Why checking a Non-nullable type against null actually compiles in C#8

Question

I am trying to use Nullable Reference Types in C# 8.0. I have set up a sample code to test it but something has baffled me!

As far as I understood, when we do not add ? to the Type it is treated as non-nullable. Then why in the Main() method the compiler does not complain that I am checking myClass against null? Shouldn't it be obvious by this point to the compiler that myClass is non-nullable?

Update:

I added a stub method to MyClass. As you see, at this point the compiler is aware that myClass is not null. Why would it allow for the next null checking? It does not make sense!

#nullable enable
using System;

namespace ConsoleApp4
{
    class Program
    {
        static void Main()
        {
            var service = new Service();
            MyClass myClass = service.GetMyClass();

            myClass.Write();

            // I am curious why this line compiles!
            if (myClass == null)
            {
                throw new ArgumentNullException();
            }
        }
    }

    public sealed class Service
    {
        private MyClass? _obj;

        public MyClass GetMyClass()
        {
            if (_obj == null)
            {
                _obj = new MyClass();
            }

            return _obj;
        }
    }

    public sealed class MyClass
    {
        public void Write()
        {

        }
    }
}

Answer 1

Shouldn't it be obvious by this point to the compiler that myClass is non-nullable?

It does, kind of. Hover over myClass on the null check, and the tool tip from static analysis it will tell you that myClass is not null there.

So, why can you check it against null? Because you can. It is still a reference type. The fact that it will not be null is not annotated in the type.

Furthermore, notice that it is not telling you that the exception is not reachable code. Why is that?

Well, C# 8.0 nullable references static analysis is not foolproof. It is good enough for most cases. However, from time to time there are patterns that they do not recognize. Needing the help of annotations and the null-forgiving operator (!). Which you can use to tell the compiler when you know better. However, these let room for telling lies to the compiler:

public MyClass GetMyClass()
{
    return _obj!;
}

Similarly, you can check for null in something that according to the static compiler cannot be null. Because you know better:

public sealed class Service
{
    private MyClass? _obj;

    public MyClass GetMyClass()
    {
        return _obj!;
    }
}

internal static class Program
{
    private static void Main()
    {
        var service = new Service();
        var myClass = service.GetMyClass();

        // I am curious why this line compiles!
        if (myClass == null)
        {
            throw new ArgumentNullException();
        }
    }
}

Of course, I would not recommend to actually go ahead and lie to the compiler. Yet, if you are receiving parameters from a third party, it is advised to still do null checks. Similarly when consuming a public API, when you cannot trust the implementation.

---

I will also remind you that C# 8.0 nullable references is a feature that was intended to be rolled out steadily across big legacy codebases. It was not intended to stop them from compiling when you had it enabled in some part and not in others. In fact, as far as I can tell, it will issue warnings at worst. Making null check not compile would go against that.

---

Why not a warning on the null check?

Well, Roslyn is deferring to the developer in this case. If you try to use myClass inside body of the conditional, and hover over it, it will tell you that it can be null there. Which, of course, would have been necessary to enter the conditional in the first place.

Furthermore, remove the throw and use myClass after the conditional. The static analysis will also say that myClass can be null after the conditional. Thus, Roslyn is using the null check as a hint that the variable is actually nullable.

Why did they opt to use it as a hint instead of issuing a warning? That I don't know. I could not find a discussion about it on dotnet/csharplang or dotnet/roslyn on github.

Answer 2

For future reference, here is how I was able to force the compiler to actually NOT compile when null-checking is redundant.

For anyone interested, please also check this:

In C# 8, how do I detect impossible null checks?

For those unfamiliar with DDD, ValueObject is the base class for our immutable objects.

#nullable enable
using System;

namespace ConsoleApp4
{
    internal static class Program
    {
        private static void Main()
        {
            var service = new Service();
            MyClass myClass = service.GetMyClass();

            myClass.Write();

            // Compiler now will not compile! mcClass can never be null!
            if (myClass == null)
            {
                throw new ArgumentNullException();
            }
        }
    }

    public sealed class Service
    {
        private MyClass? _obj;

        public MyClass GetMyClass()
        {
            return _obj ?? (_obj = new MyClass());
        }
    }

    public sealed class MyClass : ValueObject<MyClass>
    {
        public void Write()
        {
        }

        protected override bool EqualsCore(MyClass other)
        {
            throw new NotImplementedException();
        }

        protected override int GetHashCodeCore()
        {
            throw new NotImplementedException();
        }
    }
}


#nullable enable

namespace ConsoleApp4
{
    public abstract class ValueObject<T> where T : ValueObject<T>
    {
        public sealed override bool Equals(object obj)
        {
            return obj is T valueObject && EqualsCore(valueObject);
        }

        public sealed override int GetHashCode()
        {
            return GetHashCodeCore();
        }

        protected abstract bool EqualsCore(T other);

        protected abstract int GetHashCodeCore();

        public static bool operator ==(ValueObject<T> a, ValueObject<T> b)
        {
            return a.Equals(b);
        }

        public static bool operator !=(ValueObject<T> a, ValueObject<T> b)
        {
            return !(a == b);
        }
    }
}