I don't like our macro tests -- they are brittle and don't inspire confidence. I think the reason for that is that we try to unit-test them, but that is at odds with reality, where macro expansion fundamentally depends on name resolution.