Review complete. I didn't see any major issues, but I left a lot of comments on the PR (as you've already seen). I very much appreciate the new unit tests, and it looks great overall.

I didn't look too strictly at the new matcher implementation code, as I know we're trying to strike a balance between cleaning up what was in hscAstrom and not doing anything that could change its behavior.

I also didn't worry too much about the overall structure and where the astrometry.net code was still being used, as I know there's more restructuring yet to be done.

I implemented the simpler changes in a squashed push. That leaves changing the reference loader to return magnitudes instead of fluxes (which will be fairly disruptive, but is well worth doing), revisiting how dense regions (e.g. M31) are avoided in the matcher, and adding magnitude limits to the loader (which may be postponed for another ticket). These changes will be separate commits.

I have pushed more changes to afw, ip_diffIm, meas_algorithms, meas_astrom and obs_sdss, all on tickets/DM-1576, including:

• afw's Config.h has convenience functions that compute flux in Jy to and from AB magnitudes, and using those cleaned up a lot of code
• I moved the abstract base class LoadReferenceObjectsTask to meas_astrom, which is where it belongs (for now) and so that other code could take advantage of its tools for generating reference catalog schema and retrieving flux fields.
• ip_diffIm needed a few fixes due to the updated reference catalog schema.
• obs_sdss calibrate task and config overrides needed a few tweaks.
• The reference loader returns flux in Jy
• The matcher handles version 0 and version 1 source schemas. This feature will not be needed in a few weeks, when catalogs will be converted as they are read in, but I propose leaving it in for now because some tests and examples have catalogs using the old schema. It will be fairly easy to take out later. Still, if you have a better idea, please let me know. I really don't want to have to wait to merge the code until the catalog unpersistence handles the conversion automatically, but I am a bit nervous about the matcher's "sourceFluxType" config field, since more traditionally we ask the user to provide the full field name. Still...we can switch that field without much trouble, as it is not used extensively.
• obs_sdss's photometry task now works and buildbot runs to completion, though some values have changed slightly, making the final test fail
• the PhotoCal task was updated to handle reference flux in Jy

Known issues:

• The PhotoCal task does not handle flux errors fully correctly; changing reference flux to Jy highlighted the problem (the scale of the reference flux and source flux were so different that the reference flux errors were being ignored). I implemented a workaround that I would like checked. It would be even better to handle the errors fully correctly, but I don't know how to do that.
• the astrometry task's handling of schema is nonstandard in that the _init_ does not take a reference schema and expand it. There are several reasons for this: C++ code computes the schema from scratch and some of the needed information is not available at init time.

I believe I've covered everything that's changed since my last pass, with more comments appearing on the various PRs. Let me know if you think there's something I missed; I didn't look at all of meas_astrom again (instead I just tried to look at the bits of code you'd pointed me at here).

Thank you for your second review. I have merged all the code changes to master.

I still have to update the expected results files in the lsst demo, but don't yet know how to push changes to that. It will probably have to be done in two steps, since only one of the two results files is generated if the first one fails to match.