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

[THIERRY FORVEILLE]

"Frank Valdes" <valdes at noao.edu> writes
> Come on (!), talking about alternate formats is not helpful.  DECam is
> generating a huge amount of data (arguably the biggest optical data source
> in the world) on an almost daily basis.  The raw data is an MEF and uses
> INHERIT.  The calibrated data produces MEF with INHERIT.  All mosaic data
> that NOAO has in its archive, -- >10 yrs of Mosaic, > 5 yrs of NEWFIRM,
> DECAM -- is in this format.  
I don't think anybody is suggesting that the standard be altered to somehow
forbid INHERIT, or that the convention be deregistered. Reserving the keyword
so that it's not inadvertently used for something else also seems quite 
reasonable to me.

The issue on the table is whether it should become part of the standard, which 
represents a seal of approval and not merely an acknowledgement that it exists 
in the wild. Both the flaws that have been pointed out and the existence of an
alternate solution are important negative factors in that decision.

> I echo that binary tables seems more complex to
> many people and I really agree with Bill T. that people look at the primary
> header first and I've found it very useful (and used in the pipeline) to
> just check the primary header for some information.
> 
Perceptions of complexity only count on my slate if they reflect actual
complexity. On at least one significant aspect binary tables are actually 
simpler/more powerful, in that one can obtain the location in the file of 
any desired pixel after decoding just the binary table header, rather than 
have to recursively go through 35 extensions (or at least their headers) 
to find a pixel in extension 36, but overall I'd say that the two solutions 
have quite similar complexities.

> The only complications I've encountered is people that have software that
> doesn't support INHERIT.  Making an image cube for a mosaic would be a new
> convention which almost no software (e.g. DS9) would not support.
> 
The only complications I've encountered is people that have software that
doesn't support binary tables ;-) Joke aside, support in important software
like DS9 is a valid consideration, but one that should be weighted against the
downsides.