CGAL users discussion list

["Sebastien Loriot (GeometryFactory)" <>]

The piece of code you "by passed" indicates the issue. There is no 
rounding mode handling in the javascript version which makes the
filtered predicates non robust anymore.

There is a high chance that the issue is coming from here. One solution
would be to change the interval implementation to not rely on a given
rounding mode (by always be pessimistic and increase/decrease the bounds
by one ulp). No idea what would be the effect on the runtime however.

Best,

Sebastien.


On 10/21/20 11:57 AM, Alessio Mochi ( via 
cgal-discuss Mailing List) wrote:
> Hello,
> I tried to compile cgal 5.1 with emscripten and boost 1_62.
>
> I had to bypass a piece of code in Interval_nt.h
>
> #if 0
>         CGAL_assertion_msg(-CGAL_IA_MUL(-1.1, 10.1) != CGAL_IA_MUL(1.1, 
> 10.1),
>                            "Wrong rounding: did you forget the  
> -frounding-math  option if you use GCC (or  -fp-model strict  for Intel)?");
>         CGAL_assertion_msg(-CGAL_IA_DIV(-1., 10) != CGAL_IA_DIV(1., 10),
>                            "Wrong rounding: did you forget the  
> -frounding-math  option if you use GCC (or  -fp-model strict  for Intel)?");
> #endif
>
> and use boost 1_62 to get a correct compilation.
>
> Many example work, but the corefine example crash on one 
>   CGAL_triangulation_assertion( s != LEFT_TURN );
>
> CGAL-5.1/include/CGAL/Triangulation_2.h at line 928
>
> This is the code where crash happen
>
> Vertex_circulator start = incident_vertices(infinite_vertex());
>      Vertex_circulator pc(start);
>      Vertex_circulator qc(start); ++qc;
>      Vertex_circulator rc(start); ++rc; ++rc;
>      do {
>        Orientation s = orientation(pc->point(),
>                                    qc->point(),
>                                    rc->point());
>        CGAL_triangulation_assertion( s != LEFT_TURN );
>        result = result && ( s != LEFT_TURN );
>        ++pc ; ++qc ; ++rc;
>      } while(pc != start);
>
> does anyone have any idea how to fix this problem?
> Thanks in advance.
>
>
> --
> You are currently subscribed to cgal-discuss.
> To unsubscribe or access the archives, go to
> https://sympa.inria.fr/sympa/info/cgal-discuss