Struct SynQueue 

pub struct SynQueue<T> { /* private fields */ }

Fast internally syncrhonized MPMC queue.

Principle

The basic problem with MPMC queues on arrays is synchronizing access to data with shifting of head/tail pointers. To address this, a general solution (you can find in crossbeam-queue and others) is to add one atomic bit per element, which serves as a barrier for advancing the queue pointers.

SynQueue tries a different approach. There is no extra bits. Instead, we are keeping 2 representations of the queue state: wide and narrow. Pushing advances wide.head, writes the data, and then makes narrow.head to catch up. Popping advances narrow.tail, reads the data, and then makes wide.tail to catch up. Every operation is thus sequence of CAS loop, data operation, another CAS loop.

Internal invariants.

Considering an infinite sequence (without wraparounds): wide.tail <= narrow.tail <= narrow.head <= wide.head

Implementations

impl<T> SynQueue<T>

pub fn new(capacity: usize) -> Self

pub fn push(&self, value: T) -> Result<(), T>

pub fn pop(&self) -> Option<T>

Trait Implementations

impl<T> Drop for SynQueue<T>

fn drop(&mut self)

Executes the destructor for this type.

impl<T> Sync for SynQueue<T>