Fix ItemKind::TraitAlias encoding

[dlrobertson]

ItemKind::TraitAlias should be encoded just like a regular Trait.

Fixes: #56488

[rust-highfive]

r? @michaelwoerister

(rust_highfive has picked a reviewer for you, use r? to override)

[dlrobertson]

r? @Centril

[dlrobertson]

The test I wrote doesn't include all of the example in #56488 but it still causes nightly to panic.

[Centril]

r? @nikomatsakis

[nikomatsakis]

Hmm, this doesn't seem like the right fix. I think the bug is a touch deeper though (cc @alexreg). In particular, the is_trait_alias function seems to currently be thread-local:

/// Returns `true` if `def_id` is a trait alias.
pub fn is_trait_alias(tcx: TyCtxt<'_, '_, '_>, def_id: DefId) -> bool {
    if let Some(node_id) = tcx.hir.as_local_node_id(def_id) {
        if let Node::Item(item) = tcx.hir.get(node_id) {
            if let hir::ItemKind::TraitAlias(..) = item.node {
                return true;
            }
        }
    }
    false
}

but I think that it will not work correctly when you try to actually use a trait alias across crates, as a result.

i.e., can we make a test like

// Crate A
trait Foo = Send + Sync;

// Crate b:
use crate_a::Foo;
fn use_b<T: Foo>() { }
fn main() {
    use_b::<u32>(); // Expect Ok
    use_b::<Rc<u32>>(); // Expect error
}

and see what behavior is?

[nikomatsakis]

In any case, I feel like we can't just pretend a trait is a trait alias across crates.

[dlrobertson]

Hmm, this doesn't seem like the right fix... In any case, I feel like we can't just pretend a trait is a trait alias across crates.

Ah, right... good point.

[alexreg]

Thanks for having a go at this @dlrobertson!

I think @nikomatsakis is right; the is_trait_alias fn that I wrote is crate-local (I think Niko writing "thread-local" was a slip heh). I'm not sure on the best way to fix this right now, but let me give it some though.

[dlrobertson]

👍