Is the Atom Publishing Protocol the Answer?

Are Atom and APP the answer to everything?  Easy one: No.

Dare Obasanjo raised a few hackles with a provocative post (Why GData/APP Fails as a General Purpose Editing Protocol for the Web).  In a followup (GData isn't a Best Practice Implementation of the Atom Publishing Protocol) he notes that GData != APP.  DeWitt Clinton of Google follows up with a refinement of this equation to GData > APP_t where t < now in On APP and GData.

I hope this clarifies things for everybody.

There seems to be a complaint that outside of the tiny corner of the Web comprised of web pages, news stories, articles, blog posts, comments, lists of links, podcasts, online photo albums, video albums, directory listings, search results, ... Atom doesn't match some data models.  This boils down to two issues, the need to include things you don't need, and the inability of the Atom format to allow physical embedding of hierarchical data.

An atom:entry minimally needs an atom:id, either an atom:link or atom:content, atom:title, and atom:updated.  Also, if it's standalone, it needs an atom:author.  Let's say we did in fact want to embed hierarchical content and we don't really care about title or author as the data is automatically generated.  I might then propose this:

    <id>tag:a unique key</id>
    <author><name>MegaCorp LLC</name></author>
    <updated>timestamp of last DB change</updated>
    <content type="application/atom+xml">
        <feed> ... it's turtles all the way down! ... </feed>

Requestors could specify the desired inline hierarchy depth desired.  Subtrees below that node can still be linked to via content@src.  And when you get to your leaf nodes, just use whatever content type you desire.

Alternatively, if you need a completely general graph structure, there's always RDF.  Which can also be enclosed inside atom:content.

The above is about as minimal as I can think of. It does require a unique ID, so if you can't generate that you're out of luck.  I think unique IDs are kind of general.  It also requires an author, which can be awfully useful in tracking down copyright and provenance issues.  So that's pretty generic too, and small in any case. 

Of course this type of content would be fairly useless in a feed reader, but it would get carried through things like proxies, aggregators, libraries, etc.  And if you also wanted to be feedreader friendly for some reason, the marginal cost of annotating with title and summary is minimal.