Type Checking: typeof, GetType, or is?

I've seen many people use the following code:

Type t = typeof(obj1);
if (t == typeof(int))
    // Some code here

But I know you could also do this:

if (obj1.GetType() == typeof(int))
    // Some code here

Or this:

if (obj1 is int)
    // Some code here

Personally, I feel the last one is the cleanest, but is there something I'm missing? Which one is the best to use, or is it personal preference?

Solution 1:

All are different.

• typeof takes a type name (which you specify at compile time).
• GetType gets the runtime type of an instance.
• is returns true if an instance is in the inheritance tree.

Example

class Animal { } 
class Dog : Animal { }

void PrintTypes(Animal a) { 
    Console.WriteLine(a.GetType() == typeof(Animal)); // false 
    Console.WriteLine(a is Animal);                   // true 
    Console.WriteLine(a.GetType() == typeof(Dog));    // true
    Console.WriteLine(a is Dog);                      // true 
}

Dog spot = new Dog(); 
PrintTypes(spot);

What about typeof(T)? Is it also resolved at compile time?

Yes. T is always what the type of the expression is. Remember, a generic method is basically a whole bunch of methods with the appropriate type. Example:

string Foo<T>(T parameter) { return typeof(T).Name; }

Animal probably_a_dog = new Dog();
Dog    definitely_a_dog = new Dog();

Foo(probably_a_dog); // this calls Foo<Animal> and returns "Animal"
Foo<Animal>(probably_a_dog); // this is exactly the same as above
Foo<Dog>(probably_a_dog); // !!! This will not compile. The parameter expects a Dog, you cannot pass in an Animal.

Foo(definitely_a_dog); // this calls Foo<Dog> and returns "Dog"
Foo<Dog>(definitely_a_dog); // this is exactly the same as above.
Foo<Animal>(definitely_a_dog); // this calls Foo<Animal> and returns "Animal". 
Foo((Animal)definitely_a_dog); // this does the same as above, returns "Animal"

Solution 2:

Use typeof when you want to get the type at compilation time. Use GetType when you want to get the type at execution time. There are rarely any cases to use is as it does a cast and, in most cases, you end up casting the variable anyway.

There is a fourth option that you haven't considered (especially if you are going to cast an object to the type you find as well); that is to use as.

Foo foo = obj as Foo;

if (foo != null)
    // your code here

This only uses one cast whereas this approach:

if (obj is Foo)
    Foo foo = (Foo)obj;

requires two.

Update (Jan 2020):

• As of C# 7+, you can now cast inline, so the 'is' approach can now be done in one cast as well.

Example:

if(obj is Foo newLocalFoo)
{
    // For example, you can now reference 'newLocalFoo' in this local scope
    Console.WriteLine(newLocalFoo);
}

Solution 3:

1.

Type t = typeof(obj1);
if (t == typeof(int))

This is illegal, because typeof only works on types, not on variables. I assume obj1 is a variable. So, in this way typeof is static, and does its work at compile time instead of runtime.

2.

if (obj1.GetType() == typeof(int))

This is true if obj1 is exactly of type int. If obj1 derives from int, the if condition will be false.

3.

if (obj1 is int)

This is true if obj1 is an int, or if it derives from a class called int, or if it implements an interface called int.

Solution 4:

Type t = typeof(obj1);
if (t == typeof(int))
    // Some code here

This is an error. The typeof operator in C# can only take type names, not objects.

if (obj1.GetType() == typeof(int))
    // Some code here

This will work, but maybe not as you would expect. For value types, as you've shown here, it's acceptable, but for reference types, it would only return true if the type was the exact same type, not something else in the inheritance hierarchy. For instance:

class Animal{}
class Dog : Animal{}

static void Foo(){
    object o = new Dog();

    if(o.GetType() == typeof(Animal))
        Console.WriteLine("o is an animal");
    Console.WriteLine("o is something else");
}

This would print "o is something else", because the type of o is Dog, not Animal. You can make this work, however, if you use the IsAssignableFrom method of the Type class.

if(typeof(Animal).IsAssignableFrom(o.GetType())) // note use of tested type
    Console.WriteLine("o is an animal");

This technique still leaves a major problem, though. If your variable is null, the call to GetType() will throw a NullReferenceException. So to make it work correctly, you'd do:

if(o != null && typeof(Animal).IsAssignableFrom(o.GetType()))
    Console.WriteLine("o is an animal");

With this, you have equivalent behavior of the is keyword. Hence, if this is the behavior you want, you should use the is keyword, which is more readable and more efficient.

if(o is Animal)
    Console.WriteLine("o is an animal");

In most cases, though, the is keyword still isn't what you really want, because it's usually not enough just to know that an object is of a certain type. Usually, you want to actually use that object as an instance of that type, which requires casting it too. And so you may find yourself writing code like this:

if(o is Animal)
    ((Animal)o).Speak();

But that makes the CLR check the object's type up to two times. It will check it once to satisfy the is operator, and if o is indeed an Animal, we make it check again to validate the cast.

It's more efficient to do this instead:

Animal a = o as Animal;
if(a != null)
    a.Speak();

The as operator is a cast that won't throw an exception if it fails, instead returning null. This way, the CLR checks the object's type just once, and after that, we just need to do a null check, which is more efficient.

But beware: many people fall into a trap with as. Because it doesn't throw exceptions, some people think of it as a "safe" cast, and they use it exclusively, shunning regular casts. This leads to errors like this:

(o as Animal).Speak();

In this case, the developer is clearly assuming that o will always be an Animal, and as long as their assumption is correct, everything works fine. But if they're wrong, then what they end up with here is a NullReferenceException. With a regular cast, they would have gotten an InvalidCastException instead, which would have more correctly identified the problem.

Sometimes, this bug can be hard to find:

class Foo{
    readonly Animal animal;

    public Foo(object o){
        animal = o as Animal;
    }

    public void Interact(){
        animal.Speak();
    }
}

This is another case where the developer is clearly expecting o to be an Animal every time, but this isn't obvious in the constructor, where the as cast is used. It's not obvious until you get to the Interact method, where the animal field is expected to be positively assigned. In this case, not only do you end up with a misleading exception, but it isn't thrown until potentially much later than when the actual error occurred.

In summary:

• If you only need to know whether or not an object is of some type, use is.

• If you need to treat an object as an instance of a certain type, but you don't know for sure that the object will be of that type, use as and check for null.

• If you need to treat an object as an instance of a certain type, and the object is supposed to be of that type, use a regular cast.

Solution 5:

If you're using C# 7, then it is time for an update to Andrew Hare's great answer. Pattern matching has introduced a nice shortcut that gives us a typed variable within the context of the if statement, without requiring a separate declaration/cast and check:

if (obj1 is int integerValue)
{
    integerValue++;
}

This looks pretty underwhelming for a single cast like this, but really shines when you have many possible types coming into your routine. The below is the old way to avoid casting twice:

Button button = obj1 as Button;
if (button != null)
{
    // do stuff...
    return;
}
TextBox text = obj1 as TextBox;
if (text != null)
{
    // do stuff...
    return;
}
Label label = obj1 as Label;
if (label != null)
{
    // do stuff...
    return;
}
// ... and so on

Working around shrinking this code as much as possible, as well as avoiding duplicate casts of the same object has always bothered me. The above is nicely compressed with pattern matching to the following:

switch (obj1)
{
    case Button button:
        // do stuff...
        break;
    case TextBox text:
        // do stuff...
        break;
    case Label label:
        // do stuff...
        break;
    // and so on...
}

EDIT: Updated the longer new method to use a switch as per Palec's comment.