On 30/10/2023 14:55, Richard Guo wrote: > > On Thu, Oct 26, 2023 at 12:07 PM Andrei Lepikhov > <a.lepikhov@postgrespro.ru <mailto:a.lepikhov@postgrespro.ru>> wrote: > > Do you've thought about the case, fixed with the commit 1db5667? As I > see, that bugfix still isn't covered by regression tests. Could your > approach of a PARAM_EXEC slot reusing break that case? > > > Hm, I don't think so. The issue fixed by commit 1db5667 was caused by > sharing PARAM_EXEC slots between different levels of NestLoop. AFAICS > it's safe to share PARAM_EXEC slots within the same level of NestLoop. > > The change here is about sharing PARAM_EXEC slots between subquery's > subplan_params and outer-relation variables, which happens within the > same level of NestLoop. > ... > Did you notice a case that the change here breaks? > > Hi Tom, could you share your insights on this issue and the proposed > fix?
I think your patch works correctly so far. I mentioned the commit 1db5667 because, as I see, the origin of the problem was parallel workers. I have thought about pushing Memoize down to a parallel worker and couldn't imagine whether such a solution would be correct. Sorry if I disturbed you in vain.
Thanks for mentioning commit 1db5667, which brings my attention to more aspects about the PARAM_EXEC mechanism. I don't think the discussion is in vain. It helps a lot.