Note to self: What about using the EDGE mask plane for this check?

Now that I finally have some useful string output for the psfCandidates and SpatialCells and an understanding of the code, it looks like makePsfCandidates can catch some of these, because it similarly calls psfCandidate.getMaskedImage(). It doesn't always catch them, because it uses the default stamp size, whereas the code block described on this ticket uses the psfex image kernel size.

Since this doesn't seem to happen very often, I propose adding a `log.warning` to the exception block in psfexPsfDeterminer describing the failing candidate and expected stamp size, and leaving it as a try:catch. We're catching a specific exception here, so I'm not sure that replacing said exception with a check on the stamp bbox gains us anything.

Note that this argument is against what I argue in my comment on DM-27535:

Jim Bosch: do you mind reviewing these three small (~100 lines total) PRs? I did not make the change as requested here; see my comment above for my reasoning. I could still do that, but I think letting the psfCandidate handle bounds checking is more robust.

In the end, I did add useful stringification of some otherwise opaque types, so it's not a complete waste. Those str outputs should make the error messages here clearer.

afw: https://github.com/lsst/afw/pull/680
meas_algorithms: https://github.com/lsst/meas_algorithms/pull/307
meas_extensions_psfex: https://github.com/lsst/meas_extensions_psfex/pull/67

Jenkins (with ci_hsc and ci_imsim): https://ci.lsst.codes/blue/organizations/jenkins/stack-os-matrix/detail/stack-os-matrix/38074/pipeline

Additions look good, but I'm not sure the error message is actually better - maybe I'm just not familiar enough with what's wrong with the current one. A few minor comments on the PRs.

One of the things that confused me when digging into all of this is that I could not find any examples of the makePsfCandidates version of the warning, but I did find a half dozen examples of the psfex warning (the subject of this ticket) in an rc2_subset run. I believe this is because the stamp size we use in MakePsfCandidates is smaller than the one in PsfexPsfDeterminer, so a candidate can get through the first (when close to an edge) but fail the second.

That sounds related to an RFC Arun Kannawadi filed back on config fields for PSF sizes, so he may have a useful opinion on what we should do about it (if anything).

Post cleanup Jenkins run: https://ci.lsst.codes/blue/organizations/jenkins/stack-os-matrix/detail/stack-os-matrix/38097/pipeline