Allow Overrides and Implicit Interface Implementations to Override/Implement using Generically-Typed, Constrained members

[fitdev]

It would be nice if we could write code like this where generically-typed members (with appropriate constraints) can override members in the base class:

    public class ClassGenericDerived<T> : ClassBase where T : IFoo {
      //public override IFoo Foo { get; set; } //<- Works
      public override T Foo { get; set; } //<- Error: CS1715 'ClassGenericDerived<T>.Foo': type must be 'IFoo' to match overridden member 'ClassBase.Foo'
    }

    public class ClassBase {
      public virtual IFoo Foo { get; set; }
    }

    public interface IFoo { }

In the above code T is guaranteed to be of type IFoo due to generic constraint. So, why can't T Foo {get; set; } override base class member public virtual IFoo Foo { get; set; }?

Same logic would go for implicit interface implementations.

[canton7]

What happens if someone does this:

public class Foo1 : IFoo { }
public class Foo2 : IFoo { }

ClassBase c = new ClassGenericDerived<Foo1>();
c.Foo = new Foo2();

The signature of ClassBase.Foo says that someone can assign any type which implements IFoo. But your suggestion would mean that ClassGenericDerived<Foo1>.Foo is of type Foo1, so trying to assign any other implementation of IFoo wouldn't work. However that breaks the interface declared in ClassBase.

[HaloFour]

Agreed, this would make more sense if the property was get-only in which case it might qualify for covariant returns. But I don't know if that applies to properties or to generic classes like this.

[canton7]

If a getter-only property doesn't come under covariant returns, I'd think it should be!

[HaloFour]

I believe it does based on this error from SharpLab:

error CS8831: 'B.P': Target runtime doesn't support covariant types in overrides. Type must be 'object' to match overridden member 'A.P'

It also appears that the requested feature will also work with a getter-only property, based on the error message: SharpLab

Someone running the preview can test for sure. I wonder when SharpLab will support running on .NET 5.0.