Algorithm for concurrent connection resolution

[whyrusleeping]

issue #174 is (probably) a result of two nodes connecting to eachother at the same time. I have also created a test case where two nodes attempt connections to eachother at the same time which hangs and times out. We need to discuss what to do in such a scenario.

[jbenet]

Oh i'm already fixing this. I've it 80% of the way there. Works like this:

• added Conn interface
• Conn struct -> singleConn, implements Conn
• PeerConn implements Conn, has map[multiaddr string]Conn (*singleConn) objects
• PeerConn routes traffic to whatever child conn it makes sense to (i.e. always one)
• PeerConn times out child conns after a period of idleness
• Swarm has map[peerID]Conn (*PeerConn)
• Swarm uses a conn.Listener and adds singleConns to PeerConn

[whyrusleeping]

What if A and B are doubly connected and each pick the other connection? I guess traffic would still make it to each peer properly, but one connection wouldnt time out

[whyrusleeping]

I guess, how does a peer decide which connection "makes sense"?

[jbenet]

For now, we can select deterministically by ordering the addresses.

Later, we can measure latency and stuff!