Thinking about this more I am not really happy with this change. It feels like we are leaking too many implementation details by having to change the signature of TestClock.make and introducing irregularity with the constructors for the other test implementations. It looks like a good amount of the issue is that sometimes an effect is put to sleep in an uninterruptible block which by default makes the warning fiber uninterruptible. That is a very easy fix that we can do purely internally. Then any other warnings would be cases where sleep was called but adjust or its equivalents was never called, in which case sleep could just be replaced by never so we don't need to change all of our signatures to support that. I will update shortly.

@jdegoes Sorry I had some second thoughts about this that I was in the process of making.

The version we merged has creating a TestClock return a Managed, so advantage is that warning will get canceled if test completes even though sleep is used and clock is not adjusted. Disadvantage is that TestClock.make now returns a Managed unlike the constructors for all the other test implementations and is a little harder to use.

I was thinking of going with a more minimalistic change where we just made the warning fiber interruptible (it could be uninterruptible before if the clock adjustment occurred in an uninterruptible region). Disadvantage is warning could still be displayed if a test completed with a sleep and no adjustment, but in that case the sleep effect could just be replaced with never (though that could be harder if the sleep is composed into some larger effect you are calling).

I see the pluses and minuses of both and am happy either way, just want to make sure we are aligned.