Details
-
Type:
Bug
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: afw
-
Labels:None
-
Story Points:4
-
Sprint:Arch 2018-11-05, Arch 2018-11-12
-
Team:Architecture
Description
While reviewing DM-16253 I re-processed DECam data through processCCD to test the metadata fix, but found that the reported wcs of the images was occasionally catastrophically wrong. The bug appears seemingly randomly when the fits header is read to create a wcs:
for _i in range(40):
|
wcs_test = butler_test.get('calexp_wcs', {'visit': 411735, 'ccdnum': 10})
|
print(wcs_test.pixelToSky(-.5,-.5))
|
(137.919516, -1.319873)
|
(nan, nan)
|
(137.919516, -1.319873)
|
(138.379127, +87.701332)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(228.379127, -0.000000)
|
(137.919516, -1.319873)
|
(138.379127, +87.701332)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(138.379127, +87.701332)
|
(137.919516, -1.319873)
|
(228.379127, -0.000000)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(nan, nan)
|
(138.379127, +87.701332)
|
(318.379127, -87.701332)
|
(nan, nan)
|
(228.379127, -0.000000)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(137.919516, -1.319873)
|
(228.379127, -0.000000)
|
(137.919516, -1.319873)
|
(228.379127, -0.000000)
|
(228.379127, -0.000000)
|
(nan, nan)
|
(137.919516, -1.319873)
|
The above values are random each time, and the order is not repeatable.
If the full exposure is read in and the wcs is obtained from exposure.getWcs(), then it is correct each time. Also, the above command when run on data ingested in late January 2018 also produces correct results.
Attachments
Issue Links
- relates to
-
DM-16429 Ensure WCS (and other Exposure components) are retrieved properly when loaded individually
- Done
Activity
This looks great. Two small suggestions on github.
Russell Owen this evening I have had a go at using the PropertyList->FitsChan conversion directly without going through the string. It seems to work fine but it will need another review. If people prefer I merge the reviewed code and then put this improvement on a separate ticket please let me know.
(but I must remember to fix the demo – I assume I just copy in the new versions of the catalogs – it still surprises me that they are changing at all).
Merged. Jenkins passed.
The demo is currently failing on this branch because some of the declinations in the catalog have shifted by a tiny amount (suggesting that we are putting coordinates in catalogs derived at some point from WCS in headers and not the real WCS).