Rob Sanderson writes:
> On Mon, 2007-04-16 at 17:24 +0100, Mike Taylor wrote:
> > > (2) The ability to retrieve a record's metadata according to a
> > > specific metadata schema already exists. You just retrieve
> > > that record and name that schema in the record schema
> > > parameter (granted, that doesn't allow you to get both the
> > > record data and its metadata at the same time, you need two
> > > requests - not a big hardship).
> > Ha! I hadn't spotted that. Nice observation.
> True, but doesn't that argue against the need for the extension at
It's a judgement call; but I think not. I think of "give me the
metadata as well as the record" as a reasonable thing to do. Whereas
"I am a super-specific and very demanding application, and I need the
metadata in THIS format" seems much more special-casey, and suitable
to be relegated to a scenario where a separate request is needed.
> > > So I agree with you that in this extension we are developing, the
> > > ability to name a schema is overkill.
> > Good.
> Then we need to define a very thorough schema, or we'll constantly
> be fielding requests: Can you just add <insert favourite metadata
> field> please?
Yes; but that's not the end of the world. We already do this for the
various context sets, and in fact it's not a common or demanding
Or we can explicitly allow the metadata schema to include extension
elements from other namespaces.
> > > (3)
> > > > I would also prefer that the metadata extension be self-contained,
> > > > which means defining something like:
> > > > &x-info-99-metadata=1
> > > What does "1" signify here?
> > It just means "yes" or "true". Maybe it would be more explicit to say:
> > &x-info-99-metadata=true
> > ? Do we have any precedent for true/false values in SRU extension
> > parameters?
> Do we need a value at all? Why not just ...&x-info-99-metadata&... ?
I don't know. Ray's message suggests that we _can_, but it's very
rare to see URL query parameters like that, and I can't help thinking
it's bound to create problems with toolkits that make more assumptions
than they are really supposed to. I agree that omitting the value is
more elegant, but I'm inclined to think it might cause problems down
> > Then we need to define a very thorough schema, or we'll
> > constantly be fielding requests: Can you just add <insert
> > favourite metadata field> please?
> I'm tired of trying to define comprehensive schemas that cause us
> to invent scenarios to justify sticking in anything we can think of
> in the name of comprehensiveness. I want a quick-and-dirty
> solution that satisfies known needs and the ability to specify
> private schemas for private reasons.
I very strongly disagree with this -- it will kill interoperability.
It would mean that every server that wanted to be useful would need to
be able to respond to requests for metadata in many different formats;
and since not all servers would do so, carefuls client would need to
cope with requests being rejected and fall back to asking for the same
data in a less preferred format. In short, it would involve both
client and server in exactly the same kind of clodhopping dance that
we deliberately got rid of when we decided that SRU would support only
one record syntax (XML) rather than the dozen that Z39.50 supports.
So I strongly prefer One True Metadata Schema.
Also: wouldn't it be better to call this Record Data (as we do in the
Record Data content set "rec") since most of the world calls stuff
like author, title and subject by the name Metadata?
/o ) \/ Mike Taylor <[log in to unmask]> http://www.miketaylor.org.uk
)_v__/\ "There's no point in being grown up if you can't be childish
sometimes." -- Dr. Who.