Struct DocumentFieldSession 

pub struct DocumentFieldSession { /* private fields */ }

One process serving many observations against one field.

The store and field are opened once (open); every observe then reuses that open field and the derived-cache probe, so manifest/descriptor reads, their content-id BLAKE3 checks, the descriptor parse, and the typed model decodes are paid once rather than per observation.

Implementations

impl DocumentFieldSession

pub fn open( store_dir: &Path, field_hex: &str, opts: SessionOptions, ) -> Result<Self>

Open the store and the field once. Does the manifest read + content_id BLAKE3, descriptor read + Id::of BLAKE3 + Descriptor::parse, and the FieldId hex parse exactly once.

pub fn field_id(&self) -> FieldId

The field this session serves.

pub fn observe( &mut self, req: &ObserveRequest, limits: Limits, ) -> Result<(FieldAnswer, ObserveStats, FieldId)>

One observation. &mut self.store, &self.field, &self.index, and a clone of the Rc-based memo are disjoint field borrows in one call.

pub fn observe_batch( &mut self, reqs: &[ObserveRequest], limits: Limits, ) -> Vec<Result<(FieldAnswer, ObserveStats, FieldId)>>

N observations in one process, in order.

pub fn sync(&self) -> Result<()>

EntropyFS epoch barrier; no-op on the filesystem backend.

pub fn source_exact(&self, limits: Limits) -> Result<Vec<u8>>

The exact original source bytes, materialized from the field’s own exact authority (the descriptor it was built from), not by re-opening the original source file. This is the substrate a fused kernel reads: it is the same byte-authority as materialize --exact, and it is off the observation path (it is never consulted by, and never changes, observe).