[Council] proposed relationship metadata JEP
stpeter at jabber.org
Wed Dec 17 10:27:20 CST 2003
On Wed, Dec 17, 2003 at 08:57:32AM -0700, Matthew A. Miller wrote:
> For the most part, I like this better than the original setup within
> JEP-0123. There is one issue and a couple of questions I have.
> The changes to disco give me pause. The addition of the "uri" attribute
> also changes the requirements for the "jid" attribute from "required" to
> "optional". This change would break some existing implementations of
> disco, which expect there to always be a valid JID. The only thing I
> can think of for now is to add an additional element to #items to
> represent URI-based things (maybe <ref uri='some-uri' name='some name'/>
> or some such). I'm just very concerned about making breaking changes to
> disco (which I've seen used quite successfully) this late in its game.
Yes, that is one implication, which is why it needs to be discussed. The
big question is whether it makes sense to use disco to refer to things
outside the Jabber network. I'm not sure it does yet, but it's worth
thining about. It is true that a server could have an associated web
page and it might be nice to advertise that using disco#items, just as
it might be nice to advertise my weblog or code project or whatever. But
one could also advertise those things using infobit keys. So I'm not
wedded to doing this, and I don't want to break existing disco impls.
Of course, the other approach is for entity metadata and relationship
metadata to be treated in a different fashion -- entity metadata is a
disco#items node but relationship metadata is published and gathered
using a separate namespace. I like the idea of making them consistent,
but there's no use getting foolish about consistency. If we pull out the
'uri' business then I'd list relationships to other JIDs only, which is
not the end of the world but does potentially reduce the centrality of
Jabber in how I communicate with others.
> Second, the protocol for retrieving relationship metadata uses <item/>
> in an "#info" query, rather than an "#items" query. I'm guessing this
> was a just an oversight. However, I wonder if we could have
> accomplished the same result by simply making a "#items" query to the
> already (soon-to-be) registered well-known "metadata" node from
> JEP-0123. I'm not objecting to having two JEPs (so this doesn't need to
> be moved back to JEP-0123 on my account), but I wonder if we might make
> better use of our registry this way.
Yes, that's a copy-and-paste typo.
The 'metadata' node is intended to be "entity" metadata -- metadata
about me (a la vCard) not about other things. That's why I think
separating the nodes is a good idea.
More information about the Council