Ordering wrongly assumes fixed size

I stumbled again on an issue related to the determination of fixed size in the ordering (old issue #59 (closed), that was fixed in !129 (merged)).

In the old issue and the test case provided therein, we created a finite element (Q1DG) that has size 0 on elements with an even index and size c on elements with an odd index. According to the description in !129 (merged), the fixed version should work in all cases but the one where only the last element in the iteration has a non-zero number of DOFs.

In andreas.nuessing/dune-pdelab@fefe8300 I extended the test to also check the case where the number of degrees switches from 0 to c in the middle of the domain. In this case, the ordering again assumes a fixed size and the size of the ordering is incorrect. For 1D, Q1DG, 15 elements and switch after element 7 the size of the ordering should be 7*2=14 but is 15*2=30. In addition, fixedSize is true.

I did not investigate further where the problem is coming from, but could this issue be related to the exception of the original fix? If this is the case, would it be possible to force the ordering to not assume fixed size? (Similar to the old NoConstOrderingSize?)

I (still) don't grasp the whole ordering-magic, but I will try to dig deeper.