CGAL users discussion list

[Efi Fogel <>]

Hi Shen,

The entire CGAL library, and not only the "2D Arrangement" package of CGAL, follows the Exact Geometric Computation (EGC) paradigm.

EGC ensures robust computation, while only using sufficient precision. Sometimes though high precision is required.

The high precision is required for robustness (and not necessarily for accuracy).

Now, let's assume that you do not take the risk of using a limited precision floating point number type.

Indeed you need to convert the limited precision coordinates and coefficients to the high-precision number type used.

You can reduce the number of bits and perhaps expedite the computation a bit by multiplying your real number by a certain factor, say 1000, cast to an integer, and divide back by 1000:

using Kernel = CGAL::Exact_predicates_exact_construction_kernel;

using FT = Kernel::FT;

using Tr = CGAL::Fraction_traits<FT>;

FT fx(Tr::Compose()(static_cast<int>(x * 1000), 1000));

Observe that those kinds of tricks may alter the topology, which is undesired.

Cheers,

Efi

   ____  _        ____             _
  /_____/_) o    /__________  __  //
 (____ (   (    (    (_/ (_/-(-'_(/
                         _/

On Mon, 25 Jul 2022 at 17:32, ShenCiao <> wrote:

Hi, 

I'm working on a 2D computer graphics project and need for Arrangement module. The project needs to construct arrangement objects from float precision polylines. (The project link if you are interested in).

But it seems that the whole arrangement module is constructed on the assumption of need for high precision. I haven't find a geometry trait support for floating-point number. As the document of CGAL::Arr_segment_traits_2 written: 

> While selecting an inexact kernel when developing a program usually leads to shorter running times, it causes robustness problems in most cases, and thus renders the program useless. 

And the Arr_segment_traits_2 is the modal of ArrangementTraits_2. I would think that every geometry traits are the modal of ArrangementTraits_2 and do not support floating point construction.

Well, it is ok to cast floating-point number to higher precision but it is not that optimal for performance and user experience in my project.

Is it possible to make a geometry trait myself that lets the whole arrangement module support floating-point construction? If it is possible, I would appreciate if I could receive more guidance and help from you gentlemen.

Thanks

Shen Ciao.

--
You are currently subscribed to cgal-discuss.
To unsubscribe or access the archives, go to
https://sympa.inria.fr/sympa/info/cgal-discuss