Agreed. It took me a few hours to incorporate code from file(1) into the
NCSA httpd so that it can check the magic number of a file (and set the
content-type: accordingly) before sending the file out. There is no reason
for the web to continue to rely on solely filename overloading conventions of
an archaic operating system. It can provide clues as to content-type in the
absence of other, mnore reliable information, but it is by no means a robust
scheme.
Including a hard-wired content-type in a link seems to defeat the purpose of
the HTTP negotiation algorithm, although, as in overloading the filename, it
can be of utility when all else fails.
-marc