rust/compiler/rustc_hir_analysis
nils 7e90732abe
Rollup merge of #109470 - compiler-errors:gat-normalize-bound, r=jackh726
Correctly substitute GAT's type used in `normalize_param_env` in `check_type_bounds`

Given:

```rust
trait Foo {
    type Assoc<T>: PartialEq<Self::Assoc<i32>>;
}

impl Foo for () {
    type Assoc<T> = Wrapper<T>;
}

struct Wrapper<T>(T);

impl<T> PartialEq<Wrapper<i32>> for Wrapper<T> { }
```

We add an additional predicate in the `normalize_param_env` in `check_type_bounds` that is used to normalize the GAT's bounds to check them in the impl. Problematically, though, that predicate is constructed to be `for<^0> <() as Foo>::Assoc<^0> => Wrapper<T>`, instead of `for<^0> <() as Foo>::Assoc<^0> => Wrapper<^0>`.

That means `Self::Assoc<i32>` in the bounds that we're checking normalizes to `Wrapper<T>`, instead of `Wrapper<i32>`, and so the bound `Self::Assoc<T>: PartialEq<Self::Assoc<i32>>` normalizes to `Wrapper<T>: PartialEq<Wrapper<T>>`, which does not hold.

Fixes this by properly substituting the RHS of that normalizes predicate that we add to the `normalize_param_env`. That means the bound is properly normalized to `Wrapper<T>: PartialEq<Wrapper<i32>>`, which *does* hold.

---

The second commit in this PR just cleans up some substs stuff and some naming.

r? `@jackh726` cc #87900
2023-03-28 12:51:13 +02:00
..
2023-02-21 11:51:56 +11:00

For high-level intro to how type checking works in rustc, see the type checking chapter of the rustc dev guide.