So it sounds like the issue is with Postgres reporting the precision of some column types in a strange way. I have no real insight into what 35 means here, but probably we can hardcode a special case into the postgres dialect.