CGAL users discussion list

[Marc Glisse <>]

On Fri, 4 Mar 2016, Indranil Chowdhury wrote:

>   I am new to the CGAL mailing list. I am playing with CGAL triangulation
> lately, and am trying to use my own double number type to pass to CGAL and
> want all computations that the CGAL kernel does to use this only (specially
> the comparison operators). Currently I am not concerned with speed, but
> want the kernel to interprete anything less than a threshold to be zero.
>
> Code :
>
> #include <boost\operators.hpp>

Ah, windows...

> #include <CGAL/Exact_predicates_inexact_constructions_kernel.h>
>
> //a number type to overload all floating point operations
> struct myDouble : boost::operators<myDouble>
> {
>  double val;
>
>  myDouble( double f )
>  {val = f;}
>  myDouble( const myDouble &f ):
>    val(f.val){}
>
> //arithmetic operators...
>
> //logical operators
>  bool operator<(const myDouble &other ) const
>  {
>    bool isEqual = (std::abs(val-other.val) < c_zero);//c_zero is a given
> threshold
>    return ( !isEqual && val<other.val );
>  }
>  bool operator==(const myDouble &other ) const
>  {
>    return ( std::abs(val-other.val) < c_zero );
>  }
> };
>
> ///From definition of inexact kernels
> namespace CGAL
> {
>  // A modification of inexact kernel from CGAL
>  class Epick_cgal2DSurf
>    : public Filtered_kernel_adaptor<
>    Type_equality_wrapper<
> Simple_cartesian<myDouble>::Base<Epick_cgal2DSurf>::Type, Epick_cgal2DSurf
> > ,
> #ifdef CGAL_NO_STATIC_FILTERS
>    false >
> #else
>    true >
> #endif
>  {
>  };
> }
>
> I get compilation error :
>
> Error    1    error C2440: '<function-style-cast>' : cannot convert from
> 'const CgalDouble' to 'CGAL::Gmpq'
> E:\AnsysDev\3rdparty\CGAL\CGAL-4.4\include\CGAL\NT_converter.h    41
>
> and I am not sure how to get this code to compile.
>
> My questions are:
> 1. Is there any way for CGAL to accept this numberType and get it  compiled?
> 2. Is there any other way to tell CGAL that any quantity less than
> specified is zero?

It isn't clear what you are hoping to achieve. Probably you are looking 
for the kernel Simple_cartesian<myDouble>, but I am not sure. 
Filtered_kernel implements all predicates by converting coordinates to 
intervals, evaluating using intervals, and if that doesn't give the 
answer, it converts to an expensive exact type (Gmpq here) and evaluates 
again. Your number type is not convertible to intervals and Gmpq so it 
cannot use Filtered_kernel.

--
Marc Glisse