I updated an incorrect doc string and pushed back on the other suggested changes, ungrateful wretch that I am.
For the record the main area of disagreement appears to be whether there should be variants of makeMatchStatisticsInRadians/InPixels that don't take a WCS. I argue that these assume too much about the match list (especially while fitting the WCS, during which the necessary information may be unavailable, or, heaven forbid, stale), and that the existing non-WCS variant that computes statistics based on the distance field of the match is preferred once the match list is fully updated. However, Simon Krughoff has a point; it can be useful to get distance on-sky or in pixels (instead of being stuck with whatever is in the distance field) and the provided functions for that are less efficient than they could be because they rely on a WCS. At this point I relegate such additional functions to a future ticket.
Another issue I have minor concerns about is that each the loops that refine the WCS solution work differently in AstrometryTask (AT) and ANetBasicAstrometryTask (ANBAT):
- ANBAT takes a pure tangent WCS and a match list, and for each iteration it first fits a WCS and then performs a new match based on that WCS.
- AT does the opposite: it doesn't even take a match list, but for each loop it first finds one, then fits a WCS to it.
I spent a bit of time trying the AT approach in ANBAT but it did not work quite as well (perhaps because the initial WCS is pure tangent, ignoring optical distortion). The AT approach seems to work fine so I didn't want to change that task. I concluded it was safer to leave both tasks working as they were.