[jdev] Re: Google DNS SRV records

[Rory]

On Mon, Aug 29, 2005 at 11:05:31PM +0200, Tomasz Sterna wrote:
> We are trying to make XMPP accessible to everyone, not the computer geeks only.
> 
> Please try to explain the whole concept of hosts file, ssh and
> tunneling to Aunt Tillie...
> 
> I'd rather like to explain how to click one checkbox and what to enter
> into two extra fields on the Connection tab in her client.
> And I'm really glad most clients support it.
> 

Hey, I was just answering the scenario I was given. Don't actually
know any Aunt Tillie living on a network with all non-ssh traffic
blocked.  Even if she exists, clicking a checkbox and editing two
extra fields ain't gonna get her jabbering - she still needs to
set up the tunnel.

It might come as a shock, but I agree that there are situations
in which being able to specify a specific hostname/IP address is
useful or necessary.

This said, there is currently no reason why support for this
should be compulsory. However, if we were to adopt an attitude of
indifference with respect to whether or not XMPP service providers
bothered to use DNS SRV records for discovery of their service,
then we are basically making such support compulsory. And there will
be other consequeces also, in fact I'm more than a little surprised
that the alarm bells are not ringing in more heads out there.

Here is just one example of a problem. I'm currently contemplating
building a legacy bridge that would allow the non-xmpp-complient
clients to talk to any XMPP server.  It would work as follows. The
client connects to the bridge and does it's <stream:stream
to="amessage.de" ...>. The bridge thinks fine and it goes off and
asks for the DNS SRV record of '_xmpp-client._tcp.amessage.de' which
tells it that the service is hosted at c2s.aare.eniac.de:5234. So
it pretends to let the client authenticate with it, grabs
the credentials and goes and does a proper XMPP connection to
c2s.aare.eniac.de. And so on...

Clearly this won't currently work if a client connects with
<stream:stream to="gmail.com" ...>. What would you have me do -
start maintaining a list of mappings from domain-names to hostnames
for a bunch of publically available XMPP services who don't bother
using DNS SRV records?

Oh, and Yes, the irony that clients who wished to use the bridge
would need to be able to connect to an address different than the
one returned from DNS has not gone unnoticed ;)

> And of course I would also warmly welcome the SRV record for
> _xmpp-client._tcp.gmail.com, becouse that would spare me that
> explanation, making "things just work".
> 

Perhaps you understand now that there might be a little bit more 
to win here than just saving an explaination or two.

Regards

Rory