[fitswcs] double, double, toil and trouble

[Rob Seaman]

> 3.1.4. Higher Precision in Binary Tables:  Doublet Vectors
> In binary tables one may use vector pairs. The time column in
> such a table shall contain a vector of two doubles where the first
> component of the doublet contains the integer portion of the time
> value and the second one the fractional part.

This is confusing two issues, the precision required with the desired  
representation.  Two doubles (or a single quad floating point) may be  
sufficient to a particular purpose - they are not necessary.  Similar  
precision can be obtained by the prudent use of integer or fixed point  
types.

The workflow logistics of various use cases may benefit from avoiding  
floating point.  For instance, integer types compress much better.  In  
general a time series might "want" to be expressed as delta times  
rather than quasi-absolute times.  (As the document points out, all  
timekeeping is relative.)

My concern is just that such usage as described in section pi (3.1.4)  
ought be permitted by FITS, not required.

Rob