Possible regression on use of Self in generics ("mismatched types")

[rye]

The following code (also on playground) compiles fine on Stable and Beta rustc, but has started no longer compiling on latest nightly (2020-02-28, 2020-02-29 both have this issue).

pub struct Container<V>(pub V);

impl<T, V> core::ops::Mul<T> for Container<V>
where
    T: From<V> + core::ops::Mul<T, Output = T>,
{
    type Output = Container<T>;

    fn mul(self, scalar: T) -> Self::Output {
        Self(scalar * self.0.into())
    }
}

I would expect this code snippet to compile with no errors as it currently does in stable and beta Rust.

Instead, on recent nightlies (02-28 and 02-29 were tested), an error is produced:

  --> src/lib.rs:10:14
   |
3  | impl<T, V> core::ops::Mul<T> for Container<V>
   |      -  - expected type parameter
   |      |
   |      found type parameter
...
10 |         Self(scalar * self.0.into())
   |              ^^^^^^^^^^^^^^^^^^^^^^ expected type parameter `V`, found type parameter `T`
   |
   = note: expected type parameter `V`
              found type parameter `T`
   = note: a type parameter was expected, but a different one was found; you might be missing a type parameter or trait bound
   = note: for more information, visit https://doc.rust-lang.org/book/ch10-02-traits.html#traits-as-parameters

For what it's worth, the substitution

 pub struct Container<V>(pub V);

 impl<T, V> core::ops::Mul<T> for Container<V>
 where
     T: From<V> + core::ops::Mul<T, Output = T>,
 {
     type Output = Container<T>;
 
     fn mul(self, scalar: T) -> Self::Output {
-        Self(scalar * self.0.into())
+        Container(scalar * self.0.into())
     }
 }

resolves the error.

I believe this is because rustc is erroneously picking Self to be Container<V> (the type for which the impl is being specified) instead of using Container and inferring the (formerly correct) output type of Container<T>.

Meta

rustc --version --verbose:

rustc 1.43.0-nightly (d3c79346a 2020-02-29)
binary: rustc
commit-hash: d3c79346a3e7ddbb5fb417810f226ac5a9209007
commit-date: 2020-02-29
host: x86_64-apple-darwin
release: 1.43.0-nightly
LLVM version: 9.0

[rye]

Sure enough, this seems related to #69306, which was marked resolved by PR #69340. I guess I'm one of the lucky people affected by those changes. 😄

[Centril]

Yeah, this was caused by a bug fix to type checking of Self paths, which would previously wrongly introduce inference variables for type parameters in the impl. Using Self should always refer to the Self type, just as using Self {...} would. In other words, your code should never have compiled.

[Centril]

Closing as expected breakage and "wontfix".