A general library for higher-kinded data

[chris-martin]

Rel8 defines the term "higher-kinded data" as:

Higher-kinded data types are data types of the pattern:

data MyType f =
  MyType { field1 :: Column f T1 OR HK1 f
         , field2 :: Column f T2 OR HK2 f
         , ...
         , fieldN :: Column f Tn OR HKn f
         }

where Tn is any Haskell type, and HKn is any higher-kinded type.

Aside from the specific detail of the Column type family, this notion seems very general! Do y'all think there's any potential for a library that would help others replicate the pattern for other applications? (For example, I have in mind a similar sort of approach to filesystem storage, with database rows and fields corresponding to directories and files.)

[ocharles]

Maybe! Though I will note I think those libraries maybe already exist. In this space, we have:

1. https://hackage.haskell.org/package/barbies - probably the one I'd most recommend
2. https://hackage.haskell.org/package/rank2classes
3. https://hackage.haskell.org/package/hkd

and there may be others. So I think we'd have to be confident that we have something substantially different. There's a good chance that if https://ghc-proposals.readthedocs.io/en/latest/proposals/0242-unsaturated-type-families.html makes it into GHC proper, we can drop a lot of what we've had to invent in Rel8 and just use barbies - but that remains to be proven.

[ocharles]

(@chris-martin I moved this to a discussion - hope that's OK!)

[chris-martin]

(@chris-martin I moved this to a discussion - hope that's OK!)

Oh cool, I didn't even know about github discussions. Yes, it's certainly not a rel8 "issue".

[chris-martin]

The main things I really like about Rel8able that I haven't managed to replicate more generally are:

• The Column type family enumerates the library-specific type constructors that inhabit f, which makes the API much easier to understand than directly dealing with general-purpose type variable soups like (forall a. p a -> m (q a)) -> g p -> m (g q)
• The type family approach allows the Identity type to disappear
• You derive the library-specific Rel8able class, rather than something generic like Rank2.Traversable, which means if there are any mistakes in the data definition, you see the type errors immediately where the datatype is defined, rather than seeing some constraint violation off at some use site.

By the way, I'd found barbies and hkd, but thanks for pointing out rank2classes - that one is working out much better for me.