cvs pserver sig11 on 4.8-R

[Andy Newman]

arno at heho.snv.jussieu.fr wrote:
> 
> Could this be a "problem at exit" or something like that?
> 

We see the same thing from, primarially Win32 clients, talking
to a pserver on a 4.x box at work.  At worst we get failed
checkouts (which affects our build system), at best lots of
cruft in the logs.  It is certainly an @exit problem relating
to cleanup of the buffers between the parent/child CVS processes.
In hunting around mail lists people seemed to think it was related
to compression being enabled at the CVS pserver protocol level.
It was also claimed to have been fixed in 1.11.5 but we still
get it with that version.  I was going to spend some time
tracking it down but we're moving to Perforce for various
reasons.