[fitsbits] start of Public Comment Period on the INHERIT convention

[Rob Seaman]

On Jul 1, 2015, at 8:57 AM, Phil Hodge <hodge at stsci.edu> wrote:

> On 07/01/2015 11:44 AM, William Thompson wrote:
> 
>> In principle, one could have a completely empty primary header, with all the metadata being stored only in the binary table header. However, I don't think this would have bought us any friends. 
> 
> If the metadata pertain to the data in the table, the metadata should indeed be in the table header.  Why on earth would you put them in the primary header?

Many instruments have multiple chips, amplifiers or the equivalent and each of these is often put into a separate HDU.  Whether this is a table or image is an implementation detail.  Indeed if an image HDU is tile-compressed it becomes a table HDU itself.  For such multiplex instrumentation, an MEF is a natural representation and then the primary (data-less) header conveys metadata pertinent to the exposure as a whole and each extension header contains metadata specific to that amp or CCD.  One can come up with other use cases that have similar structures.

Given the noted limitations of the current ASCII headers, FITS2 might rather choose to relegate them to only their basic structural role and move the science metadata into one-or-more table extensions that offer new features to support inheritance, long keyword names and values, and all the other desired features.  In this case the metadata and data would be in separate extensions, quite likely both binary tables.  Indeed for some use cases the data and metadata might traverse completely different paths in the workflow and only be packaged back into single files of whatever structure at the end.  Utility tools would permit restructuring the files for different purposes; FPACK is an example of such a tool.

Whether or not we pursue this route there are already files in the wild with primary HDUs that are both data-less and metadata-less, for instance.  Usage as described is already conforming FITS and it is safe to say that if usage is not forbidden (and sometimes if it is) somebody, somewhere will implement it.

Rob