Mapping::get_vertices() uses GeometryInfo::vertices_per_cell

[nfehn]

What happens in the non-hyper-cube case? see also https://github.com/dealii/dealii/issues/13308

[peterrum]

Could you be a bit more precise? Mapping returns boost::container::small_vector< Point< spacedim >, GeometryInfo< dim >::vertices_per_cell >. GeometryInfo< dim >::vertices_per_cell is not the size of the vector but the max size.

[nfehn]

Yes, I mean the return type. So you want to say that we are on the save side with GeometryInfo< dim >::vertices_per_cell, even if this value is defined for hyper-cubes?

[bangerth]

That last argument is the maximal size boost::small_vector can efficiently save. You need to look at what the function actually returns, i.e., what size it sets the vector to.

[nfehn]

thanks. Ok, so there seems to be no problem in using this function for simplex for example? This information is still somewhat implicit here in this discussion.

I think it would be good to explain this in the documentation, and with your help I can also add this if you want.

More generally, I wonder why this function belongs to Mapping at all, and why it is not associated to Triangulation (and e.g. deprecated for mapping?).

[bangerth]

Would it be easier to understand if the return type was

boost::container::small_vector<Point<spacedim>, (1<<dim)>

instead? I would accept a patch to that end.

As for an explanation of what boost::small_vector does, see https://www.boost.org/doc/libs/1_62_0/doc/html/boost/container/small_vector.html

[peterrum]

boost::container::small_vector<Point, (1<<dim)>

Please not. The information (number of max vertices), should become part of ReferenceCell.

[bangerth]

So you'd like to have

  /**
   * Return the maximal number of vertices a `dim`-dimensional cell can have.
   */
  template <int dim>
  constexpr unsigned int max_vertices_per_cell () { return (1<<dim); }

in namespace ReferenceCell and use that in place of the template argument? Or make it a template variable? That would both certainly be ok with me as well.

[bangerth]

(Actually, in namespace ReferenceCells, plural.)

[nfehn]

What I want (as a minimal solution), is to write into the docu "This function assumes the maximum number of vertices of a cell is not larger than 2^dim", if I understood you correctly. Please correct me.

Of course, I would prefer a non-incremental solution even more. One should just not see GeometryInfo in the return type in my opinion, and also not something like (1<<dim). I actually thought about replacing (1<<dim) in the implementation of GeometryInfo by something readable. Why not std::vectoras return type? Why not remove this function (as written above)?

[nfehn]

Please not. The information (number of max vertices), should become part of ReferenceCell.

Would also be a solution for me since this solution is at least clear in terms of documentation.

[bangerth]

boost::small_vector can store arbitrary numbers of elements. It just happens to do so more efficiently for small numbers, where "small" means "at most as many as given by the second template argument". As a consequence, setting that template argument to max_vertices_per_cell <dim> makes sense.