Am Donnerstag, den 13.11.2008, 13:59 +0100 schrieb Sylvain Pion:

> Jörn Ungermann wrote:

> > I am using CGAL for nearest neighbor interpolation in two and three

> > dimensions. The manual got me started well enough to implement both.

> > When compiling the code with gcc 4.2.1 *with* optimizations, I get

> > warnings with respect to some array subscripts being out of bounds.

> > 2D and 3D Delauney triangulations set off different errors.

> >

> > I attached example source files that I stripped down as much as I could

> > to generate the warnings. For each file, a log is contained that gives

> > the compilation command and the resulting output.

> >

> > For 3D triangulation, simply inserting points seems to suffice, while

> > for 2D, the nearest neighbor calculation triggers it.

> >

> > As we have a policy of treating warnings as errors this may prevent me

> > from using your nice library.

> >

> > If somebody could assert that these warnings are (known) artifices of

> > gcc or even better how to get rid of them (without setting compiler

> > flags) I would be delighted.

> > I can easily supply further information about my system, if it would

> > help or do test compilations. As I am unfamiliar with the CGAL and I

> > have no clue what even triggers some of the warnings finding the cause

> > of this is a bit over my head.

>

> We are aware of this warning. Unfortunately, we have not succeeded in

> getting rid of it (despite quite some time spent on it).

> In my opinion, it is a GCC bug, with pretty high probability.

Thanks for the information!

>

> > Second, I'd like to do natural neighbor interpolations on the surface of

> > a sphere (i.e. earth surface being approximated as either sphere or

> > ellipsoid). Is something like this already contained within CGAL or easy

> > to add?

>

> Does the following fit the bill?

>

http://www.cgal.org/Manual/3.3/doc_html/cgal_manual/Interpolation/Chapter_main.html#Section_52.2>

I was looking more for the following: I need something similar to the

regular 2D triangulation/nearest neighbor identification function, but

working not on an infinite euclidean plane, but on (ideally part of) the

unit sphere (or something topographically similar). A one time high

setup cost is OK, but the interpolation needs to be reasonably fast as

we need to do that often.

Playing a bit with the example pointed out by you, I get for 2000 points

(which is actually rather small compared to what we are striving to

have) about 200 interpolations per second via

"surface_neighbor_coordinates_3" whereas for (albeit distorted

projection to) real 2D it is via a pre-calculated triangulation and

"natural_neighbor_coordinates_vertex_2" it is about 250000 (which is

still a tad slow for our purposes, but OK). The problem is that no

suitable (conform) projection exists to map the unit sphere onto any

plane...

Kind regards,

Jörn

--

Jörn Ungermann

Dipl.-Mathematiker

ICG-1, Forschungszentrum Jülich

Tel.: +49 2461 61 1840

-------------------------------------------------------------------

-------------------------------------------------------------------

Forschungszentrum Jülich GmbH

52425 Jülich

Sitz der Gesellschaft: Jülich

Eingetragen im Handelsregister des Amtsgerichts Düren Nr. HR B 3498

Vorsitzende des Aufsichtsrats: MinDir'in Bärbel Brumme-Bothe

Geschäftsführung: Prof. Dr. Achim Bachem (Vorsitzender),

Dr. Ulrich Krafft (stellv. Vorsitzender), Prof. Dr. Harald Bolt,

Dr. Sebastian M. Schmidt

-------------------------------------------------------------------

-------------------------------------------------------------------

--

You are currently subscribed to cgal-discuss.

To unsubscribe or access the archives, go to

https://lists-sop.inria.fr/wws/info/cgal-discuss