Only determine impl-trait desugaring kind once

[cjgillot]

impl-trait can either be desugared as a generic parameter (universal), as an opaque type (existential), or forbidden. This desugaring is currently on the AST in two places: def_collector and ast_lowering. The two logic are not consistent with one another. When this happens, ast_lowering is right.

dyn Trait<T: Bound> sometimes desugars to dyn Trait<T = impl Bound>, and sometimes does not.

We should only determine the desugaring once. This can be achieved by:

• saving the desugaring from def_collector into an impl_trait_context: FxHashMap<LocalDefId, ImplTraitContext> in Resolver;
• add impl_trait_id: NodeId in rustc_ast::AssocConstraint and use it to create the impl-trait's LocalDefId;
• introduce UniversalInDyn to handle the dyn Trait<T: Bound> case in def_collector and create the definition there, desugar_to_impl_trait should be replaced by resolver.opt_local_def_id(constraint.impl_trait_id).is_some();
• move the &mut Vec in rustc_ast_lowering::ImplTraitContext to LoweringContext, the new field should be saved and restored by with_hir_id_owner;
• merge rustc_ast_lowering::ImplTraitContext and rustc_resolve::ImplTraitContext, introducing a ResolverAstLowering::get_impl_trait_context(LocalDefId) -> Option<ImplTraitContext> to access the information.

Please reach out on Zulip for any question.

[randomicon00]

@rustbot claim

[vE5li]

Is anyone working on this? If not I would like to take a look

[randomicon00]

The tasks are interlinked, from my understanding, and the version that I have has one, two, and three almost done and I stopped at the last item. In fact, I am stuck there, give me like a week or two so I refresh my memory and get an exact idea about what's blocking me on that last item?