I am going to try to answer myself.

On 10/31/05, Luke Arno <[EMAIL PROTECTED]> wrote:
> On 10/31/05, Joe Gregorio <[EMAIL PROTECTED]> wrote:
> > On 10/31/05, Luke Arno <[EMAIL PROTECTED]> wrote:
> > > I think that Atom syntax provides us with what we need.
> > >
> > > Not only do I think that we can deal in valid entries, I
> > > strongly believe that we need *no* extensions for the
> > > core protocol.
> >
> > A Pace with camera ready copy that explains
> > the disposition of atom:link and atom:id on a POST
> > would be most appreciated.
> >
>
> That is a tall order and I think it requires some
> discussion as to how. :-)
>
> I would like to start with atom:id.
>
> Should it be generated client side?
>

It seems like that could work.

> Does anyone have suggestions for a client side
> algorithm?
>

That is a dumb question. Most methods of producing
an id are fine on the client side.

If a conflict comes up the server can reject the POST.

> Should the server pass them out?
>

Na. That would be pain.

> If so how?
>

See above.

> Should whole entry templates be served up?
>

Forget that. That was a straw man proposal.

> Should the "POST id" be reset by the server?
>

No. What about moving entries from one collection
to another? The id shouldn't change.

A DELETE/POST combo can be used to "move" an
entry.

> Can our use of id at POST time address our
> idempotency concerns?
>

Hmmm. That is tougher. Try this:

If identical entries (id and all) are POSTed within
X time go ahead and just issue the same 201
for each.

Would this break down?

What am I missing?

- Luke

Reply via email to