On Tue, 23 Jan 2024 at 18:56, Richard Guo <guofengli...@gmail.com> wrote: > > Similar to what we did to GROUP BY keys in 0452b461bc, I think we can do > the same to DISTINCT keys, i.e. reordering DISTINCT keys to match input > path's pathkeys, which can help reduce cost by avoiding unnecessary > re-sort, or allowing us to use incremental-sort to save efforts. For > instance,
I've not caught up on the specifics of 0452b461b, but I just wanted to highlight that there was some work done in [1] in this area. It seems Ankit didn't ever add that to a CF, so that might explain why it's been lost. Anyway, just pointing it out as there may be useful code or discussion in the corresponding threads. David [1] https://postgr.es/m/da9425ae-8ff7-33d9-23b3-2a3eb605e106%40gmail.com
