[jdev] Re: discovering extensions without disco

Robert McQueen robert.mcqueen at collabora.co.uk
Thu Apr 6 13:50:35 CDT 2006

Gary Burd wrote:
> Please let us know which of the extensions you would like to see
> documented and supported in a formal way.

Of particular interest at the moment is how we should discover your STUN
& TURN servers. Does the client have the servers hardcoded, or does it
get them sent to it during the web authentication? Would you consider
adding SRV records in the SIP style of _stun._udp.gmail.com etc for
making at least your STUN servers available to 3rd party clients?

>> according to the standard in places (<presence type="unavailable">
>> behaves like invisible, so the server sends us presence, messages and
>> iqs... something like <feature
>> var='presence-unavailable-behaves-like-invisible'/>).
> I will fix this instead of advertising the broken behavior.

Ok, cool... but fix as in remove this invisibility behaviour entirely
and make unavailable do what it says on the tin, or could you leave us a
way of becoming invisible somehow? In lieu of a simplified
jabber:iq:privacy profile at the moment, could you behave like jabberd2
and have this invisible behaviour available as <presence
type='invisible' />, and make it discoverable as <feature
var='presence-invisible' />, or would the resulting slight XMPP
transgression be too offensive? Failing that, a google:invisible would
probably suffice for the time being, if you could see it was there with
disco. :)

>> +1 too, but in the meantime, how are we supposed to block users on
>> Google Talk?
> We discouarge the use of the extension, but don't prevent you from
> using the extension.

Sure, I understand this, but it's clear (at least to me) that if I've
looked at diagnostic logs and made educated guesses at how something
works, and that this something is undocumented and lives inside your
private namespace, I really can't expect that it will always work like
this or will always be available. However, I don't think that making a
feature appear in service discovery doesn't imply a level of support for
it. It just gives anyone who is foolish enough to implement it a very
clear indication of when it's definitely won't work. It would be very
nice to see some disco responses with some of these google things in,
even if in two weeks they break or disappear. :)


More information about the JDev mailing list