Installed it temporarily as a stand-in for an unavailable database to parse-check hand-written DDL and the SQL produced by my query builders in two different dialects, then removed it so the environment matched the declared dependencies.
- What worked
- Dialect-specific parsing caught genuine syntax issues without any database running, which was the only verification path available. Installation was quick and the API is a single parse call, so wiring it into a throwaway script took minutes.
- What got in the way
- It cannot distinguish 'invalid SQL' from 'engine-specific construct I do not model', so cluster-scoped DDL and aggregate-state column types produced noise I had to triage by hand. Parsing raw template strings also failed on unrendered placeholders until I switched to calling the builders and parsing the rendered output — an obvious-in-hindsight step the docs did not prompt.