[postgis-devel] PostGIS 3D - Points to be discussed

[Oliver Courtin]

Le 1 févr. 2013 à 14:50, Oliver Courtin a écrit :

Hi Sandro,

> This is really good news, I'm excited to see CGAL getting into
> the picture. May I ask about robustness issues found with CGAL ?
> Are there known puzzles that CGAL is able to solve while GEOS 
> can't ? Does the existing testsuite pass all fine ?
> Note that I'm only talking 2D here…

Right now we are exploring CGAL, as a way to deal with 3D spatial
operations (as GEOS never planed to handle 3D)

So first new features provided by CGAL (and implemented in our first
SFCGAL stuff) are related to 3D spatial functions: ST_3DIntersects,
ST_3DIntersection, ST_3DDelaunayTriangles and so on…

Related to 3D operations, there's a need to keep arbitrary precision,
if we want to keep 3D geometries topology/consistency. 
The good news is that CGAL provide arbitrary precision (GEOS don't), 
with quite similar performances than GEOS ones for double precision 
(2D bench here)

The bad news is that serialization is not as cheap (meaning time) 
to CGAL structures than with GEOS ones, and so we explored a way to 
save (de)serializations on nested functions calls (e.g see ref_geometry)


About robustness and unit tests, the current SFCGAL implementation 
bring their one unit tests, and passed alls, (nota they are more focused 
on integration, rather than in data robustness)

We have not worked that far on 2D PostGIS/regress GEOS unit tests till now,
we know that there's at least some formatting issues (decimal handling
and so on) to deal with, before to be able to fully compare/check with 
SFCGAL 2D functions.


And even if it's not in your initial question, we don't have the feeling
that CGAL could be (in a near future) a way to not use GEOS anymore,
we are more thinking AND than OR...

O.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-devel/attachments/20130201/7212ac53/attachment.html>