On Thu, Aug 2, 2012 at 5:37 PM, Roy Stogner <email@example.com>
After the fixes in r5916 I can no longer get any bugs to crop up;
On Thu, 2 Aug 2012, Roy Stogner wrote:
On Thu, 2 Aug 2012, Cody Permann wrote:
I'll see what I can do. That rectangle example is actually the most
complicated one in the bunch! it is using subdomain restricted
terms, but I think the problem shows up even in simpler examples.
I've managed to duplicate the problem with adaptivity_ex5; a simpler
example would be greatly preferable but I think I can get a handle on
it from this if necessary.
would you see if that works for your regression tests as well?
Sure thing - I'll check and get back to you!
If not, feel free to revert to before the broken code for now;
"periodic + threading is horribly broken" would still be vastly
preferable to "periodic is subtlely broken". But if the new code is
still broken, I'd definitely still appreciate if you could come up
with a simple test case triggering the breakage.
So I tried a bit today and I really had a tough time producing the problem with square domains. The problem showed up quite a bit more with more interesting transformations. The one rectangle I sent this morning was it, but even that was solving for two variables on overlapping subdomains (don't know why that mattered). If I can come up with something simpler, I will get it to you.
Also: how helpful would it be for you guys if I find time to finish
fixing the threading+periodic combination?
This would be immensely helpful! One area that MOOSE is being used for extensively is Meso-scale modeling. Many of these sims are run on periodic domains and at this point, we have lots of physics active on those domains. The ability to run threaded+periodic bcs reliably, will be of extreme importance as we scale up these problems to run on the clusters.
My motivation here was
merely "there's an all-too-common race condition that's screwing up
our buildbot results", and since I'm swamped with work right now I
left the fix in a state of "there's a race condition that I can
see in theory and that might be risky in 3D but that never seems to
trigger in 2D tests". But 90% of the work is done and the other 10%
would be higher up on my priorities if I knew it would be put to use.
I can't stress how important this is for us, if we can support you in any way in finishing this up, we'll make that happen.
And as an aside: what's this about "when MOOSE goes open-source"? Any
solid time frame there?
Perhaps sooner than later? It's a big step for the Idaho National Laboratory. Our lab has never produced an Open-Source package before, so our legal and licensing people are fairly green in this area. We are working hard to make it a reality.