Our current approach to silencing only a subset of diagnostics in unreachable sections of code has some issues. They are documented here. In short, we still emit false positives in some cases:

if False:
    x: int = 1
else:
    x: str = "a"

if False:
    # TODO We currently emit a false positive here:
    # error: [invalid-assignment] "Object of type `Literal["a"]` is not assignable to `int`"
    other: int = x
else:
    other: str = x

We should re-evaluate if it makes more sense to silence all diagnostics (possibly with a very limited set of exceptions, like reveal_type-messages) in unreachable sections.