So, I'm just going to say one more time: I think this is an insufficiently general approach to fixing this problem. It's an interesting strategy, but it's also violating a core assumption that scalac makes in general, so I'm wary of treating it as the solution to this problem.