> On Mon, Jul 27, 2020 at 12:24:31PM +0200, Dmitry Dolgov wrote:
>
> I see, thanks for clarification. You're right, in this part of
> implementation there is no way to give up if LSN changes like
> _bt_killitems does. As far as I can see the leaf page is already pinned
> all the time between reading relevant tuples and comparing them, I only
> need to handle posting list tuples.
Here is a new version that hopefully address most of the concerns
mentioned in this thread so far. As before, first two patches are taken
from UniqueKeys thread and attached only for the reference. List of
changes includes:
* fix for index scan not being fully covered
* rebase on the latest UniqueKey patch
* taking into account posting tuples (although I must say I couldn't
produce a test that will hit this part, so I would appreciate if
someone can take a look)
* fixes suggested by Floris with adjustments as discussed in the thread
There are no changes related to EquivalenceClasses vs expressions, which
would probably be my next target. Having this in mind I must admit I'm
not super excited about possibility of including another patch as a
dependency without clear prospects and plans for it.
Thanks for the feedback folks!