SCS-1.1 internal draft available

[Markus Demleitner]

Hi Walter,

On Fri, Jul 14, 2017 at 11:50:57AM -0700, Walter Landry wrote:
> Marco Molinaro <molinaro at oats.inaf.it> wrote:
> > Dear all,
> > I drafted a first version of the SCS-1.1
> > as an internal draft for the DAL WG.
> * Do current cone services use this attribute? (talking about type="results")
> 
> I set it.  I think the question is more whether other tools use it.

It's a service to clients; I'd suspect right now almost all of them
will just just the first table they find.  But as soon as you can
have more than one resource, it's certainly a good idea to be
explicit, and of course, it's what DALI says.  Consistency is always
good.

> 
> * should we relax on the number of tables?
> 
> How would that work?  Would you specify multiple tables to search at once?

The immediate use case is to allow datalink descriptors in cone
search results.  There are many uses for that, most typically related
to provenance ("these object properties were derived from these
spectra"), but lots others are conceivable.

Further down the road, if we tell clients to look at the
@type="results" resource already now, I suppose we'll have less
growing pains if we move SCS to actually return object-relational
material ("Source DM"), for which multi-table (resource?)
serialisations will be the rule rather than the exception.

> * A solution to move to UCD1+ would be nice, but it should probably be
>   postponed to a major revision.
> 
> Is it worth doing if we do not have UCD1+?  That and allowing updated
> VOTable versions are the two main issues for me.

Oh, SCS is painful in so many minor ways that cleaning it up to the
extent that's possible without breaking clients is absolutely
worthwhile.  Also, you can of course already use UCD1+ for everything
but the identifier and the main position, and most services do that.

But I simply can't see a sensible way to get rid of the old UCDs
specified in 1.0 without breaking old clients.  There'd of course be
no problem saying you can return these "protocol" UCDs in UCD1+ *if
you know* that your client is 1.1 -- but you'd still have to produce
the old ones by default, so you'd not save anything, really.

       -- Markus