On Sat, 21 Mar 2026 at 15:47, Richard Guo <[email protected]> wrote: > Currently, the planner can remove useless left joins if the join > condition cannot match more than one RHS row, and the RHS rel is not > referenced above the join. I'd like to propose a similar optimization > for inner joins.
I tried this many years ago and it was pretty much a dead end with how the current foreign key implementation deferring the cascade of the foreign key until the end of the query. There's plenty of discussion. See [1] and [2]. In particular, read and follow along from what Heikki mentions in [3]. You should read all of that and understand it to prevent prompting the same discussions all over again. It doesn't look like your patch has anything to protect against any of the issues mentioned in [1], so I assume you weren't aware of that work. David [1] https://www.postgresql.org/message-id/flat/CAApHDvpDXXvKE%2B%3Dug1kA--nKfa%3DbjrjvK8Gp9G8UYwv6nHckVg%40mail.gmail.com#2f33cb1e196e863eba0e1376084cdf4d [2] https://www.postgresql.org/message-id/CAApHDvocUEYdt1uT%2BDLDPs2xEu%3Dv3qJGT6HeXKonQM4rY_OsSA%40mail.gmail.com [3] https://www.postgresql.org/message-id/54240C51.8040304%40vmware.com
