On Wed, Jan 26, 2011 at 5:00 PM, David Menendez <dave@zednenem.com> wrote:
On Wed, Jan 26, 2011 at 3:56 PM, Edward Kmett <ekmett@gmail.com> wrote:
> Fair enough.
> These are the observations I have about 'transformers'.

> 2.) The instance for Applicative for MaybeT doesn't really follow the other
> instances for Applicative in transformers.
> instance Applicative m => Applicative (MaybeT m) where
>    pure = MaybeT . pure . Just
>    f <*> a = MaybeT $ (<*>) <$> runMaybeT f <*> runMaybeT a
> would instead follow the convention that Applicatives depend only on
> Applicatives where possible, as opposed to the current default instance.
> This convention is followed everywhere in transformers except this module.

Then you no longer have (<*>) = ap.

This convention is also not followed for the StateT and ErrorT
transformers for the same reason.

If you want something like MaybeT f whose Applicative instance depends
only on the Applicative instance of f, try Compose f Maybe.
 
By that reasoning the existing Applicative instances for WriterT and ListT would be wrong in mtl as well.

StateT actually requires a monad, because the s -> m (a, s) monad is given rise to by sandwiching a monad in an adjunction, which isn't sufficient to  yield an Applicative instance, but the ErrorT and MaybeT can both perfectly well get by on the weaker Applicative requirement, because they simply compose.

The consistency of ap and (<*>) is assumed universally. Applicative merely allows you to optimize this case.

-Edward Kmett