Struct QueuedGate 

pub struct QueuedGate { /* private fields */ }

Per-process “already queued this pass” flags (scheduler.md §4’s literal queued_this_pass bit, lifted out of any one process’s storage into its own gate so sensitivity_match_to_enqueue can dedup across every cell a process is sensitive to, not just one).

The flags are a Vec<AtomicBool> indexed by ProcessId behind an RwLock: both methods take &self and must create a flag on first touch (no registration step pre-populates them), which needs the write lock once per process; every later touch takes only the read lock, and the flag itself is flipped and reset by a genuine atomic swap/store (SC-056’s acceptance: “the AtomicBool must be real, not a Cell<bool>”), so QueuedGate stays Send + Sync with no lock needed around it. (It was a Mutex<HashMap<ProcessId, Arc<AtomicBool>>>, whose hash, lock and Arc clone on every match cost 3-5% of a fan-out-heavy run.)

Implementations

impl QueuedGate

pub fn new() -> Self

pub fn mark_queued_if_not_already(&self, proc: ProcessId) -> bool

The exact if !process.queued_this_pass.swap(true, Ordering::AcqRel) { ... } idiom from scheduler.md §4: atomically marks proc as queued and returns true iff this call is the one that transitioned the flag from unqueued to queued — i.e. iff this call is the one that should push proc onto active_q. A second call for the same proc before the flag is reset returns false (SC-055’s “queued only once”).

pub fn reset_on_dequeue(&self, proc: ProcessId)

Clears proc’s queued flag. Callers invoke this the instant a process is popped off the ready set to run, before resume() is called — never at enqueue time (SC-056) — so a legitimate re-trigger landing in the dequeue-to-re-registration gap is correctly seen as a fresh, unqueued state and re-queues the process rather than being silently swallowed.

Trait Implementations

impl Default for QueuedGate

fn default() -> Self

Returns the “default value” for a type.