You may be interested in the pattern Jim Bosch used in DM-17048 for doing something similar with ref_cats.

The first few commits, at least.  There was a lot of extra stuff for ref_cat that shouldn't be necessary for brightObjectMask.  Maybe we should pair-program this?

Feel free to assign to yourself.  I see you filed a ticket right after I did!

It turns out that the problem is a bit more interesting than I thought, and I'm inclined to fix it by changing how the regular Gen2 ci_hsc works a bit.

The issue is that the brightObjectMasks are in the repo at $CI_HSC_DIR/DATA, and they're defined with a (tract, patch) data ID.  But there's actually no skymap defined in that repo - the skymap is added to the $CI_HSC_DIR/DATA/rerun/ci_hsc repo.  So the gen2convert is quite reasonably warning that it can't identify the skymap that defines the tract and patch data ID for brightObjectMask, and then skipping them.

This could be fixed by either defining the skymap directly in the root repo or putting the brightObjectMasks into the rerun.  I'm leaning towards the latter; please shout if you have an opinion.

We should also keep this in mind when it comes time to run gen2convert on the real repos on lsst-dev or tiger; I bet they suffer from the same problem.

Changed my mind: I'm now leaning towards putting the SkyMap in the root repo, as that will let us inherit both in Gen3-only processing later instead of having to manually transfer them from the Gen2 output repo.

Changed my mind again: makeSkyMap.py won't run unless you give it an explicit output repo, so the brightObjectMasks are going in the ci_hsc rerun.

Anyhow, in addition to those changes I've added a new Formatter class to pipe_tasks (next to the definition of the ObjectMaskCatalog class) to support reading these without lying about them being FITS the way we did in Gen2.

Tim Jenness, could you look at the daf_butler changes?  It's some straightforward config additions, along with the removal of a bad validity check as we discussed briefly on Slack.

Yusra AlSayyad, could you look at the pipe_tasks and ci_hsc changes?

Looks good. Confirmed it works (left= mask before, right= mask after)

I put two commits on pipe_tasks u/yusra/DM-17300 for you to cherry-pick: https://github.com/lsst/pipe_tasks/tree/u/yusra/DM-17300  The metadata-check commit is optional.  More info on that:

brightObjectMasks.table.getMetadata() (which in this case is ‘patch’: ‘5,4’, ‘tract’: 0}}doesn’t match the dataID which is {{'abstract_filter': 'r', 'skymap': 'ci_hsc', 'tract': 0, 'patch': 69 The code that checks the header: https://github.com/lsst/pipe_tasks/commit/a6db3fba5ff6b925e49eb8b8039fb54bb818a66c#diff-f9cf63fcf2aed29eec448bae6d243b5fR595

The headers look like:

[yusra@lsst-dev01 0]$ more BrightObjectMask-0-5,4-HSC-R.reg 

# BRIGHT STAR CATALOG: Jean Coupon

# GENERATED ON: 2017-06-04

# TRACT: 0

# PATCH: 5,4

# FILTER HSC-I

wcs; fk5

circle(320.735301,-0.407181,0.003015d) # ID: 2686366082164587136, mag: 16.787139

box(320.735301,-0.407181,0.000905d,0.009046d,0) # ID: 2686366082164587136, mag: 16.78713 

and it prints this warning:

compareWarpAssembleCoadd WARN: Expected to see abstract_filter in metadata

compareWarpAssembleCoadd WARN: Expected to see skymap in metadata

compareWarpAssembleCoadd WARN: Expected to see patch == 5,4 in metadata, saw 69

We could either:
(a) adapt the lines that check it so that it sees this as match
(b) remove the lines that check it (use the second commit)
(c) just leave it printing the warnings

Adapting the data IDs is tricky, because looking up the (camera-specific) translator for a data ID is hard to do in this context.

I think long term I'd like to move this check to whatever is responsible for ingesting these into the data repo (right now that's just gen2convert, in which it is possible to get at those translators).  Note that in Gen3 that can't just be moving the files into position, so there will have to be some kind of script that can do this.  But I don't think we need to do that now, and I don't have a strong preference between (b) and (c) in the meantime.

Merged, with (as discussed OOB) the warnings removed.