Rewrite one :discard on queue overflow test for robustness

[rkday]

How about this as a different approach to 6af1316 ? The basic idea is that you take a lock, then inject enough tasks waiting on that lock to fill up all the threads and the queue. Then you inject more tasks which should never run, unlock the lock, and shutdown the executor so that all threads run to completion and your results are predictable.

It passes UT for me, and I think this is a robust approach, but the proof of the pudding will be in the eating. If you agree witth the approach, I can definitely reuse it for the other :discard tests, and maybe some others.

[coveralls]

Coverage increased (+0.03%) when pulling 92073c1 on rkday:queue_overflow_uts into 5929b46 on ruby-concurrency:master.

[rkday]

Okay, clearly not quite robust enough yet.

[rkday]

Ah, because there's no guarantee that my tasks are being moved off the queue onto a worker thread before I put more tasks on the queue. Fixing that now.

[coveralls]

Coverage increased (+0.03%) when pulling 445146a on rkday:queue_overflow_uts into 5929b46 on ruby-concurrency:master.

[pitr-ch]

@rkday This is great, Thanks! It may be better to replace Mutex with Event though where the threads can Event#wait and be suspended.

[jdantonio]

Awesome! Thank you!

[rkday]

@pitr-ch, thanks! I hadn't come across Concurrent::Event yet but it looks useful - I've switched to it in #211 .

[jdantonio]

@rkday Concurrent::Event is a trow-back to the days of yore when I worked on banking systems using MS Visual C++. I'm surprised more languages today don't have something like it, but I'm glad people find our version useful.