> > OK. The question still remains, though. If an attribute is only used in|
> > Accounting-Request packets, does the argument against complex
> > attributes still apply?
> My preference would be to say yes, especially where the contents of
> that attribute are interpreted by *later* policies on the server.
Now that we have RFC 5176, accounting data can be kept as "state" that is
later used to construct CoA/Disconnect-Requests. Is this what you mean?