[Users] Meeting Minutes for 2015-02-09

[Matthew Turk]

Hi Roland,

On Mon, Feb 9, 2015 at 10:26 AM, Roland Haas
<roland.haas at physics.gatech.edu> wrote:
> Hello Matt,
>
>> Oh, rats, I didn't know this was on the agenda or I'd have called
>> in. Please do let me know what I can do to help out.  The last time
>> I looked at the code I'd written, it was parsing many but not all
>> cases of the datasets correctly; IIRC, Steve and Frank had it mostly
>> working with some issues at coincident processor/level boundaries?  I
>> confess I don't know where Sherwood's project was left.
> No issue. This was not listed on the official agenda since it only came
> up again while talking with someone locally today (before the phone
> call). Your code makes three attempts then? LSU, Caltech and you?
>

I think both of the other two were more advanced than what I had done,
and may have used bits of my code as a starting point.  (Although that
may not be the case.)

>> What would be the best thing for me to try to do to support the
>> efforts?
> I think most interesting (at least for me who was not involved) would be
> to see hear what the current status is and to maybe see the current code
> in case someone wants to give it another try out of shear frustration
> with other visualization software.

Ah, sure.  :)  So if I remember correctly, the biggest issue was
simply reliably and consistently identifying the full hierarchy of the
mesh.  I tried two different tactics, the first of which was parsing
the string literal saved in an HDF5 file, and the other was parsing
the hdf5 attributes for each node.  Where this became tricky was with
boundary conditions, with higher-refinement grids abutting across
processor boundaries, and with grids partially passing into coarser
grids.  I think it's all feasible and tractable, there are just a
number of edge cases that made it tricky.  For what it's worth, in the
intervening time since the last attempt last Fall, yt has gained much
better support for meshes that caused it trouble before, such as
overlapping, mixed-ratio refinement, and so on.
 *

>
> Yours,
> Roland
>