Struct BackgroundJobManager 

pub struct BackgroundJobManager { /* private fields */ }

Shared manager for background jobs.

Implementations

impl BackgroundJobManager

pub fn new() -> Self

pub fn completed_notify(&self) -> Arc<Notify>

Return a clone of the completed-notify handle.

Used by the TUI loop arbiter to select! on job completion. When a background job enters a terminal state (Completed, Failed, TimedOut), notify_one() is called. notify_one() (rather than notify_waiters()) stores a permit so a completion that happens while the arbiter is mid-turn (not yet polling) is not lost — the next notified() poll consumes the stored permit.

pub fn insert(&mut self, job: Job) -> String

Insert a new job and return its ID.

pub fn get_state(&self, id: &str) -> Option<JobState>

Get a reference to a job’s state.

pub fn clear(&mut self)

Remove all jobs immediately.

pub fn set_watch(&mut self, path: PathBuf, offset: u64)

Register a file for mid-run event wakes. offset is the byte offset to start reading from (typically the file size at registration time, so only bytes appended afterwards wake the agent). Replaces any prior watch.

pub fn clear_watch(&mut self)

Clear any registered file watch.

pub fn set_loop_control(&mut self, ctrl: LoopControl)

Set an agent-requested loop control (e.g. Stop). Replaces any prior pending control. Read and consumed by the loop arbiter between turns.

pub fn take_loop_control(&mut self) -> Option<LoopControl>

Take (consume) any pending loop-control request. Returns None when nothing is pending.

pub fn poll_watch(&mut self) -> Option<String>

Poll the registered watch file for new bytes past the stored offset. On success, advances the offset and returns the new content (capped at WATCH_CHUNK_BYTES per call; remaining bytes surface on the next poll). Returns None when no watch is set, the file is gone, or no new bytes are available. Handles log rotation (file shrank) by resetting to the start.

pub fn take_completed(&mut self) -> Option<(String, String)>

Remove and return the first completed job, if any.

Returns Some((job_id, output_string)) where output_string includes the exit code and captured stdout/stderr. Completed jobs are removed from the tracked set.

If a terminal job remains after this removal, the completion notify is re-armed. update() uses notify_one(), which stores a single permit: when ≥2 jobs reach a terminal state before a waiter polls, the second notify_one() is a no-op, so a waiter that consumes the one permit would never wake for the remaining job. Re-arming here guarantees every completed job has a scheduled wake until it is drained (goal-379).

Trait Implementations

impl Default for BackgroundJobManager

fn default() -> Self

Returns the “default value” for a type.