Non-compliance with UFC entity ordering scheme

[Jan Blechta]

UFC requires ordering of subentities of dimension d' of any entity of dimension d with d'<d as described in chapter 5 of UFC manual. Simply said, global indices of the subentities must be increasing. DOLFIN does not respect this except when d' == 0. This is demonstrated by deleting if d1 != 0: continue block in the test.

[Prof Garth Wells]

I'm a bit confused by this report.

• Has this been fixed, or not, with recent topology computation changes?
• The UFC specification was no well considered with respect to global vs local (process-wise) numbering. It is generally more convenient to work with local indices.

[Jan Blechta]

Sorry for being to terse before. It seems to me that DOLFIN (before 2016.1.0 and after the recent fix c9bad0e2065a8ae1c7ae5c9c7093fc6e985f4d29) respects UFC rule of increasing indexing for d-0 connectivity with global (process-wise) indices for any d. On the other hand the same rule for d-d' connectivity with d > d' > 0 is not, and probably never was, respected. This is all demonstrated in the test https://bitbucket.org/fenics-project/dolfin/src/2570a75ba20ce79b6fd5142e000dc0d4b0148465/test/unit/python/mesh/test_mesh.py?at=master&fileviewer=file-view-default#test_mesh.py-579.

To be honest I am not even sure whether such a numbering is guaranteed to exist. I am just throwing this for a discussion.

[Jan Blechta]

Shortly,

• no, this has not been fixed, except of special case when d' == 0,
• ok, I can imagine that the complete UFC scheme is not practical and d' == 0 case is sufficient.