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 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 };
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
['_'] = Sword
;
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 /* Matches any character whose syntax is specified. Followed by
659 a byte which contains a syntax code, e.g., Sword. */
662 /* Matches any character whose syntax is not that specified. */
666 ,before_dot
, /* Succeeds if before point. */
667 at_dot
, /* Succeeds if at point. */
668 after_dot
, /* Succeeds if after point. */
670 /* Matches any character whose category-set contains the specified
671 category. The operator is followed by a byte which contains a
672 category code (mnemonic ASCII character). */
675 /* Matches any character whose category-set does not contain the
676 specified category. The operator is followed by a byte which
677 contains the category code (mnemonic ASCII character). */
682 /* Common operations on the compiled pattern. */
684 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
686 #define STORE_NUMBER(destination, number) \
688 (destination)[0] = (number) & 0377; \
689 (destination)[1] = (number) >> 8; \
692 /* Same as STORE_NUMBER, except increment DESTINATION to
693 the byte after where the number is stored. Therefore, DESTINATION
694 must be an lvalue. */
696 #define STORE_NUMBER_AND_INCR(destination, number) \
698 STORE_NUMBER (destination, number); \
699 (destination) += 2; \
702 /* Put into DESTINATION a number stored in two contiguous bytes starting
705 #define EXTRACT_NUMBER(destination, source) \
707 (destination) = *(source) & 0377; \
708 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
712 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
714 extract_number (dest
, source
)
718 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
719 *dest
= *source
& 0377;
723 # ifndef EXTRACT_MACROS /* To debug the macros. */
724 # undef EXTRACT_NUMBER
725 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
726 # endif /* not EXTRACT_MACROS */
730 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
731 SOURCE must be an lvalue. */
733 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
735 EXTRACT_NUMBER (destination, source); \
740 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
743 extract_number_and_incr (destination
, source
)
747 extract_number (destination
, *source
);
751 # ifndef EXTRACT_MACROS
752 # undef EXTRACT_NUMBER_AND_INCR
753 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
754 extract_number_and_incr (&dest, &src)
755 # endif /* not EXTRACT_MACROS */
759 /* Store a multibyte character in three contiguous bytes starting
760 DESTINATION, and increment DESTINATION to the byte after where the
761 character is stored. Therefore, DESTINATION must be an lvalue. */
763 #define STORE_CHARACTER_AND_INCR(destination, character) \
765 (destination)[0] = (character) & 0377; \
766 (destination)[1] = ((character) >> 8) & 0377; \
767 (destination)[2] = (character) >> 16; \
768 (destination) += 3; \
771 /* Put into DESTINATION a character stored in three contiguous bytes
772 starting at SOURCE. */
774 #define EXTRACT_CHARACTER(destination, source) \
776 (destination) = ((source)[0] \
777 | ((source)[1] << 8) \
778 | ((source)[2] << 16)); \
782 /* Macros for charset. */
784 /* Size of bitmap of charset P in bytes. P is a start of charset,
785 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
786 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
788 /* Nonzero if charset P has range table. */
789 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
791 /* Return the address of range table of charset P. But not the start
792 of table itself, but the before where the number of ranges is
793 stored. `2 +' means to skip re_opcode_t and size of bitmap,
794 and the 2 bytes of flags at the start of the range table. */
795 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
797 /* Extract the bit flags that start a range table. */
798 #define CHARSET_RANGE_TABLE_BITS(p) \
799 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
800 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
802 /* Test if C is listed in the bitmap of charset P. */
803 #define CHARSET_LOOKUP_BITMAP(p, c) \
804 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
805 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
807 /* Return the address of end of RANGE_TABLE. COUNT is number of
808 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
809 is start of range and end of range. `* 3' is size of each start
811 #define CHARSET_RANGE_TABLE_END(range_table, count) \
812 ((range_table) + (count) * 2 * 3)
814 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
815 COUNT is number of ranges in RANGE_TABLE. */
816 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
819 re_wchar_t range_start, range_end; \
821 re_char *range_table_end \
822 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
824 for (p = (range_table); p < range_table_end; p += 2 * 3) \
826 EXTRACT_CHARACTER (range_start, p); \
827 EXTRACT_CHARACTER (range_end, p + 3); \
829 if (range_start <= (c) && (c) <= range_end) \
838 /* Test if C is in range table of CHARSET. The flag NOT is negated if
839 C is listed in it. */
840 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
843 /* Number of ranges in range table. */ \
845 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
847 EXTRACT_NUMBER_AND_INCR (count, range_table); \
848 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
852 /* If DEBUG is defined, Regex prints many voluminous messages about what
853 it is doing (if the variable `debug' is nonzero). If linked with the
854 main program in `iregex.c', you can enter patterns and strings
855 interactively. And if linked with the main program in `main.c' and
856 the other test files, you can run the already-written tests. */
860 /* We use standard I/O for debugging. */
863 /* It is useful to test things that ``must'' be true when debugging. */
866 static int debug
= -100000;
868 # define DEBUG_STATEMENT(e) e
869 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
870 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
871 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
872 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
873 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
874 if (debug > 0) print_partial_compiled_pattern (s, e)
875 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
876 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
879 /* Print the fastmap in human-readable form. */
882 print_fastmap (fastmap
)
885 unsigned was_a_range
= 0;
888 while (i
< (1 << BYTEWIDTH
))
894 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
910 /* Print a compiled pattern string in human-readable form, starting at
911 the START pointer into it and ending just before the pointer END. */
914 print_partial_compiled_pattern (start
, end
)
924 fprintf (stderr
, "(null)\n");
928 /* Loop over pattern commands. */
931 fprintf (stderr
, "%d:\t", p
- start
);
933 switch ((re_opcode_t
) *p
++)
936 fprintf (stderr
, "/no_op");
940 fprintf (stderr
, "/succeed");
945 fprintf (stderr
, "/exactn/%d", mcnt
);
948 fprintf (stderr
, "/%c", *p
++);
954 fprintf (stderr
, "/start_memory/%d", *p
++);
958 fprintf (stderr
, "/stop_memory/%d", *p
++);
962 fprintf (stderr
, "/duplicate/%d", *p
++);
966 fprintf (stderr
, "/anychar");
972 register int c
, last
= -100;
973 register int in_range
= 0;
974 int length
= CHARSET_BITMAP_SIZE (p
- 1);
975 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
977 fprintf (stderr
, "/charset [%s",
978 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
980 assert (p
+ *p
< pend
);
982 for (c
= 0; c
< 256; c
++)
984 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
986 /* Are we starting a range? */
987 if (last
+ 1 == c
&& ! in_range
)
989 fprintf (stderr
, "-");
992 /* Have we broken a range? */
993 else if (last
+ 1 != c
&& in_range
)
995 fprintf (stderr
, "%c", last
);
1000 fprintf (stderr
, "%c", c
);
1006 fprintf (stderr
, "%c", last
);
1008 fprintf (stderr
, "]");
1012 if (has_range_table
)
1015 fprintf (stderr
, "has-range-table");
1017 /* ??? Should print the range table; for now, just skip it. */
1018 p
+= 2; /* skip range table bits */
1019 EXTRACT_NUMBER_AND_INCR (count
, p
);
1020 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1026 fprintf (stderr
, "/begline");
1030 fprintf (stderr
, "/endline");
1033 case on_failure_jump
:
1034 extract_number_and_incr (&mcnt
, &p
);
1035 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1038 case on_failure_keep_string_jump
:
1039 extract_number_and_incr (&mcnt
, &p
);
1040 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1043 case on_failure_jump_nastyloop
:
1044 extract_number_and_incr (&mcnt
, &p
);
1045 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1048 case on_failure_jump_loop
:
1049 extract_number_and_incr (&mcnt
, &p
);
1050 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1053 case on_failure_jump_smart
:
1054 extract_number_and_incr (&mcnt
, &p
);
1055 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1059 extract_number_and_incr (&mcnt
, &p
);
1060 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1064 extract_number_and_incr (&mcnt
, &p
);
1065 extract_number_and_incr (&mcnt2
, &p
);
1066 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1070 extract_number_and_incr (&mcnt
, &p
);
1071 extract_number_and_incr (&mcnt2
, &p
);
1072 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1076 extract_number_and_incr (&mcnt
, &p
);
1077 extract_number_and_incr (&mcnt2
, &p
);
1078 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1082 fprintf (stderr
, "/wordbound");
1086 fprintf (stderr
, "/notwordbound");
1090 fprintf (stderr
, "/wordbeg");
1094 fprintf (stderr
, "/wordend");
1097 fprintf (stderr
, "/syntaxspec");
1099 fprintf (stderr
, "/%d", mcnt
);
1103 fprintf (stderr
, "/notsyntaxspec");
1105 fprintf (stderr
, "/%d", mcnt
);
1110 fprintf (stderr
, "/before_dot");
1114 fprintf (stderr
, "/at_dot");
1118 fprintf (stderr
, "/after_dot");
1122 fprintf (stderr
, "/categoryspec");
1124 fprintf (stderr
, "/%d", mcnt
);
1127 case notcategoryspec
:
1128 fprintf (stderr
, "/notcategoryspec");
1130 fprintf (stderr
, "/%d", mcnt
);
1135 fprintf (stderr
, "/begbuf");
1139 fprintf (stderr
, "/endbuf");
1143 fprintf (stderr
, "?%d", *(p
-1));
1146 fprintf (stderr
, "\n");
1149 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1154 print_compiled_pattern (bufp
)
1155 struct re_pattern_buffer
*bufp
;
1157 re_char
*buffer
= bufp
->buffer
;
1159 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1160 printf ("%ld bytes used/%ld bytes allocated.\n",
1161 bufp
->used
, bufp
->allocated
);
1163 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1165 printf ("fastmap: ");
1166 print_fastmap (bufp
->fastmap
);
1169 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1170 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1171 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1172 printf ("no_sub: %d\t", bufp
->no_sub
);
1173 printf ("not_bol: %d\t", bufp
->not_bol
);
1174 printf ("not_eol: %d\t", bufp
->not_eol
);
1175 printf ("syntax: %lx\n", bufp
->syntax
);
1177 /* Perhaps we should print the translate table? */
1182 print_double_string (where
, string1
, size1
, string2
, size2
)
1195 if (FIRST_STRING_P (where
))
1197 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1198 putchar (string1
[this_char
]);
1203 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1204 putchar (string2
[this_char
]);
1208 #else /* not DEBUG */
1213 # define DEBUG_STATEMENT(e)
1214 # define DEBUG_PRINT1(x)
1215 # define DEBUG_PRINT2(x1, x2)
1216 # define DEBUG_PRINT3(x1, x2, x3)
1217 # define DEBUG_PRINT4(x1, x2, x3, x4)
1218 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1219 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1221 #endif /* not DEBUG */
1223 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1224 also be assigned to arbitrarily: each pattern buffer stores its own
1225 syntax, so it can be changed between regex compilations. */
1226 /* This has no initializer because initialized variables in Emacs
1227 become read-only after dumping. */
1228 reg_syntax_t re_syntax_options
;
1231 /* Specify the precise syntax of regexps for compilation. This provides
1232 for compatibility for various utilities which historically have
1233 different, incompatible syntaxes.
1235 The argument SYNTAX is a bit mask comprised of the various bits
1236 defined in regex.h. We return the old syntax. */
1239 re_set_syntax (syntax
)
1240 reg_syntax_t syntax
;
1242 reg_syntax_t ret
= re_syntax_options
;
1244 re_syntax_options
= syntax
;
1247 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1249 /* This table gives an error message for each of the error codes listed
1250 in regex.h. Obviously the order here has to be same as there.
1251 POSIX doesn't require that we do anything for REG_NOERROR,
1252 but why not be nice? */
1254 static const char *re_error_msgid
[] =
1256 gettext_noop ("Success"), /* REG_NOERROR */
1257 gettext_noop ("No match"), /* REG_NOMATCH */
1258 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1259 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1260 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1261 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1262 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1263 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1264 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1265 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1266 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1267 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1268 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1269 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1270 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1271 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1272 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1275 /* Avoiding alloca during matching, to placate r_alloc. */
1277 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1278 searching and matching functions should not call alloca. On some
1279 systems, alloca is implemented in terms of malloc, and if we're
1280 using the relocating allocator routines, then malloc could cause a
1281 relocation, which might (if the strings being searched are in the
1282 ralloc heap) shift the data out from underneath the regexp
1285 Here's another reason to avoid allocation: Emacs
1286 processes input from X in a signal handler; processing X input may
1287 call malloc; if input arrives while a matching routine is calling
1288 malloc, then we're scrod. But Emacs can't just block input while
1289 calling matching routines; then we don't notice interrupts when
1290 they come in. So, Emacs blocks input around all regexp calls
1291 except the matching calls, which it leaves unprotected, in the
1292 faith that they will not malloc. */
1294 /* Normally, this is fine. */
1295 #define MATCH_MAY_ALLOCATE
1297 /* When using GNU C, we are not REALLY using the C alloca, no matter
1298 what config.h may say. So don't take precautions for it. */
1303 /* The match routines may not allocate if (1) they would do it with malloc
1304 and (2) it's not safe for them to use malloc.
1305 Note that if REL_ALLOC is defined, matching would not use malloc for the
1306 failure stack, but we would still use it for the register vectors;
1307 so REL_ALLOC should not affect this. */
1308 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1309 # undef MATCH_MAY_ALLOCATE
1313 /* Failure stack declarations and macros; both re_compile_fastmap and
1314 re_match_2 use a failure stack. These have to be macros because of
1315 REGEX_ALLOCATE_STACK. */
1318 /* Approximate number of failure points for which to initially allocate space
1319 when matching. If this number is exceeded, we allocate more
1320 space, so it is not a hard limit. */
1321 #ifndef INIT_FAILURE_ALLOC
1322 # define INIT_FAILURE_ALLOC 20
1325 /* Roughly the maximum number of failure points on the stack. Would be
1326 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1327 This is a variable only so users of regex can assign to it; we never
1328 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1329 before using it, so it should probably be a byte-count instead. */
1330 # if defined MATCH_MAY_ALLOCATE
1331 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1332 whose default stack limit is 2mb. In order for a larger
1333 value to work reliably, you have to try to make it accord
1334 with the process stack limit. */
1335 size_t re_max_failures
= 40000;
1337 size_t re_max_failures
= 4000;
1340 union fail_stack_elt
1343 /* This should be the biggest `int' that's no bigger than a pointer. */
1347 typedef union fail_stack_elt fail_stack_elt_t
;
1351 fail_stack_elt_t
*stack
;
1353 size_t avail
; /* Offset of next open position. */
1354 size_t frame
; /* Offset of the cur constructed frame. */
1357 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1358 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1361 /* Define macros to initialize and free the failure stack.
1362 Do `return -2' if the alloc fails. */
1364 #ifdef MATCH_MAY_ALLOCATE
1365 # define INIT_FAIL_STACK() \
1367 fail_stack.stack = (fail_stack_elt_t *) \
1368 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1369 * sizeof (fail_stack_elt_t)); \
1371 if (fail_stack.stack == NULL) \
1374 fail_stack.size = INIT_FAILURE_ALLOC; \
1375 fail_stack.avail = 0; \
1376 fail_stack.frame = 0; \
1379 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1381 # define INIT_FAIL_STACK() \
1383 fail_stack.avail = 0; \
1384 fail_stack.frame = 0; \
1387 # define RESET_FAIL_STACK() ((void)0)
1391 /* Double the size of FAIL_STACK, up to a limit
1392 which allows approximately `re_max_failures' items.
1394 Return 1 if succeeds, and 0 if either ran out of memory
1395 allocating space for it or it was already too large.
1397 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1399 /* Factor to increase the failure stack size by
1400 when we increase it.
1401 This used to be 2, but 2 was too wasteful
1402 because the old discarded stacks added up to as much space
1403 were as ultimate, maximum-size stack. */
1404 #define FAIL_STACK_GROWTH_FACTOR 4
1406 #define GROW_FAIL_STACK(fail_stack) \
1407 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1408 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1410 : ((fail_stack).stack \
1411 = (fail_stack_elt_t *) \
1412 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1413 (fail_stack).size * sizeof (fail_stack_elt_t), \
1414 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1415 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1416 * FAIL_STACK_GROWTH_FACTOR))), \
1418 (fail_stack).stack == NULL \
1420 : ((fail_stack).size \
1421 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1422 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1423 * FAIL_STACK_GROWTH_FACTOR)) \
1424 / sizeof (fail_stack_elt_t)), \
1428 /* Push a pointer value onto the failure stack.
1429 Assumes the variable `fail_stack'. Probably should only
1430 be called from within `PUSH_FAILURE_POINT'. */
1431 #define PUSH_FAILURE_POINTER(item) \
1432 fail_stack.stack[fail_stack.avail++].pointer = (item)
1434 /* This pushes an integer-valued item onto the failure stack.
1435 Assumes the variable `fail_stack'. Probably should only
1436 be called from within `PUSH_FAILURE_POINT'. */
1437 #define PUSH_FAILURE_INT(item) \
1438 fail_stack.stack[fail_stack.avail++].integer = (item)
1440 /* Push a fail_stack_elt_t value onto the failure stack.
1441 Assumes the variable `fail_stack'. Probably should only
1442 be called from within `PUSH_FAILURE_POINT'. */
1443 #define PUSH_FAILURE_ELT(item) \
1444 fail_stack.stack[fail_stack.avail++] = (item)
1446 /* These three POP... operations complement the three PUSH... operations.
1447 All assume that `fail_stack' is nonempty. */
1448 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1449 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1450 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1452 /* Individual items aside from the registers. */
1453 #define NUM_NONREG_ITEMS 3
1455 /* Used to examine the stack (to detect infinite loops). */
1456 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1457 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1458 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1459 #define TOP_FAILURE_HANDLE() fail_stack.frame
1462 #define ENSURE_FAIL_STACK(space) \
1463 while (REMAINING_AVAIL_SLOTS <= space) { \
1464 if (!GROW_FAIL_STACK (fail_stack)) \
1466 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1467 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1470 /* Push register NUM onto the stack. */
1471 #define PUSH_FAILURE_REG(num) \
1473 char *destination; \
1474 ENSURE_FAIL_STACK(3); \
1475 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1476 num, regstart[num], regend[num]); \
1477 PUSH_FAILURE_POINTER (regstart[num]); \
1478 PUSH_FAILURE_POINTER (regend[num]); \
1479 PUSH_FAILURE_INT (num); \
1482 /* Change the counter's value to VAL, but make sure that it will
1483 be reset when backtracking. */
1484 #define PUSH_NUMBER(ptr,val) \
1486 char *destination; \
1488 ENSURE_FAIL_STACK(3); \
1489 EXTRACT_NUMBER (c, ptr); \
1490 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1491 PUSH_FAILURE_INT (c); \
1492 PUSH_FAILURE_POINTER (ptr); \
1493 PUSH_FAILURE_INT (-1); \
1494 STORE_NUMBER (ptr, val); \
1497 /* Pop a saved register off the stack. */
1498 #define POP_FAILURE_REG_OR_COUNT() \
1500 int reg = POP_FAILURE_INT (); \
1503 /* It's a counter. */ \
1504 /* Here, we discard `const', making re_match non-reentrant. */ \
1505 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1506 reg = POP_FAILURE_INT (); \
1507 STORE_NUMBER (ptr, reg); \
1508 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1512 regend[reg] = POP_FAILURE_POINTER (); \
1513 regstart[reg] = POP_FAILURE_POINTER (); \
1514 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1515 reg, regstart[reg], regend[reg]); \
1519 /* Check that we are not stuck in an infinite loop. */
1520 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1522 int failure = TOP_FAILURE_HANDLE (); \
1523 /* Check for infinite matching loops */ \
1524 while (failure > 0 \
1525 && (FAILURE_STR (failure) == string_place \
1526 || FAILURE_STR (failure) == NULL)) \
1528 assert (FAILURE_PAT (failure) >= bufp->buffer \
1529 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1530 if (FAILURE_PAT (failure) == pat_cur) \
1535 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1536 failure = NEXT_FAILURE_HANDLE(failure); \
1538 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1541 /* Push the information about the state we will need
1542 if we ever fail back to it.
1544 Requires variables fail_stack, regstart, regend and
1545 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1548 Does `return FAILURE_CODE' if runs out of memory. */
1550 #define PUSH_FAILURE_POINT(pattern, string_place) \
1552 char *destination; \
1553 /* Must be int, so when we don't save any registers, the arithmetic \
1554 of 0 + -1 isn't done as unsigned. */ \
1556 DEBUG_STATEMENT (nfailure_points_pushed++); \
1557 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1558 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1559 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1561 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1563 DEBUG_PRINT1 ("\n"); \
1565 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1566 PUSH_FAILURE_INT (fail_stack.frame); \
1568 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1569 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1570 DEBUG_PRINT1 ("'\n"); \
1571 PUSH_FAILURE_POINTER (string_place); \
1573 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1574 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1575 PUSH_FAILURE_POINTER (pattern); \
1577 /* Close the frame by moving the frame pointer past it. */ \
1578 fail_stack.frame = fail_stack.avail; \
1581 /* Estimate the size of data pushed by a typical failure stack entry.
1582 An estimate is all we need, because all we use this for
1583 is to choose a limit for how big to make the failure stack. */
1584 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1585 #define TYPICAL_FAILURE_SIZE 20
1587 /* How many items can still be added to the stack without overflowing it. */
1588 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1591 /* Pops what PUSH_FAIL_STACK pushes.
1593 We restore into the parameters, all of which should be lvalues:
1594 STR -- the saved data position.
1595 PAT -- the saved pattern position.
1596 REGSTART, REGEND -- arrays of string positions.
1598 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1599 `pend', `string1', `size1', `string2', and `size2'. */
1601 #define POP_FAILURE_POINT(str, pat) \
1603 assert (!FAIL_STACK_EMPTY ()); \
1605 /* Remove failure points and point to how many regs pushed. */ \
1606 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1607 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1608 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1610 /* Pop the saved registers. */ \
1611 while (fail_stack.frame < fail_stack.avail) \
1612 POP_FAILURE_REG_OR_COUNT (); \
1614 pat = POP_FAILURE_POINTER (); \
1615 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1616 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1618 /* If the saved string location is NULL, it came from an \
1619 on_failure_keep_string_jump opcode, and we want to throw away the \
1620 saved NULL, thus retaining our current position in the string. */ \
1621 str = POP_FAILURE_POINTER (); \
1622 DEBUG_PRINT2 (" Popping string %p: `", str); \
1623 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1624 DEBUG_PRINT1 ("'\n"); \
1626 fail_stack.frame = POP_FAILURE_INT (); \
1627 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1629 assert (fail_stack.avail >= 0); \
1630 assert (fail_stack.frame <= fail_stack.avail); \
1632 DEBUG_STATEMENT (nfailure_points_popped++); \
1633 } while (0) /* POP_FAILURE_POINT */
1637 /* Registers are set to a sentinel when they haven't yet matched. */
1638 #define REG_UNSET(e) ((e) == NULL)
1640 /* Subroutine declarations and macros for regex_compile. */
1642 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1643 reg_syntax_t syntax
,
1644 struct re_pattern_buffer
*bufp
));
1645 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1646 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1647 int arg1
, int arg2
));
1648 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1649 int arg
, unsigned char *end
));
1650 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1651 int arg1
, int arg2
, unsigned char *end
));
1652 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1654 reg_syntax_t syntax
));
1655 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1657 reg_syntax_t syntax
));
1658 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1659 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1660 char *fastmap
, const int multibyte
));
1662 /* Fetch the next character in the uncompiled pattern, with no
1664 #define PATFETCH(c) \
1667 if (p == pend) return REG_EEND; \
1668 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1673 /* If `translate' is non-null, return translate[D], else just D. We
1674 cast the subscript to translate because some data is declared as
1675 `char *', to avoid warnings when a string constant is passed. But
1676 when we use a character as a subscript we must make it unsigned. */
1678 # define TRANSLATE(d) \
1679 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1683 /* Macros for outputting the compiled pattern into `buffer'. */
1685 /* If the buffer isn't allocated when it comes in, use this. */
1686 #define INIT_BUF_SIZE 32
1688 /* Make sure we have at least N more bytes of space in buffer. */
1689 #define GET_BUFFER_SPACE(n) \
1690 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1693 /* Make sure we have one more byte of buffer space and then add C to it. */
1694 #define BUF_PUSH(c) \
1696 GET_BUFFER_SPACE (1); \
1697 *b++ = (unsigned char) (c); \
1701 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1702 #define BUF_PUSH_2(c1, c2) \
1704 GET_BUFFER_SPACE (2); \
1705 *b++ = (unsigned char) (c1); \
1706 *b++ = (unsigned char) (c2); \
1710 /* As with BUF_PUSH_2, except for three bytes. */
1711 #define BUF_PUSH_3(c1, c2, c3) \
1713 GET_BUFFER_SPACE (3); \
1714 *b++ = (unsigned char) (c1); \
1715 *b++ = (unsigned char) (c2); \
1716 *b++ = (unsigned char) (c3); \
1720 /* Store a jump with opcode OP at LOC to location TO. We store a
1721 relative address offset by the three bytes the jump itself occupies. */
1722 #define STORE_JUMP(op, loc, to) \
1723 store_op1 (op, loc, (to) - (loc) - 3)
1725 /* Likewise, for a two-argument jump. */
1726 #define STORE_JUMP2(op, loc, to, arg) \
1727 store_op2 (op, loc, (to) - (loc) - 3, arg)
1729 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1730 #define INSERT_JUMP(op, loc, to) \
1731 insert_op1 (op, loc, (to) - (loc) - 3, b)
1733 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1734 #define INSERT_JUMP2(op, loc, to, arg) \
1735 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1738 /* This is not an arbitrary limit: the arguments which represent offsets
1739 into the pattern are two bytes long. So if 2^16 bytes turns out to
1740 be too small, many things would have to change. */
1741 /* Any other compiler which, like MSC, has allocation limit below 2^16
1742 bytes will have to use approach similar to what was done below for
1743 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1744 reallocating to 0 bytes. Such thing is not going to work too well.
1745 You have been warned!! */
1746 #if defined _MSC_VER && !defined WIN32
1747 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1748 # define MAX_BUF_SIZE 65500L
1750 # define MAX_BUF_SIZE (1L << 16)
1753 /* Extend the buffer by twice its current size via realloc and
1754 reset the pointers that pointed into the old block to point to the
1755 correct places in the new one. If extending the buffer results in it
1756 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1757 #if __BOUNDED_POINTERS__
1758 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1759 # define MOVE_BUFFER_POINTER(P) \
1760 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1761 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1764 SET_HIGH_BOUND (b); \
1765 SET_HIGH_BOUND (begalt); \
1766 if (fixup_alt_jump) \
1767 SET_HIGH_BOUND (fixup_alt_jump); \
1769 SET_HIGH_BOUND (laststart); \
1770 if (pending_exact) \
1771 SET_HIGH_BOUND (pending_exact); \
1774 # define MOVE_BUFFER_POINTER(P) (P) += incr
1775 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1777 #define EXTEND_BUFFER() \
1779 re_char *old_buffer = bufp->buffer; \
1780 if (bufp->allocated == MAX_BUF_SIZE) \
1782 bufp->allocated <<= 1; \
1783 if (bufp->allocated > MAX_BUF_SIZE) \
1784 bufp->allocated = MAX_BUF_SIZE; \
1785 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1786 if (bufp->buffer == NULL) \
1787 return REG_ESPACE; \
1788 /* If the buffer moved, move all the pointers into it. */ \
1789 if (old_buffer != bufp->buffer) \
1791 int incr = bufp->buffer - old_buffer; \
1792 MOVE_BUFFER_POINTER (b); \
1793 MOVE_BUFFER_POINTER (begalt); \
1794 if (fixup_alt_jump) \
1795 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1797 MOVE_BUFFER_POINTER (laststart); \
1798 if (pending_exact) \
1799 MOVE_BUFFER_POINTER (pending_exact); \
1801 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1805 /* Since we have one byte reserved for the register number argument to
1806 {start,stop}_memory, the maximum number of groups we can report
1807 things about is what fits in that byte. */
1808 #define MAX_REGNUM 255
1810 /* But patterns can have more than `MAX_REGNUM' registers. We just
1811 ignore the excess. */
1812 typedef int regnum_t
;
1815 /* Macros for the compile stack. */
1817 /* Since offsets can go either forwards or backwards, this type needs to
1818 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1819 /* int may be not enough when sizeof(int) == 2. */
1820 typedef long pattern_offset_t
;
1824 pattern_offset_t begalt_offset
;
1825 pattern_offset_t fixup_alt_jump
;
1826 pattern_offset_t laststart_offset
;
1828 } compile_stack_elt_t
;
1833 compile_stack_elt_t
*stack
;
1835 unsigned avail
; /* Offset of next open position. */
1836 } compile_stack_type
;
1839 #define INIT_COMPILE_STACK_SIZE 32
1841 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1842 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1844 /* The next available element. */
1845 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1847 /* Explicit quit checking is only used on NTemacs. */
1848 #if defined WINDOWSNT && defined emacs && defined QUIT
1849 extern int immediate_quit
;
1850 # define IMMEDIATE_QUIT_CHECK \
1852 if (immediate_quit) QUIT; \
1855 # define IMMEDIATE_QUIT_CHECK ((void)0)
1858 /* Structure to manage work area for range table. */
1859 struct range_table_work_area
1861 int *table
; /* actual work area. */
1862 int allocated
; /* allocated size for work area in bytes. */
1863 int used
; /* actually used size in words. */
1864 int bits
; /* flag to record character classes */
1867 /* Make sure that WORK_AREA can hold more N multibyte characters.
1868 This is used only in set_image_of_range and set_image_of_range_1.
1869 It expects WORK_AREA to be a pointer.
1870 If it can't get the space, it returns from the surrounding function. */
1872 #define EXTEND_RANGE_TABLE(work_area, n) \
1874 if (((work_area)->used + (n)) * sizeof (int) > (work_area)->allocated) \
1876 extend_range_table_work_area (work_area); \
1877 if ((work_area)->table == 0) \
1878 return (REG_ESPACE); \
1882 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1883 (work_area).bits |= (bit)
1885 /* Bits used to implement the multibyte-part of the various character classes
1886 such as [:alnum:] in a charset's range table. */
1887 #define BIT_WORD 0x1
1888 #define BIT_LOWER 0x2
1889 #define BIT_PUNCT 0x4
1890 #define BIT_SPACE 0x8
1891 #define BIT_UPPER 0x10
1892 #define BIT_MULTIBYTE 0x20
1894 /* Set a range START..END to WORK_AREA.
1895 The range is passed through TRANSLATE, so START and END
1896 should be untranslated. */
1897 #define SET_RANGE_TABLE_WORK_AREA(work_area, start, end) \
1900 tem = set_image_of_range (&work_area, start, end, translate); \
1902 FREE_STACK_RETURN (tem); \
1905 /* Free allocated memory for WORK_AREA. */
1906 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1908 if ((work_area).table) \
1909 free ((work_area).table); \
1912 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1913 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1914 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1915 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1918 /* Set the bit for character C in a list. */
1919 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1922 /* Get the next unsigned number in the uncompiled pattern. */
1923 #define GET_UNSIGNED_NUMBER(num) \
1924 do { if (p != pend) \
1928 FREE_STACK_RETURN (REG_BADBR); \
1929 while ('0' <= c && c <= '9') \
1935 num = num * 10 + c - '0'; \
1936 if (num / 10 != prev) \
1937 FREE_STACK_RETURN (REG_BADBR); \
1943 FREE_STACK_RETURN (REG_BADBR); \
1947 #if WIDE_CHAR_SUPPORT
1948 /* The GNU C library provides support for user-defined character classes
1949 and the functions from ISO C amendement 1. */
1950 # ifdef CHARCLASS_NAME_MAX
1951 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1953 /* This shouldn't happen but some implementation might still have this
1954 problem. Use a reasonable default value. */
1955 # define CHAR_CLASS_MAX_LENGTH 256
1957 typedef wctype_t re_wctype_t
;
1958 typedef wchar_t re_wchar_t
;
1959 # define re_wctype wctype
1960 # define re_iswctype iswctype
1961 # define re_wctype_to_bit(cc) 0
1963 # define CHAR_CLASS_MAX_LENGTH 9 /* Namely, `multibyte'. */
1966 /* Character classes. */
1967 typedef enum { RECC_ERROR
= 0,
1968 RECC_ALNUM
, RECC_ALPHA
, RECC_WORD
,
1969 RECC_GRAPH
, RECC_PRINT
,
1970 RECC_LOWER
, RECC_UPPER
,
1971 RECC_PUNCT
, RECC_CNTRL
,
1972 RECC_DIGIT
, RECC_XDIGIT
,
1973 RECC_BLANK
, RECC_SPACE
,
1974 RECC_MULTIBYTE
, RECC_NONASCII
,
1975 RECC_ASCII
, RECC_UNIBYTE
1978 typedef int re_wchar_t
;
1980 /* Map a string to the char class it names (if any). */
1985 const char *string
= str
;
1986 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
1987 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
1988 else if (STREQ (string
, "word")) return RECC_WORD
;
1989 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
1990 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
1991 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
1992 else if (STREQ (string
, "lower")) return RECC_LOWER
;
1993 else if (STREQ (string
, "print")) return RECC_PRINT
;
1994 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
1995 else if (STREQ (string
, "space")) return RECC_SPACE
;
1996 else if (STREQ (string
, "upper")) return RECC_UPPER
;
1997 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
1998 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
1999 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2000 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2001 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2002 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2006 /* True iff CH is in the char class CC. */
2008 re_iswctype (ch
, cc
)
2014 case RECC_ALNUM
: return ISALNUM (ch
);
2015 case RECC_ALPHA
: return ISALPHA (ch
);
2016 case RECC_BLANK
: return ISBLANK (ch
);
2017 case RECC_CNTRL
: return ISCNTRL (ch
);
2018 case RECC_DIGIT
: return ISDIGIT (ch
);
2019 case RECC_GRAPH
: return ISGRAPH (ch
);
2020 case RECC_LOWER
: return ISLOWER (ch
);
2021 case RECC_PRINT
: return ISPRINT (ch
);
2022 case RECC_PUNCT
: return ISPUNCT (ch
);
2023 case RECC_SPACE
: return ISSPACE (ch
);
2024 case RECC_UPPER
: return ISUPPER (ch
);
2025 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2026 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2027 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2028 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2029 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2030 case RECC_WORD
: return ISWORD (ch
);
2031 case RECC_ERROR
: return false;
2037 /* Return a bit-pattern to use in the range-table bits to match multibyte
2038 chars of class CC. */
2040 re_wctype_to_bit (cc
)
2045 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2046 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2047 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2048 case RECC_LOWER
: return BIT_LOWER
;
2049 case RECC_UPPER
: return BIT_UPPER
;
2050 case RECC_PUNCT
: return BIT_PUNCT
;
2051 case RECC_SPACE
: return BIT_SPACE
;
2052 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2053 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2060 /* Filling in the work area of a range. */
2062 /* Actually extend the space in WORK_AREA. */
2065 extend_range_table_work_area (work_area
)
2066 struct range_table_work_area
*work_area
;
2068 work_area
->allocated
+= 16 * sizeof (int);
2069 if (work_area
->table
)
2071 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2074 = (int *) malloc (work_area
->allocated
);
2079 /* Carefully find the ranges of codes that are equivalent
2080 under case conversion to the range start..end when passed through
2081 TRANSLATE. Handle the case where non-letters can come in between
2082 two upper-case letters (which happens in Latin-1).
2083 Also handle the case of groups of more than 2 case-equivalent chars.
2085 The basic method is to look at consecutive characters and see
2086 if they can form a run that can be handled as one.
2088 Returns -1 if successful, REG_ESPACE if ran out of space. */
2091 set_image_of_range_1 (work_area
, start
, end
, translate
)
2092 RE_TRANSLATE_TYPE translate
;
2093 struct range_table_work_area
*work_area
;
2094 re_wchar_t start
, end
;
2096 /* `one_case' indicates a character, or a run of characters,
2097 each of which is an isolate (no case-equivalents).
2098 This includes all ASCII non-letters.
2100 `two_case' indicates a character, or a run of characters,
2101 each of which has two case-equivalent forms.
2102 This includes all ASCII letters.
2104 `strange' indicates a character that has more than one
2107 enum case_type
{one_case
, two_case
, strange
};
2109 /* Describe the run that is in progress,
2110 which the next character can try to extend.
2111 If run_type is strange, that means there really is no run.
2112 If run_type is one_case, then run_start...run_end is the run.
2113 If run_type is two_case, then the run is run_start...run_end,
2114 and the case-equivalents end at run_eqv_end. */
2116 enum case_type run_type
= strange
;
2117 int run_start
, run_end
, run_eqv_end
;
2119 Lisp_Object eqv_table
;
2121 if (!RE_TRANSLATE_P (translate
))
2123 EXTEND_RANGE_TABLE (work_area
, 2);
2124 work_area
->table
[work_area
->used
++] = (start
);
2125 work_area
->table
[work_area
->used
++] = (end
);
2129 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2131 for (; start
<= end
; start
++)
2133 enum case_type this_type
;
2134 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2135 int minchar
, maxchar
;
2137 /* Classify this character */
2139 this_type
= one_case
;
2140 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2141 this_type
= two_case
;
2143 this_type
= strange
;
2146 minchar
= start
, maxchar
= eqv
;
2148 minchar
= eqv
, maxchar
= start
;
2150 /* Can this character extend the run in progress? */
2151 if (this_type
== strange
|| this_type
!= run_type
2152 || !(minchar
== run_end
+ 1
2153 && (run_type
== two_case
2154 ? maxchar
== run_eqv_end
+ 1 : 1)))
2157 Record each of its equivalent ranges. */
2158 if (run_type
== one_case
)
2160 EXTEND_RANGE_TABLE (work_area
, 2);
2161 work_area
->table
[work_area
->used
++] = run_start
;
2162 work_area
->table
[work_area
->used
++] = run_end
;
2164 else if (run_type
== two_case
)
2166 EXTEND_RANGE_TABLE (work_area
, 4);
2167 work_area
->table
[work_area
->used
++] = run_start
;
2168 work_area
->table
[work_area
->used
++] = run_end
;
2169 work_area
->table
[work_area
->used
++]
2170 = RE_TRANSLATE (eqv_table
, run_start
);
2171 work_area
->table
[work_area
->used
++]
2172 = RE_TRANSLATE (eqv_table
, run_end
);
2177 if (this_type
== strange
)
2179 /* For a strange character, add each of its equivalents, one
2180 by one. Don't start a range. */
2183 EXTEND_RANGE_TABLE (work_area
, 2);
2184 work_area
->table
[work_area
->used
++] = eqv
;
2185 work_area
->table
[work_area
->used
++] = eqv
;
2186 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2188 while (eqv
!= start
);
2191 /* Add this char to the run, or start a new run. */
2192 else if (run_type
== strange
)
2194 /* Initialize a new range. */
2195 run_type
= this_type
;
2198 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2202 /* Extend a running range. */
2204 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2208 /* If a run is still in progress at the end, finish it now
2209 by recording its equivalent ranges. */
2210 if (run_type
== one_case
)
2212 EXTEND_RANGE_TABLE (work_area
, 2);
2213 work_area
->table
[work_area
->used
++] = run_start
;
2214 work_area
->table
[work_area
->used
++] = run_end
;
2216 else if (run_type
== two_case
)
2218 EXTEND_RANGE_TABLE (work_area
, 4);
2219 work_area
->table
[work_area
->used
++] = run_start
;
2220 work_area
->table
[work_area
->used
++] = run_end
;
2221 work_area
->table
[work_area
->used
++]
2222 = RE_TRANSLATE (eqv_table
, run_start
);
2223 work_area
->table
[work_area
->used
++]
2224 = RE_TRANSLATE (eqv_table
, run_end
);
2232 /* Record the the image of the range start..end when passed through
2233 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2234 and is not even necessarily contiguous.
2235 Normally we approximate it with the smallest contiguous range that contains
2236 all the chars we need. However, for Latin-1 we go to extra effort
2239 This function is not called for ASCII ranges.
2241 Returns -1 if successful, REG_ESPACE if ran out of space. */
2244 set_image_of_range (work_area
, start
, end
, translate
)
2245 RE_TRANSLATE_TYPE translate
;
2246 struct range_table_work_area
*work_area
;
2247 re_wchar_t start
, end
;
2249 re_wchar_t cmin
, cmax
;
2252 /* For Latin-1 ranges, use set_image_of_range_1
2253 to get proper handling of ranges that include letters and nonletters.
2254 For a range that includes the whole of Latin-1, this is not necessary.
2255 For other character sets, we don't bother to get this right. */
2256 if (RE_TRANSLATE_P (translate
) && start
< 04400
2257 && !(start
< 04200 && end
>= 04377))
2264 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2274 EXTEND_RANGE_TABLE (work_area
, 2);
2275 work_area
->table
[work_area
->used
++] = (start
);
2276 work_area
->table
[work_area
->used
++] = (end
);
2278 cmin
= -1, cmax
= -1;
2280 if (RE_TRANSLATE_P (translate
))
2284 for (ch
= start
; ch
<= end
; ch
++)
2286 re_wchar_t c
= TRANSLATE (ch
);
2287 if (! (start
<= c
&& c
<= end
))
2293 cmin
= MIN (cmin
, c
);
2294 cmax
= MAX (cmax
, c
);
2301 EXTEND_RANGE_TABLE (work_area
, 2);
2302 work_area
->table
[work_area
->used
++] = (cmin
);
2303 work_area
->table
[work_area
->used
++] = (cmax
);
2310 #ifndef MATCH_MAY_ALLOCATE
2312 /* If we cannot allocate large objects within re_match_2_internal,
2313 we make the fail stack and register vectors global.
2314 The fail stack, we grow to the maximum size when a regexp
2316 The register vectors, we adjust in size each time we
2317 compile a regexp, according to the number of registers it needs. */
2319 static fail_stack_type fail_stack
;
2321 /* Size with which the following vectors are currently allocated.
2322 That is so we can make them bigger as needed,
2323 but never make them smaller. */
2324 static int regs_allocated_size
;
2326 static re_char
** regstart
, ** regend
;
2327 static re_char
**best_regstart
, **best_regend
;
2329 /* Make the register vectors big enough for NUM_REGS registers,
2330 but don't make them smaller. */
2333 regex_grow_registers (num_regs
)
2336 if (num_regs
> regs_allocated_size
)
2338 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2339 RETALLOC_IF (regend
, num_regs
, re_char
*);
2340 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2341 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2343 regs_allocated_size
= num_regs
;
2347 #endif /* not MATCH_MAY_ALLOCATE */
2349 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2353 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2354 Returns one of error codes defined in `regex.h', or zero for success.
2356 Assumes the `allocated' (and perhaps `buffer') and `translate'
2357 fields are set in BUFP on entry.
2359 If it succeeds, results are put in BUFP (if it returns an error, the
2360 contents of BUFP are undefined):
2361 `buffer' is the compiled pattern;
2362 `syntax' is set to SYNTAX;
2363 `used' is set to the length of the compiled pattern;
2364 `fastmap_accurate' is zero;
2365 `re_nsub' is the number of subexpressions in PATTERN;
2366 `not_bol' and `not_eol' are zero;
2368 The `fastmap' field is neither examined nor set. */
2370 /* Insert the `jump' from the end of last alternative to "here".
2371 The space for the jump has already been allocated. */
2372 #define FIXUP_ALT_JUMP() \
2374 if (fixup_alt_jump) \
2375 STORE_JUMP (jump, fixup_alt_jump, b); \
2379 /* Return, freeing storage we allocated. */
2380 #define FREE_STACK_RETURN(value) \
2382 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2383 free (compile_stack.stack); \
2387 static reg_errcode_t
2388 regex_compile (pattern
, size
, syntax
, bufp
)
2391 reg_syntax_t syntax
;
2392 struct re_pattern_buffer
*bufp
;
2394 /* We fetch characters from PATTERN here. */
2395 register re_wchar_t c
, c1
;
2397 /* A random temporary spot in PATTERN. */
2400 /* Points to the end of the buffer, where we should append. */
2401 register unsigned char *b
;
2403 /* Keeps track of unclosed groups. */
2404 compile_stack_type compile_stack
;
2406 /* Points to the current (ending) position in the pattern. */
2408 /* `const' makes AIX compiler fail. */
2409 unsigned char *p
= pattern
;
2411 re_char
*p
= pattern
;
2413 re_char
*pend
= pattern
+ size
;
2415 /* How to translate the characters in the pattern. */
2416 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2418 /* Address of the count-byte of the most recently inserted `exactn'
2419 command. This makes it possible to tell if a new exact-match
2420 character can be added to that command or if the character requires
2421 a new `exactn' command. */
2422 unsigned char *pending_exact
= 0;
2424 /* Address of start of the most recently finished expression.
2425 This tells, e.g., postfix * where to find the start of its
2426 operand. Reset at the beginning of groups and alternatives. */
2427 unsigned char *laststart
= 0;
2429 /* Address of beginning of regexp, or inside of last group. */
2430 unsigned char *begalt
;
2432 /* Place in the uncompiled pattern (i.e., the {) to
2433 which to go back if the interval is invalid. */
2434 re_char
*beg_interval
;
2436 /* Address of the place where a forward jump should go to the end of
2437 the containing expression. Each alternative of an `or' -- except the
2438 last -- ends with a forward jump of this sort. */
2439 unsigned char *fixup_alt_jump
= 0;
2441 /* Counts open-groups as they are encountered. Remembered for the
2442 matching close-group on the compile stack, so the same register
2443 number is put in the stop_memory as the start_memory. */
2444 regnum_t regnum
= 0;
2446 /* Work area for range table of charset. */
2447 struct range_table_work_area range_table_work
;
2449 /* If the object matched can contain multibyte characters. */
2450 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2454 DEBUG_PRINT1 ("\nCompiling pattern: ");
2457 unsigned debug_count
;
2459 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2460 putchar (pattern
[debug_count
]);
2465 /* Initialize the compile stack. */
2466 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2467 if (compile_stack
.stack
== NULL
)
2470 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2471 compile_stack
.avail
= 0;
2473 range_table_work
.table
= 0;
2474 range_table_work
.allocated
= 0;
2476 /* Initialize the pattern buffer. */
2477 bufp
->syntax
= syntax
;
2478 bufp
->fastmap_accurate
= 0;
2479 bufp
->not_bol
= bufp
->not_eol
= 0;
2481 /* Set `used' to zero, so that if we return an error, the pattern
2482 printer (for debugging) will think there's no pattern. We reset it
2486 /* Always count groups, whether or not bufp->no_sub is set. */
2489 #if !defined emacs && !defined SYNTAX_TABLE
2490 /* Initialize the syntax table. */
2491 init_syntax_once ();
2494 if (bufp
->allocated
== 0)
2497 { /* If zero allocated, but buffer is non-null, try to realloc
2498 enough space. This loses if buffer's address is bogus, but
2499 that is the user's responsibility. */
2500 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2503 { /* Caller did not allocate a buffer. Do it for them. */
2504 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2506 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2508 bufp
->allocated
= INIT_BUF_SIZE
;
2511 begalt
= b
= bufp
->buffer
;
2513 /* Loop through the uncompiled pattern until we're at the end. */
2522 if ( /* If at start of pattern, it's an operator. */
2524 /* If context independent, it's an operator. */
2525 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2526 /* Otherwise, depends on what's come before. */
2527 || at_begline_loc_p (pattern
, p
, syntax
))
2528 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2537 if ( /* If at end of pattern, it's an operator. */
2539 /* If context independent, it's an operator. */
2540 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2541 /* Otherwise, depends on what's next. */
2542 || at_endline_loc_p (p
, pend
, syntax
))
2543 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2552 if ((syntax
& RE_BK_PLUS_QM
)
2553 || (syntax
& RE_LIMITED_OPS
))
2557 /* If there is no previous pattern... */
2560 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2561 FREE_STACK_RETURN (REG_BADRPT
);
2562 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2567 /* 1 means zero (many) matches is allowed. */
2568 boolean zero_times_ok
= 0, many_times_ok
= 0;
2571 /* If there is a sequence of repetition chars, collapse it
2572 down to just one (the right one). We can't combine
2573 interval operators with these because of, e.g., `a{2}*',
2574 which should only match an even number of `a's. */
2578 if ((syntax
& RE_FRUGAL
)
2579 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2583 zero_times_ok
|= c
!= '+';
2584 many_times_ok
|= c
!= '?';
2590 || (!(syntax
& RE_BK_PLUS_QM
)
2591 && (*p
== '+' || *p
== '?')))
2593 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2596 FREE_STACK_RETURN (REG_EESCAPE
);
2597 if (p
[1] == '+' || p
[1] == '?')
2598 PATFETCH (c
); /* Gobble up the backslash. */
2604 /* If we get here, we found another repeat character. */
2608 /* Star, etc. applied to an empty pattern is equivalent
2609 to an empty pattern. */
2610 if (!laststart
|| laststart
== b
)
2613 /* Now we know whether or not zero matches is allowed
2614 and also whether or not two or more matches is allowed. */
2619 boolean simple
= skip_one_char (laststart
) == b
;
2620 unsigned int startoffset
= 0;
2622 /* Check if the loop can match the empty string. */
2623 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2624 ? on_failure_jump
: on_failure_jump_loop
;
2625 assert (skip_one_char (laststart
) <= b
);
2627 if (!zero_times_ok
&& simple
)
2628 { /* Since simple * loops can be made faster by using
2629 on_failure_keep_string_jump, we turn simple P+
2630 into PP* if P is simple. */
2631 unsigned char *p1
, *p2
;
2632 startoffset
= b
- laststart
;
2633 GET_BUFFER_SPACE (startoffset
);
2634 p1
= b
; p2
= laststart
;
2640 GET_BUFFER_SPACE (6);
2643 STORE_JUMP (ofj
, b
, b
+ 6);
2645 /* Simple * loops can use on_failure_keep_string_jump
2646 depending on what follows. But since we don't know
2647 that yet, we leave the decision up to
2648 on_failure_jump_smart. */
2649 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2650 laststart
+ startoffset
, b
+ 6);
2652 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2657 /* A simple ? pattern. */
2658 assert (zero_times_ok
);
2659 GET_BUFFER_SPACE (3);
2660 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2664 else /* not greedy */
2665 { /* I wish the greedy and non-greedy cases could be merged. */
2667 GET_BUFFER_SPACE (7); /* We might use less. */
2670 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2672 /* The non-greedy multiple match looks like
2673 a repeat..until: we only need a conditional jump
2674 at the end of the loop. */
2675 if (emptyp
) BUF_PUSH (no_op
);
2676 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2677 : on_failure_jump
, b
, laststart
);
2681 /* The repeat...until naturally matches one or more.
2682 To also match zero times, we need to first jump to
2683 the end of the loop (its conditional jump). */
2684 INSERT_JUMP (jump
, laststart
, b
);
2690 /* non-greedy a?? */
2691 INSERT_JUMP (jump
, laststart
, b
+ 3);
2693 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2710 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2712 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2714 /* Ensure that we have enough space to push a charset: the
2715 opcode, the length count, and the bitset; 34 bytes in all. */
2716 GET_BUFFER_SPACE (34);
2720 /* We test `*p == '^' twice, instead of using an if
2721 statement, so we only need one BUF_PUSH. */
2722 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2726 /* Remember the first position in the bracket expression. */
2729 /* Push the number of bytes in the bitmap. */
2730 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2732 /* Clear the whole map. */
2733 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2735 /* charset_not matches newline according to a syntax bit. */
2736 if ((re_opcode_t
) b
[-2] == charset_not
2737 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2738 SET_LIST_BIT ('\n');
2740 /* Read in characters and ranges, setting map bits. */
2743 boolean escaped_char
= false;
2744 const unsigned char *p2
= p
;
2746 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2748 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2749 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2750 So the translation is done later in a loop. Example:
2751 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2754 /* \ might escape characters inside [...] and [^...]. */
2755 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2757 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2760 escaped_char
= true;
2764 /* Could be the end of the bracket expression. If it's
2765 not (i.e., when the bracket expression is `[]' so
2766 far), the ']' character bit gets set way below. */
2767 if (c
== ']' && p2
!= p1
)
2771 /* What should we do for the character which is
2772 greater than 0x7F, but not BASE_LEADING_CODE_P?
2775 /* See if we're at the beginning of a possible character
2778 if (!escaped_char
&&
2779 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2781 /* Leave room for the null. */
2782 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2783 const unsigned char *class_beg
;
2789 /* If pattern is `[[:'. */
2790 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2795 if ((c
== ':' && *p
== ']') || p
== pend
)
2797 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2800 /* This is in any case an invalid class name. */
2805 /* If isn't a word bracketed by `[:' and `:]':
2806 undo the ending character, the letters, and
2807 leave the leading `:' and `[' (but set bits for
2809 if (c
== ':' && *p
== ']')
2814 cc
= re_wctype (str
);
2817 FREE_STACK_RETURN (REG_ECTYPE
);
2819 /* Throw away the ] at the end of the character
2823 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2825 /* Most character classes in a multibyte match
2826 just set a flag. Exceptions are is_blank,
2827 is_digit, is_cntrl, and is_xdigit, since
2828 they can only match ASCII characters. We
2829 don't need to handle them for multibyte.
2830 They are distinguished by a negative wctype. */
2833 SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work
,
2834 re_wctype_to_bit (cc
));
2836 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2838 int translated
= TRANSLATE (ch
);
2839 if (re_iswctype (btowc (ch
), cc
))
2840 SET_LIST_BIT (translated
);
2843 /* Repeat the loop. */
2848 /* Go back to right after the "[:". */
2852 /* Because the `:' may starts the range, we
2853 can't simply set bit and repeat the loop.
2854 Instead, just set it to C and handle below. */
2859 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2862 /* Discard the `-'. */
2865 /* Fetch the character which ends the range. */
2868 if (SINGLE_BYTE_CHAR_P (c
))
2870 if (! SINGLE_BYTE_CHAR_P (c1
))
2872 /* Handle a range starting with a
2873 character of less than 256, and ending
2874 with a character of not less than 256.
2875 Split that into two ranges, the low one
2876 ending at 0377, and the high one
2877 starting at the smallest character in
2878 the charset of C1 and ending at C1. */
2879 int charset
= CHAR_CHARSET (c1
);
2880 re_wchar_t c2
= MAKE_CHAR (charset
, 0, 0);
2882 SET_RANGE_TABLE_WORK_AREA (range_table_work
,
2887 else if (!SAME_CHARSET_P (c
, c1
))
2888 FREE_STACK_RETURN (REG_ERANGE
);
2891 /* Range from C to C. */
2894 /* Set the range ... */
2895 if (SINGLE_BYTE_CHAR_P (c
))
2896 /* ... into bitmap. */
2898 re_wchar_t this_char
;
2899 re_wchar_t range_start
= c
, range_end
= c1
;
2901 /* If the start is after the end, the range is empty. */
2902 if (range_start
> range_end
)
2904 if (syntax
& RE_NO_EMPTY_RANGES
)
2905 FREE_STACK_RETURN (REG_ERANGE
);
2906 /* Else, repeat the loop. */
2910 for (this_char
= range_start
; this_char
<= range_end
;
2912 SET_LIST_BIT (TRANSLATE (this_char
));
2916 /* ... into range table. */
2917 SET_RANGE_TABLE_WORK_AREA (range_table_work
, c
, c1
);
2920 /* Discard any (non)matching list bytes that are all 0 at the
2921 end of the map. Decrease the map-length byte too. */
2922 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2926 /* Build real range table from work area. */
2927 if (RANGE_TABLE_WORK_USED (range_table_work
)
2928 || RANGE_TABLE_WORK_BITS (range_table_work
))
2931 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
2933 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2934 bytes for flags, two for COUNT, and three bytes for
2936 GET_BUFFER_SPACE (4 + used
* 3);
2938 /* Indicate the existence of range table. */
2939 laststart
[1] |= 0x80;
2941 /* Store the character class flag bits into the range table.
2942 If not in emacs, these flag bits are always 0. */
2943 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
2944 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
2946 STORE_NUMBER_AND_INCR (b
, used
/ 2);
2947 for (i
= 0; i
< used
; i
++)
2948 STORE_CHARACTER_AND_INCR
2949 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
2956 if (syntax
& RE_NO_BK_PARENS
)
2963 if (syntax
& RE_NO_BK_PARENS
)
2970 if (syntax
& RE_NEWLINE_ALT
)
2977 if (syntax
& RE_NO_BK_VBAR
)
2984 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2985 goto handle_interval
;
2991 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2993 /* Do not translate the character after the \, so that we can
2994 distinguish, e.g., \B from \b, even if we normally would
2995 translate, e.g., B to b. */
3001 if (syntax
& RE_NO_BK_PARENS
)
3002 goto normal_backslash
;
3009 /* Look for a special (?...) construct */
3010 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3012 PATFETCH (c
); /* Gobble up the '?'. */
3016 case ':': shy
= 1; break;
3018 /* Only (?:...) is supported right now. */
3019 FREE_STACK_RETURN (REG_BADPAT
);
3030 if (COMPILE_STACK_FULL
)
3032 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3033 compile_stack_elt_t
);
3034 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3036 compile_stack
.size
<<= 1;
3039 /* These are the values to restore when we hit end of this
3040 group. They are all relative offsets, so that if the
3041 whole pattern moves because of realloc, they will still
3043 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3044 COMPILE_STACK_TOP
.fixup_alt_jump
3045 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3046 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3047 COMPILE_STACK_TOP
.regnum
= shy
? -regnum
: regnum
;
3050 start_memory for groups beyond the last one we can
3051 represent in the compiled pattern. */
3052 if (regnum
<= MAX_REGNUM
&& !shy
)
3053 BUF_PUSH_2 (start_memory
, regnum
);
3055 compile_stack
.avail
++;
3060 /* If we've reached MAX_REGNUM groups, then this open
3061 won't actually generate any code, so we'll have to
3062 clear pending_exact explicitly. */
3068 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3070 if (COMPILE_STACK_EMPTY
)
3072 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3073 goto normal_backslash
;
3075 FREE_STACK_RETURN (REG_ERPAREN
);
3081 /* See similar code for backslashed left paren above. */
3082 if (COMPILE_STACK_EMPTY
)
3084 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3087 FREE_STACK_RETURN (REG_ERPAREN
);
3090 /* Since we just checked for an empty stack above, this
3091 ``can't happen''. */
3092 assert (compile_stack
.avail
!= 0);
3094 /* We don't just want to restore into `regnum', because
3095 later groups should continue to be numbered higher,
3096 as in `(ab)c(de)' -- the second group is #2. */
3097 regnum_t this_group_regnum
;
3099 compile_stack
.avail
--;
3100 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3102 = COMPILE_STACK_TOP
.fixup_alt_jump
3103 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3105 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3106 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
3107 /* If we've reached MAX_REGNUM groups, then this open
3108 won't actually generate any code, so we'll have to
3109 clear pending_exact explicitly. */
3112 /* We're at the end of the group, so now we know how many
3113 groups were inside this one. */
3114 if (this_group_regnum
<= MAX_REGNUM
&& this_group_regnum
> 0)
3115 BUF_PUSH_2 (stop_memory
, this_group_regnum
);
3120 case '|': /* `\|'. */
3121 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3122 goto normal_backslash
;
3124 if (syntax
& RE_LIMITED_OPS
)
3127 /* Insert before the previous alternative a jump which
3128 jumps to this alternative if the former fails. */
3129 GET_BUFFER_SPACE (3);
3130 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3134 /* The alternative before this one has a jump after it
3135 which gets executed if it gets matched. Adjust that
3136 jump so it will jump to this alternative's analogous
3137 jump (put in below, which in turn will jump to the next
3138 (if any) alternative's such jump, etc.). The last such
3139 jump jumps to the correct final destination. A picture:
3145 If we are at `b', then fixup_alt_jump right now points to a
3146 three-byte space after `a'. We'll put in the jump, set
3147 fixup_alt_jump to right after `b', and leave behind three
3148 bytes which we'll fill in when we get to after `c'. */
3152 /* Mark and leave space for a jump after this alternative,
3153 to be filled in later either by next alternative or
3154 when know we're at the end of a series of alternatives. */
3156 GET_BUFFER_SPACE (3);
3165 /* If \{ is a literal. */
3166 if (!(syntax
& RE_INTERVALS
)
3167 /* If we're at `\{' and it's not the open-interval
3169 || (syntax
& RE_NO_BK_BRACES
))
3170 goto normal_backslash
;
3174 /* If got here, then the syntax allows intervals. */
3176 /* At least (most) this many matches must be made. */
3177 int lower_bound
= 0, upper_bound
= -1;
3182 FREE_STACK_RETURN (REG_EBRACE
);
3184 GET_UNSIGNED_NUMBER (lower_bound
);
3187 GET_UNSIGNED_NUMBER (upper_bound
);
3189 /* Interval such as `{1}' => match exactly once. */
3190 upper_bound
= lower_bound
;
3192 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3193 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3194 FREE_STACK_RETURN (REG_BADBR
);
3196 if (!(syntax
& RE_NO_BK_BRACES
))
3199 FREE_STACK_RETURN (REG_BADBR
);
3205 FREE_STACK_RETURN (REG_BADBR
);
3207 /* We just parsed a valid interval. */
3209 /* If it's invalid to have no preceding re. */
3212 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3213 FREE_STACK_RETURN (REG_BADRPT
);
3214 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3217 goto unfetch_interval
;
3220 if (upper_bound
== 0)
3221 /* If the upper bound is zero, just drop the sub pattern
3224 else if (lower_bound
== 1 && upper_bound
== 1)
3225 /* Just match it once: nothing to do here. */
3228 /* Otherwise, we have a nontrivial interval. When
3229 we're all done, the pattern will look like:
3230 set_number_at <jump count> <upper bound>
3231 set_number_at <succeed_n count> <lower bound>
3232 succeed_n <after jump addr> <succeed_n count>
3234 jump_n <succeed_n addr> <jump count>
3235 (The upper bound and `jump_n' are omitted if
3236 `upper_bound' is 1, though.) */
3238 { /* If the upper bound is > 1, we need to insert
3239 more at the end of the loop. */
3240 unsigned int nbytes
= (upper_bound
< 0 ? 3
3241 : upper_bound
> 1 ? 5 : 0);
3242 unsigned int startoffset
= 0;
3244 GET_BUFFER_SPACE (20); /* We might use less. */
3246 if (lower_bound
== 0)
3248 /* A succeed_n that starts with 0 is really a
3249 a simple on_failure_jump_loop. */
3250 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3256 /* Initialize lower bound of the `succeed_n', even
3257 though it will be set during matching by its
3258 attendant `set_number_at' (inserted next),
3259 because `re_compile_fastmap' needs to know.
3260 Jump to the `jump_n' we might insert below. */
3261 INSERT_JUMP2 (succeed_n
, laststart
,
3266 /* Code to initialize the lower bound. Insert
3267 before the `succeed_n'. The `5' is the last two
3268 bytes of this `set_number_at', plus 3 bytes of
3269 the following `succeed_n'. */
3270 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3275 if (upper_bound
< 0)
3277 /* A negative upper bound stands for infinity,
3278 in which case it degenerates to a plain jump. */
3279 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3282 else if (upper_bound
> 1)
3283 { /* More than one repetition is allowed, so
3284 append a backward jump to the `succeed_n'
3285 that starts this interval.
3287 When we've reached this during matching,
3288 we'll have matched the interval once, so
3289 jump back only `upper_bound - 1' times. */
3290 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3294 /* The location we want to set is the second
3295 parameter of the `jump_n'; that is `b-2' as
3296 an absolute address. `laststart' will be
3297 the `set_number_at' we're about to insert;
3298 `laststart+3' the number to set, the source
3299 for the relative address. But we are
3300 inserting into the middle of the pattern --
3301 so everything is getting moved up by 5.
3302 Conclusion: (b - 2) - (laststart + 3) + 5,
3303 i.e., b - laststart.
3305 We insert this at the beginning of the loop
3306 so that if we fail during matching, we'll
3307 reinitialize the bounds. */
3308 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3309 upper_bound
- 1, b
);
3314 beg_interval
= NULL
;
3319 /* If an invalid interval, match the characters as literals. */
3320 assert (beg_interval
);
3322 beg_interval
= NULL
;
3324 /* normal_char and normal_backslash need `c'. */
3327 if (!(syntax
& RE_NO_BK_BRACES
))
3329 assert (p
> pattern
&& p
[-1] == '\\');
3330 goto normal_backslash
;
3336 /* There is no way to specify the before_dot and after_dot
3337 operators. rms says this is ok. --karl */
3345 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3351 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3357 BUF_PUSH_2 (categoryspec
, c
);
3363 BUF_PUSH_2 (notcategoryspec
, c
);
3369 if (syntax
& RE_NO_GNU_OPS
)
3372 BUF_PUSH_2 (syntaxspec
, Sword
);
3377 if (syntax
& RE_NO_GNU_OPS
)
3380 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3385 if (syntax
& RE_NO_GNU_OPS
)
3391 if (syntax
& RE_NO_GNU_OPS
)
3397 if (syntax
& RE_NO_GNU_OPS
)
3399 BUF_PUSH (wordbound
);
3403 if (syntax
& RE_NO_GNU_OPS
)
3405 BUF_PUSH (notwordbound
);
3409 if (syntax
& RE_NO_GNU_OPS
)
3415 if (syntax
& RE_NO_GNU_OPS
)
3420 case '1': case '2': case '3': case '4': case '5':
3421 case '6': case '7': case '8': case '9':
3425 if (syntax
& RE_NO_BK_REFS
)
3426 goto normal_backslash
;
3430 /* Can't back reference to a subexpression before its end. */
3431 if (reg
> regnum
|| group_in_compile_stack (compile_stack
, reg
))
3432 FREE_STACK_RETURN (REG_ESUBREG
);
3435 BUF_PUSH_2 (duplicate
, reg
);
3442 if (syntax
& RE_BK_PLUS_QM
)
3445 goto normal_backslash
;
3449 /* You might think it would be useful for \ to mean
3450 not to translate; but if we don't translate it
3451 it will never match anything. */
3458 /* Expects the character in `c'. */
3460 /* If no exactn currently being built. */
3463 /* If last exactn not at current position. */
3464 || pending_exact
+ *pending_exact
+ 1 != b
3466 /* We have only one byte following the exactn for the count. */
3467 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3469 /* If followed by a repetition operator. */
3470 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3471 || ((syntax
& RE_BK_PLUS_QM
)
3472 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3473 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3474 || ((syntax
& RE_INTERVALS
)
3475 && ((syntax
& RE_NO_BK_BRACES
)
3476 ? p
!= pend
&& *p
== '{'
3477 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3479 /* Start building a new exactn. */
3483 BUF_PUSH_2 (exactn
, 0);
3484 pending_exact
= b
- 1;
3487 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3493 len
= CHAR_STRING (c
, b
);
3497 (*pending_exact
) += len
;
3502 } /* while p != pend */
3505 /* Through the pattern now. */
3509 if (!COMPILE_STACK_EMPTY
)
3510 FREE_STACK_RETURN (REG_EPAREN
);
3512 /* If we don't want backtracking, force success
3513 the first time we reach the end of the compiled pattern. */
3514 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3517 /* We have succeeded; set the length of the buffer. */
3518 bufp
->used
= b
- bufp
->buffer
;
3523 re_compile_fastmap (bufp
);
3524 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3525 print_compiled_pattern (bufp
);
3530 #ifndef MATCH_MAY_ALLOCATE
3531 /* Initialize the failure stack to the largest possible stack. This
3532 isn't necessary unless we're trying to avoid calling alloca in
3533 the search and match routines. */
3535 int num_regs
= bufp
->re_nsub
+ 1;
3537 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3539 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3541 if (! fail_stack
.stack
)
3543 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3544 * sizeof (fail_stack_elt_t
));
3547 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3549 * sizeof (fail_stack_elt_t
)));
3552 regex_grow_registers (num_regs
);
3554 #endif /* not MATCH_MAY_ALLOCATE */
3556 FREE_STACK_RETURN (REG_NOERROR
);
3557 } /* regex_compile */
3559 /* Subroutines for `regex_compile'. */
3561 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3564 store_op1 (op
, loc
, arg
)
3569 *loc
= (unsigned char) op
;
3570 STORE_NUMBER (loc
+ 1, arg
);
3574 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3577 store_op2 (op
, loc
, arg1
, arg2
)
3582 *loc
= (unsigned char) op
;
3583 STORE_NUMBER (loc
+ 1, arg1
);
3584 STORE_NUMBER (loc
+ 3, arg2
);
3588 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3589 for OP followed by two-byte integer parameter ARG. */
3592 insert_op1 (op
, loc
, arg
, end
)
3598 register unsigned char *pfrom
= end
;
3599 register unsigned char *pto
= end
+ 3;
3601 while (pfrom
!= loc
)
3604 store_op1 (op
, loc
, arg
);
3608 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3611 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3617 register unsigned char *pfrom
= end
;
3618 register unsigned char *pto
= end
+ 5;
3620 while (pfrom
!= loc
)
3623 store_op2 (op
, loc
, arg1
, arg2
);
3627 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3628 after an alternative or a begin-subexpression. We assume there is at
3629 least one character before the ^. */
3632 at_begline_loc_p (pattern
, p
, syntax
)
3633 re_char
*pattern
, *p
;
3634 reg_syntax_t syntax
;
3636 re_char
*prev
= p
- 2;
3637 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3640 /* After a subexpression? */
3641 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3642 /* After an alternative? */
3643 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3644 /* After a shy subexpression? */
3645 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3646 && prev
[-1] == '?' && prev
[-2] == '('
3647 && (syntax
& RE_NO_BK_PARENS
3648 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3652 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3653 at least one character after the $, i.e., `P < PEND'. */
3656 at_endline_loc_p (p
, pend
, syntax
)
3658 reg_syntax_t syntax
;
3661 boolean next_backslash
= *next
== '\\';
3662 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3665 /* Before a subexpression? */
3666 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3667 : next_backslash
&& next_next
&& *next_next
== ')')
3668 /* Before an alternative? */
3669 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3670 : next_backslash
&& next_next
&& *next_next
== '|');
3674 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3675 false if it's not. */
3678 group_in_compile_stack (compile_stack
, regnum
)
3679 compile_stack_type compile_stack
;
3684 for (this_element
= compile_stack
.avail
- 1;
3687 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3694 If fastmap is non-NULL, go through the pattern and fill fastmap
3695 with all the possible leading chars. If fastmap is NULL, don't
3696 bother filling it up (obviously) and only return whether the
3697 pattern could potentially match the empty string.
3699 Return 1 if p..pend might match the empty string.
3700 Return 0 if p..pend matches at least one char.
3701 Return -1 if fastmap was not updated accurately. */
3704 analyse_first (p
, pend
, fastmap
, multibyte
)
3707 const int multibyte
;
3712 /* If all elements for base leading-codes in fastmap is set, this
3713 flag is set true. */
3714 boolean match_any_multibyte_characters
= false;
3718 /* The loop below works as follows:
3719 - It has a working-list kept in the PATTERN_STACK and which basically
3720 starts by only containing a pointer to the first operation.
3721 - If the opcode we're looking at is a match against some set of
3722 chars, then we add those chars to the fastmap and go on to the
3723 next work element from the worklist (done via `break').
3724 - If the opcode is a control operator on the other hand, we either
3725 ignore it (if it's meaningless at this point, such as `start_memory')
3726 or execute it (if it's a jump). If the jump has several destinations
3727 (i.e. `on_failure_jump'), then we push the other destination onto the
3729 We guarantee termination by ignoring backward jumps (more or less),
3730 so that `p' is monotonically increasing. More to the point, we
3731 never set `p' (or push) anything `<= p1'. */
3735 /* `p1' is used as a marker of how far back a `on_failure_jump'
3736 can go without being ignored. It is normally equal to `p'
3737 (which prevents any backward `on_failure_jump') except right
3738 after a plain `jump', to allow patterns such as:
3741 10: on_failure_jump 3
3742 as used for the *? operator. */
3745 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3752 /* If the first character has to match a backreference, that means
3753 that the group was empty (since it already matched). Since this
3754 is the only case that interests us here, we can assume that the
3755 backreference must match the empty string. */
3760 /* Following are the cases which match a character. These end
3766 int c
= RE_STRING_CHAR (p
+ 1, pend
- p
);
3768 if (SINGLE_BYTE_CHAR_P (c
))
3777 /* We could put all the chars except for \n (and maybe \0)
3778 but we don't bother since it is generally not worth it. */
3779 if (!fastmap
) break;
3784 /* Chars beyond end of bitmap are possible matches.
3785 All the single-byte codes can occur in multibyte buffers.
3786 So any that are not listed in the charset
3787 are possible matches, even in multibyte buffers. */
3788 if (!fastmap
) break;
3789 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3790 j
< (1 << BYTEWIDTH
); j
++)
3794 if (!fastmap
) break;
3795 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3796 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3798 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3801 if ((not && multibyte
)
3802 /* Any character set can possibly contain a character
3803 which doesn't match the specified set of characters. */
3804 || (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3805 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3806 /* If we can match a character class, we can match
3807 any character set. */
3809 set_fastmap_for_multibyte_characters
:
3810 if (match_any_multibyte_characters
== false)
3812 for (j
= 0x80; j
< 0xA0; j
++) /* XXX */
3813 if (BASE_LEADING_CODE_P (j
))
3815 match_any_multibyte_characters
= true;
3819 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3820 && match_any_multibyte_characters
== false)
3822 /* Set fastmap[I] 1 where I is a base leading code of each
3823 multibyte character in the range table. */
3826 /* Make P points the range table. `+ 2' is to skip flag
3827 bits for a character class. */
3828 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3830 /* Extract the number of ranges in range table into COUNT. */
3831 EXTRACT_NUMBER_AND_INCR (count
, p
);
3832 for (; count
> 0; count
--, p
+= 2 * 3) /* XXX */
3834 /* Extract the start of each range. */
3835 EXTRACT_CHARACTER (c
, p
);
3836 j
= CHAR_CHARSET (c
);
3837 fastmap
[CHARSET_LEADING_CODE_BASE (j
)] = 1;
3844 if (!fastmap
) break;
3846 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
3848 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3849 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
3853 /* This match depends on text properties. These end with
3854 aborting optimizations. */
3858 case notcategoryspec
:
3859 if (!fastmap
) break;
3860 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
3862 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3863 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
3867 /* Any character set can possibly contain a character
3868 whose category is K (or not). */
3869 goto set_fastmap_for_multibyte_characters
;
3872 /* All cases after this match the empty string. These end with
3892 EXTRACT_NUMBER_AND_INCR (j
, p
);
3894 /* Backward jumps can only go back to code that we've already
3895 visited. `re_compile' should make sure this is true. */
3898 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
3900 case on_failure_jump
:
3901 case on_failure_keep_string_jump
:
3902 case on_failure_jump_loop
:
3903 case on_failure_jump_nastyloop
:
3904 case on_failure_jump_smart
:
3910 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
3911 to jump back to "just after here". */
3914 case on_failure_jump
:
3915 case on_failure_keep_string_jump
:
3916 case on_failure_jump_nastyloop
:
3917 case on_failure_jump_loop
:
3918 case on_failure_jump_smart
:
3919 EXTRACT_NUMBER_AND_INCR (j
, p
);
3921 ; /* Backward jump to be ignored. */
3923 { /* We have to look down both arms.
3924 We first go down the "straight" path so as to minimize
3925 stack usage when going through alternatives. */
3926 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
3934 /* This code simply does not properly handle forward jump_n. */
3935 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
3937 /* jump_n can either jump or fall through. The (backward) jump
3938 case has already been handled, so we only need to look at the
3939 fallthrough case. */
3943 /* If N == 0, it should be an on_failure_jump_loop instead. */
3944 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
3946 /* We only care about one iteration of the loop, so we don't
3947 need to consider the case where this behaves like an
3964 abort (); /* We have listed all the cases. */
3967 /* Getting here means we have found the possible starting
3968 characters for one path of the pattern -- and that the empty
3969 string does not match. We need not follow this path further. */
3973 /* We reached the end without matching anything. */
3976 } /* analyse_first */
3978 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3979 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3980 characters can start a string that matches the pattern. This fastmap
3981 is used by re_search to skip quickly over impossible starting points.
3983 Character codes above (1 << BYTEWIDTH) are not represented in the
3984 fastmap, but the leading codes are represented. Thus, the fastmap
3985 indicates which character sets could start a match.
3987 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3988 area as BUFP->fastmap.
3990 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3993 Returns 0 if we succeed, -2 if an internal error. */
3996 re_compile_fastmap (bufp
)
3997 struct re_pattern_buffer
*bufp
;
3999 char *fastmap
= bufp
->fastmap
;
4002 assert (fastmap
&& bufp
->buffer
);
4004 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4005 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4007 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4008 fastmap
, RE_MULTIBYTE_P (bufp
));
4009 bufp
->can_be_null
= (analysis
!= 0);
4011 } /* re_compile_fastmap */
4013 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4014 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4015 this memory for recording register information. STARTS and ENDS
4016 must be allocated using the malloc library routine, and must each
4017 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4019 If NUM_REGS == 0, then subsequent matches should allocate their own
4022 Unless this function is called, the first search or match using
4023 PATTERN_BUFFER will allocate its own register data, without
4024 freeing the old data. */
4027 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4028 struct re_pattern_buffer
*bufp
;
4029 struct re_registers
*regs
;
4031 regoff_t
*starts
, *ends
;
4035 bufp
->regs_allocated
= REGS_REALLOCATE
;
4036 regs
->num_regs
= num_regs
;
4037 regs
->start
= starts
;
4042 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4044 regs
->start
= regs
->end
= (regoff_t
*) 0;
4047 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4049 /* Searching routines. */
4051 /* Like re_search_2, below, but only one string is specified, and
4052 doesn't let you say where to stop matching. */
4055 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4056 struct re_pattern_buffer
*bufp
;
4058 int size
, startpos
, range
;
4059 struct re_registers
*regs
;
4061 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4064 WEAK_ALIAS (__re_search
, re_search
)
4066 /* Head address of virtual concatenation of string. */
4067 #define HEAD_ADDR_VSTRING(P) \
4068 (((P) >= size1 ? string2 : string1))
4070 /* End address of virtual concatenation of string. */
4071 #define STOP_ADDR_VSTRING(P) \
4072 (((P) >= size1 ? string2 + size2 : string1 + size1))
4074 /* Address of POS in the concatenation of virtual string. */
4075 #define POS_ADDR_VSTRING(POS) \
4076 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4078 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4079 virtual concatenation of STRING1 and STRING2, starting first at index
4080 STARTPOS, then at STARTPOS + 1, and so on.
4082 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4084 RANGE is how far to scan while trying to match. RANGE = 0 means try
4085 only at STARTPOS; in general, the last start tried is STARTPOS +
4088 In REGS, return the indices of the virtual concatenation of STRING1
4089 and STRING2 that matched the entire BUFP->buffer and its contained
4092 Do not consider matching one past the index STOP in the virtual
4093 concatenation of STRING1 and STRING2.
4095 We return either the position in the strings at which the match was
4096 found, -1 if no match, or -2 if error (such as failure
4100 re_search_2 (bufp
, str1
, size1
, str2
, size2
, startpos
, range
, regs
, stop
)
4101 struct re_pattern_buffer
*bufp
;
4102 const char *str1
, *str2
;
4106 struct re_registers
*regs
;
4110 re_char
*string1
= (re_char
*) str1
;
4111 re_char
*string2
= (re_char
*) str2
;
4112 register char *fastmap
= bufp
->fastmap
;
4113 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4114 int total_size
= size1
+ size2
;
4115 int endpos
= startpos
+ range
;
4116 boolean anchored_start
;
4118 /* Nonzero if we have to concern multibyte character. */
4119 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4121 /* Check for out-of-range STARTPOS. */
4122 if (startpos
< 0 || startpos
> total_size
)
4125 /* Fix up RANGE if it might eventually take us outside
4126 the virtual concatenation of STRING1 and STRING2.
4127 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4129 range
= 0 - startpos
;
4130 else if (endpos
> total_size
)
4131 range
= total_size
- startpos
;
4133 /* If the search isn't to be a backwards one, don't waste time in a
4134 search for a pattern anchored at beginning of buffer. */
4135 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4144 /* In a forward search for something that starts with \=.
4145 don't keep searching past point. */
4146 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4148 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4154 /* Update the fastmap now if not correct already. */
4155 if (fastmap
&& !bufp
->fastmap_accurate
)
4156 re_compile_fastmap (bufp
);
4158 /* See whether the pattern is anchored. */
4159 anchored_start
= (bufp
->buffer
[0] == begline
);
4162 gl_state
.object
= re_match_object
;
4164 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4166 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4170 /* Loop through the string, looking for a place to start matching. */
4173 /* If the pattern is anchored,
4174 skip quickly past places we cannot match.
4175 We don't bother to treat startpos == 0 specially
4176 because that case doesn't repeat. */
4177 if (anchored_start
&& startpos
> 0)
4179 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4180 : string2
[startpos
- size1
- 1])
4185 /* If a fastmap is supplied, skip quickly over characters that
4186 cannot be the start of a match. If the pattern can match the
4187 null string, however, we don't need to skip characters; we want
4188 the first null string. */
4189 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4191 register re_char
*d
;
4192 register re_wchar_t buf_ch
;
4194 d
= POS_ADDR_VSTRING (startpos
);
4196 if (range
> 0) /* Searching forwards. */
4198 register int lim
= 0;
4201 if (startpos
< size1
&& startpos
+ range
>= size1
)
4202 lim
= range
- (size1
- startpos
);
4204 /* Written out as an if-else to avoid testing `translate'
4206 if (RE_TRANSLATE_P (translate
))
4213 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4216 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4221 range
-= buf_charlen
;
4226 && !fastmap
[RE_TRANSLATE (translate
, *d
)])
4233 while (range
> lim
&& !fastmap
[*d
])
4239 startpos
+= irange
- range
;
4241 else /* Searching backwards. */
4243 int room
= (startpos
>= size1
4244 ? size2
+ size1
- startpos
4245 : size1
- startpos
);
4246 buf_ch
= RE_STRING_CHAR (d
, room
);
4247 buf_ch
= TRANSLATE (buf_ch
);
4249 if (! (buf_ch
>= 0400
4250 || fastmap
[buf_ch
]))
4255 /* If can't match the null string, and that's all we have left, fail. */
4256 if (range
>= 0 && startpos
== total_size
&& fastmap
4257 && !bufp
->can_be_null
)
4260 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4261 startpos
, regs
, stop
);
4262 #ifndef REGEX_MALLOC
4279 /* Update STARTPOS to the next character boundary. */
4282 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4283 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4284 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
4302 /* Update STARTPOS to the previous character boundary. */
4305 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4307 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4309 /* Find the head of multibyte form. */
4310 PREV_CHAR_BOUNDARY (p
, phead
);
4311 range
+= p0
- 1 - p
;
4315 startpos
-= p0
- 1 - p
;
4321 WEAK_ALIAS (__re_search_2
, re_search_2
)
4323 /* Declarations and macros for re_match_2. */
4325 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4327 RE_TRANSLATE_TYPE translate
,
4328 const int multibyte
));
4330 /* This converts PTR, a pointer into one of the search strings `string1'
4331 and `string2' into an offset from the beginning of that string. */
4332 #define POINTER_TO_OFFSET(ptr) \
4333 (FIRST_STRING_P (ptr) \
4334 ? ((regoff_t) ((ptr) - string1)) \
4335 : ((regoff_t) ((ptr) - string2 + size1)))
4337 /* Call before fetching a character with *d. This switches over to
4338 string2 if necessary.
4339 Check re_match_2_internal for a discussion of why end_match_2 might
4340 not be within string2 (but be equal to end_match_1 instead). */
4341 #define PREFETCH() \
4344 /* End of string2 => fail. */ \
4345 if (dend == end_match_2) \
4347 /* End of string1 => advance to string2. */ \
4349 dend = end_match_2; \
4352 /* Call before fetching a char with *d if you already checked other limits.
4353 This is meant for use in lookahead operations like wordend, etc..
4354 where we might need to look at parts of the string that might be
4355 outside of the LIMITs (i.e past `stop'). */
4356 #define PREFETCH_NOLIMIT() \
4360 dend = end_match_2; \
4363 /* Test if at very beginning or at very end of the virtual concatenation
4364 of `string1' and `string2'. If only one string, it's `string2'. */
4365 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4366 #define AT_STRINGS_END(d) ((d) == end2)
4369 /* Test if D points to a character which is word-constituent. We have
4370 two special cases to check for: if past the end of string1, look at
4371 the first character in string2; and if before the beginning of
4372 string2, look at the last character in string1. */
4373 #define WORDCHAR_P(d) \
4374 (SYNTAX ((d) == end1 ? *string2 \
4375 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4378 /* Disabled due to a compiler bug -- see comment at case wordbound */
4380 /* The comment at case wordbound is following one, but we don't use
4381 AT_WORD_BOUNDARY anymore to support multibyte form.
4383 The DEC Alpha C compiler 3.x generates incorrect code for the
4384 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4385 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4386 macro and introducing temporary variables works around the bug. */
4389 /* Test if the character before D and the one at D differ with respect
4390 to being word-constituent. */
4391 #define AT_WORD_BOUNDARY(d) \
4392 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4393 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4396 /* Free everything we malloc. */
4397 #ifdef MATCH_MAY_ALLOCATE
4398 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4399 # define FREE_VARIABLES() \
4401 REGEX_FREE_STACK (fail_stack.stack); \
4402 FREE_VAR (regstart); \
4403 FREE_VAR (regend); \
4404 FREE_VAR (best_regstart); \
4405 FREE_VAR (best_regend); \
4408 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4409 #endif /* not MATCH_MAY_ALLOCATE */
4412 /* Optimization routines. */
4414 /* If the operation is a match against one or more chars,
4415 return a pointer to the next operation, else return NULL. */
4420 switch (SWITCH_ENUM_CAST (*p
++))
4431 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4434 p
= CHARSET_RANGE_TABLE (p
- 1);
4435 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4436 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4439 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4446 case notcategoryspec
:
4458 /* Jump over non-matching operations. */
4459 static unsigned char *
4460 skip_noops (p
, pend
)
4461 unsigned char *p
, *pend
;
4466 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4475 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4486 /* Non-zero if "p1 matches something" implies "p2 fails". */
4488 mutually_exclusive_p (bufp
, p1
, p2
)
4489 struct re_pattern_buffer
*bufp
;
4490 unsigned char *p1
, *p2
;
4493 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4494 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4496 assert (p1
>= bufp
->buffer
&& p1
< pend
4497 && p2
>= bufp
->buffer
&& p2
<= pend
);
4499 /* Skip over open/close-group commands.
4500 If what follows this loop is a ...+ construct,
4501 look at what begins its body, since we will have to
4502 match at least one of that. */
4503 p2
= skip_noops (p2
, pend
);
4504 /* The same skip can be done for p1, except that this function
4505 is only used in the case where p1 is a simple match operator. */
4506 /* p1 = skip_noops (p1, pend); */
4508 assert (p1
>= bufp
->buffer
&& p1
< pend
4509 && p2
>= bufp
->buffer
&& p2
<= pend
);
4511 op2
= p2
== pend
? succeed
: *p2
;
4513 switch (SWITCH_ENUM_CAST (op2
))
4517 /* If we're at the end of the pattern, we can change. */
4518 if (skip_one_char (p1
))
4520 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4528 register re_wchar_t c
4529 = (re_opcode_t
) *p2
== endline
? '\n'
4530 : RE_STRING_CHAR (p2
+ 2, pend
- p2
- 2);
4532 if ((re_opcode_t
) *p1
== exactn
)
4534 if (c
!= RE_STRING_CHAR (p1
+ 2, pend
- p1
- 2))
4536 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4541 else if ((re_opcode_t
) *p1
== charset
4542 || (re_opcode_t
) *p1
== charset_not
)
4544 int not = (re_opcode_t
) *p1
== charset_not
;
4546 /* Test if C is listed in charset (or charset_not)
4548 if (SINGLE_BYTE_CHAR_P (c
))
4550 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4551 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4554 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4555 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4557 /* `not' is equal to 1 if c would match, which means
4558 that we can't change to pop_failure_jump. */
4561 DEBUG_PRINT1 (" No match => fast loop.\n");
4565 else if ((re_opcode_t
) *p1
== anychar
4568 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4576 if ((re_opcode_t
) *p1
== exactn
)
4577 /* Reuse the code above. */
4578 return mutually_exclusive_p (bufp
, p2
, p1
);
4580 /* It is hard to list up all the character in charset
4581 P2 if it includes multibyte character. Give up in
4583 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4585 /* Now, we are sure that P2 has no range table.
4586 So, for the size of bitmap in P2, `p2[1]' is
4587 enough. But P1 may have range table, so the
4588 size of bitmap table of P1 is extracted by
4589 using macro `CHARSET_BITMAP_SIZE'.
4591 Since we know that all the character listed in
4592 P2 is ASCII, it is enough to test only bitmap
4595 if ((re_opcode_t
) *p1
== charset
)
4598 /* We win if the charset inside the loop
4599 has no overlap with the one after the loop. */
4602 && idx
< CHARSET_BITMAP_SIZE (p1
));
4604 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4608 || idx
== CHARSET_BITMAP_SIZE (p1
))
4610 DEBUG_PRINT1 (" No match => fast loop.\n");
4614 else if ((re_opcode_t
) *p1
== charset_not
)
4617 /* We win if the charset_not inside the loop lists
4618 every character listed in the charset after. */
4619 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4620 if (! (p2
[2 + idx
] == 0
4621 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4622 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4627 DEBUG_PRINT1 (" No match => fast loop.\n");
4636 switch (SWITCH_ENUM_CAST (*p1
))
4640 /* Reuse the code above. */
4641 return mutually_exclusive_p (bufp
, p2
, p1
);
4643 /* When we have two charset_not, it's very unlikely that
4644 they don't overlap. The union of the two sets of excluded
4645 chars should cover all possible chars, which, as a matter of
4646 fact, is virtually impossible in multibyte buffers. */
4653 return ((re_opcode_t
) *p1
== syntaxspec
4654 && p1
[1] == (op2
== wordend
? Sword
: p2
[1]));
4658 return ((re_opcode_t
) *p1
== notsyntaxspec
4659 && p1
[1] == (op2
== wordend
? Sword
: p2
[1]));
4662 return (((re_opcode_t
) *p1
== notsyntaxspec
4663 || (re_opcode_t
) *p1
== syntaxspec
)
4668 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4669 case notcategoryspec
:
4670 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4682 /* Matching routines. */
4684 #ifndef emacs /* Emacs never uses this. */
4685 /* re_match is like re_match_2 except it takes only a single string. */
4688 re_match (bufp
, string
, size
, pos
, regs
)
4689 struct re_pattern_buffer
*bufp
;
4692 struct re_registers
*regs
;
4694 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
4696 # if defined C_ALLOCA && !defined REGEX_MALLOC
4701 WEAK_ALIAS (__re_match
, re_match
)
4702 #endif /* not emacs */
4705 /* In Emacs, this is the string or buffer in which we
4706 are matching. It is used for looking up syntax properties. */
4707 Lisp_Object re_match_object
;
4710 /* re_match_2 matches the compiled pattern in BUFP against the
4711 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4712 and SIZE2, respectively). We start matching at POS, and stop
4715 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4716 store offsets for the substring each group matched in REGS. See the
4717 documentation for exactly how many groups we fill.
4719 We return -1 if no match, -2 if an internal error (such as the
4720 failure stack overflowing). Otherwise, we return the length of the
4721 matched substring. */
4724 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4725 struct re_pattern_buffer
*bufp
;
4726 const char *string1
, *string2
;
4729 struct re_registers
*regs
;
4736 gl_state
.object
= re_match_object
;
4737 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4738 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4741 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4742 (re_char
*) string2
, size2
,
4744 #if defined C_ALLOCA && !defined REGEX_MALLOC
4749 WEAK_ALIAS (__re_match_2
, re_match_2
)
4751 /* This is a separate function so that we can force an alloca cleanup
4754 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4755 struct re_pattern_buffer
*bufp
;
4756 re_char
*string1
, *string2
;
4759 struct re_registers
*regs
;
4762 /* General temporaries. */
4767 /* Just past the end of the corresponding string. */
4768 re_char
*end1
, *end2
;
4770 /* Pointers into string1 and string2, just past the last characters in
4771 each to consider matching. */
4772 re_char
*end_match_1
, *end_match_2
;
4774 /* Where we are in the data, and the end of the current string. */
4777 /* Used sometimes to remember where we were before starting matching
4778 an operator so that we can go back in case of failure. This "atomic"
4779 behavior of matching opcodes is indispensable to the correctness
4780 of the on_failure_keep_string_jump optimization. */
4783 /* Where we are in the pattern, and the end of the pattern. */
4784 re_char
*p
= bufp
->buffer
;
4785 re_char
*pend
= p
+ bufp
->used
;
4787 /* We use this to map every character in the string. */
4788 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4790 /* Nonzero if we have to concern multibyte character. */
4791 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4793 /* Failure point stack. Each place that can handle a failure further
4794 down the line pushes a failure point on this stack. It consists of
4795 regstart, and regend for all registers corresponding to
4796 the subexpressions we're currently inside, plus the number of such
4797 registers, and, finally, two char *'s. The first char * is where
4798 to resume scanning the pattern; the second one is where to resume
4799 scanning the strings. */
4800 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4801 fail_stack_type fail_stack
;
4804 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4807 #if defined REL_ALLOC && defined REGEX_MALLOC
4808 /* This holds the pointer to the failure stack, when
4809 it is allocated relocatably. */
4810 fail_stack_elt_t
*failure_stack_ptr
;
4813 /* We fill all the registers internally, independent of what we
4814 return, for use in backreferences. The number here includes
4815 an element for register zero. */
4816 size_t num_regs
= bufp
->re_nsub
+ 1;
4818 /* Information on the contents of registers. These are pointers into
4819 the input strings; they record just what was matched (on this
4820 attempt) by a subexpression part of the pattern, that is, the
4821 regnum-th regstart pointer points to where in the pattern we began
4822 matching and the regnum-th regend points to right after where we
4823 stopped matching the regnum-th subexpression. (The zeroth register
4824 keeps track of what the whole pattern matches.) */
4825 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4826 re_char
**regstart
, **regend
;
4829 /* The following record the register info as found in the above
4830 variables when we find a match better than any we've seen before.
4831 This happens as we backtrack through the failure points, which in
4832 turn happens only if we have not yet matched the entire string. */
4833 unsigned best_regs_set
= false;
4834 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4835 re_char
**best_regstart
, **best_regend
;
4838 /* Logically, this is `best_regend[0]'. But we don't want to have to
4839 allocate space for that if we're not allocating space for anything
4840 else (see below). Also, we never need info about register 0 for
4841 any of the other register vectors, and it seems rather a kludge to
4842 treat `best_regend' differently than the rest. So we keep track of
4843 the end of the best match so far in a separate variable. We
4844 initialize this to NULL so that when we backtrack the first time
4845 and need to test it, it's not garbage. */
4846 re_char
*match_end
= NULL
;
4849 /* Counts the total number of registers pushed. */
4850 unsigned num_regs_pushed
= 0;
4853 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4857 #ifdef MATCH_MAY_ALLOCATE
4858 /* Do not bother to initialize all the register variables if there are
4859 no groups in the pattern, as it takes a fair amount of time. If
4860 there are groups, we include space for register 0 (the whole
4861 pattern), even though we never use it, since it simplifies the
4862 array indexing. We should fix this. */
4865 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
4866 regend
= REGEX_TALLOC (num_regs
, re_char
*);
4867 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
4868 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
4870 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
4878 /* We must initialize all our variables to NULL, so that
4879 `FREE_VARIABLES' doesn't try to free them. */
4880 regstart
= regend
= best_regstart
= best_regend
= NULL
;
4882 #endif /* MATCH_MAY_ALLOCATE */
4884 /* The starting position is bogus. */
4885 if (pos
< 0 || pos
> size1
+ size2
)
4891 /* Initialize subexpression text positions to -1 to mark ones that no
4892 start_memory/stop_memory has been seen for. Also initialize the
4893 register information struct. */
4894 for (reg
= 1; reg
< num_regs
; reg
++)
4895 regstart
[reg
] = regend
[reg
] = NULL
;
4897 /* We move `string1' into `string2' if the latter's empty -- but not if
4898 `string1' is null. */
4899 if (size2
== 0 && string1
!= NULL
)
4906 end1
= string1
+ size1
;
4907 end2
= string2
+ size2
;
4909 /* `p' scans through the pattern as `d' scans through the data.
4910 `dend' is the end of the input string that `d' points within. `d'
4911 is advanced into the following input string whenever necessary, but
4912 this happens before fetching; therefore, at the beginning of the
4913 loop, `d' can be pointing at the end of a string, but it cannot
4917 /* Only match within string2. */
4918 d
= string2
+ pos
- size1
;
4919 dend
= end_match_2
= string2
+ stop
- size1
;
4920 end_match_1
= end1
; /* Just to give it a value. */
4926 /* Only match within string1. */
4927 end_match_1
= string1
+ stop
;
4929 When we reach end_match_1, PREFETCH normally switches to string2.
4930 But in the present case, this means that just doing a PREFETCH
4931 makes us jump from `stop' to `gap' within the string.
4932 What we really want here is for the search to stop as
4933 soon as we hit end_match_1. That's why we set end_match_2
4934 to end_match_1 (since PREFETCH fails as soon as we hit
4936 end_match_2
= end_match_1
;
4939 { /* It's important to use this code when stop == size so that
4940 moving `d' from end1 to string2 will not prevent the d == dend
4941 check from catching the end of string. */
4943 end_match_2
= string2
+ stop
- size1
;
4949 DEBUG_PRINT1 ("The compiled pattern is: ");
4950 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
4951 DEBUG_PRINT1 ("The string to match is: `");
4952 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
4953 DEBUG_PRINT1 ("'\n");
4955 /* This loops over pattern commands. It exits by returning from the
4956 function if the match is complete, or it drops through if the match
4957 fails at this starting point in the input data. */
4960 DEBUG_PRINT2 ("\n%p: ", p
);
4963 { /* End of pattern means we might have succeeded. */
4964 DEBUG_PRINT1 ("end of pattern ... ");
4966 /* If we haven't matched the entire string, and we want the
4967 longest match, try backtracking. */
4968 if (d
!= end_match_2
)
4970 /* 1 if this match ends in the same string (string1 or string2)
4971 as the best previous match. */
4972 boolean same_str_p
= (FIRST_STRING_P (match_end
)
4973 == FIRST_STRING_P (d
));
4974 /* 1 if this match is the best seen so far. */
4975 boolean best_match_p
;
4977 /* AIX compiler got confused when this was combined
4978 with the previous declaration. */
4980 best_match_p
= d
> match_end
;
4982 best_match_p
= !FIRST_STRING_P (d
);
4984 DEBUG_PRINT1 ("backtracking.\n");
4986 if (!FAIL_STACK_EMPTY ())
4987 { /* More failure points to try. */
4989 /* If exceeds best match so far, save it. */
4990 if (!best_regs_set
|| best_match_p
)
4992 best_regs_set
= true;
4995 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4997 for (reg
= 1; reg
< num_regs
; reg
++)
4999 best_regstart
[reg
] = regstart
[reg
];
5000 best_regend
[reg
] = regend
[reg
];
5006 /* If no failure points, don't restore garbage. And if
5007 last match is real best match, don't restore second
5009 else if (best_regs_set
&& !best_match_p
)
5012 /* Restore best match. It may happen that `dend ==
5013 end_match_1' while the restored d is in string2.
5014 For example, the pattern `x.*y.*z' against the
5015 strings `x-' and `y-z-', if the two strings are
5016 not consecutive in memory. */
5017 DEBUG_PRINT1 ("Restoring best registers.\n");
5020 dend
= ((d
>= string1
&& d
<= end1
)
5021 ? end_match_1
: end_match_2
);
5023 for (reg
= 1; reg
< num_regs
; reg
++)
5025 regstart
[reg
] = best_regstart
[reg
];
5026 regend
[reg
] = best_regend
[reg
];
5029 } /* d != end_match_2 */
5032 DEBUG_PRINT1 ("Accepting match.\n");
5034 /* If caller wants register contents data back, do it. */
5035 if (regs
&& !bufp
->no_sub
)
5037 /* Have the register data arrays been allocated? */
5038 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5039 { /* No. So allocate them with malloc. We need one
5040 extra element beyond `num_regs' for the `-1' marker
5042 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5043 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5044 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5045 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5050 bufp
->regs_allocated
= REGS_REALLOCATE
;
5052 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5053 { /* Yes. If we need more elements than were already
5054 allocated, reallocate them. If we need fewer, just
5056 if (regs
->num_regs
< num_regs
+ 1)
5058 regs
->num_regs
= num_regs
+ 1;
5059 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5060 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5061 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5070 /* These braces fend off a "empty body in an else-statement"
5071 warning under GCC when assert expands to nothing. */
5072 assert (bufp
->regs_allocated
== REGS_FIXED
);
5075 /* Convert the pointer data in `regstart' and `regend' to
5076 indices. Register zero has to be set differently,
5077 since we haven't kept track of any info for it. */
5078 if (regs
->num_regs
> 0)
5080 regs
->start
[0] = pos
;
5081 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5084 /* Go through the first `min (num_regs, regs->num_regs)'
5085 registers, since that is all we initialized. */
5086 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5088 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5089 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5093 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5095 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5099 /* If the regs structure we return has more elements than
5100 were in the pattern, set the extra elements to -1. If
5101 we (re)allocated the registers, this is the case,
5102 because we always allocate enough to have at least one
5104 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5105 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5106 } /* regs && !bufp->no_sub */
5108 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5109 nfailure_points_pushed
, nfailure_points_popped
,
5110 nfailure_points_pushed
- nfailure_points_popped
);
5111 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5113 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5115 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5121 /* Otherwise match next pattern command. */
5122 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5124 /* Ignore these. Used to ignore the n of succeed_n's which
5125 currently have n == 0. */
5127 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5131 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5134 /* Match the next n pattern characters exactly. The following
5135 byte in the pattern defines n, and the n bytes after that
5136 are the characters to match. */
5139 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5141 /* Remember the start point to rollback upon failure. */
5144 /* This is written out as an if-else so we don't waste time
5145 testing `translate' inside the loop. */
5146 if (RE_TRANSLATE_P (translate
))
5151 int pat_charlen
, buf_charlen
;
5152 unsigned int pat_ch
, buf_ch
;
5155 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
5156 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5158 if (RE_TRANSLATE (translate
, buf_ch
)
5167 mcnt
-= pat_charlen
;
5174 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5199 /* Match any character except possibly a newline or a null. */
5205 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5208 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5209 buf_ch
= TRANSLATE (buf_ch
);
5211 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5213 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5214 && buf_ch
== '\000'))
5217 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5226 register unsigned int c
;
5227 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5230 /* Start of actual range_table, or end of bitmap if there is no
5232 re_char
*range_table
;
5234 /* Nonzero if there is a range table. */
5235 int range_table_exists
;
5237 /* Number of ranges of range table. This is not included
5238 in the initial byte-length of the command. */
5241 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5243 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5245 if (range_table_exists
)
5247 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5248 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5252 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5253 c
= TRANSLATE (c
); /* The character to match. */
5255 if (SINGLE_BYTE_CHAR_P (c
))
5256 { /* Lookup bitmap. */
5257 /* Cast to `unsigned' instead of `unsigned char' in
5258 case the bit list is a full 32 bytes long. */
5259 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5260 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5264 else if (range_table_exists
)
5266 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5268 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5269 | (class_bits
& BIT_MULTIBYTE
)
5270 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5271 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5272 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5273 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5276 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5280 if (range_table_exists
)
5281 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5283 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5285 if (!not) goto fail
;
5292 /* The beginning of a group is represented by start_memory.
5293 The argument is the register number. The text
5294 matched within the group is recorded (in the internal
5295 registers data structure) under the register number. */
5297 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5299 /* In case we need to undo this operation (via backtracking). */
5300 PUSH_FAILURE_REG ((unsigned int)*p
);
5303 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5304 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5306 /* Move past the register number and inner group count. */
5311 /* The stop_memory opcode represents the end of a group. Its
5312 argument is the same as start_memory's: the register number. */
5314 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5316 assert (!REG_UNSET (regstart
[*p
]));
5317 /* Strictly speaking, there should be code such as:
5319 assert (REG_UNSET (regend[*p]));
5320 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5322 But the only info to be pushed is regend[*p] and it is known to
5323 be UNSET, so there really isn't anything to push.
5324 Not pushing anything, on the other hand deprives us from the
5325 guarantee that regend[*p] is UNSET since undoing this operation
5326 will not reset its value properly. This is not important since
5327 the value will only be read on the next start_memory or at
5328 the very end and both events can only happen if this stop_memory
5332 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5334 /* Move past the register number and the inner group count. */
5339 /* \<digit> has been turned into a `duplicate' command which is
5340 followed by the numeric value of <digit> as the register number. */
5343 register re_char
*d2
, *dend2
;
5344 int regno
= *p
++; /* Get which register to match against. */
5345 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5347 /* Can't back reference a group which we've never matched. */
5348 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5351 /* Where in input to try to start matching. */
5352 d2
= regstart
[regno
];
5354 /* Remember the start point to rollback upon failure. */
5357 /* Where to stop matching; if both the place to start and
5358 the place to stop matching are in the same string, then
5359 set to the place to stop, otherwise, for now have to use
5360 the end of the first string. */
5362 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5363 == FIRST_STRING_P (regend
[regno
]))
5364 ? regend
[regno
] : end_match_1
);
5367 /* If necessary, advance to next segment in register
5371 if (dend2
== end_match_2
) break;
5372 if (dend2
== regend
[regno
]) break;
5374 /* End of string1 => advance to string2. */
5376 dend2
= regend
[regno
];
5378 /* At end of register contents => success */
5379 if (d2
== dend2
) break;
5381 /* If necessary, advance to next segment in data. */
5384 /* How many characters left in this segment to match. */
5387 /* Want how many consecutive characters we can match in
5388 one shot, so, if necessary, adjust the count. */
5389 if (mcnt
> dend2
- d2
)
5392 /* Compare that many; failure if mismatch, else move
5394 if (RE_TRANSLATE_P (translate
)
5395 ? bcmp_translate (d
, d2
, mcnt
, translate
, multibyte
)
5396 : memcmp (d
, d2
, mcnt
))
5401 d
+= mcnt
, d2
+= mcnt
;
5407 /* begline matches the empty string at the beginning of the string
5408 (unless `not_bol' is set in `bufp'), and after newlines. */
5410 DEBUG_PRINT1 ("EXECUTING begline.\n");
5412 if (AT_STRINGS_BEG (d
))
5414 if (!bufp
->not_bol
) break;
5419 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5423 /* In all other cases, we fail. */
5427 /* endline is the dual of begline. */
5429 DEBUG_PRINT1 ("EXECUTING endline.\n");
5431 if (AT_STRINGS_END (d
))
5433 if (!bufp
->not_eol
) break;
5437 PREFETCH_NOLIMIT ();
5444 /* Match at the very beginning of the data. */
5446 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5447 if (AT_STRINGS_BEG (d
))
5452 /* Match at the very end of the data. */
5454 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5455 if (AT_STRINGS_END (d
))
5460 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5461 pushes NULL as the value for the string on the stack. Then
5462 `POP_FAILURE_POINT' will keep the current value for the
5463 string, instead of restoring it. To see why, consider
5464 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5465 then the . fails against the \n. But the next thing we want
5466 to do is match the \n against the \n; if we restored the
5467 string value, we would be back at the foo.
5469 Because this is used only in specific cases, we don't need to
5470 check all the things that `on_failure_jump' does, to make
5471 sure the right things get saved on the stack. Hence we don't
5472 share its code. The only reason to push anything on the
5473 stack at all is that otherwise we would have to change
5474 `anychar's code to do something besides goto fail in this
5475 case; that seems worse than this. */
5476 case on_failure_keep_string_jump
:
5477 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5478 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5481 PUSH_FAILURE_POINT (p
- 3, NULL
);
5484 /* A nasty loop is introduced by the non-greedy *? and +?.
5485 With such loops, the stack only ever contains one failure point
5486 at a time, so that a plain on_failure_jump_loop kind of
5487 cycle detection cannot work. Worse yet, such a detection
5488 can not only fail to detect a cycle, but it can also wrongly
5489 detect a cycle (between different instantiations of the same
5491 So the method used for those nasty loops is a little different:
5492 We use a special cycle-detection-stack-frame which is pushed
5493 when the on_failure_jump_nastyloop failure-point is *popped*.
5494 This special frame thus marks the beginning of one iteration
5495 through the loop and we can hence easily check right here
5496 whether something matched between the beginning and the end of
5498 case on_failure_jump_nastyloop
:
5499 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5500 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5503 assert ((re_opcode_t
)p
[-4] == no_op
);
5506 CHECK_INFINITE_LOOP (p
- 4, d
);
5508 /* If there's a cycle, just continue without pushing
5509 this failure point. The failure point is the "try again"
5510 option, which shouldn't be tried.
5511 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5512 PUSH_FAILURE_POINT (p
- 3, d
);
5516 /* Simple loop detecting on_failure_jump: just check on the
5517 failure stack if the same spot was already hit earlier. */
5518 case on_failure_jump_loop
:
5520 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5521 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5525 CHECK_INFINITE_LOOP (p
- 3, d
);
5527 /* If there's a cycle, get out of the loop, as if the matching
5528 had failed. We used to just `goto fail' here, but that was
5529 aborting the search a bit too early: we want to keep the
5530 empty-loop-match and keep matching after the loop.
5531 We want (x?)*y\1z to match both xxyz and xxyxz. */
5534 PUSH_FAILURE_POINT (p
- 3, d
);
5539 /* Uses of on_failure_jump:
5541 Each alternative starts with an on_failure_jump that points
5542 to the beginning of the next alternative. Each alternative
5543 except the last ends with a jump that in effect jumps past
5544 the rest of the alternatives. (They really jump to the
5545 ending jump of the following alternative, because tensioning
5546 these jumps is a hassle.)
5548 Repeats start with an on_failure_jump that points past both
5549 the repetition text and either the following jump or
5550 pop_failure_jump back to this on_failure_jump. */
5551 case on_failure_jump
:
5552 IMMEDIATE_QUIT_CHECK
;
5553 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5554 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5557 PUSH_FAILURE_POINT (p
-3, d
);
5560 /* This operation is used for greedy *.
5561 Compare the beginning of the repeat with what in the
5562 pattern follows its end. If we can establish that there
5563 is nothing that they would both match, i.e., that we
5564 would have to backtrack because of (as in, e.g., `a*a')
5565 then we can use a non-backtracking loop based on
5566 on_failure_keep_string_jump instead of on_failure_jump. */
5567 case on_failure_jump_smart
:
5568 IMMEDIATE_QUIT_CHECK
;
5569 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5570 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5573 re_char
*p1
= p
; /* Next operation. */
5574 /* Here, we discard `const', making re_match non-reentrant. */
5575 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5576 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5578 p
-= 3; /* Reset so that we will re-execute the
5579 instruction once it's been changed. */
5581 EXTRACT_NUMBER (mcnt
, p2
- 2);
5583 /* Ensure this is a indeed the trivial kind of loop
5584 we are expecting. */
5585 assert (skip_one_char (p1
) == p2
- 3);
5586 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5587 DEBUG_STATEMENT (debug
+= 2);
5588 if (mutually_exclusive_p (bufp
, p1
, p2
))
5590 /* Use a fast `on_failure_keep_string_jump' loop. */
5591 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5592 *p3
= (unsigned char) on_failure_keep_string_jump
;
5593 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5597 /* Default to a safe `on_failure_jump' loop. */
5598 DEBUG_PRINT1 (" smart default => slow loop.\n");
5599 *p3
= (unsigned char) on_failure_jump
;
5601 DEBUG_STATEMENT (debug
-= 2);
5605 /* Unconditionally jump (without popping any failure points). */
5608 IMMEDIATE_QUIT_CHECK
;
5609 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5610 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5611 p
+= mcnt
; /* Do the jump. */
5612 DEBUG_PRINT2 ("(to %p).\n", p
);
5616 /* Have to succeed matching what follows at least n times.
5617 After that, handle like `on_failure_jump'. */
5619 /* Signedness doesn't matter since we only compare MCNT to 0. */
5620 EXTRACT_NUMBER (mcnt
, p
+ 2);
5621 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5623 /* Originally, mcnt is how many times we HAVE to succeed. */
5626 /* Here, we discard `const', making re_match non-reentrant. */
5627 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5630 PUSH_NUMBER (p2
, mcnt
);
5633 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5638 /* Signedness doesn't matter since we only compare MCNT to 0. */
5639 EXTRACT_NUMBER (mcnt
, p
+ 2);
5640 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5642 /* Originally, this is how many times we CAN jump. */
5645 /* Here, we discard `const', making re_match non-reentrant. */
5646 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5648 PUSH_NUMBER (p2
, mcnt
);
5649 goto unconditional_jump
;
5651 /* If don't have to jump any more, skip over the rest of command. */
5658 unsigned char *p2
; /* Location of the counter. */
5659 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5661 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5662 /* Here, we discard `const', making re_match non-reentrant. */
5663 p2
= (unsigned char*) p
+ mcnt
;
5664 /* Signedness doesn't matter since we only copy MCNT's bits . */
5665 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5666 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
5667 PUSH_NUMBER (p2
, mcnt
);
5673 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5674 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5676 /* We SUCCEED (or FAIL) in one of the following cases: */
5678 /* Case 1: D is at the beginning or the end of string. */
5679 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5683 /* C1 is the character before D, S1 is the syntax of C1, C2
5684 is the character at D, and S2 is the syntax of C2. */
5688 int offset
= PTR_TO_OFFSET (d
- 1);
5689 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5690 UPDATE_SYNTAX_TABLE (charpos
);
5692 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5695 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5697 PREFETCH_NOLIMIT ();
5698 c2
= RE_STRING_CHAR (d
, dend
- d
);
5701 if (/* Case 2: Only one of S1 and S2 is Sword. */
5702 ((s1
== Sword
) != (s2
== Sword
))
5703 /* Case 3: Both of S1 and S2 are Sword, and macro
5704 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5705 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5714 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5716 /* We FAIL in one of the following cases: */
5718 /* Case 1: D is at the end of string. */
5719 if (AT_STRINGS_END (d
))
5723 /* C1 is the character before D, S1 is the syntax of C1, C2
5724 is the character at D, and S2 is the syntax of C2. */
5728 int offset
= PTR_TO_OFFSET (d
);
5729 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5730 UPDATE_SYNTAX_TABLE (charpos
);
5733 c2
= RE_STRING_CHAR (d
, dend
- d
);
5736 /* Case 2: S2 is not Sword. */
5740 /* Case 3: D is not at the beginning of string ... */
5741 if (!AT_STRINGS_BEG (d
))
5743 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5745 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5749 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5751 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5758 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5760 /* We FAIL in one of the following cases: */
5762 /* Case 1: D is at the beginning of string. */
5763 if (AT_STRINGS_BEG (d
))
5767 /* C1 is the character before D, S1 is the syntax of C1, C2
5768 is the character at D, and S2 is the syntax of C2. */
5772 int offset
= PTR_TO_OFFSET (d
) - 1;
5773 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5774 UPDATE_SYNTAX_TABLE (charpos
);
5776 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5779 /* Case 2: S1 is not Sword. */
5783 /* Case 3: D is not at the end of string ... */
5784 if (!AT_STRINGS_END (d
))
5786 PREFETCH_NOLIMIT ();
5787 c2
= RE_STRING_CHAR (d
, dend
- d
);
5789 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
5793 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5795 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5803 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
5805 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
5809 int offset
= PTR_TO_OFFSET (d
);
5810 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5811 UPDATE_SYNTAX_TABLE (pos1
);
5818 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5820 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
5828 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5829 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
5834 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5835 if (PTR_BYTE_POS (d
) != PT_BYTE
)
5840 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5841 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
5846 case notcategoryspec
:
5847 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
5849 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
5855 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5857 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
5868 continue; /* Successfully executed one pattern command; keep going. */
5871 /* We goto here if a matching operation fails. */
5873 IMMEDIATE_QUIT_CHECK
;
5874 if (!FAIL_STACK_EMPTY ())
5877 /* A restart point is known. Restore to that state. */
5878 DEBUG_PRINT1 ("\nFAIL:\n");
5879 POP_FAILURE_POINT (str
, pat
);
5880 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
5882 case on_failure_keep_string_jump
:
5883 assert (str
== NULL
);
5884 goto continue_failure_jump
;
5886 case on_failure_jump_nastyloop
:
5887 assert ((re_opcode_t
)pat
[-2] == no_op
);
5888 PUSH_FAILURE_POINT (pat
- 2, str
);
5891 case on_failure_jump_loop
:
5892 case on_failure_jump
:
5895 continue_failure_jump
:
5896 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
5901 /* A special frame used for nastyloops. */
5908 assert (p
>= bufp
->buffer
&& p
<= pend
);
5910 if (d
>= string1
&& d
<= end1
)
5914 break; /* Matching at this starting point really fails. */
5918 goto restore_best_regs
;
5922 return -1; /* Failure to match. */
5925 /* Subroutine definitions for re_match_2. */
5927 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5928 bytes; nonzero otherwise. */
5931 bcmp_translate (s1
, s2
, len
, translate
, multibyte
)
5934 RE_TRANSLATE_TYPE translate
;
5935 const int multibyte
;
5937 register re_char
*p1
= s1
, *p2
= s2
;
5938 re_char
*p1_end
= s1
+ len
;
5939 re_char
*p2_end
= s2
+ len
;
5941 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
5942 different lengths, but relying on a single `len' would break this. -sm */
5943 while (p1
< p1_end
&& p2
< p2_end
)
5945 int p1_charlen
, p2_charlen
;
5946 re_wchar_t p1_ch
, p2_ch
;
5948 p1_ch
= RE_STRING_CHAR_AND_LENGTH (p1
, p1_end
- p1
, p1_charlen
);
5949 p2_ch
= RE_STRING_CHAR_AND_LENGTH (p2
, p2_end
- p2
, p2_charlen
);
5951 if (RE_TRANSLATE (translate
, p1_ch
)
5952 != RE_TRANSLATE (translate
, p2_ch
))
5955 p1
+= p1_charlen
, p2
+= p2_charlen
;
5958 if (p1
!= p1_end
|| p2
!= p2_end
)
5964 /* Entry points for GNU code. */
5966 /* re_compile_pattern is the GNU regular expression compiler: it
5967 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5968 Returns 0 if the pattern was valid, otherwise an error string.
5970 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5971 are set in BUFP on entry.
5973 We call regex_compile to do the actual compilation. */
5976 re_compile_pattern (pattern
, length
, bufp
)
5977 const char *pattern
;
5979 struct re_pattern_buffer
*bufp
;
5983 /* GNU code is written to assume at least RE_NREGS registers will be set
5984 (and at least one extra will be -1). */
5985 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5987 /* And GNU code determines whether or not to get register information
5988 by passing null for the REGS argument to re_match, etc., not by
5992 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
5996 return gettext (re_error_msgid
[(int) ret
]);
5998 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6000 /* Entry points compatible with 4.2 BSD regex library. We don't define
6001 them unless specifically requested. */
6003 #if defined _REGEX_RE_COMP || defined _LIBC
6005 /* BSD has one and only one pattern buffer. */
6006 static struct re_pattern_buffer re_comp_buf
;
6010 /* Make these definitions weak in libc, so POSIX programs can redefine
6011 these names if they don't use our functions, and still use
6012 regcomp/regexec below without link errors. */
6022 if (!re_comp_buf
.buffer
)
6023 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6024 return (char *) gettext ("No previous regular expression");
6028 if (!re_comp_buf
.buffer
)
6030 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6031 if (re_comp_buf
.buffer
== NULL
)
6032 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6033 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6034 re_comp_buf
.allocated
= 200;
6036 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6037 if (re_comp_buf
.fastmap
== NULL
)
6038 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6039 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6042 /* Since `re_exec' always passes NULL for the `regs' argument, we
6043 don't need to initialize the pattern buffer fields which affect it. */
6045 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6050 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6051 return (char *) gettext (re_error_msgid
[(int) ret
]);
6062 const int len
= strlen (s
);
6064 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6066 #endif /* _REGEX_RE_COMP */
6068 /* POSIX.2 functions. Don't define these for Emacs. */
6072 /* regcomp takes a regular expression as a string and compiles it.
6074 PREG is a regex_t *. We do not expect any fields to be initialized,
6075 since POSIX says we shouldn't. Thus, we set
6077 `buffer' to the compiled pattern;
6078 `used' to the length of the compiled pattern;
6079 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6080 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6081 RE_SYNTAX_POSIX_BASIC;
6082 `fastmap' to an allocated space for the fastmap;
6083 `fastmap_accurate' to zero;
6084 `re_nsub' to the number of subexpressions in PATTERN.
6086 PATTERN is the address of the pattern string.
6088 CFLAGS is a series of bits which affect compilation.
6090 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6091 use POSIX basic syntax.
6093 If REG_NEWLINE is set, then . and [^...] don't match newline.
6094 Also, regexec will try a match beginning after every newline.
6096 If REG_ICASE is set, then we considers upper- and lowercase
6097 versions of letters to be equivalent when matching.
6099 If REG_NOSUB is set, then when PREG is passed to regexec, that
6100 routine will report only success or failure, and nothing about the
6103 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6104 the return codes and their meanings.) */
6107 regcomp (preg
, pattern
, cflags
)
6108 regex_t
*__restrict preg
;
6109 const char *__restrict pattern
;
6114 = (cflags
& REG_EXTENDED
) ?
6115 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6117 /* regex_compile will allocate the space for the compiled pattern. */
6119 preg
->allocated
= 0;
6122 /* Try to allocate space for the fastmap. */
6123 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6125 if (cflags
& REG_ICASE
)
6130 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6131 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6132 if (preg
->translate
== NULL
)
6133 return (int) REG_ESPACE
;
6135 /* Map uppercase characters to corresponding lowercase ones. */
6136 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6137 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6140 preg
->translate
= NULL
;
6142 /* If REG_NEWLINE is set, newlines are treated differently. */
6143 if (cflags
& REG_NEWLINE
)
6144 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6145 syntax
&= ~RE_DOT_NEWLINE
;
6146 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6149 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6151 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6153 /* POSIX says a null character in the pattern terminates it, so we
6154 can use strlen here in compiling the pattern. */
6155 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6157 /* POSIX doesn't distinguish between an unmatched open-group and an
6158 unmatched close-group: both are REG_EPAREN. */
6159 if (ret
== REG_ERPAREN
)
6162 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6163 { /* Compute the fastmap now, since regexec cannot modify the pattern
6165 re_compile_fastmap (preg
);
6166 if (preg
->can_be_null
)
6167 { /* The fastmap can't be used anyway. */
6168 free (preg
->fastmap
);
6169 preg
->fastmap
= NULL
;
6174 WEAK_ALIAS (__regcomp
, regcomp
)
6177 /* regexec searches for a given pattern, specified by PREG, in the
6180 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6181 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6182 least NMATCH elements, and we set them to the offsets of the
6183 corresponding matched substrings.
6185 EFLAGS specifies `execution flags' which affect matching: if
6186 REG_NOTBOL is set, then ^ does not match at the beginning of the
6187 string; if REG_NOTEOL is set, then $ does not match at the end.
6189 We return 0 if we find a match and REG_NOMATCH if not. */
6192 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6193 const regex_t
*__restrict preg
;
6194 const char *__restrict string
;
6196 regmatch_t pmatch
[__restrict_arr
];
6200 struct re_registers regs
;
6201 regex_t private_preg
;
6202 int len
= strlen (string
);
6203 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6205 private_preg
= *preg
;
6207 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6208 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6210 /* The user has told us exactly how many registers to return
6211 information about, via `nmatch'. We have to pass that on to the
6212 matching routines. */
6213 private_preg
.regs_allocated
= REGS_FIXED
;
6217 regs
.num_regs
= nmatch
;
6218 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6219 if (regs
.start
== NULL
)
6220 return (int) REG_NOMATCH
;
6221 regs
.end
= regs
.start
+ nmatch
;
6224 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6225 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6226 was a little bit longer but still only matching the real part.
6227 This works because the `endline' will check for a '\n' and will find a
6228 '\0', correctly deciding that this is not the end of a line.
6229 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6230 a convenient '\0' there. For all we know, the string could be preceded
6231 by '\n' which would throw things off. */
6233 /* Perform the searching operation. */
6234 ret
= re_search (&private_preg
, string
, len
,
6235 /* start: */ 0, /* range: */ len
,
6236 want_reg_info
? ®s
: (struct re_registers
*) 0);
6238 /* Copy the register information to the POSIX structure. */
6245 for (r
= 0; r
< nmatch
; r
++)
6247 pmatch
[r
].rm_so
= regs
.start
[r
];
6248 pmatch
[r
].rm_eo
= regs
.end
[r
];
6252 /* If we needed the temporary register info, free the space now. */
6256 /* We want zero return to mean success, unlike `re_search'. */
6257 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6259 WEAK_ALIAS (__regexec
, regexec
)
6262 /* Returns a message corresponding to an error code, ERRCODE, returned
6263 from either regcomp or regexec. We don't use PREG here. */
6266 regerror (errcode
, preg
, errbuf
, errbuf_size
)
6268 const regex_t
*preg
;
6276 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6277 /* Only error codes returned by the rest of the code should be passed
6278 to this routine. If we are given anything else, or if other regex
6279 code generates an invalid error code, then the program has a bug.
6280 Dump core so we can fix it. */
6283 msg
= gettext (re_error_msgid
[errcode
]);
6285 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6287 if (errbuf_size
!= 0)
6289 if (msg_size
> errbuf_size
)
6291 strncpy (errbuf
, msg
, errbuf_size
- 1);
6292 errbuf
[errbuf_size
- 1] = 0;
6295 strcpy (errbuf
, msg
);
6300 WEAK_ALIAS (__regerror
, regerror
)
6303 /* Free dynamically allocated space used by PREG. */
6309 if (preg
->buffer
!= NULL
)
6310 free (preg
->buffer
);
6311 preg
->buffer
= NULL
;
6313 preg
->allocated
= 0;
6316 if (preg
->fastmap
!= NULL
)
6317 free (preg
->fastmap
);
6318 preg
->fastmap
= NULL
;
6319 preg
->fastmap_accurate
= 0;
6321 if (preg
->translate
!= NULL
)
6322 free (preg
->translate
);
6323 preg
->translate
= NULL
;
6325 WEAK_ALIAS (__regfree
, regfree
)
6327 #endif /* not emacs */
6329 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6330 (do not change this comment) */