Improve error message for explicit implementation of interface method

[thomaslevesque]

Fixes #1759

[thomaslevesque]

I have hundreds of warnings in the T4-generated files... They were gone in previous builds, and now they're back. I also have them locally.
The only thing that changed, as far as I can tell, is the .NET Core SDK.

[thomaslevesque]

Confirmed: if I force the use of the 3.0 SDK instead of 3.1, I don't have any warnings.
You had reported the problem in VS Dev Community, but it looks like they won't do anything about it... Did you also report it on Github? I can't remember

[thomaslevesque]

Oh, actually it's not the same warning!
Before, we had this:

warning CS8632: The annotation for nullable reference types should only be used in code within a ‘#nullable’ annotations context

Now we have this:

warning CS8604: Possible null reference argument for parameter 'arg' in 'object?[] Func<T1, object?[]>.Invoke(T1 arg)'.

(if I remove the #nullable enable, I get the old warning again)

[thomaslevesque]

The new warning seems legit. We have this:

return valueProducer(call.GetArgument<T1>(0)!);

valueProducer is a Func<T1, object?[]>, and GetArgument has a [return: MaybeNull] attribute. So we could be passing a null T1 argument for a non-null T1 parameter. But I don't know how to say "T1 in Func<T1, object?[]> can be null". I can't use T1?, because there's no constraint on T1.
I can just add ! after the call to GetArgument, but it's frustrating...

[thomaslevesque]

OK, so according to dotnet/roslyn#37829, it's by design that <Nullable>enabled</Nullable> is ignored in generated code.
What has changed is that the compiler now reports more nullability warnings.

[blairconrad]

Perfect, @thomaslevesque. Thanks!