Cats-effect: Add LiftIO to ce3

[djspiewak]

We should restore LiftIO, but in the core module and extending Async.

[vasilmkd]

I'd like to work on this.

[djspiewak]

It's yours!

[vasilmkd]

@djspiewak Please correct me if I'm wrong. The IO#to method uses an implicit F: Effect[F] in order to use the F#to in the implementation. Does that mean that LiftIO needs to extend Effect instead of Async? If not, how can I work around this? Thanks.

[djspiewak]

Hmm, I don't think you need to worry about any of that at all. To be clear, the definition of LiftIO should be something like this:

trait LiftIO[F[_]] extends Async[F] {
  def liftF[A](ioa: IO[A]): F[A]
}

I don't think you have to worry about to at all with that. This should be definable for anything that has a liftF (e.g. OptionT[IO, A] would have a LiftIO instance, defined by the OptionT.liftF function).

As a note, if it's easier, I think it's actually okay if we just have LiftIO be an orphan typeclass, not extending Async at all. The existing LiftIO doesn't.

[vasilmkd]

Looking at the CE2 implementation of IO#to, it uses a LiftIO, but in CE3 it's Effect. Does that even matter?

Edit:
This also prevents me from implementing LiftIO.liftK.

[djspiewak]

Looking at the CE2 implementation of IO#to, it uses a LiftIO, but in CE3 it's Effect. Does that even matter?

Ah! So the use of LiftIO in ce2's IO#to is a historical artifact. The implementation in CE3 uses Effect, and that's exactly what it needs to use. :-) LiftIO means something different than "you can convert an IO to this". It actually means "you can embed an IO within this". So it's more about monad transformers.

[vasilmkd]

So does that mean I should implement LiftIO.liftK by demanding an Effect[F]?

[djspiewak]

So does that mean I should implement LiftIO.liftK by demanding an Effect[F]?

Nope! You would just implement each instance on a one-off basis. For example, here's one for OptionT:

object LiftIO {
  // ...
  implicit def liftIOForOptionT[F[_]]: LiftIO[OptionT[F, *]] =
    new LiftIO[OptionT[F, *]] {
      val liftK = OptionT.liftK[F]
    }
}

Something like that!

[vasilmkd]

Nevermind, I just got confused by the changes to IO#to. Thanks for taking the time to explain everything!

[vasilmkd]

Again, this issue wasn't automatically closed...

[djspiewak]

So weird! GitHub must have changed something… :-/