`<functional>`: `std::not_fn` can accept non-movable function

[hewillk]

It seems that std::not_fn should reject the following code according to [func.not.fn]:

Mandates: is_constructible_v<FD, F> && is_move_constructible_v<FD> is true.

godbolt:

#include <functional>

struct OnlyCopyableFun {
  OnlyCopyableFun() = default;
  OnlyCopyableFun(const OnlyCopyableFun&) = default;
  OnlyCopyableFun(OnlyCopyableFun&&) = delete;
  bool operator()(auto) const;
};

int main() {
  OnlyCopyableFun f;
  auto nf = std::not_fn(f); // only ill-formed in libc++
}

Feel free to correct me if I've missed something.
Not quite sure why these standard call wrapper factories (std::bind_front, std::bind, etc.) require all argument types to be move-constructible, although the call wrapper produced in the above example is still move-constructible as its underlying type has a copy constructor.
It seems to me that just requiring is_constructible_v<FD, F> is enough, which is what the range adaptor object does.

[frederick-vs-ja]

I think we should file an LWG issue to reduce the Mandates and Preconditions of these functions.

[StephanTLavavej]

Casey Carter (@CaseyCarter) noted that the Standard (via the "call wrapper" phrase of power that I forgot about) requires not_fn() to produce something that's Cpp17MoveConstructible, so the Mandates on it makes sense and its presence isn't a defect.

N4958 [func.require]/3: "Every call wrapper meets the Cpp17MoveConstructible and Cpp17Destructible requirements. An argument forwarding call wrapper is a call wrapper [...]"

/4: "A perfect forwarding call wrapper is an argument forwarding call wrapper that [...]"

[func.not.fn]/4 "Returns: A perfect forwarding call wrapper"

Therefore, this is an implementation issue where we should be static_asserting explicitly, instead of assuming that the implementation will "naturally diagnose" the requirement which is not the case here.