Consider removing possibility to widen effect in Resource#useForever

[kubukoz]

Because of a mix of variance, widening and Nothing, this code doesn't compile:

object Demo extends IOApp.Simple {

  val run: IO[Unit] = {
    val resource: Resource[IO, Unit] = ???

    resource.useForever
  }

}

polymorphic expression cannot be instantiated to expected type;
 found   : [G[x] >: cats.effect.IO[x]]G[Nothing]
 required: cats.effect.IO[Unit]

If you annotate it as useForever[IO], it works. A possible solution to this might involve removing the G[x] >: F[x] parameter in useForever.

[RaasAhsan]

I've ran into this problem with other functions like allocated as well (though surprisingly they were all on Dotty). Almost every Resource function widens F to G, but how useful is that in practice? Maybe there's something in fs2, but I've never actually seen or used it before.

[kubukoz]

Resource was made covariant in #731, I think it might be useful when someone's F is IO[SpecificException, *] and they want to automatically upcast it to IO[Throwable, *], for example. See #617

I think we can remove the type annotation in useForever for some reason (the REPL allowed me to do it) but not for allocated. If we keep that variance, that is.

[RaasAhsan]

Ahh, that makes sense, thanks for the context. I'm one of those IO-only users :) I'll post some of the failing Dotty snippets here later because I think it's related but it does typecheck in 2.13.