How does VB.NET compiler choose which extension overload to run?

Question

Got an interesting oddity - thought someone might be able to help.

This came out of some fun with nullable types from this question:

How to check if an obje

Answer 1

Overload 2 will ONLY work as an extension on explicitly defined Nullable(of T)'s. For example:

    Dim y As New Nullable(Of Integer)
    y.IsNullable()

This is because extension methods extend the type (or a base type), which in this case is Nullable(of T). Calling a.IsNullable() will never call overload 2. That's the easy part to figure out. This means the real question is why would overload 2 be called instead of overload 1 as a standard overloaded method call.

The CLR will determine which Overload to use by performing a "Better Conversion" check, where it implicitly converts the value(s) passed in to the type of the parameter(s) defined in the overloaded methods and then go down a checklist of rules to determine the best method to use.

From the MSDN Better Conversion Article:

If S is T1, C1 is the better conversion.

If S is T2, C2 is the better conversion.

Puting this code into Visual Studio will shows you that Overload 2 is the better conversion because the integer a (S) is the implicitly converted Nullable(of Integer) version of a (T2).

    ' a is an integer! 
    Dim a As Integer = 123

    Dim objValueType As ValueType = 123 'Or CType(a, ValueType)
    Dim objNullable As Nullable(Of Integer) = 123 'Or CType(a, Nullable(Of Integer))

    'Oh No, a compiler error for implicit conversion done for overload 1!
    Dim bolValueTypeConversionIsBetter As Boolean = (objValueType = a)

    'No error as long as Option Strict is off and it will equal True.
    Dim bolNullableConversionIsBetter As Boolean = (objNullable = a)

Answer 2

I think this is a bug, or at least a VB.NET "feature". (I'm just not sure which of VB.NET or C# is wrong.)

I have tried in LINQPad 4 (because that's what I've got on the machine I'm using) and for C# I got False for both results, for every value type and enum except for Nullable types, of course.

Whereas for VB.NET I get the False and True for all value types and enums, except for Nullable types, and ValueType and [Enum] which return False, False because you can't have a ValueType? or [Enum]?. With Option Strict Off, Object causes late binding, and fails at runtime to locate either overload, but the second result is False also because you can't have Object?.

For completeness, Nullable types return True, True for both languages as expected.

The fact that C# is doing something different (assuming my test is correct) confirms the reference to the C# "Better Conversion" check is wrong (or being misread at least - in that C# is not doing what is being interpreted as why VB.NET is doing what it is doing).

However, I do agree that the issue is probably related to the implicit conversion to Nullable(Of T) existing and somehow being a higher priority to the implicit conversion to ValueType.

Here's my LINQPad 4 "Query" (C# Program):

void Main()
{
    Test.test();
}

// Define other methods and classes here
static class Test
{
    static bool IsNullable(this ValueType obj)
    {
        return false;
    }

    static bool IsNullable<T>(this T? obj) where T:struct
    {
        return true;
    }

    public static void test()
    {
        int x = 42;

        bool result1 = x.IsNullable();
        bool result2 = IsNullable(x);

        result1.Dump("result1");
        result2.Dump("result2");
    }
}