On Mon, Jul 1, 2019 at 8:47 PM Nikita Glukhov <n.glu...@postgrespro.ru> wrote: > On 01.07.2019 13:41, Thomas Munro wrote: > > On Tue, Mar 26, 2019 at 4:30 AM Nikita Glukhov <n.glu...@postgrespro.ru> > > wrote: > >>>> Fixed two bugs in patches 3 and 10 (see below). > >>>> Patch 3 was extracted from the main patch 5 (patch 4 in v9). > >>> This patch no longer applies so marking Waiting on Author. > >> Attached 11th version of the patches rebased onto current master. > > Hi Nikita, > > > > Since a new Commitfest is here and this doesn't apply, could we please > > have a fresh rebase? > > Attached 12th version of the patches rebased onto the current master.
I have more thoughts about planner/executor infrastructure. It appears that supporting "ORDER BY col1, col2 <-> val" is too complex for initial version of patch. Handling of "ORDER BY col" and "ORDER BY col <-> val" cases uses different code paths in optimizer. So, I'd like to limit first version of this patch to support just most simple "ORDER BY col <-> val" case. We would probably be able to enhance it even in v13 release cycle, but let's start with just simple case. I also think we can evade replacing amcanorderbyop flag with method, but introduce just new boolean flag indicating knn supports only first column. ------ Alexander Korotkov Postgres Professional: http://www.postgrespro.com The Russian Postgres Company
