HTTP header/content precedence

[Walter Underwood <wunder@xxxxxxxxxx>]

I'm quoting a message, but modifying this to a new issue.

--On Friday, July 9, 2004 10:06 PM -0400 Mark Pilgrim <pilgrim@xxxxxxxxx> wrote:
> Also note that the XML specification does specifically state that
> xml:lang (if present) overrides the external language identification.
> This is notable in that it is different in some cases from the
> precedence rules for character encoding (RFC 3023), of which we are
> all now painfully aware.

We have hit a few situations where Atom needs to resolve conflicts
between HTTP headers, content, and even the XML spec. It would be
nice to resolve all of these the same way, but that might not be
possible. Regardless, we should list them all in the spec. Here is
what I have so far.

Language: xml:lang overrides HTTP content-language

Media type and character encoding: use application/atom+xml so we
can let the XML spec rules handle it and bypass HTTP rules (override
within the rules)

Expires/refresh-rate: no decision, need to resolve HTTP caching
mechanism and historical rate-oriented RSS hints

Did I miss any? In general, it looks like content overrides
headers, which is opposite to the HTTP transmogrifying gateway
assumption. HTTP is designed for gateways which can transform
content on the fly and modify the headers to mark that decision.
These seem to be quite rare in practice, probably because they
mess with the end-to-end design principle which underlies much
of the internet design [1].

wunder

[1] <http://web.mit.edu/Saltzer/www/publications/endtoend/endtoend.pdf>
--
Walter Underwood
Principal Architect, Verity