Most object-oriented languages like C++, Java, and C#, have at least one of two somewhat similar features: interfaces and abstract classes. In principle, both allow you to specify the API for an object without providing implementations for some (or all) of the methods. Interfaces being basically pure abstract classes, that is, abstract classes where every method is abstract, not just some. So why would someone use an interface, which offers only a subset of the functionality of an abstract class?

Abstract classes are a feature of C++, Java and C#, while interfaces are only available in Java and C# (not C++). Java and C# also happen to be the only two of these three languages to only support single—not multiple—inheritance. Interfaces are used in Java and C# to get around the limitations of single inheritance. If you want an object to implement multiple APIs, only one of them can come from a base class, the others must be from interfaces.

In other words, you can make a class that implements as many interfaces as you want in Java and C#, but they can only derive from one base class. Since C++ classes can derive from as many base classes (abstract or not) as they want, interfaces are unnecessary. You can simply replace them with pure abstract classes. By limiting classes to single inheritance, Java and C# avoid the issues that can arise from multiple inheritance. When a class inherits from multiple base classes, there’s an ambiguity about how you traverse the class inheritance graph. Do you do depth-first search, breadth-first search, or something more complicated? For example, Python uses an algorithm called C3 linearization, which is similar to topological sorting. This problem is sometimes called the “diamond problem” after the diamond-shape class inheritance diagram which represents the simplest example that causes it.