John Peterson wrote (snipped):

The implementer of these functions has to guarantee that the actions do not destroy the commutativity of the CIO monad.

Sorry, but several of my variable initialisation actions involve things like starting up child processes or rapid exits from the program if unsuccessful, which certainly cannot be guaranteed to be commutative. I still think the proposal I made to this list in June http://www.haskell.org//pipermail/haskell/2004-June/014104.html is sound. It requires nothing other than a new library, which is trivial to implement using the existing unsafePerformIO. It is safe. I don't think it's especially clunky. Essentially the only serious limitation is that you can only access a bit of initialised state from an IO action, but in my experience at least this is not a serious issue. However virtually everyone seems to have their own patent solution, and we are no closer to agreement. Might it be an idea to have a Wiki or something like that devoted to discussion of this problem, and we can all try to agree on the advantages and disadvantages of the competing methods?