Re: [MinGW-dvlpr] [MSYS] - normal open on a directory.

[Max T. W. <ma...@mt...>]

"Max T. Woodbury" wrote:
> 
> Earnie Boyd wrote:
> >
> > On 5:34:48 pm 2005-06-29 "Max T. Woodbury" <ma...@mt...> wrote:
> > > "Max T. Woodbury" wrote:
> > > >
> > > >  For the record it blows up on W98SE.  Permission denied on current
> > > >  directory or words to that effect.  Virtually identical to what I
> > > >  ran aground on.
> > > >
> > > >  I've been digging through the various ways things get 'open'ed and
> > > >  all the disk stuff seems to go through fhandler_base::open,
> > > >  including opendir.  I've tried alternative 'access' and 'share'
> > > >  values there with no luck.
> > > >
> > > >  I've seen open (".", O_RDONLY) fail, but opendir ends up in the
> > > >  same code and works.  It might have to do with the form of the
> > > >  name.  Let's see... PC_FULL and path_conv::check looks like the
> > > >  way to check that in syscalls.cc (_open). Can you call 'check' a
> > > >  second time without loosing strings and stuff?  Oh well, that's
> > > >  just a bit more to read.
> > > >
> > > >  I think a program to test the various directory access routines is
> > > >  probably the next thing to put in place if this last test fails.
> > >
> > > Not quite yet but this is still in-queue.
> > >
> > > >  I've backed off on the subject line since it looks like it'll only
> > > >  take one small fix (even if what that fix is is not known at the
> > > > moment).
> > >
> > > I tried a simple minded change to fhandler_disk<whatever>::open to
> > > make it look a bit more like opendir but it didn't work.  I'm
> > > building the debug .dll at the moment.  I'm probably going to need a
> > > little advice on how to use it but I'll wait on specific questions
> > > until I've tried it once.
> 
> I kludged fhandler_disk::open to return success if fhandler::open failed,
> it is not WinNT and the path is a directory.  Even with a broken file handle
> find.exe then worked properly.  I'll check fileutils shortly.  However
> I am not completely comfortable with this kluge.  Should other things be
> checked like O_RDONLY?
> 
> And thanks Christopher for the information on Cygwin.  It seems a separate
> flag is not needed, just the fact that the handle is not valid (that is
> ~0) seems to be sufficient.

Further testing indicates that this patch introduces other problems.  It
may be possible to add more work-arounds, but it is not sufficient by
itself to get all this working.  PLEASE back this patch out, at least for
now.

mt...@us...