We need to address the error

/home/runner/work/dealii/dealii/include/deal.II/matrix_free/fe_remote_evaluation.h:1192:19: error: calling a protected constructor of class 'dealii::Quadrature<2>' [clang-diagnostic-error]

This was recently introduced by @jh66637 in #16394 - any comments on how to address this? I think it would make sense to not store quadrature formulas but merely points (depending on how the consumer site looks like), or fill the quadrature objects with one of the other constructors.

The code you point out looks like this:

                // Insert a quadrature rule of correct size into the global
                // quadrature vector. First check that each face is only
                // considered once.
                Assert(global_quadrature_vector[bface].size() == 0,
                       ExcMessage(
                         "Quadrature for given face already provided."));

                global_quadrature_vector[bface] =
                  Quadrature<dim>(phi.n_q_points);

I think it's worth pointing out that what one ends up with here is a quadrature object with phi.n_q_point quadrature points that are all at the origin, and zero weights. This is a prescription for bugs (which is how I found out about this constructor to begin with). In other places we would initialize points and weights with signaling_nans, but here we don't.

In other words, I think that the design used here is broken. The object as initialized is unusable. We shouldn't create it. I don't know where that is, but assume that points and weights are set somewhere else, and we should leave resizing to the correct size to the place where we do the initialization of points and weights.

I'm afraid I have no overview of this code. Could someone help me with this location?

I can definitely help if @jh66637 is busy, let us wait for a day or two (as my schedule is overly crowded today and tomorrow).

@bangerth @kronbichler The reason I used this is the following: We only need the size of the quadrature objects. At some point I pointed out, that we have to decide if we keep the interface similar to FERemoteEvaluationCommunicator::reinit_faces() which requires a vector of quadratures or simply hand over a vector of quadrature sizes. I didn't find the respective comment, but at that time there was no opinion on how to proceed, and I kept it the way the function was written initially.

Long story short, we can simply replace

FERemoteEvaluationCommunicator::reinit_faces(..., const std::vector<std::vector<Quadrature<dim - 1>>> &quadrature_vector)

by

FERemoteEvaluationCommunicator::reinit_faces(..., const std::vector<std::vector<unsigned int>> &quadrature_sizes)

and similar for all overloads. This functionality is quite new, was in no release and is probably not used by a lot of people, so this can be easily changed in my opinion. I can do that in a separate PR if you want to, or you can do it with this PR. What do you prefer?

Since you know the code, I'd love if you could make that change in a separate PR. Once that's merged, I can rebase the current one. Thank you in advance!

Since you know the code, I'd love if you could make that change in a separate PR.

I will try to find the time tomorrow :)

Now ready given that the one place where this failed has been addressed in #17032.

There are still failing tests:

The following tests FAILED:
61 - simplex/data_out_write_gnuplot_01.debug (Failed)
62 - simplex/data_out_write_gnuplot_02.debug (Failed)
63 - simplex/data_out_write_gnuplot_03.debug (Failed)
68 - simplex/data_out_write_vtk_01.debug (Failed)
69 - simplex/data_out_write_vtk_02.debug (Failed)
155 - simplex/step-23.debug (Failed)
156 - simplex/step-31.debug (Failed)
163 - test_dependency/simplex.step-67.debug.executable (Failed)
164 - simplex/step-67.mpirun=1.debug (Not Run)
165 - simplex/step-67.mpirun=2.debug (Not Run)

I made it to work now.