Fix Version/s: None
Sprint:AP F19-5 (October), AP F19-6 (November), AP S20-1 (December), AP S20-2 (January)
Ian Dell'Antonio (appears not to be on Jira) described a jointcal bug on slack where all of the on-sky residuals after initialization were NaN, but the tangent plane values were reasonable. The existing debug information jointcal can provide was not enough to readily narrow down the source of this; there may be something pathologically wrong with the initial model. To help debug such problems, we need a way to write out the initial model to disk. A writeInitialModel option in the jointcal config that just dumps text files for each model component would be a good start: I think all of the AstrometryTransforms have useful string representations.
- is triggering
DM-22548 jointcal astrometry fails at initialization with NaN on some DECam data
- Won't Fix
Thank you for the mostly atomic commits; they helped a lot. I indicated a few places where the commits don't seem quite atomic.
Thanks for the very thorough review. I've cleaned it up quite a bit and the output does look better now and is better tested.
Jenkins run just to be sure: https://ci.lsst.codes/blue/organizations/jenkins/stack-os-matrix/detail/stack-os-matrix/31024/pipeline
Krzysztof Findeisen: do you mind taking on this large review? The commits should be completely atomic, which might make it more tractable.
I don't have any tests of the actual stringification: the string representations can be rather large, especially of the Models. Suggestions about how to best handle that are very welcome.