Struct SyncerState 

pub struct SyncerState { /* private fields */ }

Implementations

impl SyncerState

pub fn open(dir: PathBuf) -> LeanResult<SyncerState>

pub fn marker_present(&self) -> bool

pub fn sync_tree(&self) -> LeanResult<()>

Flush the filesystem holding the tree (the state dir lives in it) so everything materialised so far is on stable storage.

pub fn write_drained(&self, seq: Option<u64>, acks: usize) -> LeanResult<()>

The drain’s attestation. seq is the manifest the drain left installed, when it knows it.

pub fn clear_drained(&self) -> LeanResult<()>

A fresh incarnation owes its own attestation; a stale one from an earlier life of this tree must not vouch for it.

pub fn drained_path(&self) -> PathBuf

pub fn write_marker(&self) -> LeanResult<()>

Written LAST at checkout: the agent-start gate.

pub fn load_baseline(&self) -> LeanResult<Baseline>

pub fn save_baseline(&self, b: &Baseline) -> LeanResult<()>

COMPACT: the baseline is the other O(files) document, rewritten on the cadence tick and re-parsed several times per barrier. The agent-facing O(1) files (gauges, capabilities, acks) stay pretty — those exist to be cat-ed.

pub fn load_scope(&self) -> LeanResult<Option<Vec<String>>>

What this workspace was admitted to hold, or None for an unscoped (whole-manifest) checkout.

fs::read directly, with NO exists() pre-check: exists() answers false for EACCES and EIO alike, and an unreadable scope answered as “unscoped” is the same class of bug as an unreadable path answered as “deleted” (barrier.rs, 14b3637c). Only NotFound means absent.

pub fn save_scope(&self, scope: Option<&[String]>) -> LeanResult<()>

Some writes the admitted set; None REMOVES the document. Removing matters: a scoped checkout that crashed before its marker leaves a scope behind, and the whole-tree checkout that replaces it must not inherit a claim it did not make.

pub fn load_scope_intent(&self) -> LeanResult<Option<ScopeIntent>>

Read a rescope in flight. None = none in flight.

pub fn save_scope_intent(&self, intent: &ScopeIntent) -> LeanResult<()>

pub fn clear_scope_intent(&self) -> LeanResult<()>

Cleared LAST, after the new scope is durable. Clearing it first would turn a crash into a workspace whose scope says one thing and whose held set says another, with nothing left to replay.

pub fn load_incarnation(&self) -> LeanResult<Option<Incarnation>>

pub fn save_incarnation(&self, i: &Incarnation) -> LeanResult<()>

pub fn load_untracked_sweep_at(&self) -> LeanResult<u64>

Unix seconds of this writer’s last untracked-upload sweep; 0 when it has never swept.

pub fn save_untracked_sweep_at(&self, at: u64) -> LeanResult<()>

pub fn load_orphan_sweep_at(&self) -> LeanResult<u64>

The orphan sweep’s clock (untracked.rs, sweep_orphans_if_due): 0 = never swept, and the first commit starts it.

pub fn save_orphan_sweep_at(&self, at: u64) -> LeanResult<()>

pub fn load_retire_cache(&self) -> BTreeMap<String, RetireLog>

The retire logs read so far, by key. A cache: unreadable reads as empty (the logs are fetched again), never as an error.

pub fn save_retire_cache( &self, c: &BTreeMap<String, RetireLog>, ) -> LeanResult<()>

pub fn load_conformance(&self) -> LeanResult<Option<StoreConformance>>

A verdict that cannot be parsed reads as ABSENT, never as a pass: the probe runs again. The expensive outcome of a corrupt cache is ten requests; the cheap one would be running the collector on a store nobody asked.

pub fn save_conformance(&self, c: &StoreConformance) -> LeanResult<()>

pub fn load_intent(&self) -> LeanResult<IntentJournal>

pub fn save_intent(&self, j: &IntentJournal) -> LeanResult<()>

pub fn clear_intent_keys(&self) -> LeanResult<()>

The barrier is over: no flush in flight. The carrier stays.

pub fn append_conflict(&self, c: &ConflictRecord) -> LeanResult<()>

pub fn conflicts_dropped(&self) -> LeanResult<u64>

Records lost to rotation. A caller rendering conflicts owes the user this number: a truncated list that does not say it was truncated reads as a complete one.

pub fn load_conflicts(&self) -> LeanResult<Vec<ConflictRecord>>

The rotated generation first, then the live one: oldest to newest, unbroken across a single rotation.