Andreas Fabri

Hi @mglisse I just added the 4D version. The user we know only constructs a convex hull, and all points are on the hull, so the triangulation class does the...

In 6D with 100K points the runtime is 207 vs 94 seconds.

If I remember correctly there is still something todo by @LeoValque .

Problem with [`nullptr`](http://en.cppreference.com/w/cpp/language/nullptr) in travis

> > > TODO: Once all tests pass, add specialization of the construction wrapper to use approx input when possible (to avoid non needed conversion to interval). After the exact...

As we do constructions always with the exact time, we get lots of mallocs and reallocs of `Gmpq`. @mglisse I saw this [newsgroup entry](https://gmplib.org/list-archives/gmp-discuss/2012-March/004950.html) and I am wondering if you...

Looking at cgal-public-dev/STL_Extension-recycle_Handle_for-glisse I am wondering if we do not speak about different things here. I thought one would have to use a pool on a lower level, namely [Custom...

The `Filtered_rational_kernel` is used as `Exact_predicates_exact_constructions_kernel` when `CGAL_USE_FILTERED_RATIONAL_KERNEL` is defined. I compared the performance of three programs - Polygon_mesh_processing/examples/Polygon_mesh_processin/corefinement_mesh_union_with_evpm.cpp takes 6.9 sec instead of 5.1 sec - Boolean_set_operations/benchmark/Boolean_set_operations/polygon_set_2_join.cpp takes 17...

Everything becomes slighlty slower. For example the fit_in_double inside the static filter runs on an interval and has to determine again and again that inf == sup , the hilbert...

We probably should focus on examples where we really make constructions.