<memory>: !_Get_ptr() in the destructors of inout_ptr_t and out_ptr_t _should_ be correct

[frederick-vs-ja]

Describe the bug

Currently !_Get_ptr() is used in the destructors of inout_ptr_t and out_ptr_t.

STL/stl/inc/memory

Lines 4409 to 4411 in a26f4ad

	if (!_Get_ptr()) {
	return;
	}

STL/stl/inc/memory

Lines 4328 to 4330 in a26f4ad

	if (!_Get_ptr()) {
	return;
	}

However, it seems that nothing in Cpp17NullablePointer ([nullablepointer.requirements]) requires !p to be semantically equivalent to p != nullptr or even to be well-formed. E.g. a type with a deleted member operator! (the NoExclamation type shown below) can still meet the Cpp17NullablePointer requirements.

#include <memory>

struct NoExclamation {
    using pointer = NoExclamation;

    void* p_{};

    NoExclamation() = default;
    constexpr NoExclamation(decltype(nullptr)) noexcept {}

    friend bool operator==(NoExclamation, NoExclamation) = default;
    friend constexpr bool operator==(NoExclamation lhs, decltype(nullptr)) noexcept
    {
        return lhs.p_ == nullptr;
    }

    constexpr explicit operator bool() const noexcept
    {
        return p_ != nullptr;
    }

    void operator!() const = delete; // HERE!
};

int main()
{
    NoExclamation ne{};
    std::out_ptr(ne);
}

Currently no implementation accepts the example due to the deleted operator!.

Command-line test case

Godbolt link

Expected behavior

Perhaps the program should compile.

STL version

a26f4ad (probably every version after #1998)

Additional context

As libc++ and libstdc++ also expect usable operator!, should there be an LWG issue to make current implementation strategies comforming?

[cpplearner]

[nullablepointer.requirements]:

An object p of type P can be contextually converted to bool. The effect shall be as if p != nullptr had been evaluated in place of p.

Pedantically, this doesn't require !p to be meaningful (nor p && p or p || p, nor even implicit conversion to bool). But the intent should be quite clear to a non-pedant.

[frederick-vs-ja]

[nullablepointer.requirements]:

An object p of type P can be contextually converted to bool. The effect shall be as if p != nullptr had been evaluated in place of p.

Pedantically, this doesn't require !p to be meaningful (nor p && p or p || p, nor even implicit conversion to bool). But the intent should be quite clear to a non-pedant.

Sounds like that contextual version of boolean-testability is wanted.

[StephanTLavavej]

We talked about this at the weekly maintainer meeting. The fact that the Majestic Three implementations all want !p to behave reasonably is a strong indication to us that Cpp17NullablePointer should be requiring boolean-testable ([concept.booleantestable]) and that we should file an LWG issue rather than attempting to contort our implementation to handle absurdly pathological types.