On Fri, 17 Feb 2023 at 01:28, Frederick Virchanza Gotham via Std-Proposals <std-proposals@lists.isocpp.org> wrote:
On Fri, Feb 17, 2023 at 2:10 AM Thiago Macieira wrote:
>
> auto Func() const && noexcept noreentry && -> S &&;

Edward Catmar poststed:
>And how do you specify which allocator to use for the stable per instance state?


I have revised the syntax. You can use normal stack variables for the
mutexes/flags by doing:

    void Func(void) const && noexcept noreentry -> T &&;

But if you want it to be movable, you do:

    void Func(void) const && noexcept noreentry<std::allocator> -> T &&;

Of course, you can put any allocator you want in there. The above line
can also be written in shorthand:

    void Func(void) const && noexcept noreentry<new> -> T &&;


Edward Catmar posted:
> Come to think of it, what happens if that function is invoked on a moved from object?

Not sure what you mean here. I don't see why this would be more
complicated than if the function were invoked on an object that hasn't
been moved.

A moved from object doesn't own a lock.

By the way in previous code examples posted by other people, I can
understand where you've used unique_ptr<atomic_flag> or
unique_ptr<mutex> in order to make the object movable. I don't however
understand how you would use shared_ptr here, because then you could
have two objects in existence but only one lock (because the two
shared_ptr's refer to the same lock). I don't understand.

So they share a lock. Presumably there are use cases for only being able to use one object at a time, for a set of objects linked in this way.