RuVector Files

ruvector-wasm 2.3.1 / @ruvector/wasm 0.2.1 — IndexedDB persistence fix

saveToIndexedDB() in 0.2.0 reported success but saved nothing, and loadFromIndexedDB was an unimplemented stub. Both now work.

Fixed

• Save is commit-gated. It resolves only after the IndexedDB transaction commits; errors and aborts (including quota exceeded) reject. The database is stored as one versioned snapshot record, so a save is all-or-nothing.
• VectorDB.loadFromIndexedDB(name) is a static async that returns a VectorDB. It rejects when nothing is saved under that name, or the payload is corrupt, an unsupported format, has mismatched dimensions or has duplicate ids. A failed load no longer leaves an empty database behind.
• Naming. New setDbName(name) and dbName getter. Names must be 1-128 chars of [A-Za-z0-9_.-]. Without setDbName, an instance saves under a random name, so choose one if you want to reload in a later session.
• The snapshot stores vectors and metadata; the HNSW graph is rebuilt on load.

Design and contract: ADR-200.

Install

• crates.io: ruvector-wasm = "2.3.1"
• npm: npm i @ruvector/wasm@0.2.1

Verification

• 16 wasm-bindgen-test browser tests pass in headless Chrome, including round-trip restore of vectors and search, missing database, invalid names, corrupt JSON, unsupported format and duplicate ids.
• A real page saved a database, was reloaded, and loaded it back intact.
• A mutation test (removing the write) turns the round-trip test red.

Known limits

• No test forces a transaction abort or quota failure; that path is covered by code review only.
• A large database is stored as a single JSON string and the index is rebuilt on load.
• The npm/wasm TypeScript wrapper was updated to the new API but is a separate package and was not published.