Concerning the text/binary distinction:

    I first met such a distinction on the Burroughs B6700.  The Burroughs
file system associated a "file type" byte with each file, there was a byte
for APL Workspaces, there was a byte for BASIC source code, there was a
byte for Algol object code, and so on.  The snag, of course, was that there
wasn't really any way to generate your _own_ file types.

    The Macintosh has the same feature, but generalised to 4 bytes.
I would have thought that it would have made more sense for a file type
to itself be a file system object, so that you could create them ad lib
and attach whatever properties you wanted to them.

    Macintosh tools like browsers and FTP agents typically "fudge" this
information from file name extensions, if they can't get it by other means.
This can have unpleasant consquences.  Windows can be told to associate
types/icons with files by extension as well, and so can some UNIX desktops.

    Since Celeste has to deal with MIME types and HTTP uses MIME types, it
might be simplest if a view of file encoding and type that was as close to
MIME as reasonable could be applied uniformly.  This might require system-
dependent backup, such as mappings between extensions and MIME types.

	Indeed within Plan 9 are the fruits of lots of man-years of experience
	with handling multi-lingual text too!
	
And that's certainly true.