Synchronous and Asynchronous Synchronization Primitives

Word-sized low-level synchronization primitives providing both asynchronous and synchronous interfaces.

Features

• No heap allocation.
• No hidden global variables.
• Provides both asynchronous and synchronous interfaces.
• lock_api support: features = ["lock_api"].
• Loom support: features = ["loom"].

Lock

saa::Lock is a low-level shared-exclusive lock providing both asynchronous and synchronous interfaces. Synchronous locking methods such as lock_sync and share_sync can be used simultaneously with their asynchronous counterparts lock_async and share_async. saa::Lock implements an allocation-free fair wait queue shared between both synchronous and asynchronous methods.

use saa::{Lock, TryLockError};

// At most `62` concurrent shared owners are allowed.
assert_eq!(Lock::MAX_SHARED_OWNERS, 62);

let lock = Lock::default();

assert!(lock.try_lock().is_ok());
assert_eq!(lock.try_lock(), Err(TryLockError::WouldBlock));
assert_eq!(lock.try_share(), Err(TryLockError::WouldBlock));

assert!(!lock.release_share());
assert!(lock.release_lock());

assert!(lock.lock_sync());

// `Lock` can be poisoned.
assert!(lock.poison_lock());
assert!(!lock.lock_sync());
assert!(lock.clear_poison());

async {
    assert!(lock.share_async().await);
    assert!(lock.release_share());
    assert!(lock.lock_async().await);
    assert!(lock.release_lock());
};

lock_api support

The lock_api feature is automatically disabled when the loom feature is enabled since loom atomic types cannot be instantiated in const contexts.

#[cfg(all(feature = "lock_api", not(feature = "loom")))]
use saa::{Mutex, RwLock, lock_async, read_async, write_async};

#[cfg(all(feature = "lock_api", not(feature = "loom")))]
fn example() {
    let mutex: Mutex<usize> = Mutex::new(0);
    let rwlock: RwLock<usize> = RwLock::new(0);

    let mut mutex_guard = mutex.lock();
    assert_eq!(*mutex_guard, 0);
    *mutex_guard += 1;
    assert_eq!(*mutex_guard, 1);
    drop(mutex_guard);

    let mut write_guard = rwlock.write();
    assert_eq!(*write_guard, 0);
    *write_guard += 1;
    drop(write_guard);

    let read_guard = rwlock.read();
    assert_eq!(*read_guard, 1);
    drop(read_guard);

    async {
        let mutex_guard = lock_async(&mutex).await;
        assert_eq!(*mutex_guard, 1);
        drop(mutex_guard);

        let mut write_guard = write_async(&rwlock).await;
        *write_guard += 1;
        drop(write_guard);

        let reader_guard = read_async(&rwlock).await;
        assert_eq!(*reader_guard, 2);
        drop(reader_guard);
    };
}

Barrier

saa::Barrier is a synchronization primitive that enables multiple tasks to start execution at the same time.

use std::sync::Arc;
use std::thread;

use saa::Barrier;

// At most `63` concurrent tasks/threads can be synchronized.
assert_eq!(Barrier::MAX_TASKS, 63);

let barrier = Arc::new(Barrier::with_count(8));

let mut threads = Vec::new();

for _ in 0..8 {
    let barrier = barrier.clone();
    threads.push(thread::spawn(move || {
        for _ in 0..4 {
            barrier.wait_sync();
        }
    }));
}

for thread in threads {
    thread.join().unwrap();
}

Semaphore

saa::Semaphore is a synchronization primitive that allows a fixed number of tasks or threads to concurrently access a resource.

use saa::Semaphore;

// At most `63` concurrent tasks/threads can be synchronized.
assert_eq!(Semaphore::MAX_PERMITS, 63);

let semaphore = Semaphore::default();

semaphore.acquire_many_sync(Semaphore::MAX_PERMITS - 1);

assert!(semaphore.try_acquire());
assert!(!semaphore.try_acquire());

assert!(semaphore.release());
assert!(!semaphore.release_many(Semaphore::MAX_PERMITS));
assert!(semaphore.release_many(Semaphore::MAX_PERMITS - 1));

async {
    semaphore.acquire_async().await;
    assert!(semaphore.release());
};

Gate

saa::Gate is an unbounded barrier that can be opened or sealed manually as needed.

use std::sync::Arc;
use std::thread;

use saa::Gate;
use saa::gate::State;

let gate = Arc::new(Gate::default());

let mut threads = Vec::new();

for _ in 0..4 {
    let gate = gate.clone();
    threads.push(thread::spawn(move || {
        assert_eq!(gate.enter_sync(), Ok(State::Controlled));
    }));
}

let mut count = 0;
while count != 4 {
    if let Ok(n) = gate.permit() {
        count += n;
    }
}

for thread in threads {
    thread.join().unwrap();
}

Pager

saa::Pager enables waiting for a resource to become available from anywhere in the program.

use std::pin::pin;

use saa::{Gate, Pager};
use saa::gate::State;

let gate = Gate::default();

let mut pinned_pager = pin!(Pager::default());

assert!(gate.register_pager(&mut pinned_pager, true));
assert_eq!(gate.open().1, 1);

assert_eq!(pinned_pager.poll_sync(), Ok(State::Open));

Notes

Using synchronous methods in an asynchronous context may lead to deadlocks. Consider a scenario where an asynchronous runtime uses two threads to execute three tasks.

• ThreadId(0): task-0: share-waiting / pending || task-1: "synchronous"-lock-waiting.
• ThreadId(1): task-2: release-lock / ready: wake-up task-0 -> task-2: lock-waiting / pending.

In this example, task-0 has logically acquired a shared lock after task-2 releases it; however, it may remain in the task queue indefinitely depending on the task scheduling policy.

Changelog