Corefinement Produces Self-Intersecting Meshes Even When Both Input Meshes Are Valid

[chudonghao]

Issue Details

I encountered a problem when performing a corefine operation on two meshes. Although both input meshes contain no self-intersections or any other anomalies, the resulting mesh after corefinement becomes invalid and contains self-intersections.

This issue can be reproduced in the CGAL Mesh Boolean Operations demo by loading two valid meshes and applying corefine/difference/union/intersection.

My questions are:

1. Why does corefinement produce a mesh with self-intersections even when both input meshes are valid?

2. How can this issue be avoided or resolved?

   • Is additional pre-processing required?
   • Are there recommended settings, repair steps, or robustness guidelines?
   • Is this a known limitation of the current corefinement implementation?

mesh.zip

[chudonghao]

Statistics:

Statistics.txt

[sloriot]

I gave a try with your meshes in the demo and I could compute the output of the union, intersection and difference successfully. The mesh I get is valid but indeed it contains self-intersections created when the exact intersection points are rounded to double values. A call to PMP::remove_almost_degenerate() (I put threshold to 100) removed all of them.

[chudonghao]

Does it mean that Epick is not used internally in the algorithm, and the precision is reduced only when outputting?

[sloriot]

For parts using only input points, EPICK is used. As soon as a construction occurs, an internal exact representation is stored and used for answering exactly predicates involving those points.