On Thu, Jul 29, 2021 at 8:21 AM Григорий Шуренков via Std-Proposals <
std-proposals_at_[hidden]> wrote:

> Hi!
>
> As far as I remember boost::optional<T&> has rather surprising assignment
> semantics when assigned from T& (rebind for empty optional and
> assign-through-reference for non-empty one).
>

It does not, there is no assign-through.


> And it's the best it can achieve if it is modeled after optional<T>.
> That's why it was not standardized.
>

There is no reason why optional<T&> couldn't rebind on assignment. Multiple
implementations of optional do this.


> Maybe we just need another type - optional_reference<T>? It can have
> different behavior more suitable for reference type.
>
> Regards,
> Gregory
>

The advantage of optional<T&> being spelled specifically optional<T&> is
that it just works in generic code. optional<decltype(f())> is valid. This
comes up a lot when you want to write operations that return a U or nothing
and you don't know whether U is a reference type or not. An example here
might be an algorithm that returns a reference (which may or may not be a
true reference) to the first element of a range, if there is one. You just
want to return optional<U>.

You don't want to return conditional_t<is_reference_v<U>,
optional_reference<remove_reference_t<U>>, optional<U>>. I mean, doing that
repeatedly would be a pain, so you'd just write something like...
real_optional<U> that is either optional_reference<remove_reference_t<U>>
or optional<U>. And then of course you'd want to ensure the same semantics
in those cases, since the user of this algorithm might also be a different
algorithm. And then why not just spell real_optional<U> as optional<U>?

Barry