Overly restrictive variance

[lcnr]

What is this

This is a design document for const generics. Any discussions about its content should be on zulip. The conclusions of these discussions should then be edited back into this issue. Please do not post any comments directly in this issue.

Content

Currently all generic parameters used by constants in the type system are constrained
to be invariant. Looking at this example, that is unnecessary:

struct UsesRef<'a> {
    actual_ref: &'a (),
    in_anon_const: [u8; std::mem::size_of::<&'a ()>()],
}

Here 'a is forced to be invariant, as it is used in the anonymous constant, but
ideally it should be covariant.

How variance can influence CTFE

The most obvious example is TypeId::of which has to differ between for<'a> fn(&'a ()) and fn(&'static ()) and for<'a> fn(fn(&'a ()) and fn(fn(&'static ()).

Higher ranked types act different during candidate selection and can therefore not simply be ignored. We currently have a future compatability lint for this in #56105 which will probably get allowed in the future.