Sorry for the delay on this. It has taken me a while to digest the comments and resolution on the RFC. I have to admit that I'm not entirely convinced the conclusion was to deprecate ScaleZeroPointTask, but rather adding a (configurable) option to allow it to be disabled. In particular, Yusra AlSayyad's last comment ends with a dangling sentence where there is a suggestion of re-working a "new" ScaleZeroPointTask. This ticket does not seem to make any official move towards deprecation of the task itself, but rather the deprecation of having the option of using the task in the context of assembleCoadd. Some clarification of the fate of the task itself vs. simply disabling the possibility of using it in assembleCoadd (and perhaps being a bit more distinctive in the use of the terms "deprecate" vs. "disable") – both here an on the RFC – would be appreciated.
Regardless, as to the question:
are there other tests you think should be done?
Uh...yup Particularly considering and in the context of one of Jim Bosch's comments on the RFC:
pipeline-output changes tend to be disruptive in unexpected ways
First, simply creating a few coadds using this branch to make sure they look reasonable (simply running ci_hsc and having it pass does not accomplish this). How different is the relative flux scaling of the final coadds (in RE many of Paul's comments)? (As another visual, creating 3-color images using the same scaling for both should result in similar looking images.) Further, I think processing all the way through multiband to make sure we get consistent output catalogs with the current behavior is an important sanity check.
I also wonder if we may see any adverse/unexpected effects in the final scaling of the variance of the warps (I appreciate this algorithm shouldn't be sensitive to any overall scaling, but I again refer back to Jim's point above!)
I would also like to see a coadd created with this branch with the config doScaleZeroPoint set to True to make sure we get bitwise-identical coadds between it and running on current master (or the latest RC2 processing run). Also, along those lines, it seems to me that HSC would still like to continue with the current behavior, in which case this ticket merge should be accompanied by a config override in obs_subaru to set doScaleZeroPoint = True. I'm not 100% confident on this, so perhaps Paul Price/Yusra AlSayyad/Jim Bosch should weigh in here.