This is kind of dangerous. We previously took [T <: Any] as an approximation of [T] since that's the best we can do. But now, if type Alias = Any we also treat [T <: Alias] as "having no upper bound", which is patently untrue.

I feel like we should sway one side or the other, not this in-between, where [T] and [T <: Any] are the same but, [T <: Any] and [T <: Alias] aren't. Maybe upper bound should be Option[Type] 😀

Edit: Actually, why don't we use NoType to differentiate between [T] and [T <: Any]... 🤔

It does not matter what the precise rules are, as long as we stick to them.

I think "aliases shouldn't affect typing" is a rule we obey. I'm not sure if it's a rule for type inference too, but I think it should be. By "stick to them" do you mean don't touch how inference works or am I interpreting too broadly?

Yes, the fact whether something has a bound or not is used in several critical pieces of type inference. We should not change that definition.