The longdesc lottery. Mark Pilgrim is now writing for the WHATWG blog. Here he makes the case for replacing the longdesc attribute with a better solution, based on ten years of developer ignorance and misuse. As always with that site, check the comments for a microcosm of the larger debate.
Sign in with 
OpenID
Seems to me the third disqualification rule is bogus. A longdesc should be allowed to point to the same URL as the src points to. You can't disqualify the URL as not containing a description unless you either request that URL with a text/* type as the most acceptable, or if you requested it before as an image and it didn't provide a "Vary: accept" header. Content Negotiation is not very practical (hard to setup on the server, a pain to work with on the client), and probably not very commonly used for longdescs, but not grounds for automatic disqualification.
On the same subject, I wonder why Firefox's default Accept string lists image/png as being a more preferable content type than text/html. It can't be that much harder to render HTML than display an image and the user experience is usually better for a full web page than a single, static image. But Firefox's awareness of the possibility of multiple content types per URL seems lacking in general, so maybe it's just an oversight.