Incorrect behavior when using alias refers to namespace in using alias

[amcasey]

extern alias Contracts1;
extern alias Contracts2;

using alias1 = Contracts1::TestContractAssembly;
using alias2 = Contracts2::TestContractAssembly;

partial class Contract : alias1.IContract, alias2.IContract
{
    int alias1.IContract.x
    {
        get
        {
            return 1;
        }
    }
    int alias2.IContract.x
    {
        get
        {
            return 1;
        }
    }
}

Both IContract types are the same - they expose an int property "x". Roslyn incorrectly reports that IContract.x has already been implemented.

This works if you inline the using aliases.

I strongly suspect that the alias points to the merged namespace symbol, rather than to the namespace symbol contained in the assembly symbol referenced by the extern alias.

[theoy]

Note, this was reported by a Microsoft team based on their real world source code, in their migration to use Roslyn.

[VladimirReshetnikov]

@amcasey @VSadov @theoy My experiments show that Dev12 compiles this code without errors, but metadata for class Contract contains 2 properties with identical signatures int32 TestContractAssembly.IContract.x() - it fails to verify and fails to load at runtime. Do we have more info about the real repro? Was it a usable assembly?

If we inline aliases, then the C# compiler includes the extern alias as part of the property name, and it works ok. In principle, we could always generate unique names, but it would be a new feature rather than a bugfix.

[VladimirReshetnikov]

@amcasey So far my tests show that aliases correctly refer to distinct non-merged namespaces with NamespaceKind.Module.

[amcasey]

The repro was reduced from an internal partner's code. I believe they care about the aliases referring to different namespace constituents but don't actually try to implement two different copies of the same interface on the same type.

[amcasey]

@VladimirReshetnikov Then why are we getting errors about implementing the same property twice?

[VladimirReshetnikov]

I'm going to close it as "By Design" for now. Let me know if you have any reasons to believe it should compile successfully, and what is the expected output.