sync_engine_drift

Drift persistence adapter for sync_engine. It provides DriftSyncStorage, a SyncStorage implementation for optimistic local entities, tombstones, and durable replica acknowledgements.

Storage model

The adapter stores all synced entity types in a shared sync_entity_table:

• entity_type and id form the primary key;
• payload holds serialized domain JSON;
• vector_clock holds serialized causal metadata;
• deleted is the tombstone flag; and
• last_modified is an integer Unix timestamp.

It also persists replica_acknowledgements and lww_frontier_entries. A causally stale saves and deletes are rejected rather than overwriting a newer durable row.

The entity write path stores one materialized winner per LWW field. Frontier rows are an application-managed acknowledgement-pruning hook; ordinary saves, updates, merges, and deletes do not create them automatically. This keeps the storage model aligned with the MVP contract documented in DESIGN.md, which does not persist multi-way LWW frontiers.

DriftSyncOutbox stores pending operations in the same database. It persists serialized payloads, retry counts, next-attempt timestamps, last errors, and dead letters so queued work survives process restarts. Transport responses are validated before an operation is removed or retried.

Acknowledgement pruning

DriftSyncStorage implements SyncAcknowledgementStorage. Frontier entries are pruned only when every member of the configured knownReplicas roster has acknowledged a clock that causally dominates the individual entry. An empty roster disables pruning: distinct acknowledgement rows alone are not a safe definition of every replica that could still need the data.

The surviving concurrent value is selected by the shared LWWRegister.winningNodeId comparator, so storage pruning and CRDT materialization use the same deterministic tie-break rule.

Drift generation and schema manifests

Enable the generator's Drift schema validation and optional table declaration in the consuming package's build.yaml:

targets:
  $default:
    builders:
      sync_engine_generator|syncable:
        options:
          generate_drift_table: true
          schema_manifest: lib/sync_engine_schema.json

Generated tables are application-side Drift declarations. DriftSyncStorage continues to use the shared sync_entity_table, so generated tables are not registered by SyncDriftDatabase or used for synchronization automatically. Applications that need the declarations must register them in their own Drift database.

For a first build, temporarily set bootstrap_schema: true, then create and review the checked-in baseline with bootstrap_schema. Later additive field changes are reviewed through update_schema; removals and rename-shaped changes fail generation.

dart run sync_engine_generator:bootstrap_schema lib/sync_engine_schema.json Task id,title
dart run sync_engine_generator:update_schema lib/sync_engine_schema.json Task id,title,completed

See DESIGN.md for the MVP migration boundary: validation is additive-name-only, and type changes require an explicit application migration.