Skip to content

WIP: Reproduce ordering errors

Christian Engwer requested to merge reproduce_ordering_errors into master

The ordering interface changed in !555 (merged) and some subtle problems surfaced later (See #185).

In this MR I (tried to) implement

  • an empty local finite element, that can be used to manage parts of a domain that are not active
  • a variable finite element map, that dynamically changes the finite element implementation per element

The test now triggers the same issue that I mentioned in #185.

  • I use a dG basis that is present in some cells (fixed order) and simply empty on other cells
  • The resulting space can (in principal) be blocked by cells
  • the old approach would have been to use ordering::Chunked, which does not work anymore, as the resulting sized reported from "below" don't match anymore
  • the EntityBlockedOrderingTag doesn't work either, as it simply assumes that every cell has the same size (which we never promised, actually we report non-fixed sizes in the finite element map)

Merge request reports