[yt-dev] Should support for octree data in the AMRKDTree be a blocker for 3.3?

Tue May 3 15:51:23 PDT 2016

I don't believe I can take this project on right now to make the VR
interface work with particle-based and/or octree-based frontends.  I don't
usually use these frontends, so I can't really justify it to myself to put
a bunch of time in on it when there is little research benefit for me, and
I've already got my plate full.  This may change in the future.  Perhaps
one of the representatives of one of the affected frontends might be
interested in taking this on to assure VR will work with their data?

Cameron

On Mon, May 2, 2016 at 5:33 PM, Matthew Turk <matthewturk at gmail.com> wrote:

> Hi Cameron,
>
> On Mon, May 2, 2016 at 7:21 PM, Cameron Hummels <chummels at gmail.com>
> wrote:
>
>> I wasn't entirely objecting to punting it, but I think it warrants
>> discussion here.  Until I tried to build some Off axis projections and VRs
>> for SPH data, I didn't realize that they weren't supported.  In fact, which
>> frontends are actually supported for doing VRs and OffAxisProjections?  If
>> Octtree-based code isn't, then that knocks out ART, ARTIO, and Ramses, and
>> without SPH, it knocks out Gadget, Tipsy, Gizmo, and Gasoline, which leaves
>> Enzo, Flash, and Athena?  Others?
>>
>> I honestly don't have any idea how much work it would take to construct a
>> KDTree from octree data.  I figured this was in the realm of CS and would
>> probably have been solved elsewhere, but maybe not.
>>
>
> Yes, it absolutely has.  When you asked me about this over private message
> or public message on Slack a while back (I forget which) I sent back
> information about how to do this.  I think it would be relatively
> straightforward to implement.  You need to insert a couple "fake" levels in
> the octree so that it is a BSP rather than the way it functions now, but
> these cutting planes can be inserted at the existing cell refinement
> divisions.  It would also be straightforward to implement a ray traversal
> that natively worked with Octrees, which would only require mocking
> PartitionedGrid objects and sorting the octs in the right order for the
> traversal.  (This would result in an identical operation, I think.)
>
> My understanding is that when over_refine_factor > 1, it works for
> particle data sets.  Desika had success with this.  I may no longer be
> right.
>
> Even without any new development, I have not heard of anyone attempting to
> debug why and where the segfault occurs.  My suspicion is that it's a
> fencepost error that shows up because of the binary division in the
> octree.  That would be a good first place to look.
>
>
>>
>> Anyway, I'm not trying to be too obstructionist here, just trying to have
>> a discussion, since there were at least 4 other issues filed regarding this
>> in the past, so it seems relevant for a lot of users.
>>
>
> I think this would be a great use of resources, but I don't think it
> should block 3.3.  I do not know when I would be able to devote the time to
> it; if someone else is willing to volunteer that time, I think it would be
> great.  Would you be willing to put the time in to ensure that it could
> happen?  If so, I think making it a blocker on 3.3 and setting a deadline
> for it would be a good compromise; if it isn't done by that deadline, punt
> until the next release.
>
> -Matt
>
>
>>
>> Cameron
>>
>> On Mon, May 2, 2016 at 4:26 PM, Nathan Goldbaum <nathan12343 at gmail.com>
>> wrote:
>>
>>> Hi all,
>>>
>>> I'm moving some discussion from the comments in issue #1212 over to the
>>> mailing list:
>>>
>>> https://bitbucket.org/yt_analysis/yt/issues/1212
>>>
>>> I wanted to check with the list since Cameron objected a bit in the
>>> issue discussion.
>>>
>>> The issue Cameron identifies is pretty serious, but fully fixing it will
>>> require pretty substantial work - adding support for octree data to the
>>> AMRKDTree.
>>>
>>> I've gone ahead and marked it for version 3.4 - i.e. that it shouldn't
>>> be a blocker for the 3.3 release. I've also created issue 1215:
>>>
>>> https://bitbucket.org/yt_analysis/yt/issues/1215
>>>
>>> which will be fixed when yt produces a useful error message rather than
>>> seg faulting when trying to construct an AMRKDTree for unsupported data.
>>>
>>> I think that requiring someone to add support for octree data to the
>>> AMRKDTree before we release 3.3 is effectively delaying the 3.3 release
>>> arbitrarily far into the future, since no one is currently willing to work
>>> on that project. That would be unfortunate, because there's lots of cool
>>> stuff people have worked on that deserves a stable release.
>>>
>>> -Nathan
>>>
>>> _______________________________________________
>>> yt-dev mailing list
>>> yt-dev at lists.spacepope.org
>>> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>>>
>>>
>>
>>
>> --
>> Cameron Hummels
>> NSF Postdoctoral Fellow
>> Department of Astronomy
>> California Institute of Technology
>> http://chummels.org
>>
>> _______________________________________________
>> yt-dev mailing list
>> yt-dev at lists.spacepope.org
>> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>>
>>
>
> _______________________________________________
> yt-dev mailing list
> yt-dev at lists.spacepope.org
> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>
>

-- 
Cameron Hummels
NSF Postdoctoral Fellow
Department of Astronomy
California Institute of Technology
http://chummels.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.spacepope.org/pipermail/yt-dev-spacepope.org/attachments/20160503/15e9c263/attachment.htm>