[iaufwg] Final wording of 64-bit image support

[Lucio Chiappetti]

On Wed, 26 Apr 2006, William Pence wrote:

> This 6 month period ends in June, so if anyone (especially those that voted
> 'No') have any suggestions for modifying the wording of the proposal that
> might be acceptable to everyone, now is the time to make them.

I can just repeat what I proposed in my e-mail of 12 December 2005, i.e. 
that one should/could add somewhere a "discouragement" statement. If we 
agree on that, the next steps will be :

 - to agree the text of the statement
 - to agree WHERE to put it in the standards document
 - would a formal vote be necessary ?

Concerning the WHERE, I had a look to the material supplied by Bill and to 
the whole 2.1b standard. It seems difficult to insert a discouragement 
statement in one of the present subsections which are more of a 
technically prescriptive nature (specially those in chapter 6 like 6.2.4) 
than usage recommendations. The possibilities seem to me to be : insert a 
new subsection at the end of either chapter 5 or 6, or insert a new 
appendix (personally I feel an appendix is "weaker", and would prefer 
something in the main text, but would not object to any location).

Concerning the text, I report below the proposal I drafted in Dec 2005.
It is rather maximal (it does not concern only BITPIX 64 in primary HDU) 
and verbose (gives examples).

I would not be scandalized at all if the statement inserted in the 
standard is shorter and limited to the "controversial" BITPIX 64 issue. 
Just it seemed appropriate to me to address the issue of "proper" data 
types in full.

   It is recommended to choose the most appropriate data type both for a 
   FITS primary array (or image data) as well as for columns in a binary 
   table, i.e. in the choice of BITPIX or TFORMn. The choice should be 
   based on scientific usage and technical issues like storage size, 
   range, accuracy and number of significant figures.  In this respect it 
   is felt that most needs can be covered for a continuous quantity by 
   32-bit floating point values (BITPIX=-32, TFORMn=rE) and for a discrete 
   quantity by 32-bit integer values (BITPIX=32, TFORMn=rJ) which 
   constitute the "preferred data types". Other data types are discouraged 
   unless there are valid reasons to use them. These might include 
   considerations on the (saving in the) amount of space needed for 
   storage (which could justify BITPIX=8,16 or TFORMn=L,X,B,I), on the 
   range covered by the data (if the values are beyond the range of a 
   32-bit floating point value, or require a larger number of significant 
   figures, then BITPIX=-64 or TFORMn=rD are justified), or on the 
   accuracy needed (e.g. a TFORMn=rK is justified for a high resolution 
   time tag). We currently fail to see justifications for other types.

----------------------------------------------------------------------------
Lucio Chiappetti - INAF/IASF - via Bassini 15 - I-20133 Milano (Italy)
For more info : http://www.iasf-milano.inaf.it/~lucio/personal.html
----------------------------------------------------------------------------