Re: Fix rescan costing of single-batch hash joins

Hi Baji, Etsuro,

Reproduced both repro scripts on two PG18.4 builds (same REL_18_4 tag, one
unpatched, one with just the costsize.c hunk). Output attached.

repro_plain.sql: plan shapes match your report exactly (Hash Join -> Nested
Loop+Memoize).

repro_e2e.sql: plan_cache_mode=auto, no config changes --
unpatched: generic_plans=0, custom_plans=10, ~65 ms/call
patched: generic_plans=5, custom_plans=5, ~1.8 ms/call

Confirms the undercosted custom plan was defeating the plan cache's own
generic-vs-custom check.

Thanks,
Nazneen

On Thu, Oct 22, 2026 at 6:32 AM Baji Shaik <baji(dot)pgdev(at)gmail(dot)com> wrote:

> On Thu, Oct 22, 2026 at 6:47 AM Etsuro Fujita <etsuro(dot)fujita(at)gmail(dot)com>
> wrote:
>
>> I agree that this is for master-only as well. Could you add it to the
>> upcoming commitfest. I'd like to review it next month.
>
>
> Thank you for looking into this, Etsuro.
> Added to the commitfest: https://commitfest.postgresql.org/patch/7502/
>
> Thanks,
> Baji Shaik.
>