On Thu, 27 Aug 2026 at 19:36, Richard Guo <[email protected]> wrote: > For the same reason as those commits, I'd also leave this master-only. > It is a v17 regression, but planner fixes risk destabilizing plans in > stable branches, and this doesn't seem severe enough to be an > exception.
IMO, disabling a node type is quite a clear request that you only want that node type used if there's no alternative. I don't think the normal reasons for not wanting to destabilise plans here apply to this. If you leave this in, it'll mean there could be an inability in what happens based on how many partitions are pruned. It's also hard to imagine someone complaining that we stopped picking the plan with the disabled node type and started choosing the next best alternative. If they don't like the fact that the bug got fixed because they want the plan with the disabled node, then why do they have that node disabled in the first place? It just feels more likely someone will encounter undesired behaviour as a result of this rather than accidentally getting something good out of it. David
