Struct ChannelSupervisor 

pub struct ChannelSupervisor { /* private fields */ }

Runtime supervisor for the approval-transport channels. Held as an Arc on crate::session::ServerState (lazy-initialized at boot by spawn_channel_pollers) so the host-gated messaging.config.set handler can reach it to spawn a channel’s watcher the instant the user enables it — no daemon/app restart (U1).

Implementations

impl ChannelSupervisor

pub fn new( host: Arc<HostState>, cancel_tx: Sender<bool>, spawn: Box<dyn Fn(ChannelId, &Arc<HostState>, &Arc<Mutex<HashMap<ChannelId, ChannelLiveness>>>, Receiver<bool>) -> Result<(), String> + Sync + Send>, ) -> ChannelSupervisor

Build a supervisor over the given host, cancel signal, and spawn closure. The boot path (spawn_channel_pollers) constructs it, records the channels it already spawned via Self::mark_spawned, and stores it on ServerState.

pub fn liveness(&self) -> Arc<Mutex<HashMap<ChannelId, ChannelLiveness>>>

The shared liveness map (so the boot path can hand the SAME Arc to the adapters it spawns, and messaging.status can read it).

pub fn cancel_all(&self)

Stop every spawned adapter loop (boot + runtime) by flipping the shared cancel signal. Called by the daemon shutdown path. Idempotent.

pub fn mark_spawned(&self, channel: ChannelId)

Record that channel’s watcher is already running (called by the boot path for each channel it spawned at startup). Idempotent.

pub fn is_spawned(&self, channel: ChannelId) -> bool

Whether channel’s watcher is currently spawned. Read by messaging.status for the watcher_running field (U2).

pub fn ensure_spawned(&self, channel: ChannelId) -> Result<(), String>

Ensure channel’s watcher is spawned (U1 — the invisible-restart fix). Idempotent: if the channel is already live, this is a no-op and returns Ok(()). Otherwise it calls the spawn closure with a fresh cancel receiver, records the channel as live on success, and returns the closure’s result. A spawn error leaves the channel NOT marked live (so a later retry can re-attempt).

Called from messaging.config.set after a successful off→on transition. Disable does NOT abort the loop (KTD1) — poll_once/observe_and_notify already gate on the enabled flag, so a disabled channel does zero work per tick; aborting would only add lifecycle complexity.

pub fn record_send_success(&self, channel: ChannelId)

Record a successful send for channel into the shared liveness (U3). Stamps now, marks ok, clears any prior error.

pub fn record_send_failure(&self, channel: ChannelId, reason: impl Into<String>)

Record a failed send for channel (hard error OR soft sent:false) into the shared liveness (U3). Stamps now, marks not-ok, stores the reason.

pub fn liveness_snapshot(&self, channel: ChannelId) -> ChannelLiveness

Snapshot channel’s liveness (default if no send recorded yet). Read by messaging.status.