Rethrow only on node v0.8.

[mcollina]

This pull-request uses process.versions.node to detect if we are on node v0.8 and only in that case we rethrow after the 'uncaughtException' handler.
This solves the problem in #14 in a much simpler way.

[mcollina]

done!

[mcollina]

@raszi could you please review this?

[raszi]

Actually I wanted to change the whole behavior on this. I started to reimplement the graceful cleanup with domains but it did not help.

[raszi]

So I'll accept this I finish the work with that.

[raszi]

Released as v0.0.22, thanks!

[mcollina]

Thank you! :)

[wibblymat]

Correct me if I'm wrong... but doesn't this mean that if I use tmp in my project then any uncaught exception will just disappear without a trace in node 0.10? If so then this is much worse behaviour than before. Debugging will be a nightmare.

[mcollina]

Ok, I saw the problem. It didn't happen in my code because I use mocha for testing (which have its own uncaughtException handler) and domains (which use uncaughtException). So, the error now it's not reported if there is no uncaughtException handler.
This assumption is based on this comment: #14 (comment).

[mcollina]

The fix is changing that line into:

  if (process.versions.node.indexOf('0.8') == 0 || process.listeners('uncaughtException').length === 1)

[mcollina]

However, that fix do not change the issue of bad error reporting.

I've noted in the code that there are both a 'uncaughtException' and 'exit' event handlers that do almost the same thing. Why do we need both? How about removing the 'uncaughtException' once and for all? All remove operations are sync.

[wibblymat]

See #21. In node 0.8 the exit event won't fire if you are exiting with an exception so you still need uncaughtException for that case.