[LSST|dm-users 510] Possible DM bug

[Andrew Becker]

Hi all - We have been using the LSST stack to reduce CFHT data at UW, 
and have come across a potential bug in the DM software.  I don't see 
how this could be something intrinsic to the data, instead it seems like 
it could be a bug in the software triggered by its application to new data.

I have a how to reproduce at NCSA at /lsst/home/becker/CFHT.  Follow 
commands in the SETUP file to get an up-to-date version of the stack 
(b449) and a modified version of ip_isr (a hack to avoid overscan 
correction) and a private branch of obs_cfht (u/krughoff/ossos_support).

If you run the commands in SOURCE_ME it will run the processCcd on a 
single CCD, in 2 modes, yielding 2 output directories.  The second call 
merely sets the bin sizes to be the same amongst all the various 
background subtractions, but is sufficient to trigger the bug.  A basic 
script included in the directory differences the 2 output calexps (a 
straight -=) and ds9 is used to view the diff.  If you pan to e.g. 
1845,546 you'll see a feature like attached in the difference (out1 on 
the left, out2 in the middle, diff on the right).  This can be seen in 
various other parts of the image, as irregular annuli around bright 
blended objects.  It suggests to me that during measurement some 
substamp manipulations are leaking into the original image.  But you can 
kind of see in the center image that the second run seems to trigger the 
issue.

Let me know if I can clarify at all.  It seems like a general enough 
problem that it probably should be looked at sooner rather than later.

Thanks,
Andy
-------------- next part --------------
A non-text attachment was scrubbed...
Name: diff.png
Type: image/png
Size: 471405 bytes
Desc: not available
URL: <https://lists.lsst.org/pipermail/dm-users/attachments/20141113/6d683044/attachment-0001.png>