On Wed, 23 Mar 2011, Andrea Hawkins-Daarud wrote:
> It appears that this topic came up before, but I did not see a
> resolution. Anyway, I have a few mesh and solution xda files that I
> created back in (about) August that do not seem to read in correctly.
> I don't know what version of libMesh I was using then, but the headers
> of these files say libMesh-0.7.0+ . I just checked out the most
> recent version r4284 and am still seeing the problem.
> It appears that the mesh file is being read correctly but that the
> solution file is not matching up to the element numbering or
> Is there a backward compatibility issue?
Search through the svn logs for libHilbert, which we use to get
partitioning-independent node numbering in mesh files. There was a
bug in our earlier interface to libHilbert which was actually creating
corrupted outputs in some cases. IIRC we switched the renumbering off
to avoid triggering the bug, then switched it back on once the bug was
tracked down and fixed. If you've got files that read fine with an
older libMesh version but get scrambled today, they were probably
written with libHilbert disabled. If you've got files that don't read
fine with any libMesh version, you probably got hit by the bug...
although I think in this case the mesh doesn't read correctly either
so you're probably in the previous case.
I don't know if we came up with a way to convert such files... Maybe
comment out the globally_renumber_nodes_and_elements() call in the
EquationSystems read, but leave it enabled to write the data back out?