I am running edge collapse on a mesh with 134 million faces. I am using the example code from here:

https://doc.cgal.org/latest/Surface_mesh_simplification/index.html#title20

“Example using Garland-Heckbert policies”

With this particular mesh, I now encounter the following error:

terminate called after throwing an instance of 'CGAL::Precondition_exception'
  what():  CGAL ERROR: precondition violation!
Expr: !get(vcm(), profile.v0()).isZero(0)
File: /nix/store/6x9gp50r6njnbb3whhn5d49v01ig016c-cgal-5.5.2/include/CGAL/Surface_mesh_simplification/Policies/Edge_collapse/internal/GarlandHeckbert_policy_base.h
Line: 227

This is the broader context of the precondition violation:

  // class GarlandHeckbert_cost_and_placement

  // Cost
  template <typename Profile>
  boost::optional<typename Profile::FT>
  operator()(const Profile& profile,
             const boost::optional<typename Profile::Point>& placement) const
  {
    typedef boost::optional<typename Profile::FT>                              Optional_FT;

    if(!placement)
      return boost::optional<typename Profile::FT>();

    CGAL_precondition(!get(vcm(), profile.v0()).isZero(0));
    CGAL_precondition(!get(vcm(), profile.v1()).isZero(0));

    const Mat_4 combined_matrix = combine_matrices(get(vcm(), profile.v0()),
                                                   get(vcm(), profile.v1()));
    const Col_4 pt = point_to_homogenous_column(*placement);
    const Optional_FT cost = (pt.transpose() * combined_matrix * pt)(0, 0);

    return cost;
  }

in file Surface_mesh_simplification/include/CGAL/Surface_mesh_simplification/Policies/Edge_collapse/internal/GarlandHeckbert_policy_base.h

This doesn’t happen on every mesh, but already on a couple meshes.
I can’t provide a sample mesh for easy reproduction yet (client data).

I prefer using Garland-Heckbert over Lindstrom-Turk because of this issue with Lindstrom-Turk.

The way I understand the failed precondition:

• profile.v0() and profile.v1() are two vertices of some edge.
• get makes a lookup in the “vertex cost map”/vcm, which probably gives the cost associated with the deletion of each vertex
• this cost can’t be exactly zero

Any idea how to go about debugging this? It would be interesting to know if this condition is linked to a certain characteristic in my input mesh.

There is no obvious weirdness in my input mesh by just looking at it.