[Standards-JIG] proto-JEP: Smart Presence Distribution

Joe Hildebrand hildjj at gmail.com
Wed May 24 06:48:19 UTC 2006


On May 18, 2006, at 1:27 AM, Philipp Hancke wrote:

> Joe Hildebrand wrote:
>>> so if you think XMPP _HAS_ to have a to= field, not because it  
>>> serves
>>> a purpose, but only because there once was a document that said that
>>> it has to have one, then sure we can have a to='servername' in  
>>> there.
>>> i'd rather update the XMPP spec, but please we can do it your way.
>> No, we're saying that the spec has this restriction for good   
>> reasons.  There's no way in many implementations to get these  
>> packets  delivered to the right places if there isn't a to  
>> address.  There are  problems on both the send side as well as the  
>> receive side.
>
> Under the following assumptions
> * direct connection between the two servers without any intermediate
>   devices (http://www.xmpp.org/specs/rfc3920.html#arch-overview)
> * XMPP 1.0 stream (for stream:features)
> * stream:to and stream:from present (see xmppbis notes)
> * no piggybacking (it seems odd to me to send and receive for domains
>                    other than the two contained in the stream header)
>
> what problems arise from omitting 'to' on the wire, e.g. the ---
> between S1 and S2?
> You can strip it immediately before sending and add it again directly
> after receiving.

Oh.  I didn't realize that what was being proposed.  I bet the CPU  
for looking for this swamps any effect on a  compressed stream,  
bandwidth-wise, but hey, if that's what you're trying to optimize  
for, shrug.

In many implementations that allow multiple domains on the same  
installation, the to address will have to be there in order to make  
it to the s2s component, and the to address will *certainly* have to  
be there in order for the server as a whole to figure out which piece  
of itself the stanza needs to go to.

>> The spec is right to require a to address, based on real   
>> implementations.  I don't understand why you're so opposed to  
>> just  putting the domain name of the server in the to address...  
>> I've got other problems with the approach, but this seems like an  
>> odd place
> > to get stuck.
>
> Indeed. If that makes you happier we can put the domain name of the  
> remote server in the 'to'. Possibly adding 'distributeitplease'
> resource not to break anything.

Sigh.  I18N.

> In turn, would you please start to put the domain name of your local
> server in the 'to' field if you intend to do a presence broadcast? ;-)

Huh.  That doesn't make sense to me, unless you wanted this to mean  
"please distribute this only to the people on the local server".   
Which, hey, I guess could be cool.  It would just be a natural  
consequence of the approach that is being discussed.



-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 1883 bytes
Desc: not available
URL: <http://mail.jabber.org/pipermail/standards/attachments/20060524/21c975be/attachment.bin>


More information about the Standards mailing list