Reference types vs Nullable types ToString()

Question

Could someone please be kind enough to explain why calling ToString() on an empty reference type causes an exception (which in my mind makes perfect sense, you cant invoke a method on nothing!) but calling ToString() on an empty Nullable(Of T) returns String.Empty? This was quite a surprise to me as I assumed the behaviour would be consistent across types.

Nullable<Guid> value = null;
Stock stock = null;
string result = value.ToString(); //Returns empty string
string result1 = stock.ToString(); //Causes a NullReferenceException

Answer 1

Nullable<T> is actually a struct that has some compiler support and implementation support to behave like a null without actually being null.

What you are seeing is the collision between the implementation allowing you to treat it naturally as a null as you would any other reference type, but allowing the method call to happen because the Nullable<T> isn't actually null, the value inside it is null.

Visually it looks like it shouldn't work, this is simply because you cannot see what is done in the background for you.

Other such visual trickery can be seen when you call an extension method on a null reference type... the call works (against visual expectation) because under the hood it is resolved into a static method call passing your null instance as a parameter.

How does a Nullable<T> type work behind the scenes?

Answer 2

Nullable is a value type and the assignment to null causes it to be initialized with Value=null and HasValue=false.

Further, Nullable.ToString() is implement as follows:

public override string ToString()
{
    if (!this.HasValue)
    {
        return "";
    }
    return this.value.ToString();
}

So what you are seeing is expected.

Answer 3

It is a bit tricky with nullable types. When you set it to null it is actualy not null cause it is not reference type (it is value type). When you initialize such variable with null it creates new sctructure instance where HasValue property is false and it's Value is null, so when you call ToString method it works well on structure instance.

Answer 4

The exception raised by calling default(object).ToString() is called NullReferenceException for a reason, it's calling a method on a null reference. default(int?) on the other hand, is not a null reference, because it's not a reference; it is a value type with a value that is equivalent to null.

The big practical point, is that if this was done, then the following would fail:

default(int?).HasValue // should return false, or throw an exception?

It would also screw-up the way we have some ability to mix nullables and non-nullables:

((int?)null).Equals(1) // should return false, or throw an exception?

And the following becomes completely useless:

default(int?).GetValueOrDefault(-1);

We could get rid of HasValue and force comparison with null, but then what if the equality override of the value-type that is made nullable can return true when compared to null in some cases. That may not be a great idea, but it can be done and the language has to cope.

Let's think back to why nullable types are introduced. The possibility that a reference type can be null, is inherent in the concept of reference types unless effort is taken to enforce non-nullability: Reference types are types that refer to something, and that implies the possibility of one not referring to anything, which we call null.

While a nuisance in many cases, we can make use of this in a variety of cases, such as representing "unknown value", "no valid value" and so on (we can use it for what null means in databases, for example).

At this point, we've given null a meaning in a given context, beyond the simple fact that a given reference doesn't refer to any object.

Since this is useful, we could therefore want to set an int or DateTime to null, but we can't because they aren't types that refer to something else, and hence can't be in a state of not referring to anything any more than I as a mammal can lose my feathers.

The nullable types introduced with 2.0 give us a form of value types that can have the semantic null, through a different mechanism than that of reference types. Most of this you could code yourself if it didn't exist, but special boxing and promotion rules allow for more sensible boxing and operator use.

Okay. Now let's consider why NullReferenceExceptions happen in the first place. Two are inevitable, and one was a design decision in C# (and doesn't apply to all of .NET).

1. You try to call a virtual method or property, or access a field on a null reference. This has to fail, because there's no way to look up what override should be called, and no such field.
2. You call a non-virtual method or property on a null reference which in turn calls a virtual method or property, or accesses a field. This is obviously a variant on point one, but the design decision we're coming to next has the advantage of guaranteeing this fails at the start, rather than part-way through (which could be confusing and have long-term side-effects).
3. You call a non-virtual method or property on a null reference which does not call a virtual method or property, or access a field. There's no inherent reason why this should not be allowed, and some languages allow it, but in C# they decided to use callvirt rather than call to force a NullReferenceException for the sake of consistency (can't say I agree, but there you go).

None of these cases apply in any way to a nullable value type. It is impossible to put a nullable value type into a condition in which there is no way to know which field or method override to access. The whole concept of NullReferenceException just doesn't make sense here.

In all, not throwing a NullReferenceException is consistent with the other types - types through it if and only if a null reference is used.

Note that there is a case where calling on a null nullable-type throws, it does so with GetType(), because GetType() is not virtual, and when called on a value-type there is always an implied boxing. This is true of other value types so:

(1).GetType()

is treated as:

((object)1).GetType()

But in the case of nullable types, boxing turns those with a false HasValue into null, and hence:

default(int?).GetType()

being treated as:

((object)default(int?)).GetType()

which results in GetType() being called on a null object, and hence throwing.

This incidentally brings us to why not faking NullReferenceType was the more sensible design decision - people who need that behaviour can always box. If you want it to through then use ((object)myNullableValue).GetString() so there's no need for the language to treat it as a special case to force the exception.

EDIT

Oh, I forgot to mention the mechanics behind NullReferenceException.

The test for NullReferenceException is very cheap, because it mostly just ignores the problem, and then catches the exception from the OS if it happens. In other words, there is no test.

See What is the CLR implementation behind raising/generating a null reference exception? and note how none of that would work with nullable value types.

Answer 5

If you investigate Nullable<> definition, there is an override ToString definition. In this function, ToString is overriden to return String.Empty.

    // Summary:
    //     Returns the text representation of the value of the current System.Nullable<T>
    //     object.
    //
    // Returns:
    //     The text representation of the value of the current System.Nullable<T> object
    //     if the System.Nullable<T>.HasValue property is true, or an empty string ("")
    //     if the System.Nullable<T>.HasValue property is false.
    public override string ToString();

On the otherhand, Stock is a custom class, which I assume ToString is not overriden. Thus it returns NullReferenceException since it uses default behaviour.

Answer 6

As per MSDN Remarks

Guid.ToSTring() method Returns a string representation of the value of this Guid instance, according to the provided format specifier.

As per MSDN Remarks on Nullable

A type is said to be nullable if it can be assigned a value or can be assigned null, which means the type has no value whatsoever. Consequently, a nullable type can express a value, or that no value exists. For example, a reference type such as String is nullable, whereas a value type such as Int32 is not. A value type cannot be nullable because it has enough capacity to express only the values appropriate for that type; it does not have the additional capacity required to express a value of null.