Dare Obasanjo wrote:
--- Robert Sayre <[EMAIL PROTECTED]> wrote:
On 4 Nov 2004, at 12:33 pm, Robert Sayre wrote:
It uses a new a verb so you can't screw up the
XML on purpose and send
requests to random POST handlers. Using a new
verb means that the
resource actually has to be an Atom error
handler.
????
HTTP/1.1 200 OK Content-Type: application/atom+xml Atom-Error: http://example.com/manager/admin <?xml version="1.0" encoding="UTF-8"?> <feed> ... <entry>...</entry>
POST /manager/admin HTTP/1.1
I still don't get it. How does using a new HTTP method
somehow prevent whatever problems you claim exist by
using POST. Also you need to do a better job of
explaining what these supposed problems are.
I could see an argument as follows
- Error notification semantics are sufficiently crisp or distinct that they are justified in being broken out from POST*.
- Only servers built with this HTTP extension in mind will need to deal with it (for now that's Atom aware. but over time we should be wary about calling it an "Atom method")
- Intermediaries, such as security applications, can reason about the method independent of other things going through POST.
- Any caching subtleties can be reasoned about independent of POST.
Nonetheless, my main issue is bifurcation of resources** into those that know about ERR and those that don't. Unless the case is cast-iron I think it hurts Atom adoption to be extending HTTP like this. Finally, that case probably needs to incorporate a solid argument that doing error notifs over POST is not just broken or inelegant, but harmful.
cheers Bill
* This is where Mark Baker and I will not start a sidethread about POST's genericism ;)
** Only web services wankers worry about services.
