Struct Loopback

pub struct Loopback {
    pub(crate) id: AdapterId,
    pub(crate) engine: Arc<Engine>,
    pub(crate) next_sub: AtomicU64,
}

In-process handle onto the engine.

Holds its own Engine so tests can Self::subscribe and Self::publish without going through Adapter::run. Two handles must not share an AdapterId: Self::shutdown drops every subscription under that id.

Fields

id: AdapterId

engine: Arc<Engine>

next_sub: AtomicU64

Implementations

impl Loopback

pub fn new(engine: Arc<Engine>, id: impl Into<AdapterId>) -> Self

Binds this handle to engine under id.

Does not subscribe and does not start Adapter::run. The engine is stored here; run ignores the engine the host passes in.

pub fn engine(&self) -> &Arc<Engine>

The engine this handle was built with.

pub async fn subscribe( &self, route: RouteKey, ) -> Result<Receiver<Envelope>, EngineError>

Subscribes and returns a receiver of matching Envelopes.

Delivery::sub_id is stripped; this handle assigns lb-N ids internally. Dropping the receiver stops the forwarder, but the engine subscription stays until Self::shutdown. Further publishes then count as dropped.

Errors

Returns EngineError::DuplicateSub if an internal id collides, which does not happen on a single handle.

pub async fn publish(&self, envelope: Envelope) -> PublishOutcome

Publishes through the shared engine. Same fan-out and drop rules as Engine::publish.

pub async fn shutdown(&self) -> usize

Drops every subscription this handle owns.

Returns how many were removed. Does not cancel Adapter::run; the host’s cancellation token does that.

Trait Implementations

impl Adapter for Loopback

fn run<'async_trait>( self: Arc<Self>, _engine: Arc<Engine>, shutdown: CancellationToken, ) -> Pin<Box<dyn Future<Output = Result<(), AdapterError>> + Send + 'async_trait>>

where Self: 'async_trait,

Waits for shutdown, then drops this handle’s subscriptions.

There is no socket. Tests that only call Loopback::subscribe and Loopback::publish never need this.

Unlike OWP, STOMP, and DDS, this has no session to retry and nothing in it panics in normal operation — there is no I/O and no protocol state, only a wait on shutdown — so it carries no on_panic/reconnect config. That is deliberate, not a gap: panic supervision on a function that cannot fail would be config with nothing to affect.

Errors

Does not fail. Ok after the token fires.

fn id(&self) -> &AdapterId

Stable id for this adapter instance.