Struct RefClosure 

pub struct RefClosure { /* private fields */ }

The ref-table-closure alkblobs::LivenessSource — the first alkgit-owned liveness source (substrate.md §GC; ADR-019 §6). See the module docs for the walk, the bridge, and the invalidation decision.

Implementations

impl RefClosure

pub fn new( front: ObjectFront, rows: Arc<dyn RefRows>, writes: Arc<AtomicU64>, ) -> Arc<Self>

The closure source over the object front, the refs rows seam, and the composition’s shared write counter (bumped by the refs writes — crate::refs::VersionedRefTable on the sqlite assembly).

pub fn register(self: &Arc<Self>, store: &Store) -> Result<Registration, Error>

Register the source with the store — the live-shared seam (Store::register_liveness_source; alkblobs store-api.md §Lifecycle). Registered at the composition’s construction; deregistration/re-registration around a direct delete is the ADR-012 §6.4 embedder correction flow (out of scope here).

The composition carries this source for the store’s lifetime (registration once, at construction — the returned token is deregister-at-most-once and is the composition’s to keep).

pub fn keys_builds(&self) -> u64

How many closure walks have run — the memoization pin (a fingerprint hit consumes no walk; the arbitration consult never triggers one).

Trait Implementations

impl LivenessSource for RefClosure

fn live_keys(&self) -> Vec<Key>

The mark-time consult — the whole live set of the refs closure, bridged through the front’s async-context posture (multi-thread runtime floor). Fingerprint hit = no walk; a failed snapshot consult answers empty (the next mark corrects it). The delete leg is not widened by it: arbitration consults this impl’s trait-method contains, which answers against the same memo the failed mark left untouched — the staged pool-wide candidates are guarded by the stale-epoch answer, so the deletion set is the residual class documented in the module docs, not the snapshot failure’s. The alternative postures (blocking or panicking the sweep’s mark) are worse.

fn contains(&self, key: &Key) -> bool

The delete-time arbitration consult — under the store’s arbitration lock, quick, indexed, never blocking (no bridge, no reads, no recompute):

• an index hit answers exactly (true — the key was live at the memo’s build);
• a cold (never-built) memo answers true — the safe default (unknown is live);
• an index miss after an in-process refs write answers true too — the write counter moved past the build, so the candidate’s coverage state is unknown: cancel it now, let the next mark’s rebuild decide. A ref delete between mark and arbitration lands here the safe way too (kept one more sweep, reclaimed after the rebuilt mark confirms).