LWG-4015 LWG-3973 broke `const` overloads of `std::optional` monadic operations

[StephanTLavavej]

LWG-4015 LWG-3973 broke const overloads of std::optional monadic operations

It looks like this may be unobservable for us right now, especially because in #4158 I noted that the earlier LWG-3973 was "Implemented without comments". However, I do observe variation between our current implementation and LWG-4015's resolution. For example, our converting move constructor says _STD move(*_Right) (i.e. deref optional, then move):

STL/stl/inc/optional

Lines 292 to 296 in ab6dec6

	template <class _Ty2, enable_if_t<conjunction_v<_AllowUnwrapping<_Ty2>, is_constructible<_Ty, _Ty2>>, int> = 0>
	_CONSTEXPR20 explicit(!is_convertible_v<_Ty2, _Ty>) optional(optional<_Ty2>&& _Right)
	noexcept(is_nothrow_constructible_v<_Ty, _Ty2>) /* strengthened */ {
	if (_Right) {
	this->_Construct(_STD move(*_Right));

But LWG-4015 specifies std::move(rhs).operator*(), i.e. move optional, then deref.

I'll note this issue in #5607 tracking optional<T&>, but I think we can clean up divergence here before implementing that feature.

[khaledS20]

Hi,
I'd like to work on this issue.

[frederick-vs-ja]

Just do it. You can test this with ADL-found operator* like, e.g.

namespace bad {
    template<class T>
    void operator*(const optional<T>&) = delete;

    struct S { /* ... */ };
} // namespace bad

// testing optional<bad::S>