[fitsbits] Bintable proposals

[Robert Lupton the Good]

Although tapes are on their way out, writing the results of database
queries to sockets is probably on its way in.

The SDSS is currently thinking about how to support requests for
binary tables sent from a database server to a local machine.  One
proposal (that I for one like) has been to write them as FITS BINTABLEs
with an unknown number of rows, then include the number of rows
at the end -- rather like encapsulated postscript's ATEND directive.
This could be done by setting NAXIS2 == -1 if we wanted to keep it
as a numerical field.

No, this isn't allowed by the current FITS standard, but I don't
see why history automatically should dictate current practice. You
cannot trivially skip over such a file --- but I don't expect that
this approach will ever be used for multiple HDUs in a file.  More
worryingly, you have to think a little about how to handle heap;
but I think that that can be done with ATEND (-1?) PCOUNT and THEAP fields.


As has been pointed out, you can easily patch up such files to be 
regular FITS once you've got them.  Splitting them up into
little pieces makes life harder for everyone --- including the 
recipient.  In fact, if I got a file that was made up of a number
of sub-tables, the very first thing that I'd do was to rewrite it
into a single table, and that is much more of a pain than moving
some ATEND fields to the header.





					R
-- 

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Robert Lupton                                           rhl at astro.princeton.edu
(609) 466 2431 (Home)					(609) 258 3811 (Work)