Re: AW: AW: BUG #18147: ERROR: invalid perminfoindex 0 in RTE with relid xxxxx

On Mon, Oct 23, 2023 at 12:01 PM Peter Geoghegan <pg(at)bowt(dot)ie> wrote:
> On Mon, Oct 23, 2023 at 11:55 AM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> > Yeah, there are comments within index_unchanged_by_update about those
> > things. What I'm unhappy about is that indexam.sgml's discussion of
> > the indexUnchanged flag makes it sound far more trustworthy than it
> > actually is. Somebody who just read that doco and didn't scour the
> > underlying code would be badly misled.
>
> I understand. I'll come up with a doc patch for that later on today.

What do you think of the attached?

If there are no objections I'll commit this soon, backpatching to v14.

Separately, I wonder if index_unchanged_by_update should actually just
always give the hint with a non-HOT update, regardless of the
specifics for each index/its columns -- just like on the v14 branch.

In general I'm more concerned about the danger of not giving the hint
when we should, rather than giving the hint too often. Most individual
btinsert() calls do nothing with the hint already (because there is
still enough free space for the incoming item on the page). The hint
is inherently nothing more than a signal that bottom-up index deletion
might be a good idea, iff we're just about out of better options for
affected leaf pages.

--
Peter Geoghegan