Jules Bean wrote:
Well, mzero isn't a return value in the IO monad, it's an exception. But yes, I agree with you that the (plausible) laws I have seen for MonadPlus seem to say that mzero should ignore the actions. But this in practice is not how IO behaves.
Jules
I can see three possible solutions: 1) IO is not an instance of MonadPlus (but may still be an instance of MonadError) 2) Side effects are ignored (or state is ignored) and IO may be an instance of MonadPlus 3) bind (>>=) is redefined for IO. As the IO Monad is a function which resturns a computation, bindIO can be changed such that (a >> mzero === mzero). In other words if the RHS is mzero, the LHS is not included in the final result (and its actions would not get executed), however this must be inconsistent if we consider: f = getChar >>= (\a -> if a == "F" then mzero else return a) In this case if the LHS returns "F" the LHS should not have been run... this contradicts itself, so this is a non option I guess. Acutally looking at GHC CVS libraries, there is not a definition for MonadPlus on the state or IO monads... Keean.