sendable

The sendable crate defines types to facilitate sending data between threads:

• SendRc, a single-threaded reference-counting pointer that can be sent between threads. You can think of it as a variant of Rc<T> that is Send if T is Send. This is unlike Rc<T> which is never Send, and also unlike Arc<T>, which requires T: Send + Sync to be Send.
• SendOption, which holds an Option<T> and is Send even if T is not Send.

When is SendRc useful?

You might consider SendRc if:

• your values form an acyclic graph or a hierarchy with cross-references;
• you build and use the hierarchy from a single thread;
• need to occasionally move the whole thing to another thread.

Within the confines of a single thread, using Rc and RefCell to represent acyclic graphs and data sharing is ergonomic and safe. It is also efficient because single-threaded manipulation doesn't require atomics or locks, makes deref() trivial, and allows the compiler to inline borrow() and borrow_mut() and even optimize them away where they are not globally observable.

In programs that process many such graphs it comes in very useful to be able to create them in one thread and ship them to another for processing (and possibly to a third one for destruction). After all, types like RefCell and Cell are Send - they provide interior mutability, but no sharing. The trouble is with Rc, which is neither Send nor Sync, and for good reason. Even when it would be perfectly safe to move an entire hierarchy of Rc<RefCell<T>>s from one thread to another, the borrow checker doesn't allow it because it cannot statically prove that you have moved all of them. If some Rcs pointing to the data that was moved to a new thread remained in the original thread, the unsynchronized manipulation to the contents and the reference counts would exhibit undefined behavior and wreak havoc.

If there were a way to demonstrate to Rust that you've sent all pointers to a particular allocation to a different thread, there would be no problem in moving Rc<T> instances to a different thread, provided that T itself were Send. SendRc does exactly that.

How does SendRc work?

When a SendRc is constructed, it stores the current thread id next to the value and the reference count. On access to the value, and before manipulating the reference count through clone() and drop(), it checks that the SendRc is still in the thread it was created in.

When SendRcs needs to be moved to a different thread, each pointer is explicitly marked for sending using the API provided for that purpose. Once thus marked, access to underlying data from that pointer is prohibited, even in the original thread. When all pointers to an allocation disabled, they can be sent across the thread boundary, and explicitly re-enabled in the new thread. In a simple case of two SendRcs, the process looks like this:

// create two SendRcs pointing to the same allocation
let mut r1 = SendRc::new(RefCell::new(1));
let mut r2 = SendRc::clone(&r1);

// prepare to ship them off to a different thread
let mut pre_send = SendRc::pre_send();
pre_send.disable(&mut r1); // r1 is unusable from this point
pre_send.disable(&mut r2); // r2 is unusable from this point
// ready() would panic on un-disabled SendRcs pointing to the allocation of r1/r2
let mut post_send = pre_send.ready();

// move everything to a different thread
std::thread::spawn(move || {
    // both pointers are unusable here
    post_send.enable(&mut r1); // r1 is usable from this point
    post_send.enable(&mut r2); // r2 is usable from this point
    *r1.borrow_mut() += 1;
    assert_eq!(*r2.borrow(), 2);
})
.join()
.unwrap();

If the SendRcs are edges in a graph, you'll need to visit the whole graph before and after the migration to the new thread. In the pre-send phase you'll need to disable the pointer after visiting its neighbors, whereas in the post-send step you'll need to first re-enable the pointer and then visit the neighbors.

Why not just use Arc?

Arc indeed allows moves between threads, but it fundamentally assumes that the underlying value will be shared between threads. Arc requires T: Send + Sync in order for Arc<T> to be Send because if it only required T: Send, you could create an Arc<RefCell<u32>>, clone it, send the clone to a different thread, and call borrow_mut() from two threads on the same RefCell without synchronization. That is forbidden, and is why Arc<RefCell<T>> is not a thing in Rust.

SendRc can get away with allowing this because it requires proof that all access to the allocated value in the previous thread was relinquished before allowing the value to be pinned to a new thread. SendRc<RefCell<u32>> is sound because if you clone it and send the clone to a different thread, you won't be able to access the data, nor clone or even drop it - any of these would result in a panic.

One could fix the issue by using the full-blown Arc<Mutex<T>> or Arc<RwLock<T>>. However, that slows down access to data because it requires atomics, poison checks, and calls into the pthread API. It also increases the memory overhead because it requires an extra allocation for the system mutex. Even the most efficient mutex implementations like parking_lot don't come for free, and bear the cost of atomic synchronization. But even disregarding the cost, the issue is also conceptual: it is simply wrong to use Arc<Mutex<T>> if neither Arc nor Mutex is actually needed because the code doesn't access the value of T from multiple threads in parallel.

In summary, SendRc<T> is Send, with certains guarantees enforced at run time, the same way an Arc<Mutex<T>> is Send + Sync, with certain guarantees enforced at run time. They just serve different purposes.

Why not use an arena? Or unsafe?

An arena would be an acceptable solution, but to make it Send, it requires the whole design to be devoted to that idea from the ground up. A simple solution of replacing Rc with an arena id doesn't really work because in addition to the id, the object needs a reference to the arena. It can't have an Option<&Arena> field because it would make it non-Send for an arena that contains non-Sync cells. Since we need Arena to contain RefCell, this doesn't work.

There are arena-based designs that do work, but require more radical changes, such as decoupling storage of values from access and sharing. All data is then in the arena, and the accessors are created on-the-fly and have a lifetime connected to the lifetime of the arena. This requires dealing with the lifetime everywhere and is not easy to get right for non-experts.

Finally, one can avoid the arena by just using unsafe impl Send on a root type that is used to send the whole world to the new thread, and borrow checker be damned. That solution is hacky and gives up the guarantees afforded by Rust. If you make a mistake, say by leaving an Rc clone in the original thread, you're back to core dumps like in C++. In Rust we hope to do better, and SendRc is an attempt to make such a sound solution that addresses this scenario.

What about SendOption?

SendOption is a related proposition: a type that holds Option<T> and is always Send, regardless of whether T is Send. Surely that can't be safe?

What makes it work is that SendOption requires you to set the value to None before sending it to another thread. If the inner Option<T> is None, it doesn't matter if T is !Send because no T is actually getting sent anywhere. If you do send a non-None SendOption<T> into another thread, SendOption will use panic to prevent you from accessing it in any way (including by dropping it). Failure to abide by the rules results in a T that was effectively never "sent" to another thread, only its bits were shallow-copied and forgotten, and that's safe.

SendOption is designed for types which are composed of Send data, except for an optional field of a non-send type. The field is set and used only inside a particular thread, and will be None while sent across threads, but since Rust can't prove that, a field of Option<NonSendType> makes the entire outer type not Send. For example, a field with a SendOption<Rc<Arena>> could be used to create a Send type that refers to a single-threaded arena.

Is this really safe?

As with any crate that involves unsafe, one can never be 100% certain that there is no soundness bug. The code is fairly straightforward in implementing the design outlined above. I went through several iterations of the design and the implementation before settling on the current approach and, while I did find the occasional issue, the underlying idea held up under scrutiny. MIRI finds no undefined behavior while running the tests.

You are invited to review at the code - it is not large - and report any issues you encounter.

Are the run-time checks expensive?

While run-time checks are certainly more expensive than in case of Rc and Option which don't need any, they are still quite cheap.

SendRc::deref() just checks that the SendRc was not disabled (by comparing to a constant) and compares the id of the pinned-to thread fetched with a relaxed atomic load with the current thread. The relaxed atomic load compiles to an ordinary load on Intel, which is as cheap as it gets, and if you're worried, you can hold on to the reference to avoid repeating the checks. (The borrow checker will prevent you from sending the SendRc to another thread while there is an outstanding reference.) SendRc::clone() and SendRc::drop() do the same kind of check.

SendOption::deref() and SendOption::deref_mut() only check that the current thread is the pinned-to thread, the same as in SendRc.

Regarding memory usage, SendRc's heap overhead is two machine words, the same as that of an Rc (but SendRc doesn't support weak references). Additoinally, each individual SendRc is two machine words wide because it has to carry an identity of the pointer. SendOption stores a u64 alongside the underlying option.

License

sendable is distributed under the terms of both the MIT license and the Apache License (Version 2.0). See LICENSE-APACHE and LICENSE-MIT for details. Contributing changes is assumed to signal agreement with these licensing terms.