Dave Cridland wrote: > On Tue Oct 14 02:35:08 2008, Peter Saint-Andre wrote: >> Right, as long as you never send stream management data unless you first >> learn that the other side supports the protocol, this seems fine. >> >> > But that's not what happens. > > What happens is that the client (initiator) declares the namespace first > off, without knowing whether the server (responder) supports the > protocol.
Ah, I was speaking only of the usage once the stream has been negotiated, not inclusion of the namespace in the stream header. > Now, should the responder: > > a) Assume that the unknown namespace is not an extension namespace, and > not bother checking for illegal uses of the prefix within stanzas. Yes, please. :) > b) Assume that the unknown namespace *is* an extension namespace, and > reject the stream as illegal. (The client is violating a MUST NOT). > > c) Assume that the unknown namespace might be an extension namespace, > and check each stanza for its usage, thus potentially incurring > significantly increased processing time. > > >> > I say that we should push the 0198 namespace to the <stream:stream>, it >> > seems wasteful not to do it. >> >> It's a small thing really, but I suppose saving a few bytes per ack is a >> nice thing. >> >> > Sure. Which is why I suggest namespacing the namespaces we use in this > way, to force the (a) case. What exactly do you mean by "namespacing the namespaces"? I assume you mean "sender MUST include the namespace declaration in the stream header" and "receiver MUST check the stream header for namespace declarations and remember the stream management prefix" or somesuch. Peter -- Peter Saint-Andre https://stpeter.im/
