OpenSSH 5.1: call for testing

[Corinna Vinschen]

On Jul  8 16:41, Corinna Vinschen wrote:
> On Jul  9 00:18, Damien Miller wrote:
> > On Mon, 7 Jul 2008, Corinna Vinschen wrote:
> > 
> > > Other than that:
> > > 
> > > - session.c, line 427:
> > > 
> > >     #define USE_PIPES
> > > 
> > >   Is that planned or just left over fomr some test?
> > 
> > No, that is planned. We now unconditionally use pipes for communicating
> > with the session subprocesses because they seem to give better semantics
> > for half-closed channels (cf. https://bugzilla.mindrot.org/b/85).
> 
> Ok.  Given that USE_PIPES is now only used in sftp.c, wouldn't it 
> be easier now to always use pipes, even in sftp.c?  That would drop
> the whole test and a bunch of conditional code.
> 
> > > - The following testcases fail on Cygwin 1.5.25:
> > > 
> > >   - addrmatch.sh  tries to run IPv6 tests even though IPv6 is not
> > >                   available.
> > 
> > I'm not sure of a good way to determine at runtime whether IPv6 is
> > available on a platform. Perhaps these tests should be disabled in
> > portable or made non-fatal.
> 
> Ack.  Unfortunately `ssh -6' falls silently back to IPv4 instead of
> complaining on platforms not supporting IPv6.  Complaining would allow
> to use this as a test.

Oh, btw.  Here's another testsuite problem.  Even after I fixed the IPv6
problems in Cygwin I mentioned in the other mail, there's nothing I can
do against this error on Windows up to and including Windows Server 2003:

forwarding.sh:

  exit on -L forward failure, proto 1
  connection not termintated, but should (0)
  [...]
  exit on -L forward failure, proto 2
  connection not termintated, but should (0)

This problem can only be fixed for Windows Vista and Server 2008.  The
underlying problem is a bug in SO_REUSEADDR handling in WinSock.
Usually it should be impossible to reuse a complete duplicate of a local
TCP address (same IP, same port), even if SO_REUSEADDR has been set.
That's unfortunately possible in WinSock and, instead of fixing this
bug, has been defined as "that's how it works here", despite a lot of
other changes in this area over the time.

I have a workaround for IPv4 for a while now, but working around this
for IPv6 requires the existance of a function which is only available
since Windows Vista/2008.  No chance on earlier Windows versions.

I don't know if that fact should really be taken into account in the
testsuite, but I thought I should mention it for the records.


Corinna

-- 
Corinna Vinschen
Cygwin Project Co-Leader
Red Hat