I think I was assuming that the forced photometry tasks would explicitly set the coord fields from the reference catalog, but it appears that's not the case - reference catalog centroids are now added to the forced catalogs via the base_TransformedCentroid plugins in meas_base, which basically act like a centroiding algorithm that just looks up the coord fields from the reference catalog and copies it into the centroid fields in the forced catalog. It appears that in the past we were actually relying on copyColumns to do that.
So, one way to fix this would be to just add these to copyColumns in the ForcedMeasurementTask config defaults. Another would be to add a forced-measurement version of the base_SkyCoord algorithm in meas_base, which is what we use in single-frame measurement to fill the coord fields. Or it might be best to have base_TransformedCentroid do that itself; I can't think of a reason why it shouldn't. I think any of these is fine - they're all a bit ugly, IMO, and maybe copyColumns is worst just because it can lead to the kinds of problems we saw in DM-8210 if we use it to overwrite existing columns - but the ultimate reason this is difficutl is because I think it's intrinsically tricky (but of course incredibly useful) to have the coord columns at all in what is otherwise a table containing only raw, uncalibrated quantities.
I'm guessing this originates in
DM-8210, where, on 448121a4, the copyColumns config which included coord_ra and coord_dec was removed as a duplicate. We do still have a copyColumns in ForcedMeasurementConfig, but it doesn't mention these fields.Reading
DM-8210, I'm confused about what's actually supposed to be going on here. Jim Bosch, maybe you could clarify the intention?