implement Cats typeclass instances for Ask and Local

[bpholt]

This lets users map/flatMap (for Ask) or imap (for Local) the environment of a given instance and make the mapped value available in a new implicit scope.

This came up for me when trying to provide a Local[F, natchez.Span[F]], where F[_] is Kleisli[IO, Span[IO], *]. Local.baseLocalForKleisli gives us Local[Kleisli[IO, Span[IO], *], Span[IO]], which is not quite the right shape for an app with an implicit ev: Local[F, Span[F]] (which becomes ev: Local[Kleisli[IO, Span[IO], *], Span[Kleisli[IO, Span[IO], *]]]). In other words, baseLocalForKleisli gives us a local environment of Span[IO] when we need Span[Kleisli[IO, Span[IO], *]].

Given Invariant[Local[F, *]], we can do

Local[Kleisli[IO, Span[IO], *], Span[IO]].imap(Kleisli.liftK)(Kleisli.applyK(NoopSpan()))

to get the Local[Kleisli[IO, Span[IO], *], Span[Kleisli[IO, Span[IO], *]]] needed for the app.

[rossabaker]

Interesting. I was thinking about your Eq[Ask], and it hit me: is there anywhere in the Cats ecosystem where we have an instance for an instance? I can see where this is useful, but it's novel enough to have me thinking why it hasn't done.

[bpholt]

I have a vague memory of suggesting similar sorts of instances in the past, but I'm having trouble finding those PRs. Maybe it was in chat, not sure. I don't remember there being specific reasons not to do it, mostly just "huh that's different". But it's been a while; maybe I'm misremembering.

I considered just adding the map/flatMap/imap methods to Ask and Local directly, but after using cats-tagless for a while, it's been my experience that it's occasionally frustrating to have types with the methods for Cats instances, that don't actually implement the cats instances. Implementing the functionality using the typeclasses felt like it would be a little more flexible, and all else being equal, that's a good thing.

Implementing this way also lets us take advantage of the laws, which would be harder to test otherwise. Now that I think about it, that actually might be the stronger argument for doing it this way. 🙂

[armanbilge]

where we have an instance for an instance

Do the Invariant[Eq] instances and stuff count? They're not higher-kinded, but they are still instances right?

https://github.com/typelevel/cats/blob/024dd3d13ac9725ea663a4f914a89aa41bc3baea/core/src/main/scala/cats/Invariant.scala#L171-L255

[armanbilge]

👍 I really like this idea!