Better diagnostics wanted for use of nullable type in switch pattern

[gafter]

We currently produce suboptimal parse errors for an attempt to use a nullable type in a switch pattern. We interpret it as a ?: expression, producing syntax errors, rather than letting the user know that a nullable type may not be used in a type pattern. For example:

        [Fact, WorkItem(16559, "https://github.com/dotnet/roslyn/issues/16559")]
        public void CasePatternVariableUsedInCaseExpression()
        {
            var program = @"
public class Program5815
{
    public static void Main(object o)
    {
        switch (o)
        {
            case Color Color:
            case Color? Color2:
                break;
        }
    }
    private static object M() => null;
}";
            CreateCompilationWithMscorlib45(program).VerifyDiagnostics(
                // (9,32): error CS1525: Invalid expression term 'break'
                //             case Color? Color2:
                Diagnostic(ErrorCode.ERR_InvalidExprTerm, "").WithArguments("break").WithLocation(9, 32),
                // (9,32): error CS1003: Syntax error, ':' expected
                //             case Color? Color2:
                Diagnostic(ErrorCode.ERR_SyntaxError, "").WithArguments(":", "break").WithLocation(9, 32),
                // (8,18): error CS0118: 'Color' is a variable but is used like a type
                //             case Color Color:
                Diagnostic(ErrorCode.ERR_BadSKknown, "Color").WithArguments("Color", "variable", "type").WithLocation(8, 18),
                // (9,25): error CS0103: The name 'Color2' does not exist in the current context
                //             case Color? Color2:
                Diagnostic(ErrorCode.ERR_NameNotInContext, "Color2").WithArguments("Color2").WithLocation(9, 25)
                );
        }

[svick]

Using a nullable type in an is pattern has a very similar issue. The code if (o is int? i) produces:

error CS1003: Syntax error, ':' expected
error CS1525: Invalid expression term ')'
error CS0103: The name 'i' does not exist in the current context

I think it should behave the same as if (o is Nullable<int> i):

error CS8116: It is not legal to use nullable type 'int?' in a pattern; use the underlying type 'int' instead.

[lindexi]

In the code o is int? means (o is int) ? a : b not if (o is Nullable<int> i) See #20156