`SupervisorSuite` timeout on Windows

[djspiewak]

Looks like VM slowness to me:

[error] ==> X cats.effect.std.SupervisorSuite.concurrent - restart / cancel race  20.056s cats.effect.TestTimeoutException: test timed out after 20 seconds
[error]     at cats.effect.Runners.$anonfun$timeout$1(Runners.scala:151)
[error]     at cats.effect.unsafe.WorkStealingThreadPool$$anon$1.apply(WorkStealingThreadPool.scala:695)
[error]     at cats.effect.unsafe.WorkStealingThreadPool$$anon$1.apply(WorkStealingThreadPool.scala:692)
[error]     at cats.effect.unsafe.WorkerThread.run(WorkerThread.scala:908)

[durban]

Which branch?

Possibly related: #4446.

[djspiewak]

Almost positive this was series/3.x

[djspiewak]

@durban You know, I wonder if your spinlock fix resolves this (and the related issues).

[durban]

I think it does. I've even mentioned it in the PR. (Okay, I've mentioned MutexSpec, but I think this is probably the same.)

(I could reproduce #4448 locally by putting Thread.yields into the WSTP and running MutexSpec, and MutexSpec didn't fnish running. With the fix, it worked fine.)

[djspiewak]

Absolutely fascinating.

I wonder if there's a way we can improve on our test and/or benchmark coverage in this area. I don't think there are easy answers but it's worth pondering. I noticed when we did the Native port that it was kind of disturbingly easy to perturb the implementation in a way which didn't cause any changes in the tests.

[durban]

Somewhat related issue: #3442.