What's the problem with this code (hyperbolic Delaunay)?

[stla]

Hello,

Here is my code. I call it from R and it works fine. However, when it is checked with Valgrind (a tool to check memory problems in R packages), there are some issues, such as:

==539271== Conditional jump or move depends on uninitialised value(s)
==539271==    at 0x5785AC92: operator() (R-devel/site-library/RcppCGAL/include/CGAL/Hyperbolic_triangulation_2/internal/Hyperbolic_Delaunay_triangulation_traits_2_functions.h:180)
==539271==    by 0x5785AC92: mark_face (R-devel/site-library/RcppCGAL/include/CGAL/Hyperbolic_Delaunay_triangulation_2.h:802)
==539271==    by 0x5785AC92: CGAL::Hyperbolic_Delaunay_triangulation_2<CGAL::Hyperbolic_Delaunay_triangulation_traits_2<CGAL::Epick>, CGAL::Triangulation_data_structure_2<CGAL::Triangulation_vertex_base_with_id_2<CGAL::Hyperbolic_Delaunay_triangulation_traits_2<CGAL::Epick>, CGAL::Triangulation_vertex_base_2<CGAL::Hyperbolic_Delaunay_triangulation_traits_2<CGAL::Epick>, CGAL::Triangulation_ds_vertex_base_2<void> > >, CGAL::Hyperbolic_triangulation_face_base_2<CGAL::Hyperbolic_Delaunay_triangulation_traits_2<CGAL::Epick>, CGAL::Triangulation_ds_face_base_2<void> > > >::mark_finite_non_hyperbolic_faces() const (R-devel/site-library/RcppCGAL/include/CGAL/Hyperbolic_Delaunay_triangulation_2.h:734)
==539271==    by 0x578644CD: insert<__gnu_cxx::__normal_iterator<CGAL::Point_2<CGAL::Epick>*, std::vector<CGAL::Point_2<CGAL::Epick>, std::allocator<CGAL::Point_2<CGAL::Epick> > > > > (R-devel/site-library/RcppCGAL/include/CGAL/Hyperbolic_Delaunay_triangulation_2.h:450)
==539271==    by 0x578644CD: Rcpp::Vector<19, Rcpp::PreserveStorage> hdelaunay_cpp<CGAL::Hyperbolic_Delaunay_triangulation_2<CGAL::Hyperbolic_Delaunay_triangulation_traits_2<CGAL::Epick>, CGAL::Triangulation_data_structure_2<CGAL::Triangulation_vertex_base_with_id_2<CGAL::Hyperbolic_Delaunay_triangulation_traits_2<CGAL::Epick>, CGAL::Triangulation_vertex_base_2<CGAL::Hyperbolic_Delaunay_triangulation_traits_2<CGAL::Epick>, CGAL::Triangulation_ds_vertex_base_2<void> > >, CGAL::Hyperbolic_triangulation_face_base_2<CGAL::Hyperbolic_Delaunay_triangulation_traits_2<CGAL::Epick>, CGAL::Triangulation_ds_face_base_2<void> > > >, CGAL::Point_2<CGAL::Epick> >(Rcpp::Matrix<14, Rcpp::PreserveStorage>) (packages/tests-vg/gyro/src/hdelaunay.cpp:60)
==539271==    by 0x57830422: hdelaunay_K(Rcpp::Matrix<14, Rcpp::PreserveStorage>) (packages/tests-vg/gyro/src/hdelaunay.cpp:118)

Because of this problem I can't publish my package :-(. Do you have an idea?

[stla]

I found another way to proceed. Hopefully this will work now.

[stla]

Does not work. Valgrind complains at the line calling insert:

hdt.insert(hpts.begin(), hpts.end());

I wanted to try reserve(hdt, nvertices, nfaces, nedges) but I don't understand how to call this function. What is its namespace?

[afabri]

What do you refer to with reserve() ?

[stla]

Hello,
I'm not sure to understand your question. Valgrind seems to complain because my hyperbolic Delaunay triangulation (hdt) object is not initialized. So I'm trying to use reserve which is defined for the MutableFaceGraphs.

[afabri]

Triangulations are not mutable. At least that is not stated here.

[stla]

Ah ok. How could I allocate memory? Can I use malloc? Everything works fine, that's just Valgrind who complains.

[stla]

It is claimed in the doc that push_back is equivalent to insert. However my code compiles with insert but not with push_back (error: Vertex_handle inaccessible within this context).

[afabri]

You write that because it should be inherited from the base class Delaunay_triangulation_2?

[stla]

Yes, I'm refering to this point in the doc.

[afabri]

It would be easy to add. It is also well possible that other functions of the base class are missing.

[afabri]

When I compile the example with the address sanitizer of VC++ enabled I also get error messages. So we should maybe have a closer look.

[afabri]

Well, I do not have it with clang and its address sanitizer.

[stla]

I've done a minimal example:

#include <CGAL/Exact_predicates_inexact_constructions_kernel.h>
#include <CGAL/Triangulation_data_structure_2.h>
#include <CGAL/Hyperbolic_Delaunay_triangulation_2.h>
#include <CGAL/Hyperbolic_Delaunay_triangulation_traits_2.h>
#include <CGAL/Triangulation_vertex_base_with_id_2.h>

typedef CGAL::Exact_predicates_inexact_constructions_kernel K;
typedef CGAL::Hyperbolic_Delaunay_triangulation_traits_2<K> HDtt;
typedef HDtt::Point_2 HPoint;
typedef CGAL::Triangulation_data_structure_2<
  CGAL::Triangulation_vertex_base_with_id_2<HDtt>,
  CGAL::Hyperbolic_triangulation_face_base_2<HDtt>>
    HTds;
typedef CGAL::Hyperbolic_Delaunay_triangulation_2<HDtt, HTds> HDt;

int main(){
  HDt hdt;
  hdt.insert(HPoint(0.1, 0.1));
  hdt.insert(HPoint(0.8, 0.2));
  hdt.insert(HPoint(0.4, 0.6));
  hdt.insert(HPoint(-0.5, -0.5));
  return 0;
} 

Here are the Valgrind complains. Functions whose name start with R are from R.

So it really seems the problem is not from my code (this time), this is from CGAL.

I said something wrong here:

Valgrind (a tool to check memory problems in R packages)

Valgrind has nothing to do with R. This is a tool to check C/C++ code. There's also UBSAN or ASAN which are similar tools, that you can use via Docker.

[stla]

I checked the Euclidean Delaunay triangulation with Valgrind. No issue, clean.