Skip to content

NEXT_TASK.md

Latest commit

 

History

History
53 lines (42 loc) · 3.14 KB

NEXT_TASK.md

File metadata and controls

53 lines (42 loc) · 3.14 KB
Raw

CoCreateDB — Remaining Work

Implemented: server-owned CRUD, supplied/generated $update patches, resolved client replies, peer patch application, index CRUD naming, disk, separate local and peer distribution, connection inventory exchange, accumulated API errors, flat UUID record/key files, permanent root.json, and peer file identity transfer.

Index layout migration

  • Migrate existing datasets to flat UUID record files and root.json before using an existing dataset. The code never automatically migrates live files; dataset backup/bootstrap is a separate operation.
  • Coordinate peer versions: old nested partition paths are not the new wire format.

Persistence workflow

  • Wire the agreed persistence envelope and persistence.create/read/update/delete operations to the configured workflow. The retained hook still uses storage and external workflow writes are disabled.
  • Add file-level fallback after memory, disk, and known peers. Accept parsed objects or serialized JSON file values and use the existing acquisition paths; cache locally when disk is enabled.
  • Keep response timing and durability explicit. Do not silently require external acknowledgment before notifying users; operation ordering is pending discussion.
  • Preserve disabled persistence hooks and legacy complete-object recovery until their replacement is implemented and removal is explicitly decided.

Follow-up decisions

  • Cross-node operation ordering, conflict resolution, deletion policy, and incremental operation deduplication. No distributed LWW/CRDT policy is implemented.
  • Client IndexedDB generation/application of $update and client acknowledgment.
  • Index update semantics beyond the current insertion alias, tier-level index addressing, and optional custom definitions/uniqueness.
  • Batching large inventory exchanges and handling legacy local ownership when the server ID had already changed before stable local markers were introduced.
  • Worker/client recipient removal and reconstruction after disconnect/restart.
  • Multi-file rollback, write-behind failure handling, and crash recovery. Existing baseline failures are not evidence of durable transactions.

Known checks in test/write-behind.test.js still need attention: the failed-indexing rollback test, the sequential-persistence test (old public-ID filename and snapshot assumptions), the deletion-marker test (old nested path), and workflow acknowledgment while the external hook remains disabled. These expectations need explicit reconciliation with the current contract.

The entity-holder directory is currently used for metadata and exact-object acquisition. Keep it; replacing it with index-trail lookup requires a separate architecture decision. Inventory exchange does not enumerate document payloads.

Constraints

Read AGENTS.md and ARCHITECTURE.md before changing the engine. Preserve scoped identities, complete acquired documents, shared query/update semantics, reference ownership, and unrelated workspace changes. Use targeted peer reads with timeouts. Do not persist peer inventories or remove deferred features merely for having no active callers. Report what was tested and what remains unsupported.