Re: ssh ControlMaster problem, workaround

[Alan Barrett]
> svn kills the ssh process with SIGKILL, which cannot be trapped.
> This is very rude, giving the ssh process (or a wrapper script) no
> chance to clean up after itself.

Yes, that's one reason the situation I described fails visibly. The
correct solution to _that_ is, IMO, what Debian did in Feb 2006:
patched OpenSSH _not_ to print a message when exiting with a signal.
There's no reason ssh should print the message, as almost no Unix
programs ever do. Some time after, I patched Debian subversion to use
TERM instead of KILL, as the reason for KILL no longer applied to our
systems.

> My recommended fix is described in
> <http://svn.haxx.se/dev/archive-2006-03/0984.shtml>.

Waiting potentially a few seconds after a short svn command-line
transaction would be annoying for client users. Also, if you do have
to send the signal, you still get that unsightly warning message, thus
not addressing the reason the KILL hack was used in the first place.

-- 
Peter Samuelson | org-tld!p12n!peter | http://p12n.org/