On balance, I've found PostgreSQL lack of hints to be a good thing. I can recall a single genuine case where the planner could not make a good plan, and 1000 w

On balance, I've found PostgreSQL lack of hints to be a good thing. I can recall a single genuine case where the planner could not make a good plan, and 1000 where a colleague would have created some unmaintainable time bomb; instead they were driven to understand an actual underlying issue that resulted in a more accurate expression of the query. That one case, as the author says, is PostgreSQL statistics are per-table and do not understand correlation in the data. How different would the implementation be for the hints to express these correlations instead?