When I first used MonadState instance for ParserT I had ParserT s (State s') a and I was expecting it to be defined as in most of transformers:

(MonadState s m) => MonadState s (ParserT s' m)

but instead it's defined as:

(Monad m) => MonadState (ParseState s) (ParserT s m)
-- `MonadThrow` and `MonadError` are defined in same fashion.

So my question is, why it's like that? I think If we had normal definitions of Monad* then ParseT would be more useful in transformer stacks (we could expose functions from this typeclasses with different name, for users to still be able to manipulate ParserT's Internal structure).