The final specifier in C++ marks a class or virtual member function as one which cannot be derived from or overriden. For example, consider the following code:   struct base {    virtual void f() const = 0;  };    struct derived final : base {    void f() const override {}  };
  • addieEnglish
    7·
    1 year ago

    tl:dr; making a class final allows the compiler to save a vtable lookup in some (limited, slightly artificial) places where it couldn’t before. No actual performance measurements.

    I am sceptical. I’d understood vtable lookups to be extremely cheap on modern architecture; usually cheaper than the if / switch statement you’d have to write as an alternative if you weren’t using inheritance. And for really performance-sensitive code, you would never have used virtualized classes anyway. I’d like to see some proper performance results before mangling code and making maintenance difficult.

    • lysdexic@programming.devOPMEnglish
      4·
      1 year ago

      I am sceptical. I’d understood vtable lookups to be extremely cheap on modern architecture; usually cheaper than the if / switch statement you’d have to write as an alternative if you weren’t using inheritance.

      They are relatively cheap, but I think cheap does not mean free and some references mention a performance hit that can be as high as 7%.

      And for really performance-sensitive code, you would never have used virtualized classes anyway.

      Yeah, because of this performance hit.

      One of the cool things about this article is the fact that it points out how adding ‘final’ can magically get rid of that performance hit.