[postgis-devel] Bug in default B-Tree operator class for geography

[Peter Geoghegan]

On Wed, Aug 24, 2016 at 8:25 AM, Paul Ramsey <pramsey at cleverelephant.ca> wrote:
> So, I don't think that fixing them to be deterministically ordered should
> cause any especially bad user-facing effects (by the same token, the fact
> that they are broken probably hasn't bitten anyone, particularly since they
> are broken on empties, which also show up rarely in practice).

It might -- I think it will change the behavior of "SELECT
COUNT(DISTINCT foo) FROM bar", in some cases. Although, the answers
that will be given to queries like that (where foo is a geography
column with empty geometries) will probably be different based on
accidentally factors like how quicksort leaves things ordered from its
input (quicksort is not stable).


-- 
Peter Geoghegan