++ docs/guides/migration.md
- All write and search operations now validate embedding length against the table
dimension. Existing databases created with
create_tableare supported automatically, but manual schemas must follow thefloat[dim]declaration used bysqlite-vec. - Action: ensure any custom tooling or fixtures produce embeddings with the expected dimension before calling client methods.
import_from_jsonandimport_from_csvhonourskip_duplicates, skipping records whose rowids already exist.- Importers now require embeddings to be present; CSV sources exported without the
embeddingcolumn raise a descriptive error. - Action: export backups with
include_embeddings=Trueif you intend to re-import them.
- New
backup()andrestore()helpers wrap JSONL/CSV workflows and log the format being used. Prefer these helpers for consistent backup scripts.
- The CI pipeline now uploads
coverage.xmlas an artifact. Configure downstream tooling (Codecov, Sonar, etc.) to consume the artifact if you need external reporting.
- Always run
pytest --cov=sqlite_vec_client --cov-report=xmlbefore publishing. - Keep
requirements-dev.txtup-to-date locally to build the documentation site withmkdocs serve.