Excerpts from aditya siram's message of Mon Aug 09 15:05:14 -0400 2010:
Until test is called in 'main' we don't know the order of monads. In fact even the base monad is not know. All we know is that it uses the State and Writer monad. In each call to 'test' in main we can determine the stacking order and the base monad yielding different results. This seems to be a more flexible way of using monad transformers but I haven't seen this in code before so is there anything wrong with this style?
There are two points here: * You have observed that running the monads in different orders results in different values: however, in the case of the examples you've chosen, the essential character of the result is still the same. This is not universally true: only some monads "commute", so to speak. So you definitely do not want to leave the ordering implicit if order matters. But when the monads are commutative... * The general feeling on this issue, as I've heard from Don, is that generic types for specifying monad stacks are too fragile: due to the extra generality, it's a lot easier to write code that typechecks but doesn't do what you actually want it to. A decent compromise might be to use a concrete stack, and offer a utility function that "lifts" it into the generic monad typeclasses. Cheers, Edward