CGAL users discussion list

[Ben Supnik <>]

Hi Laurent,

Two thoughts:

- Make the kernel numeric requirements a lot more obvious...I would 
suggest that the minimum kernel requirements be:

1. On the package overview page so that for any given package I can see 
"can I use this, or will I need a new kernel that's more expensive and 
powerful than what I've got?".  (For example, if I'm using CGAL for 
triangulations I might have an exact pred/inexact ctor kernel, but would 
have to "upgrade" to use boolean ops.)

2. Similarly, looking at the ref and user's manuals for both arrangement 
and boolean ops, the kernel requirement is sort of 'burried'.  I would 
suggest that it be at the top of any class definition that can be 
templated directly on the kernel and in the intro to the package.

I can see how this might be frustrating to the developers of CGAL (why 
don't those users read the manuals) but perhaps it is an indication of 
the power of CGAL that we (the users) can use these immensely powerful 
and efficient computational geometry algorithms without knowing enough 
about how they work to realize when we're in kernel-trouble.

(And sometimes it's not entirely obvious.  If you didn't know that the 
boolean operations construct an arrangement and may do so repeatedly in 
several steps when doing a join of many polygons, it might not be 
obvious why exact constructions and not just exact predicates are required.)

- My other idea is: simply use some C++ tagging in the kernel to cause 
the compiler to catch requirement failures.  For example, in my 
experience arrangement_2 is almost completely unusable without exact 
constructions...it wouldn't be a loss to require an exact constructions 
kernel.

cheers
Ben


Laurent Rineau (GeometryFactory) wrote:
> Le jeudi 03 septembre 2009 16:33:53, Bernd Gaertner a écrit :
> > Genís wrote:
> > > CGAL error:
> > > precondition violation!
> > > Expr: comp_f(object, nodeP->object) != LARGER
> > > File: /usr/include/CGAL/Multiset.h
> > > Line: 214
> > The cause is most likely this line:
> > >     typedef CGAL::Cartesian<double>    Kernel;
> > This is an inexact kernel that comes with no guarantee and whose usage
> > may lead to the indicated failures. CGAL has other kernels that do come
> > with guarantees, please see the FAQ item at
> > http://www.cgal.org/FAQ.html#inexact_NT, and the philosophy page at
> > http://www.cgal.org/philosophy.html.
>
> Genis, as you are not at all the only user of CGAL that made that error. Can 
> you tell us how you have happened to use that kernel? Did you see occurrences 
> of that error-prone kernel in the manual? Where? Have you copied an example 
> from CGAL? Which one?
>
> There are only a few cases where the use of CGAL::Cartesian<double> is the 
> right choice, and we, CGAL developers, should try to enhance the CGAL manual 
> and example so that our users are not guided to make such errors. That is why 
> I ask for feedback.

--
Scenery Home Page: http://scenery.x-plane.com/
Scenery blog: http://xplanescenery.blogspot.com/
Plugin SDK: http://www.xsquawkbox.net/xpsdk/
X-Plane Wiki: http://wiki.x-plane.com/
Scenery mailing list: 

Developer mailing list: