Release v0.4.0 · hfmsio/dbxlite

BigQuery onboarding wizard: staged setup with live post-connect preflight, a minimal OAuth scope set, and a paste-a-token auth mode for users who can't run the OAuth flow.
Schema-aware autocomplete: suggestions come from your connected catalog across DuckDB, BigQuery, and Snowflake, resolving CTEs, aliases, quoted identifiers, and multi-level catalogs. Cloud and inline DuckDB file/URL columns load lazily on first reference, with a "Loading columns" placeholder that fills in when the fetch completes.
Cancel control for BigQuery sign-in so a closed or errored Google popup returns immediately instead of waiting on a timeout.
Export cost confirmation: a pre-run dialog surfaces row/scope and potential warehouse cost before a full-result export, so large pulls are never triggered silently.
Full-result export with no row cap: exports always re-run the query for the complete result set. OPFS-streamed Parquet and streamed CSV/JSON, with a buffered fallback when OPFS is unavailable.
Parquet compression setting (default ZSTD).
ESC-to-cancel exports with a confirmation prompt to guard against accidental cancellation of long-running downloads.
Catalog-aware engine auto-detection: switches on definitive single-engine syntax and on table names present in the attached DuckDB catalog.

Decomposed the streaming-query-service god object into individually tested collaborators (connector registry, pagination planner, query executor, row-count estimator, abort registry, file VFS, per-connector lifecycles) behind an unchanged public facade.
Replaced ten UI state-sync polling loops with a connector event surface plus ref-counted remote probes and focus/visibility permission rechecks.
DuckDB attach introspects each schema with a fixed set of batched queries and estimated row counts instead of per-table column and COUNT(*) queries, keeping attach time flat on large catalogs.
xlsx sources are read with all_varchar=true, so a column that sniffs numeric but holds text further down no longer aborts the read.
Autocomplete alias resolution is scoped to the statement under the cursor, so a reused alias in another statement no longer offers the wrong table's columns.

BigQuery / Google OAuth: closing or cancelling the popup now returns a graceful error instead of hanging until the timeout.
Attached DuckDB databases appear in the in-browser explorer again (they were being routed to the server-mode path).
Autocomplete inserts columns as identifiers (bare or double-quoted) rather than single-quoted string literals.
xlsx queries with mixed-type columns no longer fail with a type-conversion error.
BigQuery project listing no longer strands users on a disabled API (projects.list primary, Cloud Resource Manager fallback).
BigQuery dot-completion (alias. / table.) resolves columns on demand.
DuckDB Parquet export reports the real exported row count instead of 0.

Synced ARCHITECTURE, README, CHANGELOG, and test coverage to the current codebase.

Leaner dbxlite-ui npm package: roughly 3x smaller to install (10 MB vs 30 MB tarball) after dropping the DuckDB WASM bundles the app never loads.

Navigation Menu