Kyotaro HORIGUCHI wrote:
> Hello, I've totally refactored the series of patches and cut out the
appropriate portion as 'unique (and non-nullable) index stuff'.
> As the discussion before, it got rid of path distinctness. This patch
works only on index 'full-orederedness', i.e., unique index on non-nullable
columns.
This is interesting!
I took a quick look at the patch. Here is my initial comment. I don't
think it's so good to set the pathkeys of a unique-index path to
query_pathkeys after the scan/join optimization because it can't exploit the
full-orderedness property in implementing mergejoins, i.e., it can't skip
doing an explicit sort that is considered unnecessary from the property.
So, I think the path's pathkeys should be set to query_pathkeys before the
scan/join optimization, i.e., at the index-path creation time.
If you wouldn't mind, I'd like to rework on the patch.
Thanks,
Best regards,
Etsuro Fujita
