In the design of Control.Monad.Accum, I noticed that for a new Monad transformer t, one typically has to define the instance via LiftingAccum. However, I believe it would be more straightforward to define an automatic lifting mechanism that traverses the outer Monad transformers.

For instance, consider the following definition:

instance {-# OVERLAPPABLE #-}
  (MonadTrans t, Monad m, MonadAccum w m) =>
    MonadAccum w (t m) where
  accum :: (MonadTrans t, Monad m, MonadAccum w m) => (w -> (a, w)) -> t m a
  accum = lift . accum

I think this approach would be more convenient for users who want to adopt a new Monad Transformer (say, LogicT). Is there a reason that precluded the designer of MonadAccum from adopting this easy implementation?

The implementation was compiled and worked as expected.