feat(database): layerTest, a Database with no server for unit tests - #93

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Merge-readiness pass (71e82cf):

• Merged main into the branch (it was based on c5fa679). The merge had no textual conflicts.
• Fixed the build on Effect 4.0.6. SqlClient.make now requires a driver name, so the fake client passes driver: "effect-orm-test". Without it the typecheck failed.
• Added tests for the dialects that are new on main. sqliteDialect sends ? placeholders, decodes integer booleans and records BEGIN/SELECT/COMMIT. d1Dialect fails a transaction with TransactionUnsupported and sends nothing.
• Added a test for afterCommit (feat(database): afterCommit runs work after the outermost COMMIT #100): Db.afterCommit hooks run after the recorded COMMIT and never after a ROLLBACK.
• Updated the TestDatabaseOptions.dialect doc and docs/testing.md to list all four dialects and mention afterCommit and D1.

Verified locally:

• bun run build and bun run typecheck are clean.
• bun run test against live ClickHouse 26.8 with PGlite and node:sqlite, including fuzz at FUZZ_SCALE=0.3, passed 74 files and 1441 tests. The doc citation, export catalog and doc example checks also pass, with 27 doc example files including the new one.
• bun run test:package passes the tarball and CommonJS consumer checks.

CI run 41472238736 started without needing approval. If it fails only on the known fuzz flakes, they are fixed separately in #131: ClickHouse/Postgres division by zero in generated fixtures …, or compiled row codec retains generated values.

Maple review

🟡 Confidence 6/10 · needs attention. Held at 6 because a correctness warning is open.
_{1 warning −1 · tests partial −1 · risk medium}

Adds Db.layerTest, a Database with no server for unit tests: a fake SqlClient records every statement and answers with queued rows or a driver SqlError, so compile, decode, tenant scope and transactions run unchanged. The mechanism is sound and covered for Postgres, ClickHouse, SQLite and D1.

• layerTest provides Database | TestDatabase over a fake SqlClient connection
• TestDatabase records statements and queues rows or driver SqlErrors
• TestReply, TestStatement and the layer are exported from /database

Findings

Skip the queue in `answer` for the transaction-settings statement (SQL starting with `SET TRANSACTION`), the way `control` already does for BEGIN and COMMIT.

Findings from an automated review of commit 71e82cfebd2dff94a936d495c46af349beecb035. Verify each one against the current code before changing anything, fix only those that still apply, and keep each fix to the lines it names.

---

F1 · Warning · correctness · src/database/test-layer.ts:87-88
`SET TRANSACTION` takes the reply queued for the transaction body's query
With `postgresDialect`, `Db.transaction(body, { isolationLevel })` first runs `SET TRANSACTION` through `Database.execute` (`src/database/database.ts:392`), which is the prepared path, so it pops the next queued reply. The body's query then gets `respond` or no rows, and the test sees `[]` instead of what it queued.
Suggested fix: Skip the queue in `answer` for the transaction-settings statement (SQL starting with `SET TRANSACTION`), the way `control` already does for BEGIN and COMMIT.

_{71e82cf · What was checked, coverage and telemetry details are on the Maple / review check. Reply "won't fix" to dismiss a finding, or mention @maple-review-bot to ask about one.}