Friday, October 28, 2005, 12:59:04 AM, you wrote:

> What are the key remaining issues that need to be settled?

> 1. Introspection
> 2. Publishing controls (including pub:draft and pub:significant)
> 3. Error reporting format
> 4. The value of atom:id on posted entries
> 5. Traversal/List/View templates
> 6. Any thing else?

Are we interested in doing anything about idempotent POSTing and
preventing duplicates when the client has to retry? It has been raised
a few times eg [1].

Assuming "The network is reliable", is a poor practice even in browser
applications [2]. I really think that we should try to do something to
tackle this in Atom. With mobile applications it is likely to be
particular problematic.

Even if we don't make such a feature core, we should consider it and
ensure that it is possible to add as an extension to the protocol.


It is quite a tricky problem. Some techniques to avoid it require the
client to contact the server to reserve a posting URL to PUT to. This
only works when posting URLs are a cheap resource (eg a GUID in query
string), but Atom allows servers to servers to provide pretty URLs
based on the Name header, and such URL's are not cheap resources.

Another option is to use a Message-Id header or field.  This raises
the question of who generates the Message-Id.  A server generated id
is more reliable, but it requires an extra round trip.  Clients might
use naive algorithms if they are left to do it.  A compromise could be
to use a hybrid approach where the server generates a prefix for the
client, and the client adds a suffix to produce its own ids.  The
client could then keep the prefix between sessions if it wanted, which
would avoid much concern about extra round trips.

[1] http://www.imc.org/atom-protocol/mail-archive/msg01686.html
[2] http://www.mnot.net/blog/2003/09/13/click_submit_only_once


-- 
Dave

Reply via email to