BUG : EM streams messed up when one of many in-script connections closed.

[mistergibson]

Background:

I am attempting to support mutliple xmpp client objects running in the same script. I need, for my use case, for this to work without conflicts. I can get the client objects to connect under their respective jids and send chat messages. However, if I disconnect just ONE of those client objects -- ALL traffic of ALL clients gets utterly hosed in EM. In fact, it appears that EM is actually shut down for no good reason, and I have to hacky-hack a restart - not that it helps the messed up streams any.

So:

I'm wondering if there is a method I should override for alternate behaviour when closing streams such that EM will be left running happily along as before? I will need that situation to sustain support for the other connected client objects. Also, do you know of a smoother way to restart EM if it is buggered for any reason (just wondering).

[paradisaeidae]

XMPP connections are opened as bi-synchronous/bidirectional xml streams.
Opened at each end, the root level.
If the xml is broken, the stream breaks, connection unwinds.
Very strong level of connection rigor.
This requires a fine granularity of processing with a connection.
I'm unsure of the finer detail myself. Still learning here.
Though seeing the language architecture of Erlang produce robust xmpp server the 'Erlang application' is a fine grained encapsulation of such.
The EM listener is a port level listener.
I've not read the code though it appears Erlang beams are able to listen per connection.
Might be each connection has a stack?
So, pushing the abilities of Ruby to get this.
Some have fun trying!
https://github.com/igrigorik/em-synchrony
There may be somewhere there a rescue might be placed.
Find the line 'This is dangerous territory - you've been warned.'

[mistergibson]

Well, as a work-around I overrode unbind method to do everything but shutdown EM, and it works fine now:

    def unbind
        # Fix bug where closing connection stops EM for ALL Client Instances.
        # call_handler_for(:disconnected, nil) || (EM.reactor_running? && EM.stop)
        call_handler_for(:disconnected, nil)
        true
    end