WRT DecamMapper.paf,
the call: butler.get(flat, dataid(visit=n, ccdnum=m)) will look up data in the
'flat' dataset type. In the case where the information is not available, need
to consult the 'reference' dataset type, and look it up there, like so:
1. dataid is used to generate a location using the raw dataset type. (because it
doesn't exist in flat, and we wish policy:reference was "raw", but actually it's
raw_visit which is a subset of the raw table that is optimized to go faster but
is different.
2. generate the location, it should be a fits file. in this case it has a bracket
extension. call lsst.afw.image.readmetadata (or similar in astroypy) on the
location (with brackets or not) and it will get the right fits header.
3. depending on the camera, the header may need to be converted (numeric to string
or vice versa). so we need a hook for the mapper subclass to be able to do
something (transform) the header value.
4. then it gets fed into the existing calibrationMapping subclass that does the
query.
CalibrationMapping gets the data via the registry (that I need to get via
metadata) in lookup, at line 348:
lookups = self.refRegistry.executeQuery(columns, self.reference,
where, None, values)
it passes it to Mapping.lookup and THAT uses the validity lookup - using
self.obsTimeName and self.range which is set in the CalibrationMapping ctor,
based on the policy.
to create the test:
1. modify the policy to look more like the obs_decam policy:
subpolicy needs template, validTime, etc
policy would ask for the MJD-OBS - the julian date since (a very long time ago),
it's modified because it's had a (fixed, known) number subtracted from it.
or, try using https://github.com/lsst/obs_cfht

Robert Lupton says in HipChat that we should start with just the "standard" TIME-OBS header (which is actually a combination of DATE-OBS and TIME-OBS only if present) to do calibration rendezvous, adding capabilities for additional field names later.

See http://heasarc.gsfc.nasa.gov/docs/fcg/common_dict.html

There is a standard (without quotes) for time headers: http://dx.doi.org/10.1051/0004-6361/201424653

That suggests DATE-BEG and DATE-END (or MJD equivalents) and many other keywords. I think what we really need is an internal metadata standard and a header translation infrastructure (see e.g. https://github.com/Starlink/perl-Astro-FITS-HdrTrans)

I'm going to go through all the butler/daf_persistence tickets when daf_butler takes over.

Gen2 is dead.