Parsec is indeed a pain in the behind -- it could have been written as a monad transformer (with the current version implemented as a transformation of the Identity monad). It takes a bit of work to make a monad transforming version out of the Parsec source code, a bit more if you want to generalize it out to support parsing byte vectors (as opposed to linked lists of characters) or other things, and overall it's just a pain, and you end up with three monad transformers atop one another, if you do it "right." So then you think, "ok, I'll just make a general parsing typeclass so I can newtype derive solid parsing types," and then boilerplate instances begin, and that takes O(n^2) code overall, where n is the number of monad transformer typeclasses you want to have playing nice with each other. And don't get me started on functional dependencies.