To clarify some of the points in this discussion: Alastair is right in that the finalizer thread isn't necessarily run to completion if the main thread exits. GHC does run any outstanding finalizers at the end of the program, but it doesn't necessarily complete any existing finalizer threads that have already started. This is clearly inconsistent, and we were planning to make it consistent by not guaranteeing to run finalizers at program exit at all. There are good reasons (at least, good justifications) for this[1]. If your finalizer results in <<loop>>, it is probably because you tried to write to stdout in the finalizer, and the finalizer ran after stdout had already been finalized. Try calling something lower-level if you want to do some output from a finalizer. We don't implement any kind of finalizer ordering or dependencies between finalizers. There are good reasons for this[1] :-) Cheers, Simon [1] http://www.hpl.hp.com/techreports/2002/HPL-2002-335.html, especially Appendix A.