dart_db_server 0.1.1

dart_db_server: ^0.1.1 copied to clipboard

• Docs-only release. Adds a doc/ folder with real-world vector-database recipes: RAG semantic search, multi-tenant search, hybrid vector + BM25, recommendations, duplicate detection, index selection, and production operations.
• README: index of the new recipes.

Initial pub.flutter-io.cn release.

Highlights #

• Pure-Dart SQL engine — SQLite-inspired surface: DDL, DML, joins, subqueries, CTEs, window functions, aggregate ORDER BY / FILTER, RETURNING, upsert, common table expressions, and transactions with savepoints.
• Three storage backends — in-memory (JSON-persistable), out-of-core paged (USING paged), and SQLite on-disk format (.sqlite files round-trip through the real SQLite format).
• Vector database built into the SQL surface:
  • Six index kinds: flat, hnsw, ivf, lsh, pq, ivfpq.
  • Four metrics: l2, l2sq, inner_product, cosine.
  • DDL: BLOB VECTOR(dim=..., kind=..., metric=..., ...) inline attribute or CREATE VIRTUAL TABLE ... USING vector_index(...).
  • Table-valued functions: vec_search, vec_search_batch, vec_search_filtered, vec_search_filtered_batch, vec_range_search, vec_range_search_batch, vec_hybrid_search, vec_hybrid_search_batch, vec_search_join, vec_batch_insert, vec_import_csv.
  • Payload-filter pruning: declare filter_cols='tenant,kind' on the index and the query engine intersects payload buckets before touching the vector index.
  • Fast-path planner: SELECT ... ORDER BY VEC_L2(col, const) LIMIT k is served by the built index automatically; EXPLAIN QUERY PLAN surfaces the choice.
  • Admin PRAGMAs: vector_index_list, vector_index_stats, vector_index_rebuild[_all], vector_index_warm[_all], vector_verify[_all], vector_analyze, fts5_warm.
• FTS5-lite — BM25 ranking, MATCH operator, and hybrid retrieval via Reciprocal Rank Fusion.
• TCP server — JSON line protocol on any port + interactive REPL.
• MySQL wire compatibility — a subset of the MySQL 8 auth / query protocol so ORMs like mysql_client can talk to the server unchanged.
• 1500+ tests covering every surface above.

See example/vector_semantic_search.dart for a runnable end-to-end semantic-search walkthrough.

• Initial version.