[Standards] XEP-0115 redux

[Joe Hildebrand]

On Jan 18, 2008, at 3:25 PM, Peter Saint-Andre wrote:
> So caps would include everything from XEP-0092 except name and OS.  
> Since I agree with you that including the OS is problematic (it  
> unnecessarily opens a security hole), why not just include the name  
> in caps and be done with it? Does n='Psi' add that many bytes to  
> worry about? Or are you concerned about the fact that the name is  
> not localized?

Among other things, yes.

> Sure, clients could send an iq:version request once upon receiving  
> caps for a given node+ver+v from a given JID, then cache that across  
> sessions. But I can hear the complaints coming already: "I develop a  
> web/mobile/whatever client and I can't cache across sessions so I'll  
> have to send iq:version requests every time". And I think this  
> objection will be common enough that we might as well include  
> n='client' (probably about an extra 10 bytes on average) in every  
> presence stanza.

Perhaps I didn't show enough of my work.  The idea is that if you  
cache on node+ver+v, you'll disco#info the same number of times as in  
version 1.4, and get the exact same identity data.  No need for n=, no  
need for iq:version requests.

> If including 'n=' enables us to finally finally finally put XEP-0115  
> to bed, I'm all for it. It is mildly annoying and it offends the  
> desire we all have for clean protocols, but we've already held our  
> noses over caps so many times now that I've lost count. ;-)

I guess I'm the i18n police today.  Sigh.

-- 
Joe Hildebrand