path: root/posix/fnmatch_loop.c

/* Copyright (C) 1991-2020 Free Software Foundation, Inc.
   This file is part of the GNU C Library.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Lesser General Public
   License as published by the Free Software Foundation; either
   version 2.1 of the License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with the GNU C Library; if not, see
   <https://www.gnu.org/licenses/>.  */

#include <stdint.h>

struct STRUCT
{
  const CHAR *pattern;
  const CHAR *string;
  int no_leading_period;
};

/* Match STRING against the filename pattern PATTERN, returning zero if
   it matches, nonzero if not.  */
static int FCT (const CHAR *pattern, const CHAR *string,
		const CHAR *string_end, int no_leading_period, int flags,
		struct STRUCT *ends, size_t alloca_used);
static int EXT (INT opt, const CHAR *pattern, const CHAR *string,
		const CHAR *string_end, int no_leading_period, int flags,
		size_t alloca_used);
static const CHAR *END (const CHAR *patternp);

static int
FCT (const CHAR *pattern, const CHAR *string, const CHAR *string_end,
     int no_leading_period, int flags, struct STRUCT *ends, size_t alloca_used)
{
  const CHAR *p = pattern, *n = string;
  UCHAR c;
#ifdef _LIBC
# if WIDE_CHAR_VERSION
  const char *collseq = (const char *)
    _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
# else
  const UCHAR *collseq = (const UCHAR *)
    _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQMB);
# endif
#endif

  while ((c = *p++) != L('\0'))
    {
      int new_no_leading_period = 0;
      c = FOLD (c);

      switch (c)
	{
	case L('?'):
	  if (__builtin_expect (flags & FNM_EXTMATCH, 0) && *p == '(')
	    {
	      int res = EXT (c, p, n, string_end, no_leading_period,
			     flags, alloca_used);
	      if (res != -1)
		return res;
	    }

	  if (n == string_end)
	    return FNM_NOMATCH;
	  else if (*n == L('/') && (flags & FNM_FILE_NAME))
	    return FNM_NOMATCH;
	  else if (*n == L('.') && no_leading_period)
	    return FNM_NOMATCH;
	  break;

	case L('\\'):
	  if (!(flags & FNM_NOESCAPE))
	    {
	      c = *p++;
	      if (c == L('\0'))
		/* Trailing \ loses.  */
		return FNM_NOMATCH;
	      c = FOLD (c);
	    }
	  if (n == string_end || FOLD ((UCHAR) *n) != c)
	    return FNM_NOMATCH;
	  break;

	case L('*'):
	  if (__builtin_expect (flags & FNM_EXTMATCH, 0) && *p == '(')
	    {
	      int res = EXT (c, p, n, string_end, no_leading_period,
			     flags, alloca_used);
	      if (res != -1)
		return res;
	    }
	  else if (ends != NULL)
	    {
	      ends->pattern = p - 1;
	      ends->string = n;
	      ends->no_leading_period = no_leading_period;
	      return 0;
	    }

	  if (n != string_end && *n == L('.') && no_leading_period)
	    return FNM_NOMATCH;

	  for (c = *p++; c == L('?') || c == L('*'); c = *p++)
	    {
	      if (*p == L('(') && (flags & FNM_EXTMATCH) != 0)
		{
		  const CHAR *endp = END (p);
		  if (endp != p)
		    {
		      /* This is a pattern.  Skip over it.  */
		      p = endp;
		      continue;
		    }
		}

	      if (c == L('?'))
		{
		  /* A ? needs to match one character.  */
		  if (n == string_end)
		    /* There isn't another character; no match.  */
		    return FNM_NOMATCH;
		  else if (*n == L('/')
			   && __builtin_expect (flags & FNM_FILE_NAME, 0))
		    /* A slash does not match a wildcard under
		       FNM_FILE_NAME.  */
		    return FNM_NOMATCH;
		  else
		    /* One character of the string is consumed in matching
		       this ? wildcard, so *??? won't match if there are
		       less than three characters.  */
		    ++n;
		}
	    }

	  if (c == L('\0'))
	    /* The wildcard(s) is/are the last element of the pattern.
	       If the name is a file name and contains another slash
	       this means it cannot match, unless the FNM_LEADING_DIR
	       flag is set.  */
	    {
	      int result = (flags & FNM_FILE_NAME) == 0 ? 0 : FNM_NOMATCH;

	      if (flags & FNM_FILE_NAME)
		{
		  if (flags & FNM_LEADING_DIR)
		    result = 0;
		  else
		    {
		      if (MEMCHR (n, L('/'), string_end - n) == NULL)
			result = 0;
		    }
		}

	      return result;
	    }
	  else
	    {
	      const CHAR *endp;
	      struct STRUCT end;

	      end.pattern = NULL;
	      endp = MEMCHR (n, (flags & FNM_FILE_NAME) ? L('/') : L('\0'),
			     string_end - n);
	      if (endp == NULL)
		endp = string_end;

	      if (c == L('[')
		  || (__builtin_expect (flags & FNM_EXTMATCH, 0) != 0
		      && (c == L('@') || c == L('+') || c == L('!'))
		      && *p == L('(')))
		{
		  int flags2 = ((flags & FNM_FILE_NAME)
				? flags : (flags & ~FNM_PERIOD));

		  for (--p; n < endp; ++n, no_leading_period = 0)
		    if (FCT (p, n, string_end, no_leading_period, flags2,
			     &end, alloca_used) == 0)
		      goto found;
		}
	      else if (c == L('/') && (flags & FNM_FILE_NAME))
		{
		  while (n < string_end && *n != L('/'))
		    ++n;
		  if (n < string_end && *n == L('/')
		      && (FCT (p, n + 1, string_end, flags & FNM_PERIOD, flags,
			       NULL, alloca_used) == 0))
		    return 0;
		}
	      else
		{
		  int flags2 = ((flags & FNM_FILE_NAME)
				? flags : (flags & ~FNM_PERIOD));

		  if (c == L('\\') && !(flags & FNM_NOESCAPE))
		    c = *p;
		  c = FOLD (c);
		  for (--p; n < endp; ++n, no_leading_period = 0)
		    if (FOLD ((UCHAR) *n) == c
			&& (FCT (p, n, string_end, no_leading_period, flags2,
				 &end, alloca_used) == 0))
		      {
		      found:
			if (end.pattern == NULL)
			  return 0;
			break;
		      }
		  if (end.pattern != NULL)
		    {
		      p = end.pattern;
		      n = end.string;
		      no_leading_period = end.no_leading_period;
		      continue;
		    }
		}
	    }

	  /* If we come here no match is possible with the wildcard.  */
	  return FNM_NOMATCH;

	case L('['):
	  {
	    /* Nonzero if the sense of the character class is inverted.  */
	    const CHAR *p_init = p;
	    const CHAR *n_init = n;
	    int not;
	    CHAR cold;
	    UCHAR fn;

	    if (posixly_correct == 0)
	      posixly_correct = getenv ("POSIXLY_CORRECT") != NULL ? 1 : -1;

	    if (n == string_end)
	      return FNM_NOMATCH;

	    if (*n == L('.') && no_leading_period)
	      return FNM_NOMATCH;

	    if (*n == L('/') && (flags & FNM_FILE_NAME))
	      /* `/' cannot be matched.  */
	      return FNM_NOMATCH;

	    not = (*p == L('!') || (posixly_correct < 0 && *p == L('^')));
	    if (not)
	      ++p;

	    fn = FOLD ((UCHAR) *n);

	    c = *p++;
	    for (;;)
	      {
		if (!(flags & FNM_NOESCAPE) && c == L('\\'))
		  {
		    if (*p == L('\0'))
		      return FNM_NOMATCH;
		    c = FOLD ((UCHAR) *p);
		    ++p;

		    goto normal_bracket;
		  }
		else if (c == L('[') && *p == L(':'))
		  {
		    /* Leave room for the null.  */
		    CHAR str[CHAR_CLASS_MAX_LENGTH + 1];
		    size_t c1 = 0;
#if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
		    wctype_t wt;
#endif
		    const CHAR *startp = p;

		    for (;;)
		      {
			if (c1 == CHAR_CLASS_MAX_LENGTH)
			  /* The name is too long and therefore the pattern
			     is ill-formed.  */
			  return FNM_NOMATCH;

			c = *++p;
			if (c == L(':') && p[1] == L(']'))
			  {
			    p += 2;
			    break;
			  }
			if (c < L('a') || c >= L('z'))
			  {
			    /* This cannot possibly be a character class name.
			       Match it as a normal range.  */
			    p = startp;
			    c = L('[');
			    goto normal_bracket;
			  }
			str[c1++] = c;
		      }
		    str[c1] = L('\0');

#if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
		    wt = IS_CHAR_CLASS (str);
		    if (wt == 0)
		      /* Invalid character class name.  */
		      return FNM_NOMATCH;

# if defined _LIBC && ! WIDE_CHAR_VERSION
		    /* The following code is glibc specific but does
		       there a good job in speeding up the code since
		       we can avoid the btowc() call.  */
		    if (_ISCTYPE ((UCHAR) *n, wt))
		      goto matched;
# else
		    if (ISWCTYPE (BTOWC ((UCHAR) *n), wt))
		      goto matched;
# endif
#else
		    if ((STREQ (str, L("alnum")) && ISALNUM ((UCHAR) *n))
			|| (STREQ (str, L("alpha")) && ISALPHA ((UCHAR) *n))
			|| (STREQ (str, L("blank")) && ISBLANK ((UCHAR) *n))
			|| (STREQ (str, L("cntrl")) && ISCNTRL ((UCHAR) *n))
			|| (STREQ (str, L("digit")) && ISDIGIT ((UCHAR) *n))
			|| (STREQ (str, L("graph")) && ISGRAPH ((UCHAR) *n))
			|| (STREQ (str, L("lower")) && ISLOWER ((UCHAR) *n))
			|| (STREQ (str, L("print")) && ISPRINT ((UCHAR) *n))
			|| (STREQ (str, L("punct")) && ISPUNCT ((UCHAR) *n))
			|| (STREQ (str, L("space")) && ISSPACE ((UCHAR) *n))
			|| (STREQ (str, L("upper")) && ISUPPER ((UCHAR) *n))
			|| (STREQ (str, L("xdigit")) && ISXDIGIT ((UCHAR) *n)))
		      goto matched;
#endif
		    c = *p++;
		  }
#ifdef _LIBC
		else if (c == L('[') && *p == L('='))
		  {
		    /* It's important that STR be a scalar variable rather
		       than a one-element array, because GCC (at least 4.9.2
		       -O2 on x86-64) can be confused by the array and
		       diagnose a "used initialized" in a dead branch in the
		       findidx function.  */
		    UCHAR str;
		    uint32_t nrules =
		      _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
		    const CHAR *startp = p;

		    c = *++p;
		    if (c == L('\0'))
		      {
			p = startp;
			c = L('[');
			goto normal_bracket;
		      }
		    str = c;

		    c = *++p;
		    if (c != L('=') || p[1] != L(']'))
		      {
			p = startp;
			c = L('[');
			goto normal_bracket;
		      }
		    p += 2;

		    if (nrules == 0)
		      {
			if ((UCHAR) *n == str)
			  goto matched;
		      }
		    else
		      {
			const int32_t *table;
# if WIDE_CHAR_VERSION
			const int32_t *weights;
			const wint_t *extra;
# else
			const unsigned char *weights;
			const unsigned char *extra;
# endif
			const int32_t *indirect;
			int32_t idx;
			const UCHAR *cp = (const UCHAR *) &str;

# if WIDE_CHAR_VERSION
			table = (const int32_t *)
			  _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
			weights = (const int32_t *)
			  _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
			extra = (const wint_t *)
			  _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
			indirect = (const int32_t *)
			  _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
# else
			table = (const int32_t *)
			  _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
			weights = (const unsigned char *)
			  _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
			extra = (const unsigned char *)
			  _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
			indirect = (const int32_t *)
			  _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
# endif

			idx = FINDIDX (table, indirect, extra, &cp, 1);
			if (idx != 0)
			  {
			    /* We found a table entry.  Now see whether the
			       character we are currently at has the same
			       equivalance class value.  */
			    int len = weights[idx & 0xffffff];
			    int32_t idx2;
			    const UCHAR *np = (const UCHAR *) n;

			    idx2 = FINDIDX (table, indirect, extra,
					    &np, string_end - n);
			    if (idx2 != 0
				&& (idx >> 24) == (idx2 >> 24)
				&& len == weights[idx2 & 0xffffff])
			      {
				int cnt = 0;

				idx &= 0xffffff;
				idx2 &= 0xffffff;

				while (cnt < len
				       && (weights[idx + 1 + cnt]
					   == weights[idx2 + 1 + cnt]))
				  ++cnt;

				if (cnt == len)
				  goto matched;
			      }
			  }
		      }

		    c = *p++;
		  }
#endif
		else if (c == L('\0'))
		  {
		    /* [ unterminated, treat as normal character.  */
		    p = p_init;
		    n = n_init;
		    c = L('[');
		    goto normal_match;
		  }
		else
		  {
		    int is_range = 0;

#ifdef _LIBC
		    int is_seqval = 0;

		    if (c == L('[') && *p == L('.'))
		      {
			uint32_t nrules =
			  _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
			const CHAR *startp = p;
			size_t c1 = 0;

			while (1)
			  {
			    c = *++p;
			    if (c == L('.') && p[1] == L(']'))
			      {
				p += 2;
				break;
			      }
			    if (c == '\0')
			      return FNM_NOMATCH;
			    ++c1;
			  }

			/* We have to handling the symbols differently in
			   ranges since then the collation sequence is
			   important.  */
			is_range = *p == L('-') && p[1] != L('\0');

			if (nrules == 0)
			  {
			    /* There are no names defined in the collation
			       data.  Therefore we only accept the trivial
			       names consisting of the character itself.  */
			    if (c1 != 1)
			      return FNM_NOMATCH;

			    if (!is_range && *n == startp[1])
			      goto matched;

			    cold = startp[1];
			    c = *p++;
			  }
			else
			  {
			    int32_t table_size;
			    const int32_t *symb_table;
			    const unsigned char *extra;
			    int32_t idx;
			    int32_t elem;
# if WIDE_CHAR_VERSION
			    CHAR *wextra;
# endif

			    table_size =
			      _NL_CURRENT_WORD (LC_COLLATE,
						_NL_COLLATE_SYMB_HASH_SIZEMB);
			    symb_table = (const int32_t *)
			      _NL_CURRENT (LC_COLLATE,
					   _NL_COLLATE_SYMB_TABLEMB);
			    extra = (const unsigned char *)
			      _NL_CURRENT (LC_COLLATE,
					   _NL_COLLATE_SYMB_EXTRAMB);

			    for (elem = 0; elem < table_size; elem++)
			      if (symb_table[2 * elem] != 0)
				{
				  idx = symb_table[2 * elem + 1];
				  /* Skip the name of collating element.  */
				  idx += 1 + extra[idx];
# if WIDE_CHAR_VERSION
				  /* Skip the byte sequence of the
				     collating element.  */
				  idx += 1 + extra[idx];
				  /* Adjust for the alignment.  */
				  idx = (idx + 3) & ~3;

				  wextra = (CHAR *) &extra[idx + 4];

				  if (/* Compare the length of the sequence.  */
				      c1 == wextra[0]
				      /* Compare the wide char sequence.  */
				      && WMEMCMP (startp + 1, &wextra[1],
						  c1) == 0)
				    /* Yep, this is the entry.  */
				    break;
# else
				  if (/* Compare the length of the sequence.  */
				      c1 == extra[idx]
				      /* Compare the byte sequence.  */
				      && memcmp (startp + 1,
						 &extra[idx + 1], c1) == 0)
				    /* Yep, this is the entry.  */
				    break;
# endif
				}

			    if (elem < table_size)
			      {
				/* Compare the byte sequence but only if
				   this is not part of a range.  */
				if (! is_range

# if WIDE_CHAR_VERSION
				    && WMEMCMP (n, &wextra[1], c1) == 0
# else
				    && memcmp (n, &extra[idx + 1], c1) == 0
# endif
				    )
				  {
				    n += c1 - 1;
				    goto matched;
				  }

				/* Get the collation sequence value.  */
				is_seqval = 1;
# if WIDE_CHAR_VERSION
				cold = wextra[1 + wextra[0]];
# else
				idx += 1 + extra[idx];
				/* Adjust for the alignment.  */
				idx = (idx + 3) & ~3;
				cold = *((int32_t *) &extra[idx]);
# endif

				c = *p++;
			      }
			    else if (c1 == 1)
			      {
				/* No valid character.  Match it as a
				   single byte.  */
				if (!is_range && *n == startp[1])
				  goto matched;

				cold = startp[1];
				c = *p++;
			      }
			    else
			      return FNM_NOMATCH;
			  }
		      }
		    else
#endif
		      {
			c = FOLD (c);
		      normal_bracket:

			/* We have to handling the symbols differently in
			   ranges since then the collation sequence is
			   important.  */
			is_range = (*p == L('-') && p[1] != L('\0')
				    && p[1] != L(']'));

			if (!is_range && c == fn)
			  goto matched;

			/* This is needed if we goto normal_bracket; from
			   outside of is_seqval's scope.  */
			is_seqval = 0;
			cold = c;
			c = *p++;
		      }

		    if (c == L('-') && *p != L(']'))
		      {
#if _LIBC
			/* We have to find the collation sequence
			   value for C.  Collation sequence is nothing
			   we can regularly access.  The sequence
			   value is defined by the order in which the
			   definitions of the collation values for the
			   various characters appear in the source
			   file.  A strange concept, nowhere
			   documented.  */
			uint32_t fcollseq;
			uint32_t lcollseq;
			UCHAR cend = *p++;

# if WIDE_CHAR_VERSION
			/* Search in the `names' array for the characters.  */
			fcollseq = __collseq_table_lookup (collseq, fn);
			if (fcollseq == ~((uint32_t) 0))
			  /* XXX We don't know anything about the character
			     we are supposed to match.  This means we are
			     failing.  */
			  goto range_not_matched;

			if (is_seqval)
			  lcollseq = cold;
			else
			  lcollseq = __collseq_table_lookup (collseq, cold);
# else
			fcollseq = collseq[fn];
			lcollseq = is_seqval ? cold : collseq[(UCHAR) cold];
# endif

			is_seqval = 0;
			if (cend == L('[') && *p == L('.'))
			  {
			    uint32_t nrules =
			      _NL_CURRENT_WORD (LC_COLLATE,
						_NL_COLLATE_NRULES);
			    const CHAR *startp = p;
			    size_t c1 = 0;

			    while (1)
			      {
				c = *++p;
				if (c == L('.') && p[1] == L(']'))
				  {
				    p += 2;
				    break;
				  }
				if (c == '\0')
				  return FNM_NOMATCH;
				++c1;
			      }

			    if (nrules == 0)
			      {
				/* There are no names defined in the
				   collation data.  Therefore we only
				   accept the trivial names consisting
				   of the character itself.  */
				if (c1 != 1)
				  return FNM_NOMATCH;

				cend = startp[1];
			      }
			    else
			      {
				int32_t table_size;
				const int32_t *symb_table;
				const unsigned char *extra;
				int32_t idx;
				int32_t elem;
# if WIDE_CHAR_VERSION
				CHAR *wextra;
# endif

				table_size =
				  _NL_CURRENT_WORD (LC_COLLATE,
						    _NL_COLLATE_SYMB_HASH_SIZEMB);
				symb_table = (const int32_t *)
				  _NL_CURRENT (LC_COLLATE,
					       _NL_COLLATE_SYMB_TABLEMB);
				extra = (const unsigned char *)
				  _NL_CURRENT (LC_COLLATE,
					       _NL_COLLATE_SYMB_EXTRAMB);

				for (elem = 0; elem < table_size; elem++)
				  if (symb_table[2 * elem] != 0)
				    {
				      idx = symb_table[2 * elem + 1];
				      /* Skip the name of collating
					 element.  */
				      idx += 1 + extra[idx];
# if WIDE_CHAR_VERSION
				      /* Skip the byte sequence of the
					 collating element.  */
				      idx += 1 + extra[idx];
				      /* Adjust for the alignment.  */
				      idx = (idx + 3) & ~3;

				      wextra = (CHAR *) &extra[idx + 4];

				      if (/* Compare the length of the
					     sequence.  */
					  c1 == wextra[0]
					  /* Compare the wide char sequence.  */
					  && WMEMCMP (startp + 1, &wextra[1],
						      c1) == 0)
					/* Yep, this is the entry.  */
					break;
# else
				      if (/* Compare the length of the
					     sequence.  */
					  c1 == extra[idx]
					  /* Compare the byte sequence.  */
					  && memcmp (startp + 1,
						     &extra[idx + 1], c1) == 0)
					/* Yep, this is the entry.  */
					break;
# endif
				    }

				if (elem < table_size)
				  {
				    /* Get the collation sequence value.  */
				    is_seqval = 1;
# if WIDE_CHAR_VERSION
				    cend = wextra[1 + wextra[0]];
# else
				    idx += 1 + extra[idx];
				    /* Adjust for the alignment.  */
				    idx = (idx + 3) & ~3;
				    cend = *((int32_t *) &extra[idx]);
# endif
				  }
				else if (c1 == 1)
				  {
				    cend = startp[1];
				    c = *p++;
				  }
				else
				  return FNM_NOMATCH;
			      }
			  }
			else
			  {
			    if (!(flags & FNM_NOESCAPE) && cend == L('\\'))
			      cend = *p++;
			    if (cend == L('\0'))
			      return FNM_NOMATCH;
			    cend = FOLD (cend);
			  }

			/* XXX It is not entirely clear to me how to handle
			   characters which are not mentioned in the
			   collation specification.  */
			if (
# if WIDE_CHAR_VERSION
			    lcollseq == 0xffffffff ||
# endif
			    lcollseq <= fcollseq)
			  {
			    /* We have to look at the upper bound.  */
			    uint32_t hcollseq;

			    if (is_seqval)
			      hcollseq = cend;
			    else
			      {
# if WIDE_CHAR_VERSION
				hcollseq =
				  __collseq_table_lookup (collseq, cend);
				if (hcollseq == ~((uint32_t) 0))
				  {
				    /* Hum, no information about the upper
				       bound.  The matching succeeds if the
				       lower bound is matched exactly.  */
				    if (lcollseq != fcollseq)
				      goto range_not_matched;

				    goto matched;
				  }
# else
				hcollseq = collseq[cend];
# endif
			      }

			    if (lcollseq <= hcollseq && fcollseq <= hcollseq)
			      goto matched;
			  }
# if WIDE_CHAR_VERSION
		      range_not_matched:
# endif
#else
			/* We use a boring value comparison of the character
			   values.  This is better than comparing using
			   `strcoll' since the latter would have surprising
			   and sometimes fatal consequences.  */
			UCHAR cend = *p++;

			if (!(flags & FNM_NOESCAPE) && cend == L('\\'))
			  cend = *p++;
			if (cend == L('\0'))
			  return FNM_NOMATCH;

			/* It is a range.  */
			if (cold <= fn && fn <= cend)
			  goto matched;
#endif

			c = *p++;
		      }
		  }

		if (c == L(']'))
		  break;
	      }

	    if (!not)
	      return FNM_NOMATCH;
	    break;

	  matched:
	    /* Skip the rest of the [...] that already matched.  */
	    while ((c = *p++) != L (']'))
	      {
		if (c == L('\0'))
		  /* [... (unterminated) loses.  */
		  return FNM_NOMATCH;

		if (!(flags & FNM_NOESCAPE) && c == L('\\'))
		  {
		    if (*p == L('\0'))
		      return FNM_NOMATCH;
		    /* XXX 1003.2d11 is unclear if this is right.  */
		    ++p;
		  }
		else if (c == L('[') && *p == L(':'))
		  {
		    int c1 = 0;
		    const CHAR *startp = p;

		    while (1)
		      {
			c = *++p;
			if (++c1 == CHAR_CLASS_MAX_LENGTH)
			  return FNM_NOMATCH;

			if (*p == L(':') && p[1] == L(']'))
			  break;

			if (c < L('a') || c >= L('z'))
			  {
			    p = startp - 2;
			    break;
			  }
		      }
		    p += 2;
		  }
		else if (c == L('[') && *p == L('='))
		  {
		    c = *++p;
		    if (c == L('\0'))
		      return FNM_NOMATCH;
		    c = *++p;
		    if (c != L('=') || p[1] != L(']'))
		      return FNM_NOMATCH;
		    p += 2;
		  }
		else if (c == L('[') && *p == L('.'))
		  {
		    while (1)
		      {
			c = *++p;
			if (c == L('\0'))
			  return FNM_NOMATCH;

			if (c == L('.') && p[1] == L(']'))
			  break;
		      }
		    p += 2;
		  }
	      }
	    if (not)
	      return FNM_NOMATCH;
	  }
	  break;

	case L('+'):
	case L('@'):
	case L('!'):
	  if (__builtin_expect (flags & FNM_EXTMATCH, 0) && *p == '(')
	    {
	      int res = EXT (c, p, n, string_end, no_leading_period, flags,
			     alloca_used);
	      if (res != -1)
		return res;
	    }
	  goto normal_match;

	case L('/'):
	  if (NO_LEADING_PERIOD (flags))
	    {
	      if (n == string_end || c != (UCHAR) *n)
		return FNM_NOMATCH;

	      new_no_leading_period = 1;
	      break;
	    }
	  /* FALLTHROUGH */
	default:
	normal_match:
	  if (n == string_end || c != FOLD ((UCHAR) *n))
	    return FNM_NOMATCH;
	}

      no_leading_period = new_no_leading_period;
      ++n;
    }

  if (n == string_end)
    return 0;

  if ((flags & FNM_LEADING_DIR) && n != string_end && *n == L('/'))
    /* The FNM_LEADING_DIR flag says that "foo*" matches "foobar/frobozz".  */
    return 0;

  return FNM_NOMATCH;
}


static const CHAR *
END (const CHAR *pattern)
{
  const CHAR *p = pattern;

  while (1)
    if (*++p == L('\0'))
      /* This is an invalid pattern.  */
      return pattern;
    else if (*p == L('['))
      {
	/* Handle brackets special.  */
	if (posixly_correct == 0)
	  posixly_correct = getenv ("POSIXLY_CORRECT") != NULL ? 1 : -1;

	/* Skip the not sign.  We have to recognize it because of a possibly
	   following ']'.  */
	if (*++p == L('!') || (posixly_correct < 0 && *p == L('^')))
	  ++p;
	/* A leading ']' is recognized as such.  */
	if (*p == L(']'))
	  ++p;
	/* Skip over all characters of the list.  */
	while (*p != L(']'))
	  if (*p++ == L('\0'))
	    /* This is no valid pattern.  */
	    return pattern;
      }
    else if ((*p == L('?') || *p == L('*') || *p == L('+') || *p == L('@')
	      || *p == L('!')) && p[1] == L('('))
      {
	p = END (p + 1);
	if (*p == L('\0'))
	  /* This is an invalid pattern.  */
	  return pattern;
      }
    else if (*p == L(')'))
      break;

  return p + 1;
}


static int
EXT (INT opt, const CHAR *pattern, const CHAR *string, const CHAR *string_end,
     int no_leading_period, int flags, size_t alloca_used)
{
  const CHAR *startp;
  int level;
  struct patternlist
  {
    struct patternlist *next;
    CHAR malloced;
    CHAR str[0];
  } *list = NULL;
  struct patternlist **lastp = &list;
  size_t pattern_len = STRLEN (pattern);
  int any_malloced = 0;
  const CHAR *p;
  const CHAR *rs;
  int retval = 0;

  /* Parse the pattern.  Store the individual parts in the list.  */
  level = 0;
  for (startp = p = pattern + 1; level >= 0; ++p)
    if (*p == L('\0'))
      {
	/* This is an invalid pattern.  */
	retval = -1;
	goto out;
      }
    else if (*p == L('['))
      {
	/* Handle brackets special.  */
	if (posixly_correct == 0)
	  posixly_correct = getenv ("POSIXLY_CORRECT") != NULL ? 1 : -1;

	/* Skip the not sign.  We have to recognize it because of a possibly
	   following ']'.  */
	if (*++p == L('!') || (posixly_correct < 0 && *p == L('^')))
	  ++p;
	/* A leading ']' is recognized as such.  */
	if (*p == L(']'))
	  ++p;
	/* Skip over all characters of the list.  */
	while (*p != L(']'))
	  if (*p++ == L('\0'))
	    {
	      /* This is no valid pattern.  */
	      retval = -1;
	      goto out;
	    }
      }
    else if ((*p == L('?') || *p == L('*') || *p == L('+') || *p == L('@')
	      || *p == L('!')) && p[1] == L('('))
      /* Remember the nesting level.  */
      ++level;
    else if (*p == L(')'))
      {
	if (level-- == 0)
	  {
	    /* This means we found the end of the pattern.  */
#define NEW_PATTERN \
	    struct patternlist *newp;					      \
	    size_t slen = (opt == L('?') || opt == L('@')		      \
			   ? pattern_len : (p - startp + 1));		      \
	    slen = sizeof (struct patternlist) + (slen * sizeof (CHAR));      \
	    int malloced = ! __libc_use_alloca (alloca_used + slen);	      \
	    if (__builtin_expect (malloced, 0))				      \
	      {								      \
		newp = malloc (slen);					      \
		if (newp == NULL)					      \
		  {							      \
		    retval = -2;					      \
		    goto out;						      \
		  }							      \
		any_malloced = 1;					      \
	      }								      \
	    else							      \
	      newp = alloca_account (slen, alloca_used);		      \
	    newp->next = NULL;						      \
	    newp->malloced = malloced;					      \
	    *((CHAR *) MEMPCPY (newp->str, startp, p - startp)) = L('\0');    \
	    *lastp = newp;						      \
	    lastp = &newp->next
	    NEW_PATTERN;
	  }
      }
    else if (*p == L('|'))
      {
	if (level == 0)
	  {
	    NEW_PATTERN;
	    startp = p + 1;
	  }
      }
  assert (list != NULL);
  assert (p[-1] == L(')'));
#undef NEW_PATTERN

  switch (opt)
    {
    case L('*'):
      if (FCT (p, string, string_end, no_leading_period, flags, NULL,
	       alloca_used) == 0)
	goto success;
      /* FALLTHROUGH */

    case L('+'):
      do
	{
	  for (rs = string; rs <= string_end; ++rs)
	    /* First match the prefix with the current pattern with the
	       current pattern.  */
	    if (FCT (list->str, string, rs, no_leading_period,
		     flags & FNM_FILE_NAME ? flags : flags & ~FNM_PERIOD,
		     NULL, alloca_used) == 0
		/* This was successful.  Now match the rest with the rest
		   of the pattern.  */
		&& (FCT (p, rs, string_end,
			 rs == string
			 ? no_leading_period
			 : rs[-1] == '/' && NO_LEADING_PERIOD (flags) ? 1 : 0,
			 flags & FNM_FILE_NAME
			 ? flags : flags & ~FNM_PERIOD, NULL, alloca_used) == 0
		    /* This didn't work.  Try the whole pattern.  */
		    || (rs != string
			&& FCT (pattern - 1, rs, string_end,
				rs == string
				? no_leading_period
				: (rs[-1] == '/' && NO_LEADING_PERIOD (flags)
				   ? 1 : 0),
				flags & FNM_FILE_NAME
				? flags : flags & ~FNM_PERIOD, NULL,
				alloca_used) == 0)))
	      /* It worked.  Signal success.  */
	      goto success;
	}
      while ((list = list->next) != NULL);

      /* None of the patterns lead to a match.  */
      retval = FNM_NOMATCH;
      break;

    case L('?'):
      if (FCT (p, string, string_end, no_leading_period, flags, NULL,
	       alloca_used) == 0)
	goto success;
      /* FALLTHROUGH */

    case L('@'):
      do
	/* I cannot believe it but `strcat' is actually acceptable
	   here.  Match the entire string with the prefix from the
	   pattern list and the rest of the pattern following the
	   pattern list.  */
	if (FCT (STRCAT (list->str, p), string, string_end,
		 no_leading_period,
		 flags & FNM_FILE_NAME ? flags : flags & ~FNM_PERIOD,
		 NULL, alloca_used) == 0)
	  /* It worked.  Signal success.  */
	  goto success;
      while ((list = list->next) != NULL);

      /* None of the patterns lead to a match.  */
      retval = FNM_NOMATCH;
      break;

    case L('!'):
      for (rs = string; rs <= string_end; ++rs)
	{
	  struct patternlist *runp;

	  for (runp = list; runp != NULL; runp = runp->next)
	    if (FCT (runp->str, string, rs,  no_leading_period,
		     flags & FNM_FILE_NAME ? flags : flags & ~FNM_PERIOD,
		     NULL, alloca_used) == 0)
	      break;

	  /* If none of the patterns matched see whether the rest does.  */
	  if (runp == NULL
	      && (FCT (p, rs, string_end,
		       rs == string
		       ? no_leading_period
		       : rs[-1] == '/' && NO_LEADING_PERIOD (flags) ? 1 : 0,
		       flags & FNM_FILE_NAME ? flags : flags & ~FNM_PERIOD,
		       NULL, alloca_used) == 0))
	    /* This is successful.  */
	    goto success;
	}

      /* None of the patterns together with the rest of the pattern
	 lead to a match.  */
      retval = FNM_NOMATCH;
      break;

    default:
      assert (! "Invalid extended matching operator");
      retval = -1;
      break;
    }

 success:
 out:
  if (any_malloced)
    while (list != NULL)
      {
	struct patternlist *old = list;
	list = list->next;
	if (old->malloced)
	  free (old);
      }

  return retval;
}


#undef FOLD
#undef CHAR
#undef UCHAR
#undef INT
#undef FCT
#undef EXT
#undef END
#undef STRUCT
#undef MEMPCPY
#undef MEMCHR
#undef STRCOLL
#undef STRLEN
#undef STRCAT
#undef L
#undef BTOWC
#undef WIDE_CHAR_VERSION
#undef FINDIDX