splitting TensorProductElements does not work after UFL PR #122

[JHopeCollins]

UFL PR #122 changed the logic of how objects are handled in split.

It used to be that the original object was returned if it was not from a MixedFunctionSpace, but now it is only returned if ufl_element().num_sub_elements == 0. This is not the case for TensorProductElements (where num_sub_elements is the number of elements in the product) so the original object is not returned and it fails a test further down.

MFE in firedrake with an extruded mesh:

import firedrake as fd

mesh2D = fd.UnitSquareMesh(4, 4, quadrilateral=True)

mesh1D = fd.UnitIntervalMesh(4)
meshEx = fd.ExtrudedMesh(mesh1D, 4, layer_height=0.25)

V = fd.FunctionSpace(mesh2D, "CG", 1)
Ve = fd.FunctionSpace(meshEx, "CG", 1)

v = fd.Function(V)
ve = fd.Function(Ve)

fd.split(v)   # works                                                                                                                                                                                                                                         
fd.split(ve)  # breaks

Both the split calls work if FEniCS#122 is reverted.

[dham]

Sounds like TensorProductElement is abusing .num_sub_elements. That attribute probably needs to be split into .num_sub_elements and .num_factor_elements.

[JHopeCollins]

Ok. It's breaking some of the CI tests for asQ, would you suggest any workarounds until that is changed? Other than reverting the UFL commit on the CI or just waiting.

[dham]

I don't have an immediate suggestion. I also don't know when this will get fixed or by whom.

[JHopeCollins]

Ok. We have other tests that use ExtrudedMesh that still pass, so we can just xfail the failing ones for now.