[Standards] stream restarts

Gaston Dombiak gato at jivesoftware.com
Tue Apr 29 20:51:13 UTC 2008

Hey Peter,

When I implemented "stream restarts" I was really following the spec without seeing the reason for the restarts. I was not sure if I was missing the point about something or not. In our case, the server was not keeping any information about the session so we really were not forgetting anything with the stream restart. Maybe I'm still missing the point here. :)

Anyway, I'm all good with avoiding the stream restart and just use one stream header where TLS, SASL, etc. are negotiated.

BTW, what's the reason for doing this change now? Is it to reduce the amount of traffic and make it more appealing to mobile clients?


  -- Gato

-----Original Message-----
From: standards-bounces at xmpp.org [mailto:standards-bounces at xmpp.org] On Behalf Of Peter Saint-Andre
Sent: Tuesday, April 29, 2008 12:57 PM
To: XMPP Extension Discussion List
Subject: [Standards] stream restarts

A few weeks ago I got to talking with Joe Hildebrand and Travis Shirk at
Jabber Inc. about stream restarts. Once upon a time we thought we needed
them (e.g., so that the server would be sure to forget about any data it
received before STARTTLS completed), but now we realize that was a
misunderstanding of the TLS and SASL specs. So it seems that we could
redefine the stream negotiation process to get rid of stream restarts
after STARTTLS and SASL negotiation. The conclusion that Joe and Travis
and I came to is that we could do this by defining new features for
STARTTLS and SASL negotiation. So a server that supports old STARTTLS
and "STARTTLS2" would advertise both features. If you choose STARTTLS2,
you would not restart the stream and the server would not expect you to
do so. But if you support STARTTLS you would use that and both sides
would expect the stream restarts. IMHO the new features would use
namespaces like urn:xmpp:starttls instead of the namespaces in the IETF
tree, but that's a minor detail (the important point is that the xmlns
would be different).

If there are no objections to this idea, I'll write up a little XEP or
two about this.


Peter Saint-Andre

More information about the Standards mailing list