On Tuesday 29 April 2008 1:55 pm, Peter Saint-Andre wrote: > Right. There is also the fact that using one stream header will enable > us to quiet those who say that XMPP is not XML. :P
An XML document isn't defined as being the entirety of a TCP connection anyway, so I don't see how stream restarts lead to the argument that "XMPP is not XML". Yeah, I'm slightly sad we can't just point an XML parser at a TCP socket and be done with it, but those days were over the moment we wanted encryption. Personally, I think removing the stream restarts might make XMPP more /unlike/ XML, because then the document is no longer continuous. A single XML parsing state has to bridge between parsing unencrypted to parsing encrypted for the same document. Contrast this to the current practice, where when the switchover occurs, you work on a fresh document with a new parsing instance. However, I am interested in optimizing away round trips. I wonder if perhaps a more clean (and compatible) way would be to write Tony Finch's "psychic stream restarts" right into the RFC. -Justin
