Enable Quadratic Elements in NetGenMeshInterface

[miaoyinb]

closes #4505

[roystgnr]

Are we supporting TET14 here? There's clearly some partial support (which I'd say leave in even if we don't get to total support yet), but it looks to me like we're only handling mid-edge node movement, not mid-face nodes.

If the answer is "yes", could we add a unit test?

If "no", could we take it out of the if() before that libmesh_not_implemented()?

Otherwise this looks great.

[miaoyinb]

Ah, yes. I missed the face centroid of TRI7 for TET14, will add it.

[moosebuild]

Job Coverage, step Generate coverage on f0a4b59 wanted to post the following:

Coverage

	5112d2	#4506 f0a4b5
	Total	Total	+/-	New
Rate	65.72%	65.74%	+0.01%	100.00%
Hits	78898	78962	+64	71
Misses	41151	41158	+7	0

Diff coverage report

Full coverage report

This comment will be updated on new commits.

[miaoyinb]

Added full TET14 support

[roystgnr]

Come again? I'm fine with merging this as-is, since you're just working around this problem, not creating it, but I'd love to know what's happening exactly and get a reproducer. If there's some conditions under which broadcast() does not result in an identical mesh on all ranks, that's a bug, and I want to get that replicated and fixed, not just worked around!

[miaoyinb]

Thanks, I think we can merge this for now. Here are more details about the issue (testNetGenQuadratic can be a good reproducer at np=2)

• On rank 0, triangulate() deletes the input surface elements (the octahedron's 8 boundary TRIs) and adds the 4 new TETs. delete_elem() on a ReplicatedMesh leaves null holes in the _elements storage vector and does not shrink it, so rank 0 ends up with 4 live elements occupying an id space of size 8 (max_elem_id()=8).
• broadcast() on the receivers does mesh.clear() and then unpacks only the live entities, preserving their ids. So rank 1 gets a freshly-rebuilt, compact mesh (max_elem_id()=4).
• broadcast()'s own postcondition asserts verify(n_elem()) / verify(n_nodes()), which hold on both ranks. It does not assert max_elem_id(), which is why the id-space-size difference slips through.

So the assertion that trips is all_increased_order_range's comm().verify(mesh.max_elem_id())