4.x: Fix timed delay hangs with dotCover and the DefaultScheduler

[akarnokd]

For some reason, dotCover would hang in various timed test which use the DefaultScheduler. I don't know why, but removing that empty try-catch along with the atomic dispose management solves the issue.

In addition, the Virtual_ThreadSafety() was running extremely slow in its original form (more than 5 minutes). The reduced time amount and a more close thread-synchronization should be faster now while preserving the behavior under test.

[danielcweber]

I see a race here where _state could be set null but the timer will execute, passing the wrong state (it doesn't help that _action is set to Stubs.Ignore before, that might be rearranged). Maybe timer._action should only be called when TryDispose was true, setting _action and _state by itself (or by some DisposeCore method)?

[danielcweber]

Do you know anything about the issue or bug here? Seems that the SpinWait would loop forever?

[akarnokd]

I don't know, I can't switch to debug while running with coverage enabled. Maybe there is a way to trigger a thread dump externally?

[danielcweber]

I could give it a try.

[danielcweber]

Do you know what process will eventually host our code? Everything I debug doesn't run anything form System.Reactive.

[akarnokd]

Either a dotnet.exe or dottest.exe. Usually the one with high CPU usage but here, it probably hangs thus there is no CPU usage.

[danielcweber]

I can't reproduce the hanging tests under coverage. Virtual_ThreadSafety() running slow is understandable, it takes a lot of loops and AdvanceBy to actually trigger the virtual scheduler.

[akarnokd]

Were you running dotCover or microsoft's own coverage tool of VS Enterprise?

[danielcweber]

DotCover within VS

[akarnokd]

There could be a probabilistic signal loss from the timer.

You could try and add a loop around the body of Delay_TimeSpan_DefaultScheduler for example, usually 1.000 - 10.000 iterations.

        [Fact]
        public void Delay_TimeSpan_DefaultScheduler()
        {
            for (int i = 0;  i < 10000; i++)    
                Assert.True(Observable.Return(1).Delay(TimeSpan.FromMilliseconds(1))
                   .ToEnumerable().SequenceEqual(new[] { 1 }));
        }

[danielcweber]

Can't really reproduce it but the fix is straightforward and the SpinWait-solution is wacky enough that this should be merged.