Comments on: More on human targetting

By: Mark Baker

Mark Baker — Wed, 26 Sep 2007 01:48:06 +0000

I’ve used Accept in the wild, but you need to keep in mind that servers are free to ignore it (and virtually all of them do), so you still need to manage unknown content types.

By: Bill Burke

Bill Burke — Fri, 21 Sep 2007 03:56:49 +0000

You talk a bit here about the client setting up handlers to process responses, but what about having the client request the response type with Accept header? From your experience, is this header generally feasible or not?

Thanks….

By: Mark Baker

Mark Baker — Mon, 26 Feb 2007 13:15:25 +0000

Actually, practically all services that use the two */xml media types do so incorrectly, because they assume the recipient will interpret the root namespace as the real indicator of the vocabulary, which isn’t licensed by RFC 3023 (I wrote the section that warns about this interpretation). Those services should be using more specific media types, like, say (the fictitious) application/stockquote+xml.

By: Stelios Sfakianakis

Stelios Sfakianakis — Mon, 26 Feb 2007 09:11:09 +0000

On the self descriptiveness of HTTP messages: Looking only to the content-type is not enough otherwise Google would just say that GData queries return text/xml (or application/xml) instead of specifying all the details e.g. as in http://code.google.com/apis/gdata/elements.html .

So the description of the messages format either in human readable form (as in GData) or in a machine friendly form (as in WSDL) is required before the client can do something really useful with your application. Of course in the case of REST you can have some basic interaction supported from the beginning even without knowing the details (format, “semantics”,…) of the returned messages e.g. you can use your browser to inspect the messages or have a generic HTTP client (e.g. Joe’s httplib2) download and store locally the representations for you…