Crash when computing a centroid

[stla]

Hello,

There's a crash of the R session when I compute the centroid of the mesh in the attached file. I'm going to prepare a reproducible C++ code in order that you can try.

crash_centroid.off.txt

[stla]

The code:

#include <CGAL/Exact_predicates_exact_constructions_kernel.h>
#include <CGAL/Kernel/global_functions.h>
#include <CGAL/Surface_mesh.h>
#include <CGAL/number_utils.h>
#include <CGAL/Polygon_mesh_processing/measure.h>
#include <CGAL/Polygon_mesh_processing/IO/polygon_mesh_io.h>

typedef CGAL::Exact_predicates_exact_constructions_kernel EK;
typedef EK::Point_3                                       EPoint3;
typedef CGAL::Surface_mesh<EPoint3>                       EMesh3;

namespace PMP = CGAL::Polygon_mesh_processing;

int main() {
  EMesh3 mesh;
  const bool ok = PMP::IO::read_polygon_mesh(
    "crash_centroid.off", mesh, CGAL::parameters::verbose(true)
  );
  if(!ok) {
    std::cout << "Reading failure";
    std::exit(1);
  }
  const bool valid = mesh.is_valid(false);
  if(!valid) {
    std::cout << "The mesh is not valid.";
    std::exit(1);
  }
  const EPoint3 centroid = PMP::centroid(mesh);
  std::cout << CGAL::to_double<EK::FT>(centroid.x()) << " "
            << CGAL::to_double<EK::FT>(centroid.y()) << " "
            << CGAL::to_double<EK::FT>(centroid.z()) << "\n";
  return 0;
}

Not sure I included all the required .h files (I don't know how to run this code).

[stla]

I get:

Erreur : pile C débordée vers le haut

[lrineau]

PMP::centroid is probably an issue similar to the one that was fixed by @janetournois with that commit 5b41ca0.

The issue is that the CGAL::Epeck kernel maintains a graph of construction, and probably PMP::centroid creates a long chain of constructions, by accumulating values.

Jane, do you remember that issue?

[sloriot]

@afabri since EPECK is used I suspect that's the same kind of issues with iterative computations (like what we fixed in area for example)

[lrineau]

We answered almost the same answer, almost at the same moment. See #7005 (reply in thread).

[stla]

Oddly, there's no crash with Valgrind.

If this is due to EPECK I will convert to EPICK before the calculation.

[janetournois]

Note we have already talked about similar issues #2066 and #2073

[stla]

Works with EPICK. Thanks.