feat: run on drizzle databases and their transactions - #23

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

ℹ️ No critical issues. Two small suggestions: one inline, one below.

Reviewed changes

This review covers how Reylo now runs on Drizzle databases, transactions and savepoints, plus the postgres.js JSON change that came with it. I checked underneath() against the published drizzle-orm@0.45.4 source. Only libsql/session.js keeps a tx field. Every other session, PlanetScale's and TiDB's included, puts the transaction's connection on client, so session.tx ?? session.client picks the right one everywhere.

• Unwrapping: underneath() swaps a Drizzle database, transaction or savepoint for the driver client its session holds. Both run and runSync call it first, before they pick a path.
• Types: DrizzleDatabase<C> is matched by $client. Each dialect's transaction is matched by a method only it has. runSync takes Drizzle over a synchronous SQLite driver.
• postgres.js JSON: every client with typed() (all but Bun.sql) now gets a JSON parameter as typed(JSON.stringify(v), 705). Drizzle replaces the client's json/jsonb serializers with ones that pass the value through unchanged.
• Error hints: the old "pass db.$client" hint is gone. New hints tell Kysely and Prisma users which client to pass, including from inside a transaction.
• Lab and docs: five Drizzle driver-matrix jobs, Drizzle type-check cases, and Drizzle tabs on the Supported drivers page (checked by twoslash). The dialect pages and the README mention Drizzle.

ℹ️ The transaction types accept a Drizzle transaction from any driver of the dialect

DrizzleMysqlTransaction and the other transaction types match by dialect, not by driver. So .run(tx) also type-checks for a transaction from a Drizzle driver Reylo doesn't run on, even though $client rejects that driver's database. Most such drivers fail loudly at run time. A drizzle-orm/tidb-serverless transaction fails silently: its session.client is TiDB's native transaction, which has execute() but no query(), so the call goes to planetscale(). TiDB returns a plain array of rows, and planetscale() returns [] whenever the result has no .rows. This affects only a niche driver, so you may decide it's not worth handling.

# A Drizzle tidb-serverless transaction type-checks and returns no rows

## Affected sites
- src/execute.ts:118 — `DrizzleMysqlTransaction` matches every Drizzle `MySqlTransaction`, `TiDBServerlessTransaction` included.
- src/decode-mysql.ts:284 — `if (!Array.isArray(result?.rows)) return [];` turns TiDB's row array into no rows.

## Required outcome
- A client whose `execute()` answers in a shape `planetscale()` doesn't recognize should fail with an error, not return no rows. Alternatively, document that only the Drizzle drivers listed in the docs are supported.

## Open questions for the human
- Is a run-time refusal enough, or should the types try to narrow the transaction to its driver (probably not feasible structurally)?

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using claude-opus-5-5 ｜ 𝕏}

✅ No new issues found. Both notes from the last review are addressed.

Reviewed changes

This run covers what changed since the last review at 814a2d9. The branch was rebased onto main, and the new work is in the postgres.js JSON tests and in the MySQL planetscale() path. tests/execute.test.ts passes locally (42 of 42).

• postgres.js JSON coverage: added a { unsafe, typed } fake to the array/JSON test and to the jsonQuery test. They assert that each JSON value, jsonb[] included, arrives as { value: <JSON text>, type: 705 } and that null is still bound as null. Updated the comment above json() and the test title to match.
• planetscale() refuses results without rows: an execute() answer with no rows array now throws a TypeError that describes what came back. Before, it was read as no rows. So a drizzle-orm/tidb-serverless transaction now fails loudly instead of returning [], and a new test covers this. Existing MySQL fakes that meant mysql2 now have query beside execute, so they still take the mysql2 path.

  ^{｜ View workflow run ｜ Using claude-opus-5-5 ｜ 𝕏}