From: Cody P. <cod...@gm...> - 2012-10-23 19:44:37
|
Great - I'll give it a shot! On Tue, Oct 23, 2012 at 1:41 PM, Roy Stogner <roy...@ic...>wrote: > > On Fri, 14 Sep 2012, Cody Permann wrote: > > Well, it turns out that this issue is rather difficult to find, but >> super easy to replicate! I'm using introduction_ex1 to locate the >> problem. Right now we can't even read a nemesis mesh with the >> current head in pure libMesh code. In optimized mode - it may hang >> your machine as it gobbles up an insane amount of memory very >> quickly, in debug mode. It trips this assert: >> Assertion `par_max_node_id == mesh.max_node_id()' failed. >> > > So here's a sad thing: I've been putting off replicating this for a > month, partly because I've been so swamped, but partly because > distributed mesh failures are always so hard to investigate. > > This one was not. Sorry I didn't get to it right away. > > The attached three-line change ends both the dbg-mode assertion > failures and the opt-mode segfaults for me with both the test meshes > you sent, and I think it's the solution to another ParallelMesh bug > that I'd been struggling to find. Thanks for the nice test cases! > > I wasn't able to replicate any "gobbles up an insane amount of memory" > behavior, though. Try it with your full test suite? There were two > separate bugs at work here, so I wouldn't swear there isn't a third > still hiding. > --- > Roy |