We're not warning against fn ptr and wide raw ptr nested inside ADTs

[RalfJung]

I found a gap in our match forward-compatibility warnings:

type T = (*const [i32], i32);
const C: T = (unsafe { std::mem::transmute((0usize, 0usize)) }, 0);

fn test(x: T) -> bool {
    matches!(x, C)
}

This should warn but doesn't.

The reason is that here we are only checking the top-level type to be a wide raw ptr or fn ptr, we are not recursing.

The fix IMO is to reject those types in compiler/rustc_trait_selection/src/traits/structural_match.rs, which already recurses the type.

Cc @oli-obk @lcnr

[oli-obk]

The fix IMO is to reject those types in compiler/rustc_trait_selection/src/traits/structural_match.rs, which already recurses the type.

That would warn on Option<String>, which can be used at None value

[RalfJung]

That would warn on Option, which can be used at None value

It would? That's surprising, since valtree construction will succeed for None: Option<String>.
So that could be used as a signal not to lint.