On Sat, Sep 27, 2014 at 1:18 AM, Heikki Linnakangas
<hlinnakangas@vmware.com> wrote:
> Sort support for text with strxfrm() poor man's keys
> ---
>
> Peter: Are you waiting for Robert to review this? Robert, could you review
> the latest patch, please? Peter: Could you try to get rid of the extra
> SortSupport object that Robert didn't like?
> (http://www.postgresql.org/message-id/CA+TgmobDe+YDFnHTS0GWpT54-er8BPT3vx8rPshD+98CTDo25g@mail.gmail.com).
> I think it would speed up the process if you did that, instead of waiting
> for Robert to find the time.
I am not waiting on Robert to spend the time, FWIW. The question that
resolving if we should not have an extra sortsupport object is
blocking on is the need to have a consistent sorttuple.datum1
representation for the benefit of having comparetup_heap() know that
it's either always abbreviated keys or always pointers to text. My
view is that it's not worth going back to fix up datum1 to always be a
pointer to text when we abort abbreviation - I think we should just
forget about datum1 on the rare occasion that happens (due to the
costs involved, as well as the complexity implied).
I think that it will be necessary for me to rigorously prove that
view, as with the "memcmp() == 0" thing. So I'm looking at that.
--
Peter Geoghegan