Hence the c++20 [[likely]] attribute…

Hey we argue recently about thoses if(){return} guards with my colleagues and there was an argument that rose up in the discussion : performance.
thanks to a compiled code viewer (such as https://godbolt.org/ for instance) we saw that a code with a return at the beginning has 2 JMP instructions for the “normal” case, whereas if the code is inside the “if” statement the normal case uses no JMP at all.
And even if the compiler can optimize it for simple cases, we cannot be sure it will be properly done for more complex ones.
Hence the instruction is to not use those guards in function which are called a lot (in update loop for example).

Well, what I have in mind is a style of programming I haven’t seen anywhere: instead of returns, functions emit events, and callers can hook continuations to those events.

It’s not strictly related to guards though, as a topic.

Hey Achilleas, could you please clarify for me the link between early returns and event-based programming?

double applyDiscount(double price)
{
    return reductionType_ == Percentage
        ? price * (1 – reductionValue_)
        : return price – reductionValue_
        ;
}

Event-based programming. Unfortunately it does not seem to be easy to do with C++.

🙂 !