On Wed, 26 Mar 2025 at 01:11, David Rowley <dgrowle...@gmail.com> wrote: > I'm happy to proceed with the v7 version. I'm also happy to credit you > as the primary author of it, given that you came up with the v2b > patch. It might be best to extract the performance stuff that's in v7 > and apply that separately from the bug fix. If you're still interested > and have time in the next 7 hours to do it, would you be able to split > the v7 as described, making v8-0001 as the bug fix and v8-0002 as the > performance stuff? If so, could you also add the total_jumble_len to > v8-0001 like Michael's last patch had, but adjust so it's only > maintained for Assert builds. Michael is quite keen on having more > assurance that custom jumble functions don't decide to forego any > jumbling. > > If you don't have time, I'll do it in my morning (UTC+13).
Here is the v8 version with the bug fix and performance stuff separated out. I also added the Assert code to ensure we always add something to the jumble buffer when jumbling a node. I didn't include the regression tests I saw in the v2b patch [1]. I felt these were just marking the fact that there used to be a bug here. David [1] https://postgr.es/m/5ac172e0b77a4baba50671cd1a15285f%40localhost.localdomain
v8-0001-Fix-query-jumbling-to-account-for-NULL-nodes.patch
Description: Binary data
v8-0002-Optimize-Query-jumble.patch
Description: Binary data
