c.typecheck is unreliable

[xeno-by]

Our initial implementation of macro annotations includes YOLO-support for c.typecheck: we do our best effort to call the typechecker so that it doesn't crash, occasionally losing precision because the typechecker resists our efforts.

When I released 2.0.0-SNAPSHOT with this limitation, my hope was that our experience with c.typecheck will let us understand it better and come up with a strategy that is consistent, reasonable and implementable in scalac. Unfortunately, it looks that my hope was ungrounded, because it's been two years and nothing satisfying has come up.

Related issues are:

• c.typeCheck doesn't see declarations from the current scope #14
• c.typecheck causes spurious cyclic errors when used in a method annotation #49
• c.typecheck leads to a weird NPE #50
• missleading error message (cyclic reference) with Macro-Annotations when typecheck doesn't work in the given context #57
• typechecking annottees inside a macroTransform crashes the compiler #67

[xeno-by]

I will keep this issue open in case something comes up, but the current situation is:

1. I don't know how to fix this in 2.x.
2. An upcoming refresh of our macro system may let us view this question from a different angle.