ScheduledObserver Ordering

[mairbek]

Fixing issue #233.
Blocking solution for now, thinking about non-blocking alternatives.

[cloudbees-pull-request-builder]

RxJava-pull-requests #90 FAILURE
Looks like there's a problem with this pull request

[mairbek]

this might grow the stack size, should be reimplemented.

[cloudbees-pull-request-builder]

RxJava-pull-requests #91 FAILURE
Looks like there's a problem with this pull request

[cloudbees-pull-request-builder]

RxJava-pull-requests #92 ABORTED

[cloudbees-pull-request-builder]

RxJava-pull-requests #93 SUCCESS
This pull request looks good

[benjchristensen]

This looks like an elegant solution @mairbek

I can't quite figure out if there's a chance for a race condition on the counter but am merging it for now since I can't prove one way or another and this is way better than what we have.

I'm manually merging via #238 since I added some unit tests.

[mairbek]

I'm not sure if this is a really working solution as well, looks too simple.
Need to think about use cases to break this code.

[benjchristensen]

Thanks, I'd appreciate that as I haven't had time to think through those.

[mairbek]

Okay, this seems to be working.

Assuming that underlying observable implemented correctly, this means that OnNext, OnCompleted and OnError events will be called sequentially. The solution itself is inspired by the way a CurrentThreadScheduler works except that it relies on concurrent primitives instead of ThreadLocal.

I can see 4 cases:

1. Enqueue when queue was empty. Counter is set to 1 and queue processing is started. Since counter is atomic two processes cannot start.
2. Enqueue when queue is not empty. Counter is set to 1, queue processing is already going on. Queue processing cannot stop since counter is atomic.
3. Dequeue when queue has more than one element. This would recursively schedule queue processing.
4. Dequeue when queue has one item. When item is processed counter would be set to 0. Queue processing stopped. Next enqueue will schedule out new queue processing.

@benjchristensen what do you think?

[abersnaze]

I think that as long as nothing makes any calls to the underlying observer after this line it should be ok.