Question

Is it a valid recipe to say that if I write final (to a member function) one should not write virtual?

In a base class, final methods would make no sense:

struct Driver {
    virtual void print();
};

If one would add final to print() this would defy the reason for polymorphism in the first place. So that would be useless (though possible).

When I derive from this class I can detect errors with final, but only without virtual:

struct KeyboardDriver : public Driver {
    virtual void prynt() final;    // Oops: typo, but compiler-ok
};

struct MouseDriver : public Driver {
    void prynt() final;    // Error: Hooray, compiler found my typo
};

The additional final for KeyboardDriver::prynt was legal. Because final only requires the member function to be virtual — the compiler lets this pass (FDIS 9.2p9).

But when I leave out the virtual the typo makes this function non-virtual — it overrides nothing, i.e. no virtual function. Therefore final without virtual serves the same purpose to this respect as override does.

Update: Is my analysis correct? Does the functionality of final without virtual include that one of override?

Answer 1

Your analysis is correct. Not marking a final member as virtual avoids declaring a new, non-overriden member and catches mistakes. However, since that’s the purpose of override, it may be simpler to just use final override; then whether the member is marked virtual or not won’t matter at all.