternative to "here".
2394 The space for the jump has already been allocated. */
2395 #define FIXUP_ALT_JUMP() \
2397 if (fixup_alt_jump) \
2398 STORE_JUMP (jump, fixup_alt_jump, b); \
2402 /* Return, freeing storage we allocated. */
2403 #define FREE_STACK_RETURN(value) \
2405 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2406 free (compile_stack.stack); \
2410 static reg_errcode_t
2411 regex_compile (re_char
*pattern
, size_t size
,
2413 # define syntax RE_SYNTAX_EMACS
2414 bool posix_backtracking
,
2415 const char *whitespace_regexp
,
2417 reg_syntax_t syntax
,
2418 # define posix_backtracking (!(syntax & RE_NO_POSIX_BACKTRACKING))
2420 struct re_pattern_buffer
*bufp
)
2422 /* We fetch characters from PATTERN here. */
2423 register re_wchar_t c
, c1
;
2425 /* Points to the end of the buffer, where we should append. */
2426 register unsigned char *b
;
2428 /* Keeps track of unclosed groups. */
2429 compile_stack_type compile_stack
;
2431 /* Points to the current (ending) position in the pattern. */
2433 /* `const' makes AIX compiler fail. */
2434 unsigned char *p
= pattern
;
2436 re_char
*p
= pattern
;
2438 re_char
*pend
= pattern
+ size
;
2440 /* How to translate the characters in the pattern. */
2441 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2443 /* Address of the count-byte of the most recently inserted `exactn'
2444 command. This makes it possible to tell if a new exact-match
2445 character can be added to that command or if the character requires
2446 a new `exactn' command. */
2447 unsigned char *pending_exact
= 0;
2449 /* Address of start of the most recently finished expression.
2450 This tells, e.g., postfix * where to find the start of its
2451 operand. Reset at the beginning of groups and alternatives. */
2452 unsigned char *laststart
= 0;
2454 /* Address of beginning of regexp, or inside of last group. */
2455 unsigned char *begalt
;
2457 /* Place in the uncompiled pattern (i.e., the {) to
2458 which to go back if the interval is invalid. */
2459 re_char
*beg_interval
;
2461 /* Address of the place where a forward jump should go to the end of
2462 the containing expression. Each alternative of an `or' -- except the
2463 last -- ends with a forward jump of this sort. */
2464 unsigned char *fixup_alt_jump
= 0;
2466 /* Work area for range table of charset. */
2467 struct range_table_work_area range_table_work
;
2469 /* If the object matched can contain multibyte characters. */
2470 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2473 /* Nonzero if we have pushed down into a subpattern. */
2474 int in_subpattern
= 0;
2476 /* These hold the values of p, pattern, and pend from the main
2477 pattern when we have pushed into a subpattern. */
2479 re_char
*main_pattern
;
2485 DEBUG_PRINT ("\nCompiling pattern: ");
2488 unsigned debug_count
;
2490 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2491 putchar (pattern
[debug_count
]);
2496 /* Initialize the compile stack. */
2497 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2498 if (compile_stack
.stack
== NULL
)
2501 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2502 compile_stack
.avail
= 0;
2504 range_table_work
.table
= 0;
2505 range_table_work
.allocated
= 0;
2507 /* Initialize the pattern buffer. */
2509 bufp
->syntax
= syntax
;
2511 bufp
->fastmap_accurate
= 0;
2512 bufp
->not_bol
= bufp
->not_eol
= 0;
2513 bufp
->used_syntax
= 0;
2515 /* Set `used' to zero, so that if we return an error, the pattern
2516 printer (for debugging) will think there's no pattern. We reset it
2520 /* Always count groups, whether or not bufp->no_sub is set. */
2523 #if !defined emacs && !defined SYNTAX_TABLE
2524 /* Initialize the syntax table. */
2525 init_syntax_once ();
2528 if (bufp
->allocated
== 0)
2531 { /* If zero allocated, but buffer is non-null, try to realloc
2532 enough space. This loses if buffer's address is bogus, but
2533 that is the user's responsibility. */
2534 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2537 { /* Caller did not allocate a buffer. Do it for them. */
2538 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2540 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2542 bufp
->allocated
= INIT_BUF_SIZE
;
2545 begalt
= b
= bufp
->buffer
;
2547 /* Loop through the uncompiled pattern until we're at the end. */
2553 /* If this is the end of an included regexp,
2554 pop back to the main regexp and try again. */
2558 pattern
= main_pattern
;
2564 /* If this is the end of the main regexp, we are done. */
2577 /* If there's no special whitespace regexp, treat
2578 spaces normally. And don't try to do this recursively. */
2579 if (!whitespace_regexp
|| in_subpattern
)
2582 /* Peek past following spaces. */
2589 /* If the spaces are followed by a repetition op,
2590 treat them normally. */
2592 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2593 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2596 /* Replace the spaces with the whitespace regexp. */
2600 main_pattern
= pattern
;
2601 p
= pattern
= (re_char
*) whitespace_regexp
;
2602 pend
= p
+ strlen (whitespace_regexp
);
2609 if ( /* If at start of pattern, it's an operator. */
2611 /* If context independent, it's an operator. */
2612 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2613 /* Otherwise, depends on what's come before. */
2614 || at_begline_loc_p (pattern
, p
, syntax
))
2615 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2624 if ( /* If at end of pattern, it's an operator. */
2626 /* If context independent, it's an operator. */
2627 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2628 /* Otherwise, depends on what's next. */
2629 || at_endline_loc_p (p
, pend
, syntax
))
2630 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2639 if ((syntax
& RE_BK_PLUS_QM
)
2640 || (syntax
& RE_LIMITED_OPS
))
2645 /* If there is no previous pattern... */
2648 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2649 FREE_STACK_RETURN (REG_BADRPT
);
2650 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2655 /* 1 means zero (many) matches is allowed. */
2656 boolean zero_times_ok
= 0, many_times_ok
= 0;
2659 /* If there is a sequence of repetition chars, collapse it
2660 down to just one (the right one). We can't combine
2661 interval operators with these because of, e.g., `a{2}*',
2662 which should only match an even number of `a's. */
2666 if ((syntax
& RE_FRUGAL
)
2667 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2671 zero_times_ok
|= c
!= '+';
2672 many_times_ok
|= c
!= '?';
2678 || (!(syntax
& RE_BK_PLUS_QM
)
2679 && (*p
== '+' || *p
== '?')))
2681 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2684 FREE_STACK_RETURN (REG_EESCAPE
);
2685 if (p
[1] == '+' || p
[1] == '?')
2686 PATFETCH (c
); /* Gobble up the backslash. */
2692 /* If we get here, we found another repeat character. */
2696 /* Star, etc. applied to an empty pattern is equivalent
2697 to an empty pattern. */
2698 if (!laststart
|| laststart
== b
)
2701 /* Now we know whether or not zero matches is allowed
2702 and also whether or not two or more matches is allowed. */
2707 boolean simple
= skip_one_char (laststart
) == b
;
2708 size_t startoffset
= 0;
2710 /* Check if the loop can match the empty string. */
2711 (simple
|| !analyze_first (laststart
, b
, NULL
, 0))
2712 ? on_failure_jump
: on_failure_jump_loop
;
2713 assert (skip_one_char (laststart
) <= b
);
2715 if (!zero_times_ok
&& simple
)
2716 { /* Since simple * loops can be made faster by using
2717 on_failure_keep_string_jump, we turn simple P+
2718 into PP* if P is simple. */
2719 unsigned char *p1
, *p2
;
2720 startoffset
= b
- laststart
;
2721 GET_BUFFER_SPACE (startoffset
);
2722 p1
= b
; p2
= laststart
;
2728 GET_BUFFER_SPACE (6);
2731 STORE_JUMP (ofj
, b
, b
+ 6);
2733 /* Simple * loops can use on_failure_keep_string_jump
2734 depending on what follows. But since we don't know
2735 that yet, we leave the decision up to
2736 on_failure_jump_smart. */
2737 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2738 laststart
+ startoffset
, b
+ 6);
2740 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2745 /* A simple ? pattern. */
2746 assert (zero_times_ok
);
2747 GET_BUFFER_SPACE (3);
2748 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2752 else /* not greedy */
2753 { /* I wish the greedy and non-greedy cases could be merged. */
2755 GET_BUFFER_SPACE (7); /* We might use less. */
2758 boolean emptyp
= analyze_first (laststart
, b
, NULL
, 0);
2760 /* The non-greedy multiple match looks like
2761 a repeat..until: we only need a conditional jump
2762 at the end of the loop. */
2763 if (emptyp
) BUF_PUSH (no_op
);
2764 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2765 : on_failure_jump
, b
, laststart
);
2769 /* The repeat...until naturally matches one or more.
2770 To also match zero times, we need to first jump to
2771 the end of the loop (its conditional jump). */
2772 INSERT_JUMP (jump
, laststart
, b
);
2778 /* non-greedy a?? */
2779 INSERT_JUMP (jump
, laststart
, b
+ 3);
2781 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2800 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2802 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2804 /* Ensure that we have enough space to push a charset: the
2805 opcode, the length count, and the bitset; 34 bytes in all. */
2806 GET_BUFFER_SPACE (34);
2810 /* We test `*p == '^' twice, instead of using an if
2811 statement, so we only need one BUF_PUSH. */
2812 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2816 /* Remember the first position in the bracket expression. */
2819 /* Push the number of bytes in the bitmap. */
2820 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2822 /* Clear the whole map. */
2823 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2825 /* charset_not matches newline according to a syntax bit. */
2826 if ((re_opcode_t
) b
[-2] == charset_not
2827 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2828 SET_LIST_BIT ('\n');
2830 /* Read in characters and ranges, setting map bits. */
2833 boolean escaped_char
= false;
2834 const unsigned char *p2
= p
;
2838 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2840 /* See if we're at the beginning of a possible character
2842 if (syntax
& RE_CHAR_CLASSES
&&
2843 (cc
= re_wctype_parse(&p
, pend
- p
)) != -1)
2846 FREE_STACK_RETURN (REG_ECTYPE
);
2849 FREE_STACK_RETURN (REG_EBRACK
);
2852 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2853 if (re_iswctype (btowc (ch
), cc
))
2856 if (c
< (1 << BYTEWIDTH
))
2860 /* Most character classes in a multibyte match just set
2861 a flag. Exceptions are is_blank, is_digit, is_cntrl, and
2862 is_xdigit, since they can only match ASCII characters.
2863 We don't need to handle them for multibyte. */
2865 /* Setup the gl_state object to its buffer-defined value.
2866 This hardcodes the buffer-global syntax-table for ASCII
2867 chars, while the other chars will obey syntax-table
2868 properties. It's not ideal, but it's the way it's been
2870 SETUP_BUFFER_SYNTAX_TABLE ();
2872 for (c
= 0; c
< 0x80; ++c
)
2873 if (re_iswctype (c
, cc
))
2879 if (ASCII_CHAR_P (c1
))
2881 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
2884 SET_RANGE_TABLE_WORK_AREA_BIT
2885 (range_table_work
, re_wctype_to_bit (cc
));
2887 /* In most cases the matching rule for char classes only
2888 uses the syntax table for multibyte chars, so that the
2889 content of the syntax-table is not hardcoded in the
2890 range_table. SPACE and WORD are the two exceptions. */
2891 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2892 bufp
->used_syntax
= 1;
2894 /* Repeat the loop. */
2898 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2899 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2900 So the translation is done later in a loop. Example:
2901 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2904 /* \ might escape characters inside [...] and [^...]. */
2905 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2907 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2910 escaped_char
= true;
2914 /* Could be the end of the bracket expression. If it's
2915 not (i.e., when the bracket expression is `[]' so
2916 far), the ']' character bit gets set way below. */
2917 if (c
== ']' && p2
!= p1
)
2921 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2924 /* Discard the `-'. */
2927 /* Fetch the character which ends the range. */
2930 if (CHAR_BYTE8_P (c1
)
2931 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
2932 /* Treat the range from a multibyte character to
2933 raw-byte character as empty. */
2938 /* Range from C to C. */
2943 if (syntax
& RE_NO_EMPTY_RANGES
)
2944 FREE_STACK_RETURN (REG_ERANGEX
);
2945 /* Else, repeat the loop. */
2950 /* Set the range into bitmap */
2951 for (; c
<= c1
; c
++)
2954 if (ch
< (1 << BYTEWIDTH
))
2961 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
2963 if (CHAR_BYTE8_P (c1
))
2964 c
= BYTE8_TO_CHAR (128);
2968 if (CHAR_BYTE8_P (c
))
2970 c
= CHAR_TO_BYTE8 (c
);
2971 c1
= CHAR_TO_BYTE8 (c1
);
2972 for (; c
<= c1
; c
++)
2977 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
2981 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
2988 /* Discard any (non)matching list bytes that are all 0 at the
2989 end of the map. Decrease the map-length byte too. */
2990 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2994 /* Build real range table from work area. */
2995 if (RANGE_TABLE_WORK_USED (range_table_work
)
2996 || RANGE_TABLE_WORK_BITS (range_table_work
))
2999 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3001 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3002 bytes for flags, two for COUNT, and three bytes for
3004 GET_BUFFER_SPACE (4 + used
* 3);
3006 /* Indicate the existence of range table. */
3007 laststart
[1] |= 0x80;
3009 /* Store the character class flag bits into the range table.
3010 If not in emacs, these flag bits are always 0. */
3011 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3012 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3014 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3015 for (i
= 0; i
< used
; i
++)
3016 STORE_CHARACTER_AND_INCR
3017 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3024 if (syntax
& RE_NO_BK_PARENS
)
3031 if (syntax
& RE_NO_BK_PARENS
)
3038 if (syntax
& RE_NEWLINE_ALT
)
3045 if (syntax
& RE_NO_BK_VBAR
)
3052 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3053 goto handle_interval
;
3059 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3061 /* Do not translate the character after the \, so that we can
3062 distinguish, e.g., \B from \b, even if we normally would
3063 translate, e.g., B to b. */
3069 if (syntax
& RE_NO_BK_PARENS
)
3070 goto normal_backslash
;
3075 regnum_t regnum
= 0;
3078 /* Look for a special (?...) construct */
3079 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3081 PATFETCH (c
); /* Gobble up the '?'. */
3087 case ':': shy
= 1; break;
3089 /* An explicitly specified regnum must start
3092 FREE_STACK_RETURN (REG_BADPAT
);
3094 case '1': case '2': case '3': case '4':
3095 case '5': case '6': case '7': case '8': case '9':
3096 regnum
= 10*regnum
+ (c
- '0'); break;
3098 /* Only (?:...) is supported right now. */
3099 FREE_STACK_RETURN (REG_BADPAT
);
3106 regnum
= ++bufp
->re_nsub
;
3108 { /* It's actually not shy, but explicitly numbered. */
3110 if (regnum
> bufp
->re_nsub
)
3111 bufp
->re_nsub
= regnum
;
3112 else if (regnum
> bufp
->re_nsub
3113 /* Ideally, we'd want to check that the specified
3114 group can't have matched (i.e. all subgroups
3115 using the same regnum are in other branches of
3116 OR patterns), but we don't currently keep track
3117 of enough info to do that easily. */
3118 || group_in_compile_stack (compile_stack
, regnum
))
3119 FREE_STACK_RETURN (REG_BADPAT
);
3122 /* It's really shy. */
3123 regnum
= - bufp
->re_nsub
;
3125 if (COMPILE_STACK_FULL
)
3127 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3128 compile_stack_elt_t
);
3129 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3131 compile_stack
.size
<<= 1;
3134 /* These are the values to restore when we hit end of this
3135 group. They are all relative offsets, so that if the
3136 whole pattern moves because of realloc, they will still
3138 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3139 COMPILE_STACK_TOP
.fixup_alt_jump
3140 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3141 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3142 COMPILE_STACK_TOP
.regnum
= regnum
;
3144 /* Do not push a start_memory for groups beyond the last one
3145 we can represent in the compiled pattern. */
3146 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3147 BUF_PUSH_2 (start_memory
, regnum
);
3149 compile_stack
.avail
++;
3154 /* If we've reached MAX_REGNUM groups, then this open
3155 won't actually generate any code, so we'll have to
3156 clear pending_exact explicitly. */
3162 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3164 if (COMPILE_STACK_EMPTY
)
3166 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3167 goto normal_backslash
;
3169 FREE_STACK_RETURN (REG_ERPAREN
);
3175 /* See similar code for backslashed left paren above. */
3176 if (COMPILE_STACK_EMPTY
)
3178 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3181 FREE_STACK_RETURN (REG_ERPAREN
);
3184 /* Since we just checked for an empty stack above, this
3185 ``can't happen''. */
3186 assert (compile_stack
.avail
!= 0);
3188 /* We don't just want to restore into `regnum', because
3189 later groups should continue to be numbered higher,
3190 as in `(ab)c(de)' -- the second group is #2. */
3193 compile_stack
.avail
--;
3194 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3196 = COMPILE_STACK_TOP
.fixup_alt_jump
3197 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3199 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3200 regnum
= COMPILE_STACK_TOP
.regnum
;
3201 /* If we've reached MAX_REGNUM groups, then this open
3202 won't actually generate any code, so we'll have to
3203 clear pending_exact explicitly. */
3206 /* We're at the end of the group, so now we know how many
3207 groups were inside this one. */
3208 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3209 BUF_PUSH_2 (stop_memory
, regnum
);
3214 case '|': /* `\|'. */
3215 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3216 goto normal_backslash
;
3218 if (syntax
& RE_LIMITED_OPS
)
3221 /* Insert before the previous alternative a jump which
3222 jumps to this alternative if the former fails. */
3223 GET_BUFFER_SPACE (3);
3224 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3228 /* The alternative before this one has a jump after it
3229 which gets executed if it gets matched. Adjust that
3230 jump so it will jump to this alternative's analogous
3231 jump (put in below, which in turn will jump to the next
3232 (if any) alternative's such jump, etc.). The last such
3233 jump jumps to the correct final destination. A picture:
3239 If we are at `b', then fixup_alt_jump right now points to a
3240 three-byte space after `a'. We'll put in the jump, set
3241 fixup_alt_jump to right after `b', and leave behind three
3242 bytes which we'll fill in when we get to after `c'. */
3246 /* Mark and leave space for a jump after this alternative,
3247 to be filled in later either by next alternative or
3248 when know we're at the end of a series of alternatives. */
3250 GET_BUFFER_SPACE (3);
3259 /* If \{ is a literal. */
3260 if (!(syntax
& RE_INTERVALS
)
3261 /* If we're at `\{' and it's not the open-interval
3263 || (syntax
& RE_NO_BK_BRACES
))
3264 goto normal_backslash
;
3268 /* If got here, then the syntax allows intervals. */
3270 /* At least (most) this many matches must be made. */
3271 int lower_bound
= 0, upper_bound
= -1;
3275 GET_INTERVAL_COUNT (lower_bound
);
3278 GET_INTERVAL_COUNT (upper_bound
);
3280 /* Interval such as `{1}' => match exactly once. */
3281 upper_bound
= lower_bound
;
3284 || (0 <= upper_bound
&& upper_bound
< lower_bound
))
3285 FREE_STACK_RETURN (REG_BADBR
);
3287 if (!(syntax
& RE_NO_BK_BRACES
))
3290 FREE_STACK_RETURN (REG_BADBR
);
3292 FREE_STACK_RETURN (REG_EESCAPE
);
3297 FREE_STACK_RETURN (REG_BADBR
);
3299 /* We just parsed a valid interval. */
3301 /* If it's invalid to have no preceding re. */
3304 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3305 FREE_STACK_RETURN (REG_BADRPT
);
3306 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3309 goto unfetch_interval
;
3312 if (upper_bound
== 0)
3313 /* If the upper bound is zero, just drop the sub pattern
3316 else if (lower_bound
== 1 && upper_bound
== 1)
3317 /* Just match it once: nothing to do here. */
3320 /* Otherwise, we have a nontrivial interval. When
3321 we're all done, the pattern will look like:
3322 set_number_at <jump count> <upper bound>
3323 set_number_at <succeed_n count> <lower bound>
3324 succeed_n <after jump addr> <succeed_n count>
3326 jump_n <succeed_n addr> <jump count>
3327 (The upper bound and `jump_n' are omitted if
3328 `upper_bound' is 1, though.) */
3330 { /* If the upper bound is > 1, we need to insert
3331 more at the end of the loop. */
3332 unsigned int nbytes
= (upper_bound
< 0 ? 3
3333 : upper_bound
> 1 ? 5 : 0);
3334 unsigned int startoffset
= 0;
3336 GET_BUFFER_SPACE (20); /* We might use less. */
3338 if (lower_bound
== 0)
3340 /* A succeed_n that starts with 0 is really a
3341 a simple on_failure_jump_loop. */
3342 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3348 /* Initialize lower bound of the `succeed_n', even
3349 though it will be set during matching by its
3350 attendant `set_number_at' (inserted next),
3351 because `re_compile_fastmap' needs to know.
3352 Jump to the `jump_n' we might insert below. */
3353 INSERT_JUMP2 (succeed_n
, laststart
,
3358 /* Code to initialize the lower bound. Insert
3359 before the `succeed_n'. The `5' is the last two
3360 bytes of this `set_number_at', plus 3 bytes of
3361 the following `succeed_n'. */
3362 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3367 if (upper_bound
< 0)
3369 /* A negative upper bound stands for infinity,
3370 in which case it degenerates to a plain jump. */
3371 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3374 else if (upper_bound
> 1)
3375 { /* More than one repetition is allowed, so
3376 append a backward jump to the `succeed_n'
3377 that starts this interval.
3379 When we've reached this during matching,
3380 we'll have matched the interval once, so
3381 jump back only `upper_bound - 1' times. */
3382 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3386 /* The location we want to set is the second
3387 parameter of the `jump_n'; that is `b-2' as
3388 an absolute address. `laststart' will be
3389 the `set_number_at' we're about to insert;
3390 `laststart+3' the number to set, the source
3391 for the relative address. But we are
3392 inserting into the middle of the pattern --
3393 so everything is getting moved up by 5.
3394 Conclusion: (b - 2) - (laststart + 3) + 5,
3395 i.e., b - laststart.
3397 We insert this at the beginning of the loop
3398 so that if we fail during matching, we'll
3399 reinitialize the bounds. */
3400 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3401 upper_bound
- 1, b
);
3406 beg_interval
= NULL
;
3411 /* If an invalid interval, match the characters as literals. */
3412 assert (beg_interval
);
3414 beg_interval
= NULL
;
3416 /* normal_char and normal_backslash need `c'. */
3419 if (!(syntax
& RE_NO_BK_BRACES
))
3421 assert (p
> pattern
&& p
[-1] == '\\');
3422 goto normal_backslash
;
3436 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3442 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3448 BUF_PUSH_2 (categoryspec
, c
);
3454 BUF_PUSH_2 (notcategoryspec
, c
);
3460 if (syntax
& RE_NO_GNU_OPS
)
3463 BUF_PUSH_2 (syntaxspec
, Sword
);
3468 if (syntax
& RE_NO_GNU_OPS
)
3471 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3476 if (syntax
& RE_NO_GNU_OPS
)
3483 if (syntax
& RE_NO_GNU_OPS
)
3490 if (syntax
& RE_NO_GNU_OPS
)
3499 FREE_STACK_RETURN (REG_BADPAT
);
3503 if (syntax
& RE_NO_GNU_OPS
)
3505 BUF_PUSH (wordbound
);
3509 if (syntax
& RE_NO_GNU_OPS
)
3511 BUF_PUSH (notwordbound
);
3515 if (syntax
& RE_NO_GNU_OPS
)
3521 if (syntax
& RE_NO_GNU_OPS
)
3526 case '1': case '2': case '3': case '4': case '5':
3527 case '6': case '7': case '8': case '9':
3531 if (syntax
& RE_NO_BK_REFS
)
3532 goto normal_backslash
;
3536 if (reg
> bufp
->re_nsub
|| reg
< 1
3537 /* Can't back reference to a subexp before its end. */
3538 || group_in_compile_stack (compile_stack
, reg
))
3539 FREE_STACK_RETURN (REG_ESUBREG
);
3542 BUF_PUSH_2 (duplicate
, reg
);
3549 if (syntax
& RE_BK_PLUS_QM
)
3552 goto normal_backslash
;
3556 /* You might think it would be useful for \ to mean
3557 not to translate; but if we don't translate it
3558 it will never match anything. */
3565 /* Expects the character in `c'. */
3567 /* If no exactn currently being built. */
3570 /* If last exactn not at current position. */
3571 || pending_exact
+ *pending_exact
+ 1 != b
3573 /* We have only one byte following the exactn for the count. */
3574 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3576 /* If followed by a repetition operator. */
3577 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3578 || ((syntax
& RE_BK_PLUS_QM
)
3579 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3580 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3581 || ((syntax
& RE_INTERVALS
)
3582 && ((syntax
& RE_NO_BK_BRACES
)
3583 ? p
!= pend
&& *p
== '{'
3584 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3586 /* Start building a new exactn. */
3590 BUF_PUSH_2 (exactn
, 0);
3591 pending_exact
= b
- 1;
3594 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3601 len
= CHAR_STRING (c
, b
);
3606 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3607 if (! CHAR_BYTE8_P (c1
))
3609 re_wchar_t c2
= TRANSLATE (c1
);
3611 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3617 (*pending_exact
) += len
;
3622 } /* while p != pend */
3625 /* Through the pattern now. */
3629 if (!COMPILE_STACK_EMPTY
)
3630 FREE_STACK_RETURN (REG_EPAREN
);
3632 /* If we don't want backtracking, force success
3633 the first time we reach the end of the compiled pattern. */
3634 if (!posix_backtracking
)
3637 /* We have succeeded; set the length of the buffer. */
3638 bufp
->used
= b
- bufp
->buffer
;
3643 re_compile_fastmap (bufp
);
3644 DEBUG_PRINT ("\nCompiled pattern: \n");
3645 print_compiled_pattern (bufp
);
3650 #ifndef MATCH_MAY_ALLOCATE
3651 /* Initialize the failure stack to the largest possible stack. This
3652 isn't necessary unless we're trying to avoid calling alloca in
3653 the search and match routines. */
3655 int num_regs
= bufp
->re_nsub
+ 1;
3657 if (fail_stack
.size
< emacs_re_max_failures
* TYPICAL_FAILURE_SIZE
)
3659 fail_stack
.size
= emacs_re_max_failures
* TYPICAL_FAILURE_SIZE
;
3660 falk_stack
.stack
= realloc (fail_stack
.stack
,
3661 fail_stack
.size
* sizeof *falk_stack
.stack
);
3664 regex_grow_registers (num_regs
);
3666 #endif /* not MATCH_MAY_ALLOCATE */
3668 FREE_STACK_RETURN (REG_NOERROR
);
3673 # undef posix_backtracking
3675 } /* regex_compile */
3677 /* Subroutines for `regex_compile'. */
3679 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3682 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3684 *loc
= (unsigned char) op
;
3685 STORE_NUMBER (loc
+ 1, arg
);
3689 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3692 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3694 *loc
= (unsigned char) op
;
3695 STORE_NUMBER (loc
+ 1, arg1
);
3696 STORE_NUMBER (loc
+ 3, arg2
);
3700 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3701 for OP followed by two-byte integer parameter ARG. */
3704 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3706 register unsigned char *pfrom
= end
;
3707 register unsigned char *pto
= end
+ 3;
3709 while (pfrom
!= loc
)
3712 store_op1 (op
, loc
, arg
);
3716 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3719 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3721 register unsigned char *pfrom
= end
;
3722 register unsigned char *pto
= end
+ 5;
3724 while (pfrom
!= loc
)
3727 store_op2 (op
, loc
, arg1
, arg2
);
3731 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3732 after an alternative or a begin-subexpression. We assume there is at
3733 least one character before the ^. */
3736 at_begline_loc_p (re_char
*pattern
, re_char
*p
, reg_syntax_t syntax
)
3738 re_char
*prev
= p
- 2;
3739 boolean odd_backslashes
;
3741 /* After a subexpression? */
3743 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3745 /* After an alternative? */
3746 else if (*prev
== '|')
3747 odd_backslashes
= (syntax
& RE_NO_BK_VBAR
) == 0;
3749 /* After a shy subexpression? */
3750 else if (*prev
== ':' && (syntax
& RE_SHY_GROUPS
))
3752 /* Skip over optional regnum. */
3753 while (prev
- 1 >= pattern
&& prev
[-1] >= '0' && prev
[-1] <= '9')
3756 if (!(prev
- 2 >= pattern
3757 && prev
[-1] == '?' && prev
[-2] == '('))
3760 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3765 /* Count the number of preceding backslashes. */
3767 while (prev
- 1 >= pattern
&& prev
[-1] == '\\')
3769 return (p
- prev
) & odd_backslashes
;
3773 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3774 at least one character after the $, i.e., `P < PEND'. */
3777 at_endline_loc_p (re_char
*p
, re_char
*pend
, reg_syntax_t syntax
)
3780 boolean next_backslash
= *next
== '\\';
3781 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3784 /* Before a subexpression? */
3785 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3786 : next_backslash
&& next_next
&& *next_next
== ')')
3787 /* Before an alternative? */
3788 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3789 : next_backslash
&& next_next
&& *next_next
== '|');
3793 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3794 false if it's not. */
3797 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3799 ssize_t this_element
;
3801 for (this_element
= compile_stack
.avail
- 1;
3804 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3811 If fastmap is non-NULL, go through the pattern and fill fastmap
3812 with all the possible leading chars. If fastmap is NULL, don't
3813 bother filling it up (obviously) and only return whether the
3814 pattern could potentially match the empty string.
3816 Return 1 if p..pend might match the empty string.
3817 Return 0 if p..pend matches at least one char.
3818 Return -1 if fastmap was not updated accurately. */
3821 analyze_first (re_char
*p
, re_char
*pend
, char *fastmap
,
3822 const int multibyte
)
3827 /* If all elements for base leading-codes in fastmap is set, this
3828 flag is set true. */
3829 boolean match_any_multibyte_characters
= false;
3833 /* The loop below works as follows:
3834 - It has a working-list kept in the PATTERN_STACK and which basically
3835 starts by only containing a pointer to the first operation.
3836 - If the opcode we're looking at is a match against some set of
3837 chars, then we add those chars to the fastmap and go on to the
3838 next work element from the worklist (done via `break').
3839 - If the opcode is a control operator on the other hand, we either
3840 ignore it (if it's meaningless at this point, such as `start_memory')
3841 or execute it (if it's a jump). If the jump has several destinations
3842 (i.e. `on_failure_jump'), then we push the other destination onto the
3844 We guarantee termination by ignoring backward jumps (more or less),
3845 so that `p' is monotonically increasing. More to the point, we
3846 never set `p' (or push) anything `<= p1'. */
3850 /* `p1' is used as a marker of how far back a `on_failure_jump'
3851 can go without being ignored. It is normally equal to `p'
3852 (which prevents any backward `on_failure_jump') except right
3853 after a plain `jump', to allow patterns such as:
3856 10: on_failure_jump 3
3857 as used for the *? operator. */
3866 /* If the first character has to match a backreference, that means
3867 that the group was empty (since it already matched). Since this
3868 is the only case that interests us here, we can assume that the
3869 backreference must match the empty string. */
3874 /* Following are the cases which match a character. These end
3880 /* If multibyte is nonzero, the first byte of each
3881 character is an ASCII or a leading code. Otherwise,
3882 each byte is a character. Thus, this works in both
3887 /* For the case of matching this unibyte regex
3888 against multibyte, we must set a leading code of
3889 the corresponding multibyte character. */
3890 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
3892 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
3899 /* We could put all the chars except for \n (and maybe \0)
3900 but we don't bother since it is generally not worth it. */
3901 if (!fastmap
) break;
3906 if (!fastmap
) break;
3908 /* Chars beyond end of bitmap are possible matches. */
3909 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3910 j
< (1 << BYTEWIDTH
); j
++)
3915 if (!fastmap
) break;
3916 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3917 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3919 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3923 if (/* Any leading code can possibly start a character
3924 which doesn't match the specified set of characters. */
3927 /* If we can match a character class, we can match any
3928 multibyte characters. */
3929 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3930 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3933 if (match_any_multibyte_characters
== false)
3935 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3936 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3938 match_any_multibyte_characters
= true;
3942 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3943 && match_any_multibyte_characters
== false)
3945 /* Set fastmap[I] to 1 where I is a leading code of each
3946 multibyte character in the range table. */
3948 unsigned char lc1
, lc2
;
3950 /* Make P points the range table. `+ 2' is to skip flag
3951 bits for a character class. */
3952 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3954 /* Extract the number of ranges in range table into COUNT. */
3955 EXTRACT_NUMBER_AND_INCR (count
, p
);
3956 for (; count
> 0; count
--, p
+= 3)
3958 /* Extract the start and end of each range. */
3959 EXTRACT_CHARACTER (c
, p
);
3960 lc1
= CHAR_LEADING_CODE (c
);
3962 EXTRACT_CHARACTER (c
, p
);
3963 lc2
= CHAR_LEADING_CODE (c
);
3964 for (j
= lc1
; j
<= lc2
; j
++)
3973 if (!fastmap
) break;
3975 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
3977 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3978 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
3982 /* This match depends on text properties. These end with
3983 aborting optimizations. */
3987 case notcategoryspec
:
3988 if (!fastmap
) break;
3989 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
3991 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
3992 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
3995 /* Any leading code can possibly start a character which
3996 has or doesn't has the specified category. */
3997 if (match_any_multibyte_characters
== false)
3999 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4000 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4002 match_any_multibyte_characters
= true;
4006 /* All cases after this match the empty string. These end with
4026 EXTRACT_NUMBER_AND_INCR (j
, p
);
4028 /* Backward jumps can only go back to code that we've already
4029 visited. `re_compile' should make sure this is true. */
4034 case on_failure_jump
:
4035 case on_failure_keep_string_jump
:
4036 case on_failure_jump_loop
:
4037 case on_failure_jump_nastyloop
:
4038 case on_failure_jump_smart
:
4044 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4045 to jump back to "just after here". */
4047 case on_failure_jump
:
4048 case on_failure_keep_string_jump
:
4049 case on_failure_jump_nastyloop
:
4050 case on_failure_jump_loop
:
4051 case on_failure_jump_smart
:
4052 EXTRACT_NUMBER_AND_INCR (j
, p
);
4054 ; /* Backward jump to be ignored. */
4056 { /* We have to look down both arms.
4057 We first go down the "straight" path so as to minimize
4058 stack usage when going through alternatives. */
4059 int r
= analyze_first (p
, pend
, fastmap
, multibyte
);
4067 /* This code simply does not properly handle forward jump_n. */
4068 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4070 /* jump_n can either jump or fall through. The (backward) jump
4071 case has already been handled, so we only need to look at the
4072 fallthrough case. */
4076 /* If N == 0, it should be an on_failure_jump_loop instead. */
4077 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4079 /* We only care about one iteration of the loop, so we don't
4080 need to consider the case where this behaves like an
4097 abort (); /* We have listed all the cases. */
4100 /* Getting here means we have found the possible starting
4101 characters for one path of the pattern -- and that the empty
4102 string does not match. We need not follow this path further. */
4106 /* We reached the end without matching anything. */
4109 } /* analyze_first */
4111 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4112 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4113 characters can start a string that matches the pattern. This fastmap
4114 is used by re_search to skip quickly over impossible starting points.
4116 Character codes above (1 << BYTEWIDTH) are not represented in the
4117 fastmap, but the leading codes are represented. Thus, the fastmap
4118 indicates which character sets could start a match.
4120 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4121 area as BUFP->fastmap.
4123 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4126 Returns 0 if we succeed, -2 if an internal error. */
4129 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4131 char *fastmap
= bufp
->fastmap
;
4134 assert (fastmap
&& bufp
->buffer
);
4136 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4137 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4139 analysis
= analyze_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4140 fastmap
, RE_MULTIBYTE_P (bufp
));
4141 bufp
->can_be_null
= (analysis
!= 0);
4143 } /* re_compile_fastmap */
4145 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4146 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4147 this memory for recording register information. STARTS and ENDS
4148 must be allocated using the malloc library routine, and must each
4149 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4151 If NUM_REGS == 0, then subsequent matches should allocate their own
4154 Unless this function is called, the first search or match using
4155 PATTERN_BUFFER will allocate its own register data, without
4156 freeing the old data. */
4159 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4163 bufp
->regs_allocated
= REGS_REALLOCATE
;
4164 regs
->num_regs
= num_regs
;
4165 regs
->start
= starts
;
4170 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4172 regs
->start
= regs
->end
= 0;
4175 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4177 /* Searching routines. */
4179 /* Like re_search_2, below, but only one string is specified, and
4180 doesn't let you say where to stop matching. */
4183 re_search (struct re_pattern_buffer
*bufp
, const char *string
, size_t size
,
4184 ssize_t startpos
, ssize_t range
, struct re_registers
*regs
)
4186 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4189 WEAK_ALIAS (__re_search
, re_search
)
4191 /* Head address of virtual concatenation of string. */
4192 #define HEAD_ADDR_VSTRING(P) \
4193 (((P) >= size1 ? string2 : string1))
4195 /* Address of POS in the concatenation of virtual string. */
4196 #define POS_ADDR_VSTRING(POS) \
4197 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4199 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4200 virtual concatenation of STRING1 and STRING2, starting first at index
4201 STARTPOS, then at STARTPOS + 1, and so on.
4203 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4205 RANGE is how far to scan while trying to match. RANGE = 0 means try
4206 only at STARTPOS; in general, the last start tried is STARTPOS +
4209 In REGS, return the indices of the virtual concatenation of STRING1
4210 and STRING2 that matched the entire BUFP->buffer and its contained
4213 Do not consider matching one past the index STOP in the virtual
4214 concatenation of STRING1 and STRING2.
4216 We return either the position in the strings at which the match was
4217 found, -1 if no match, or -2 if error (such as failure
4221 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, size_t size1
,
4222 const char *str2
, size_t size2
, ssize_t startpos
, ssize_t range
,
4223 struct re_registers
*regs
, ssize_t stop
)
4226 re_char
*string1
= (re_char
*) str1
;
4227 re_char
*string2
= (re_char
*) str2
;
4228 register char *fastmap
= bufp
->fastmap
;
4229 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4230 size_t total_size
= size1
+ size2
;
4231 ssize_t endpos
= startpos
+ range
;
4232 boolean anchored_start
;
4233 /* Nonzero if we are searching multibyte string. */
4234 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4236 /* Check for out-of-range STARTPOS. */
4237 if (startpos
< 0 || startpos
> total_size
)
4240 /* Fix up RANGE if it might eventually take us outside
4241 the virtual concatenation of STRING1 and STRING2.
4242 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4244 range
= 0 - startpos
;
4245 else if (endpos
> total_size
)
4246 range
= total_size
- startpos
;
4248 /* If the search isn't to be a backwards one, don't waste time in a
4249 search for a pattern anchored at beginning of buffer. */
4250 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4259 /* In a forward search for something that starts with \=.
4260 don't keep searching past point. */
4261 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4263 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4269 /* Update the fastmap now if not correct already. */
4270 if (fastmap
&& !bufp
->fastmap_accurate
)
4271 re_compile_fastmap (bufp
);
4273 /* See whether the pattern is anchored. */
4274 anchored_start
= (bufp
->buffer
[0] == begline
);
4277 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4279 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4281 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4285 /* Loop through the string, looking for a place to start matching. */
4288 /* If the pattern is anchored,
4289 skip quickly past places we cannot match.
4290 We don't bother to treat startpos == 0 specially
4291 because that case doesn't repeat. */
4292 if (anchored_start
&& startpos
> 0)
4294 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4295 : string2
[startpos
- size1
- 1])
4300 /* If a fastmap is supplied, skip quickly over characters that
4301 cannot be the start of a match. If the pattern can match the
4302 null string, however, we don't need to skip characters; we want
4303 the first null string. */
4304 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4306 register re_char
*d
;
4307 register re_wchar_t buf_ch
;
4309 d
= POS_ADDR_VSTRING (startpos
);
4311 if (range
> 0) /* Searching forwards. */
4313 ssize_t irange
= range
, lim
= 0;
4315 if (startpos
< size1
&& startpos
+ range
>= size1
)
4316 lim
= range
- (size1
- startpos
);
4318 /* Written out as an if-else to avoid testing `translate'
4320 if (RE_TRANSLATE_P (translate
))
4327 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4328 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4329 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4332 range
-= buf_charlen
;
4338 register re_wchar_t ch
, translated
;
4341 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4342 translated
= RE_TRANSLATE (translate
, ch
);
4343 if (translated
!= ch
4344 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4346 if (fastmap
[buf_ch
])
4359 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4360 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4362 range
-= buf_charlen
;
4366 while (range
> lim
&& !fastmap
[*d
])
4372 startpos
+= irange
- range
;
4374 else /* Searching backwards. */
4378 buf_ch
= STRING_CHAR (d
);
4379 buf_ch
= TRANSLATE (buf_ch
);
4380 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4385 register re_wchar_t ch
, translated
;
4388 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4389 translated
= TRANSLATE (ch
);
4390 if (translated
!= ch
4391 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4393 if (! fastmap
[TRANSLATE (buf_ch
)])
4399 /* If can't match the null string, and that's all we have left, fail. */
4400 if (range
>= 0 && startpos
== total_size
&& fastmap
4401 && !bufp
->can_be_null
)
4404 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4405 startpos
, regs
, stop
);
4418 /* Update STARTPOS to the next character boundary. */
4421 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4422 int len
= BYTES_BY_CHAR_HEAD (*p
);
4440 /* Update STARTPOS to the previous character boundary. */
4443 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4445 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4447 /* Find the head of multibyte form. */
4448 PREV_CHAR_BOUNDARY (p
, phead
);
4449 range
+= p0
- 1 - p
;
4453 startpos
-= p0
- 1 - p
;
4459 WEAK_ALIAS (__re_search_2
, re_search_2
)
4461 /* Declarations and macros for re_match_2. */
4463 static int bcmp_translate (re_char
*s1
, re_char
*s2
,
4464 register ssize_t len
,
4465 RE_TRANSLATE_TYPE translate
,
4466 const int multibyte
);
4468 /* This converts PTR, a pointer into one of the search strings `string1'
4469 and `string2' into an offset from the beginning of that string. */
4470 #define POINTER_TO_OFFSET(ptr) \
4471 (FIRST_STRING_P (ptr) \
4473 : (ptr) - string2 + (ptrdiff_t) size1)
4475 /* Call before fetching a character with *d. This switches over to
4476 string2 if necessary.
4477 Check re_match_2_internal for a discussion of why end_match_2 might
4478 not be within string2 (but be equal to end_match_1 instead). */
4479 #define PREFETCH() \
4482 /* End of string2 => fail. */ \
4483 if (dend == end_match_2) \
4485 /* End of string1 => advance to string2. */ \
4487 dend = end_match_2; \
4490 /* Call before fetching a char with *d if you already checked other limits.
4491 This is meant for use in lookahead operations like wordend, etc..
4492 where we might need to look at parts of the string that might be
4493 outside of the LIMITs (i.e past `stop'). */
4494 #define PREFETCH_NOLIMIT() \
4498 dend = end_match_2; \
4501 /* Test if at very beginning or at very end of the virtual concatenation
4502 of `string1' and `string2'. If only one string, it's `string2'. */
4503 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4504 #define AT_STRINGS_END(d) ((d) == end2)
4506 /* Disabled due to a compiler bug -- see comment at case wordbound */
4508 /* The comment at case wordbound is following one, but we don't use
4509 AT_WORD_BOUNDARY anymore to support multibyte form.
4511 The DEC Alpha C compiler 3.x generates incorrect code for the
4512 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4513 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4514 macro and introducing temporary variables works around the bug. */
4517 /* Test if D points to a character which is word-constituent. We have
4518 two special cases to check for: if past the end of string1, look at
4519 the first character in string2; and if before the beginning of
4520 string2, look at the last character in string1. */
4521 #define WORDCHAR_P(d) \
4522 (SYNTAX ((d) == end1 ? *string2 \
4523 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4526 /* Test if the character before D and the one at D differ with respect
4527 to being word-constituent. */
4528 #define AT_WORD_BOUNDARY(d) \
4529 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4530 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4533 /* Free everything we malloc. */
4534 #ifdef MATCH_MAY_ALLOCATE
4535 # define FREE_VAR(var) \
4543 # define FREE_VARIABLES() \
4545 REGEX_FREE_STACK (fail_stack.stack); \
4546 FREE_VAR (regstart); \
4547 FREE_VAR (regend); \
4548 FREE_VAR (best_regstart); \
4549 FREE_VAR (best_regend); \
4550 REGEX_SAFE_FREE (); \
4553 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4554 #endif /* not MATCH_MAY_ALLOCATE */
4557 /* Optimization routines. */
4559 /* If the operation is a match against one or more chars,
4560 return a pointer to the next operation, else return NULL. */
4562 skip_one_char (re_char
*p
)
4575 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4578 p
= CHARSET_RANGE_TABLE (p
- 1);
4579 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4580 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4583 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4590 case notcategoryspec
:
4602 /* Jump over non-matching operations. */
4604 skip_noops (re_char
*p
, re_char
*pend
)
4618 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4629 /* Test if C matches charset op. *PP points to the charset or charset_not
4630 opcode. When the function finishes, *PP will be advanced past that opcode.
4631 C is character to test (possibly after translations) and CORIG is original
4632 character (i.e. without any translations). UNIBYTE denotes whether c is
4633 unibyte or multibyte character. */
4635 execute_charset (re_char
**pp
, unsigned c
, unsigned corig
, bool unibyte
)
4637 re_char
*p
= *pp
, *rtp
= NULL
;
4638 bool not = (re_opcode_t
) *p
== charset_not
;
4640 if (CHARSET_RANGE_TABLE_EXISTS_P (p
))
4643 rtp
= CHARSET_RANGE_TABLE (p
);
4644 EXTRACT_NUMBER_AND_INCR (count
, rtp
);
4645 *pp
= CHARSET_RANGE_TABLE_END ((rtp
), (count
));
4648 *pp
+= 2 + CHARSET_BITMAP_SIZE (p
);
4650 if (unibyte
&& c
< (1 << BYTEWIDTH
))
4651 { /* Lookup bitmap. */
4652 /* Cast to `unsigned' instead of `unsigned char' in
4653 case the bit list is a full 32 bytes long. */
4654 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (p
) * BYTEWIDTH
)
4655 && p
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4661 int class_bits
= CHARSET_RANGE_TABLE_BITS (p
);
4662 re_wchar_t range_start
, range_end
;
4664 /* Sort tests by the most commonly used classes with some adjustment to which
4665 tests are easiest to perform. Take a look at comment in re_wctype_parse
4666 for table with frequencies of character class names. */
4668 if ((class_bits
& BIT_MULTIBYTE
) ||
4669 (class_bits
& BIT_ALNUM
&& ISALNUM (c
)) ||
4670 (class_bits
& BIT_ALPHA
&& ISALPHA (c
)) ||
4671 (class_bits
& BIT_SPACE
&& ISSPACE (c
)) ||
4672 (class_bits
& BIT_BLANK
&& ISBLANK (c
)) ||
4673 (class_bits
& BIT_WORD
&& ISWORD (c
)) ||
4674 ((class_bits
& BIT_UPPER
) &&
4675 (ISUPPER (c
) || (corig
!= c
&&
4676 c
== downcase (corig
) && ISLOWER (c
)))) ||
4677 ((class_bits
& BIT_LOWER
) &&
4678 (ISLOWER (c
) || (corig
!= c
&&
4679 c
== upcase (corig
) && ISUPPER(c
)))) ||
4680 (class_bits
& BIT_PUNCT
&& ISPUNCT (c
)) ||
4681 (class_bits
& BIT_GRAPH
&& ISGRAPH (c
)) ||
4682 (class_bits
& BIT_PRINT
&& ISPRINT (c
)))
4685 for (p
= *pp
; rtp
< p
; rtp
+= 2 * 3)
4687 EXTRACT_CHARACTER (range_start
, rtp
);
4688 EXTRACT_CHARACTER (range_end
, rtp
+ 3);
4689 if (range_start
<= c
&& c
<= range_end
)
4697 /* Non-zero if "p1 matches something" implies "p2 fails". */
4699 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, re_char
*p1
,
4703 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4704 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4706 assert (p1
>= bufp
->buffer
&& p1
< pend
4707 && p2
>= bufp
->buffer
&& p2
<= pend
);
4709 /* Skip over open/close-group commands.
4710 If what follows this loop is a ...+ construct,
4711 look at what begins its body, since we will have to
4712 match at least one of that. */
4713 p2
= skip_noops (p2
, pend
);
4714 /* The same skip can be done for p1, except that this function
4715 is only used in the case where p1 is a simple match operator. */
4716 /* p1 = skip_noops (p1, pend); */
4718 assert (p1
>= bufp
->buffer
&& p1
< pend
4719 && p2
>= bufp
->buffer
&& p2
<= pend
);
4721 op2
= p2
== pend
? succeed
: *p2
;
4727 /* If we're at the end of the pattern, we can change. */
4728 if (skip_one_char (p1
))
4730 DEBUG_PRINT (" End of pattern: fast loop.\n");
4738 register re_wchar_t c
4739 = (re_opcode_t
) *p2
== endline
? '\n'
4740 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4742 if ((re_opcode_t
) *p1
== exactn
)
4744 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4746 DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4751 else if ((re_opcode_t
) *p1
== charset
4752 || (re_opcode_t
) *p1
== charset_not
)
4754 if (!execute_charset (&p1
, c
, c
, !multibyte
|| IS_REAL_ASCII (c
)))
4756 DEBUG_PRINT (" No match => fast loop.\n");
4760 else if ((re_opcode_t
) *p1
== anychar
4763 DEBUG_PRINT (" . != \\n => fast loop.\n");
4771 if ((re_opcode_t
) *p1
== exactn
)
4772 /* Reuse the code above. */
4773 return mutually_exclusive_p (bufp
, p2
, p1
);
4775 /* It is hard to list up all the character in charset
4776 P2 if it includes multibyte character. Give up in
4778 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4780 /* Now, we are sure that P2 has no range table.
4781 So, for the size of bitmap in P2, `p2[1]' is
4782 enough. But P1 may have range table, so the
4783 size of bitmap table of P1 is extracted by
4784 using macro `CHARSET_BITMAP_SIZE'.
4786 In a multibyte case, we know that all the character
4787 listed in P2 is ASCII. In a unibyte case, P1 has only a
4788 bitmap table. So, in both cases, it is enough to test
4789 only the bitmap table of P1. */
4791 if ((re_opcode_t
) *p1
== charset
)
4794 /* We win if the charset inside the loop
4795 has no overlap with the one after the loop. */
4798 && idx
< CHARSET_BITMAP_SIZE (p1
));
4800 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4804 || idx
== CHARSET_BITMAP_SIZE (p1
))
4806 DEBUG_PRINT (" No match => fast loop.\n");
4810 else if ((re_opcode_t
) *p1
== charset_not
)
4813 /* We win if the charset_not inside the loop lists
4814 every character listed in the charset after. */
4815 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4816 if (! (p2
[2 + idx
] == 0
4817 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4818 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4823 DEBUG_PRINT (" No match => fast loop.\n");
4836 /* Reuse the code above. */
4837 return mutually_exclusive_p (bufp
, p2
, p1
);
4839 /* When we have two charset_not, it's very unlikely that
4840 they don't overlap. The union of the two sets of excluded
4841 chars should cover all possible chars, which, as a matter of
4842 fact, is virtually impossible in multibyte buffers. */
4848 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4850 return ((re_opcode_t
) *p1
== syntaxspec
4851 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4853 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4856 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4858 return ((re_opcode_t
) *p1
== notsyntaxspec
4859 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4861 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4864 return (((re_opcode_t
) *p1
== notsyntaxspec
4865 || (re_opcode_t
) *p1
== syntaxspec
)
4870 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4871 case notcategoryspec
:
4872 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4884 /* Matching routines. */
4886 #ifndef emacs /* Emacs never uses this. */
4887 /* re_match is like re_match_2 except it takes only a single string. */
4890 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
4891 size_t size
, ssize_t pos
, struct re_registers
*regs
)
4893 regoff_t result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
,
4894 size
, pos
, regs
, size
);
4897 WEAK_ALIAS (__re_match
, re_match
)
4898 #endif /* not emacs */
4900 /* re_match_2 matches the compiled pattern in BUFP against the
4901 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4902 and SIZE2, respectively). We start matching at POS, and stop
4905 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4906 store offsets for the substring each group matched in REGS. See the
4907 documentation for exactly how many groups we fill.
4909 We return -1 if no match, -2 if an internal error (such as the
4910 failure stack overflowing). Otherwise, we return the length of the
4911 matched substring. */
4914 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
,
4915 size_t size1
, const char *string2
, size_t size2
, ssize_t pos
,
4916 struct re_registers
*regs
, ssize_t stop
)
4922 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4923 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4924 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4927 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4928 (re_char
*) string2
, size2
,
4932 WEAK_ALIAS (__re_match_2
, re_match_2
)
4935 /* This is a separate function so that we can force an alloca cleanup
4938 re_match_2_internal (struct re_pattern_buffer
*bufp
, re_char
*string1
,
4939 size_t size1
, re_char
*string2
, size_t size2
,
4940 ssize_t pos
, struct re_registers
*regs
, ssize_t stop
)
4942 /* General temporaries. */
4946 /* Just past the end of the corresponding string. */
4947 re_char
*end1
, *end2
;
4949 /* Pointers into string1 and string2, just past the last characters in
4950 each to consider matching. */
4951 re_char
*end_match_1
, *end_match_2
;
4953 /* Where we are in the data, and the end of the current string. */
4956 /* Used sometimes to remember where we were before starting matching
4957 an operator so that we can go back in case of failure. This "atomic"
4958 behavior of matching opcodes is indispensable to the correctness
4959 of the on_failure_keep_string_jump optimization. */
4962 /* Where we are in the pattern, and the end of the pattern. */
4963 re_char
*p
= bufp
->buffer
;
4964 re_char
*pend
= p
+ bufp
->used
;
4966 /* We use this to map every character in the string. */
4967 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4969 /* Nonzero if BUFP is setup from a multibyte regex. */
4970 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4972 /* Nonzero if STRING1/STRING2 are multibyte. */
4973 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4975 /* Failure point stack. Each place that can handle a failure further
4976 down the line pushes a failure point on this stack. It consists of
4977 regstart, and regend for all registers corresponding to
4978 the subexpressions we're currently inside, plus the number of such
4979 registers, and, finally, two char *'s. The first char * is where
4980 to resume scanning the pattern; the second one is where to resume
4981 scanning the strings. */
4982 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4983 fail_stack_type fail_stack
;
4985 #ifdef DEBUG_COMPILES_ARGUMENTS
4986 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4989 #if defined REL_ALLOC && defined REGEX_MALLOC
4990 /* This holds the pointer to the failure stack, when
4991 it is allocated relocatably. */
4992 fail_stack_elt_t
*failure_stack_ptr
;
4995 /* We fill all the registers internally, independent of what we
4996 return, for use in backreferences. The number here includes
4997 an element for register zero. */
4998 size_t num_regs
= bufp
->re_nsub
+ 1;
5000 /* Information on the contents of registers. These are pointers into
5001 the input strings; they record just what was matched (on this
5002 attempt) by a subexpression part of the pattern, that is, the
5003 regnum-th regstart pointer points to where in the pattern we began
5004 matching and the regnum-th regend points to right after where we
5005 stopped matching the regnum-th subexpression. (The zeroth register
5006 keeps track of what the whole pattern matches.) */
5007 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5008 re_char
**regstart
, **regend
;
5011 /* The following record the register info as found in the above
5012 variables when we find a match better than any we've seen before.
5013 This happens as we backtrack through the failure points, which in
5014 turn happens only if we have not yet matched the entire string. */
5015 unsigned best_regs_set
= false;
5016 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5017 re_char
**best_regstart
, **best_regend
;
5020 /* Logically, this is `best_regend[0]'. But we don't want to have to
5021 allocate space for that if we're not allocating space for anything
5022 else (see below). Also, we never need info about register 0 for
5023 any of the other register vectors, and it seems rather a kludge to
5024 treat `best_regend' differently than the rest. So we keep track of
5025 the end of the best match so far in a separate variable. We
5026 initialize this to NULL so that when we backtrack the first time
5027 and need to test it, it's not garbage. */
5028 re_char
*match_end
= NULL
;
5030 #ifdef DEBUG_COMPILES_ARGUMENTS
5031 /* Counts the total number of registers pushed. */
5032 unsigned num_regs_pushed
= 0;
5035 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5037 REGEX_USE_SAFE_ALLOCA
;
5041 #ifdef MATCH_MAY_ALLOCATE
5042 /* Do not bother to initialize all the register variables if there are
5043 no groups in the pattern, as it takes a fair amount of time. If
5044 there are groups, we include space for register 0 (the whole
5045 pattern), even though we never use it, since it simplifies the
5046 array indexing. We should fix this. */
5049 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5050 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5051 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5052 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5054 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5062 /* We must initialize all our variables to NULL, so that
5063 `FREE_VARIABLES' doesn't try to free them. */
5064 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5066 #endif /* MATCH_MAY_ALLOCATE */
5068 /* The starting position is bogus. */
5069 if (pos
< 0 || pos
> size1
+ size2
)
5075 /* Initialize subexpression text positions to -1 to mark ones that no
5076 start_memory/stop_memory has been seen for. Also initialize the
5077 register information struct. */
5078 for (reg
= 1; reg
< num_regs
; reg
++)
5079 regstart
[reg
] = regend
[reg
] = NULL
;
5081 /* We move `string1' into `string2' if the latter's empty -- but not if
5082 `string1' is null. */
5083 if (size2
== 0 && string1
!= NULL
)
5090 end1
= string1
+ size1
;
5091 end2
= string2
+ size2
;
5093 /* `p' scans through the pattern as `d' scans through the data.
5094 `dend' is the end of the input string that `d' points within. `d'
5095 is advanced into the following input string whenever necessary, but
5096 this happens before fetching; therefore, at the beginning of the
5097 loop, `d' can be pointing at the end of a string, but it cannot
5101 /* Only match within string2. */
5102 d
= string2
+ pos
- size1
;
5103 dend
= end_match_2
= string2
+ stop
- size1
;
5104 end_match_1
= end1
; /* Just to give it a value. */
5110 /* Only match within string1. */
5111 end_match_1
= string1
+ stop
;
5113 When we reach end_match_1, PREFETCH normally switches to string2.
5114 But in the present case, this means that just doing a PREFETCH
5115 makes us jump from `stop' to `gap' within the string.
5116 What we really want here is for the search to stop as
5117 soon as we hit end_match_1. That's why we set end_match_2
5118 to end_match_1 (since PREFETCH fails as soon as we hit
5120 end_match_2
= end_match_1
;
5123 { /* It's important to use this code when stop == size so that
5124 moving `d' from end1 to string2 will not prevent the d == dend
5125 check from catching the end of string. */
5127 end_match_2
= string2
+ stop
- size1
;
5133 DEBUG_PRINT ("The compiled pattern is: ");
5134 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5135 DEBUG_PRINT ("The string to match is: \"");
5136 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5137 DEBUG_PRINT ("\"\n");
5139 /* This loops over pattern commands. It exits by returning from the
5140 function if the match is complete, or it drops through if the match
5141 fails at this starting point in the input data. */
5144 DEBUG_PRINT ("\n%p: ", p
);
5148 /* End of pattern means we might have succeeded. */
5149 DEBUG_PRINT ("end of pattern ... ");
5151 /* If we haven't matched the entire string, and we want the
5152 longest match, try backtracking. */
5153 if (d
!= end_match_2
)
5155 /* True if this match is the best seen so far. */
5159 /* True if this match ends in the same string (string1
5160 or string2) as the best previous match. */
5161 bool same_str_p
= (FIRST_STRING_P (match_end
)
5162 == FIRST_STRING_P (d
));
5164 /* AIX compiler got confused when this was combined
5165 with the previous declaration. */
5167 best_match_p
= d
> match_end
;
5169 best_match_p
= !FIRST_STRING_P (d
);
5172 DEBUG_PRINT ("backtracking.\n");
5174 if (!FAIL_STACK_EMPTY ())
5175 { /* More failure points to try. */
5177 /* If exceeds best match so far, save it. */
5178 if (!best_regs_set
|| best_match_p
)
5180 best_regs_set
= true;
5183 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5185 for (reg
= 1; reg
< num_regs
; reg
++)
5187 best_regstart
[reg
] = regstart
[reg
];
5188 best_regend
[reg
] = regend
[reg
];
5194 /* If no failure points, don't restore garbage. And if
5195 last match is real best match, don't restore second
5197 else if (best_regs_set
&& !best_match_p
)
5200 /* Restore best match. It may happen that `dend ==
5201 end_match_1' while the restored d is in string2.
5202 For example, the pattern `x.*y.*z' against the
5203 strings `x-' and `y-z-', if the two strings are
5204 not consecutive in memory. */
5205 DEBUG_PRINT ("Restoring best registers.\n");
5208 dend
= ((d
>= string1
&& d
<= end1
)
5209 ? end_match_1
: end_match_2
);
5211 for (reg
= 1; reg
< num_regs
; reg
++)
5213 regstart
[reg
] = best_regstart
[reg
];
5214 regend
[reg
] = best_regend
[reg
];
5217 } /* d != end_match_2 */
5220 DEBUG_PRINT ("Accepting match.\n");
5222 /* If caller wants register contents data back, do it. */
5223 if (regs
&& !bufp
->no_sub
)
5225 /* Have the register data arrays been allocated? */
5226 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5227 { /* No. So allocate them with malloc. We need one
5228 extra element beyond `num_regs' for the `-1' marker
5230 regs
->num_regs
= max (RE_NREGS
, num_regs
+ 1);
5231 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5232 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5233 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5238 bufp
->regs_allocated
= REGS_REALLOCATE
;
5240 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5241 { /* Yes. If we need more elements than were already
5242 allocated, reallocate them. If we need fewer, just
5244 if (regs
->num_regs
< num_regs
+ 1)
5246 regs
->num_regs
= num_regs
+ 1;
5247 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5248 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5249 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5258 /* These braces fend off a "empty body in an else-statement"
5259 warning under GCC when assert expands to nothing. */
5260 assert (bufp
->regs_allocated
== REGS_FIXED
);
5263 /* Convert the pointer data in `regstart' and `regend' to
5264 indices. Register zero has to be set differently,
5265 since we haven't kept track of any info for it. */
5266 if (regs
->num_regs
> 0)
5268 regs
->start
[0] = pos
;
5269 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5272 /* Go through the first `min (num_regs, regs->num_regs)'
5273 registers, since that is all we initialized. */
5274 for (reg
= 1; reg
< min (num_regs
, regs
->num_regs
); reg
++)
5276 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5277 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5280 regs
->start
[reg
] = POINTER_TO_OFFSET (regstart
[reg
]);
5281 regs
->end
[reg
] = POINTER_TO_OFFSET (regend
[reg
]);
5285 /* If the regs structure we return has more elements than
5286 were in the pattern, set the extra elements to -1. If
5287 we (re)allocated the registers, this is the case,
5288 because we always allocate enough to have at least one
5290 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5291 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5292 } /* regs && !bufp->no_sub */
5294 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5295 nfailure_points_pushed
, nfailure_points_popped
,
5296 nfailure_points_pushed
- nfailure_points_popped
);
5297 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed
);
5299 ptrdiff_t dcnt
= POINTER_TO_OFFSET (d
) - pos
;
5301 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt
);
5307 /* Otherwise match next pattern command. */
5310 /* Ignore these. Used to ignore the n of succeed_n's which
5311 currently have n == 0. */
5313 DEBUG_PRINT ("EXECUTING no_op.\n");
5317 DEBUG_PRINT ("EXECUTING succeed.\n");
5320 /* Match the next n pattern characters exactly. The following
5321 byte in the pattern defines n, and the n bytes after that
5322 are the characters to match. */
5325 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt
);
5327 /* Remember the start point to rollback upon failure. */
5331 /* This is written out as an if-else so we don't waste time
5332 testing `translate' inside the loop. */
5333 if (RE_TRANSLATE_P (translate
))
5337 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5357 /* The cost of testing `translate' is comparatively small. */
5358 if (target_multibyte
)
5361 int pat_charlen
, buf_charlen
;
5366 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5369 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5372 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5374 if (TRANSLATE (buf_ch
) != pat_ch
)
5382 mcnt
-= pat_charlen
;
5394 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5395 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5402 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5403 if (! CHAR_BYTE8_P (buf_ch
))
5405 buf_ch
= TRANSLATE (buf_ch
);
5406 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5412 if (buf_ch
!= pat_ch
)
5425 /* Match any character except possibly a newline or a null. */
5430 reg_syntax_t syntax
;
5432 DEBUG_PRINT ("EXECUTING anychar.\n");
5435 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5437 buf_ch
= TRANSLATE (buf_ch
);
5440 syntax
= RE_SYNTAX_EMACS
;
5442 syntax
= bufp
->syntax
;
5445 if ((!(syntax
& RE_DOT_NEWLINE
) && buf_ch
== '\n')
5446 || ((syntax
& RE_DOT_NOT_NULL
) && buf_ch
== '\000'))
5449 DEBUG_PRINT (" Matched \"%d\".\n", *d
);
5458 register unsigned int c
, corig
;
5461 /* Whether matching against a unibyte character. */
5462 boolean unibyte_char
= false;
5464 DEBUG_PRINT ("EXECUTING charset%s.\n",
5465 (re_opcode_t
) *(p
- 1) == charset_not
? "_not" : "");
5468 corig
= c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5469 if (target_multibyte
)
5474 c1
= RE_CHAR_TO_UNIBYTE (c
);
5477 unibyte_char
= true;
5483 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5485 if (! CHAR_BYTE8_P (c1
))
5487 c1
= TRANSLATE (c1
);
5488 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5491 unibyte_char
= true;
5496 unibyte_char
= true;
5500 if (!execute_charset (&p
, c
, corig
, unibyte_char
))
5508 /* The beginning of a group is represented by start_memory.
5509 The argument is the register number. The text
5510 matched within the group is recorded (in the internal
5511 registers data structure) under the register number. */
5513 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p
);
5515 /* In case we need to undo this operation (via backtracking). */
5516 PUSH_FAILURE_REG (*p
);
5519 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5520 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart
[*p
]));
5522 /* Move past the register number and inner group count. */
5527 /* The stop_memory opcode represents the end of a group. Its
5528 argument is the same as start_memory's: the register number. */
5530 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p
);
5532 assert (!REG_UNSET (regstart
[*p
]));
5533 /* Strictly speaking, there should be code such as:
5535 assert (REG_UNSET (regend[*p]));
5536 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5538 But the only info to be pushed is regend[*p] and it is known to
5539 be UNSET, so there really isn't anything to push.
5540 Not pushing anything, on the other hand deprives us from the
5541 guarantee that regend[*p] is UNSET since undoing this operation
5542 will not reset its value properly. This is not important since
5543 the value will only be read on the next start_memory or at
5544 the very end and both events can only happen if this stop_memory
5548 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend
[*p
]));
5550 /* Move past the register number and the inner group count. */
5555 /* \<digit> has been turned into a `duplicate' command which is
5556 followed by the numeric value of <digit> as the register number. */
5559 register re_char
*d2
, *dend2
;
5560 int regno
= *p
++; /* Get which register to match against. */
5561 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno
);
5563 /* Can't back reference a group which we've never matched. */
5564 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5567 /* Where in input to try to start matching. */
5568 d2
= regstart
[regno
];
5570 /* Remember the start point to rollback upon failure. */
5573 /* Where to stop matching; if both the place to start and
5574 the place to stop matching are in the same string, then
5575 set to the place to stop, otherwise, for now have to use
5576 the end of the first string. */
5578 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5579 == FIRST_STRING_P (regend
[regno
]))
5580 ? regend
[regno
] : end_match_1
);
5585 /* If necessary, advance to next segment in register
5589 if (dend2
== end_match_2
) break;
5590 if (dend2
== regend
[regno
]) break;
5592 /* End of string1 => advance to string2. */
5594 dend2
= regend
[regno
];
5596 /* At end of register contents => success */
5597 if (d2
== dend2
) break;
5599 /* If necessary, advance to next segment in data. */
5602 /* How many characters left in this segment to match. */
5605 /* Want how many consecutive characters we can match in
5606 one shot, so, if necessary, adjust the count. */
5607 if (dcnt
> dend2
- d2
)
5610 /* Compare that many; failure if mismatch, else move
5612 if (RE_TRANSLATE_P (translate
)
5613 ? bcmp_translate (d
, d2
, dcnt
, translate
, target_multibyte
)
5614 : memcmp (d
, d2
, dcnt
))
5619 d
+= dcnt
, d2
+= dcnt
;
5625 /* begline matches the empty string at the beginning of the string
5626 (unless `not_bol' is set in `bufp'), and after newlines. */
5628 DEBUG_PRINT ("EXECUTING begline.\n");
5630 if (AT_STRINGS_BEG (d
))
5632 if (!bufp
->not_bol
) break;
5637 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5641 /* In all other cases, we fail. */
5645 /* endline is the dual of begline. */
5647 DEBUG_PRINT ("EXECUTING endline.\n");
5649 if (AT_STRINGS_END (d
))
5651 if (!bufp
->not_eol
) break;
5655 PREFETCH_NOLIMIT ();
5662 /* Match at the very beginning of the data. */
5664 DEBUG_PRINT ("EXECUTING begbuf.\n");
5665 if (AT_STRINGS_BEG (d
))
5670 /* Match at the very end of the data. */
5672 DEBUG_PRINT ("EXECUTING endbuf.\n");
5673 if (AT_STRINGS_END (d
))
5678 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5679 pushes NULL as the value for the string on the stack. Then
5680 `POP_FAILURE_POINT' will keep the current value for the
5681 string, instead of restoring it. To see why, consider
5682 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5683 then the . fails against the \n. But the next thing we want
5684 to do is match the \n against the \n; if we restored the
5685 string value, we would be back at the foo.
5687 Because this is used only in specific cases, we don't need to
5688 check all the things that `on_failure_jump' does, to make
5689 sure the right things get saved on the stack. Hence we don't
5690 share its code. The only reason to push anything on the
5691 stack at all is that otherwise we would have to change
5692 `anychar's code to do something besides goto fail in this
5693 case; that seems worse than this. */
5694 case on_failure_keep_string_jump
:
5695 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5696 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5699 PUSH_FAILURE_POINT (p
- 3, NULL
);
5702 /* A nasty loop is introduced by the non-greedy *? and +?.
5703 With such loops, the stack only ever contains one failure point
5704 at a time, so that a plain on_failure_jump_loop kind of
5705 cycle detection cannot work. Worse yet, such a detection
5706 can not only fail to detect a cycle, but it can also wrongly
5707 detect a cycle (between different instantiations of the same
5709 So the method used for those nasty loops is a little different:
5710 We use a special cycle-detection-stack-frame which is pushed
5711 when the on_failure_jump_nastyloop failure-point is *popped*.
5712 This special frame thus marks the beginning of one iteration
5713 through the loop and we can hence easily check right here
5714 whether something matched between the beginning and the end of
5716 case on_failure_jump_nastyloop
:
5717 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5718 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5721 assert ((re_opcode_t
)p
[-4] == no_op
);
5724 CHECK_INFINITE_LOOP (p
- 4, d
);
5726 /* If there's a cycle, just continue without pushing
5727 this failure point. The failure point is the "try again"
5728 option, which shouldn't be tried.
5729 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5730 PUSH_FAILURE_POINT (p
- 3, d
);
5734 /* Simple loop detecting on_failure_jump: just check on the
5735 failure stack if the same spot was already hit earlier. */
5736 case on_failure_jump_loop
:
5738 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5739 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5743 CHECK_INFINITE_LOOP (p
- 3, d
);
5745 /* If there's a cycle, get out of the loop, as if the matching
5746 had failed. We used to just `goto fail' here, but that was
5747 aborting the search a bit too early: we want to keep the
5748 empty-loop-match and keep matching after the loop.
5749 We want (x?)*y\1z to match both xxyz and xxyxz. */
5752 PUSH_FAILURE_POINT (p
- 3, d
);
5757 /* Uses of on_failure_jump:
5759 Each alternative starts with an on_failure_jump that points
5760 to the beginning of the next alternative. Each alternative
5761 except the last ends with a jump that in effect jumps past
5762 the rest of the alternatives. (They really jump to the
5763 ending jump of the following alternative, because tensioning
5764 these jumps is a hassle.)
5766 Repeats start with an on_failure_jump that points past both
5767 the repetition text and either the following jump or
5768 pop_failure_jump back to this on_failure_jump. */
5769 case on_failure_jump
:
5770 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5771 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5774 PUSH_FAILURE_POINT (p
-3, d
);
5777 /* This operation is used for greedy *.
5778 Compare the beginning of the repeat with what in the
5779 pattern follows its end. If we can establish that there
5780 is nothing that they would both match, i.e., that we
5781 would have to backtrack because of (as in, e.g., `a*a')
5782 then we can use a non-backtracking loop based on
5783 on_failure_keep_string_jump instead of on_failure_jump. */
5784 case on_failure_jump_smart
:
5785 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5786 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5789 re_char
*p1
= p
; /* Next operation. */
5790 /* Here, we discard `const', making re_match non-reentrant. */
5791 unsigned char *p2
= (unsigned char *) p
+ mcnt
; /* Jump dest. */
5792 unsigned char *p3
= (unsigned char *) p
- 3; /* opcode location. */
5794 p
-= 3; /* Reset so that we will re-execute the
5795 instruction once it's been changed. */
5797 EXTRACT_NUMBER (mcnt
, p2
- 2);
5799 /* Ensure this is indeed the trivial kind of loop
5800 we are expecting. */
5801 assert (skip_one_char (p1
) == p2
- 3);
5802 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5803 DEBUG_STATEMENT (debug
+= 2);
5804 if (mutually_exclusive_p (bufp
, p1
, p2
))
5806 /* Use a fast `on_failure_keep_string_jump' loop. */
5807 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5808 *p3
= (unsigned char) on_failure_keep_string_jump
;
5809 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5813 /* Default to a safe `on_failure_jump' loop. */
5814 DEBUG_PRINT (" smart default => slow loop.\n");
5815 *p3
= (unsigned char) on_failure_jump
;
5817 DEBUG_STATEMENT (debug
-= 2);
5821 /* Unconditionally jump (without popping any failure points). */
5825 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5826 DEBUG_PRINT ("EXECUTING jump %d ", mcnt
);
5827 p
+= mcnt
; /* Do the jump. */
5828 DEBUG_PRINT ("(to %p).\n", p
);
5832 /* Have to succeed matching what follows at least n times.
5833 After that, handle like `on_failure_jump'. */
5835 /* Signedness doesn't matter since we only compare MCNT to 0. */
5836 EXTRACT_NUMBER (mcnt
, p
+ 2);
5837 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt
);
5839 /* Originally, mcnt is how many times we HAVE to succeed. */
5842 /* Here, we discard `const', making re_match non-reentrant. */
5843 unsigned char *p2
= (unsigned char *) p
+ 2; /* counter loc. */
5846 PUSH_NUMBER (p2
, mcnt
);
5849 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5854 /* Signedness doesn't matter since we only compare MCNT to 0. */
5855 EXTRACT_NUMBER (mcnt
, p
+ 2);
5856 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt
);
5858 /* Originally, this is how many times we CAN jump. */
5861 /* Here, we discard `const', making re_match non-reentrant. */
5862 unsigned char *p2
= (unsigned char *) p
+ 2; /* counter loc. */
5864 PUSH_NUMBER (p2
, mcnt
);
5865 goto unconditional_jump
;
5867 /* If don't have to jump any more, skip over the rest of command. */
5874 unsigned char *p2
; /* Location of the counter. */
5875 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5877 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5878 /* Here, we discard `const', making re_match non-reentrant. */
5879 p2
= (unsigned char *) p
+ mcnt
;
5880 /* Signedness doesn't matter since we only copy MCNT's bits. */
5881 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5882 DEBUG_PRINT (" Setting %p to %d.\n", p2
, mcnt
);
5883 PUSH_NUMBER (p2
, mcnt
);
5890 boolean
not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5891 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5893 /* We SUCCEED (or FAIL) in one of the following cases: */
5895 /* Case 1: D is at the beginning or the end of string. */
5896 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5900 /* C1 is the character before D, S1 is the syntax of C1, C2
5901 is the character at D, and S2 is the syntax of C2. */
5906 ssize_t offset
= PTR_TO_OFFSET (d
- 1);
5907 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5908 UPDATE_SYNTAX_TABLE (charpos
);
5910 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5913 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5915 PREFETCH_NOLIMIT ();
5916 GET_CHAR_AFTER (c2
, d
, dummy
);
5919 if (/* Case 2: Only one of S1 and S2 is Sword. */
5920 ((s1
== Sword
) != (s2
== Sword
))
5921 /* Case 3: Both of S1 and S2 are Sword, and macro
5922 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5923 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5933 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5935 /* We FAIL in one of the following cases: */
5937 /* Case 1: D is at the end of string. */
5938 if (AT_STRINGS_END (d
))
5942 /* C1 is the character before D, S1 is the syntax of C1, C2
5943 is the character at D, and S2 is the syntax of C2. */
5948 ssize_t offset
= PTR_TO_OFFSET (d
);
5949 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5950 UPDATE_SYNTAX_TABLE (charpos
);
5953 GET_CHAR_AFTER (c2
, d
, dummy
);
5956 /* Case 2: S2 is not Sword. */
5960 /* Case 3: D is not at the beginning of string ... */
5961 if (!AT_STRINGS_BEG (d
))
5963 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5965 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5969 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5971 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5978 DEBUG_PRINT ("EXECUTING wordend.\n");
5980 /* We FAIL in one of the following cases: */
5982 /* Case 1: D is at the beginning of string. */
5983 if (AT_STRINGS_BEG (d
))
5987 /* C1 is the character before D, S1 is the syntax of C1, C2
5988 is the character at D, and S2 is the syntax of C2. */
5993 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
5994 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5995 UPDATE_SYNTAX_TABLE (charpos
);
5997 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6000 /* Case 2: S1 is not Sword. */
6004 /* Case 3: D is not at the end of string ... */
6005 if (!AT_STRINGS_END (d
))
6007 PREFETCH_NOLIMIT ();
6008 GET_CHAR_AFTER (c2
, d
, dummy
);
6010 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6014 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6016 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6023 DEBUG_PRINT ("EXECUTING symbeg.\n");
6025 /* We FAIL in one of the following cases: */
6027 /* Case 1: D is at the end of string. */
6028 if (AT_STRINGS_END (d
))
6032 /* C1 is the character before D, S1 is the syntax of C1, C2
6033 is the character at D, and S2 is the syntax of C2. */
6037 ssize_t offset
= PTR_TO_OFFSET (d
);
6038 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6039 UPDATE_SYNTAX_TABLE (charpos
);
6042 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6045 /* Case 2: S2 is neither Sword nor Ssymbol. */
6046 if (s2
!= Sword
&& s2
!= Ssymbol
)
6049 /* Case 3: D is not at the beginning of string ... */
6050 if (!AT_STRINGS_BEG (d
))
6052 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6054 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6058 /* ... and S1 is Sword or Ssymbol. */
6059 if (s1
== Sword
|| s1
== Ssymbol
)
6066 DEBUG_PRINT ("EXECUTING symend.\n");
6068 /* We FAIL in one of the following cases: */
6070 /* Case 1: D is at the beginning of string. */
6071 if (AT_STRINGS_BEG (d
))
6075 /* C1 is the character before D, S1 is the syntax of C1, C2
6076 is the character at D, and S2 is the syntax of C2. */
6080 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6081 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6082 UPDATE_SYNTAX_TABLE (charpos
);
6084 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6087 /* Case 2: S1 is neither Ssymbol nor Sword. */
6088 if (s1
!= Sword
&& s1
!= Ssymbol
)
6091 /* Case 3: D is not at the end of string ... */
6092 if (!AT_STRINGS_END (d
))
6094 PREFETCH_NOLIMIT ();
6095 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6097 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6101 /* ... and S2 is Sword or Ssymbol. */
6102 if (s2
== Sword
|| s2
== Ssymbol
)
6111 boolean
not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6113 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6118 ssize_t offset
= PTR_TO_OFFSET (d
);
6119 ssize_t pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6120 UPDATE_SYNTAX_TABLE (pos1
);
6127 GET_CHAR_AFTER (c
, d
, len
);
6128 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6137 DEBUG_PRINT ("EXECUTING at_dot.\n");
6138 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6143 case notcategoryspec
:
6145 boolean
not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6147 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6148 not ? "not" : "", mcnt
);
6154 GET_CHAR_AFTER (c
, d
, len
);
6155 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6167 continue; /* Successfully executed one pattern command; keep going. */
6170 /* We goto here if a matching operation fails. */
6173 if (!FAIL_STACK_EMPTY ())
6176 /* A restart point is known. Restore to that state. */
6177 DEBUG_PRINT ("\nFAIL:\n");
6178 POP_FAILURE_POINT (str
, pat
);
6181 case on_failure_keep_string_jump
:
6182 assert (str
== NULL
);
6183 goto continue_failure_jump
;
6185 case on_failure_jump_nastyloop
:
6186 assert ((re_opcode_t
)pat
[-2] == no_op
);
6187 PUSH_FAILURE_POINT (pat
- 2, str
);
6189 case on_failure_jump_loop
:
6190 case on_failure_jump
:
6193 continue_failure_jump
:
6194 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6199 /* A special frame used for nastyloops. */
6206 assert (p
>= bufp
->buffer
&& p
<= pend
);
6208 if (d
>= string1
&& d
<= end1
)
6212 break; /* Matching at this starting point really fails. */
6216 goto restore_best_regs
;
6220 return -1; /* Failure to match. */
6223 /* Subroutine definitions for re_match_2. */
6225 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6226 bytes; nonzero otherwise. */
6229 bcmp_translate (re_char
*s1
, re_char
*s2
, ssize_t len
,
6230 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6232 re_char
*p1
= s1
, *p2
= s2
;
6233 re_char
*p1_end
= s1
+ len
;
6234 re_char
*p2_end
= s2
+ len
;
6236 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6237 different lengths, but relying on a single `len' would break this. -sm */
6238 while (p1
< p1_end
&& p2
< p2_end
)
6240 int p1_charlen
, p2_charlen
;
6241 re_wchar_t p1_ch
, p2_ch
;
6243 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6244 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6246 if (RE_TRANSLATE (translate
, p1_ch
)
6247 != RE_TRANSLATE (translate
, p2_ch
))
6250 p1
+= p1_charlen
, p2
+= p2_charlen
;
6253 if (p1
!= p1_end
|| p2
!= p2_end
)
6259 /* Entry points for GNU code. */
6261 /* re_compile_pattern is the GNU regular expression compiler: it
6262 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6263 Returns 0 if the pattern was valid, otherwise an error string.
6265 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6266 are set in BUFP on entry.
6268 We call regex_compile to do the actual compilation. */
6271 re_compile_pattern (const char *pattern
, size_t length
,
6273 bool posix_backtracking
, const char *whitespace_regexp
,
6275 struct re_pattern_buffer
*bufp
)
6279 /* GNU code is written to assume at least RE_NREGS registers will be set
6280 (and at least one extra will be -1). */
6281 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6283 /* And GNU code determines whether or not to get register information
6284 by passing null for the REGS argument to re_match, etc., not by
6288 ret
= regex_compile ((re_char
*) pattern
, length
,
6299 return gettext (re_error_msgid
[(int) ret
]);
6301 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6303 /* Entry points compatible with 4.2 BSD regex library. We don't define
6304 them unless specifically requested. */
6306 #if defined _REGEX_RE_COMP || defined _LIBC
6308 /* BSD has one and only one pattern buffer. */
6309 static struct re_pattern_buffer re_comp_buf
;
6313 /* Make these definitions weak in libc, so POSIX programs can redefine
6314 these names if they don't use our functions, and still use
6315 regcomp/regexec below without link errors. */
6318 re_comp (const char *s
)
6324 if (!re_comp_buf
.buffer
)
6325 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6326 return (char *) gettext ("No previous regular expression");
6330 if (!re_comp_buf
.buffer
)
6332 re_comp_buf
.buffer
= malloc (200);
6333 if (re_comp_buf
.buffer
== NULL
)
6334 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6335 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6336 re_comp_buf
.allocated
= 200;
6338 re_comp_buf
.fastmap
= malloc (1 << BYTEWIDTH
);
6339 if (re_comp_buf
.fastmap
== NULL
)
6340 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6341 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6344 /* Since `re_exec' always passes NULL for the `regs' argument, we
6345 don't need to initialize the pattern buffer fields which affect it. */
6347 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6352 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6353 return (char *) gettext (re_error_msgid
[(int) ret
]);
6361 re_exec (const char *s
)
6363 const size_t len
= strlen (s
);
6364 return re_search (&re_comp_buf
, s
, len
, 0, len
, 0) >= 0;
6366 #endif /* _REGEX_RE_COMP */
6368 /* POSIX.2 functions. Don't define these for Emacs. */
6372 /* regcomp takes a regular expression as a string and compiles it.
6374 PREG is a regex_t *. We do not expect any fields to be initialized,
6375 since POSIX says we shouldn't. Thus, we set
6377 `buffer' to the compiled pattern;
6378 `used' to the length of the compiled pattern;
6379 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6380 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6381 RE_SYNTAX_POSIX_BASIC;
6382 `fastmap' to an allocated space for the fastmap;
6383 `fastmap_accurate' to zero;
6384 `re_nsub' to the number of subexpressions in PATTERN.
6386 PATTERN is the address of the pattern string.
6388 CFLAGS is a series of bits which affect compilation.
6390 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6391 use POSIX basic syntax.
6393 If REG_NEWLINE is set, then . and [^...] don't match newline.
6394 Also, regexec will try a match beginning after every newline.
6396 If REG_ICASE is set, then we considers upper- and lowercase
6397 versions of letters to be equivalent when matching.
6399 If REG_NOSUB is set, then when PREG is passed to regexec, that
6400 routine will report only success or failure, and nothing about the
6403 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6404 the return codes and their meanings.) */
6407 regcomp (regex_t
*_Restrict_ preg
, const char *_Restrict_ pattern
,
6412 = (cflags
& REG_EXTENDED
) ?
6413 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6415 /* regex_compile will allocate the space for the compiled pattern. */
6417 preg
->allocated
= 0;
6420 /* Try to allocate space for the fastmap. */
6421 preg
->fastmap
= malloc (1 << BYTEWIDTH
);
6423 if (cflags
& REG_ICASE
)
6427 preg
->translate
= malloc (CHAR_SET_SIZE
* sizeof *preg
->translate
);
6428 if (preg
->translate
== NULL
)
6429 return (int) REG_ESPACE
;
6431 /* Map uppercase characters to corresponding lowercase ones. */
6432 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6433 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6436 preg
->translate
= NULL
;
6438 /* If REG_NEWLINE is set, newlines are treated differently. */
6439 if (cflags
& REG_NEWLINE
)
6440 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6441 syntax
&= ~RE_DOT_NEWLINE
;
6442 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6445 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6447 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6449 /* POSIX says a null character in the pattern terminates it, so we
6450 can use strlen here in compiling the pattern. */
6451 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6453 /* POSIX doesn't distinguish between an unmatched open-group and an
6454 unmatched close-group: both are REG_EPAREN. */
6455 if (ret
== REG_ERPAREN
)
6458 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6459 { /* Compute the fastmap now, since regexec cannot modify the pattern
6461 re_compile_fastmap (preg
);
6462 if (preg
->can_be_null
)
6463 { /* The fastmap can't be used anyway. */
6464 free (preg
->fastmap
);
6465 preg
->fastmap
= NULL
;
6470 WEAK_ALIAS (__regcomp
, regcomp
)
6473 /* regexec searches for a given pattern, specified by PREG, in the
6476 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6477 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6478 least NMATCH elements, and we set them to the offsets of the
6479 corresponding matched substrings.
6481 EFLAGS specifies `execution flags' which affect matching: if
6482 REG_NOTBOL is set, then ^ does not match at the beginning of the
6483 string; if REG_NOTEOL is set, then $ does not match at the end.
6485 We return 0 if we find a match and REG_NOMATCH if not. */
6488 regexec (const regex_t
*_Restrict_ preg
, const char *_Restrict_ string
,
6489 size_t nmatch
, regmatch_t pmatch
[_Restrict_arr_
], int eflags
)
6492 struct re_registers regs
;
6493 regex_t private_preg
;
6494 size_t len
= strlen (string
);
6495 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6497 private_preg
= *preg
;
6499 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6500 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6502 /* The user has told us exactly how many registers to return
6503 information about, via `nmatch'. We have to pass that on to the
6504 matching routines. */
6505 private_preg
.regs_allocated
= REGS_FIXED
;
6509 regs
.num_regs
= nmatch
;
6510 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6511 if (regs
.start
== NULL
)
6513 regs
.end
= regs
.start
+ nmatch
;
6516 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6517 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6518 was a little bit longer but still only matching the real part.
6519 This works because the `endline' will check for a '\n' and will find a
6520 '\0', correctly deciding that this is not the end of a line.
6521 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6522 a convenient '\0' there. For all we know, the string could be preceded
6523 by '\n' which would throw things off. */
6525 /* Perform the searching operation. */
6526 ret
= re_search (&private_preg
, string
, len
,
6527 /* start: */ 0, /* range: */ len
,
6528 want_reg_info
? ®s
: 0);
6530 /* Copy the register information to the POSIX structure. */
6537 for (r
= 0; r
< nmatch
; r
++)
6539 pmatch
[r
].rm_so
= regs
.start
[r
];
6540 pmatch
[r
].rm_eo
= regs
.end
[r
];
6544 /* If we needed the temporary register info, free the space now. */
6548 /* We want zero return to mean success, unlike `re_search'. */
6549 return ret
>= 0 ? REG_NOERROR
: REG_NOMATCH
;
6551 WEAK_ALIAS (__regexec
, regexec
)
6554 /* Returns a message corresponding to an error code, ERR_CODE, returned
6555 from either regcomp or regexec. We don't use PREG here.
6557 ERR_CODE was previously called ERRCODE, but that name causes an
6558 error with msvc8 compiler. */
6561 regerror (int err_code
, const regex_t
*preg
, char *errbuf
, size_t errbuf_size
)
6567 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6568 /* Only error codes returned by the rest of the code should be passed
6569 to this routine. If we are given anything else, or if other regex
6570 code generates an invalid error code, then the program has a bug.
6571 Dump core so we can fix it. */
6574 msg
= gettext (re_error_msgid
[err_code
]);
6576 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6578 if (errbuf_size
!= 0)
6580 if (msg_size
> errbuf_size
)
6582 memcpy (errbuf
, msg
, errbuf_size
- 1);
6583 errbuf
[errbuf_size
- 1] = 0;
6586 strcpy (errbuf
, msg
);
6591 WEAK_ALIAS (__regerror
, regerror
)
6594 /* Free dynamically allocated space used by PREG. */
6597 regfree (regex_t
*preg
)
6599 free (preg
->buffer
);
6600 preg
->buffer
= NULL
;
6602 preg
->allocated
= 0;
6605 free (preg
->fastmap
);
6606 preg
->fastmap
= NULL
;
6607 preg
->fastmap_accurate
= 0;
6609 free (preg
->translate
);
6610 preg
->translate
= NULL
;
6612 WEAK_ALIAS (__regfree
, regfree
)
6614 #endif /* not emacs */