Re: cvs 1.11.9 for windows hanging

[Paul Edwards]

"Paul Edwards" <kerravon@nosppaam.w3.to> wrote in message 
news:LLNob.173917$bo1.30094@news-server.bigpond.net.au...
> > Well, the problem appears to be with the windows client, but as an aid
> > to tracking down where exactly things are going wrong may you may find
> > that replicating the problem locally will let you debug the problem
> > more easily.
>
> I have tracked the problem down to CVS is hanging in buffer.c
> while waiting for the child process.
>
> That child process happens to be SSH, which is running a
> command "cvs server", presumably on the remote site.
>
> Therefore it is out of scope.  The problem is on the CVS
> server, or potentially in SSH.  Or Windows 98 etc etc.

Ok, I've now tracked down on the CVS 1.12.2 side, where it
is hanging.  It is hanging on the getc() in function
stdio_buffer_input().

The reason things work with compression is because the
compression has its own special status for "end of input",
which doesn't rely on the transport layer.

On the client side, in stdio_buffer_shutdown, after the
fclose(fp) is done, I assume that that is meant to signal to
the external processor that it is end of file.  I don't know
how this mechanism works.  I assume that they share the
file pointers or something, so that a close on one should
trigger ssh to realise it has lost its communication partner.

Either way its outside the scope of being a CVS code
error, as far as I am aware, and into the transport layer.
As such, I don't know what the solution is.  Maybe it is a
bug in ssh.  pserver presumably works because windows
handles the fclose() properly.

There is of course the SHUTDOWN_SERVER() macro
if a hack is to be inserted, but the hack would potentially
interfere with correct pserver operations.

Thanks Mark for providing the system to test this on.  After
setting it up a good testing environment on the Unix system,
tonight I was able to debug in peace, all from the DOS
prompt using micro-emacs without having to stuff around
with the weird terminal problems (not solved with
stty -ocrnl, the problem is to do with scrolling, I'm getting
cr for text at the top of the screen).  :-)

BFN.  Paul.