Re: [Haskell-cafe] semantics of concurrent program depends on -O level, -f[no-]omit-yields

so thread scheduling would be more obvious with individual characters being output instead of a single flush triggered by the final putChar.

Yes but in my example program, there is no contention for stdout, as only one thread is using it. I am inclined to enter this into the GHC issue tracker as it seems there's no obvious explanation, and "lost TVars confusing the STM machinery" was mentioned. Do you mean that this a known thing? Searching the tracker for "lost TVar" does not turn up anything. - J.W.