On Tue, 28 May 2002, Martin Husemann wrote:

> > The problem is we have to do this while treating the cookie we get from
> > the NFS layer as an opaque object. :-|
>
> The point I'm still missing here is: how can a linux system as a client to
> the same server possibly work at all?

Depends on what the server is. I expect that linux's client works on their
server. So our code must work when talking to a linux server.

If it didn't work, I'd expect this problem would have been fixed long ago.
:-)

> If it doesn't, we're done - our emulation is perfect.
>
> If it works (due to kernel hacks, as has been suggested) we should look at
> those before continuing this thread.

I think it works because of a fundamental difference of definition. We
(and the NFS spec) define the cookie as opaque, whereas Linux does not.
Given their definition, what glibc does isn't wrong. So it's kinda hard
calling what their kernel does a "hack." Their own thing to be sure...

Take care,

Bill