Implement RcUninit (#112566)

[kstrafe]

RcUninit allows the user to construct cyclic data structures without using Rc::new_cyclic, which allows cyclic constructions across await points. It also allows you to create long linked lists without overflowing the stack.

This is an alternative to UniqueRc. While UniqueRc does allow for cyclic data structures to be created, it must be done by mutating the UniqueRc. Mutation is prone to creating reference cycles. Construction-only assignment of fields, without any mutation to "set" the struct afterwards cannot generate reference cycles. It's also more cumbersome to work with.

For instance, if we have objects A, B, and C, and we want these to connect as A => B => C -> A (=> being strong, -> being weak), then we must do something along the following lines.

    let mut a_uniq = UniqueRc::new(A::new());
    let a_weak = UniqueRc::downgrade(&a_uniq);

    let c = Rc::new(C::new(a_weak));
    let b = Rc::new(B::new(c));

    a_uniq.set_b(b);
    let a = a_uniq.into_rc();

To implement A::set_b, the field A::b must either be

• Option<Rc<B>>: Requiring unwrap/clone for each access.
• MaybeUninit<Rc<B>>: Requiring unsafe.
• Weak<B>: Requiring upgrade for every access.

The above also makes it easier to make mistakes in more complex programs where we don't have the full picture. It is not hard to change the above into Rc<RefCell<A>>, and then provide this pointer to C, which would cause a reference cycle to be created once a.borrow_mut().set_b(b) gets called.

On the other hand RcUninit doesn't have this problem, since initialization is deferred. The equivalent would look like the following.

    let a_uninit = RcUninit::new();
    let b_uninit = RcUninit::new();
    let c_uninit = RcUninit::new();

    let c = c_uninit.init(C::new(a_uninit.weak()));
    let b = b_uninit.init(B::new(c));
    let a = a_uninit.init(b);

This creates the structure (A => B => C -> A)

Tracking Issue

#112566

RcUninit was discussed in
rust-lang/libs-team#90
https://internals.rust-lang.org/t/an-alternative-to-rc-new-cyclic/22849/6
#112566

[rustbot]

r? @thomcc

rustbot has assigned @thomcc.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[rust-log-analyzer]

The job mingw-check failed! Check out the build log: (web) (plain)

Click to see the possible cause of the failure (guessed by this bot)

[thomcc]

I'm going to reroll this one, I don't have all the context and am pretty busy at the moment.

r? libs

[ibraheemdev]

r? libs-api

Just a heads up that this will likely need an ACP as it's different from the original proposal.

[kstrafe]

r? libs-api

Just a heads up that this will likely need an FCP as it's different from the original proposal.

@ibraheemdev any guidance on how to proceed? I've not contributed to Rust yet.

[ibraheemdev]

You can create an ACP proposal on the libs-team repository, similar to the original proposal (rust-lang/libs-team#90). If that is approved by libs-api then this PR can be merged (module any implementation issues).