RE: tag: uri scheme

I disagree with the assertion that, for HTTP, "by and large 
the whole thing doesn't work very well)".

First, it usually isn't "authors" who personally assign MIME types
to anything. Content is written by software applications, usually,
and software applications generally are set to at least generate
file types or file extensions where the file extension is for the
locally appropriate file type -- otherwise the software wouldn't
function for the authors when they went to reopen the content.

MIME types are generally assigned by the HTTP servers, of which
Apache and IIS are the most popular. Perhaps you might want to
argue the number of "Major web servers" vs the number of 
"Major browsers", but I think there are more browser instances
than there are server instances, and the statistics are actually
much more skewed as to sites and pages served (a small number
of sites are responsible for a large proportion of pages retrieved,
while it isn't true that a small number of browser users are
responsible for a large proportion of pages retrieved.)

This is important, because the difficulties experienced with
MIME type assignment are mainly ones of configuration, not
software capability. There were some earlier versions of Apache
that would serve unknown file extensions as text/plain instead
of application/octet-stream, but that was a configuration error.

In general, it is fruitless to write standards that try to mandate
behavior for software, organizations, or configurations which have
not previously followed standards, because there is no indication
whatsoever that they would follow the new standards any more than
the old ones; if you merely write standards to describe current
behavior, there's no guarantee that the current behavior won't
continue to drift, since the organizations involved have no more
incentive to keep to the new standards any more than they did the
old ones.

So the issue isn't "authors", it is "software that authors use",
and there's no reason to believe that package-generating
software would do any worse generating correct MIME types than
they would generating correct ZIP files.

Larry
-- 
http://larry.masinter.net


=============================
Larry Masinter wrote:
> I'm not sure about 'authoring might be more complicated',
> though. The author/sender/creator of a package has a lot more
> insight about the types of the components of the package
> than the recipient, and if there's any guesswork to be
> done, putting the burden on the author would seem to be more
> stable and effective for the overall communication system.

That strongly depends on the relative numbers of authors and recipients, 
their relative cluefulness, and their relative resource availability... 
  For example, if your authors will largely tend to get their MIME types 
wrong and there are lots of them and there are only three possible 
recipients, all of whom are willing to put in the sort of work the 
authors aren't, the tradeoff might lie on just having the recipients deal.

This is not to say that the tradeoff might not fall the other way too, 
but authoring certainly _is_ more complicated if authors have to choose 
their types themselves (witness HTTP, where by and large the whole thing 
doesn't work very well).  That might be a sacrifice that's worth it in 
the interests of other things, naturally.

-Boris

Received on Thursday, 22 January 2009 04:03:32 UTC