YADIS as an abstraction layer

[Martin Atkins]

Lukas Leander Rosenstock wrote:
> Hello!
> When reading all your posts I wondered why nobody got the idea of 
> agreeing on a filename for identity meta data. Let's call this file, 
> whatever it may be at this point, "yadis.xml".
[snip]

The HTML decoupling thing was debated for OpenID as well:
<http://comments.gmane.org/gmane.comp.web.openid.general/718>

Unfortunately, I cannot find the discussion Brad mentions where we 
decided not to use "magic" filenames, but I have my own set of drawbacks 
caused by this approach:

* How do you handle URLs that don't end with a trailing slash? Where is 
the auto-discovery document for <http://joe.mysite.com/aboutme.html>?
* How do you handle URLs with query components? Sadly, some sites 
present horrible identity URLs like 
<http://www.mysite.com/identity.php?user=fred>.
* What about sites powered by software that is simply incapable of 
presenting content at that URL? I know at least one content management 
system that takes over the entire URLspace and uses URLs like 
<http://www.mysite.com/section/section/section/>, reflecting the 
structure of the site and always including a trailing slash. Not all 
sites are delivered from static files on disk.

In short, "magic" filenames open a can of worms since URLs were never 
intended to work like that. Even IE's "magic" favicon.ico is overridable 
by a LINK element, presumably because the designers of that feature 
realised that it is not appropriate nor possible to dictate a particular 
URL structure for all sites.