- From: Mark Baker <distobj@acm.org>
- Date: Fri, 12 Nov 2004 16:49:43 -0500
- To: noah_mendelsohn@us.ibm.com
- Cc: Hugo Haas <hugo@w3.org>, www-ws-desc@w3.org, xml-dist-app@w3.org
I agree that's a reasonable ex post facto explanation of the current state of affairs, but I don't recall simplicity being an impetus for the removal of the other response codes from the state transition table; as I recall, the lone reason was "protocol independence". I looked for the discussion around this issue, but couldn't find it, and it's now late on Friday. FWIW though, I did find this[1] from Henrik which addresses an important consideration; if 202 was used, the client can't expect the one-way MEP, it can only learn of it after receiving the 202. So if a 200 was received instead, then the client would get the "real" response. I don't know how the MEP model would deal with this, but I'm not a big fan of it anyhow, so I'll let WSD worry about that. 8-) Cheers! [1] http://lists.w3.org/Archives/Public/xml-dist-app/2002Jan/0352.html Mark. On Fri, Nov 12, 2004 at 04:15:27PM -0500, noah_mendelsohn@us.ibm.com wrote: > Mark Baker writes: > > >> Sweet, sweet irony. 8-) > > Or maybe a good example of keeping it simple. Nothing prevents anyone > from publishing a specification for a different/enhanced HTTP binding that > provides richer features and more complex MEPs, which I believe this use > case to be. Basing core interop on simple request/response seems to me to > be a good 80/20 point, not an oversight. I have no problem with someone > doing another binding aimed more at HTTP exploitation, if there's user > demand. I do think there is a question about about how any > application-level response is modeled. It's one thing to have an > acknowledged one-way transmission, which is what my naive understanding > says you get with a 202. It's another to have the response delivered > through another connection, through addressing that has to be worked out, > etc. The key thing that a request/response MEP gives you, IMO, is that > the response is implicitly addressed to the requester. Usually there is > also some sharing of connection infrastructure, particularly in the case > of a short-lived request/response. That's what we're modelling with the > existing HTTP binding and I think it's a good 80/20 point. > > -------------------------------------- > Noah Mendelsohn > IBM Corporation > One Rogers Street > Cambridge, MA 02142 > 1-617-693-4036 > -------------------------------------- > > > > > > > > > Mark Baker <distobj@acm.org> > 11/12/2004 04:11 PM > > > To: noah_mendelsohn@us.ibm.com > cc: Hugo Haas <hugo@w3.org>, www-ws-desc@w3.org, xml-dist-app@w3.org > Subject: Re: Seeking clarification about the use of the HTTP binding > > > On Fri, Nov 12, 2004 at 02:47:01PM -0500, noah_mendelsohn@us.ibm.com > wrote: > > Hmm. So an interesting question is whether the HTTP binding ever sends > a > > 202. > > Ah, good point. > > It's unfortunate we (XMLP) chose to declare the state transition on > "200" rather than "2xx", but IIRC, there was considerable debate about > this point, as those promoting "protocol independence" feared that > exposing too much of HTTP to the application was a bad idea. Support > for 2xx used to be there[1], but was removed and replaced by [2] as a > result (IIRC). > > Sweet, sweet irony. 8-) > > [1] http://www.w3.org/TR/2001/WD-soap12-part2-20011217/#NFDC > [2] http://www.w3.org/TR/2003/PR-soap12-part2-20030507/#httpoptionality > > Mark. > -- > Mark Baker. Ottawa, Ontario, CANADA. http://www.markbaker.ca > > > -- Mark Baker. Ottawa, Ontario, CANADA. http://www.markbaker.ca
Received on Friday, 12 November 2004 21:47:35 UTC