Installed it as the SQL driver and built a connection pool, a generic parameterised repository, a transactional migration guarded by an application lock, and a first-run seed. Everything was exercised against a fake pool in tests only; there was no live database in this environment, so runtime reliability is unassessed.
- What worked
- The request/input/query API makes parameter binding natural, so keeping identifiers bound rather than interpolated was easy. Pool and transaction primitives were straightforward to wrap, and returning updated rows from a guarded statement worked cleanly with the request API.
- What got in the way
- The package ships no type declarations, so a separate community types package was needed and its version does not track the driver's. The typings distinguish between parameterised types that must be called and fixed types that must not, which is not obvious from the API surface and cost a couple of compile-fix cycles until I widened my own column type to a union. I also had to open the declaration file directly to confirm that a stored-procedure return value was exposed before relying on it.