Bug on find_active_cell_around_point for p::d::t.

[luca-heltai]

PR #10174 reminded us that there is a problem in the fast algorithm, so we had to resort to the slow function

          return find_active_cell_around_point(mapping,
                                               mesh,
                                               p,
                                               marked_vertices);

in case the initial search did not succeed. This should be investigated and fixed.

[kronbichler]

Adjusting to the 9.3 milestone now that we have the temporary solution.

[luca-heltai]

So, this has been around for a long time: #7576. I think it is time we think of a permanent solution... (see also #10459, and #10506)

[nfehn]

I currently encounter problems with the fast GridTools::find_active_cell_around_point() function. My question is therefore: To which extent can one expect that the slow and fast versions give the same result? If the algorithm returns the first cell that is found, it might be that the slow and fast functions return different cells, is that correct? However, I call GridTools::find_all_active_cells_around_point() afterwards to find all cells, so I would expect that in this case the slow and fast versions (with a subsequent call of find_all_...()) must give the same result?

[nfehn]

So, this has been around for a long time: #7576. I think it is time we think of a permanent solution... (see also #10459, and #10506)

@luca-heltai what is the relation to #7930, which you did not mention above?

[nfehn]

The problem mentioned above will be solved by PR #10540

[bangerth]

@nfehn You mention that this should have been fixed by a patch a couple of years ago. Can you confirm that this is so?

[kronbichler]

I think we made several improvements during the last release. However, in exadg/exadg#438 we observed another problematic case. I am not sure if we can fix this entirely for the upcoming 9.5 release, but I think we also need to look over the interfaces, because there are too many functions doing similar things in GridTools. Let us keep this on the milestone for now.

[kronbichler]

I'm closing this issue, given the progress in #15233. We might have additional bugs or optimizations, but we should make those based on up-to-date test cases and/or code restructuring.