Re: [HACKERS] create_unique_path and GEQO

On Thu, Nov 30, 2017 at 8:19 PM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> On Fri, Mar 24, 2017 at 9:50 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> Yeah. I think the code in mark_dummy_rel is newer and better-thought-out
>> than what's in create_unique_path. It probably makes sense to change over.
>
> I did a bit of archaeology here. create_unique_path() first appears
> in commit bdfbfde1b168b3332c4cdac34ac86a80aaf4d442 (vintage 2003),
> where it used GetMemoryChunkContext(rel). Commit
> f41803bb39bc2949db200116a609fd242d0ec221 (vintage 2007) changed it to
> use root->planner_cxt, but neither the comment changes in that patch
> nor the commit message give any hint as to the motivation for the
> change. The comments do mention that it should be kept in sync with
> best_inner_indexscan(), which was also switched to use
> root->planner_cxt by that commit, again without any real explanation
> as to why, and best_inner_indexscan() continued to use
> root->planner_cxt until its demise in commit
> e2fa76d80ba571d4de8992de6386536867250474 (vintage 2012). Meanwhile,
> mark_dummy_rel() didn't switch contexts at all until commit
> eca75a12a27d28b972fc269c1c8813cd8eb15441 (vintage 2011) at which point
> it began using GetMemoryChunkContext(rel).

Thanks a lot for digging into the history.

>
> All of which, I think, is a long-winded way of saying that I'm going
> to go commit this patch.

Thanks a lot for the commit.

--
Best Wishes,
Ashutosh Bapat
EnterpriseDB Corporation
The Postgres Database Company