Shouldn't these datasets be defined in obs_base rather than obs_subaru, so we don't have to manually add them to every obs package?

They already are defined in obs_base. But, as usual, the templates are instrument-specific.

Ah, the templates are instrument-specific because they depend on the visit. I was thinking they're like coadds, but they're not.

Hi Paul Price, since you seem to be responsible for a lot of the other entries in the mapper file, could you take a look and see if my additions are reasonable? Thanks!

The existing templates (e.g., for deepCoadd_tempExp_diffsrc) suggest that the first element of the path should be deepCoadd-diff instead of deepDiff.
Why isn't the tract part of the path? Aren't you using coadds as templates, in which case you need to disambiguate between different (but overlapping) templates?
deepDiff_metadata needs a leading element in the filename, because we should never rely on merely the directory structure to identify a file. Along the same lines, deepDiff_matchedExp's leading filename element matches one that already exists.

These are (AFAIK) difference images, not coadds. So tract would not make sense (and is not used in e.g. the corresponding DECam templates). I don't know if that affects your answer for deepCoadd-diff, but I was under the impression that they were unrelated datasets.

Can you explain what you mean by leading element? For example, there's:

processCcd_metadata:

    template: '%(pointing)05d/%(filter)s/processCcd_metadata/%(visit)07d-%(ccd)03d.boost'

isr_metadata:

    template: '%(pointing)05d/%(filter)s/isr_metadata/%(visit)07d-%(ccd)03d.boost'

What is the template used for the difference images? If it's a coadd, then tract needs to be in the template for the difference image.

By "leading element of the filename", I mean the part after the directory structure, but before the standard %(visit)07d-%(ccd)03d. The two examples you give above do not have any leading element, which I don't think is good (but they are existing and therefore harder to change than something being added now).

What is the template used for the difference images? If it's a coadd, then tract needs to be in the template for the difference image.

At the very least, it can be either a coadd or a calexp. There may be other modes as well.

Then that would need to have multiple datasets, because those have different requirements on the filename template.

I am not aware of any requirements (or lack thereof) on the output of ImageDifferenceTask, but reworking it seems a bit out of scope for this ticket. What do you suggest for the time being?

Well, I guess we just follow the example of obs_decam and proceed with the broken option of not including the tract, and hope this gets fixed.

I've added or updated the leading elements as requested, but not changed the name of the top-level directory. (I still don't know what a deepCoadd_*diffsrc is, but it seems to be HSC-specific and therefore not related to a generic task like ImageDifferenceTask.)

Thanks for the review!