Let make_links_to_me_local handle fake neighbors #2362
Conversation
|
Assuming this passes the standard CI I'm going to want to do some DistributedMesh sweeps before we merge. |
|
@arovinelli, what was the answer to "Wait, are you planning to move these element pairs apart after displacement?" If it's "yes" then I might as well just cancel this PR; we'll definitely have to just communicate more topology data if we can't use nodal ids or nodal coordinates to reconstruct it. |
|
@roystgnr sorry I didn’t noticed this. Yes the plan is to torn it apart, which again, already works for replicated cases
…Sent from my iPhone
On Dec 6, 2019, at 1:13 PM, roystgnr <[email protected]> wrote:
@arovinelli<https://github.com/arovinelli>, what was the answer to "Wait, are you planning to move these element pairs apart after displacement?" If it's "yes" then I might as well just cancel this PR; we'll definitely have to just communicate more topology data if we can't use nodal ids or nodal coordinates to reconstruct it.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub<#2362?email_source=notifications&email_token=AIIMUU7SB5EFBNKZMJOQDB3QXKP37A5CNFSM4JW6HRVKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEGFCF5Y#issuecomment-562701047>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AIIMUU6KILM6GOFHDFYAWP3QXKP37ANCNFSM4JW6HRVA>.
|
@roystgnr thinking on node id lines, what might help here is a map with fake-neighbor elem,side pairs. Or, could we subclass Node or add a method to keep track of intially coincident nodes and compare by such list in the test?? |
|
Okay, I'll see what else I can do then. |
|
I feel like this might have been the right solution since we always have an undisplaced reference mesh in every MOOSE calculation (and this work was done for MOOSE) |
|
Also if it is related, for adaptive meshing we always undisplace the displaced mesh beforehand. This was done way back in 2014 by @jwpeterson in idaholab/moose@f9561331 |
Compare with centroids instead of global node ids. This is extremely useful when comparing side elements that are coincident but may not actually share the same nodes. See idaholab/moose#12033 for application and libMesh#2362 for more inspiration. This probably won't work but let's give it a shot
This fixes the test case in idaholab/moose#14456 for me, although as discussed in that ticket there are still a few other parts of libMesh that I'd expect fake neighbor links to break.