Link to wrong header file in Docu of MappingQ?

[nfehn]

The docu of MappingQ https://www.dealii.org/developer/doxygen/deal.II/classMappingQ.html links to the header file

#include <deal.II/fe/mapping_manifold.h>

but it should be "#include <deal.II/fe/mapping_q.h>" instead, shouldn't it?

[kronbichler]

That is a good remark, I never noticed it. It seems Doxygen has difficulties in picking up the right include (it does have a few problems also elsewhere) sometimes. In this particular case, it seems it picks up a forward declaration here

dealii/include/deal.II/fe/mapping_manifold.h

Lines 32 to 33 in c80b699

	template <int, int>
	class MappingQ;

rather than the actual declaration of MappingQ. Does anyone have an idea (or do we already have a strategy) to prevent doxygen from picking up the wrong link? Of course, we could #ifndef DOXYGEN out all places we find, but maybe there is something more generic.

[nfehn]

Is the forward declaration in mapping_manifold.h actually needed? The one in mapping_q.h should at least not be needed.

I will try it out and maybe open a PR

[nfehn]

This should work, see #13330

[nfehn]

MappingQCache also has a wrong link due to a forward declaration in mapping_q.h.

The problem is more involved since it is defined a friend in MappingQ. I am currently trying to remove the friend declaration, so that we get rid of the forward declaration of MappingQCache as well.

[nfehn]

MappingQCache also has a wrong link due to a forward declaration in mapping_q.h.

The problem is more involved since it is defined a friend in MappingQ. I am currently trying to remove the friend declaration, so that we get rid of the forward declaration of MappingQCache as well.

This does not work. MappingQCache derives from MappingQ, but an independent object mapping_q of MappingQ calls the protected function compute_mapping_support_points().

Independently of the problem of wrong header files, I think that this should be resolved somehow, as this is difficult to understand (one would expect that a protected function can be called in a derived class without the need to declare it a friend class). One solution is obviously to make the function public, but there might also be reasons against it ... I guess someone else should resolve this. At least the documentation should be made more precise regarding the friend declaration in my opinion.

[nfehn]

For the problem with MappingQCache, I decided to follow the suggestion here #13330 (review)

The friend topic should be discussed elsewhere.