Add an extra class to Control.Applicative.Alternative

[Tom Ellis]

On Thu, May 01, 2014 at 01:19:50PM +0200, S D Swierstra wrote:
> On 30 Apr 2014, at 15:50 , Edward Kmett <ekmett at gmail.com> wrote:
> 
> > A few thoughts:
> > 
> > * Splittable has been used as a name historically for splitting random number generators and the like, so the name conflict (especially from somewhere so prominent) would be unfortunate. That can of course be fixed by a simple bikeshedding exercise.
> > 
> > * You should be able to implement this in one pass. e.g. f a -> Either a (f a) rather than as two combinators, but it isn't clear to me what the getNonPure branch can do other than return the original when the scan fails or some partially zonked/expanded version of it.
> 
> I have probably not been clear enough. The relationship that should hold between getPure and getNonPure is as follows:
> 
> case (getPure p, genNonPure p) of
>   (Nothing, Nothing)       -> "should not happen since p should have at least a pure or a nonpure part"
>   (Just pp,  Just npp)        -> pure pp <|> npp
>   (Just np,  Nothing)        -> pure np
>   (Nothing, Just npp)       -> npp

If (Nothing, Nothing) is impossible why not encode it in the type?

    data OneOrBoth a b = One a | Other b | Both a b

    class Splittable f where                                                                                                     
        split :: f a -> OneOrBoth (f a) a

Tom