UCD elements

[Martin Hill]

Patricio F. Ortiz wrote:

> On Thu, 1 Jul 2004, Martin Hill wrote:
> 
> 
>>At the moment we have an element called <UCD> that contains UCD v1 values. We
>>are now talking about UCD1+ which are different; what element name should we
>>use for these?  <UCD_PLUS> ?
>>
>>It would be Bad to use the same element, as the values are really of different
>>types (and so interpreting it will be ambiguous).  However we might be able to
>>get away with it this once as it is still early days, and I gather no-one has
>>data lying around in VO formats that they intend to keep for long!
> 
> 
> IMHO we should stick to <UCD>. I assume that very few services have
> implemented UCD1+ yet, (or v1 for that matter).

Happy with that as long as it's true...

> Perhaps a way to leave a
> door open to reflect differences would be an attribute "version", which by
> default should be 1+. 

That sounds like a good idea, as we ought to think about UCD v2 too. It strikes 
me that we are likely to want to allow both (or more) at once in our documents, 
so they can be submitted both to old services (which will only understand older 
UCDs) and newer ones (where we may want to take advantage of the newer ones).

> There are some syntax differences which could allow a
> service to distinguish one version from the other, at least in principle.

Hmmm... We shouldn't have to tell the users of our formats that they need to 
mess about parsing the values to find out what type they are.  That's what 
element types are for :-)

Imagine what would happen if we sent a UCD2 document to an old UCD1-only 
service; we need to make sure that the old service, which knows nothing of the 
new UCDs, doesn't assume the UCD2 values are the only UCDs it knows about.  That 
doesn't matter so much now as UCD1+ is very different from UCD1s (and so we can 
expect the service to just fail) but we can't count on that in the future.

Roy Williams wrote:

 > I think the element name UCD will be enough for now. That keeps it simple.

It keeps the table *looking* simple, but it makes the interpretation so much 
more complicated!

I believe there are a fair number of datasets out there now that include UCD1s 
in their metadata.  What is going to happen during the transition?  Can we 
assume the transition will be fairly quick?   Does it really not matter if some 
VOTable results contain UCD1s in the header whereas others contain UCD1+s?  This 
is fine if the UCDs are not used much, and won't be used until all datasets have 
switched to the new ones...

(Shall we continue this in the Registry thread only?  Or VOTable?  Or is it in 
fact a different answer for each of those?)

Cheers,

MC


-- 
Martin Hill
www.mchill.net
+44 7901 55 24 66