From: Roy S. <roy...@ic...> - 2009-11-13 19:01:17
|
On Fri, 13 Nov 2009, MARRO MASSIMO wrote: > We attach a short test case. This looks like it's tripping a bug that another user ran into recently, and that we haven't figured out a good fix for yet - the Hilbert curve library that we use to get a partitioning-independent ordering on output files gave him a collision which caused libMesh to give two nodes the same number on highly refined meshes. Would you try commenting out the calls to globally_renumber_nodes_and_elements, around lines 292 and 415 of equation_systems_io.C, and see if this fixes the problem for you? Alternatively, configuring with --disable-libHilbert should turn this functionality off. I'm going to change the libHilbert integration to be off by default in the SVN head, whether it's the only problem you've encountered or not. We should have done that already; getting a partitioning-independent node numbering 99.99% of the time isn't worth getting a corrupt data file 0.01% of the time. A suggestion if you run into further problems: try building and running with METHOD=dbg. That won't cause the problem to go away, but it will often cause the problem to trigger earlier, with a more informative error when an assertion fails. Providing a full test case is just as helpful, but was more work on your part, and the less work our users have to do to report bugs, the better. --- Roy |