First Atom Publishing Protocol Interop Notes

So the first Atom Publishing Protocol Interop meeting took place over the past two days. The event was hosted by Google, and I'd like to extend a big Thank You to DeWitt and the rest of the Googlers for hosting this event. The food and facilities were great, though it appears 20 guests in a single room seemed to tromp the guest wi-fi network.

It was great to meet so many people face to face after all this time. See Tim's post for a good set of pictures, plus write-ups from Keith Fahlgren, Dave Johnson, Joe Cheng, James Snell, and John Panzer. Hat tip to John, with Sam Ruby not able to make it, he and I were the only two people from the original Atom meeting almost four years ago.

Down to interop. The executive summary: It's all good.

The slightly longer summary:

We had interop a majority of the time. The fact that at the very first interop meeting you could pick any two implementations at random and have a good chance that they'd work together is fantastic.

Here is the interop grid, with checkmarks where we had interop between a client and a server. Note that there are notes in the grid for both successful and unsuccessful pairings. With 14 serves and 9 clients it just wasn't going to be possible to fill in all 126 combinations in two days. That fact that we had that many servers and clients before even getting an RFC number is very encouraging.

While we did have some cases where we didn't achieve interop, the best news is that none of the problems was due to an ambiguity in the spec.

There were several areas that were consistent sticking points.

Authentication: There was a much wider variety of authentication schemes than most people expected. I'll correct James here and point out that the APP Test Client is built on httplib2, which supported all the authentication mechanisms there, including WSSE.
Relative URIs: My server implementation was considered "evil" by some since I used relative URIs everywhere I could. The "evil" was all in joking and I have to agree with Tim about the attitude in the room. We had a general discussion about relative URIs and xml:base and everyone who didn't support them went on to add support right there in the room, or to put it on their to-do list.
Semicolons, who knew?: The semicolon I added to my service document URI was not intended in anyway to be "evil", but it caused at least one client to fail "spectacularly", to quote that clients author.
XHTML: My APP Test Client can send summary and content as XHTML, which caused problems for two servers that I know of, with one of them dropping all the markup while preserving the text. Back to the attitude, this was logged as a bug, and it will be fixed.

In all, the meeting was incredibly useful and productive and everyone agreed that we should have these face to face sessions every six months to a year, and that we should also have "virtual" interop sessions once a month. Thankfully DeWitt volunteered to herd those cats.

What next?

This was a question I got alot; "After APP gets an RFC number, what comes next?"

Part of the answer is logistical and it was great to have Lisa Dusseault there to answer those questions. The WG will cease, but the mailing list will continue to operate. Extensions can be brought forward as individual submissions and discussed on the list before asking the IETF to raise them to Informational, Experimental, or Proposed Standard. As for the future of APP itself I see useful work going forward on:

Extensions: I think there are a bunch of useful ones that could be added, such as a standard way to link from an OpenSearch document from a service document. Navigating around a small collection via paging is okay, but when you get to collections with members in the thousands or millions, search is going to be primary means of navigating the collection, and OpenSearch is the drop dead obvious way to do that.
Profiles: now there seem to be two schools of thought on this, either a profile for a 'workspace', i.e. these collections represent a blog, or a more focused profile for collections, i.e. this is a blog entry collection, this is a collection for links, this is 'trash', etc.
Relationships: Also there was a lot of interest in links among collections, i.e. entry 'A' has a link to collection 'B', where 'B' is a collection of all the comments on entry 'A'.

All in all it was a great meeting and I look forward to the next one.