TransLift expressiveness is limited

[djspiewak]

As discussed with @stew, capturing the type constructor to be lifted in the enclosing typeclass is neat in that it does allow the constraint to vary, but it also causes some problems at the call site. It also unfortunately makes it impossible for cases like shims, since I cannot abstract over both TransLift and scalaz's Hoist, and I'm fairly certain even TransLift alone would not be usable for emm.

As a sidebar, I suspect that the insertion of the type constructor into the typeclass also gives rise to problems related to SI-2712 for users writing functions polymorphic in that effect type; problems which wouldn't occur with the type attached to liftM.

We proposed adjusting the type of TransLift to something more like the following signature:

trait TransLift[MT[_[_], _]] {
  type TC[M[_]]
  def liftT[M[_], A](ma: M[A])(implicit M: TC[M]): MT[M,A]
}

This would retain the power of the variable constraint by binding that power to the transformer itself, not the pair of the transformer and the inner effect. @ceedubs noted that he had tried this approach in the past and ran into some problems with Unapply cases, but I think it's worth exploring a bit further.

[ceedubs]

I started working on cleaning up my stash of this approach that I had. I remembered that a strange thing about the proposed change is that for types that don't require any constraint to liftT (like Kleisli), I think you are forced to write a dummy typeclass for which there is a trivial implicit instance.

[djspiewak]

I started working on cleaning up my stash of this approach that I had. I remembered that a strange thing about the proposed change is that for types that don't require any constraint to liftT (like Kleisli), I think you are forced to write a dummy typeclass for which there is a trivial implicit instance.

You can do that, or you can cheat with something ugly like I did and set the typeclass to Lambda[X[_] => Unit =:= Unit], which will obviously trivially resolve.

[adelbertc]

set the typeclass to Lambda[X[_] => Unit =:= Unit]

inches away

[ceedubs]

@djspiewak clever. I like it.

[adelbertc]

@djspiewak can we call this closed?

[djspiewak]

Yep