Simplify superinstance constraints for ErrorT

[paf31]

@garyb @joneshf Mind having a look at this one? I'd be helpful for the latest book chapter.

This new Apply instance enables parallel asynchronous computations with ErrorT on top of a Parallel newtype for ContT which is quite nice.

[joneshf]

I puzzled it for a bit, but I don't understand it. Care to elaborate?

[garyb]

Nice 👍

Functor m, Monad m => ... was redundant anyway, I thought I'd already gone through and updated all those after we introduced superclasses, but I guess not. I wouldn't be surprised if there are other constraints throughout transformers that can be relaxed or simplified in a similar way to this.

[paf31]

@joneshf ErrorT e m a is isomorphic to m (Either e a) so if we unwrap the error constructor, we can use regular lift2 to define <*>. The problem is the extra Either e wrapper. We have an Either a (a -> b) and an Either e a and want an Either e b. We just have to lift lift2 itself. Make sense?

This is beneficial, because obviously we prefer to use the Apply instance over Monad, especially if we don't have a Monad, as is the case for Parallel in the new book chapter.

[paf31]

Ok to merge then?

[joneshf]

Ah, yes, go for it!

[paf31]

Cool, thanks.