Geometry::input_global_indices might be needed here to avoid hardcoding values

Geometry::input_global_indices might be needed here to avoid hardcoding values

Thanks for pointing me to this function. I think to make proper use of this and to simplify the tests in this and the related PR's we probably want to split the testing of the generation (everything prior to call to create_mesh) and the mesh generation testing (after redistributing, creating ghost nodes). So we need an interface change. How about something like impl::create_interval_cells -> {x, cells} (and equally for the other mesh types - this should allow for a shared code path for the mesh generation in generation.h all together) that is then called within create_interval to facilitate this?

What remains an open problem with this restructuring would be the question of how to check the mesh parallelization effectively (controlling in test cases which process owns which node and where ghost nodes are generated).

Is this ready, or does it need changes based on @jpdean's comment?

I'd say its good to go, and we can adapt later on. Addressing the before mentioned issue will require more changes, than only the tests, so maybe a good idea to separate it from this.

In principle these tests could be moved to python, I think. Nevertheless I would push for keeping them here in the cpp part. We are strictly unit testing here and thus should avoid as many additional layers as possible.

For example the related tests with parallel cases (in other PRs) currently are not transferable to python due to create_cell_partitioner not working in python as noted in #3362.

Superseded by #3359