[Serusers] PRACK

[Jiri Kuthan]

Greg,

thanks for updating me. I maintain ser is not guilty, snom is. 
Replies bellow.

On Tue, 4 Mar 2003, Greg Fausak wrote:
[...]
> INVITE -> *
> 401 Unauthorized <- *
> ACK -> *
> INVITE (with credentials) -> *
> 100 Trying <- *
> 183 Ringing <-
> PRACK -> *
> 200 OK <- * (this is OK for the PRACK)
> 200 OK <-
> 
> This is where things get off track, the SNOM phone wants to send the
> ACK to the GATEWAY instead of the PROXY.  All of the messages
> above that have an * means no record-route is present in the message.
> 
> What SNOM is saying is that the *first* OK from the PRACK sets up
> Who to send the message back to....that doesn't make sense to me.  The
> *second* OK from the INVITE is that SNOM is responding to, so why don't
> they
> use the record-route that is in that message.
> 
> The long winded answer is yes, they think SER should be providing a
> 'Record-route' in the first OK.

Record-route is mirrored in replies from requests by phones, not by proxy
servers. If it is missing in replies to record-routed requests, it is 
phone's fault. 

The proper behaviour is imho as follows:
- INVITE is record-routed by proxy; 18x replies must mirror record-routes
  (if they don't, it's phone's UAS bug)
- subsequent transactions must use route set learned from the INVITE 
  transaction; that means, PRACK should set Routes, and so should later
  on the ACK too; (if that does not happen, it's phone's UAC bug)

So I think the snom-phone bugs encountered here are:
- route learned from INVITE 18x is not set for PRACK and ACK
- record-route is not mirrored in replies to PRACK

-Jiri