The current implementation of MockClock could be subject to a race condition in some circumstances. Consider the following snippet:

for {
  mockClock <- MockClock.makeMock(DefaultData)
  fiber     <- mockClock.sleep(2.millis).zipPar(mockClock.sleep(1.millis)).fork
  _         <- mockClock.adjust(2.millis)
  _         <- fiber.join
  result    <- mockClock.fiberState.get
} yield result.nanoTime == 2000000L

In the current implementation this would be flaky. When we adjust the MockClock we wake up both the left and the right effects in zipPar. Although we wake up the right effect first, since the effects are on separate fibers their order of completion is indeterminate. So if the left effect completes first, the right effect will set the fiber time to 1 millisecond, leading to data corruption.

With the new functionality for specifying custom logic for inheriting FiberRef instances implemented in #1879 we can address this by always using the maximum fiber time when combining data from two fibers.