On 6/1/10 7:42 PM, Roy Stogner wrote: > > On Tue, 1 Jun 2010, Boyce Griffith wrote: > >> On 6/1/10 7:19 PM, Roy Stogner wrote: >>> >>> On Tue, 1 Jun 2010, Boyce Griffith wrote: >>> >>>> I guess there must be some other necessary DOFs in the default >>>> send_list? >>> >>> Not for enforce_constraints_exactly(), there shouldn't... >> >> Well, using the full send list seems to work correctly, at least in >> the sense that the assertion does not fail. > > What's the full stack trace for when it does fail?
I'm now away from the system where I've been running these tests, and it is not easily accessed from outside of the firewall. I will let you know tomorrow... > I'm looking for where the problem could be, but not seeing it. If > your code wasn't patterned after the enforce_constraints_exactly() > implementation, it might as well have been based on how close they > are, and I'm not sure how one loop could be hitting a dependency index > that the other missed. > >> Only non-standard ghost DOFs --- I am leaving the local DOFs alone (at >> least for now). This is for a problem which uses AMR, but this >> assertion failure occurs before any mesh refinement has taken place, >> so the mesh is a simple 3D uniform grid. > > Hmm... that's simple enough that walking through it in the debugger or > spewing to cout might not be the nightmare I was imagining. At the > very least it might be interesting to see what node(s) hit the > assertion failure. > > But that also makes the problem stranger. The only dofs that should > be in the standard send_list but not in your send_list are dofs on > neighbors and nodal neighbors, none of which should matter for a > problem whose only constraints are between distant periodic sides. > >> Are there methods I could call to try to ensure that all of the >> DofConstraints stuff has been (re-)initialized? > > Not exactly, but you might see what happens when the problem is > simplified even further. If you only use one periodic boundary pair > instead of three, for instance, then there should be no recursive > constraints and we can pin down or rule out a problem there. In fact, there is only one pair of periodic boundaries for this problem geometry. Tomorrow I will try to see if I can figure out which DOFs are missing from the constraint DOF list which are present in the standard send_list. -- Boyce ------------------------------------------------------------------------------ _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
