New connection parser

[mcollina]

I was able to get some better performance numbers from our poor Connection class.

@adamvr could you please review it?

I'm getting something near 10% increase.

[mcollina]

@adamvr, If it is not a problem I'll release it as 0.3.0 over the weekend

[adamvr]

@mcollina I'm unsure about the stream builder thing. Is there actually a problem with reusing the same socket?

[mcollina]

You were doing it wrong before. Node.js sockets retain same state when they close, and they cannot be piped again. The connect method is not really supposed to be used that way :(.

[adamvr]

Aha, okay. I guess I'll defer to you then. Make it so, Mr. Collina!

[adamvr]

p.s. node_redis just calls connect again on the socket as far as I can tell. See https://github.com/mranney/node_redis/blob/master/index.js#L479-494

[mcollina]

Maybe it is a node v0.10/stream2 thing, I am not sure.
Still, you cannot pipe a stream after it has been closed :(.

[mcollina]

Do you think it is better to release it as 0.3.0, or just as the next in the 0.2.x series?

[adamvr]

Is it actually necessary to have a full pipe, though? I mean, the only information the parser needs to know from the stream is the data coming out of it, not whether or not the stream is connected, disconnected, closing, piping or whatever. I'm having good (passing tests) results just doing something like:

this.stream.on('readable', function() {
   var chunk = that.stream.read();
   that.parser.write(chunk);
}

Any chance you could gist your benchmarks so I can test that against them?

As to the version number, I don't know. I would probably put it in the 0.2.x since it isn't an API changing change (externally, anyway. I'm sure the few if any people who do new MqttClient are going to be right confused), but you're right that we probably should have bumped reconnection/offline into 0.3.0.

[mcollina]

The real difference is that by using pipe you handle the "pressure" on the stream, e.g. if it has not completed then it's not called again, and there is some good old buffering (which is good, btw).
My benches are there: https://github.com/mcollina/mosca/blob/master/benchmarks/bombing.js and https://github.com/mcollina/mosca/blob/master/benchmarks/throughputCounter.js

[mcollina]

Already included in #118.