What is consteval?

Question

Apparently, consteval is going to be a keyword in C++20. The cppreference page for it is currently blank. What is it going to be and how does it relate to constexpr?

Beginner

"Apparently, consteval is going to be a keyword in C++20" - provide a link to support this assertion. – Beauvais

@NeilButterworth The cppreference page I already linked to claims it to be. – Beginner

decltype(std::declval<f()>)'ish but .. fun stuff. – Electromotive

The C++20 Standard has not yet been published, so cppreference (or anyone else) can't say what will appear in it and what the semantics of it will be - if it exists at all. – Beauvais

@NeilButterworth: By that reasoning, nobody can even call it "C++20" or say that there will even be a next version of C++. It is not unreasonable to ask about upcoming features that have been approved in accord with WG21 procedures at the various meetings. Like the most recent one. – Contagium

cppreference now provides some info – Nosography

Answer 1

It declares immediate functions, that is, functions that must be evaluated at compile time to produce a constant. (It used to be spelled constexpr! in a previous revision of the paper.) In contrast, constexpr functions may be evaluated at compile time or run time, and need not produce a constant in all cases.

The adopted paper is P1073R3, which is not yet publicly available, but a previous revision is available and the introductory (motivation and high-level description) portion is about the same (except that the "Source Locations" section is deleted in R3).

Wendeline

Can we overload based on consteval-ness? – Cresting

No, you can't, just like constexpr. – Wendeline

It might be worth pointing out that the motivation is to have functions that rely on compiler data structures that need not be preserved in the binary. – Worley

@Cresting You don't need to, but instead of overloading you can test an expression on compile-timeness by usage either noexcept operator (for msvc or gcc) or through the __builtin_constant_p operator (for clang or gcc). Read these for details: * #13299894 * reddit.com/r/cpp/comments/7c208c/… – Armenta

@Andry: I don't think that these are 100% suited for the task. noexcept has its problems, as far as I remember. And __builtin_constant_p is not usable with gcc, as its output depends on optimization level. – Cresting

@Cresting If you don't beleave then you can always test the code generation in godbolt.org with disabled optimization (-Od in gcc or /Ox in msvc). Hint: you can pass the result of noexcept or __builtin_constant_p to a class template bool argument to guarantee the compile-timeness. – Armenta

@Andry: it's not just about compile-timedness. It's about that the result is wrong. Check out: godbolt.org/z/12aZCH. The result depends on optimization level. – Cresting

@Cresting may be particularly for gcc you have to use noexcept instead of __builtin_constant_p because it works in clang as expected: godbolt.org/z/e4V41F (just switch to clang and see for yourself) – Armenta

@Andry: about noexcept, check out the accepted answer at the question you've linked. I'm not sure about the current (c++17/c++20) state of this trick. – Cresting

@Cresting accepted answers on the stackoverflow is not always the best, especially after the time have pass – Armenta

A few useful concrete examples can be found at awfulcode.io/2019/01/26/immediate-functions-in-c20 – Forewarn

Does this mean consteval functions aren't available at runtime? – Culpable

That's correct, @MarcusJ, consteval functions won't be available at runtime because there's no way to call them unless all their parameters are known at compile-time. And of course, if that's true, the compiler can figure out (at compile-time) what those routines return, with no need to actually generate code for them. – Unprepared