Hover on constants should render the display/debug impl if available

[Veykril]

Might be a crazy idea but I'll take the bitflags crate as an example here (although we fail to calculate the layout for it in const eval due to the projection used I believe, pasting the expansion and fixing that up makes it work):

bitflags::bitflags! {
    #[derive(Default, Debug, Clone, Copy)]
    pub(crate) struct TypeVariableFlags: u8 {
        const DIVERGING = 1 << 0;
        const INTEGER = 1 << 1;
        const FLOAT = 1 << 2;
        const FOO = TypeVariableFlags::FLOAT.bits() | TypeVariableFlags::INTEGER.bits();
    }
}

Hovering FOO renders (again <layout-error> currently) the following:

hir_ty::infer::unify::x
const Foo: TypeVariableFlags = TypeVariableFlags(InternalBitFlags(3))

which is obviously not that useful, meanwhile the debug implementation would render as:

TypeVariableFlags(INTEGER | FLOAT)

Noticed this in a project of mine that works with bitflags where our hover is practically useless

[Veykril]

cc @HKalbasi maybe you already had this idea in your head yourself :^) Would be good to know if we can pursue this in the (near) future or not as I think this would be very useful

[HKalbasi]

I think it is very doable now. We just need to run this:

::core::fmt::format(::core::fmt::Arguments::new_v1(
        &[""],
        &[::core::fmt::ArgumentV1::new(&(THE_CONST), ::core::fmt::Debug::fmt)],
    ));

And show its result. The only challanges are:

• There is no body containing that code, so we need to manually call the interpreter internals (it won't be much problem I think)
• Changes in the fmt infrastructure can break us. We can use the current displayer as a fallback, and those changes will break us in other ways anyway.

We also need something similar for displaying panic messages I think.

I can do this, but this looks like a good opportunity for increasing the "bus factor" of mir, maybe?

[Veykril]

I'm fine with trying to tackle this myself (once I have time, I'm currently somewhat occupied with things).

We should fallback to our current approach no matter what if we have problems running the debug/display implementation.

[Veykril]

I don't think I'll have the capacity to put the work into figuring out our MIR stuff like this in the coming time. So feel free to implement this yourself if you want, I'd look into the PR to catch up on how it was done then (which should teach me more in a shorter time I think)