Thoughs / observations:

1. Do we know any real-world examples where Nothing is inferred and the warning seems useful? The two examples I added as test case here are synthetic. Other than that, I only saw examples where the desired warning is about inferring Any. I think the idea of warning for Nothing came from #4401 by symmetry, but there's no example there (or in #11053).

2. There's an existing warning for inferring Any in 2.13.16, but it doesn't trigger if the expected type is Any:

class Bar
class Test {
  def f[F[_], A](v: F[A]) = v
  implicit def barToList(b: Bar): List[Int] = List(42)
  def t1 = f(new Bar)      // warns in 2.13.16, two warnings in 2.13.17
  def t2: Any = f(new Bar) // no warning in 2.13.16, new 'kind-polymorphic' warning in 2.13.17
}

That's because the Any makes it into the type var bounds, affecting !tvar.constr.hiBounds.exists(t => topTypes.contains(t.typeSymbol)). I think this needs to stay (#9452).

3. The object retronym test case added in #11053 doesn't actually trigger any warning. None of test1 / test2 (after un-commenting) / test3 lead to inferring Any for F[_]. So inference improved it seems, maybe due to SI-2712.

So I think we should revert the new warning for the Nothing case.

Perhaps also revert it for the Any case? We know a few examples where the warning shows up due to weak inference of Scala 2 (example, example), but it's hard to know how many people this affects and how useful it really is.

Yea, the ratio of kind-checking saving the day vs posing issues does not seem great, so I'm supportive of just reverting the whole thing.

#11162