> (opaque URI-based) rel so I have proposed class="action" be added for
<link rel="stop@occi"
where "occi" is registered as for use by occi.
IOW allow the link registry to define rel scopes. The url or clash thing sucks and I don't want someone rolling up with a curies/qnames or some other XML macrocrap proposal for Atom.
Thanks for the feedback - this is an interesting idea, however it's not immediately clear to me how this helps us as one of our main motivations is that others can create their own actions without bothering us or having to be registered.
Basically we'd be trading one opaque identifier that doesn't need to be registered for a shorter one that does and unless we were to accept petitions from the public it would still not be possible to identify which relations are "actions" anyway (consider "stop@occi" vs "quiesce@vendor"). The use case is the client saying "ok all these things are actions so even if I don't understand what "quiesce" means I can still show a button to the user in the right place".
A reversed, dotted domain format ala "org.occi-wg.action.stop" was something else we considered but it too would not enjoy backwards compatibility with existing implementations. It would however provide a simple solution to the registry "problem" and could be easily differentiated from "short" relations by the presence of "." and URIs by ":" (or some other regexps). That is to say, it's compatible with what we have today, infinitely scalable, distributed, concise, basically everything you want in an identifier.
Sam
Sam Johnston wrote:
Another requirement I've stumbled on is the ability to group links.
For OCCI for example we want to give users the ability to create their
own actions (eg start, stop, restart) which will be advertised in the
HTTP headers and/or HTML HEAD. Currently each action has it's own
(opaque URI-based) rel so I have proposed class="action" be added for
grouping. Alternatively we could do rel="[http://purl.org/occi/]
action" and refine the action with one or more custom attributes (type
is not really appropriate here, nor when advertising protocol
endpoints like SSH and RDP which is anoter problem we ran into).
Pagination (eg first, last, next, previous) is another potentially
interesting group. Though the semantics are mostly predefined one
could envisage links like "next 100", "next 1000".
Sam on iPhone
On 12/11/2009, at 7:18, Joe Gregorio <joe@xxxxxxxxxxxxxx> wrote:
On Thu, Sep 17, 2009 at 5:26 PM, Mark Nottingham <mnot@xxxxxxxx>
wrote:
On 16/09/2009, at 7:46 PM, Sam Johnston wrote:
Would it be possible then to support multiple references so that
people
can see at a glance that a given relation is implemented as
described in
multiple formats (rather than just the first format that happened to
register it)? May well not be worth the maintenance effort.
How about adding a new field for references to more information
about how a
relation is used in a particular context (scoped by context media
type)?
E.g.,
References regarding use in specific contexts:
text/html: [HTML5]
application/atom+xml: [RFC4287]
Yes, that sounds like a great idea. And vaguely familiar:
http://lists.w3.org/Archives/Public/ietf-http-wg/2009JulSep/
0699.html
Thanks,
-joe