Interface binding - again...

I've just joined the list,  and I've been reading some of the
archives.  One thing that struck me was a similarity to an
idea I've had for some time - namely the way the interface
binding works.

What I've been thinking of was to remove _all_ code for
binding to IPC other than local pipe / unix sockets from
the server.

Then an extension would be written such that open connections
could be handed to the server - fd passing.  This would allow
a helper app to listen to a TCP (IPv4/IPv6 as required) socket
only when required,  and upon deciding to accept the connection
(this could even involve user interaction,  plus initial protocol
manipulation ala SSH behaviour) pass the open descriptor to the
server using this extension.

The fd passing part is simply an enhancement,  as obviously the
connection helper app could simply sit there and shuffle bytes.

However I suspect the extension would still be needed such that
the connection could be sensibly 'named' to the server,  as
opposed to simply looking like a local one.  (I'm not sure on
this,  not having looked too deeping at the can of worms that
is xtrans).

Thoughts?

DF