[Haskell-cafe] MPTC or functional dependencies?

[Roman Cheplyaka]

* Petr P <petr.mvd at gmail.com> [2012-12-21 13:38:21+0100]
> Dear Haskellers,
> 
> I'm working on a small library for representing semigroup (or monoid)
> actions on a set <http://hackage.haskell.org/package/semigroups-actions>.
> The MultiParamTypeClasses extension seems to be best suited for the task,
> as a group can act on many sets, and a set can be acted on by different
> groups:
> 
>     -- | Represents an action of semigroup @g@ to set @a at .
>     --
>     -- Laws: @'Endo' . 'act'@ must be a homomorphism of semigroups.
>     class Semigroup g => SemigroupAct g a where
>         act :: g -> (a -> a)
> 
> But soon I realized that with MPTC the compiler has problems inferring
> types and I had to explicitly specify types when using `act` in many
> places. Because it seems that in most cases a set will have only a single
> group acting on it, I was thinking about using FDs:
> 
>     class Semigroup g => SemigroupAct g a | a -> g where
> 
> But on the other hand, this can limit the generality of the type class. I
> cannot decide which one I should choose.
> 
> What would you suggest? According to your experience, would you choose
> plain MPTC or FD?

Another option to consider is not to make this a type class at all.

Something like

  newtype SemigroupAction g a = SemigroupAction { act :: g -> a -> a }

Instead of specifying types you'll be passing SemigroupAction values
(which is arguably less tedious). Plus you don't have to do newtype
wrapping of types and can have as many actions as you want even for the
same pairs of types.

Roman