[RFC] Add Data.Functor.Rep.TH

[bgamari]

This is a first cut at deriving Rep instances with Template Haskell. Does this interface look sane?

[bgamari]

I should add that @ekmett has suggested using generics for this but I'm a bit too dense to see a way of doing this that provides the same ease of use and safety as the Template Haskell option.

[bgamari]

One open question surrounding this patch is what set of instances the representation type should be equipped with. Currently this set includes Typeable which is a bit unfortunate as it requires -XDeriveDataTypeable.

[ekmett]

I may take a bit to merge this one, but I'm sold on the general idea.

[sjoerdvisscher]

I would expect the generics based code to look like this:

import qualified GHC.Generics as Gen

class Distributive f => Representable f where
  type Rep f :: *
  type Rep f = Rep (Gen.Rep1 f)

  tabulate :: (Rep f -> a) -> f a
  default tabulate :: (Gen.Generic1 f, Rep (Gen.Rep1 f) ~ Rep f, Representable (Gen.Rep1 f)) => (Rep f -> a) -> f a
  tabulate = Gen.to1 . tabulate

  index    :: f a -> Rep f -> a
  default index :: (Gen.Generic1 f, Rep (Gen.Rep1 f) ~ Rep f, Representable (Gen.Rep1 f)) => f a -> Rep f -> a
  index = index . Gen.from1

and also a lot of instances for Functor, Distributive and Representable for all the GHC.Generics data types except V1 and :+:.

And then you can do

instance Representable A

instead of

deriveRep 'A

But I'd first start with derving instances for Distributive actually. Although Distributive has the problem that all methods already have a default implementation, so when you replace one method with a generic implementation, anyone who doesn't use generics would have to implement that one and would no longer be able to choose between implementing distribute or collect.

[ekmett]

I'm open to including both GHC.Generics and template-haskell based versions as the former is 'zero touch' for the easy cases and the latter yields nicer representations.

On Thu, Apr 3, 2014 at 5:26 PM, Sjoerd Visscher notifications@github.comwrote:

I would expect the generics based code to look like this:

import qualified GHC.Generics as Gen

class Distributive f => Representable f where type Rep f :: * type Rep f = Rep (Gen.Rep1 f)

tabulate :: (Rep f -> a) -> f a default tabulate :: (Gen.Generic1 f, Rep (Gen.Rep1 f) ~ Rep f, Representable (Gen.Rep1 f)) => (Rep f -> a) -> f a tabulate = Gen.to1 . tabulate

index :: f a -> Rep f -> a default index :: (Gen.Generic1 f, Rep (Gen.Rep1 f) ~ Rep f, Representable (Gen.Rep1 f)) => f a -> Rep f -> a index = index . Gen.from1

and also a lot of instances for Functor, Distributive and Representablefor all the GHC.Generics data types except V1 and :+:.

And then you can do

instance Representable A

instead of

deriveRep 'A

But I'd first start with derving instances for Distributive actually. Although Distributive has the problem that all methods already have a default implementation, so when you replace one method with a generic implementation, anyone who doesn't use generics would have to implement that one and would no longer be able to choose between implementing distribute or collect.

— Reply to this email directly or view it on GitHubhttps://github.com/ekmett/adjunctions/pull/8#issuecomment-39507018 .

[bgamari]

@sjoerdvisscher, regarding Distributive you might want to have a look at https://github.com/ekmett/distributive/pull/4. It was only after working through this that I realized how one might derive Representable from Generic as you documented. This is now on my to-do list for an idle weekend but it could be a while until such a weekend appears.

[bgamari]

@sjoerdvisscher the trouble with this Generic approach is that you require Representable (Rep1 f) which isn't possible as Rep1 f is not a functor as would be required by the Distributive f class bound. Perhaps you meant GRepresentable (Rep1 f) here?

[bgamari]

@sjoerdvisscher oh dear, I completely overlooked your mention of introducing Functor instances for the generic types. Forget about the above.