Re: ACTION-472: New Mime-web-info draft

Nathan wrote:
> 
> Thus, a media type would always be identified by a URI, and when 
> referring to a media type one could use the token for registered 
> types, and full URI for unregistered.
> 

Just to point out again, there's no such thing as an "unregistered
media type"; there are strings which *are* media types, as they appear
in the registry, and there are strings which *are not* media types, as
only what appears in the registry is, by definition, a media type.  So
a URI will never be a media type, because media types aren't first-
class objects (they're registry entries).

Allowing URIs as tokens will immediately result in Content-Type
referring to data types, instead of processing models, by insinuating
that media types are first-class objects on the network.  As Web
architecture stands, media types simply don't work like that, so they
can't be assigned URIs.  What processing model is described by this?

Content-Type: http://tools.ietf.org/html/rfc3023

Even if you give a media type a URI, since media types aren't first-
class objects, it will always be just as ambiguous as using the URI of
the data type.  Unless the entire existing system is scrap-heaped and
rewritten such that media type definitions like RFC 3023 are only
allowed to reference one media type, or change things such that data
types have 1:1 identifiers instead of being part of a family -- except I
thought the purpose here was fixing the registry, not re-architecting
the Internet.  ;-)

The best way to make the problem of *anything* appearing in Content-
Type _worse_, is to allow URIs as tokens, since that requires media
types to be recast as first-class objects -- which they're not.

-Eric

Received on Monday, 31 January 2011 19:14:06 UTC