Format of concatenated UCD's

[Norman Gray]

Rob,

On 2005 Jun 9 , at 22.19, Rob Seaman wrote:

> Is there some technical reason that Rick's example can't become 
> "con:event.burst ucd:em.opt"?  Let's try to avoid getting lost in the 
> semantic underbrush.  Please tell me why we can't have separate 
> namespaces?

Myself, I don't think there's anything wrong with 
con:event.burst;ivoa:em.opt.  The strong nervousness which others have 
about the use of UCD namespaces, as expressed for example in the UCD1+ 
document, I understand but do not fully share.

There seem to be two issues.

1. (semantic) If you have a namespaced UCD atom, then not all software 
will understand it.  This is true, but it's also rather in the nature 
of namespaces that they are an acknowledgement that the namespaced 
thing is special or non-standard, so that it will only be useful to 
certain processors.  They also avoid the atom being confused with other 
atoms which happen to have the same name (this is of course The Point), 
but _usually_ give some indication of where a fuller explanation may be 
found.  That is, they're a complication, but avoid a worse problem.

2. (syntactic) How do you associate a namespace prefix with a URI (for 
that is the only sane way of pinning down a namespace)?  In the XML 
context, this is straightforward, since you can simply use the same 
namespace resolution algorithm as exists in XML (I believe it would be 
bad to specify a different one).  That has the problems that it makes 
namespaced UCDs somewhat context-dependent, and that the namespace 
algorithm has some confusing edge cases.  It represents more 
complication, but is perfectly manageable, I think.

How do you do the association in non-XML contexts, such as FITS 
headers, or in a database?  That's more complicated, and is probably 
inevitably messy.  One possibility, I suppose, would be to follow the 
unstandardised but common XML habit of writing 
{namespace-uri}namespaced-thing, as in 
'{http://www.ivoa.net/ucd}em.opt'.  Not pretty, and it could easily get 
very long-winded.

All this is a long way of saying that there are technical worries about 
using namespaces, but that they don't seem to me to be killing ones.

The advantage of using namespaces is that they allow sets of `ucds' 
like the VOConcepts to be defined, which don't clutter the generic UCD 
vocabulary, but which do inherit the precision in the UCD spec, and any 
updates follow it.  ...it seems to me.

All the best,

Norman


-- 
----------------------------------------------------------------------
Norman Gray  :  Physics & Astronomy, Glasgow University, UK
http://www.astro.gla.ac.uk/users/norman/  :  www.starlink.ac.uk