In perusing the pipe_analysis outputs of the RC2 reprocessing runs to check that our results are all looking as they should, I noticed that the aperture corrections were comparing differently between reprocessing runs (e.g. w42 vs. w41). While, in principle, this could happen as a result of code changes, this did not appear to be the case as the flux comparisons were all identical. Digging into it, I discovered that the key offsets point to a different place for different patches/visits for all aperture correction fields within the same processing run. I.e. the catalog schemas don't match each other, nor the one defined in the schema files in the repo.

E.g. in /datasets/hsc/repo/rerun/RC/w_2018_42/DM-16095/, the meas cat schema in schema/deepCoadd_meas.fits has:

base_GaussianFlux_apCorr Key<D>(offset=1832, nElements=1)

base_PsfFlux_apCorr Key<D>(offset=1992, nElements=1)

But for two patches, we find:

dataId = {"patch": "8,0", "tract": 9615, "filter":"HSC-I"}

base_GaussianFlux_apCorr Key<D>(offset=2040, nElements=1)

base_PsfFlux_apCorr Key<D>(offset=1992, nElements=1)

dataId = {"patch": "3,7", "tract": 9615, "filter":"HSC-I"}

base_GaussianFlux_apCorr Key<D>(offset=1880, nElements=1)

base_PsfFlux_apCorr Key<D>(offset=1928, nElements=1)

This is most undesirable as it precludes concatenating all patches in one tract into a single catalog (which requires them all to have identical schemas).  The fix may be as simple as looping over a sorted list (as opposed to one where order is not guaranteed) when adding the apCorr keys to the schema.