One site-meta to rule them all

Eran & Mark's site-meta proposal is... interesting. I have a gut feel that it's a bit like democracy: The worst method for whole-site discovery, except for all the others. The killer app for this, IMHO, is the ability to solve things like "lookup metadata for user x on domain y" in a general way. An immediate practical application is translating things like email addresses and Jabber IDs, all of the form (user@domain), to something you can perform discovery on.

Other than the embarrassment of hacking another hard-coded magic name alongside favicon.ico and robots.txt, I really only have one issue with the proposal: It requires a directory lookup via an XML document. I have nothing against XML, but it seems like overkill for this purpose.

An alternative would be to use a very, very simple text-based format that is NOT very extensible. Fortunately, there's already a proposal for this type of format from Mark, for a Link header:

Link: http://example.org/ch rel="previous";
title="previous chapter"
Just for simplicity, we can take the same format and embed it in an application/site-meta document. The sample site-meta XML would then transform into something like this (with newlines delimiting each type of metadata):
/robots.txt rel="robots"
/p3p.xml rel="privacy"
http://other.example.net/example rel="http://example.com/rel"
We lose the ability to use namespaces and inline (embedded) metadata in this site-meta document. Or alternatively, we gain the ability to ignore namespaces and we don't need to download inline metadata we don't care about.

One closing thought: A persistent issue with XML, unfortunately, is problems with cryptographic signatures, primarily due to complexity of signing and verifying something whose canonical representation is really an Infoset rather than a text stream. This hypothetical format, or any of a number close to it in configuration space, could solve that problem easily:
/robots.txt rel="robots"
/p3p.xml rel="privacy"
http://other.example.net/example rel="http://example.com/rel"
application/x-pkcs7-signature;base64,iVBORA...rkJggg== rel="signature"
...with some simple rules to determine which octets get signed.


  1. +1 especially since if sites are going to have a single /site-meta file for all their discovery, then as soon as one person fat-fingers invalid syntax for an entry, it will break discovery for the entire site. XML is very brittle. But with one-line-per-entry syntax, presumably the damage would be limited and more easy to ignore.

  2. Hey John,

    I like the democracy comparison -- well said.

    WRT XML - you're not the first person to raise this, and I agree that it adds some risk.

    The intent, however, is to address as many of these use cases as possible, and I often run across people for who the extra roundtrip is unacceptable (e.g., in P3P, this concern drove the whole compact policy design, which IMO is pretty broken). Since site-meta is aimed at addressing as many of these uses as possible, that's why XML was chosen.

    However, if there's emerging consensus that a line-oriented format would be better, I'm all for it; the whole point here is to build momentum behind one approach, and make it successful. So, get people to make some noise!

    P.S. I'm re-submitting this comment, because the first time the openid dialog ate the comment text. Grr...

  3. This comment has been removed by a blog administrator.


Suspended by the Baby Boss at Twitter

Well!  I'm now suspended from Twitter for stating that Elon's jet was in London recently.  (It was flying in the air to Qatar at the...