Clarification on DataLink service descriptor input parameters

[Markus Demleitner]

Hi Adrian,

On Mon, Sep 26, 2022 at 07:53:14PM +0000, Damian, Adrian wrote:
> <PARAM arraysize="2" datatype="double" name="BAND"
>            ucd="em.wl" unit="m" value=""
>            xtype="interval">
>       <DESCRIPTION>Spectral cutout interval</DESCRIPTION>
>       <VALUES>
>         <MIN value="3.52631986e-07"/>
>         <MAX value="9.21500998e-07"/>
>       </VALUES>
>  </PARAM>
> 
> The VOTable parser in astropy does not like the syntax - it
> expects, and I would say logically so, that MIN and MAX have the
> type of the BAND param, i.e. an interval. Is there a reason not to
> have the above MIN and MAX expressed as the maximum interval, i.e.
> <MAX value="3.52631986e-07 9.21500998e-07”/>?

Well, this has always been a very sore spot (previously:
http://mail.ivoa.net/pipermail/dal/2020-April/008330.html) -- and I
think it needs to be discussed over on Apps, as it is really a
VOTable problem.

The meaning of MIN and MAX with arrays has never been properly
defined.  Now, given that arrays are supposed to be "homogeneous"
data structures -- all members "mean" about the same thing --, one
can argue that arraysize should be ignored for MIN and MAX, i.e.,
that they are always scalar.  I think that would certainly be the
intent for OPTIONS and for VALUES/null (the latter of which is
admittedly deprecated).

Also, one cannot order vectors (respecting arithmetic); hence, a
really well-defined MIN and MAX for arrays is not possible anyway.

On the other hand, having component-wise MIN and MAX is of course
possible (although that would have about as silly consequences of
char arrays as the "ignore arraysize" convention).  But even then one
would have to ask what such a multi-component MIN/MAX would mean with
arraysize="*".

This latter interpretation is what we introduced in soda (against my
mild resistence, if I may say so,
http://mail.ivoa.net/pipermail/dal/2016-November/007633.html).

Well -- I'd say it's time to clarify this whole thing in VOTable.
I'd kind of like the "give up to arraysize elements; when fewer
values are specified in MIN/MAX than are in the array, the last value
is repeated" I had proposed in April 2020, I must say on re-reading
it.  This would help with the variable-length arrays and would indeed
redeem the datalink example above.  Of course, it's a bit painful in
implementation, but perhaps not overly so.

Whatever we go for: someone would have to put it (and something on
OPTIONS and VALUES/@null) into the VOTable spec.  I'd help out with
it, but I don't want to push it along...

           -- Markus