marcin mank wrote:
> You are correct, I think, though this does not solve the division problem:
>
As a first goal I'm just attempting to reduce the EPSILON from 1.0E-6
down to 1.0E-015 (give or take). The current regression test suite works
fine down to 1.0E-09. At 1.0E-10 errors appear, not in the geometry
sections, but in the select_view test of all things. This is most likely
due to postgresql now giving the more correct (hence different) answers.
A real test suite is needed for this. Setting up PostGIS + MySQL +
OtherCommerical for comparison purposes. The other problem is many of
the basic geometric operators in postgres, such a left of, above, etc,
are so incorrectly implemented, they are not even wrong.