Suggestion

We should add an extra type-checking step to rule-tester when testing type-aware lint rules. Right now, we have some test cases that don't pass, most notably because they have undeclared variables (which becomeany). However we don't guarantee linting behavior when there are type errors (because the result is likelyany, which either created new lint errors or masks actual lint errors). Such type errors may be hard to catch in more complex situations.

Brad mentioned there may be a perf penalty, but it would be useful to experiment nonetheless.