These features are hard to get right, and we have seen some edge cases pop up over the last week.

As far as I understood, the current way of supplying the hover info by walking the parse-tree up is never going to be perfect, so the tests have to be permissive and be more of a warning of where the current approach is not up to the task. Still, seeing the overall picture should be highly beneficial, and should help both make compromises with the current design, as well as assist in integrating the (foreshadowed) type/semantic analysis.

So yeah we need tests.