Negative blanket OIBIT impl treated as positive

[bluss]

impl !Pod for .. {} looks like it produces a positive impl.

(playpen)

#![feature(optin_builtin_traits)]

unsafe trait Pod {}

unsafe impl !Pod for .. {}

struct Foo<X, Y>(X, Y);

fn is_pod<T: Pod>(_: &T) { }

fn main() {
    is_pod(&Foo(0, 0));
}

cc @nikomatsakis

OIBIT tracker issue: #13231

[arielb1]

Negative impls are only supposed to work if they are nominal (impl<U,V,W> !Foo for Bar<U,V,W>). We should however decide what this means.

[nikomatsakis]

impl !Pod for .. is certainly not legal, in any case.

[lcnr]

What exactly has to be done to close this issue?

The example given here doesn't seem to compile.

[jonas-schievink]

impl Trait for .. syntax doesn't exist anymore, so it's impossible to write this sort of impl now.

[nikomatsakis]

Yeah, that syntax never made sense, and has been replaced with "auto trait" (or is it #[auto] trait). Probably this issue is just "outmoded".

One thing is that I think we should not permit a negative impl like impl<T> !Foo for T -- i.e., negative impls should always be implemented for specific types. The truth is that we have to write out the specific rules at some point.