Use a NIO framework

[Julien Herr]

Again, it's a question about the architecture.

Why don't you use a NIO framework like netty, Apache MINA, Grizzly, ...?

[Christian Schudt]

What would be the benefits? I don't use it because I am not familiar with them and I want to keep this library lighweight. Java's Socket is perfectly fine I think. AFAIK those frameworks are useful for server architectures, which need to handle several thousand connections, but this scenario is unlikely for a client. Also, non-blocking IO would likely increase the complexity.

[Julien Herr]

Yes, I understand the lighweight argument.

And you right about the server needs too. But does it mean it won't be a "good" idea to build a XMPP server or proxy (from/to another protocol by example) with babbler?

The most benefits I see for this kind of framework is to use non-blocking IO without "complexity". But I'm not enough familiar with them to determine if it could be a good or bad idea (that's why I asked).

FYI, Async-http-client is a http client which has an netty and grizzly implementation (but right, its main goal is to be async).

[Christian Schudt]

It's not intentend to act as a server or proxy, it's only a client library. Theoretically you could use the xmpp-core project and build your own server/proxy, I guess.

It has only the "model" classes, which are mapped to XML and which are shared between server and client, e.g. Message, Presence, ...

The whole business logic needed for a server is not part of this project, though.

[Julien Herr]

Ok! If I understand well there is no use case where a "client" will be connected to multiple servers?

[Christian Schudt]

You can do that of course. E.g. Adium client allows for multiple concurrent accounts. But the amount of concurrent sessions is usually only a few, if at all. In this case you can connect each session in its own thread. I can only hardly imagine a client, which wants to manage thousands of concurrent connections to different servers. The connect() and login() methods are blocking methods anyway. Doing this asynchronously would mean a lot of work, a whole new concept and probably also a lot of more complexity to the user of these methods.

[Julien Herr]

So, it seems my idea wasn't so good after all! Just a sledge hammer to kill a ant :p

Thank for you feedback :)

[Christian Schudt]

Maybe one day, far in the future, we can reconsider this ;-)

[Christian Schudt]

I told you, maybe one day this issue will be reconsidered.

After 3 years, that day has finally come!

Implementing NIO support was a lot of work, but I finally did it: a9ce40d (among many other commits).

There will be a new connection method using NIO (Netty) starting with version 0.8.0.

[Julien HERR]

I'm not playing with xmpp anymore but it's awesome!!! Great job :)

Did you show perf improvments?

[Christian Schudt]

I didn't do real performance tests. I've created 100 concurrent sessions with blocking and non-blocking IO and stopped the time, but both approaches performed very similar. That doesn't surprise me, afaik NIO doesn't perform faster, it just safes resources (threads).

[Christian Schudt]

A Netty implementation is available since 0.8.0