RE: I-D Action:draft-snell-http-prefer-03.txt

On 2011-03-30 at 05:14:11, Cyrus Daboo wrote:
> Hi Poul-Henning,
> 
> --On March 29, 2011 10:12:34 AM +0000 Poul-Henning Kamp 
> <phk@phk.freebsd.dk> wrote:
> 
> >> The goal seems to be to ask for the server's result of a PUT or 
> >> POST to be returned as part of the same action instead of requiring 
> >> the client to make an additional GET request.
> >
> > And standardizing a header to kindly request but not demand this, 
> > would help how ?
> >
> > Is the hope that all browsers will send this by default ?
> 
> Browsers are not the only HTTP clients around.
> 
> In the CalDAV (RFC4791) world we do have servers immediately modifying 
> data PUT by clients with the requirement that clients then have to 
> immediately do a GET. This happens because the server typically does 
> take immediate action to do some form of scheduling - that may simply 
> be to add an indicator to the data that a scheduling operation is 
> pending (and that operation then happens asynchronously). Avoiding the 
> extra roundtrip would be beneficial in this case particularly as 
> mobile devices make use of this service.

Having a PUT and GET merged sounds like a useful idea.

There's an inherent assumption in this that I don't think you've considered.  The response to a PUT might already be an entity with some purpose (e.g. feedback about the change of a different nature).  For one, that means that - unlike a regular GET - the entity in the response cannot be assumed to be a representation of the resource.  That doesn't mean that you can't merge the two, but I'd have to suggest a specific indication.  It might be as simple as setting Content-Location.

--Martin