Inconsistent behaviour of mesh refinement in serial

[Chris Richardson]

When uniformly refining in serial, there is no choice of algorithm. Should be possible to choose e.g. plaza algorithm if desired.

[Chris Richardson]

Fixed in master

[Martin Sandve Alnæs]

Removing milestone: 1.6 (automated comment)

[Bartlomiej Siudeja]

I am sorry to revive an old solved issue, but I wanted to make sure this is what I encountered in 1.5 and 1.6.

When refining a mesh without any markers in 1.4 the midpoint splitting was used (as in UniformMeshRefinement.cpp). Starting with 1.5, it seems that the long edge refinement is always applied, with or without markers.

Assuming this was the issue I described, which official version of FEniCS will include the fix?

[Chris Richardson]

Yes, so now it is consistent. What needs fixing?

[Bartlomiej Siudeja]

By consistent, do you mean that when no markers are given then midpoint splitting happens (uniform)?

If so, then there is nothing to fix, except that the fix was not included in Mac 1.6 version when I installed it. Should I reinstall? Or wait for 1.7?

[Chris Richardson]

Consistent means same behaviour in serial and parallel... what do you mean by midpoint splitting? All refinement is done by midpoint splitting of edges, anyway.

[Bartlomiej Siudeja]

I should have explained it better. I meant all sides split in the middle to get 4 similar triangles, as opposed to just one side split for long edge refinement.