`<functional>`: `move_only_function` behaves differently when constructed with null function pointer in slightly different ways

[AlexGuteniev]

The following program:

#include <functional>
#include <print>

int main() {
    std::move_only_function<int(int)> fn1(static_cast<int(*)(int)>(nullptr));
    std::move_only_function<int(int)> fn2(std::in_place_type<int(*)(int)>, nullptr);
    std::print("fn1: {}, fn2: {}", static_cast<bool>(fn1), static_cast<bool>(fn2));
}

Prints:

fn1: false, fn2: true

This is what the Standard is apparently asking for.

Isn't this stupid?

[frederick-vs-ja]

I think this deserves an LWG issue.

Perhaps we want to say (adding to [func.wrap.move.ctor]/14)

Let rf be an xvalue of type VT referring to the target object that would be originally created. If move_only_function(rf) would construct a move_only_function with no target object, the target object is not created and *this has no target object.

(same for copyable_function)

[AlexGuteniev]

You can go ahead and try to file one, as you have vision on the proposed resolution phrasing.
But I'm afraid it would be closed with "paper needed" resolution, as this seem to change the intention.