[PATCH 2/1][BZ #866] glob() should match dangling symlinks.

[Ondřej Bílka]

On Thu, Jan 02, 2014 at 05:04:29PM -0500, Mike Frysinger wrote:
> On Saturday 12 October 2013 04:25:29 Ondřej Bílka wrote:
> > This is another bug in glob with patch in bugzilla from Tom Lee ready,
> > see https://sourceware.org/bugzilla/show_bug.cgi?id=866
> > 
> > Patch itself is ok, what is left is decide if listing dangling symlinks
> > is desired behavior.
> 
> it is desired behavior
> 
> > --- a/posix/glob.c
> > +++ b/posix/glob.c
> > @@ -204,7 +204,7 @@ static const char *next_brace_sub (const char *begin,
> > int flags) __THROW;
> > 
> >  static int glob_in_dir (const char *pattern, const char *directory,
> >  			int flags, int (*errfunc) (const char *, int),
> > -			glob_t *pglob, size_t alloca_used);
> > +			glob_t *pglob, size_t alloca_used, int chklnks);
> >  extern int __glob_pattern_type (const char *pattern, int quote)
> >      attribute_hidden;
> > 
> > @@ -1078,7 +1078,7 @@ glob (pattern, flags, errfunc, pglob)
> >  	  status = glob_in_dir (filename, dirs.gl_pathv[i],
> >  				((flags | GLOB_APPEND)
> >  				 & ~(GLOB_NOCHECK | GLOB_NOMAGIC)),
> > -				errfunc, pglob, alloca_used);
> > +				errfunc, pglob, alloca_used, 1);
> 
> what is this scenario where we're still checking the links ?
>
Rationale was not match dangling symlinks when matching directories like
glob ("*/", ...)

I do not remember exact logic now, glob code takes quite bit of effort
to track what happens there.
 
> also, globtest.sh needs updating to make sure this new behavior doesn't 
> regress
> -mike

I send a updated testcase in followup thread, glob is also tested in
tst-gnuglob.c bug-glob[123].c and I modified bug-glob3.c which
previously failed to match current behaviour.