I think this is the exact same situation with characters, words, and
SGML tags. What does the character 'A' reference? Or the word
'bunny'? Or <H1>? How these are translated into images or interfaces
depends on your client. Some clients might read the text to you, some
will present words on a display with a serif or san-serif font. How
the client deals with a reference to 'directory icon' should depend on
some negotiation between the client, the display environment of the
client, and the organization that defines 'directory icon'.
> Viewing standard entities as URNs wouldn't change anything in the
> browsers, of course. A browser that accidentally tries to do a GET on
> such a URN would receive a placeholder, just as the entity definitions
> in the SGML standard are just placeholders.
I don't see why the browser that does a 'GET' on a URN for 'directory
icon' (however that URN is written) shouldn't in fact get the bitmap
for the directory icon. Maybe 'GET' needs more arguments, that's all.