DataForge

A neo-brutalist database IDE inside VS Code. Phase 1 covers the full connection-to-results loop; Phase 2 adds a data editor and CSV export; Phase 3 adds completion, formatting, history, EXPLAIN and explicit transactions; Phase 4 adds a MySQL adapter behind the same driver ports:

• Connections — store PostgreSQL connection profiles, test them, connect and disconnect from the DataForge sidebar or the workbench.
• Schema browser — lazily expand a connection into schemas, then tables and views, then columns with types, primary keys, nullability and defaults.
• SQL workbench — run a statement, see the command tag, row count, duration and notices in a dense result grid. Results are capped (default 500 rows) so a stray select * cannot flood the extension host.
• Data editor — open a table from the tree, edit cells, add and mark rows for deletion, then commit or roll back the whole batch. Paging and sorting run in PostgreSQL; writes are parameterised and sent as one transaction.
• Optimistic locking — every row carries its xmin; if another session changed it, the commit fails with the conflicting primary keys and writes nothing.
• CSV export — stream a whole relation to a file you pick; the webview never touches the filesystem.
• Safety — passwords never leave the extension host: they live in SecretStorage, never in settings, logs or webview messages. Mark a profile read-only and only select/with/show/explain statements are sent, and the data editor refuses to commit.
• Cancellation — every query runs with a cancel button, a statement timeout and a hard row cap. Commits and CSV exports register an abort controller too: cancelling rolls the transaction back and deletes the partial file.
• Completion and formatting — typing in a .sql editor completes keywords, schemas, tables and columns from the host's schema cache. DataForge: Format SQL formats the selection (or the whole document) through a WorkspaceEdit, so undo keeps working.
• History and timing — the last 200 statements live in globalState; DataForge: Query History reloads one into the editor. The status bar shows the duration of the last run and slow queries (dataforge.slowQueryMs) are logged.
• EXPLAIN — one read-only statement at a time, rendered as the JSON plan PostgreSQL returns.
• Transactions — Begin transaction opens a dedicated backend; later runs go through it until you commit or roll back, and the host rolls the session back on panel close, disconnect or deactivate.

Requirements

• VS Code 1.115 or newer.
• A PostgreSQL server reachable from the machine running VS Code (MySQL connections only need queries and metadata).

Getting started

Open the DataForge activity bar view, press New, fill in host, port, database and user, use Test to verify, then Create. Expand the saved connection to browse schemas and tables, or press Run (Ctrl/⌘ + Enter) in the workbench to execute SQL. Select part of the editor before pressing Run to execute only the selection.

Settings

Development

npm install
npm run build        # extension bundle + webview assets
npm run type-check   # both workspaces
npm run test:unit    # host and webview unit tests

Press F5 ("Run DataForge Extension") to start an Extension Development Host. The workbench HTML is written by the extension host so the CSP nonce stays under its control; the webview bundle is built into extension/dist/webview/assets/.

Architecture

• shared/protocol.ts — dependency-free message contract shared by both sides.
• extension/src/domain — profiles, errors, driver ports. No vscode, no pg.
• extension/src/application — connection, query, schema, data-editing, export, history, format and transaction-session services plus ports.
• extension/src/infrastructure — PostgreSQL and MySQL drivers, SQL builder, CSV writer, SecretStorage, Memento stores.
• extension/src/presentation — tree view, webview panel, message router, completion provider, diagnostics, logging.
• webview/src — Vue 3 + Pinia + Tailwind workbench; it only ever talks to the host through HostBridge, never to a database.