Comments on individual proposals:
1) If you deprecate kernelSize in the base task, do the child tasks still have that field, with their own definitions of it, or is it supposed to be replaced with the `stampSize` defined in 4)?
2) Did you mean `makePsfCandidates`? And by redefine, do you mean "change the docsstring"?
3) Does this also include modifying psfex and piff to use `samplingSize` in the same way?
4) You refer to `kernelSize` here, but you deprecated it in 1)?
5) I assume pca and psfex both keep these min/max configs because they both do use them to check the calculated kernelSize?

Hey John,

1) kernelSize is to be deprecated from BasePsfDeterminer and all its children tasks. Some PSF determiners implicitly redefine kernelSize, which will be replaced with stampSize as you mentioned.

2) Oops, sorry, I meant makePsfCandidate (now changed in the original proposal above). By redefine, I mean changing the docstring and ensuring that the usage of the config is in accordance with the docstring.

3) Yes

4) makePsfCandidates will continue to have kernelSize. Only its namesake in BasePsfDeterminer that I propose to deprecate.

5) Not in practice. The config values that are set at the instrument level, and by default in the algorithm do not trigger that behavior where `kernelSizeMin` and `kernelSizeMax` are used. We might essentially consider them as unused and remove those parts altogether.

This all sounds reasonable to me, and a clear improvement on the status quo. Seems like deprecating those min/max configs John asked about is worth doing at the same time.

Arun Kannawadi are you intending to adopt this RFC or withdraw it?

Adopting the RFC with DM-36071

Thoughts that came up while implementing this RFC in DM-36071:

3) Does it make sense to promote `samplingSize` parameter to the Base class, given that PCA PSF determiner does not use this? Or does it make sense because this will ensure consistency between PSFEx and Piff PSF determiners that do use `samplingSize`?

4) Because the PSF candidates (in particular, their bounding boxes) are global, modifying them in one part of the pipeline changes them for others as well. Piff and PCA Psfs do this. Either, these algorithms should be modified to make a copy internally, or `stampSize` should become a mandatory config field. The latter is not possible (by design) since until DM-36311, when kernelSize is completely removed (1). So making an internal copy seems to me the best way forward.

I don't have a very strong opinion on (3), but I lean towards keeping it out of the base class; it's not something anyone downstream of the PSF determiner should depend on, and I can envision an PSF determiner implementation that might have multiple samplingSize parameters for different aspects of the model (perhaps even a Piff-based one) for which a single config option might be awkward.

For (4), sounds like copy does seem like the best approach, and while I doubt those copies will matter in the big picture, let's add a note to DM-36311 to avoid them there if still possible.

I also vote to keep samplingSize out of the base class to avoid designing ourselves into a corner in the future.  We should definitely still make it consistent in the various classes where it is used though.