Covariance and Contravariance FAQ

Covariance and Contravariance FAQ

Rate This
  • Comments 35

In this post I’ll try to answer the most common questions I find on forums and in documentation feedback about C# covariance and contravariance. It’s a big topic for a single blog post, so expect to see a lot of “more information” links.

Special thanks to Eric Lippert and Chris Burrows for reviewing and providing helpful comments.

What are covariance and contravariance?

In C#, covariance and contravariance enable implicit reference conversion for array types, delegate types, and generic type arguments.Covariance preserves assignment compatibility and contravariance reverses it.

The following code demonstrates the difference between assignment compatibility, covariance, and contravariance.

// Assignment compatibility.
string str = "test";
// An object of a more derived type is assigned to an object of a less derived type.
object obj = str;

// Covariance.
IEnumerable<string> strings = new List<string>();
// An object that is instantiated with a more derived type argument
// is assigned to an object instantiated with a less derived type argument.
// Assignment compatibility is preserved.
IEnumerable<object> objects = strings;

// Contravariance.           
// Assume that I have this method:
// static void SetObject(object o) { }
Action<object> actObject = SetObject;
// An object that is instantiated with a less derived type argument
// is assigned to an object instantiated with a more derived type argument.
// Assignment compatibility is reversed.
Action<string> actString = actObject;

In C#, variance is supported in the following scenarios:

  1. Covariance in arrays (since C# 1.0)
  2. Covariance and contravariance in delegates, also known as “method group variance” (since C# 2.0)
  3. Variance for generic type parameters in interfaces and delegates (since C# 4.0)

What is array covariance?

Arrays are covariant since C# 1.0. You can always do the following:

object[] obj = new String[10];

In the above code, I assigned an array of strings to an array of objects. So I used a more derived type than that originally specified, which is covariance.
Covariance in arrays is considered “not safe,” because you can also do this:

obj[0] = 5;

This code compiles, but it throws an exception at run time because obj is in fact an array of strings and cannot contain integers.

What is delegate, or method group, variance?

This feature was added in C# 2.0. When you instantiate a delegate, you can assign it a method that has a more derived return type than that specified in the delegate (covariance). You can also assign a method that has parameter types less derived than those in the delegate (contravariance).

Here’s a quick code example illustrating the feature and some of its limitations.

static object GetObject() { return null; }
static void SetObject(object obj) { }

static string GetString() { return ""; }
static void SetString(string str) { }

static void Main()
{
    // Covariance. A delegate specifies a return type as object,
    // but I can assign a method that returns a string.
    Func<object> del = GetString;

    // Contravariance. A delegate specifies a parameter type as string,
    // but I can assign a method that takes an object.
    Action<string> del2 = SetObject;

    // But implicit conversion between generic delegates is not supported until C# 4.0.
    Func<string> del3 = GetString;
    Func<object> del4 = del3; // Compiler error here until C# 4.0.
}

By the way, this feature works for all delegates, both generic and non-generic, not just for Func and Action delegates.

For more information and examples, see Covariance and Contravariance in Delegates on MSDN and Eric Lippert’s post Covariance and Contravariance in C#, Part Three: Method Group Conversion Variance.

What is variance for generic type parameters?

This is a new feature in C# 4.0. Now, when creating a generic interface, you can specify whether there is an implicit conversion between interface instances that have different type arguments. For example, you can use an interface instance that has methods with more derived return types than originally specified (covariance) or that has methods with less derived parameter types (contravariance). The same rules are applied to generic delegates.

While you can create variant interfaces and delegates yourself, this is not the main purpose for this feature. What is more important is that a set of interfaces and delegates in .NET Framework 4 have been updated to become variant.
Here’s the list of updated interfaces:

And the list of updated delegates:

The most frequent scenario for most users is expected to be something like this one:

IEnumerable<Object> objects = new List<String>();

While this code doesn’t look that impressive, it allows you to reuse a lot of methods that accept IEnumerable objects.

class Program
{
    // The method has a parameter of the IEnumerable<Person> type.
    public static void PrintFullName(IEnumerable<Person> persons)
    {
        // The method iterates through a sequence and prints some info.
    }

    public static void Main()
    {
        List<Employee> employees = new List<Employee>();

        // I can pass List<Employee>, which is in fact IEnumerable<Employee>,
        // although the method expects IEnumerable<Person>.
        PrintFullName(employees);
    }
}

A couple of important rules to remember:

  • This feature works only for generic interfaces and delegates. If you implement a variant generic interface, the implementing class is still invariant. Classes and structs do not support variance in C# 4.0.
    So the following doesn’t compile:
    // List<T> implements the covariant interface
    // IEnumerable<out T>. But classes are invariant.
    List<Person> list = new List<Employee>(); // Compiler error here.
  • Variance is supported only if a type parameter is a reference type. Variance is not supported for value types.
    The following doesn’t compile either:
    // int is a value type, so the code doesn't compile.
    IEnumerable<Object> objects = new List<int>(); // Compiler error here.

Where can I find more examples of using covariance and contravariance?

I wrote a couple of MSDN topics that show how you can benefit from this new feature. They might help you better understand the principles of covariance and contravariance:

Also, take a look at the video How Do I: Use Covariance and Contravariance in VS 2010 Part I? by Eric Lippert.

How can I create variant generic interfaces and delegates myself?

The out keyword marks a type parameter as covariant, and the in keyword marks it as contravariant. The two most important rules to remember:

  • You can mark a generic type parameter as covariant if it is used only as a method return type and is not used as a type of formal method parameters.
  • And vice versa, you can mark a type as contravariant if it is used only as a type of formal method parameters and not used as a method return type.

For more information about variance validation, read Creating Variant Generic Interfaces and Variance in Delegates on MSDN and Eric Lippert’s post Exact rules for variance validity.

This example shows how to create a variant generic interface:

interface IVariant<out R, in A>
{
    // These methods satisfy the rules.
    R GetR();
    void SetA(A sampleArg);
    R GetRSetA(A sampleArg);

    // And these don’t.
    // A GetA();
    // void SetR(R sampleArg);
    // A GetASetR(R sampleArg);
}

If you extend a variant interface, the extending interface is invariant by default. You must explicitly specify whether the type parameters are covariant or contravariant by using the out or in keyword. Here is a quick example from MSDN:

interface ICovariant<out T> { }

// This interface is invariant because I didn't use the "out" keyword.
interface IInvariant<T> : ICovariant<T> { }

// And this one is covariant because I explicitly specified this.
interface IExtCovariant<out T> : ICovariant<T> { }

And once again, this feature is supported for generic interfaces and delegates only. So the following doesn’t compile:

class Sample<out T> { }  // Compiler error here.

For more examples, take a look at the video How Do I: Use Covariance and Contravariance in VS 2010 Part II? by Eric Lippert.

Where can I find more in-depth information about covariance and contravariance?

This is the MSDN root topic: Covariance and Contravariance.

And, of course, read Eric Lippert’s blog. He designed this feature for C# 4.0, so who knows more about it?

Leave a Comment
  • Please add 2 and 6 and type the answer here:
  • Post
  • "Wouldn't it be possible to define IList<T> as a specialization of IList<in Tin, out Tout>?"

    You could define an IList<in TIn, out TOut>, minus the indexer property (which is read/write in IList<T>). But it seems to me the usefulness of such an interface is dubious. For reading, we'll already have the covariant IEnumerable<out T> implemented by IList<T>, which is half the battle. So that leaves writing; but does that really come up that much? I'm having a hard time imagining a practical scenario where IList<in TIn, out TOut> actually makes the code better.

    And of course, if by "as a specialization of", you mean that IList<T> would implement IList<in TIn, out TOut>, well since we already have a definition of IList<T> in .NET, I don't see how that can be added without breaking all the existing code that currently implements IList<T>.

  • @pete.d

    You are right about IList<T>. I even deleted my original comment, since yours is more relevant.

    As for variance on classes - just curios what scenarios do you have in mind? I see that the most often request is variance for collections, and it is not safe, unless collection is read-only.

    @James

    I am not sure what you mean here. I am not an expert in C++. You can create IMyList<in Tin, out TOut>, the syntax is supported. But what exactly do you want to achieve here?

  • @Alexandra Rusina: "As for variance on classes - just curios what scenarios do you have in mind?"

    Sorry if I wasn't clear. Your comment is exactly my point. The likelihood of having a class that actually _can_ be variant is low IMHO. In fact, even a read-only collection has to get the data into it _somehow_ (e.g. the constructor).

    A variant class is certainly theoretically possible. A read-only one that creates output from scratch, or accepts input from some method that's not using the generic type, for example. Likewise a write-only class that only sends data somewhere, never to be seen again. But even if one assumes that there are valid scenarios where a variant class might be useful, surely these scenarios are exceedingly rare, and can be addressed using interfaces anyway, and thus probably do not justify the cost in adding such a feature to the language.

    (I take it for granted that we all remember, a new language feature has to be more than just useful in some way. It has to be _enough_ useful to justify the cost involved in implementing it).

  • I think I may be missing something, I do not think this is correct, as the SetObject returns void, and is not called with anything? Can you clarify what is going on here?

    // Assume that I have this method:

    // static void SetObject(object o) { }

    Action<object> actObject = SetObject;

  • @Shawn

    Yes, SetObject doesn't return anything but sets some value somewhere based on the "object o" parameter.

    This is why I can initialize Action<object> delegate with this method. Action<delegate> represents a method that returns void and has one parameter of type "object".

    I don't call a method here, only initialize a delegate with it, this is why I don't pass any parameters in this example.

    Does this answer your question?

  • Great article.  I spotted another minor typo.

    "A couple important rules to remember:"

    Should read:

    "A couple of important rules to remember:"

    (note the insertion of the word "of" between couple and important. :-) )

  • @Jason

    Fixed. Thanks.

  • Isnt it just hidden casting?

    IEnumerable<Object> objects = (IEnumerable<Object>) new List<String>();

  • @admg

    In c#3.5 your statement would not compile. List<string> implements IEnumerable<string> which can not be cast to IEnumerable<object>. C#4.0 allows you to cast IEnumerable<string> to IEnumerable<object>.

  • @admg

    The problems that C# had prior to version 4.0 due to the lack of variance support in generics are well-described here:

    http://msdn.microsoft.com/en-us/library/ms228359%28VS.90%29.aspx

    Note that this document is for C# 3.0, because now you should use variance in interfaces instead.

  • I had translate this article to chinese.

    我已将此文章翻译成中文:

    www.cnblogs.com/.../covariance-and-contravariance-faq.html

  • @Alexandra Rusina: I would be glad if out-Parameters could be declared as covariant.

    Currently it seems as if out and ref Parameters are both treated invariant.

    Consider the bool TryGetValue(TKey key, out TValue value) Methods of IDictionary:

    Here the Type 'value' should be declared as covariant.

    Currently I work around by switching both Return Values:

    TValue TryGetValue(TKey key, out bool isValid)

    But using this Signature is not as appealing AND the Parameter Type cannot be derived from a Return Value.

    Will this be supported in the foreseeable Future?

  • Rusina i could't understand what does this mean "// An object of a more derived type is assigned to an object of a less derived type."? what u mean more derived type and less derived type?  which is more derived type and which is less derived type? i am waiting eagerly for this answer so that i can learn this variance :( pls hlp

  • Object is less derived whereas any other type derived from object such as int or strings or such are more derived.

  • May be I am not right, but the first lines of code demonstrates the boxing not assigning of a more derived type to an object of a less derived type. Please correct me if I am wrong.

    string str = "test";

    // An object of a more derived type is assigned to an object of a less derived type.

    object obj = str;

Page 2 of 3 (35 items) 123