[Standards] XEP-0258 and XEP-0060

Ralph Meijer ralphm at ik.nu
Tue Nov 22 14:06:50 UTC 2011


On Tue, 2011-11-22 at 13:40 +0000, Dave Cridland wrote:
> On Tue Nov 22 13:33:36 2011, Kurt Zeilenga wrote:
> > 
> > On Nov 21, 2011, at 2:49 PM, Dave Cridland wrote:
> > > One label per publish suits me.

Agreed, but see below.

> > Though that can be made to work with digital signatures, it would  
> > be slightly odd.
> 
> You think? I think it'd be very difficult to manage multi-level  
> publish requests in any kind of real environment. What happens if  
> only part of the publish request can pass through a link? I feel very  
> uncomfortable partially sending, so that suggests the whole thing has  
> to be rejected.

In version 1.13, batch publishing was first removed and then restored
pending further discussion. I see there are still remnants of the
removal, and the spec contradicts itself in a few places. That should be
cleaned up.

That said, a publish-subscribe service with support for security labels
could simply disallow batch publishing for the whole service or just
nodes with policies. Then, the semantics could be made unambiguous
without the need to rely on ordering. You could also enforce that
notifications always contain a single item.

I believe the problem remains for retrieving items, though, where it is
likely you get multiple items back, possibly with different labels.
 
Another approach would be to mark-up the label to refer to the item
identifier of the item it is associated with (with a 'for' attribute,
like in HTML?). This also avoids ordering issues. 

--
ralphm





More information about the Standards mailing list