Handle implicit Sized bounds

[flodiebold]

We currently don't add the implicit Sized bounds to type parameters, which leads to some inaccuracies in trait solving, so we need to fix this.

Rough steps:

• find out the exact conditions for the implicit Sized bound (is it simply all type parameters and associated types without ?Sized except trait Self? not sure)
• during lowering of generic predicates, add the Sized bounds for all type parameters that don't have ?Sized (I feel it makes more sense to add them at this step, rather than in hir_def, but I'm not 100% sure) except for trait Self
• similarly, add the bounds for associated types; their bounds are currently lowered directly in the Chalk integration, which we should change at some point (maybe now!)
• make sure we don't suddenly print + Sized everywhere, e.g. when displaying argument impl Trait types
• the coerce_unsize_apit test should now work correctly; maybe add some explicit tests with Sized bounds in trait impls
• check analysis-stats performance before/after

[bjorn3]

is it simply all type parameters without ?Sized except trait Self?

As far as I am aware that is the case.

[flodiebold]

Hmm, associated types get it as well, don't they?

[bjorn3]

They would have to for Self::Foo to be acceptable as argument and return type.

[iDawer]

If nobody working on this yet, I'd like to claim it.

Nice opportunity to learn Chalk!

Got some intermediate results

//#[test]
//fn play() {
//    check_infer(
//        r#"
#[lang = "sized"]
trait Sized {}

trait Foo {
    fn foo() -> u8;
}
trait Trait<T: Foo + ?Sized, U: Foo> {
    fn bar() {
        T::foo;
    }
}
//"#,
//        expect_test::expect![[]],
//    );
//}

dbg!() output:

running 1 test
generic_predicates_query: def = TraitId(TraitId(1)), trait Foo
[crates/hir_ty/src/lower.rs:1096] &explicitly_unsized_tys = {}
[crates/hir_ty/src/lower.rs:1121] predicates = []
generic_predicates_query: def = TraitId(TraitId(2)), trait Trait
[crates/hir_ty/src/lower.rs:1096] &explicitly_unsized_tys = {
    ^0.1,
}
[crates/hir_ty/src/lower.rs:1121] predicates = [
    for<type, type, type> for<> Implemented(^1.1: TraitId(1)),
    for<type, type, type> for<> Implemented(^1.2: TraitId(1)),
    for<type, type, type> for<> Implemented(^1.2: TraitId(0)),
]
generic_predicates_query: def = TraitId(TraitId(0)), trait Sized
[crates/hir_ty/src/lower.rs:1096] &explicitly_unsized_tys = {}
[crates/hir_ty/src/lower.rs:1121] predicates = []

[jonas-schievink]

@iDawer Are you still working on this? If not I can pick this up

[flodiebold]

@jonas-schievink If you do, note that they did run into some problems: https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/implicit.20Sized.20bounds

[iDawer]

I apologize for being missing without a notify. I'm back and will continue on this