Equality in C#


In this article I’m going to discuss the notion of equality in the C# programming language. In particular, I’m going to show you how it’s very difficult to correctly implement comparisons of objects in the presence of inheritance. This will lead up to the conclusion that it’s usually best to stick to implementing equality comparisons only for value types (i.e. structs) or classes with value-like semantics (i.e. sealed classes, usually with immutable fields). These conclusions cover all comparisons which are implemented by extending a struct or class such as:

Comparisons are often best left to the various mechanisms offered elsewhere in the .NET Framework such as:

A quick note before you proceed: make sure you understand why the following program behaves the way it does:

Now on to the discussion proper. Caveat: My code probably has subtle bugs in it. This is kind of the point.

Here’s our base class, imaginatively named BaseClass:

We’ve decided, for whatever reason, that BaseClass is to be value-like:

Other features of note:

Most discussions of IEquatable<T> state that this interface is provided for more strongly-typed comparisons as well as to allow a more efficient implementation. While it can achieve the former, it cannot always achieve the latter. If the type T is a struct, then we certainly can provide a more efficient implementation than the one implemented for Equals(object) that is generated for us by the C# compiler (which performs the comparison by reflecting over the type T to enumerate all of its fields—I need to verify this claim at some point). However, providing a custom behaviour for Equals(T) without providing a consistent implementation for Equals(object) will break comparisons. Weird behaviour will ensue if objects are ever compared via virtual Equals(object) calls instead of static Equals(T) calls: you wouldn’t want to live in a world where 123.Equals(123) yielded a different result from ((object)123).Equals((object)123). In practice, therefore, you’ll want to delegate your Equals(T) call to your Equals(object) implementation so that they are guaranteed to be equivalent.

In the case of a class like this which attempts to handle inheritance, no performance enhancement is gained by implementing IEquatable<T> since all comparison methods must dynamically dispatch via virtual method calls.

Here’s a derived class named DerivedClass:

We’ve decided, for whatever reason, that DerivedClass is also to be value-like:

Of note:

Pretty ugly, huh?

To drive the point home, let’s define another class, MoreDerivedClass, that derives from DerivedClass:

We’ve decided that MoreDerivedClass like its ascendants is also to be value-like:

Well, this all sucks. Here are some tests:

Note that these tests are by no means exhaustive: it covers only a subset of the various Equals overrides and does not test GetHashCode at all.

This is a lot of work and any given implementation is likely to contain subtle bugs. Some of this complexity and bugginess could be eliminated through the use of code generation since each of these overload methods is more-or-less mechanically generatable. This is roughly what @EqualsAndHashCode in Lombok does. Writing this code by hand is generally tedious and pointless. Life is simpler if we stick to only comparing value or properly value-like types. Doing this kind of work requires that the developer spend some time thinking about why he or she is contemplating it. Often it will be to take advantage of language or framework features such as hash tables and sets etc. There will often require a value-like object as key and developers will be tempted to slap an Equals or IComparable on a class to make this work. This is usually the Wrong Thing to do. A more principled approach to designing systems like this will be to try to determine what the key in such a system is and then probably implement around that: these keys will typically be stable, immutable values or value-like objects.

I’ll probably talk more on this subject in the future. Bye for now!

Update 2020-03-12

Here’s an interesting, related project from my co-worker Jay Bazuzi: ValueTypeAssertions.

Related posts

Example of a serializable exception in C#



Content © 2010–2020 Richard Cook. All rights reserved.