On Mon, 15 May 2006 [EMAIL PROTECTED] wrote:
> On Sun, 14 May 2006, Elliot Foster wrote:
>
> > For shame, thinking that I sullied Danga.
> >
> > The root cause appears to have been postfix being configured with a
> > (relatively) low timeout on helo/connect. It was not so much that the
> > server
> > was speaking early as much as it was the remote server closing the
> > connection.
> >
> > Perhaps we could put some effort into differentiating an early talker from a
> > "quitter before I say hello?"
>
> You'd have to catch $SIG{PIPE} for this I think. Do we do that already?
If you're going to disconnect the client anyway for being an early talker,
and you know the socket is readable, just try and read a byte to see if
you're at eof or not. One of the parties is hanging up in either case, so
no need preserving that read for parties down the chain.
- Brad
