Should you use final?

Today, let’s discuss a bit about the not too frequently used final specicfier. I think there is not a lot of reason to use which is supported by the core guidelines. But even when you’d use it, probably you should not.

Today, let’s discuss the not-too-frequently-used final specifier. I think there aren’t many good reasons to use it — something the C++ Core Guidelines also suggests. And even when you could use it, you probably shouldn’t.

final is not a keyword, “only” an “identifier with special meaning”. Technically, it can be used as a name for objects or functions, but it can have a special meaning if used at the right place. If you want readable code though, don’t use it as an identifier!

final is not technically a keyword, it’s an “identifier with special meaning”. It can be used as a name for objects or functions, but if used in the right place, it gains special behavior. For readable code, though, don’t use it as an identifier!

Let’s start with a quick recap of the final specifier, then move on to a nasty example I recently ran into.

What is final

In C++, final is a specifier used to prevent further derivation or overriding. It can be applied to both classes and virtual functions.

final classes

When final is applied to a class, it prevents other classes from inheriting from it.

1
2
3
4
5
6

class Base {};

class Derived final : public Base {};  // OK

class AnotherDerived : public Derived {};  // Error! Derived is final

final virtual functions

When final is applied to a virtual function, it prevents derived classes from overriding it.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

class Base {
public:
    virtual void foo() {} 

    // void bar() final {} // error: only virtual member functions can be marked 'final'
};

class Derived : public Base {
public:
    void foo() final {}
};

class AnotherDerived : public Derived {
	// void foo() override {}  // error: foo() is marked final in Derived
};  

It’s worth noting that you cannot mark a non-virtual function as final.

When to use it?

Don’t use final a lot. More than that, use it less than you probably would! The Core Guidelines talks about final usage in two rules.

C.139 says that you should use final on classes sparingly. The reason? It’s rarely needed for logical reasons and can hurt the extensibility of your class hierarchies.

Some claim that marking classes final improves performance, but that’s more of an urban legend — or at least only true in specific cases. As always: measure.

In the next section, I’ll show how a final class can be too limiting.

Before that, let’s briefly mention the other rule, C.128. It says that (virtual) functions should specify exactly one of virtual, override or final. Using only one is explicit and precise:

• virtual -> “this is a new virtual function”.
• override -> “this is a non-final overrider”
• final -> “this is a final overrider.”

Though you should use final on functions just as sparignly as on classes.

Don’t use final on mocks!

Recently, I was cleaning up the outputs of a test harness. To reduce noise, I decided to wrap a mock class with NiceMock<T>.

If you set no expectations on a mocked call, but it’s invoked, the test harness will report “uninteresting calls”. Using NiceMock<T> suppresses those. If you use StrictMock<T>, uninteresting calls will cause test failures. Look for more info here.

When I tried to compile, it failed! After some wondering and digging, I discovered that NiceMock<T> and StrictMock<T> both inherit from T. In fact, they use the CRTP design pattern. So, if you try to use NiceMock<T> on a class T that has been marked final then the compilation will fail, because NiceMock cannot inherit from T.

Too bad…

If you have the rights to change the source code of T, good for you. Go ahead and consider remove final from class declaration. If you cannot do that, you might ask the owners to do so or you simply have to give up on using NiceMock.

Nevertheless, if you are using gmock and you are writing some mock classes, don’t mark then final. Don’t make life harder for your users or for yourself.

Conclusion

In this article, we talked about the final specifier. You can use it on classes and member functions to prevent inheritance and overrides, respectively.

While it may seem useful at first glance, it doesn’t offer performance benefits in most cases—and it often just adds unnecessary constraints. That’s why the Core Guidelines recommend using it sparingly.

We saw a concrete example where marking a mock class as final prevented its use with NiceMock<T> or StrictMock<T>, since they inherit from the mock class behind the scenes.

Connect deeper

If you liked this article, please

• hit on the like button,
• subscribe to my newsletter