This was a bit complicated. I've added a wrapper type for now that does the
LocalSyntaxPtr <-> ExprId translation; we might want to get rid of that or give
it a nicer interface.
370: Self params & type r=matklad a=flodiebold
This implements type inference for `self`, so field completion for methods taking `self` works now.
- rename `IMPL_ITEM` to `IMPL_BLOCK` -- rustc calls the methods etc. inside an impl `ImplItem`s, and the impl itself doesn't define an item, so I thought this name was clearer.
- add HIR for impl blocks -- we collect all impls in a crate at once, so we can go from methods to containing impls, and since we will later also need to find all impls for a certain type (which may be anywhere in the crate, I think?). We could be more lazy here, but I don't know if it's worth the complexity.
- resolve `self` and `Self` during type inference
- refactor a bit in ty.rs as well
Co-authored-by: Florian Diebold <flodiebold@gmail.com>
Since we need to be able to go from def to containing impl block, as well as the
other direction, and to find all impls for a certain type, a design similar to
the one for modules, where we collect all impls for the whole crate and keep
them in an arena, seemed fitting. The ImplBlock type, which provides the public
interface, then consists only of an Arc to the arena containing all impls, and
the index into it.
This will really become necessary when we implement generics, but even now, it
allows us to reason 'backwards' to infer types of expressions that we didn't
understand for some reason.
We use ena, the union-find implementation extracted from rustc, to keep track of
type variables.
