Talking points

[David Recordon]

I don't want to be rash and think that there won't be efforts
incompatible with YADIS, but in the end it is a pretty simple concept.
Here is a URL, it points to an XML document, that XML document describes
where to perform queries against supported services the owner of this
URL cares about.

Sure, most of this could be achieved using <link> tags in a HTML
document.  That doesn't however account for various cases such as
someone's URL not being an HTML document.  What if the URL they use is
their resume in PDF format?

>From a service providers view, it is easier and the outcome is better
most of the time to be dealing with an XML document with a known format
than trying to parse HTML.  Especially when we're talking about John
Doe's personal homepage with malformed HTML, his <link> tags in his
second head section, etc.

Sure, XML makes it more complicated, but that also means that John Doe
isn't going to fuss around with creating his own document.  Rather he'll
go to some website that makes the document for him when he enters some
values in a web form, or his host or service provider will provide it
for him.  We've already seen services in the past to generate documents
like this for other protocols.

If for example a basic caching scheme relied on fetching the HTML
document, md5'ing the contents, and comparing it to the previous hash
value to see if it changed.  Their cache would have a pretty low hit
rate since as you've pointed out these technologies are largely being
adopted by bloggers.  Rather pulling it out into a separate document
will increase cache hit rates and reduce processing time.  This also
makes it possible to have multiple different URLs, for your various
personas, point to one service document.  Of course it is up to the user
to decide if correlation is far to easy at that point. ;)

We've already seen libraries, thanks to Jan Rain and NetMesh, in
multiple languages for both OpenID and YADIS.  We want to keep this
stuff simple, easy to understand, easy to implement, and make a bunch of
individual pieces each that have a unique task in weaving together some
part of this metasystem.  I strongly believe that mixing authentication
and reputation, and now service discovery, into one protocol is not the
best way to approach this problem.  Rather keeping these as separate
pieces, with a service that ties them all together, is what will be the
most flexible and ultimately the most successful.

I'd love to see us find convergence on something so simple.  While I
know OpenID and YADIS are not perfect, Six Apart has shown that even
when there may not be agreement on something, it is quite powerful to
enable over eight-million users as we did when OpenID launched. Today we
estimate this number to be close to double that; we plan to do the same
with YADIS.

--David


-----Original Message-----
From: yadis-bounces at lists.danga.com
[mailto:yadis-bounces at lists.danga.com] On Behalf Of Dick Hardt
Sent: Thursday, February 16, 2006 6:34 PM
To: Eugene Eric Kim
Cc: YADIS List
Subject: Re: Talking points


On 16-Feb-06, at 10:57 AM, Eugene Eric Kim wrote:

>   * Enables you to support OpenID, LID, and i-names in a fairly
>     transparent way.  Hopefully others (like SXIP) will jump on board;
>     at minimum, they're participating in the conversation.

No one has addressed my questions at this point.

>   * With YADIS, you don't have to wait for further convergence.
>     Implement now.  Should be forward-compatible with future specs.

I would disagree with this.

I think there is some convergence on this list of people that see the
same part of the elephant, but there are lots of other parts of the
elephant IMO that are incompatible with directions here in YADIS.