Re: Refactoring Semigroup/Monoid (was: instance Applicative Data.Map)

On Sun, 17 Mar 2013, Roman Cheplyaka wrote:

* Henning Thielemann [2012-11-20 00:26:17+0100]

...

On Sat, 17 Nov 2012, Edward Kmett wrote:

...

Honestly, the main issue is that even if you have the ability to describe default definitions for how to implement superclasses, it isn't really all that useful. :( e.g. Every monad transformer still needs each of its instances crafted by hand. This even applies to simpler types:

Given

instance (Traversable f, Traversable g) => Traversable (Compose f g)

you don't want the derived instances for Functor and Foldable, 

instance (Traversable f, Traversable g) => Functor (Compose f g) instance (Traversable f, Traversable g) => Foldable (Compose f g)

you want the more permissive ones you can roll by hand.

instance (Functor f, Functor g) => Functor (Compose f g) instance (Functor f, Functor g) => Foldable (Compose f g)

Good example. Frequently I think that we should replace all these type class instance extensions by a generic way to program instances.

What do you mean? Don't we already have TH and plenty of generic programming libraries?

That was last year, how shall I know today what I meant? :-) The various meta-programming utilites can be used to _implement_ type class instances. But I think there should be a way to unify all the UndecidableInstances, OverlappingInstances, FlexibleInstances, FlexibleContexts, TypeSynonymInstances, that is there might a programmable way to _select_ a class dictionary for some given concrete types.