Re: Issue 133, and permitting no body

I don't disagree with what you say (although the subtleties of your
last paragraph are still washing over me... I'm still a little
stunned by repeated references to the arcana of distributed object
systems), but I'd note that one of the biggest reasons PUT never
really saw its full power realised was the failure of implementations
to support content negotiation. PUT combined with conneg is a very
powerful tool.

Of course, POST is pretty nifty too.

Cheers,


On Tue, Feb 05, 2002 at 05:08:49PM -0800, John J. Barton wrote:
> At 05:12 PM 2/5/2002 -0500, noah_mendelsohn@us.ibm.com wrote:
> >... I particularly understand
> >it's applicability to the many situations in which there are
> >stateful resources, and a desire to create/update/access them in a
> >consistent manner.
> 
> Resources are things you GET representations from.  This barrier --
> you don't get resources but only representations -- has to work in
> the opposite direction.  So you can't create/update resources.  You
> can only send in representations.
> 
> That's why the Web didn't buy PUT.  By that I mean we did a big
> experiment and the results are in: POST wins by a landslide.  Yes
> there are some confusing factors, but POST has exactly the property
> of sending in a representation (ok, just some funny data) with an
> impact on later GETs That impact is determined by the server in
> some encapsulated way, to preserve its barrier between
> representations and resources.  That barrier is key to web server
> sanity.
> 
> The other magic of POST has not (yet?) been realized in web
> services. Critical to the progress of web technology has been the
> ability of web site developers to instantly upgrade their system by
> changing the content of forms downloaded by GET then POSTed.
> Legendary are the annals of distributed systems developed, hyped,
> deployed, and then frozen in time.  Any system that does not
> incorporate an upgrade solution as effective as web apps will join
> them.  Web services included.
> 
> Returning to Stuart's thread, this line of reasoning says: "don't
> worry about the POST response, worry about how the client formed
> the message". Semantic ignorance is bliss on the client side: the
> less it knows the more we can accomplish.  If web service requires
> simultaneous upgrade (homogeneity and central admin), we have
> created a lot of angle brackets for nothing.

-- 
Mark Nottingham
http://www.mnot.net/

Received on Tuesday, 5 February 2002 20:12:47 UTC