> I'd say your patch can be useful only in case the dispatcher subclass
> doesn't send() neither recv() any data, in which case the connection
> is supposed to remain open forever.

There are some cases where it is important to detect that the remote
end is disconnected even if there is no data to send. Say a logger
connected to a data collector that sends data every few minutes. The
data collector dies, the logger may have to take actions on this
event: connect to a backup collector, raise an alarm, whatever... It
should not have to depend on the fact that data needs to be sent to
learn of the disconnection.

> Perhaps you could provide more info about why you needed to do this
> in the first place.

See also issue 12498 and the message 146653. When the remote end
performs a half-duplex disconnection, you may send data without
detecting the close event in send(), so you must rely on recv() to
detect it.