Updated.
[glibc.git] / posix / fnmatch_loop.c
blob4c6168ae9fe44db73e3c7e4df7210a90ac2acff3
1 /* Copyright (C) 1991-1993, 1996-2000, 2001 Free Software Foundation, Inc.
2 This file is part of the GNU C Library.
4 The GNU C Library is free software; you can redistribute it and/or
5 modify it under the terms of the GNU Lesser General Public
6 License as published by the Free Software Foundation; either
7 version 2.1 of the License, or (at your option) any later version.
9 The GNU C Library is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 Lesser General Public License for more details.
14 You should have received a copy of the GNU Lesser General Public
15 License along with the GNU C Library; if not, write to the Free
16 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
17 02111-1307 USA. */
19 /* Match STRING against the filename pattern PATTERN, returning zero if
20 it matches, nonzero if not. */
21 static int FCT (const CHAR *pattern, const CHAR *string,
22 const CHAR *string_end, int no_leading_period, int flags)
23 internal_function;
24 static int EXT (INT opt, const CHAR *pattern, const CHAR *string,
25 const CHAR *string_end, int no_leading_period, int flags)
26 internal_function;
27 static const CHAR *END (const CHAR *patternp) internal_function;
29 static int
30 internal_function
31 FCT (pattern, string, string_end, no_leading_period, flags)
32 const CHAR *pattern;
33 const CHAR *string;
34 const CHAR *string_end;
35 int no_leading_period;
36 int flags;
38 register const CHAR *p = pattern, *n = string;
39 register UCHAR c;
40 #ifdef _LIBC
41 # if WIDE_CHAR_VERSION
42 const char *collseq = (const char *)
43 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
44 # else
45 const UCHAR *collseq = (const UCHAR *)
46 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQMB);
47 # endif
48 #endif
50 while ((c = *p++) != L('\0'))
52 int new_no_leading_period = 0;
53 c = FOLD (c);
55 switch (c)
57 case L('?'):
58 if (__builtin_expect (flags & FNM_EXTMATCH, 0) && *p == '(')
60 int res;
62 res = EXT (c, p, n, string_end, no_leading_period,
63 flags);
64 if (res != -1)
65 return res;
68 if (n == string_end)
69 return FNM_NOMATCH;
70 else if (*n == L('/') && (flags & FNM_FILE_NAME))
71 return FNM_NOMATCH;
72 else if (*n == L('.') && no_leading_period)
73 return FNM_NOMATCH;
74 break;
76 case L('\\'):
77 if (!(flags & FNM_NOESCAPE))
79 c = *p++;
80 if (c == L('\0'))
81 /* Trailing \ loses. */
82 return FNM_NOMATCH;
83 c = FOLD (c);
85 if (n == string_end || FOLD ((UCHAR) *n) != c)
86 return FNM_NOMATCH;
87 break;
89 case L('*'):
90 if (__builtin_expect (flags & FNM_EXTMATCH, 0) && *p == '(')
92 int res;
94 res = EXT (c, p, n, string_end, no_leading_period,
95 flags);
96 if (res != -1)
97 return res;
100 if (n != string_end && *n == L('.') && no_leading_period)
101 return FNM_NOMATCH;
103 for (c = *p++; c == L('?') || c == L('*'); c = *p++)
105 if (*p == L('(') && (flags & FNM_EXTMATCH) != 0)
107 const CHAR *endp = END (p);
108 if (endp != p)
110 /* This is a pattern. Skip over it. */
111 p = endp;
112 continue;
116 if (c == L('?'))
118 /* A ? needs to match one character. */
119 if (n == string_end)
120 /* There isn't another character; no match. */
121 return FNM_NOMATCH;
122 else if (*n == L('/')
123 && __builtin_expect (flags & FNM_FILE_NAME, 0))
124 /* A slash does not match a wildcard under
125 FNM_FILE_NAME. */
126 return FNM_NOMATCH;
127 else
128 /* One character of the string is consumed in matching
129 this ? wildcard, so *??? won't match if there are
130 less than three characters. */
131 ++n;
135 if (c == L('\0'))
136 /* The wildcard(s) is/are the last element of the pattern.
137 If the name is a file name and contains another slash
138 this means it cannot match, unless the FNM_LEADING_DIR
139 flag is set. */
141 int result = (flags & FNM_FILE_NAME) == 0 ? 0 : FNM_NOMATCH;
143 if (flags & FNM_FILE_NAME)
145 if (flags & FNM_LEADING_DIR)
146 result = 0;
147 else
149 if (MEMCHR (n, L('/'), string_end - n) == NULL)
150 result = 0;
154 return result;
156 else
158 const CHAR *endp;
160 endp = MEMCHR (n, (flags & FNM_FILE_NAME) ? L('/') : L('\0'),
161 string_end - n);
162 if (endp == NULL)
163 endp = string_end;
165 if (c == L('[')
166 || (__builtin_expect (flags & FNM_EXTMATCH, 0) != 0
167 && (c == L('@') || c == L('+') || c == L('!'))
168 && *p == L('(')))
170 int flags2 = ((flags & FNM_FILE_NAME)
171 ? flags : (flags & ~FNM_PERIOD));
172 int no_leading_period2 = no_leading_period;
174 for (--p; n < endp; ++n, no_leading_period2 = 0)
175 if (FCT (p, n, string_end, no_leading_period2, flags2)
176 == 0)
177 return 0;
179 else if (c == L('/') && (flags & FNM_FILE_NAME))
181 while (n < string_end && *n != L('/'))
182 ++n;
183 if (n < string_end && *n == L('/')
184 && (FCT (p, n + 1, string_end, flags & FNM_PERIOD, flags)
185 == 0))
186 return 0;
188 else
190 int flags2 = ((flags & FNM_FILE_NAME)
191 ? flags : (flags & ~FNM_PERIOD));
192 int no_leading_period2 = no_leading_period;
194 if (c == L('\\') && !(flags & FNM_NOESCAPE))
195 c = *p;
196 c = FOLD (c);
197 for (--p; n < endp; ++n, no_leading_period2 = 0)
198 if (FOLD ((UCHAR) *n) == c
199 && (FCT (p, n, string_end, no_leading_period2, flags2)
200 == 0))
201 return 0;
205 /* If we come here no match is possible with the wildcard. */
206 return FNM_NOMATCH;
208 case L('['):
210 /* Nonzero if the sense of the character class is inverted. */
211 register int not;
212 CHAR cold;
213 UCHAR fn;
215 if (posixly_correct == 0)
216 posixly_correct = getenv ("POSIXLY_CORRECT") != NULL ? 1 : -1;
218 if (n == string_end)
219 return FNM_NOMATCH;
221 if (*n == L('.') && no_leading_period)
222 return FNM_NOMATCH;
224 if (*n == L('/') && (flags & FNM_FILE_NAME))
225 /* `/' cannot be matched. */
226 return FNM_NOMATCH;
228 not = (*p == L('!') || (posixly_correct < 0 && *p == L('^')));
229 if (not)
230 ++p;
232 fn = FOLD ((UCHAR) *n);
234 c = *p++;
235 for (;;)
237 if (!(flags & FNM_NOESCAPE) && c == L('\\'))
239 if (*p == L('\0'))
240 return FNM_NOMATCH;
241 c = FOLD ((UCHAR) *p);
242 ++p;
244 if (c == fn)
245 goto matched;
247 else if (c == L('[') && *p == L(':'))
249 /* Leave room for the null. */
250 CHAR str[CHAR_CLASS_MAX_LENGTH + 1];
251 size_t c1 = 0;
252 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
253 wctype_t wt;
254 #endif
255 const CHAR *startp = p;
257 for (;;)
259 if (c1 == CHAR_CLASS_MAX_LENGTH)
260 /* The name is too long and therefore the pattern
261 is ill-formed. */
262 return FNM_NOMATCH;
264 c = *++p;
265 if (c == L(':') && p[1] == L(']'))
267 p += 2;
268 break;
270 if (c < L('a') || c >= L('z'))
272 /* This cannot possibly be a character class name.
273 Match it as a normal range. */
274 p = startp;
275 c = L('[');
276 goto normal_bracket;
278 str[c1++] = c;
280 str[c1] = L('\0');
282 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
283 wt = IS_CHAR_CLASS (str);
284 if (wt == 0)
285 /* Invalid character class name. */
286 return FNM_NOMATCH;
288 # if defined _LIBC && ! WIDE_CHAR_VERSION
289 /* The following code is glibc specific but does
290 there a good job in speeding up the code since
291 we can avoid the btowc() call. */
292 if (_ISCTYPE ((UCHAR) *n, wt))
293 goto matched;
294 # else
295 if (ISWCTYPE (BTOWC ((UCHAR) *n), wt))
296 goto matched;
297 # endif
298 #else
299 if ((STREQ (str, L("alnum")) && ISALNUM ((UCHAR) *n))
300 || (STREQ (str, L("alpha")) && ISALPHA ((UCHAR) *n))
301 || (STREQ (str, L("blank")) && ISBLANK ((UCHAR) *n))
302 || (STREQ (str, L("cntrl")) && ISCNTRL ((UCHAR) *n))
303 || (STREQ (str, L("digit")) && ISDIGIT ((UCHAR) *n))
304 || (STREQ (str, L("graph")) && ISGRAPH ((UCHAR) *n))
305 || (STREQ (str, L("lower")) && ISLOWER ((UCHAR) *n))
306 || (STREQ (str, L("print")) && ISPRINT ((UCHAR) *n))
307 || (STREQ (str, L("punct")) && ISPUNCT ((UCHAR) *n))
308 || (STREQ (str, L("space")) && ISSPACE ((UCHAR) *n))
309 || (STREQ (str, L("upper")) && ISUPPER ((UCHAR) *n))
310 || (STREQ (str, L("xdigit")) && ISXDIGIT ((UCHAR) *n)))
311 goto matched;
312 #endif
313 c = *p++;
315 #ifdef _LIBC
316 else if (c == L('[') && *p == L('='))
318 UCHAR str[1];
319 uint32_t nrules =
320 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
321 const CHAR *startp = p;
323 c = *++p;
324 if (c == L('\0'))
326 p = startp;
327 c = L('[');
328 goto normal_bracket;
330 str[0] = c;
332 c = *++p;
333 if (c != L('=') || p[1] != L(']'))
335 p = startp;
336 c = L('[');
337 goto normal_bracket;
339 p += 2;
341 if (nrules == 0)
343 if ((UCHAR) *n == str[0])
344 goto matched;
346 else
348 const int32_t *table;
349 # if WIDE_CHAR_VERSION
350 const int32_t *weights;
351 const int32_t *extra;
352 # else
353 const unsigned char *weights;
354 const unsigned char *extra;
355 # endif
356 const int32_t *indirect;
357 int32_t idx;
358 const UCHAR *cp = (const UCHAR *) str;
360 /* This #include defines a local function! */
361 # if WIDE_CHAR_VERSION
362 # include <locale/weightwc.h>
363 # else
364 # include <locale/weight.h>
365 # endif
367 # if WIDE_CHAR_VERSION
368 table = (const int32_t *)
369 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
370 weights = (const int32_t *)
371 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
372 extra = (const int32_t *)
373 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
374 indirect = (const int32_t *)
375 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
376 # else
377 table = (const int32_t *)
378 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
379 weights = (const unsigned char *)
380 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
381 extra = (const unsigned char *)
382 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
383 indirect = (const int32_t *)
384 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
385 # endif
387 idx = findidx (&cp);
388 if (idx != 0)
390 /* We found a table entry. Now see whether the
391 character we are currently at has the same
392 equivalance class value. */
393 int len = weights[idx];
394 int32_t idx2;
395 const UCHAR *np = (const UCHAR *) n;
397 idx2 = findidx (&np);
398 if (idx2 != 0 && len == weights[idx2])
400 int cnt = 0;
402 while (cnt < len
403 && (weights[idx + 1 + cnt]
404 == weights[idx2 + 1 + cnt]))
405 ++cnt;
407 if (cnt == len)
408 goto matched;
413 c = *p++;
415 #endif
416 else if (c == L('\0'))
417 /* [ (unterminated) loses. */
418 return FNM_NOMATCH;
419 else
421 int is_range = 0;
423 #ifdef _LIBC
424 int is_seqval = 0;
426 if (c == L('[') && *p == L('.'))
428 uint32_t nrules =
429 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
430 const CHAR *startp = p;
431 size_t c1 = 0;
433 while (1)
435 c = *++p;
436 if (c == L('.') && p[1] == L(']'))
438 p += 2;
439 break;
441 if (c == '\0')
442 return FNM_NOMATCH;
443 ++c1;
446 /* We have to handling the symbols differently in
447 ranges since then the collation sequence is
448 important. */
449 is_range = *p == L('-') && p[1] != L('\0');
451 if (nrules == 0)
453 /* There are no names defined in the collation
454 data. Therefore we only accept the trivial
455 names consisting of the character itself. */
456 if (c1 != 1)
457 return FNM_NOMATCH;
459 if (!is_range && *n == startp[1])
460 goto matched;
462 cold = startp[1];
463 c = *p++;
465 else
467 int32_t table_size;
468 const int32_t *symb_table;
469 # ifdef WIDE_CHAR_VERSION
470 char str[c1];
471 unsigned int strcnt;
472 # else
473 # define str (startp + 1)
474 # endif
475 const unsigned char *extra;
476 int32_t idx;
477 int32_t elem;
478 int32_t second;
479 int32_t hash;
481 # ifdef WIDE_CHAR_VERSION
482 /* We have to convert the name to a single-byte
483 string. This is possible since the names
484 consist of ASCII characters and the internal
485 representation is UCS4. */
486 for (strcnt = 0; strcnt < c1; ++strcnt)
487 str[strcnt] = startp[1 + strcnt];
488 #endif
490 table_size =
491 _NL_CURRENT_WORD (LC_COLLATE,
492 _NL_COLLATE_SYMB_HASH_SIZEMB);
493 symb_table = (const int32_t *)
494 _NL_CURRENT (LC_COLLATE,
495 _NL_COLLATE_SYMB_TABLEMB);
496 extra = (const unsigned char *)
497 _NL_CURRENT (LC_COLLATE,
498 _NL_COLLATE_SYMB_EXTRAMB);
500 /* Locate the character in the hashing table. */
501 hash = elem_hash (str, c1);
503 idx = 0;
504 elem = hash % table_size;
505 second = hash % (table_size - 2);
506 while (symb_table[2 * elem] != 0)
508 /* First compare the hashing value. */
509 if (symb_table[2 * elem] == hash
510 && c1 == extra[symb_table[2 * elem + 1]]
511 && memcmp (str,
512 &extra[symb_table[2 * elem + 1]
513 + 1], c1) == 0)
515 /* Yep, this is the entry. */
516 idx = symb_table[2 * elem + 1];
517 idx += 1 + extra[idx];
518 break;
521 /* Next entry. */
522 elem += second;
525 if (symb_table[2 * elem] != 0)
527 /* Compare the byte sequence but only if
528 this is not part of a range. */
529 # ifdef WIDE_CHAR_VERSION
530 int32_t *wextra;
532 idx += 1 + extra[idx];
533 /* Adjust for the alignment. */
534 idx = (idx + 3) & ~3;
536 wextra = (int32_t *) &extra[idx + 4];
537 # endif
539 if (! is_range)
541 # ifdef WIDE_CHAR_VERSION
542 for (c1 = 0; c1 < wextra[idx]; ++c1)
543 if (n[c1] != wextra[1 + c1])
544 break;
546 if (c1 == wextra[idx])
547 goto matched;
548 # else
549 for (c1 = 0; c1 < extra[idx]; ++c1)
550 if (n[c1] != extra[1 + c1])
551 break;
553 if (c1 == extra[idx])
554 goto matched;
555 # endif
558 /* Get the collation sequence value. */
559 is_seqval = 1;
560 # ifdef WIDE_CHAR_VERSION
561 cold = wextra[1 + wextra[idx]];
562 # else
563 /* Adjust for the alignment. */
564 idx += 1 + extra[idx];
565 idx = (idx + 3) & ~4;
566 cold = *((int32_t *) &extra[idx]);
567 # endif
569 c = *p++;
571 else if (c1 == 1)
573 /* No valid character. Match it as a
574 single byte. */
575 if (!is_range && *n == str[0])
576 goto matched;
578 cold = str[0];
579 c = *p++;
581 else
582 return FNM_NOMATCH;
585 else
586 # undef str
587 #endif
589 c = FOLD (c);
590 normal_bracket:
592 /* We have to handling the symbols differently in
593 ranges since then the collation sequence is
594 important. */
595 is_range = *p == L('-') && p[1] != L('\0');
597 if (!is_range && c == fn)
598 goto matched;
600 cold = c;
601 c = *p++;
604 if (c == L('-') && *p != L(']'))
606 #if _LIBC
607 /* We have to find the collation sequence
608 value for C. Collation sequence is nothing
609 we can regularly access. The sequence
610 value is defined by the order in which the
611 definitions of the collation values for the
612 various characters appear in the source
613 file. A strange concept, nowhere
614 documented. */
615 uint32_t fcollseq;
616 uint32_t lcollseq;
617 UCHAR cend = *p++;
619 # ifdef WIDE_CHAR_VERSION
620 /* Search in the `names' array for the characters. */
621 fcollseq = collseq_table_lookup (collseq, fn);
622 if (fcollseq == ~((uint32_t) 0))
623 /* XXX We don't know anything about the character
624 we are supposed to match. This means we are
625 failing. */
626 goto range_not_matched;
628 if (is_seqval)
629 lcollseq = cold;
630 else
631 lcollseq = collseq_table_lookup (collseq, cold);
632 # else
633 fcollseq = collseq[fn];
634 lcollseq = is_seqval ? cold : collseq[(UCHAR) cold];
635 # endif
637 is_seqval = 0;
638 if (cend == L('[') && *p == L('.'))
640 uint32_t nrules =
641 _NL_CURRENT_WORD (LC_COLLATE,
642 _NL_COLLATE_NRULES);
643 const CHAR *startp = p;
644 size_t c1 = 0;
646 while (1)
648 c = *++p;
649 if (c == L('.') && p[1] == L(']'))
651 p += 2;
652 break;
654 if (c == '\0')
655 return FNM_NOMATCH;
656 ++c1;
659 if (nrules == 0)
661 /* There are no names defined in the
662 collation data. Therefore we only
663 accept the trivial names consisting
664 of the character itself. */
665 if (c1 != 1)
666 return FNM_NOMATCH;
668 cend = startp[1];
670 else
672 int32_t table_size;
673 const int32_t *symb_table;
674 # ifdef WIDE_CHAR_VERSION
675 char str[c1];
676 unsigned int strcnt;
677 # else
678 # define str (startp + 1)
679 # endif
680 const unsigned char *extra;
681 int32_t idx;
682 int32_t elem;
683 int32_t second;
684 int32_t hash;
686 # ifdef WIDE_CHAR_VERSION
687 /* We have to convert the name to a single-byte
688 string. This is possible since the names
689 consist of ASCII characters and the internal
690 representation is UCS4. */
691 for (strcnt = 0; strcnt < c1; ++strcnt)
692 str[strcnt] = startp[1 + strcnt];
693 # endif
695 table_size =
696 _NL_CURRENT_WORD (LC_COLLATE,
697 _NL_COLLATE_SYMB_HASH_SIZEMB);
698 symb_table = (const int32_t *)
699 _NL_CURRENT (LC_COLLATE,
700 _NL_COLLATE_SYMB_TABLEMB);
701 extra = (const unsigned char *)
702 _NL_CURRENT (LC_COLLATE,
703 _NL_COLLATE_SYMB_EXTRAMB);
705 /* Locate the character in the hashing
706 table. */
707 hash = elem_hash (str, c1);
709 idx = 0;
710 elem = hash % table_size;
711 second = hash % (table_size - 2);
712 while (symb_table[2 * elem] != 0)
714 /* First compare the hashing value. */
715 if (symb_table[2 * elem] == hash
716 && (c1
717 == extra[symb_table[2 * elem + 1]])
718 && memcmp (str,
719 &extra[symb_table[2 * elem + 1]
720 + 1], c1) == 0)
722 /* Yep, this is the entry. */
723 idx = symb_table[2 * elem + 1];
724 idx += 1 + extra[idx];
725 break;
728 /* Next entry. */
729 elem += second;
732 if (symb_table[2 * elem] != 0)
734 /* Compare the byte sequence but only if
735 this is not part of a range. */
736 # ifdef WIDE_CHAR_VERSION
737 int32_t *wextra;
739 idx += 1 + extra[idx];
740 /* Adjust for the alignment. */
741 idx = (idx + 3) & ~4;
743 wextra = (int32_t *) &extra[idx + 4];
744 # endif
745 /* Get the collation sequence value. */
746 is_seqval = 1;
747 # ifdef WIDE_CHAR_VERSION
748 cend = wextra[1 + wextra[idx]];
749 # else
750 /* Adjust for the alignment. */
751 idx += 1 + extra[idx];
752 idx = (idx + 3) & ~4;
753 cend = *((int32_t *) &extra[idx]);
754 # endif
756 else if (symb_table[2 * elem] != 0 && c1 == 1)
758 cend = str[0];
759 c = *p++;
761 else
762 return FNM_NOMATCH;
764 # undef str
766 else
768 if (!(flags & FNM_NOESCAPE) && cend == L('\\'))
769 cend = *p++;
770 if (cend == L('\0'))
771 return FNM_NOMATCH;
772 cend = FOLD (cend);
775 /* XXX It is not entirely clear to me how to handle
776 characters which are not mentioned in the
777 collation specification. */
778 if (
779 # ifdef WIDE_CHAR_VERSION
780 lcollseq == 0xffffffff ||
781 # endif
782 lcollseq <= fcollseq)
784 /* We have to look at the upper bound. */
785 uint32_t hcollseq;
787 if (is_seqval)
788 hcollseq = cend;
789 else
791 # ifdef WIDE_CHAR_VERSION
792 hcollseq =
793 collseq_table_lookup (collseq, cend);
794 if (hcollseq == ~((uint32_t) 0))
796 /* Hum, no information about the upper
797 bound. The matching succeeds if the
798 lower bound is matched exactly. */
799 if (lcollseq != fcollseq)
800 goto range_not_matched;
802 goto matched;
804 # else
805 hcollseq = collseq[cend];
806 # endif
809 if (lcollseq <= hcollseq && fcollseq <= hcollseq)
810 goto matched;
812 # ifdef WIDE_CHAR_VERSION
813 range_not_matched:
814 # endif
815 #else
816 /* We use a boring value comparison of the character
817 values. This is better than comparing using
818 `strcoll' since the latter would have surprising
819 and sometimes fatal consequences. */
820 UCHAR cend = *p++;
822 if (!(flags & FNM_NOESCAPE) && cend == L('\\'))
823 cend = *p++;
824 if (cend == L('\0'))
825 return FNM_NOMATCH;
827 /* It is a range. */
828 if (cold <= fn && fn <= cend)
829 goto matched;
830 #endif
832 c = *p++;
836 if (c == L(']'))
837 break;
840 if (!not)
841 return FNM_NOMATCH;
842 break;
844 matched:
845 /* Skip the rest of the [...] that already matched. */
848 ignore_next:
849 c = *p++;
851 if (c == L('\0'))
852 /* [... (unterminated) loses. */
853 return FNM_NOMATCH;
855 if (!(flags & FNM_NOESCAPE) && c == L('\\'))
857 if (*p == L('\0'))
858 return FNM_NOMATCH;
859 /* XXX 1003.2d11 is unclear if this is right. */
860 ++p;
862 else if (c == L('[') && *p == L(':'))
864 int c1 = 0;
865 const CHAR *startp = p;
867 while (1)
869 c = *++p;
870 if (++c1 == CHAR_CLASS_MAX_LENGTH)
871 return FNM_NOMATCH;
873 if (*p == L(':') && p[1] == L(']'))
874 break;
876 if (c < L('a') || c >= L('z'))
878 p = startp;
879 goto ignore_next;
882 p += 2;
883 c = *p++;
885 else if (c == L('[') && *p == L('='))
887 c = *++p;
888 if (c == L('\0'))
889 return FNM_NOMATCH;
890 c = *++p;
891 if (c != L('=') || p[1] != L(']'))
892 return FNM_NOMATCH;
893 p += 2;
894 c = *p++;
896 else if (c == L('[') && *p == L('.'))
898 ++p;
899 while (1)
901 c = *++p;
902 if (c == '\0')
903 return FNM_NOMATCH;
905 if (*p == L('.') && p[1] == L(']'))
906 break;
908 p += 2;
909 c = *p++;
912 while (c != L(']'));
913 if (not)
914 return FNM_NOMATCH;
916 break;
918 case L('+'):
919 case L('@'):
920 case L('!'):
921 if (__builtin_expect (flags & FNM_EXTMATCH, 0) && *p == '(')
923 int res;
925 res = EXT (c, p, n, string_end, no_leading_period, flags);
926 if (res != -1)
927 return res;
929 goto normal_match;
931 case L('/'):
932 if (NO_LEADING_PERIOD (flags))
934 if (n == string_end || c != *n)
935 return FNM_NOMATCH;
937 new_no_leading_period = 1;
938 break;
940 /* FALLTHROUGH */
941 default:
942 normal_match:
943 if (n == string_end || c != FOLD ((UCHAR) *n))
944 return FNM_NOMATCH;
947 no_leading_period = new_no_leading_period;
948 ++n;
951 if (n == string_end)
952 return 0;
954 if ((flags & FNM_LEADING_DIR) && n != string_end && *n == L('/'))
955 /* The FNM_LEADING_DIR flag says that "foo*" matches "foobar/frobozz". */
956 return 0;
958 return FNM_NOMATCH;
962 static const CHAR *
963 internal_function
964 END (const CHAR *pattern)
966 const CHAR *p = pattern;
968 while (1)
969 if (*++p == L('\0'))
970 /* This is an invalid pattern. */
971 return pattern;
972 else if (*p == L('['))
974 /* Handle brackets special. */
975 if (posixly_correct == 0)
976 posixly_correct = getenv ("POSIXLY_CORRECT") != NULL ? 1 : -1;
978 /* Skip the not sign. We have to recognize it because of a possibly
979 following ']'. */
980 if (*++p == L('!') || (posixly_correct < 0 && *p == L('^')))
981 ++p;
982 /* A leading ']' is recognized as such. */
983 if (*p == L(']'))
984 ++p;
985 /* Skip over all characters of the list. */
986 while (*p != L(']'))
987 if (*p++ == L('\0'))
988 /* This is no valid pattern. */
989 return pattern;
991 else if ((*p == L('?') || *p == L('*') || *p == L('+') || *p == L('@')
992 || *p == L('!')) && p[1] == L('('))
993 p = END (p + 1);
994 else if (*p == L(')'))
995 break;
997 return p + 1;
1001 static int
1002 internal_function
1003 EXT (INT opt, const CHAR *pattern, const CHAR *string, const CHAR *string_end,
1004 int no_leading_period, int flags)
1006 const CHAR *startp;
1007 int level;
1008 struct patternlist
1010 struct patternlist *next;
1011 CHAR str[0];
1012 } *list = NULL;
1013 struct patternlist **lastp = &list;
1014 size_t pattern_len = STRLEN (pattern);
1015 const CHAR *p;
1016 const CHAR *rs;
1018 /* Parse the pattern. Store the individual parts in the list. */
1019 level = 0;
1020 for (startp = p = pattern + 1; level >= 0; ++p)
1021 if (*p == L('\0'))
1022 /* This is an invalid pattern. */
1023 return -1;
1024 else if (*p == L('['))
1026 /* Handle brackets special. */
1027 if (posixly_correct == 0)
1028 posixly_correct = getenv ("POSIXLY_CORRECT") != NULL ? 1 : -1;
1030 /* Skip the not sign. We have to recognize it because of a possibly
1031 following ']'. */
1032 if (*++p == L('!') || (posixly_correct < 0 && *p == L('^')))
1033 ++p;
1034 /* A leading ']' is recognized as such. */
1035 if (*p == L(']'))
1036 ++p;
1037 /* Skip over all characters of the list. */
1038 while (*p != L(']'))
1039 if (*p++ == L('\0'))
1040 /* This is no valid pattern. */
1041 return -1;
1043 else if ((*p == L('?') || *p == L('*') || *p == L('+') || *p == L('@')
1044 || *p == L('!')) && p[1] == L('('))
1045 /* Remember the nesting level. */
1046 ++level;
1047 else if (*p == L(')'))
1049 if (level-- == 0)
1051 /* This means we found the end of the pattern. */
1052 #define NEW_PATTERN \
1053 struct patternlist *newp; \
1055 if (opt == L('?') || opt == L('@')) \
1056 newp = alloca (sizeof (struct patternlist) \
1057 + (pattern_len * sizeof (CHAR))); \
1058 else \
1059 newp = alloca (sizeof (struct patternlist) \
1060 + ((p - startp + 1) * sizeof (CHAR))); \
1061 *((CHAR *) MEMPCPY (newp->str, startp, p - startp)) = L('\0'); \
1062 newp->next = NULL; \
1063 *lastp = newp; \
1064 lastp = &newp->next
1065 NEW_PATTERN;
1068 else if (*p == L('|'))
1070 if (level == 0)
1072 NEW_PATTERN;
1073 startp = p + 1;
1076 assert (list != NULL);
1077 assert (p[-1] == L(')'));
1078 #undef NEW_PATTERN
1080 switch (opt)
1082 case L('*'):
1083 if (FCT (p, string, string_end, no_leading_period, flags) == 0)
1084 return 0;
1085 /* FALLTHROUGH */
1087 case L('+'):
1090 for (rs = string; rs <= string_end; ++rs)
1091 /* First match the prefix with the current pattern with the
1092 current pattern. */
1093 if (FCT (list->str, string, rs, no_leading_period,
1094 flags & FNM_FILE_NAME ? flags : flags & ~FNM_PERIOD) == 0
1095 /* This was successful. Now match the rest with the rest
1096 of the pattern. */
1097 && (FCT (p, rs, string_end,
1098 rs == string
1099 ? no_leading_period
1100 : rs[-1] == '/' && NO_LEADING_PERIOD (flags) ? 1 : 0,
1101 flags & FNM_FILE_NAME
1102 ? flags : flags & ~FNM_PERIOD) == 0
1103 /* This didn't work. Try the whole pattern. */
1104 || (rs != string
1105 && FCT (pattern - 1, rs, string_end,
1106 rs == string
1107 ? no_leading_period
1108 : (rs[-1] == '/' && NO_LEADING_PERIOD (flags)
1109 ? 1 : 0),
1110 flags & FNM_FILE_NAME
1111 ? flags : flags & ~FNM_PERIOD) == 0)))
1112 /* It worked. Signal success. */
1113 return 0;
1115 while ((list = list->next) != NULL);
1117 /* None of the patterns lead to a match. */
1118 return FNM_NOMATCH;
1120 case L('?'):
1121 if (FCT (p, string, string_end, no_leading_period, flags) == 0)
1122 return 0;
1123 /* FALLTHROUGH */
1125 case L('@'):
1127 /* I cannot believe it but `strcat' is actually acceptable
1128 here. Match the entire string with the prefix from the
1129 pattern list and the rest of the pattern following the
1130 pattern list. */
1131 if (FCT (STRCAT (list->str, p), string, string_end,
1132 no_leading_period,
1133 flags & FNM_FILE_NAME ? flags : flags & ~FNM_PERIOD) == 0)
1134 /* It worked. Signal success. */
1135 return 0;
1136 while ((list = list->next) != NULL);
1138 /* None of the patterns lead to a match. */
1139 return FNM_NOMATCH;
1141 case L('!'):
1142 for (rs = string; rs <= string_end; ++rs)
1144 struct patternlist *runp;
1146 for (runp = list; runp != NULL; runp = runp->next)
1147 if (FCT (runp->str, string, rs, no_leading_period,
1148 flags & FNM_FILE_NAME ? flags : flags & ~FNM_PERIOD) == 0)
1149 break;
1151 /* If none of the patterns matched see whether the rest does. */
1152 if (runp == NULL
1153 && (FCT (p, rs, string_end,
1154 rs == string
1155 ? no_leading_period
1156 : rs[-1] == '/' && NO_LEADING_PERIOD (flags) ? 1 : 0,
1157 flags & FNM_FILE_NAME ? flags : flags & ~FNM_PERIOD)
1158 == 0))
1159 /* This is successful. */
1160 return 0;
1163 /* None of the patterns together with the rest of the pattern
1164 lead to a match. */
1165 return FNM_NOMATCH;
1167 default:
1168 assert (! "Invalid extended matching operator");
1169 break;
1172 return -1;
1176 #undef FOLD
1177 #undef CHAR
1178 #undef UCHAR
1179 #undef INT
1180 #undef FCT
1181 #undef EXT
1182 #undef END
1183 #undef MEMPCPY
1184 #undef MEMCHR
1185 #undef STRCOLL
1186 #undef STRLEN
1187 #undef STRCAT
1188 #undef L
1189 #undef BTOWC