CGAL users discussion list

[Phil Weir <>]

On 30/10/13 17:28, Phil Weir wrote:
> On 10/17/2013 05:57 PM, Garth N. Wells wrote:
> > What I'm doing is surface meshing a sphere, converting the surface 
> > mesh to a polyhedron, and trying to generate a volume mesh of the 
> > polyhedron (the motivation is to add/subtract polyhedra to create 
> > more complicated domains). The code that reproduces the error (using 
> > CGAL 4.2 and GCC 4.8) is attached. The code is basically a 
> > concatenation of the demos in sections 48.3.1 and 51.3.2. Any 
> > suggestions to get this working?
> >
> > Garth
>
> Actually, I have just had a very similar problem the last couple of 
> days, same abort message - I rebuilt CGAL 4.2, with identical source 
> to the working version, and it appeared. The old binary still runs 
> fine. CGAL 4.2 and 4.3 show the same thing with the 
> mesh_polyhedral_domain demo. A fresh compile of CGAL 4.3 on an Ubuntu 
> 12.04 virtual box has no problems. As such, I suspect there may be 
> something with library versions.
A related problem with the same code turned out to be this bug in the 
version of the STL libraries 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58800 that is now default on 
Ubuntu 13.10. I have not confirmed whether the underlying cause is 
related to the problem above too, but this particular one was occurring 
during conversion of a polyhedron to a Nef polyhedron - K3 tree calls 
std::nth_element, which in turn calls an introselect routine.

This routine runs past the iterator end in a small percentage of cases 
where it picks out a (critically) bad pivot. As K3 tree hits this 
routine a lot of times for a large polyhedron, it seems to fairly 
consistently kill my code for non-trivial tests - the comparison 
operator dereferences the iter and CGAL::assign fails to cast it. The 
upshot is a segfault below K3_tree.h:116-8 when trying to pull out the 
point value of a vertex for comparison - if anybody else sees this, that 
may be why.

I don't know if anybody else is currently trying CGAL on Ubuntu 13.10, 
but I could imagine that causing non-obvious bugs in other 
geometry-building places too.

Best,
PTW