I do have a feedback on the implementation. The code tries to support older PostgreSQL server versions when declarative partitions were not supported before version 10 (relkind value of 'p'). Those versions will never return any result from the query being built. So I would suggest an early return from the function. The upside would be that the query building would be simpler. I can make patch implementing that suggestion if you want.
This is question - maybe we can support older partitioning based on only inheritance - and the query can be more exact on PostgreSQL 10 and newer.
Please, send any patch. You are welcome.
In my very humble opinion, I would restrict the definition of partitions to declarative partitioning. My justification would be that partitions all use inheritance, but not all inheritance is a partition (how would you handle multiple inheritance).
See patch attached that fails (in a way similar to other features) when connected to servers with version earlier than 10.0.
