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) 1993,94,95,96,97,98,99,2000,04 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 2, 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, write to the Free Software
19 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
23 - structure the opcode space into opcode+flag.
24 - merge with glibc's regex.[ch].
25 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
26 need to modify the compiled regexp so that re_match can be reentrant.
27 - get rid of on_failure_jump_smart by doing the optimization in re_comp
28 rather than at run-time, so that re_match can be reentrant.
31 /* AIX requires this to be the first thing in the file. */
32 #if defined _AIX && !defined REGEX_MALLOC
40 #if defined STDC_HEADERS && !defined emacs
43 /* We need this for `regex.h', and perhaps for the Emacs include files. */
44 # include <sys/types.h>
47 /* Whether to use ISO C Amendment 1 wide char functions.
48 Those should not be used for Emacs since it uses its own. */
50 #define WIDE_CHAR_SUPPORT 1
52 #define WIDE_CHAR_SUPPORT \
53 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
56 /* For platform which support the ISO C amendement 1 functionality we
57 support user defined character classes. */
59 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
65 /* We have to keep the namespace clean. */
66 # define regfree(preg) __regfree (preg)
67 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
68 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
69 # define regerror(errcode, preg, errbuf, errbuf_size) \
70 __regerror(errcode, preg, errbuf, errbuf_size)
71 # define re_set_registers(bu, re, nu, st, en) \
72 __re_set_registers (bu, re, nu, st, en)
73 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
74 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
75 # define re_match(bufp, string, size, pos, regs) \
76 __re_match (bufp, string, size, pos, regs)
77 # define re_search(bufp, string, size, startpos, range, regs) \
78 __re_search (bufp, string, size, startpos, range, regs)
79 # define re_compile_pattern(pattern, length, bufp) \
80 __re_compile_pattern (pattern, length, bufp)
81 # define re_set_syntax(syntax) __re_set_syntax (syntax)
82 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
83 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
84 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
86 /* Make sure we call libc's function even if the user overrides them. */
87 # define btowc __btowc
88 # define iswctype __iswctype
89 # define wctype __wctype
91 # define WEAK_ALIAS(a,b) weak_alias (a, b)
93 /* We are also using some library internals. */
94 # include <locale/localeinfo.h>
95 # include <locale/elem-hash.h>
96 # include <langinfo.h>
98 # define WEAK_ALIAS(a,b)
101 /* This is for other GNU distributions with internationalized messages. */
102 #if HAVE_LIBINTL_H || defined _LIBC
103 # include <libintl.h>
105 # define gettext(msgid) (msgid)
109 /* This define is so xgettext can find the internationalizable
111 # define gettext_noop(String) String
114 /* The `emacs' switch turns on certain matching commands
115 that make sense only in Emacs. */
121 /* Make syntax table lookup grant data in gl_state. */
122 # define SYNTAX_ENTRY_VIA_PROPERTY
125 # include "charset.h"
126 # include "category.h"
131 # define malloc xmalloc
135 # define realloc xrealloc
141 /* Converts the pointer to the char to BEG-based offset from the start. */
142 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
143 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
145 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
146 # define RE_STRING_CHAR(p, s) \
147 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
148 # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \
149 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
151 /* Set C a (possibly multibyte) character before P. P points into a
152 string which is the virtual concatenation of STR1 (which ends at
153 END1) or STR2 (which ends at END2). */
154 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
158 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
159 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
160 re_char *d0 = dtemp; \
161 PREV_CHAR_BOUNDARY (d0, dlimit); \
162 c = STRING_CHAR (d0, dtemp - d0); \
165 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
169 #else /* not emacs */
171 /* If we are not linking with Emacs proper,
172 we can't use the relocating allocator
173 even if config.h says that we can. */
176 # if defined STDC_HEADERS || defined _LIBC
183 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
184 If nothing else has been done, use the method below. */
185 # ifdef INHIBIT_STRING_HEADER
186 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
187 # if !defined bzero && !defined bcopy
188 # undef INHIBIT_STRING_HEADER
193 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
194 This is used in most programs--a few other programs avoid this
195 by defining INHIBIT_STRING_HEADER. */
196 # ifndef INHIBIT_STRING_HEADER
197 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
201 # define bzero(s, n) (memset (s, '\0', n), (s))
203 # define bzero(s, n) __bzero (s, n)
207 # include <strings.h>
209 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
212 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
217 /* Define the syntax stuff for \<, \>, etc. */
219 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
220 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
222 # ifdef SWITCH_ENUM_BUG
223 # define SWITCH_ENUM_CAST(x) ((int)(x))
225 # define SWITCH_ENUM_CAST(x) (x)
228 /* Dummy macros for non-Emacs environments. */
229 # define BASE_LEADING_CODE_P(c) (0)
230 # define CHAR_CHARSET(c) 0
231 # define CHARSET_LEADING_CODE_BASE(c) 0
232 # define MAX_MULTIBYTE_LENGTH 1
233 # define RE_MULTIBYTE_P(x) 0
234 # define WORD_BOUNDARY_P(c1, c2) (0)
235 # define CHAR_HEAD_P(p) (1)
236 # define SINGLE_BYTE_CHAR_P(c) (1)
237 # define SAME_CHARSET_P(c1, c2) (1)
238 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
239 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
240 # define STRING_CHAR(p, s) (*(p))
241 # define RE_STRING_CHAR STRING_CHAR
242 # define CHAR_STRING(c, s) (*(s) = (c), 1)
243 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
244 # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH
245 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
246 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
247 # define MAKE_CHAR(charset, c1, c2) (c1)
248 #endif /* not emacs */
251 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
252 # define RE_TRANSLATE_P(TBL) (TBL)
255 /* Get the interface, including the syntax bits. */
258 /* isalpha etc. are used for the character classes. */
263 /* 1 if C is an ASCII character. */
264 # define IS_REAL_ASCII(c) ((c) < 0200)
266 /* 1 if C is a unibyte character. */
267 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
269 /* The Emacs definitions should not be directly affected by locales. */
271 /* In Emacs, these are only used for single-byte characters. */
272 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
273 # define ISCNTRL(c) ((c) < ' ')
274 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
275 || ((c) >= 'a' && (c) <= 'f') \
276 || ((c) >= 'A' && (c) <= 'F'))
278 /* This is only used for single-byte characters. */
279 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
281 /* The rest must handle multibyte characters. */
283 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
284 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
287 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
288 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
291 # define ISALNUM(c) (IS_REAL_ASCII (c) \
292 ? (((c) >= 'a' && (c) <= 'z') \
293 || ((c) >= 'A' && (c) <= 'Z') \
294 || ((c) >= '0' && (c) <= '9')) \
295 : SYNTAX (c) == Sword)
297 # define ISALPHA(c) (IS_REAL_ASCII (c) \
298 ? (((c) >= 'a' && (c) <= 'z') \
299 || ((c) >= 'A' && (c) <= 'Z')) \
300 : SYNTAX (c) == Sword)
302 # define ISLOWER(c) (LOWERCASEP (c))
304 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
305 ? ((c) > ' ' && (c) < 0177 \
306 && !(((c) >= 'a' && (c) <= 'z') \
307 || ((c) >= 'A' && (c) <= 'Z') \
308 || ((c) >= '0' && (c) <= '9'))) \
309 : SYNTAX (c) != Sword)
311 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
313 # define ISUPPER(c) (UPPERCASEP (c))
315 # define ISWORD(c) (SYNTAX (c) == Sword)
317 #else /* not emacs */
319 /* Jim Meyering writes:
321 "... Some ctype macros are valid only for character codes that
322 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
323 using /bin/cc or gcc but without giving an ansi option). So, all
324 ctype uses should be through macros like ISPRINT... If
325 STDC_HEADERS is defined, then autoconf has verified that the ctype
326 macros don't need to be guarded with references to isascii. ...
327 Defining isascii to 1 should let any compiler worth its salt
328 eliminate the && through constant folding."
329 Solaris defines some of these symbols so we must undefine them first. */
332 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
333 # define ISASCII(c) 1
335 # define ISASCII(c) isascii(c)
338 /* 1 if C is an ASCII character. */
339 # define IS_REAL_ASCII(c) ((c) < 0200)
341 /* This distinction is not meaningful, except in Emacs. */
342 # define ISUNIBYTE(c) 1
345 # define ISBLANK(c) (ISASCII (c) && isblank (c))
347 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
350 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
352 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
356 # define ISPRINT(c) (ISASCII (c) && isprint (c))
357 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
358 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
359 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
360 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
361 # define ISLOWER(c) (ISASCII (c) && islower (c))
362 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
363 # define ISSPACE(c) (ISASCII (c) && isspace (c))
364 # define ISUPPER(c) (ISASCII (c) && isupper (c))
365 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
367 # define ISWORD(c) ISALPHA(c)
370 # define TOLOWER(c) _tolower(c)
372 # define TOLOWER(c) tolower(c)
375 /* How many characters in the character set. */
376 # define CHAR_SET_SIZE 256
380 extern char *re_syntax_table
;
382 # else /* not SYNTAX_TABLE */
384 static char re_syntax_table
[CHAR_SET_SIZE
];
395 bzero (re_syntax_table
, sizeof re_syntax_table
);
397 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
399 re_syntax_table
[c
] = Sword
;
401 re_syntax_table
['_'] = Ssymbol
;
406 # endif /* not SYNTAX_TABLE */
408 # define SYNTAX(c) re_syntax_table[(c)]
410 #endif /* not emacs */
413 # define NULL (void *)0
416 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
417 since ours (we hope) works properly with all combinations of
418 machines, compilers, `char' and `unsigned char' argument types.
419 (Per Bothner suggested the basic approach.) */
420 #undef SIGN_EXTEND_CHAR
422 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
423 #else /* not __STDC__ */
424 /* As in Harbison and Steele. */
425 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
428 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
429 use `alloca' instead of `malloc'. This is because using malloc in
430 re_search* or re_match* could cause memory leaks when C-g is used in
431 Emacs; also, malloc is slower and causes storage fragmentation. On
432 the other hand, malloc is more portable, and easier to debug.
434 Because we sometimes use alloca, some routines have to be macros,
435 not functions -- `alloca'-allocated space disappears at the end of the
436 function it is called in. */
440 # define REGEX_ALLOCATE malloc
441 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
442 # define REGEX_FREE free
444 #else /* not REGEX_MALLOC */
446 /* Emacs already defines alloca, sometimes. */
449 /* Make alloca work the best possible way. */
451 # define alloca __builtin_alloca
452 # else /* not __GNUC__ */
455 # endif /* HAVE_ALLOCA_H */
456 # endif /* not __GNUC__ */
458 # endif /* not alloca */
460 # define REGEX_ALLOCATE alloca
462 /* Assumes a `char *destination' variable. */
463 # define REGEX_REALLOCATE(source, osize, nsize) \
464 (destination = (char *) alloca (nsize), \
465 memcpy (destination, source, osize))
467 /* No need to do anything to free, after alloca. */
468 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
470 #endif /* not REGEX_MALLOC */
472 /* Define how to allocate the failure stack. */
474 #if defined REL_ALLOC && defined REGEX_MALLOC
476 # define REGEX_ALLOCATE_STACK(size) \
477 r_alloc (&failure_stack_ptr, (size))
478 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
479 r_re_alloc (&failure_stack_ptr, (nsize))
480 # define REGEX_FREE_STACK(ptr) \
481 r_alloc_free (&failure_stack_ptr)
483 #else /* not using relocating allocator */
487 # define REGEX_ALLOCATE_STACK malloc
488 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
489 # define REGEX_FREE_STACK free
491 # else /* not REGEX_MALLOC */
493 # define REGEX_ALLOCATE_STACK alloca
495 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
496 REGEX_REALLOCATE (source, osize, nsize)
497 /* No need to explicitly free anything. */
498 # define REGEX_FREE_STACK(arg) ((void)0)
500 # endif /* not REGEX_MALLOC */
501 #endif /* not using relocating allocator */
504 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
505 `string1' or just past its end. This works if PTR is NULL, which is
507 #define FIRST_STRING_P(ptr) \
508 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
510 /* (Re)Allocate N items of type T using malloc, or fail. */
511 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
512 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
513 #define RETALLOC_IF(addr, n, t) \
514 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
515 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
517 #define BYTEWIDTH 8 /* In bits. */
519 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
523 #define MAX(a, b) ((a) > (b) ? (a) : (b))
524 #define MIN(a, b) ((a) < (b) ? (a) : (b))
526 /* Type of source-pattern and string chars. */
527 typedef const unsigned char re_char
;
529 typedef char boolean
;
533 static int re_match_2_internal
_RE_ARGS ((struct re_pattern_buffer
*bufp
,
534 re_char
*string1
, int size1
,
535 re_char
*string2
, int size2
,
537 struct re_registers
*regs
,
540 /* These are the command codes that appear in compiled regular
541 expressions. Some opcodes are followed by argument bytes. A
542 command code can specify any interpretation whatsoever for its
543 arguments. Zero bytes may appear in the compiled regular expression. */
549 /* Succeed right away--no more backtracking. */
552 /* Followed by one byte giving n, then by n literal bytes. */
555 /* Matches any (more or less) character. */
558 /* Matches any one char belonging to specified set. First
559 following byte is number of bitmap bytes. Then come bytes
560 for a bitmap saying which chars are in. Bits in each byte
561 are ordered low-bit-first. A character is in the set if its
562 bit is 1. A character too large to have a bit in the map is
563 automatically not in the set.
565 If the length byte has the 0x80 bit set, then that stuff
566 is followed by a range table:
567 2 bytes of flags for character sets (low 8 bits, high 8 bits)
568 See RANGE_TABLE_WORK_BITS below.
569 2 bytes, the number of pairs that follow (upto 32767)
570 pairs, each 2 multibyte characters,
571 each multibyte character represented as 3 bytes. */
574 /* Same parameters as charset, but match any character that is
575 not one of those specified. */
578 /* Start remembering the text that is matched, for storing in a
579 register. Followed by one byte with the register number, in
580 the range 0 to one less than the pattern buffer's re_nsub
584 /* Stop remembering the text that is matched and store it in a
585 memory register. Followed by one byte with the register
586 number, in the range 0 to one less than `re_nsub' in the
590 /* Match a duplicate of something remembered. Followed by one
591 byte containing the register number. */
594 /* Fail unless at beginning of line. */
597 /* Fail unless at end of line. */
600 /* Succeeds if at beginning of buffer (if emacs) or at beginning
601 of string to be matched (if not). */
604 /* Analogously, for end of buffer/string. */
607 /* Followed by two byte relative address to which to jump. */
610 /* Followed by two-byte relative address of place to resume at
611 in case of failure. */
614 /* Like on_failure_jump, but pushes a placeholder instead of the
615 current string position when executed. */
616 on_failure_keep_string_jump
,
618 /* Just like `on_failure_jump', except that it checks that we
619 don't get stuck in an infinite loop (matching an empty string
621 on_failure_jump_loop
,
623 /* Just like `on_failure_jump_loop', except that it checks for
624 a different kind of loop (the kind that shows up with non-greedy
625 operators). This operation has to be immediately preceded
627 on_failure_jump_nastyloop
,
629 /* A smart `on_failure_jump' used for greedy * and + operators.
630 It analyses the loop before which it is put and if the
631 loop does not require backtracking, it changes itself to
632 `on_failure_keep_string_jump' and short-circuits the loop,
633 else it just defaults to changing itself into `on_failure_jump'.
634 It assumes that it is pointing to just past a `jump'. */
635 on_failure_jump_smart
,
637 /* Followed by two-byte relative address and two-byte number n.
638 After matching N times, jump to the address upon failure.
639 Does not work if N starts at 0: use on_failure_jump_loop
643 /* Followed by two-byte relative address, and two-byte number n.
644 Jump to the address N times, then fail. */
647 /* Set the following two-byte relative address to the
648 subsequent two-byte number. The address *includes* the two
652 wordbeg
, /* Succeeds if at word beginning. */
653 wordend
, /* Succeeds if at word end. */
655 wordbound
, /* Succeeds if at a word boundary. */
656 notwordbound
, /* Succeeds if not at a word boundary. */
658 symbeg
, /* Succeeds if at symbol beginning. */
659 symend
, /* Succeeds if at symbol end. */
661 /* Matches any character whose syntax is specified. Followed by
662 a byte which contains a syntax code, e.g., Sword. */
665 /* Matches any character whose syntax is not that specified. */
669 ,before_dot
, /* Succeeds if before point. */
670 at_dot
, /* Succeeds if at point. */
671 after_dot
, /* Succeeds if after point. */
673 /* Matches any character whose category-set contains the specified
674 category. The operator is followed by a byte which contains a
675 category code (mnemonic ASCII character). */
678 /* Matches any character whose category-set does not contain the
679 specified category. The operator is followed by a byte which
680 contains the category code (mnemonic ASCII character). */
685 /* Common operations on the compiled pattern. */
687 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
689 #define STORE_NUMBER(destination, number) \
691 (destination)[0] = (number) & 0377; \
692 (destination)[1] = (number) >> 8; \
695 /* Same as STORE_NUMBER, except increment DESTINATION to
696 the byte after where the number is stored. Therefore, DESTINATION
697 must be an lvalue. */
699 #define STORE_NUMBER_AND_INCR(destination, number) \
701 STORE_NUMBER (destination, number); \
702 (destination) += 2; \
705 /* Put into DESTINATION a number stored in two contiguous bytes starting
708 #define EXTRACT_NUMBER(destination, source) \
710 (destination) = *(source) & 0377; \
711 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
715 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
717 extract_number (dest
, source
)
721 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
722 *dest
= *source
& 0377;
726 # ifndef EXTRACT_MACROS /* To debug the macros. */
727 # undef EXTRACT_NUMBER
728 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
729 # endif /* not EXTRACT_MACROS */
733 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
734 SOURCE must be an lvalue. */
736 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
738 EXTRACT_NUMBER (destination, source); \
743 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
746 extract_number_and_incr (destination
, source
)
750 extract_number (destination
, *source
);
754 # ifndef EXTRACT_MACROS
755 # undef EXTRACT_NUMBER_AND_INCR
756 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
757 extract_number_and_incr (&dest, &src)
758 # endif /* not EXTRACT_MACROS */
762 /* Store a multibyte character in three contiguous bytes starting
763 DESTINATION, and increment DESTINATION to the byte after where the
764 character is stored. Therefore, DESTINATION must be an lvalue. */
766 #define STORE_CHARACTER_AND_INCR(destination, character) \
768 (destination)[0] = (character) & 0377; \
769 (destination)[1] = ((character) >> 8) & 0377; \
770 (destination)[2] = (character) >> 16; \
771 (destination) += 3; \
774 /* Put into DESTINATION a character stored in three contiguous bytes
775 starting at SOURCE. */
777 #define EXTRACT_CHARACTER(destination, source) \
779 (destination) = ((source)[0] \
780 | ((source)[1] << 8) \
781 | ((source)[2] << 16)); \
785 /* Macros for charset. */
787 /* Size of bitmap of charset P in bytes. P is a start of charset,
788 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
789 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
791 /* Nonzero if charset P has range table. */
792 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
794 /* Return the address of range table of charset P. But not the start
795 of table itself, but the before where the number of ranges is
796 stored. `2 +' means to skip re_opcode_t and size of bitmap,
797 and the 2 bytes of flags at the start of the range table. */
798 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
800 /* Extract the bit flags that start a range table. */
801 #define CHARSET_RANGE_TABLE_BITS(p) \
802 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
803 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
805 /* Test if C is listed in the bitmap of charset P. */
806 #define CHARSET_LOOKUP_BITMAP(p, c) \
807 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
808 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
810 /* Return the address of end of RANGE_TABLE. COUNT is number of
811 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
812 is start of range and end of range. `* 3' is size of each start
814 #define CHARSET_RANGE_TABLE_END(range_table, count) \
815 ((range_table) + (count) * 2 * 3)
817 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
818 COUNT is number of ranges in RANGE_TABLE. */
819 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
822 re_wchar_t range_start, range_end; \
824 re_char *range_table_end \
825 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
827 for (p = (range_table); p < range_table_end; p += 2 * 3) \
829 EXTRACT_CHARACTER (range_start, p); \
830 EXTRACT_CHARACTER (range_end, p + 3); \
832 if (range_start <= (c) && (c) <= range_end) \
841 /* Test if C is in range table of CHARSET. The flag NOT is negated if
842 C is listed in it. */
843 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
846 /* Number of ranges in range table. */ \
848 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
850 EXTRACT_NUMBER_AND_INCR (count, range_table); \
851 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
855 /* If DEBUG is defined, Regex prints many voluminous messages about what
856 it is doing (if the variable `debug' is nonzero). If linked with the
857 main program in `iregex.c', you can enter patterns and strings
858 interactively. And if linked with the main program in `main.c' and
859 the other test files, you can run the already-written tests. */
863 /* We use standard I/O for debugging. */
866 /* It is useful to test things that ``must'' be true when debugging. */
869 static int debug
= -100000;
871 # define DEBUG_STATEMENT(e) e
872 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
873 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
874 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
875 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
876 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
877 if (debug > 0) print_partial_compiled_pattern (s, e)
878 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
879 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
882 /* Print the fastmap in human-readable form. */
885 print_fastmap (fastmap
)
888 unsigned was_a_range
= 0;
891 while (i
< (1 << BYTEWIDTH
))
897 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
913 /* Print a compiled pattern string in human-readable form, starting at
914 the START pointer into it and ending just before the pointer END. */
917 print_partial_compiled_pattern (start
, end
)
927 fprintf (stderr
, "(null)\n");
931 /* Loop over pattern commands. */
934 fprintf (stderr
, "%d:\t", p
- start
);
936 switch ((re_opcode_t
) *p
++)
939 fprintf (stderr
, "/no_op");
943 fprintf (stderr
, "/succeed");
948 fprintf (stderr
, "/exactn/%d", mcnt
);
951 fprintf (stderr
, "/%c", *p
++);
957 fprintf (stderr
, "/start_memory/%d", *p
++);
961 fprintf (stderr
, "/stop_memory/%d", *p
++);
965 fprintf (stderr
, "/duplicate/%d", *p
++);
969 fprintf (stderr
, "/anychar");
975 register int c
, last
= -100;
976 register int in_range
= 0;
977 int length
= CHARSET_BITMAP_SIZE (p
- 1);
978 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
980 fprintf (stderr
, "/charset [%s",
981 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
984 fprintf (stderr
, " !extends past end of pattern! ");
986 for (c
= 0; c
< 256; c
++)
988 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
990 /* Are we starting a range? */
991 if (last
+ 1 == c
&& ! in_range
)
993 fprintf (stderr
, "-");
996 /* Have we broken a range? */
997 else if (last
+ 1 != c
&& in_range
)
999 fprintf (stderr
, "%c", last
);
1004 fprintf (stderr
, "%c", c
);
1010 fprintf (stderr
, "%c", last
);
1012 fprintf (stderr
, "]");
1016 if (has_range_table
)
1019 fprintf (stderr
, "has-range-table");
1021 /* ??? Should print the range table; for now, just skip it. */
1022 p
+= 2; /* skip range table bits */
1023 EXTRACT_NUMBER_AND_INCR (count
, p
);
1024 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1030 fprintf (stderr
, "/begline");
1034 fprintf (stderr
, "/endline");
1037 case on_failure_jump
:
1038 extract_number_and_incr (&mcnt
, &p
);
1039 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1042 case on_failure_keep_string_jump
:
1043 extract_number_and_incr (&mcnt
, &p
);
1044 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1047 case on_failure_jump_nastyloop
:
1048 extract_number_and_incr (&mcnt
, &p
);
1049 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1052 case on_failure_jump_loop
:
1053 extract_number_and_incr (&mcnt
, &p
);
1054 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1057 case on_failure_jump_smart
:
1058 extract_number_and_incr (&mcnt
, &p
);
1059 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1063 extract_number_and_incr (&mcnt
, &p
);
1064 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1068 extract_number_and_incr (&mcnt
, &p
);
1069 extract_number_and_incr (&mcnt2
, &p
);
1070 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1074 extract_number_and_incr (&mcnt
, &p
);
1075 extract_number_and_incr (&mcnt2
, &p
);
1076 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1080 extract_number_and_incr (&mcnt
, &p
);
1081 extract_number_and_incr (&mcnt2
, &p
);
1082 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1086 fprintf (stderr
, "/wordbound");
1090 fprintf (stderr
, "/notwordbound");
1094 fprintf (stderr
, "/wordbeg");
1098 fprintf (stderr
, "/wordend");
1102 fprintf (stderr
, "/symbeg");
1106 fprintf (stderr
, "/symend");
1110 fprintf (stderr
, "/syntaxspec");
1112 fprintf (stderr
, "/%d", mcnt
);
1116 fprintf (stderr
, "/notsyntaxspec");
1118 fprintf (stderr
, "/%d", mcnt
);
1123 fprintf (stderr
, "/before_dot");
1127 fprintf (stderr
, "/at_dot");
1131 fprintf (stderr
, "/after_dot");
1135 fprintf (stderr
, "/categoryspec");
1137 fprintf (stderr
, "/%d", mcnt
);
1140 case notcategoryspec
:
1141 fprintf (stderr
, "/notcategoryspec");
1143 fprintf (stderr
, "/%d", mcnt
);
1148 fprintf (stderr
, "/begbuf");
1152 fprintf (stderr
, "/endbuf");
1156 fprintf (stderr
, "?%d", *(p
-1));
1159 fprintf (stderr
, "\n");
1162 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1167 print_compiled_pattern (bufp
)
1168 struct re_pattern_buffer
*bufp
;
1170 re_char
*buffer
= bufp
->buffer
;
1172 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1173 printf ("%ld bytes used/%ld bytes allocated.\n",
1174 bufp
->used
, bufp
->allocated
);
1176 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1178 printf ("fastmap: ");
1179 print_fastmap (bufp
->fastmap
);
1182 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1183 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1184 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1185 printf ("no_sub: %d\t", bufp
->no_sub
);
1186 printf ("not_bol: %d\t", bufp
->not_bol
);
1187 printf ("not_eol: %d\t", bufp
->not_eol
);
1188 printf ("syntax: %lx\n", bufp
->syntax
);
1190 /* Perhaps we should print the translate table? */
1195 print_double_string (where
, string1
, size1
, string2
, size2
)
1208 if (FIRST_STRING_P (where
))
1210 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1211 putchar (string1
[this_char
]);
1216 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1217 putchar (string2
[this_char
]);
1221 #else /* not DEBUG */
1226 # define DEBUG_STATEMENT(e)
1227 # define DEBUG_PRINT1(x)
1228 # define DEBUG_PRINT2(x1, x2)
1229 # define DEBUG_PRINT3(x1, x2, x3)
1230 # define DEBUG_PRINT4(x1, x2, x3, x4)
1231 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1232 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1234 #endif /* not DEBUG */
1236 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1237 also be assigned to arbitrarily: each pattern buffer stores its own
1238 syntax, so it can be changed between regex compilations. */
1239 /* This has no initializer because initialized variables in Emacs
1240 become read-only after dumping. */
1241 reg_syntax_t re_syntax_options
;
1244 /* Specify the precise syntax of regexps for compilation. This provides
1245 for compatibility for various utilities which historically have
1246 different, incompatible syntaxes.
1248 The argument SYNTAX is a bit mask comprised of the various bits
1249 defined in regex.h. We return the old syntax. */
1252 re_set_syntax (syntax
)
1253 reg_syntax_t syntax
;
1255 reg_syntax_t ret
= re_syntax_options
;
1257 re_syntax_options
= syntax
;
1260 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1262 /* This table gives an error message for each of the error codes listed
1263 in regex.h. Obviously the order here has to be same as there.
1264 POSIX doesn't require that we do anything for REG_NOERROR,
1265 but why not be nice? */
1267 static const char *re_error_msgid
[] =
1269 gettext_noop ("Success"), /* REG_NOERROR */
1270 gettext_noop ("No match"), /* REG_NOMATCH */
1271 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1272 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1273 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1274 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1275 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1276 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1277 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1278 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1279 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1280 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1281 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1282 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1283 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1284 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1285 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1288 /* Avoiding alloca during matching, to placate r_alloc. */
1290 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1291 searching and matching functions should not call alloca. On some
1292 systems, alloca is implemented in terms of malloc, and if we're
1293 using the relocating allocator routines, then malloc could cause a
1294 relocation, which might (if the strings being searched are in the
1295 ralloc heap) shift the data out from underneath the regexp
1298 Here's another reason to avoid allocation: Emacs
1299 processes input from X in a signal handler; processing X input may
1300 call malloc; if input arrives while a matching routine is calling
1301 malloc, then we're scrod. But Emacs can't just block input while
1302 calling matching routines; then we don't notice interrupts when
1303 they come in. So, Emacs blocks input around all regexp calls
1304 except the matching calls, which it leaves unprotected, in the
1305 faith that they will not malloc. */
1307 /* Normally, this is fine. */
1308 #define MATCH_MAY_ALLOCATE
1310 /* When using GNU C, we are not REALLY using the C alloca, no matter
1311 what config.h may say. So don't take precautions for it. */
1316 /* The match routines may not allocate if (1) they would do it with malloc
1317 and (2) it's not safe for them to use malloc.
1318 Note that if REL_ALLOC is defined, matching would not use malloc for the
1319 failure stack, but we would still use it for the register vectors;
1320 so REL_ALLOC should not affect this. */
1321 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1322 # undef MATCH_MAY_ALLOCATE
1326 /* Failure stack declarations and macros; both re_compile_fastmap and
1327 re_match_2 use a failure stack. These have to be macros because of
1328 REGEX_ALLOCATE_STACK. */
1331 /* Approximate number of failure points for which to initially allocate space
1332 when matching. If this number is exceeded, we allocate more
1333 space, so it is not a hard limit. */
1334 #ifndef INIT_FAILURE_ALLOC
1335 # define INIT_FAILURE_ALLOC 20
1338 /* Roughly the maximum number of failure points on the stack. Would be
1339 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1340 This is a variable only so users of regex can assign to it; we never
1341 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1342 before using it, so it should probably be a byte-count instead. */
1343 # if defined MATCH_MAY_ALLOCATE
1344 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1345 whose default stack limit is 2mb. In order for a larger
1346 value to work reliably, you have to try to make it accord
1347 with the process stack limit. */
1348 size_t re_max_failures
= 40000;
1350 size_t re_max_failures
= 4000;
1353 union fail_stack_elt
1356 /* This should be the biggest `int' that's no bigger than a pointer. */
1360 typedef union fail_stack_elt fail_stack_elt_t
;
1364 fail_stack_elt_t
*stack
;
1366 size_t avail
; /* Offset of next open position. */
1367 size_t frame
; /* Offset of the cur constructed frame. */
1370 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1371 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1374 /* Define macros to initialize and free the failure stack.
1375 Do `return -2' if the alloc fails. */
1377 #ifdef MATCH_MAY_ALLOCATE
1378 # define INIT_FAIL_STACK() \
1380 fail_stack.stack = (fail_stack_elt_t *) \
1381 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1382 * sizeof (fail_stack_elt_t)); \
1384 if (fail_stack.stack == NULL) \
1387 fail_stack.size = INIT_FAILURE_ALLOC; \
1388 fail_stack.avail = 0; \
1389 fail_stack.frame = 0; \
1392 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1394 # define INIT_FAIL_STACK() \
1396 fail_stack.avail = 0; \
1397 fail_stack.frame = 0; \
1400 # define RESET_FAIL_STACK() ((void)0)
1404 /* Double the size of FAIL_STACK, up to a limit
1405 which allows approximately `re_max_failures' items.
1407 Return 1 if succeeds, and 0 if either ran out of memory
1408 allocating space for it or it was already too large.
1410 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1412 /* Factor to increase the failure stack size by
1413 when we increase it.
1414 This used to be 2, but 2 was too wasteful
1415 because the old discarded stacks added up to as much space
1416 were as ultimate, maximum-size stack. */
1417 #define FAIL_STACK_GROWTH_FACTOR 4
1419 #define GROW_FAIL_STACK(fail_stack) \
1420 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1421 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1423 : ((fail_stack).stack \
1424 = (fail_stack_elt_t *) \
1425 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1426 (fail_stack).size * sizeof (fail_stack_elt_t), \
1427 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1428 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1429 * FAIL_STACK_GROWTH_FACTOR))), \
1431 (fail_stack).stack == NULL \
1433 : ((fail_stack).size \
1434 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1435 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1436 * FAIL_STACK_GROWTH_FACTOR)) \
1437 / sizeof (fail_stack_elt_t)), \
1441 /* Push a pointer value onto the failure stack.
1442 Assumes the variable `fail_stack'. Probably should only
1443 be called from within `PUSH_FAILURE_POINT'. */
1444 #define PUSH_FAILURE_POINTER(item) \
1445 fail_stack.stack[fail_stack.avail++].pointer = (item)
1447 /* This pushes an integer-valued item onto the failure stack.
1448 Assumes the variable `fail_stack'. Probably should only
1449 be called from within `PUSH_FAILURE_POINT'. */
1450 #define PUSH_FAILURE_INT(item) \
1451 fail_stack.stack[fail_stack.avail++].integer = (item)
1453 /* Push a fail_stack_elt_t value onto the failure stack.
1454 Assumes the variable `fail_stack'. Probably should only
1455 be called from within `PUSH_FAILURE_POINT'. */
1456 #define PUSH_FAILURE_ELT(item) \
1457 fail_stack.stack[fail_stack.avail++] = (item)
1459 /* These three POP... operations complement the three PUSH... operations.
1460 All assume that `fail_stack' is nonempty. */
1461 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1462 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1463 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1465 /* Individual items aside from the registers. */
1466 #define NUM_NONREG_ITEMS 3
1468 /* Used to examine the stack (to detect infinite loops). */
1469 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1470 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1471 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1472 #define TOP_FAILURE_HANDLE() fail_stack.frame
1475 #define ENSURE_FAIL_STACK(space) \
1476 while (REMAINING_AVAIL_SLOTS <= space) { \
1477 if (!GROW_FAIL_STACK (fail_stack)) \
1479 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1480 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1483 /* Push register NUM onto the stack. */
1484 #define PUSH_FAILURE_REG(num) \
1486 char *destination; \
1487 ENSURE_FAIL_STACK(3); \
1488 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1489 num, regstart[num], regend[num]); \
1490 PUSH_FAILURE_POINTER (regstart[num]); \
1491 PUSH_FAILURE_POINTER (regend[num]); \
1492 PUSH_FAILURE_INT (num); \
1495 /* Change the counter's value to VAL, but make sure that it will
1496 be reset when backtracking. */
1497 #define PUSH_NUMBER(ptr,val) \
1499 char *destination; \
1501 ENSURE_FAIL_STACK(3); \
1502 EXTRACT_NUMBER (c, ptr); \
1503 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1504 PUSH_FAILURE_INT (c); \
1505 PUSH_FAILURE_POINTER (ptr); \
1506 PUSH_FAILURE_INT (-1); \
1507 STORE_NUMBER (ptr, val); \
1510 /* Pop a saved register off the stack. */
1511 #define POP_FAILURE_REG_OR_COUNT() \
1513 int reg = POP_FAILURE_INT (); \
1516 /* It's a counter. */ \
1517 /* Here, we discard `const', making re_match non-reentrant. */ \
1518 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1519 reg = POP_FAILURE_INT (); \
1520 STORE_NUMBER (ptr, reg); \
1521 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1525 regend[reg] = POP_FAILURE_POINTER (); \
1526 regstart[reg] = POP_FAILURE_POINTER (); \
1527 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1528 reg, regstart[reg], regend[reg]); \
1532 /* Check that we are not stuck in an infinite loop. */
1533 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1535 int failure = TOP_FAILURE_HANDLE (); \
1536 /* Check for infinite matching loops */ \
1537 while (failure > 0 \
1538 && (FAILURE_STR (failure) == string_place \
1539 || FAILURE_STR (failure) == NULL)) \
1541 assert (FAILURE_PAT (failure) >= bufp->buffer \
1542 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1543 if (FAILURE_PAT (failure) == pat_cur) \
1548 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1549 failure = NEXT_FAILURE_HANDLE(failure); \
1551 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1554 /* Push the information about the state we will need
1555 if we ever fail back to it.
1557 Requires variables fail_stack, regstart, regend and
1558 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1561 Does `return FAILURE_CODE' if runs out of memory. */
1563 #define PUSH_FAILURE_POINT(pattern, string_place) \
1565 char *destination; \
1566 /* Must be int, so when we don't save any registers, the arithmetic \
1567 of 0 + -1 isn't done as unsigned. */ \
1569 DEBUG_STATEMENT (nfailure_points_pushed++); \
1570 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1571 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1572 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1574 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1576 DEBUG_PRINT1 ("\n"); \
1578 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1579 PUSH_FAILURE_INT (fail_stack.frame); \
1581 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1582 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1583 DEBUG_PRINT1 ("'\n"); \
1584 PUSH_FAILURE_POINTER (string_place); \
1586 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1587 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1588 PUSH_FAILURE_POINTER (pattern); \
1590 /* Close the frame by moving the frame pointer past it. */ \
1591 fail_stack.frame = fail_stack.avail; \
1594 /* Estimate the size of data pushed by a typical failure stack entry.
1595 An estimate is all we need, because all we use this for
1596 is to choose a limit for how big to make the failure stack. */
1597 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1598 #define TYPICAL_FAILURE_SIZE 20
1600 /* How many items can still be added to the stack without overflowing it. */
1601 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1604 /* Pops what PUSH_FAIL_STACK pushes.
1606 We restore into the parameters, all of which should be lvalues:
1607 STR -- the saved data position.
1608 PAT -- the saved pattern position.
1609 REGSTART, REGEND -- arrays of string positions.
1611 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1612 `pend', `string1', `size1', `string2', and `size2'. */
1614 #define POP_FAILURE_POINT(str, pat) \
1616 assert (!FAIL_STACK_EMPTY ()); \
1618 /* Remove failure points and point to how many regs pushed. */ \
1619 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1620 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1621 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1623 /* Pop the saved registers. */ \
1624 while (fail_stack.frame < fail_stack.avail) \
1625 POP_FAILURE_REG_OR_COUNT (); \
1627 pat = POP_FAILURE_POINTER (); \
1628 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1629 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1631 /* If the saved string location is NULL, it came from an \
1632 on_failure_keep_string_jump opcode, and we want to throw away the \
1633 saved NULL, thus retaining our current position in the string. */ \
1634 str = POP_FAILURE_POINTER (); \
1635 DEBUG_PRINT2 (" Popping string %p: `", str); \
1636 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1637 DEBUG_PRINT1 ("'\n"); \
1639 fail_stack.frame = POP_FAILURE_INT (); \
1640 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1642 assert (fail_stack.avail >= 0); \
1643 assert (fail_stack.frame <= fail_stack.avail); \
1645 DEBUG_STATEMENT (nfailure_points_popped++); \
1646 } while (0) /* POP_FAILURE_POINT */
1650 /* Registers are set to a sentinel when they haven't yet matched. */
1651 #define REG_UNSET(e) ((e) == NULL)
1653 /* Subroutine declarations and macros for regex_compile. */
1655 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1656 reg_syntax_t syntax
,
1657 struct re_pattern_buffer
*bufp
));
1658 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1659 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1660 int arg1
, int arg2
));
1661 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1662 int arg
, unsigned char *end
));
1663 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1664 int arg1
, int arg2
, unsigned char *end
));
1665 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1667 reg_syntax_t syntax
));
1668 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1670 reg_syntax_t syntax
));
1671 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1672 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1673 char *fastmap
, const int multibyte
));
1675 /* Fetch the next character in the uncompiled pattern, with no
1677 #define PATFETCH(c) \
1680 if (p == pend) return REG_EEND; \
1681 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1686 /* If `translate' is non-null, return translate[D], else just D. We
1687 cast the subscript to translate because some data is declared as
1688 `char *', to avoid warnings when a string constant is passed. But
1689 when we use a character as a subscript we must make it unsigned. */
1691 # define TRANSLATE(d) \
1692 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1696 /* Macros for outputting the compiled pattern into `buffer'. */
1698 /* If the buffer isn't allocated when it comes in, use this. */
1699 #define INIT_BUF_SIZE 32
1701 /* Make sure we have at least N more bytes of space in buffer. */
1702 #define GET_BUFFER_SPACE(n) \
1703 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1706 /* Make sure we have one more byte of buffer space and then add C to it. */
1707 #define BUF_PUSH(c) \
1709 GET_BUFFER_SPACE (1); \
1710 *b++ = (unsigned char) (c); \
1714 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1715 #define BUF_PUSH_2(c1, c2) \
1717 GET_BUFFER_SPACE (2); \
1718 *b++ = (unsigned char) (c1); \
1719 *b++ = (unsigned char) (c2); \
1723 /* As with BUF_PUSH_2, except for three bytes. */
1724 #define BUF_PUSH_3(c1, c2, c3) \
1726 GET_BUFFER_SPACE (3); \
1727 *b++ = (unsigned char) (c1); \
1728 *b++ = (unsigned char) (c2); \
1729 *b++ = (unsigned char) (c3); \
1733 /* Store a jump with opcode OP at LOC to location TO. We store a
1734 relative address offset by the three bytes the jump itself occupies. */
1735 #define STORE_JUMP(op, loc, to) \
1736 store_op1 (op, loc, (to) - (loc) - 3)
1738 /* Likewise, for a two-argument jump. */
1739 #define STORE_JUMP2(op, loc, to, arg) \
1740 store_op2 (op, loc, (to) - (loc) - 3, arg)
1742 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1743 #define INSERT_JUMP(op, loc, to) \
1744 insert_op1 (op, loc, (to) - (loc) - 3, b)
1746 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1747 #define INSERT_JUMP2(op, loc, to, arg) \
1748 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1751 /* This is not an arbitrary limit: the arguments which represent offsets
1752 into the pattern are two bytes long. So if 2^15 bytes turns out to
1753 be too small, many things would have to change. */
1754 # define MAX_BUF_SIZE (1L << 15)
1756 #if 0 /* This is when we thought it could be 2^16 bytes. */
1757 /* Any other compiler which, like MSC, has allocation limit below 2^16
1758 bytes will have to use approach similar to what was done below for
1759 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1760 reallocating to 0 bytes. Such thing is not going to work too well.
1761 You have been warned!! */
1762 #if defined _MSC_VER && !defined WIN32
1763 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1764 # define MAX_BUF_SIZE 65500L
1766 # define MAX_BUF_SIZE (1L << 16)
1770 /* Extend the buffer by twice its current size via realloc and
1771 reset the pointers that pointed into the old block to point to the
1772 correct places in the new one. If extending the buffer results in it
1773 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1774 #if __BOUNDED_POINTERS__
1775 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1776 # define MOVE_BUFFER_POINTER(P) \
1777 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1778 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1781 SET_HIGH_BOUND (b); \
1782 SET_HIGH_BOUND (begalt); \
1783 if (fixup_alt_jump) \
1784 SET_HIGH_BOUND (fixup_alt_jump); \
1786 SET_HIGH_BOUND (laststart); \
1787 if (pending_exact) \
1788 SET_HIGH_BOUND (pending_exact); \
1791 # define MOVE_BUFFER_POINTER(P) (P) += incr
1792 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1794 #define EXTEND_BUFFER() \
1796 re_char *old_buffer = bufp->buffer; \
1797 if (bufp->allocated == MAX_BUF_SIZE) \
1799 bufp->allocated <<= 1; \
1800 if (bufp->allocated > MAX_BUF_SIZE) \
1801 bufp->allocated = MAX_BUF_SIZE; \
1802 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1803 if (bufp->buffer == NULL) \
1804 return REG_ESPACE; \
1805 /* If the buffer moved, move all the pointers into it. */ \
1806 if (old_buffer != bufp->buffer) \
1808 int incr = bufp->buffer - old_buffer; \
1809 MOVE_BUFFER_POINTER (b); \
1810 MOVE_BUFFER_POINTER (begalt); \
1811 if (fixup_alt_jump) \
1812 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1814 MOVE_BUFFER_POINTER (laststart); \
1815 if (pending_exact) \
1816 MOVE_BUFFER_POINTER (pending_exact); \
1818 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1822 /* Since we have one byte reserved for the register number argument to
1823 {start,stop}_memory, the maximum number of groups we can report
1824 things about is what fits in that byte. */
1825 #define MAX_REGNUM 255
1827 /* But patterns can have more than `MAX_REGNUM' registers. We just
1828 ignore the excess. */
1829 typedef int regnum_t
;
1832 /* Macros for the compile stack. */
1834 /* Since offsets can go either forwards or backwards, this type needs to
1835 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1836 /* int may be not enough when sizeof(int) == 2. */
1837 typedef long pattern_offset_t
;
1841 pattern_offset_t begalt_offset
;
1842 pattern_offset_t fixup_alt_jump
;
1843 pattern_offset_t laststart_offset
;
1845 } compile_stack_elt_t
;
1850 compile_stack_elt_t
*stack
;
1852 unsigned avail
; /* Offset of next open position. */
1853 } compile_stack_type
;
1856 #define INIT_COMPILE_STACK_SIZE 32
1858 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1859 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1861 /* The next available element. */
1862 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1864 /* Explicit quit checking is only used on NTemacs. */
1865 #if defined WINDOWSNT && defined emacs && defined QUIT
1866 extern int immediate_quit
;
1867 # define IMMEDIATE_QUIT_CHECK \
1869 if (immediate_quit) QUIT; \
1872 # define IMMEDIATE_QUIT_CHECK ((void)0)
1875 /* Structure to manage work area for range table. */
1876 struct range_table_work_area
1878 int *table
; /* actual work area. */
1879 int allocated
; /* allocated size for work area in bytes. */
1880 int used
; /* actually used size in words. */
1881 int bits
; /* flag to record character classes */
1884 /* Make sure that WORK_AREA can hold more N multibyte characters.
1885 This is used only in set_image_of_range and set_image_of_range_1.
1886 It expects WORK_AREA to be a pointer.
1887 If it can't get the space, it returns from the surrounding function. */
1889 #define EXTEND_RANGE_TABLE(work_area, n) \
1891 if (((work_area)->used + (n)) * sizeof (int) > (work_area)->allocated) \
1893 extend_range_table_work_area (work_area); \
1894 if ((work_area)->table == 0) \
1895 return (REG_ESPACE); \
1899 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1900 (work_area).bits |= (bit)
1902 /* Bits used to implement the multibyte-part of the various character classes
1903 such as [:alnum:] in a charset's range table. */
1904 #define BIT_WORD 0x1
1905 #define BIT_LOWER 0x2
1906 #define BIT_PUNCT 0x4
1907 #define BIT_SPACE 0x8
1908 #define BIT_UPPER 0x10
1909 #define BIT_MULTIBYTE 0x20
1911 /* Set a range START..END to WORK_AREA.
1912 The range is passed through TRANSLATE, so START and END
1913 should be untranslated. */
1914 #define SET_RANGE_TABLE_WORK_AREA(work_area, start, end) \
1917 tem = set_image_of_range (&work_area, start, end, translate); \
1919 FREE_STACK_RETURN (tem); \
1922 /* Free allocated memory for WORK_AREA. */
1923 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1925 if ((work_area).table) \
1926 free ((work_area).table); \
1929 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1930 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1931 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1932 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1935 /* Set the bit for character C in a list. */
1936 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1939 /* Get the next unsigned number in the uncompiled pattern. */
1940 #define GET_UNSIGNED_NUMBER(num) \
1941 do { if (p != pend) \
1945 FREE_STACK_RETURN (REG_BADBR); \
1946 while ('0' <= c && c <= '9') \
1952 num = num * 10 + c - '0'; \
1953 if (num / 10 != prev) \
1954 FREE_STACK_RETURN (REG_BADBR); \
1960 FREE_STACK_RETURN (REG_BADBR); \
1964 #if ! WIDE_CHAR_SUPPORT
1966 /* Map a string to the char class it names (if any). */
1971 const char *string
= str
;
1972 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
1973 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
1974 else if (STREQ (string
, "word")) return RECC_WORD
;
1975 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
1976 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
1977 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
1978 else if (STREQ (string
, "lower")) return RECC_LOWER
;
1979 else if (STREQ (string
, "print")) return RECC_PRINT
;
1980 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
1981 else if (STREQ (string
, "space")) return RECC_SPACE
;
1982 else if (STREQ (string
, "upper")) return RECC_UPPER
;
1983 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
1984 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
1985 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
1986 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
1987 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
1988 else if (STREQ (string
, "blank")) return RECC_BLANK
;
1992 /* True iff CH is in the char class CC. */
1994 re_iswctype (ch
, cc
)
2000 case RECC_ALNUM
: return ISALNUM (ch
);
2001 case RECC_ALPHA
: return ISALPHA (ch
);
2002 case RECC_BLANK
: return ISBLANK (ch
);
2003 case RECC_CNTRL
: return ISCNTRL (ch
);
2004 case RECC_DIGIT
: return ISDIGIT (ch
);
2005 case RECC_GRAPH
: return ISGRAPH (ch
);
2006 case RECC_LOWER
: return ISLOWER (ch
);
2007 case RECC_PRINT
: return ISPRINT (ch
);
2008 case RECC_PUNCT
: return ISPUNCT (ch
);
2009 case RECC_SPACE
: return ISSPACE (ch
);
2010 case RECC_UPPER
: return ISUPPER (ch
);
2011 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2012 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2013 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2014 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2015 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2016 case RECC_WORD
: return ISWORD (ch
);
2017 case RECC_ERROR
: return false;
2023 /* Return a bit-pattern to use in the range-table bits to match multibyte
2024 chars of class CC. */
2026 re_wctype_to_bit (cc
)
2031 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2032 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2033 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2034 case RECC_LOWER
: return BIT_LOWER
;
2035 case RECC_UPPER
: return BIT_UPPER
;
2036 case RECC_PUNCT
: return BIT_PUNCT
;
2037 case RECC_SPACE
: return BIT_SPACE
;
2038 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2039 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2046 /* Filling in the work area of a range. */
2048 /* Actually extend the space in WORK_AREA. */
2051 extend_range_table_work_area (work_area
)
2052 struct range_table_work_area
*work_area
;
2054 work_area
->allocated
+= 16 * sizeof (int);
2055 if (work_area
->table
)
2057 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2060 = (int *) malloc (work_area
->allocated
);
2065 /* Carefully find the ranges of codes that are equivalent
2066 under case conversion to the range start..end when passed through
2067 TRANSLATE. Handle the case where non-letters can come in between
2068 two upper-case letters (which happens in Latin-1).
2069 Also handle the case of groups of more than 2 case-equivalent chars.
2071 The basic method is to look at consecutive characters and see
2072 if they can form a run that can be handled as one.
2074 Returns -1 if successful, REG_ESPACE if ran out of space. */
2077 set_image_of_range_1 (work_area
, start
, end
, translate
)
2078 RE_TRANSLATE_TYPE translate
;
2079 struct range_table_work_area
*work_area
;
2080 re_wchar_t start
, end
;
2082 /* `one_case' indicates a character, or a run of characters,
2083 each of which is an isolate (no case-equivalents).
2084 This includes all ASCII non-letters.
2086 `two_case' indicates a character, or a run of characters,
2087 each of which has two case-equivalent forms.
2088 This includes all ASCII letters.
2090 `strange' indicates a character that has more than one
2093 enum case_type
{one_case
, two_case
, strange
};
2095 /* Describe the run that is in progress,
2096 which the next character can try to extend.
2097 If run_type is strange, that means there really is no run.
2098 If run_type is one_case, then run_start...run_end is the run.
2099 If run_type is two_case, then the run is run_start...run_end,
2100 and the case-equivalents end at run_eqv_end. */
2102 enum case_type run_type
= strange
;
2103 int run_start
, run_end
, run_eqv_end
;
2105 Lisp_Object eqv_table
;
2107 if (!RE_TRANSLATE_P (translate
))
2109 EXTEND_RANGE_TABLE (work_area
, 2);
2110 work_area
->table
[work_area
->used
++] = (start
);
2111 work_area
->table
[work_area
->used
++] = (end
);
2115 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2117 for (; start
<= end
; start
++)
2119 enum case_type this_type
;
2120 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2121 int minchar
, maxchar
;
2123 /* Classify this character */
2125 this_type
= one_case
;
2126 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2127 this_type
= two_case
;
2129 this_type
= strange
;
2132 minchar
= start
, maxchar
= eqv
;
2134 minchar
= eqv
, maxchar
= start
;
2136 /* Can this character extend the run in progress? */
2137 if (this_type
== strange
|| this_type
!= run_type
2138 || !(minchar
== run_end
+ 1
2139 && (run_type
== two_case
2140 ? maxchar
== run_eqv_end
+ 1 : 1)))
2143 Record each of its equivalent ranges. */
2144 if (run_type
== one_case
)
2146 EXTEND_RANGE_TABLE (work_area
, 2);
2147 work_area
->table
[work_area
->used
++] = run_start
;
2148 work_area
->table
[work_area
->used
++] = run_end
;
2150 else if (run_type
== two_case
)
2152 EXTEND_RANGE_TABLE (work_area
, 4);
2153 work_area
->table
[work_area
->used
++] = run_start
;
2154 work_area
->table
[work_area
->used
++] = run_end
;
2155 work_area
->table
[work_area
->used
++]
2156 = RE_TRANSLATE (eqv_table
, run_start
);
2157 work_area
->table
[work_area
->used
++]
2158 = RE_TRANSLATE (eqv_table
, run_end
);
2163 if (this_type
== strange
)
2165 /* For a strange character, add each of its equivalents, one
2166 by one. Don't start a range. */
2169 EXTEND_RANGE_TABLE (work_area
, 2);
2170 work_area
->table
[work_area
->used
++] = eqv
;
2171 work_area
->table
[work_area
->used
++] = eqv
;
2172 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2174 while (eqv
!= start
);
2177 /* Add this char to the run, or start a new run. */
2178 else if (run_type
== strange
)
2180 /* Initialize a new range. */
2181 run_type
= this_type
;
2184 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2188 /* Extend a running range. */
2190 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2194 /* If a run is still in progress at the end, finish it now
2195 by recording its equivalent ranges. */
2196 if (run_type
== one_case
)
2198 EXTEND_RANGE_TABLE (work_area
, 2);
2199 work_area
->table
[work_area
->used
++] = run_start
;
2200 work_area
->table
[work_area
->used
++] = run_end
;
2202 else if (run_type
== two_case
)
2204 EXTEND_RANGE_TABLE (work_area
, 4);
2205 work_area
->table
[work_area
->used
++] = run_start
;
2206 work_area
->table
[work_area
->used
++] = run_end
;
2207 work_area
->table
[work_area
->used
++]
2208 = RE_TRANSLATE (eqv_table
, run_start
);
2209 work_area
->table
[work_area
->used
++]
2210 = RE_TRANSLATE (eqv_table
, run_end
);
2218 /* Record the the image of the range start..end when passed through
2219 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2220 and is not even necessarily contiguous.
2221 Normally we approximate it with the smallest contiguous range that contains
2222 all the chars we need. However, for Latin-1 we go to extra effort
2225 This function is not called for ASCII ranges.
2227 Returns -1 if successful, REG_ESPACE if ran out of space. */
2230 set_image_of_range (work_area
, start
, end
, translate
)
2231 RE_TRANSLATE_TYPE translate
;
2232 struct range_table_work_area
*work_area
;
2233 re_wchar_t start
, end
;
2235 re_wchar_t cmin
, cmax
;
2238 /* For Latin-1 ranges, use set_image_of_range_1
2239 to get proper handling of ranges that include letters and nonletters.
2240 For a range that includes the whole of Latin-1, this is not necessary.
2241 For other character sets, we don't bother to get this right. */
2242 if (RE_TRANSLATE_P (translate
) && start
< 04400
2243 && !(start
< 04200 && end
>= 04377))
2250 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2260 EXTEND_RANGE_TABLE (work_area
, 2);
2261 work_area
->table
[work_area
->used
++] = (start
);
2262 work_area
->table
[work_area
->used
++] = (end
);
2264 cmin
= -1, cmax
= -1;
2266 if (RE_TRANSLATE_P (translate
))
2270 for (ch
= start
; ch
<= end
; ch
++)
2272 re_wchar_t c
= TRANSLATE (ch
);
2273 if (! (start
<= c
&& c
<= end
))
2279 cmin
= MIN (cmin
, c
);
2280 cmax
= MAX (cmax
, c
);
2287 EXTEND_RANGE_TABLE (work_area
, 2);
2288 work_area
->table
[work_area
->used
++] = (cmin
);
2289 work_area
->table
[work_area
->used
++] = (cmax
);
2296 #ifndef MATCH_MAY_ALLOCATE
2298 /* If we cannot allocate large objects within re_match_2_internal,
2299 we make the fail stack and register vectors global.
2300 The fail stack, we grow to the maximum size when a regexp
2302 The register vectors, we adjust in size each time we
2303 compile a regexp, according to the number of registers it needs. */
2305 static fail_stack_type fail_stack
;
2307 /* Size with which the following vectors are currently allocated.
2308 That is so we can make them bigger as needed,
2309 but never make them smaller. */
2310 static int regs_allocated_size
;
2312 static re_char
** regstart
, ** regend
;
2313 static re_char
**best_regstart
, **best_regend
;
2315 /* Make the register vectors big enough for NUM_REGS registers,
2316 but don't make them smaller. */
2319 regex_grow_registers (num_regs
)
2322 if (num_regs
> regs_allocated_size
)
2324 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2325 RETALLOC_IF (regend
, num_regs
, re_char
*);
2326 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2327 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2329 regs_allocated_size
= num_regs
;
2333 #endif /* not MATCH_MAY_ALLOCATE */
2335 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2339 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2340 Returns one of error codes defined in `regex.h', or zero for success.
2342 Assumes the `allocated' (and perhaps `buffer') and `translate'
2343 fields are set in BUFP on entry.
2345 If it succeeds, results are put in BUFP (if it returns an error, the
2346 contents of BUFP are undefined):
2347 `buffer' is the compiled pattern;
2348 `syntax' is set to SYNTAX;
2349 `used' is set to the length of the compiled pattern;
2350 `fastmap_accurate' is zero;
2351 `re_nsub' is the number of subexpressions in PATTERN;
2352 `not_bol' and `not_eol' are zero;
2354 The `fastmap' field is neither examined nor set. */
2356 /* Insert the `jump' from the end of last alternative to "here".
2357 The space for the jump has already been allocated. */
2358 #define FIXUP_ALT_JUMP() \
2360 if (fixup_alt_jump) \
2361 STORE_JUMP (jump, fixup_alt_jump, b); \
2365 /* Return, freeing storage we allocated. */
2366 #define FREE_STACK_RETURN(value) \
2368 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2369 free (compile_stack.stack); \
2373 static reg_errcode_t
2374 regex_compile (pattern
, size
, syntax
, bufp
)
2377 reg_syntax_t syntax
;
2378 struct re_pattern_buffer
*bufp
;
2380 /* We fetch characters from PATTERN here. */
2381 register re_wchar_t c
, c1
;
2383 /* A random temporary spot in PATTERN. */
2386 /* Points to the end of the buffer, where we should append. */
2387 register unsigned char *b
;
2389 /* Keeps track of unclosed groups. */
2390 compile_stack_type compile_stack
;
2392 /* Points to the current (ending) position in the pattern. */
2394 /* `const' makes AIX compiler fail. */
2395 unsigned char *p
= pattern
;
2397 re_char
*p
= pattern
;
2399 re_char
*pend
= pattern
+ size
;
2401 /* How to translate the characters in the pattern. */
2402 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2404 /* Address of the count-byte of the most recently inserted `exactn'
2405 command. This makes it possible to tell if a new exact-match
2406 character can be added to that command or if the character requires
2407 a new `exactn' command. */
2408 unsigned char *pending_exact
= 0;
2410 /* Address of start of the most recently finished expression.
2411 This tells, e.g., postfix * where to find the start of its
2412 operand. Reset at the beginning of groups and alternatives. */
2413 unsigned char *laststart
= 0;
2415 /* Address of beginning of regexp, or inside of last group. */
2416 unsigned char *begalt
;
2418 /* Place in the uncompiled pattern (i.e., the {) to
2419 which to go back if the interval is invalid. */
2420 re_char
*beg_interval
;
2422 /* Address of the place where a forward jump should go to the end of
2423 the containing expression. Each alternative of an `or' -- except the
2424 last -- ends with a forward jump of this sort. */
2425 unsigned char *fixup_alt_jump
= 0;
2427 /* Counts open-groups as they are encountered. Remembered for the
2428 matching close-group on the compile stack, so the same register
2429 number is put in the stop_memory as the start_memory. */
2430 regnum_t regnum
= 0;
2432 /* Work area for range table of charset. */
2433 struct range_table_work_area range_table_work
;
2435 /* If the object matched can contain multibyte characters. */
2436 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2440 DEBUG_PRINT1 ("\nCompiling pattern: ");
2443 unsigned debug_count
;
2445 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2446 putchar (pattern
[debug_count
]);
2451 /* Initialize the compile stack. */
2452 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2453 if (compile_stack
.stack
== NULL
)
2456 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2457 compile_stack
.avail
= 0;
2459 range_table_work
.table
= 0;
2460 range_table_work
.allocated
= 0;
2462 /* Initialize the pattern buffer. */
2463 bufp
->syntax
= syntax
;
2464 bufp
->fastmap_accurate
= 0;
2465 bufp
->not_bol
= bufp
->not_eol
= 0;
2467 /* Set `used' to zero, so that if we return an error, the pattern
2468 printer (for debugging) will think there's no pattern. We reset it
2472 /* Always count groups, whether or not bufp->no_sub is set. */
2475 #if !defined emacs && !defined SYNTAX_TABLE
2476 /* Initialize the syntax table. */
2477 init_syntax_once ();
2480 if (bufp
->allocated
== 0)
2483 { /* If zero allocated, but buffer is non-null, try to realloc
2484 enough space. This loses if buffer's address is bogus, but
2485 that is the user's responsibility. */
2486 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2489 { /* Caller did not allocate a buffer. Do it for them. */
2490 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2492 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2494 bufp
->allocated
= INIT_BUF_SIZE
;
2497 begalt
= b
= bufp
->buffer
;
2499 /* Loop through the uncompiled pattern until we're at the end. */
2508 if ( /* If at start of pattern, it's an operator. */
2510 /* If context independent, it's an operator. */
2511 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2512 /* Otherwise, depends on what's come before. */
2513 || at_begline_loc_p (pattern
, p
, syntax
))
2514 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2523 if ( /* If at end of pattern, it's an operator. */
2525 /* If context independent, it's an operator. */
2526 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2527 /* Otherwise, depends on what's next. */
2528 || at_endline_loc_p (p
, pend
, syntax
))
2529 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2538 if ((syntax
& RE_BK_PLUS_QM
)
2539 || (syntax
& RE_LIMITED_OPS
))
2543 /* If there is no previous pattern... */
2546 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2547 FREE_STACK_RETURN (REG_BADRPT
);
2548 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2553 /* 1 means zero (many) matches is allowed. */
2554 boolean zero_times_ok
= 0, many_times_ok
= 0;
2557 /* If there is a sequence of repetition chars, collapse it
2558 down to just one (the right one). We can't combine
2559 interval operators with these because of, e.g., `a{2}*',
2560 which should only match an even number of `a's. */
2564 if ((syntax
& RE_FRUGAL
)
2565 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2569 zero_times_ok
|= c
!= '+';
2570 many_times_ok
|= c
!= '?';
2576 || (!(syntax
& RE_BK_PLUS_QM
)
2577 && (*p
== '+' || *p
== '?')))
2579 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2582 FREE_STACK_RETURN (REG_EESCAPE
);
2583 if (p
[1] == '+' || p
[1] == '?')
2584 PATFETCH (c
); /* Gobble up the backslash. */
2590 /* If we get here, we found another repeat character. */
2594 /* Star, etc. applied to an empty pattern is equivalent
2595 to an empty pattern. */
2596 if (!laststart
|| laststart
== b
)
2599 /* Now we know whether or not zero matches is allowed
2600 and also whether or not two or more matches is allowed. */
2605 boolean simple
= skip_one_char (laststart
) == b
;
2606 unsigned int startoffset
= 0;
2608 /* Check if the loop can match the empty string. */
2609 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2610 ? on_failure_jump
: on_failure_jump_loop
;
2611 assert (skip_one_char (laststart
) <= b
);
2613 if (!zero_times_ok
&& simple
)
2614 { /* Since simple * loops can be made faster by using
2615 on_failure_keep_string_jump, we turn simple P+
2616 into PP* if P is simple. */
2617 unsigned char *p1
, *p2
;
2618 startoffset
= b
- laststart
;
2619 GET_BUFFER_SPACE (startoffset
);
2620 p1
= b
; p2
= laststart
;
2626 GET_BUFFER_SPACE (6);
2629 STORE_JUMP (ofj
, b
, b
+ 6);
2631 /* Simple * loops can use on_failure_keep_string_jump
2632 depending on what follows. But since we don't know
2633 that yet, we leave the decision up to
2634 on_failure_jump_smart. */
2635 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2636 laststart
+ startoffset
, b
+ 6);
2638 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2643 /* A simple ? pattern. */
2644 assert (zero_times_ok
);
2645 GET_BUFFER_SPACE (3);
2646 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2650 else /* not greedy */
2651 { /* I wish the greedy and non-greedy cases could be merged. */
2653 GET_BUFFER_SPACE (7); /* We might use less. */
2656 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2658 /* The non-greedy multiple match looks like
2659 a repeat..until: we only need a conditional jump
2660 at the end of the loop. */
2661 if (emptyp
) BUF_PUSH (no_op
);
2662 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2663 : on_failure_jump
, b
, laststart
);
2667 /* The repeat...until naturally matches one or more.
2668 To also match zero times, we need to first jump to
2669 the end of the loop (its conditional jump). */
2670 INSERT_JUMP (jump
, laststart
, b
);
2676 /* non-greedy a?? */
2677 INSERT_JUMP (jump
, laststart
, b
+ 3);
2679 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2696 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2698 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2700 /* Ensure that we have enough space to push a charset: the
2701 opcode, the length count, and the bitset; 34 bytes in all. */
2702 GET_BUFFER_SPACE (34);
2706 /* We test `*p == '^' twice, instead of using an if
2707 statement, so we only need one BUF_PUSH. */
2708 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2712 /* Remember the first position in the bracket expression. */
2715 /* Push the number of bytes in the bitmap. */
2716 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2718 /* Clear the whole map. */
2719 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2721 /* charset_not matches newline according to a syntax bit. */
2722 if ((re_opcode_t
) b
[-2] == charset_not
2723 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2724 SET_LIST_BIT ('\n');
2726 /* Read in characters and ranges, setting map bits. */
2729 boolean escaped_char
= false;
2730 const unsigned char *p2
= p
;
2732 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2734 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2735 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2736 So the translation is done later in a loop. Example:
2737 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2740 /* \ might escape characters inside [...] and [^...]. */
2741 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2743 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2746 escaped_char
= true;
2750 /* Could be the end of the bracket expression. If it's
2751 not (i.e., when the bracket expression is `[]' so
2752 far), the ']' character bit gets set way below. */
2753 if (c
== ']' && p2
!= p1
)
2757 /* What should we do for the character which is
2758 greater than 0x7F, but not BASE_LEADING_CODE_P?
2761 /* See if we're at the beginning of a possible character
2764 if (!escaped_char
&&
2765 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2767 /* Leave room for the null. */
2768 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2769 const unsigned char *class_beg
;
2775 /* If pattern is `[[:'. */
2776 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2781 if ((c
== ':' && *p
== ']') || p
== pend
)
2783 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2786 /* This is in any case an invalid class name. */
2791 /* If isn't a word bracketed by `[:' and `:]':
2792 undo the ending character, the letters, and
2793 leave the leading `:' and `[' (but set bits for
2795 if (c
== ':' && *p
== ']')
2800 cc
= re_wctype (str
);
2803 FREE_STACK_RETURN (REG_ECTYPE
);
2805 /* Throw away the ] at the end of the character
2809 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2811 /* Most character classes in a multibyte match
2812 just set a flag. Exceptions are is_blank,
2813 is_digit, is_cntrl, and is_xdigit, since
2814 they can only match ASCII characters. We
2815 don't need to handle them for multibyte.
2816 They are distinguished by a negative wctype. */
2819 SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work
,
2820 re_wctype_to_bit (cc
));
2822 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2824 int translated
= TRANSLATE (ch
);
2825 if (re_iswctype (btowc (ch
), cc
))
2826 SET_LIST_BIT (translated
);
2829 /* Repeat the loop. */
2834 /* Go back to right after the "[:". */
2838 /* Because the `:' may starts the range, we
2839 can't simply set bit and repeat the loop.
2840 Instead, just set it to C and handle below. */
2845 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2848 /* Discard the `-'. */
2851 /* Fetch the character which ends the range. */
2854 if (SINGLE_BYTE_CHAR_P (c
))
2856 if (! SINGLE_BYTE_CHAR_P (c1
))
2858 /* Handle a range starting with a
2859 character of less than 256, and ending
2860 with a character of not less than 256.
2861 Split that into two ranges, the low one
2862 ending at 0377, and the high one
2863 starting at the smallest character in
2864 the charset of C1 and ending at C1. */
2865 int charset
= CHAR_CHARSET (c1
);
2866 re_wchar_t c2
= MAKE_CHAR (charset
, 0, 0);
2868 SET_RANGE_TABLE_WORK_AREA (range_table_work
,
2873 else if (!SAME_CHARSET_P (c
, c1
))
2874 FREE_STACK_RETURN (REG_ERANGE
);
2877 /* Range from C to C. */
2880 /* Set the range ... */
2881 if (SINGLE_BYTE_CHAR_P (c
))
2882 /* ... into bitmap. */
2884 re_wchar_t this_char
;
2885 re_wchar_t range_start
= c
, range_end
= c1
;
2887 /* If the start is after the end, the range is empty. */
2888 if (range_start
> range_end
)
2890 if (syntax
& RE_NO_EMPTY_RANGES
)
2891 FREE_STACK_RETURN (REG_ERANGE
);
2892 /* Else, repeat the loop. */
2896 for (this_char
= range_start
; this_char
<= range_end
;
2898 SET_LIST_BIT (TRANSLATE (this_char
));
2902 /* ... into range table. */
2903 SET_RANGE_TABLE_WORK_AREA (range_table_work
, c
, c1
);
2906 /* Discard any (non)matching list bytes that are all 0 at the
2907 end of the map. Decrease the map-length byte too. */
2908 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2912 /* Build real range table from work area. */
2913 if (RANGE_TABLE_WORK_USED (range_table_work
)
2914 || RANGE_TABLE_WORK_BITS (range_table_work
))
2917 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
2919 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2920 bytes for flags, two for COUNT, and three bytes for
2922 GET_BUFFER_SPACE (4 + used
* 3);
2924 /* Indicate the existence of range table. */
2925 laststart
[1] |= 0x80;
2927 /* Store the character class flag bits into the range table.
2928 If not in emacs, these flag bits are always 0. */
2929 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
2930 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
2932 STORE_NUMBER_AND_INCR (b
, used
/ 2);
2933 for (i
= 0; i
< used
; i
++)
2934 STORE_CHARACTER_AND_INCR
2935 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
2942 if (syntax
& RE_NO_BK_PARENS
)
2949 if (syntax
& RE_NO_BK_PARENS
)
2956 if (syntax
& RE_NEWLINE_ALT
)
2963 if (syntax
& RE_NO_BK_VBAR
)
2970 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2971 goto handle_interval
;
2977 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2979 /* Do not translate the character after the \, so that we can
2980 distinguish, e.g., \B from \b, even if we normally would
2981 translate, e.g., B to b. */
2987 if (syntax
& RE_NO_BK_PARENS
)
2988 goto normal_backslash
;
2995 /* Look for a special (?...) construct */
2996 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
2998 PATFETCH (c
); /* Gobble up the '?'. */
3002 case ':': shy
= 1; break;
3004 /* Only (?:...) is supported right now. */
3005 FREE_STACK_RETURN (REG_BADPAT
);
3016 if (COMPILE_STACK_FULL
)
3018 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3019 compile_stack_elt_t
);
3020 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3022 compile_stack
.size
<<= 1;
3025 /* These are the values to restore when we hit end of this
3026 group. They are all relative offsets, so that if the
3027 whole pattern moves because of realloc, they will still
3029 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3030 COMPILE_STACK_TOP
.fixup_alt_jump
3031 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3032 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3033 COMPILE_STACK_TOP
.regnum
= shy
? -regnum
: regnum
;
3036 start_memory for groups beyond the last one we can
3037 represent in the compiled pattern. */
3038 if (regnum
<= MAX_REGNUM
&& !shy
)
3039 BUF_PUSH_2 (start_memory
, regnum
);
3041 compile_stack
.avail
++;
3046 /* If we've reached MAX_REGNUM groups, then this open
3047 won't actually generate any code, so we'll have to
3048 clear pending_exact explicitly. */
3054 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3056 if (COMPILE_STACK_EMPTY
)
3058 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3059 goto normal_backslash
;
3061 FREE_STACK_RETURN (REG_ERPAREN
);
3067 /* See similar code for backslashed left paren above. */
3068 if (COMPILE_STACK_EMPTY
)
3070 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3073 FREE_STACK_RETURN (REG_ERPAREN
);
3076 /* Since we just checked for an empty stack above, this
3077 ``can't happen''. */
3078 assert (compile_stack
.avail
!= 0);
3080 /* We don't just want to restore into `regnum', because
3081 later groups should continue to be numbered higher,
3082 as in `(ab)c(de)' -- the second group is #2. */
3083 regnum_t this_group_regnum
;
3085 compile_stack
.avail
--;
3086 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3088 = COMPILE_STACK_TOP
.fixup_alt_jump
3089 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3091 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3092 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
3093 /* If we've reached MAX_REGNUM groups, then this open
3094 won't actually generate any code, so we'll have to
3095 clear pending_exact explicitly. */
3098 /* We're at the end of the group, so now we know how many
3099 groups were inside this one. */
3100 if (this_group_regnum
<= MAX_REGNUM
&& this_group_regnum
> 0)
3101 BUF_PUSH_2 (stop_memory
, this_group_regnum
);
3106 case '|': /* `\|'. */
3107 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3108 goto normal_backslash
;
3110 if (syntax
& RE_LIMITED_OPS
)
3113 /* Insert before the previous alternative a jump which
3114 jumps to this alternative if the former fails. */
3115 GET_BUFFER_SPACE (3);
3116 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3120 /* The alternative before this one has a jump after it
3121 which gets executed if it gets matched. Adjust that
3122 jump so it will jump to this alternative's analogous
3123 jump (put in below, which in turn will jump to the next
3124 (if any) alternative's such jump, etc.). The last such
3125 jump jumps to the correct final destination. A picture:
3131 If we are at `b', then fixup_alt_jump right now points to a
3132 three-byte space after `a'. We'll put in the jump, set
3133 fixup_alt_jump to right after `b', and leave behind three
3134 bytes which we'll fill in when we get to after `c'. */
3138 /* Mark and leave space for a jump after this alternative,
3139 to be filled in later either by next alternative or
3140 when know we're at the end of a series of alternatives. */
3142 GET_BUFFER_SPACE (3);
3151 /* If \{ is a literal. */
3152 if (!(syntax
& RE_INTERVALS
)
3153 /* If we're at `\{' and it's not the open-interval
3155 || (syntax
& RE_NO_BK_BRACES
))
3156 goto normal_backslash
;
3160 /* If got here, then the syntax allows intervals. */
3162 /* At least (most) this many matches must be made. */
3163 int lower_bound
= 0, upper_bound
= -1;
3168 FREE_STACK_RETURN (REG_EBRACE
);
3170 GET_UNSIGNED_NUMBER (lower_bound
);
3173 GET_UNSIGNED_NUMBER (upper_bound
);
3175 /* Interval such as `{1}' => match exactly once. */
3176 upper_bound
= lower_bound
;
3178 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3179 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3180 FREE_STACK_RETURN (REG_BADBR
);
3182 if (!(syntax
& RE_NO_BK_BRACES
))
3185 FREE_STACK_RETURN (REG_BADBR
);
3191 FREE_STACK_RETURN (REG_BADBR
);
3193 /* We just parsed a valid interval. */
3195 /* If it's invalid to have no preceding re. */
3198 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3199 FREE_STACK_RETURN (REG_BADRPT
);
3200 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3203 goto unfetch_interval
;
3206 if (upper_bound
== 0)
3207 /* If the upper bound is zero, just drop the sub pattern
3210 else if (lower_bound
== 1 && upper_bound
== 1)
3211 /* Just match it once: nothing to do here. */
3214 /* Otherwise, we have a nontrivial interval. When
3215 we're all done, the pattern will look like:
3216 set_number_at <jump count> <upper bound>
3217 set_number_at <succeed_n count> <lower bound>
3218 succeed_n <after jump addr> <succeed_n count>
3220 jump_n <succeed_n addr> <jump count>
3221 (The upper bound and `jump_n' are omitted if
3222 `upper_bound' is 1, though.) */
3224 { /* If the upper bound is > 1, we need to insert
3225 more at the end of the loop. */
3226 unsigned int nbytes
= (upper_bound
< 0 ? 3
3227 : upper_bound
> 1 ? 5 : 0);
3228 unsigned int startoffset
= 0;
3230 GET_BUFFER_SPACE (20); /* We might use less. */
3232 if (lower_bound
== 0)
3234 /* A succeed_n that starts with 0 is really a
3235 a simple on_failure_jump_loop. */
3236 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3242 /* Initialize lower bound of the `succeed_n', even
3243 though it will be set during matching by its
3244 attendant `set_number_at' (inserted next),
3245 because `re_compile_fastmap' needs to know.
3246 Jump to the `jump_n' we might insert below. */
3247 INSERT_JUMP2 (succeed_n
, laststart
,
3252 /* Code to initialize the lower bound. Insert
3253 before the `succeed_n'. The `5' is the last two
3254 bytes of this `set_number_at', plus 3 bytes of
3255 the following `succeed_n'. */
3256 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3261 if (upper_bound
< 0)
3263 /* A negative upper bound stands for infinity,
3264 in which case it degenerates to a plain jump. */
3265 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3268 else if (upper_bound
> 1)
3269 { /* More than one repetition is allowed, so
3270 append a backward jump to the `succeed_n'
3271 that starts this interval.
3273 When we've reached this during matching,
3274 we'll have matched the interval once, so
3275 jump back only `upper_bound - 1' times. */
3276 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3280 /* The location we want to set is the second
3281 parameter of the `jump_n'; that is `b-2' as
3282 an absolute address. `laststart' will be
3283 the `set_number_at' we're about to insert;
3284 `laststart+3' the number to set, the source
3285 for the relative address. But we are
3286 inserting into the middle of the pattern --
3287 so everything is getting moved up by 5.
3288 Conclusion: (b - 2) - (laststart + 3) + 5,
3289 i.e., b - laststart.
3291 We insert this at the beginning of the loop
3292 so that if we fail during matching, we'll
3293 reinitialize the bounds. */
3294 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3295 upper_bound
- 1, b
);
3300 beg_interval
= NULL
;
3305 /* If an invalid interval, match the characters as literals. */
3306 assert (beg_interval
);
3308 beg_interval
= NULL
;
3310 /* normal_char and normal_backslash need `c'. */
3313 if (!(syntax
& RE_NO_BK_BRACES
))
3315 assert (p
> pattern
&& p
[-1] == '\\');
3316 goto normal_backslash
;
3322 /* There is no way to specify the before_dot and after_dot
3323 operators. rms says this is ok. --karl */
3331 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3337 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3343 BUF_PUSH_2 (categoryspec
, c
);
3349 BUF_PUSH_2 (notcategoryspec
, c
);
3355 if (syntax
& RE_NO_GNU_OPS
)
3358 BUF_PUSH_2 (syntaxspec
, Sword
);
3363 if (syntax
& RE_NO_GNU_OPS
)
3366 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3371 if (syntax
& RE_NO_GNU_OPS
)
3377 if (syntax
& RE_NO_GNU_OPS
)
3383 if (syntax
& RE_NO_GNU_OPS
)
3392 FREE_STACK_RETURN (REG_BADPAT
);
3396 if (syntax
& RE_NO_GNU_OPS
)
3398 BUF_PUSH (wordbound
);
3402 if (syntax
& RE_NO_GNU_OPS
)
3404 BUF_PUSH (notwordbound
);
3408 if (syntax
& RE_NO_GNU_OPS
)
3414 if (syntax
& RE_NO_GNU_OPS
)
3419 case '1': case '2': case '3': case '4': case '5':
3420 case '6': case '7': case '8': case '9':
3424 if (syntax
& RE_NO_BK_REFS
)
3425 goto normal_backslash
;
3429 /* Can't back reference to a subexpression before its end. */
3430 if (reg
> regnum
|| group_in_compile_stack (compile_stack
, reg
))
3431 FREE_STACK_RETURN (REG_ESUBREG
);
3434 BUF_PUSH_2 (duplicate
, reg
);
3441 if (syntax
& RE_BK_PLUS_QM
)
3444 goto normal_backslash
;
3448 /* You might think it would be useful for \ to mean
3449 not to translate; but if we don't translate it
3450 it will never match anything. */
3457 /* Expects the character in `c'. */
3459 /* If no exactn currently being built. */
3462 /* If last exactn not at current position. */
3463 || pending_exact
+ *pending_exact
+ 1 != b
3465 /* We have only one byte following the exactn for the count. */
3466 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3468 /* If followed by a repetition operator. */
3469 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3470 || ((syntax
& RE_BK_PLUS_QM
)
3471 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3472 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3473 || ((syntax
& RE_INTERVALS
)
3474 && ((syntax
& RE_NO_BK_BRACES
)
3475 ? p
!= pend
&& *p
== '{'
3476 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3478 /* Start building a new exactn. */
3482 BUF_PUSH_2 (exactn
, 0);
3483 pending_exact
= b
- 1;
3486 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3492 len
= CHAR_STRING (c
, b
);
3496 (*pending_exact
) += len
;
3501 } /* while p != pend */
3504 /* Through the pattern now. */
3508 if (!COMPILE_STACK_EMPTY
)
3509 FREE_STACK_RETURN (REG_EPAREN
);
3511 /* If we don't want backtracking, force success
3512 the first time we reach the end of the compiled pattern. */
3513 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3516 /* We have succeeded; set the length of the buffer. */
3517 bufp
->used
= b
- bufp
->buffer
;
3522 re_compile_fastmap (bufp
);
3523 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3524 print_compiled_pattern (bufp
);
3529 #ifndef MATCH_MAY_ALLOCATE
3530 /* Initialize the failure stack to the largest possible stack. This
3531 isn't necessary unless we're trying to avoid calling alloca in
3532 the search and match routines. */
3534 int num_regs
= bufp
->re_nsub
+ 1;
3536 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3538 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3540 if (! fail_stack
.stack
)
3542 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3543 * sizeof (fail_stack_elt_t
));
3546 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3548 * sizeof (fail_stack_elt_t
)));
3551 regex_grow_registers (num_regs
);
3553 #endif /* not MATCH_MAY_ALLOCATE */
3555 FREE_STACK_RETURN (REG_NOERROR
);
3556 } /* regex_compile */
3558 /* Subroutines for `regex_compile'. */
3560 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3563 store_op1 (op
, loc
, arg
)
3568 *loc
= (unsigned char) op
;
3569 STORE_NUMBER (loc
+ 1, arg
);
3573 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3576 store_op2 (op
, loc
, arg1
, arg2
)
3581 *loc
= (unsigned char) op
;
3582 STORE_NUMBER (loc
+ 1, arg1
);
3583 STORE_NUMBER (loc
+ 3, arg2
);
3587 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3588 for OP followed by two-byte integer parameter ARG. */
3591 insert_op1 (op
, loc
, arg
, end
)
3597 register unsigned char *pfrom
= end
;
3598 register unsigned char *pto
= end
+ 3;
3600 while (pfrom
!= loc
)
3603 store_op1 (op
, loc
, arg
);
3607 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3610 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3616 register unsigned char *pfrom
= end
;
3617 register unsigned char *pto
= end
+ 5;
3619 while (pfrom
!= loc
)
3622 store_op2 (op
, loc
, arg1
, arg2
);
3626 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3627 after an alternative or a begin-subexpression. We assume there is at
3628 least one character before the ^. */
3631 at_begline_loc_p (pattern
, p
, syntax
)
3632 re_char
*pattern
, *p
;
3633 reg_syntax_t syntax
;
3635 re_char
*prev
= p
- 2;
3636 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3639 /* After a subexpression? */
3640 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3641 /* After an alternative? */
3642 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3643 /* After a shy subexpression? */
3644 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3645 && prev
[-1] == '?' && prev
[-2] == '('
3646 && (syntax
& RE_NO_BK_PARENS
3647 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3651 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3652 at least one character after the $, i.e., `P < PEND'. */
3655 at_endline_loc_p (p
, pend
, syntax
)
3657 reg_syntax_t syntax
;
3660 boolean next_backslash
= *next
== '\\';
3661 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3664 /* Before a subexpression? */
3665 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3666 : next_backslash
&& next_next
&& *next_next
== ')')
3667 /* Before an alternative? */
3668 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3669 : next_backslash
&& next_next
&& *next_next
== '|');
3673 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3674 false if it's not. */
3677 group_in_compile_stack (compile_stack
, regnum
)
3678 compile_stack_type compile_stack
;
3683 for (this_element
= compile_stack
.avail
- 1;
3686 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3693 If fastmap is non-NULL, go through the pattern and fill fastmap
3694 with all the possible leading chars. If fastmap is NULL, don't
3695 bother filling it up (obviously) and only return whether the
3696 pattern could potentially match the empty string.
3698 Return 1 if p..pend might match the empty string.
3699 Return 0 if p..pend matches at least one char.
3700 Return -1 if fastmap was not updated accurately. */
3703 analyse_first (p
, pend
, fastmap
, multibyte
)
3706 const int multibyte
;
3711 /* If all elements for base leading-codes in fastmap is set, this
3712 flag is set true. */
3713 boolean match_any_multibyte_characters
= false;
3717 /* The loop below works as follows:
3718 - It has a working-list kept in the PATTERN_STACK and which basically
3719 starts by only containing a pointer to the first operation.
3720 - If the opcode we're looking at is a match against some set of
3721 chars, then we add those chars to the fastmap and go on to the
3722 next work element from the worklist (done via `break').
3723 - If the opcode is a control operator on the other hand, we either
3724 ignore it (if it's meaningless at this point, such as `start_memory')
3725 or execute it (if it's a jump). If the jump has several destinations
3726 (i.e. `on_failure_jump'), then we push the other destination onto the
3728 We guarantee termination by ignoring backward jumps (more or less),
3729 so that `p' is monotonically increasing. More to the point, we
3730 never set `p' (or push) anything `<= p1'. */
3734 /* `p1' is used as a marker of how far back a `on_failure_jump'
3735 can go without being ignored. It is normally equal to `p'
3736 (which prevents any backward `on_failure_jump') except right
3737 after a plain `jump', to allow patterns such as:
3740 10: on_failure_jump 3
3741 as used for the *? operator. */
3744 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3751 /* If the first character has to match a backreference, that means
3752 that the group was empty (since it already matched). Since this
3753 is the only case that interests us here, we can assume that the
3754 backreference must match the empty string. */
3759 /* Following are the cases which match a character. These end
3765 int c
= RE_STRING_CHAR (p
+ 1, pend
- p
);
3767 if (SINGLE_BYTE_CHAR_P (c
))
3776 /* We could put all the chars except for \n (and maybe \0)
3777 but we don't bother since it is generally not worth it. */
3778 if (!fastmap
) break;
3783 /* Chars beyond end of bitmap are possible matches.
3784 All the single-byte codes can occur in multibyte buffers.
3785 So any that are not listed in the charset
3786 are possible matches, even in multibyte buffers. */
3787 if (!fastmap
) break;
3788 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3789 j
< (1 << BYTEWIDTH
); j
++)
3793 if (!fastmap
) break;
3794 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3795 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3797 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3800 if ((not && multibyte
)
3801 /* Any character set can possibly contain a character
3802 which doesn't match the specified set of characters. */
3803 || (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3804 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3805 /* If we can match a character class, we can match
3806 any character set. */
3808 set_fastmap_for_multibyte_characters
:
3809 if (match_any_multibyte_characters
== false)
3811 for (j
= 0x80; j
< 0xA0; j
++) /* XXX */
3812 if (BASE_LEADING_CODE_P (j
))
3814 match_any_multibyte_characters
= true;
3818 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3819 && match_any_multibyte_characters
== false)
3821 /* Set fastmap[I] 1 where I is a base leading code of each
3822 multibyte character in the range table. */
3825 /* Make P points the range table. `+ 2' is to skip flag
3826 bits for a character class. */
3827 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3829 /* Extract the number of ranges in range table into COUNT. */
3830 EXTRACT_NUMBER_AND_INCR (count
, p
);
3831 for (; count
> 0; count
--, p
+= 2 * 3) /* XXX */
3833 /* Extract the start of each range. */
3834 EXTRACT_CHARACTER (c
, p
);
3835 j
= CHAR_CHARSET (c
);
3836 fastmap
[CHARSET_LEADING_CODE_BASE (j
)] = 1;
3843 if (!fastmap
) break;
3845 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
3847 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3848 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
3852 /* This match depends on text properties. These end with
3853 aborting optimizations. */
3857 case notcategoryspec
:
3858 if (!fastmap
) break;
3859 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
3861 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3862 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
3866 /* Any character set can possibly contain a character
3867 whose category is K (or not). */
3868 goto set_fastmap_for_multibyte_characters
;
3871 /* All cases after this match the empty string. These end with
3893 EXTRACT_NUMBER_AND_INCR (j
, p
);
3895 /* Backward jumps can only go back to code that we've already
3896 visited. `re_compile' should make sure this is true. */
3899 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
3901 case on_failure_jump
:
3902 case on_failure_keep_string_jump
:
3903 case on_failure_jump_loop
:
3904 case on_failure_jump_nastyloop
:
3905 case on_failure_jump_smart
:
3911 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
3912 to jump back to "just after here". */
3915 case on_failure_jump
:
3916 case on_failure_keep_string_jump
:
3917 case on_failure_jump_nastyloop
:
3918 case on_failure_jump_loop
:
3919 case on_failure_jump_smart
:
3920 EXTRACT_NUMBER_AND_INCR (j
, p
);
3922 ; /* Backward jump to be ignored. */
3924 { /* We have to look down both arms.
3925 We first go down the "straight" path so as to minimize
3926 stack usage when going through alternatives. */
3927 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
3935 /* This code simply does not properly handle forward jump_n. */
3936 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
3938 /* jump_n can either jump or fall through. The (backward) jump
3939 case has already been handled, so we only need to look at the
3940 fallthrough case. */
3944 /* If N == 0, it should be an on_failure_jump_loop instead. */
3945 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
3947 /* We only care about one iteration of the loop, so we don't
3948 need to consider the case where this behaves like an
3965 abort (); /* We have listed all the cases. */
3968 /* Getting here means we have found the possible starting
3969 characters for one path of the pattern -- and that the empty
3970 string does not match. We need not follow this path further. */
3974 /* We reached the end without matching anything. */
3977 } /* analyse_first */
3979 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3980 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3981 characters can start a string that matches the pattern. This fastmap
3982 is used by re_search to skip quickly over impossible starting points.
3984 Character codes above (1 << BYTEWIDTH) are not represented in the
3985 fastmap, but the leading codes are represented. Thus, the fastmap
3986 indicates which character sets could start a match.
3988 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3989 area as BUFP->fastmap.
3991 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3994 Returns 0 if we succeed, -2 if an internal error. */
3997 re_compile_fastmap (bufp
)
3998 struct re_pattern_buffer
*bufp
;
4000 char *fastmap
= bufp
->fastmap
;
4003 assert (fastmap
&& bufp
->buffer
);
4005 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4006 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4008 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4009 fastmap
, RE_MULTIBYTE_P (bufp
));
4010 bufp
->can_be_null
= (analysis
!= 0);
4012 } /* re_compile_fastmap */
4014 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4015 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4016 this memory for recording register information. STARTS and ENDS
4017 must be allocated using the malloc library routine, and must each
4018 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4020 If NUM_REGS == 0, then subsequent matches should allocate their own
4023 Unless this function is called, the first search or match using
4024 PATTERN_BUFFER will allocate its own register data, without
4025 freeing the old data. */
4028 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4029 struct re_pattern_buffer
*bufp
;
4030 struct re_registers
*regs
;
4032 regoff_t
*starts
, *ends
;
4036 bufp
->regs_allocated
= REGS_REALLOCATE
;
4037 regs
->num_regs
= num_regs
;
4038 regs
->start
= starts
;
4043 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4045 regs
->start
= regs
->end
= (regoff_t
*) 0;
4048 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4050 /* Searching routines. */
4052 /* Like re_search_2, below, but only one string is specified, and
4053 doesn't let you say where to stop matching. */
4056 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4057 struct re_pattern_buffer
*bufp
;
4059 int size
, startpos
, range
;
4060 struct re_registers
*regs
;
4062 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4065 WEAK_ALIAS (__re_search
, re_search
)
4067 /* Head address of virtual concatenation of string. */
4068 #define HEAD_ADDR_VSTRING(P) \
4069 (((P) >= size1 ? string2 : string1))
4071 /* End address of virtual concatenation of string. */
4072 #define STOP_ADDR_VSTRING(P) \
4073 (((P) >= size1 ? string2 + size2 : string1 + size1))
4075 /* Address of POS in the concatenation of virtual string. */
4076 #define POS_ADDR_VSTRING(POS) \
4077 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4079 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4080 virtual concatenation of STRING1 and STRING2, starting first at index
4081 STARTPOS, then at STARTPOS + 1, and so on.
4083 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4085 RANGE is how far to scan while trying to match. RANGE = 0 means try
4086 only at STARTPOS; in general, the last start tried is STARTPOS +
4089 In REGS, return the indices of the virtual concatenation of STRING1
4090 and STRING2 that matched the entire BUFP->buffer and its contained
4093 Do not consider matching one past the index STOP in the virtual
4094 concatenation of STRING1 and STRING2.
4096 We return either the position in the strings at which the match was
4097 found, -1 if no match, or -2 if error (such as failure
4101 re_search_2 (bufp
, str1
, size1
, str2
, size2
, startpos
, range
, regs
, stop
)
4102 struct re_pattern_buffer
*bufp
;
4103 const char *str1
, *str2
;
4107 struct re_registers
*regs
;
4111 re_char
*string1
= (re_char
*) str1
;
4112 re_char
*string2
= (re_char
*) str2
;
4113 register char *fastmap
= bufp
->fastmap
;
4114 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4115 int total_size
= size1
+ size2
;
4116 int endpos
= startpos
+ range
;
4117 boolean anchored_start
;
4119 /* Nonzero if we have to concern multibyte character. */
4120 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4122 /* Check for out-of-range STARTPOS. */
4123 if (startpos
< 0 || startpos
> total_size
)
4126 /* Fix up RANGE if it might eventually take us outside
4127 the virtual concatenation of STRING1 and STRING2.
4128 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4130 range
= 0 - startpos
;
4131 else if (endpos
> total_size
)
4132 range
= total_size
- startpos
;
4134 /* If the search isn't to be a backwards one, don't waste time in a
4135 search for a pattern anchored at beginning of buffer. */
4136 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4145 /* In a forward search for something that starts with \=.
4146 don't keep searching past point. */
4147 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4149 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4155 /* Update the fastmap now if not correct already. */
4156 if (fastmap
&& !bufp
->fastmap_accurate
)
4157 re_compile_fastmap (bufp
);
4159 /* See whether the pattern is anchored. */
4160 anchored_start
= (bufp
->buffer
[0] == begline
);
4163 gl_state
.object
= re_match_object
;
4165 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4167 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4171 /* Loop through the string, looking for a place to start matching. */
4174 /* If the pattern is anchored,
4175 skip quickly past places we cannot match.
4176 We don't bother to treat startpos == 0 specially
4177 because that case doesn't repeat. */
4178 if (anchored_start
&& startpos
> 0)
4180 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4181 : string2
[startpos
- size1
- 1])
4186 /* If a fastmap is supplied, skip quickly over characters that
4187 cannot be the start of a match. If the pattern can match the
4188 null string, however, we don't need to skip characters; we want
4189 the first null string. */
4190 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4192 register re_char
*d
;
4193 register re_wchar_t buf_ch
;
4195 d
= POS_ADDR_VSTRING (startpos
);
4197 if (range
> 0) /* Searching forwards. */
4199 register int lim
= 0;
4202 if (startpos
< size1
&& startpos
+ range
>= size1
)
4203 lim
= range
- (size1
- startpos
);
4205 /* Written out as an if-else to avoid testing `translate'
4207 if (RE_TRANSLATE_P (translate
))
4214 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4217 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4222 range
-= buf_charlen
;
4227 && !fastmap
[RE_TRANSLATE (translate
, *d
)])
4234 while (range
> lim
&& !fastmap
[*d
])
4240 startpos
+= irange
- range
;
4242 else /* Searching backwards. */
4244 int room
= (startpos
>= size1
4245 ? size2
+ size1
- startpos
4246 : size1
- startpos
);
4247 buf_ch
= RE_STRING_CHAR (d
, room
);
4248 buf_ch
= TRANSLATE (buf_ch
);
4250 if (! (buf_ch
>= 0400
4251 || fastmap
[buf_ch
]))
4256 /* If can't match the null string, and that's all we have left, fail. */
4257 if (range
>= 0 && startpos
== total_size
&& fastmap
4258 && !bufp
->can_be_null
)
4261 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4262 startpos
, regs
, stop
);
4263 #ifndef REGEX_MALLOC
4280 /* Update STARTPOS to the next character boundary. */
4283 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4284 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4285 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
4303 /* Update STARTPOS to the previous character boundary. */
4306 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4308 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4310 /* Find the head of multibyte form. */
4311 PREV_CHAR_BOUNDARY (p
, phead
);
4312 range
+= p0
- 1 - p
;
4316 startpos
-= p0
- 1 - p
;
4322 WEAK_ALIAS (__re_search_2
, re_search_2
)
4324 /* Declarations and macros for re_match_2. */
4326 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4328 RE_TRANSLATE_TYPE translate
,
4329 const int multibyte
));
4331 /* This converts PTR, a pointer into one of the search strings `string1'
4332 and `string2' into an offset from the beginning of that string. */
4333 #define POINTER_TO_OFFSET(ptr) \
4334 (FIRST_STRING_P (ptr) \
4335 ? ((regoff_t) ((ptr) - string1)) \
4336 : ((regoff_t) ((ptr) - string2 + size1)))
4338 /* Call before fetching a character with *d. This switches over to
4339 string2 if necessary.
4340 Check re_match_2_internal for a discussion of why end_match_2 might
4341 not be within string2 (but be equal to end_match_1 instead). */
4342 #define PREFETCH() \
4345 /* End of string2 => fail. */ \
4346 if (dend == end_match_2) \
4348 /* End of string1 => advance to string2. */ \
4350 dend = end_match_2; \
4353 /* Call before fetching a char with *d if you already checked other limits.
4354 This is meant for use in lookahead operations like wordend, etc..
4355 where we might need to look at parts of the string that might be
4356 outside of the LIMITs (i.e past `stop'). */
4357 #define PREFETCH_NOLIMIT() \
4361 dend = end_match_2; \
4364 /* Test if at very beginning or at very end of the virtual concatenation
4365 of `string1' and `string2'. If only one string, it's `string2'. */
4366 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4367 #define AT_STRINGS_END(d) ((d) == end2)
4370 /* Test if D points to a character which is word-constituent. We have
4371 two special cases to check for: if past the end of string1, look at
4372 the first character in string2; and if before the beginning of
4373 string2, look at the last character in string1. */
4374 #define WORDCHAR_P(d) \
4375 (SYNTAX ((d) == end1 ? *string2 \
4376 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4379 /* Disabled due to a compiler bug -- see comment at case wordbound */
4381 /* The comment at case wordbound is following one, but we don't use
4382 AT_WORD_BOUNDARY anymore to support multibyte form.
4384 The DEC Alpha C compiler 3.x generates incorrect code for the
4385 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4386 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4387 macro and introducing temporary variables works around the bug. */
4390 /* Test if the character before D and the one at D differ with respect
4391 to being word-constituent. */
4392 #define AT_WORD_BOUNDARY(d) \
4393 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4394 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4397 /* Free everything we malloc. */
4398 #ifdef MATCH_MAY_ALLOCATE
4399 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4400 # define FREE_VARIABLES() \
4402 REGEX_FREE_STACK (fail_stack.stack); \
4403 FREE_VAR (regstart); \
4404 FREE_VAR (regend); \
4405 FREE_VAR (best_regstart); \
4406 FREE_VAR (best_regend); \
4409 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4410 #endif /* not MATCH_MAY_ALLOCATE */
4413 /* Optimization routines. */
4415 /* If the operation is a match against one or more chars,
4416 return a pointer to the next operation, else return NULL. */
4421 switch (SWITCH_ENUM_CAST (*p
++))
4432 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4435 p
= CHARSET_RANGE_TABLE (p
- 1);
4436 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4437 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4440 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4447 case notcategoryspec
:
4459 /* Jump over non-matching operations. */
4461 skip_noops (p
, pend
)
4467 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4476 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4487 /* Non-zero if "p1 matches something" implies "p2 fails". */
4489 mutually_exclusive_p (bufp
, p1
, p2
)
4490 struct re_pattern_buffer
*bufp
;
4494 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4495 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4497 assert (p1
>= bufp
->buffer
&& p1
< pend
4498 && p2
>= bufp
->buffer
&& p2
<= pend
);
4500 /* Skip over open/close-group commands.
4501 If what follows this loop is a ...+ construct,
4502 look at what begins its body, since we will have to
4503 match at least one of that. */
4504 p2
= skip_noops (p2
, pend
);
4505 /* The same skip can be done for p1, except that this function
4506 is only used in the case where p1 is a simple match operator. */
4507 /* p1 = skip_noops (p1, pend); */
4509 assert (p1
>= bufp
->buffer
&& p1
< pend
4510 && p2
>= bufp
->buffer
&& p2
<= pend
);
4512 op2
= p2
== pend
? succeed
: *p2
;
4514 switch (SWITCH_ENUM_CAST (op2
))
4518 /* If we're at the end of the pattern, we can change. */
4519 if (skip_one_char (p1
))
4521 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4529 register re_wchar_t c
4530 = (re_opcode_t
) *p2
== endline
? '\n'
4531 : RE_STRING_CHAR (p2
+ 2, pend
- p2
- 2);
4533 if ((re_opcode_t
) *p1
== exactn
)
4535 if (c
!= RE_STRING_CHAR (p1
+ 2, pend
- p1
- 2))
4537 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4542 else if ((re_opcode_t
) *p1
== charset
4543 || (re_opcode_t
) *p1
== charset_not
)
4545 int not = (re_opcode_t
) *p1
== charset_not
;
4547 /* Test if C is listed in charset (or charset_not)
4549 if (SINGLE_BYTE_CHAR_P (c
))
4551 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4552 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4555 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4556 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4558 /* `not' is equal to 1 if c would match, which means
4559 that we can't change to pop_failure_jump. */
4562 DEBUG_PRINT1 (" No match => fast loop.\n");
4566 else if ((re_opcode_t
) *p1
== anychar
4569 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4577 if ((re_opcode_t
) *p1
== exactn
)
4578 /* Reuse the code above. */
4579 return mutually_exclusive_p (bufp
, p2
, p1
);
4581 /* It is hard to list up all the character in charset
4582 P2 if it includes multibyte character. Give up in
4584 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4586 /* Now, we are sure that P2 has no range table.
4587 So, for the size of bitmap in P2, `p2[1]' is
4588 enough. But P1 may have range table, so the
4589 size of bitmap table of P1 is extracted by
4590 using macro `CHARSET_BITMAP_SIZE'.
4592 Since we know that all the character listed in
4593 P2 is ASCII, it is enough to test only bitmap
4596 if ((re_opcode_t
) *p1
== charset
)
4599 /* We win if the charset inside the loop
4600 has no overlap with the one after the loop. */
4603 && idx
< CHARSET_BITMAP_SIZE (p1
));
4605 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4609 || idx
== CHARSET_BITMAP_SIZE (p1
))
4611 DEBUG_PRINT1 (" No match => fast loop.\n");
4615 else if ((re_opcode_t
) *p1
== charset_not
)
4618 /* We win if the charset_not inside the loop lists
4619 every character listed in the charset after. */
4620 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4621 if (! (p2
[2 + idx
] == 0
4622 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4623 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4628 DEBUG_PRINT1 (" No match => fast loop.\n");
4637 switch (SWITCH_ENUM_CAST (*p1
))
4641 /* Reuse the code above. */
4642 return mutually_exclusive_p (bufp
, p2
, p1
);
4644 /* When we have two charset_not, it's very unlikely that
4645 they don't overlap. The union of the two sets of excluded
4646 chars should cover all possible chars, which, as a matter of
4647 fact, is virtually impossible in multibyte buffers. */
4653 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4655 return ((re_opcode_t
) *p1
== syntaxspec
4656 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4658 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4661 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4663 return ((re_opcode_t
) *p1
== notsyntaxspec
4664 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4666 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4669 return (((re_opcode_t
) *p1
== notsyntaxspec
4670 || (re_opcode_t
) *p1
== syntaxspec
)
4675 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4676 case notcategoryspec
:
4677 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4689 /* Matching routines. */
4691 #ifndef emacs /* Emacs never uses this. */
4692 /* re_match is like re_match_2 except it takes only a single string. */
4695 re_match (bufp
, string
, size
, pos
, regs
)
4696 struct re_pattern_buffer
*bufp
;
4699 struct re_registers
*regs
;
4701 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
4703 # if defined C_ALLOCA && !defined REGEX_MALLOC
4708 WEAK_ALIAS (__re_match
, re_match
)
4709 #endif /* not emacs */
4712 /* In Emacs, this is the string or buffer in which we
4713 are matching. It is used for looking up syntax properties. */
4714 Lisp_Object re_match_object
;
4717 /* re_match_2 matches the compiled pattern in BUFP against the
4718 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4719 and SIZE2, respectively). We start matching at POS, and stop
4722 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4723 store offsets for the substring each group matched in REGS. See the
4724 documentation for exactly how many groups we fill.
4726 We return -1 if no match, -2 if an internal error (such as the
4727 failure stack overflowing). Otherwise, we return the length of the
4728 matched substring. */
4731 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4732 struct re_pattern_buffer
*bufp
;
4733 const char *string1
, *string2
;
4736 struct re_registers
*regs
;
4743 gl_state
.object
= re_match_object
;
4744 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4745 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4748 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4749 (re_char
*) string2
, size2
,
4751 #if defined C_ALLOCA && !defined REGEX_MALLOC
4756 WEAK_ALIAS (__re_match_2
, re_match_2
)
4758 /* This is a separate function so that we can force an alloca cleanup
4761 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4762 struct re_pattern_buffer
*bufp
;
4763 re_char
*string1
, *string2
;
4766 struct re_registers
*regs
;
4769 /* General temporaries. */
4774 /* Just past the end of the corresponding string. */
4775 re_char
*end1
, *end2
;
4777 /* Pointers into string1 and string2, just past the last characters in
4778 each to consider matching. */
4779 re_char
*end_match_1
, *end_match_2
;
4781 /* Where we are in the data, and the end of the current string. */
4784 /* Used sometimes to remember where we were before starting matching
4785 an operator so that we can go back in case of failure. This "atomic"
4786 behavior of matching opcodes is indispensable to the correctness
4787 of the on_failure_keep_string_jump optimization. */
4790 /* Where we are in the pattern, and the end of the pattern. */
4791 re_char
*p
= bufp
->buffer
;
4792 re_char
*pend
= p
+ bufp
->used
;
4794 /* We use this to map every character in the string. */
4795 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4797 /* Nonzero if we have to concern multibyte character. */
4798 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4800 /* Failure point stack. Each place that can handle a failure further
4801 down the line pushes a failure point on this stack. It consists of
4802 regstart, and regend for all registers corresponding to
4803 the subexpressions we're currently inside, plus the number of such
4804 registers, and, finally, two char *'s. The first char * is where
4805 to resume scanning the pattern; the second one is where to resume
4806 scanning the strings. */
4807 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4808 fail_stack_type fail_stack
;
4811 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4814 #if defined REL_ALLOC && defined REGEX_MALLOC
4815 /* This holds the pointer to the failure stack, when
4816 it is allocated relocatably. */
4817 fail_stack_elt_t
*failure_stack_ptr
;
4820 /* We fill all the registers internally, independent of what we
4821 return, for use in backreferences. The number here includes
4822 an element for register zero. */
4823 size_t num_regs
= bufp
->re_nsub
+ 1;
4825 /* Information on the contents of registers. These are pointers into
4826 the input strings; they record just what was matched (on this
4827 attempt) by a subexpression part of the pattern, that is, the
4828 regnum-th regstart pointer points to where in the pattern we began
4829 matching and the regnum-th regend points to right after where we
4830 stopped matching the regnum-th subexpression. (The zeroth register
4831 keeps track of what the whole pattern matches.) */
4832 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4833 re_char
**regstart
, **regend
;
4836 /* The following record the register info as found in the above
4837 variables when we find a match better than any we've seen before.
4838 This happens as we backtrack through the failure points, which in
4839 turn happens only if we have not yet matched the entire string. */
4840 unsigned best_regs_set
= false;
4841 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4842 re_char
**best_regstart
, **best_regend
;
4845 /* Logically, this is `best_regend[0]'. But we don't want to have to
4846 allocate space for that if we're not allocating space for anything
4847 else (see below). Also, we never need info about register 0 for
4848 any of the other register vectors, and it seems rather a kludge to
4849 treat `best_regend' differently than the rest. So we keep track of
4850 the end of the best match so far in a separate variable. We
4851 initialize this to NULL so that when we backtrack the first time
4852 and need to test it, it's not garbage. */
4853 re_char
*match_end
= NULL
;
4856 /* Counts the total number of registers pushed. */
4857 unsigned num_regs_pushed
= 0;
4860 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4864 #ifdef MATCH_MAY_ALLOCATE
4865 /* Do not bother to initialize all the register variables if there are
4866 no groups in the pattern, as it takes a fair amount of time. If
4867 there are groups, we include space for register 0 (the whole
4868 pattern), even though we never use it, since it simplifies the
4869 array indexing. We should fix this. */
4872 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
4873 regend
= REGEX_TALLOC (num_regs
, re_char
*);
4874 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
4875 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
4877 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
4885 /* We must initialize all our variables to NULL, so that
4886 `FREE_VARIABLES' doesn't try to free them. */
4887 regstart
= regend
= best_regstart
= best_regend
= NULL
;
4889 #endif /* MATCH_MAY_ALLOCATE */
4891 /* The starting position is bogus. */
4892 if (pos
< 0 || pos
> size1
+ size2
)
4898 /* Initialize subexpression text positions to -1 to mark ones that no
4899 start_memory/stop_memory has been seen for. Also initialize the
4900 register information struct. */
4901 for (reg
= 1; reg
< num_regs
; reg
++)
4902 regstart
[reg
] = regend
[reg
] = NULL
;
4904 /* We move `string1' into `string2' if the latter's empty -- but not if
4905 `string1' is null. */
4906 if (size2
== 0 && string1
!= NULL
)
4913 end1
= string1
+ size1
;
4914 end2
= string2
+ size2
;
4916 /* `p' scans through the pattern as `d' scans through the data.
4917 `dend' is the end of the input string that `d' points within. `d'
4918 is advanced into the following input string whenever necessary, but
4919 this happens before fetching; therefore, at the beginning of the
4920 loop, `d' can be pointing at the end of a string, but it cannot
4924 /* Only match within string2. */
4925 d
= string2
+ pos
- size1
;
4926 dend
= end_match_2
= string2
+ stop
- size1
;
4927 end_match_1
= end1
; /* Just to give it a value. */
4933 /* Only match within string1. */
4934 end_match_1
= string1
+ stop
;
4936 When we reach end_match_1, PREFETCH normally switches to string2.
4937 But in the present case, this means that just doing a PREFETCH
4938 makes us jump from `stop' to `gap' within the string.
4939 What we really want here is for the search to stop as
4940 soon as we hit end_match_1. That's why we set end_match_2
4941 to end_match_1 (since PREFETCH fails as soon as we hit
4943 end_match_2
= end_match_1
;
4946 { /* It's important to use this code when stop == size so that
4947 moving `d' from end1 to string2 will not prevent the d == dend
4948 check from catching the end of string. */
4950 end_match_2
= string2
+ stop
- size1
;
4956 DEBUG_PRINT1 ("The compiled pattern is: ");
4957 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
4958 DEBUG_PRINT1 ("The string to match is: `");
4959 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
4960 DEBUG_PRINT1 ("'\n");
4962 /* This loops over pattern commands. It exits by returning from the
4963 function if the match is complete, or it drops through if the match
4964 fails at this starting point in the input data. */
4967 DEBUG_PRINT2 ("\n%p: ", p
);
4970 { /* End of pattern means we might have succeeded. */
4971 DEBUG_PRINT1 ("end of pattern ... ");
4973 /* If we haven't matched the entire string, and we want the
4974 longest match, try backtracking. */
4975 if (d
!= end_match_2
)
4977 /* 1 if this match ends in the same string (string1 or string2)
4978 as the best previous match. */
4979 boolean same_str_p
= (FIRST_STRING_P (match_end
)
4980 == FIRST_STRING_P (d
));
4981 /* 1 if this match is the best seen so far. */
4982 boolean best_match_p
;
4984 /* AIX compiler got confused when this was combined
4985 with the previous declaration. */
4987 best_match_p
= d
> match_end
;
4989 best_match_p
= !FIRST_STRING_P (d
);
4991 DEBUG_PRINT1 ("backtracking.\n");
4993 if (!FAIL_STACK_EMPTY ())
4994 { /* More failure points to try. */
4996 /* If exceeds best match so far, save it. */
4997 if (!best_regs_set
|| best_match_p
)
4999 best_regs_set
= true;
5002 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5004 for (reg
= 1; reg
< num_regs
; reg
++)
5006 best_regstart
[reg
] = regstart
[reg
];
5007 best_regend
[reg
] = regend
[reg
];
5013 /* If no failure points, don't restore garbage. And if
5014 last match is real best match, don't restore second
5016 else if (best_regs_set
&& !best_match_p
)
5019 /* Restore best match. It may happen that `dend ==
5020 end_match_1' while the restored d is in string2.
5021 For example, the pattern `x.*y.*z' against the
5022 strings `x-' and `y-z-', if the two strings are
5023 not consecutive in memory. */
5024 DEBUG_PRINT1 ("Restoring best registers.\n");
5027 dend
= ((d
>= string1
&& d
<= end1
)
5028 ? end_match_1
: end_match_2
);
5030 for (reg
= 1; reg
< num_regs
; reg
++)
5032 regstart
[reg
] = best_regstart
[reg
];
5033 regend
[reg
] = best_regend
[reg
];
5036 } /* d != end_match_2 */
5039 DEBUG_PRINT1 ("Accepting match.\n");
5041 /* If caller wants register contents data back, do it. */
5042 if (regs
&& !bufp
->no_sub
)
5044 /* Have the register data arrays been allocated? */
5045 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5046 { /* No. So allocate them with malloc. We need one
5047 extra element beyond `num_regs' for the `-1' marker
5049 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5050 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5051 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5052 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5057 bufp
->regs_allocated
= REGS_REALLOCATE
;
5059 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5060 { /* Yes. If we need more elements than were already
5061 allocated, reallocate them. If we need fewer, just
5063 if (regs
->num_regs
< num_regs
+ 1)
5065 regs
->num_regs
= num_regs
+ 1;
5066 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5067 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5068 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5077 /* These braces fend off a "empty body in an else-statement"
5078 warning under GCC when assert expands to nothing. */
5079 assert (bufp
->regs_allocated
== REGS_FIXED
);
5082 /* Convert the pointer data in `regstart' and `regend' to
5083 indices. Register zero has to be set differently,
5084 since we haven't kept track of any info for it. */
5085 if (regs
->num_regs
> 0)
5087 regs
->start
[0] = pos
;
5088 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5091 /* Go through the first `min (num_regs, regs->num_regs)'
5092 registers, since that is all we initialized. */
5093 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5095 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5096 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5100 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5102 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5106 /* If the regs structure we return has more elements than
5107 were in the pattern, set the extra elements to -1. If
5108 we (re)allocated the registers, this is the case,
5109 because we always allocate enough to have at least one
5111 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5112 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5113 } /* regs && !bufp->no_sub */
5115 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5116 nfailure_points_pushed
, nfailure_points_popped
,
5117 nfailure_points_pushed
- nfailure_points_popped
);
5118 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5120 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5122 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5128 /* Otherwise match next pattern command. */
5129 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5131 /* Ignore these. Used to ignore the n of succeed_n's which
5132 currently have n == 0. */
5134 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5138 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5141 /* Match the next n pattern characters exactly. The following
5142 byte in the pattern defines n, and the n bytes after that
5143 are the characters to match. */
5146 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5148 /* Remember the start point to rollback upon failure. */
5151 /* This is written out as an if-else so we don't waste time
5152 testing `translate' inside the loop. */
5153 if (RE_TRANSLATE_P (translate
))
5158 int pat_charlen
, buf_charlen
;
5159 unsigned int pat_ch
, buf_ch
;
5162 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
5163 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5165 if (RE_TRANSLATE (translate
, buf_ch
)
5174 mcnt
-= pat_charlen
;
5181 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5206 /* Match any character except possibly a newline or a null. */
5212 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5215 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5216 buf_ch
= TRANSLATE (buf_ch
);
5218 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5220 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5221 && buf_ch
== '\000'))
5224 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5233 register unsigned int c
;
5234 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5237 /* Start of actual range_table, or end of bitmap if there is no
5239 re_char
*range_table
;
5241 /* Nonzero if there is a range table. */
5242 int range_table_exists
;
5244 /* Number of ranges of range table. This is not included
5245 in the initial byte-length of the command. */
5248 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5250 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5252 if (range_table_exists
)
5254 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5255 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5259 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5260 c
= TRANSLATE (c
); /* The character to match. */
5262 if (SINGLE_BYTE_CHAR_P (c
))
5263 { /* Lookup bitmap. */
5264 /* Cast to `unsigned' instead of `unsigned char' in
5265 case the bit list is a full 32 bytes long. */
5266 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5267 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5271 else if (range_table_exists
)
5273 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5275 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5276 | (class_bits
& BIT_MULTIBYTE
)
5277 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5278 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5279 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5280 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5283 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5287 if (range_table_exists
)
5288 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5290 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5292 if (!not) goto fail
;
5299 /* The beginning of a group is represented by start_memory.
5300 The argument is the register number. The text
5301 matched within the group is recorded (in the internal
5302 registers data structure) under the register number. */
5304 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5306 /* In case we need to undo this operation (via backtracking). */
5307 PUSH_FAILURE_REG ((unsigned int)*p
);
5310 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5311 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5313 /* Move past the register number and inner group count. */
5318 /* The stop_memory opcode represents the end of a group. Its
5319 argument is the same as start_memory's: the register number. */
5321 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5323 assert (!REG_UNSET (regstart
[*p
]));
5324 /* Strictly speaking, there should be code such as:
5326 assert (REG_UNSET (regend[*p]));
5327 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5329 But the only info to be pushed is regend[*p] and it is known to
5330 be UNSET, so there really isn't anything to push.
5331 Not pushing anything, on the other hand deprives us from the
5332 guarantee that regend[*p] is UNSET since undoing this operation
5333 will not reset its value properly. This is not important since
5334 the value will only be read on the next start_memory or at
5335 the very end and both events can only happen if this stop_memory
5339 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5341 /* Move past the register number and the inner group count. */
5346 /* \<digit> has been turned into a `duplicate' command which is
5347 followed by the numeric value of <digit> as the register number. */
5350 register re_char
*d2
, *dend2
;
5351 int regno
= *p
++; /* Get which register to match against. */
5352 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5354 /* Can't back reference a group which we've never matched. */
5355 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5358 /* Where in input to try to start matching. */
5359 d2
= regstart
[regno
];
5361 /* Remember the start point to rollback upon failure. */
5364 /* Where to stop matching; if both the place to start and
5365 the place to stop matching are in the same string, then
5366 set to the place to stop, otherwise, for now have to use
5367 the end of the first string. */
5369 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5370 == FIRST_STRING_P (regend
[regno
]))
5371 ? regend
[regno
] : end_match_1
);
5374 /* If necessary, advance to next segment in register
5378 if (dend2
== end_match_2
) break;
5379 if (dend2
== regend
[regno
]) break;
5381 /* End of string1 => advance to string2. */
5383 dend2
= regend
[regno
];
5385 /* At end of register contents => success */
5386 if (d2
== dend2
) break;
5388 /* If necessary, advance to next segment in data. */
5391 /* How many characters left in this segment to match. */
5394 /* Want how many consecutive characters we can match in
5395 one shot, so, if necessary, adjust the count. */
5396 if (mcnt
> dend2
- d2
)
5399 /* Compare that many; failure if mismatch, else move
5401 if (RE_TRANSLATE_P (translate
)
5402 ? bcmp_translate (d
, d2
, mcnt
, translate
, multibyte
)
5403 : memcmp (d
, d2
, mcnt
))
5408 d
+= mcnt
, d2
+= mcnt
;
5414 /* begline matches the empty string at the beginning of the string
5415 (unless `not_bol' is set in `bufp'), and after newlines. */
5417 DEBUG_PRINT1 ("EXECUTING begline.\n");
5419 if (AT_STRINGS_BEG (d
))
5421 if (!bufp
->not_bol
) break;
5426 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5430 /* In all other cases, we fail. */
5434 /* endline is the dual of begline. */
5436 DEBUG_PRINT1 ("EXECUTING endline.\n");
5438 if (AT_STRINGS_END (d
))
5440 if (!bufp
->not_eol
) break;
5444 PREFETCH_NOLIMIT ();
5451 /* Match at the very beginning of the data. */
5453 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5454 if (AT_STRINGS_BEG (d
))
5459 /* Match at the very end of the data. */
5461 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5462 if (AT_STRINGS_END (d
))
5467 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5468 pushes NULL as the value for the string on the stack. Then
5469 `POP_FAILURE_POINT' will keep the current value for the
5470 string, instead of restoring it. To see why, consider
5471 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5472 then the . fails against the \n. But the next thing we want
5473 to do is match the \n against the \n; if we restored the
5474 string value, we would be back at the foo.
5476 Because this is used only in specific cases, we don't need to
5477 check all the things that `on_failure_jump' does, to make
5478 sure the right things get saved on the stack. Hence we don't
5479 share its code. The only reason to push anything on the
5480 stack at all is that otherwise we would have to change
5481 `anychar's code to do something besides goto fail in this
5482 case; that seems worse than this. */
5483 case on_failure_keep_string_jump
:
5484 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5485 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5488 PUSH_FAILURE_POINT (p
- 3, NULL
);
5491 /* A nasty loop is introduced by the non-greedy *? and +?.
5492 With such loops, the stack only ever contains one failure point
5493 at a time, so that a plain on_failure_jump_loop kind of
5494 cycle detection cannot work. Worse yet, such a detection
5495 can not only fail to detect a cycle, but it can also wrongly
5496 detect a cycle (between different instantiations of the same
5498 So the method used for those nasty loops is a little different:
5499 We use a special cycle-detection-stack-frame which is pushed
5500 when the on_failure_jump_nastyloop failure-point is *popped*.
5501 This special frame thus marks the beginning of one iteration
5502 through the loop and we can hence easily check right here
5503 whether something matched between the beginning and the end of
5505 case on_failure_jump_nastyloop
:
5506 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5507 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5510 assert ((re_opcode_t
)p
[-4] == no_op
);
5513 CHECK_INFINITE_LOOP (p
- 4, d
);
5515 /* If there's a cycle, just continue without pushing
5516 this failure point. The failure point is the "try again"
5517 option, which shouldn't be tried.
5518 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5519 PUSH_FAILURE_POINT (p
- 3, d
);
5523 /* Simple loop detecting on_failure_jump: just check on the
5524 failure stack if the same spot was already hit earlier. */
5525 case on_failure_jump_loop
:
5527 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5528 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5532 CHECK_INFINITE_LOOP (p
- 3, d
);
5534 /* If there's a cycle, get out of the loop, as if the matching
5535 had failed. We used to just `goto fail' here, but that was
5536 aborting the search a bit too early: we want to keep the
5537 empty-loop-match and keep matching after the loop.
5538 We want (x?)*y\1z to match both xxyz and xxyxz. */
5541 PUSH_FAILURE_POINT (p
- 3, d
);
5546 /* Uses of on_failure_jump:
5548 Each alternative starts with an on_failure_jump that points
5549 to the beginning of the next alternative. Each alternative
5550 except the last ends with a jump that in effect jumps past
5551 the rest of the alternatives. (They really jump to the
5552 ending jump of the following alternative, because tensioning
5553 these jumps is a hassle.)
5555 Repeats start with an on_failure_jump that points past both
5556 the repetition text and either the following jump or
5557 pop_failure_jump back to this on_failure_jump. */
5558 case on_failure_jump
:
5559 IMMEDIATE_QUIT_CHECK
;
5560 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5561 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5564 PUSH_FAILURE_POINT (p
-3, d
);
5567 /* This operation is used for greedy *.
5568 Compare the beginning of the repeat with what in the
5569 pattern follows its end. If we can establish that there
5570 is nothing that they would both match, i.e., that we
5571 would have to backtrack because of (as in, e.g., `a*a')
5572 then we can use a non-backtracking loop based on
5573 on_failure_keep_string_jump instead of on_failure_jump. */
5574 case on_failure_jump_smart
:
5575 IMMEDIATE_QUIT_CHECK
;
5576 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5577 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5580 re_char
*p1
= p
; /* Next operation. */
5581 /* Here, we discard `const', making re_match non-reentrant. */
5582 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5583 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5585 p
-= 3; /* Reset so that we will re-execute the
5586 instruction once it's been changed. */
5588 EXTRACT_NUMBER (mcnt
, p2
- 2);
5590 /* Ensure this is a indeed the trivial kind of loop
5591 we are expecting. */
5592 assert (skip_one_char (p1
) == p2
- 3);
5593 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5594 DEBUG_STATEMENT (debug
+= 2);
5595 if (mutually_exclusive_p (bufp
, p1
, p2
))
5597 /* Use a fast `on_failure_keep_string_jump' loop. */
5598 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5599 *p3
= (unsigned char) on_failure_keep_string_jump
;
5600 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5604 /* Default to a safe `on_failure_jump' loop. */
5605 DEBUG_PRINT1 (" smart default => slow loop.\n");
5606 *p3
= (unsigned char) on_failure_jump
;
5608 DEBUG_STATEMENT (debug
-= 2);
5612 /* Unconditionally jump (without popping any failure points). */
5615 IMMEDIATE_QUIT_CHECK
;
5616 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5617 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5618 p
+= mcnt
; /* Do the jump. */
5619 DEBUG_PRINT2 ("(to %p).\n", p
);
5623 /* Have to succeed matching what follows at least n times.
5624 After that, handle like `on_failure_jump'. */
5626 /* Signedness doesn't matter since we only compare MCNT to 0. */
5627 EXTRACT_NUMBER (mcnt
, p
+ 2);
5628 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5630 /* Originally, mcnt is how many times we HAVE to succeed. */
5633 /* Here, we discard `const', making re_match non-reentrant. */
5634 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5637 PUSH_NUMBER (p2
, mcnt
);
5640 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5645 /* Signedness doesn't matter since we only compare MCNT to 0. */
5646 EXTRACT_NUMBER (mcnt
, p
+ 2);
5647 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5649 /* Originally, this is how many times we CAN jump. */
5652 /* Here, we discard `const', making re_match non-reentrant. */
5653 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5655 PUSH_NUMBER (p2
, mcnt
);
5656 goto unconditional_jump
;
5658 /* If don't have to jump any more, skip over the rest of command. */
5665 unsigned char *p2
; /* Location of the counter. */
5666 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5668 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5669 /* Here, we discard `const', making re_match non-reentrant. */
5670 p2
= (unsigned char*) p
+ mcnt
;
5671 /* Signedness doesn't matter since we only copy MCNT's bits . */
5672 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5673 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
5674 PUSH_NUMBER (p2
, mcnt
);
5680 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5681 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5683 /* We SUCCEED (or FAIL) in one of the following cases: */
5685 /* Case 1: D is at the beginning or the end of string. */
5686 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5690 /* C1 is the character before D, S1 is the syntax of C1, C2
5691 is the character at D, and S2 is the syntax of C2. */
5695 int offset
= PTR_TO_OFFSET (d
- 1);
5696 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5697 UPDATE_SYNTAX_TABLE (charpos
);
5699 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5702 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5704 PREFETCH_NOLIMIT ();
5705 c2
= RE_STRING_CHAR (d
, dend
- d
);
5708 if (/* Case 2: Only one of S1 and S2 is Sword. */
5709 ((s1
== Sword
) != (s2
== Sword
))
5710 /* Case 3: Both of S1 and S2 are Sword, and macro
5711 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5712 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5721 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5723 /* We FAIL in one of the following cases: */
5725 /* Case 1: D is at the end of string. */
5726 if (AT_STRINGS_END (d
))
5730 /* C1 is the character before D, S1 is the syntax of C1, C2
5731 is the character at D, and S2 is the syntax of C2. */
5735 int offset
= PTR_TO_OFFSET (d
);
5736 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5737 UPDATE_SYNTAX_TABLE (charpos
);
5740 c2
= RE_STRING_CHAR (d
, dend
- d
);
5743 /* Case 2: S2 is not Sword. */
5747 /* Case 3: D is not at the beginning of string ... */
5748 if (!AT_STRINGS_BEG (d
))
5750 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5752 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5756 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5758 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5765 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5767 /* We FAIL in one of the following cases: */
5769 /* Case 1: D is at the beginning of string. */
5770 if (AT_STRINGS_BEG (d
))
5774 /* C1 is the character before D, S1 is the syntax of C1, C2
5775 is the character at D, and S2 is the syntax of C2. */
5779 int offset
= PTR_TO_OFFSET (d
) - 1;
5780 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5781 UPDATE_SYNTAX_TABLE (charpos
);
5783 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5786 /* Case 2: S1 is not Sword. */
5790 /* Case 3: D is not at the end of string ... */
5791 if (!AT_STRINGS_END (d
))
5793 PREFETCH_NOLIMIT ();
5794 c2
= RE_STRING_CHAR (d
, dend
- d
);
5796 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
5800 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5802 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5809 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
5811 /* We FAIL in one of the following cases: */
5813 /* Case 1: D is at the end of string. */
5814 if (AT_STRINGS_END (d
))
5818 /* C1 is the character before D, S1 is the syntax of C1, C2
5819 is the character at D, and S2 is the syntax of C2. */
5823 int offset
= PTR_TO_OFFSET (d
);
5824 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5825 UPDATE_SYNTAX_TABLE (charpos
);
5828 c2
= RE_STRING_CHAR (d
, dend
- d
);
5831 /* Case 2: S2 is neither Sword nor Ssymbol. */
5832 if (s2
!= Sword
&& s2
!= Ssymbol
)
5835 /* Case 3: D is not at the beginning of string ... */
5836 if (!AT_STRINGS_BEG (d
))
5838 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5840 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5844 /* ... and S1 is Sword or Ssymbol. */
5845 if (s1
== Sword
|| s1
== Ssymbol
)
5852 DEBUG_PRINT1 ("EXECUTING symend.\n");
5854 /* We FAIL in one of the following cases: */
5856 /* Case 1: D is at the beginning of string. */
5857 if (AT_STRINGS_BEG (d
))
5861 /* C1 is the character before D, S1 is the syntax of C1, C2
5862 is the character at D, and S2 is the syntax of C2. */
5866 int offset
= PTR_TO_OFFSET (d
) - 1;
5867 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5868 UPDATE_SYNTAX_TABLE (charpos
);
5870 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5873 /* Case 2: S1 is neither Ssymbol nor Sword. */
5874 if (s1
!= Sword
&& s1
!= Ssymbol
)
5877 /* Case 3: D is not at the end of string ... */
5878 if (!AT_STRINGS_END (d
))
5880 PREFETCH_NOLIMIT ();
5881 c2
= RE_STRING_CHAR (d
, dend
- d
);
5883 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
5887 /* ... and S2 is Sword or Ssymbol. */
5888 if (s2
== Sword
|| s2
== Ssymbol
)
5896 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
5898 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
5902 int offset
= PTR_TO_OFFSET (d
);
5903 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5904 UPDATE_SYNTAX_TABLE (pos1
);
5911 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5913 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
5921 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5922 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
5927 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5928 if (PTR_BYTE_POS (d
) != PT_BYTE
)
5933 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5934 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
5939 case notcategoryspec
:
5940 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
5942 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
5948 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5950 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
5961 continue; /* Successfully executed one pattern command; keep going. */
5964 /* We goto here if a matching operation fails. */
5966 IMMEDIATE_QUIT_CHECK
;
5967 if (!FAIL_STACK_EMPTY ())
5970 /* A restart point is known. Restore to that state. */
5971 DEBUG_PRINT1 ("\nFAIL:\n");
5972 POP_FAILURE_POINT (str
, pat
);
5973 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
5975 case on_failure_keep_string_jump
:
5976 assert (str
== NULL
);
5977 goto continue_failure_jump
;
5979 case on_failure_jump_nastyloop
:
5980 assert ((re_opcode_t
)pat
[-2] == no_op
);
5981 PUSH_FAILURE_POINT (pat
- 2, str
);
5984 case on_failure_jump_loop
:
5985 case on_failure_jump
:
5988 continue_failure_jump
:
5989 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
5994 /* A special frame used for nastyloops. */
6001 assert (p
>= bufp
->buffer
&& p
<= pend
);
6003 if (d
>= string1
&& d
<= end1
)
6007 break; /* Matching at this starting point really fails. */
6011 goto restore_best_regs
;
6015 return -1; /* Failure to match. */
6018 /* Subroutine definitions for re_match_2. */
6020 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6021 bytes; nonzero otherwise. */
6024 bcmp_translate (s1
, s2
, len
, translate
, multibyte
)
6027 RE_TRANSLATE_TYPE translate
;
6028 const int multibyte
;
6030 register re_char
*p1
= s1
, *p2
= s2
;
6031 re_char
*p1_end
= s1
+ len
;
6032 re_char
*p2_end
= s2
+ len
;
6034 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6035 different lengths, but relying on a single `len' would break this. -sm */
6036 while (p1
< p1_end
&& p2
< p2_end
)
6038 int p1_charlen
, p2_charlen
;
6039 re_wchar_t p1_ch
, p2_ch
;
6041 p1_ch
= RE_STRING_CHAR_AND_LENGTH (p1
, p1_end
- p1
, p1_charlen
);
6042 p2_ch
= RE_STRING_CHAR_AND_LENGTH (p2
, p2_end
- p2
, p2_charlen
);
6044 if (RE_TRANSLATE (translate
, p1_ch
)
6045 != RE_TRANSLATE (translate
, p2_ch
))
6048 p1
+= p1_charlen
, p2
+= p2_charlen
;
6051 if (p1
!= p1_end
|| p2
!= p2_end
)
6057 /* Entry points for GNU code. */
6059 /* re_compile_pattern is the GNU regular expression compiler: it
6060 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6061 Returns 0 if the pattern was valid, otherwise an error string.
6063 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6064 are set in BUFP on entry.
6066 We call regex_compile to do the actual compilation. */
6069 re_compile_pattern (pattern
, length
, bufp
)
6070 const char *pattern
;
6072 struct re_pattern_buffer
*bufp
;
6076 /* GNU code is written to assume at least RE_NREGS registers will be set
6077 (and at least one extra will be -1). */
6078 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6080 /* And GNU code determines whether or not to get register information
6081 by passing null for the REGS argument to re_match, etc., not by
6085 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6089 return gettext (re_error_msgid
[(int) ret
]);
6091 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6093 /* Entry points compatible with 4.2 BSD regex library. We don't define
6094 them unless specifically requested. */
6096 #if defined _REGEX_RE_COMP || defined _LIBC
6098 /* BSD has one and only one pattern buffer. */
6099 static struct re_pattern_buffer re_comp_buf
;
6103 /* Make these definitions weak in libc, so POSIX programs can redefine
6104 these names if they don't use our functions, and still use
6105 regcomp/regexec below without link errors. */
6115 if (!re_comp_buf
.buffer
)
6116 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6117 return (char *) gettext ("No previous regular expression");
6121 if (!re_comp_buf
.buffer
)
6123 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6124 if (re_comp_buf
.buffer
== NULL
)
6125 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6126 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6127 re_comp_buf
.allocated
= 200;
6129 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6130 if (re_comp_buf
.fastmap
== NULL
)
6131 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6132 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6135 /* Since `re_exec' always passes NULL for the `regs' argument, we
6136 don't need to initialize the pattern buffer fields which affect it. */
6138 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6143 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6144 return (char *) gettext (re_error_msgid
[(int) ret
]);
6155 const int len
= strlen (s
);
6157 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6159 #endif /* _REGEX_RE_COMP */
6161 /* POSIX.2 functions. Don't define these for Emacs. */
6165 /* regcomp takes a regular expression as a string and compiles it.
6167 PREG is a regex_t *. We do not expect any fields to be initialized,
6168 since POSIX says we shouldn't. Thus, we set
6170 `buffer' to the compiled pattern;
6171 `used' to the length of the compiled pattern;
6172 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6173 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6174 RE_SYNTAX_POSIX_BASIC;
6175 `fastmap' to an allocated space for the fastmap;
6176 `fastmap_accurate' to zero;
6177 `re_nsub' to the number of subexpressions in PATTERN.
6179 PATTERN is the address of the pattern string.
6181 CFLAGS is a series of bits which affect compilation.
6183 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6184 use POSIX basic syntax.
6186 If REG_NEWLINE is set, then . and [^...] don't match newline.
6187 Also, regexec will try a match beginning after every newline.
6189 If REG_ICASE is set, then we considers upper- and lowercase
6190 versions of letters to be equivalent when matching.
6192 If REG_NOSUB is set, then when PREG is passed to regexec, that
6193 routine will report only success or failure, and nothing about the
6196 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6197 the return codes and their meanings.) */
6200 regcomp (preg
, pattern
, cflags
)
6201 regex_t
*__restrict preg
;
6202 const char *__restrict pattern
;
6207 = (cflags
& REG_EXTENDED
) ?
6208 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6210 /* regex_compile will allocate the space for the compiled pattern. */
6212 preg
->allocated
= 0;
6215 /* Try to allocate space for the fastmap. */
6216 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6218 if (cflags
& REG_ICASE
)
6223 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6224 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6225 if (preg
->translate
== NULL
)
6226 return (int) REG_ESPACE
;
6228 /* Map uppercase characters to corresponding lowercase ones. */
6229 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6230 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6233 preg
->translate
= NULL
;
6235 /* If REG_NEWLINE is set, newlines are treated differently. */
6236 if (cflags
& REG_NEWLINE
)
6237 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6238 syntax
&= ~RE_DOT_NEWLINE
;
6239 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6242 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6244 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6246 /* POSIX says a null character in the pattern terminates it, so we
6247 can use strlen here in compiling the pattern. */
6248 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6250 /* POSIX doesn't distinguish between an unmatched open-group and an
6251 unmatched close-group: both are REG_EPAREN. */
6252 if (ret
== REG_ERPAREN
)
6255 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6256 { /* Compute the fastmap now, since regexec cannot modify the pattern
6258 re_compile_fastmap (preg
);
6259 if (preg
->can_be_null
)
6260 { /* The fastmap can't be used anyway. */
6261 free (preg
->fastmap
);
6262 preg
->fastmap
= NULL
;
6267 WEAK_ALIAS (__regcomp
, regcomp
)
6270 /* regexec searches for a given pattern, specified by PREG, in the
6273 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6274 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6275 least NMATCH elements, and we set them to the offsets of the
6276 corresponding matched substrings.
6278 EFLAGS specifies `execution flags' which affect matching: if
6279 REG_NOTBOL is set, then ^ does not match at the beginning of the
6280 string; if REG_NOTEOL is set, then $ does not match at the end.
6282 We return 0 if we find a match and REG_NOMATCH if not. */
6285 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6286 const regex_t
*__restrict preg
;
6287 const char *__restrict string
;
6289 regmatch_t pmatch
[__restrict_arr
];
6293 struct re_registers regs
;
6294 regex_t private_preg
;
6295 int len
= strlen (string
);
6296 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6298 private_preg
= *preg
;
6300 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6301 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6303 /* The user has told us exactly how many registers to return
6304 information about, via `nmatch'. We have to pass that on to the
6305 matching routines. */
6306 private_preg
.regs_allocated
= REGS_FIXED
;
6310 regs
.num_regs
= nmatch
;
6311 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6312 if (regs
.start
== NULL
)
6313 return (int) REG_NOMATCH
;
6314 regs
.end
= regs
.start
+ nmatch
;
6317 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6318 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6319 was a little bit longer but still only matching the real part.
6320 This works because the `endline' will check for a '\n' and will find a
6321 '\0', correctly deciding that this is not the end of a line.
6322 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6323 a convenient '\0' there. For all we know, the string could be preceded
6324 by '\n' which would throw things off. */
6326 /* Perform the searching operation. */
6327 ret
= re_search (&private_preg
, string
, len
,
6328 /* start: */ 0, /* range: */ len
,
6329 want_reg_info
? ®s
: (struct re_registers
*) 0);
6331 /* Copy the register information to the POSIX structure. */
6338 for (r
= 0; r
< nmatch
; r
++)
6340 pmatch
[r
].rm_so
= regs
.start
[r
];
6341 pmatch
[r
].rm_eo
= regs
.end
[r
];
6345 /* If we needed the temporary register info, free the space now. */
6349 /* We want zero return to mean success, unlike `re_search'. */
6350 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6352 WEAK_ALIAS (__regexec
, regexec
)
6355 /* Returns a message corresponding to an error code, ERRCODE, returned
6356 from either regcomp or regexec. We don't use PREG here. */
6359 regerror (errcode
, preg
, errbuf
, errbuf_size
)
6361 const regex_t
*preg
;
6369 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6370 /* Only error codes returned by the rest of the code should be passed
6371 to this routine. If we are given anything else, or if other regex
6372 code generates an invalid error code, then the program has a bug.
6373 Dump core so we can fix it. */
6376 msg
= gettext (re_error_msgid
[errcode
]);
6378 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6380 if (errbuf_size
!= 0)
6382 if (msg_size
> errbuf_size
)
6384 strncpy (errbuf
, msg
, errbuf_size
- 1);
6385 errbuf
[errbuf_size
- 1] = 0;
6388 strcpy (errbuf
, msg
);
6393 WEAK_ALIAS (__regerror
, regerror
)
6396 /* Free dynamically allocated space used by PREG. */
6402 if (preg
->buffer
!= NULL
)
6403 free (preg
->buffer
);
6404 preg
->buffer
= NULL
;
6406 preg
->allocated
= 0;
6409 if (preg
->fastmap
!= NULL
)
6410 free (preg
->fastmap
);
6411 preg
->fastmap
= NULL
;
6412 preg
->fastmap_accurate
= 0;
6414 if (preg
->translate
!= NULL
)
6415 free (preg
->translate
);
6416 preg
->translate
= NULL
;
6418 WEAK_ALIAS (__regfree
, regfree
)
6420 #endif /* not emacs */
6422 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6423 (do not change this comment) */