[switch] Refactor TransportManager to avoid storing listeners directly on the Transport object

[dirkmc]

TransportManager stores listeners on the underlying Transport object itself. This causes strange things to happen if, for example, the Transport implementation extends EventEmitter (as it does in peer-base) because EventEmitter has a method called listeners() which gets clobbered.

Currently Switch has an object called transports whose keys are transport names (eg 'tcp') and values are the Transport implementations themselves. I suggest adding a level of indirection such that the values have an object like this:

{
  transport: <Transport implementation>,
  listeners: [<Listener>, <Listener>, ...]
}

The only complicating factor would be if there are libraries that depend on libp2p-switch and reference Switch.transports. As far as I can tell there are no references elsewhere in libp2p.

[jacobheun]

This would probably be a good opportunity to look at deprecating Switch.transports in favor of doing everything through the TransportManager. Having switch.transports and switch.transport is confusing. Pushing the logic and access to only the transport manager would be cleaner imo, and as part of that shift we could do a better job of not modifying the transport, especially when those properties aren't part of the interface by doing something along the lines of what you're suggesting.

[dirkmc]

Good point, that sounds like the right solution