Profiler durations are sometimes incorrect

[bvaughn]

While working on the DevTools profiler, I've noticed that durations are sometimes incorrect. This seems to be vaguely related to high/low priority state updates. Need to track this down more so I'm filing an issue for myself.

[bvaughn]

Anecdotally, it seems the totalElapsedPauseTime value is the underlying cause for at least some of the cases when the actualDuration is unexpectedly long. Still digging into a minimal repro.

Edit It seems that this might not always be the case. Or at least that it might be caused by the delta between totalElapsedPauseTime when a fiber is marked and when it is later recorded. (The totalElapsedPauseTime might be unexpected at either point.)

[bvaughn]

Interestingly, the more I simplify the test application– the harder it becomes to reproduce this. I started by profiling the full IO "suspense" demo and it could be triggered relatively reliably. Then I began stripping non-essential components to reduce the console noise, and it became significantly harder.

Edit This might have been a misleading observation. By reducing the number of fibers in the app, I reduced the number of possible bad measurements. I also made the app "faster" and so maybe I needed to also reduce my threshold for logging an error.