- From: <noah_mendelsohn@us.ibm.com>
- Date: Thu, 12 Feb 2004 14:18:09 -0500
- To: Jacek Kopecky <jacek.kopecky@systinet.com>
- Cc: XMLP Dist App <xml-dist-app@w3.org>
Jacek Kopecky writes: >> it is my opinion that we should not prohibit the >> use of XML 1.1 with SOAP (or the HTTP binding in >> particular), even if in some cases there may be >> problems. I can see the value in allowing this, but I remain nervous on several fronts: I think it is a very valuable to have a set of design principle that holds for SOAP 1.2 today: a) The legal content of a SOAP envelope is the same everywhere, independent of node. b) Every binding specification must provide for moving any such envelope from one node to the next (except perhaps in special cases where you know, for example, that your binding is used to connect only to a fixed named node). Independent of the HTTP binding, your proposal changes that. We enter a world in which every specification for a header or body must indicate the restrictions on its content. Implementations must take the union of the capabilities required and choose an XML format that should be 1.0 in the case that no new characters are used in names or content, but 1.1 in the case that they are. That might vary from message to message. Now we have a failure mode in which the same message easily transits 3 hops, but breaks when an intermediary using a header in the new format is introduced along with a corresponding binding. Furthermore, if you do what seems best and use XML 1.1 only when truly needed, you run the risk that nodes will successfully commincate for weeks or months, then fail on a particular message. Of course, you could go the other way and use 1.1 unconditionally, which would force the failure earlier. The downside of that is that you'll get lots of messages labeled as 1.1 when in fact they could have been 1.0. It will be a lot of burden on downstream 1.0 software to get them back in a form that's processable. So, I remain nervous about the direction you're suggesting. Perhaps you're right, but I'd like to think through what header specs are going to look like, what's going to flow on the wire in realistic scenarios, what's going to break when, and whether users will be happy with our choice. It may well be that such an analysis would support your position, but I'm reluctant to commit until we've proven it's tractable. Thank you! -------------------------------------- Noah Mendelsohn IBM Corporation One Rogers Street Cambridge, MA 02142 1-617-693-4036 --------------------------------------
Received on Thursday, 12 February 2004 14:19:06 UTC