StaticAbstractMembersInInterfaces - Roslyn does not warn that a most-derived static interface method implementation will not be used

[daiplusplus]

StaticAbstractMembersInInterfaces - Roslyn does not warn that a most-derived static interface method implementation will not be used

[CyrusNajmabadi]

I expected the call to T.DoSomething() in GenericMethod(); to be routed to static void SubclassAB.DoSomething() but instead it's routed to SubclassA..

Because you created a method unrelated to the interface. It's not an interface impl, and it's def in its own slot (which is why if you remove new we'll warn you of that).

I assume the reason is that the generic-method-instantiation logic in the CLR performs a top-down search for binding static methods - instead of bottom-up

Neither of these are incorrect. This is a non-virtual method. So there is no 'search'ing happening at all. The method being called is determined statically at compile time and the CLR just invokes the method requested by you.

in which case I expect Roslyn to raise a diagnostic (i.e. a warning) at the GenericMethod(); call-site, with a message like: "Static methods in SubclassAB will not be invoked.

We can't warn here. This is an entirely reasonable piece of code to write. If you'd like a custom warning here, you'd need to write your own analyzer.

both with and without the new modifier

The 'new' modifier does not impact semantics at all. it's just a way for a user to say they understand that one method hides another, and has no virtual inheritance relationship with it. This is often a bug, so new is needed to indicate that "yes, that hiding was intended".

[daiplusplus]

So there is no 'search'ing happening at all.

The JIT surely has to search from SubclassAB to ClassA to find the IMyInterface implementation and bind the static call-sitesa, no?

This is an entirely reasonable piece of code to write

Why is it reasonable to specify a more-derived type as a type-parameter when the effect is the same as the less-derived type? When would anyone do that?

The 'new' modifier does not impact semantics at all.

Right - I wasn't suggesting that it does - but I think it's odd that shadow methods were deemed confusing enough to warrant a compiler warning, but subclassed static interface methods were not. It just feels abitrary to me.

...especially as this isn't documented at all right now. Hence it took me by surprise.

[CyrusNajmabadi]

Why is it reasonable to specify a more-derived type as a type-parameter

For any number of reasons. Including that the code in question might care about that type-parameter versus the base type.

when the effect is the same as the less-derived type?

There's no concept of "the effect is hte same". The compiler has no idea what happens inside a method body (remember that they can come from metadata, and we want uniform compilation behavior for source vs metadata).

[CyrusNajmabadi]

The JIT surely has to search from SubclassAB to ClassA to find the IMyInterface implementation and bind the static call-sitesa, no?

For the static-virtuals, yes. For the non-virtuals, no. If you use static-virtuals (and have a more derived impl, it will be called).

[daiplusplus]

If you use static-virtuals (and have a more derived impl, it will be called).

So regardless of the use of static abstract members, given a class A : IInterface and a class B : A, the class B is not considered a more-derived-implementation of IInterface because only A implements IInterface unless it's class B : A, IInterface?

[CyrusNajmabadi]

Correct. And we even warn you about that. but you suppressed the warning by putting 'new' on teh method :)

[daiplusplus]

Correct.

Thank you, that does clarify and explain things.

And we even warn you about that. but you suppressed the warning by putting 'new' on teh method :)

The new warning is the same for both class SubclassAB and class SubclassABI though.

warning CS0108: 'SubclassAB.DoSomething()' hides inherited member 'ClassA.DoSomething()'. Use the new keyword if hiding was intended.
warning CS0108: 'SubclassABI.DoSomething()' hides inherited member 'ClassA.DoSomething()'. Use the new keyword if hiding was intended.

...but only SubclassABI.DoSomething() is considered an implementation of IMyInterface.DoSomething.

[CyrusNajmabadi]

yup. :)