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

Exactly.  The Content-Type header can contain *anything*, so there
needs to be some list of tokens which are officially recognized as
media types, which correspond to processing models which have been
vetted by peer review (particularly for security considerations, which
requires a trust model).

If we used URIs, we'd still need some sort of list of those URIs which
actually point to conforming media type definitions, which correspond
to processing models which have been vetted by peer review (also to
prevent naming collisions, like with application/rss+xml).  But, URIs
are too opaque, preventing us from knowing at a glance whether the
payload is an image, or an XML derivative, etc.

So, if we do away with the registry in favor of URIs, how do I know
that the URI I'm looking at actually resolves to a document that
defines a media type?  Whose word can I take as to the security
considerations, aside from the author?  If the URI points to multiple
processing models, how do I know which one to choose?  How can I search
for all XML-derived media types?

What it comes down to is this:  If you're going to have headers in a
protocol, you're going to need registries, because you'll never prevent
folks from sending *anything*, or prevent other folks from needing to
know whether *anything* is valid.

>
> Surely we could just use the URI of the spec (dated/versioned)?
>

No, because data-type identifiers aren't processing models.  What's the
processing model for this spec, and how would I express an intent that
it be processed as plaintext?

http://www.w3.org/TR/xhtml1/

Some sort of token representing a processing model that applies to one
or more *families* of data types, is required by the architecture.  I'm
not in favor of any proposed alternative to the registry, which removes
the feature of being able to send HTML as plaintext -- do you see how
bypassing media types in favor of links to data formats, winds up being
a different architecture altogether?

I could go on for, well, months apparently...

-Eric

ashok malhotra wrote:
>
> I believe a registry implies some oversight.
> All the best, Ashok
> 
> Jonathan Rees wrote:
> > Nathan wrote:
> >> This is possibly a stupid question, but why even have a registry,
> >> surely we could just use the URI of the spec (dated/versioned)?
> > Not a stupid question at all.  The TAG even has a name for the
> > question: ISSUE-50. (Oddly, the issue's home page has a URI, but the
> > issue itself doesn't have a URI.)
> >
> > Jonathan
> >
>

Received on Monday, 31 January 2011 03:48:45 UTC