Dear Herbert,
I've been experimenting with an alternative implementation of 'System.Timeout.timeout'[1] which avoids the overhead of spawning a new thread for each invocation.
Be warned that timeouts are very intricate. We had a lengthy discussion on the topic 2 years ago, starting at http://www.haskell.org/pipermail/libraries/2011-February/015876.html There was even an IO manager based proposal similar to yours: http://hackage.haskell.org/trac/ghc/ticket/4963 (What's the busyWontTimeout benchmark mentioned there?) http://www.haskell.org/pipermail/libraries/2011-February/015953.html The main trouble with the IO manager based approach is that even after unregisterTimeout finished, the timeout may still be invoked, and additional work is needed to protect against that. (I have more to say on this, but will postpone it until later. A lot of it has already been said in the earlier thread anyway.) Best regards, Bertram