Fix stale wakeups from Scheduler#unblock interrupting other operations

[tavianator]

Fixes: #467

Types of Changes

• Bug fix.

Contribution

• I added tests for my changes.
• I tested my changes locally.
• I agree to the Developer's Certificate of Origin 1.1.

[ioquatix]

Thanks for this work.

My initial feeling is, this is compensating for a bug that can manifest in other ways.

Spurious wakeups are a reality of high peformance concurrency - sometimes it's hard to track enough state, efficiently, to not wake something up incorrectly. It's usually more efficient to deal with the spurious wakeup (e.g. go back to sleep) rather than try to fix the source.

That being said, I'll review this PR in more detail soon.

[tavianator]

Yeah that's fair, I think a reasonable alternative fix would be to have wait_*able loop and call io_wait again if it returns false before the timeout

[ioquatix]

@tavianator I think that's the correct solution, but feel free to figure it out.

Eventually, I'd like to transition io-event and async to a ready-list based scheduler, so spurious wakeups may become more common.

[tavianator]

@ioquatix Wrote up that alternative as #469. I like it a lot better except it doesn't fix sleep() waking up early (since Kernel#sleep(5) and ConditionVariable#wait(mutex, 5) are indistinguishable, but an unblock has to wake up the CV wait)