I don't think this is relevant: http should be kept simple and data-type
independent, leave out the higher level semantics. Then 0 based
addressing is the most sensible. Even for chapters, the argument is
weak: what chapter number is the title page? What chapter number
applies to appendicies? Does the number then need to be a string that
names a sub-entity? This is a Pandora's Box that should stay closed.
Any thought on how this should interect with dynamic computed documents
(CGI-bin scripts)? Supporting range addressing of computed documents
would require either re-computation on each fetch, or caching. If
re-computed, how do you guarantee consistancy? Imagine fetching a
document one byte at a time that contains the server's load average.