[Yt-dev] RMS mass overdensity method
Matthew Turk
matthewturk at gmail.com
Wed Mar 4 11:00:53 PST 2009
> I've been working on trying to narrow down the focus, and it seems that cutting a sphere out many times causes problems. Below, between lines 67 and 73 you can see that I've de-randomized the center, so the sphere is the same every time. This eliminates potential problems of reading data outside each sub-box. I am also not retrieving ParticleMassMsun at all. It still crashes at variable points, one time similar to a standard run, one time a bit further.
Okay, so if it's a problem with cutting out a sphere too many times,
we should be able to replicate it with something like:
for i in xrange(1000):
print i
sp = pf.h.sphere([0.5, 0.5, 0.5], 1.0)
and that should kill it, right?
I think it would really help if we could eliminate all of the comments
and unused code and just strip this down into absolutely the smallest
thing that crashes. I am now getting increasingly concerned this is a
bug in the broader YT codebase that you are the first person to run
into, but we need to be able to verify this.
> Do you have ideas of what I should try next? Is there anywhere I can put some prints that would help us debug this? Should I switch tatics and instead extract the particle data out and use a KDTree?
Don't use the KDTree, it will be slower and will require more
long-term persistent loading of data. :) For now, let's strip this
down to the 100% barest minimum. I have RD0035 of the L7 run, so if
we can get this as tiny as possible I can run it on our machines here.
I think that is our tactic. Remove all comments, just the smallest
thing you think should crash.
-Matt
More information about the yt-dev
mailing list