Creating a value tuple with a non-existent field results in a confusing CS8619

[Tragetaschen]

Version Used:

17.3.0 Preview 1.0

Steps to Reproduce:

public static class S
{
    public const string A = "a";
}

public class C
{
    public (string, string) M()
    {
        return ("a", S.B); // S.B does not exist
    }
}

Expected Behavior:

I expect the error that S.B does not exist and a squiggle just for that. Derived errors should be suppressed.

Actual Behavior:

CS8619 Nullability of reference types in value of type '(string, ? B)' doesn't match target type '(string, string)'.
and then
CS0117 'S' does not contain a definition for 'B'

Notes
CS8619 comes before CS0117 in the Error List and the visible squiggle covers the entire tuple expression.
You need to carefully hover the squiggle to see the actual error.

[RikkiGibson]

I think this could be addressed by treating an error type as equal to any other type when we do nullable variance checks.