Sam Ruby has posted some thoughts on content negotiation, in part he says:

I'm also concerned about the implications of this to other caches. Perhaps Aggie should have an option to turn this feature off in case someone is faced with a less than intelligent proxy which caches?

This brought to mind another problem I had encountered with content negotiaion in the context of XML. There is a danger with burying multiple types of documents behing a single URL. For example, if you want to take my RSS feed and transform it using an XSLT stylesheet using the W3Cs XSLT Service how do you tell it to do the content-negotiation? That is, how do I tell the XSLT Service what type of document to negotiate for when it does it's GET to do the transformation. As an extreme example, and only for the purposes of illustration, I could have the URL http://bitworking.org/news/ return, alternately: HTML, RSS and the XSLT sheet needed to convert the RSS into HTML. Now when I went to the W3C Service I would put the same URL (http://bitworking.org/news) for both the XSL URL and the XML URL. Now the W3C Service might do the correct content negotiation. But even if it got this scenario correct, what if I wanted to convert my XSLT style sheet into an HTML page via a third XSLT sheet. As you can see, there are myriad ways these files can be hooked together and without the ability to guarantee the content type you'll eventually end up in trouble.

The last problem with content negotiation is with a lack of a standard way to override the servers selection. For example on RESTLog you can append ?xml or ?html to coerce the return type into XML or HTML respectively. Now adding the query parameter creates a new URL so you aren't doing content negotiation at that point. The lack of a standard or at least a common idiom on how to override this in the URL was driven home to me recently when I went to a site and kept getting the text version of a specification when I really wanted the HTML version. There was no documentation on the site on how to 'force' a content type. There are no common idioms that would guide me in guessing an override syntax. Finally, in desperation, I pulled the HTML version out of Google's cache.

Now none of these problems has stopped me from adding content negotiation support to RESTLog, nor do I think it should be dropped from Aggie. First because RESTLog has the ability to specify the content type via query parameter. The second is that although content negotiation may not be a viable strategy in the long run, I think it's still early and if we provide tools for experimentation, someone may discover a clever solution to the problems I outlined above.