Thanks for looking. On Thu, 23 Apr 2026 at 16:31, Chengpeng Yan <[email protected]> wrote: > One possible improvement, though not > a blocker, would be to defer the lhs-NULL handling until we actually > encounter the first NULL on the lhs. That could avoid a bit of extra > work in the common case where the lhs contains no NULLs.
I thought of it, but didn't do it as it meant having to keep a bit more state to track if we've filled the cache yet, plus the extra costs incurred to check if we've done it yet that would have to be paid for every NULL lookup. We currently have to check if the hash table has been set up already, so I felt more comfortable installing the new code in with that. David
