Quoting Johannes Waldmann (2018-11-29 13:42:42)
These differences already occur with the last two lines replaced by "forever $ return ()", so the STM stuff may be inessential here. But that's the context where I came across the problem.
There's another thread right now with a subject line of "Timing out a pure evaluation of an expression I did not write myself," that seems like it might be related: I would expect forever $ return () to not allocate, which would mean it would never hit any yields, and thus never be rescheduled, and hogging the CPU. I've been able to reproduce your results, and if I change the last line to: forever $ do yield atomically $ writeTVar x True ..it always prints -- so the culprit is definitely a failure to yield. -Ian