> Date: Fri, 15 Jan 93 22:47:39 CST
> From: Dan Connolly <connolly@pixel.convex.com>
>
> I thought NNTP could replace HTTP wholesale too. The irreconcilable
> difference between a news article and a WWW node is that a WWW node
> is editable. It may change over time. [I don't like this strateby,
> bit that's what's in practice.]
Irreconcilable? There are a few other problems with using NNTP as
the caching machine, the most basic being that there is no way of
getting from the news article ID to its archive site.
> Thus you have the question of versions, locking, the "home address"
> of a document, etc.
Suppose we adopt a model of having names/addresses for BOTH the fixed
article-like version AND for the logical live document. [The version
relationships can be expressed nicely with the LINK element and a
link type in HTML -- do you wnat to move this into HTTP?]
> Perhaps we could model WWW nodes as sets of articles -- a thread,
> for example, so that each time you edit a node, you generate a
> new ID.
Yes exactly.
> The problem then is, what do you use for the name of the thing?
> (for linking purposes, that is.)
Well, you can use whichever you mean. Sometimes I migth want to
refer to a particular version, sometimes to the live document its
genericity.
You need a server function to turn a link to an anchor in a specific
version into an anchor in the live document.
You need a server function to return a list of available versions of
a live document. An exaple of this is an FYI which is live and an RFC
which is dead. At any one time, an FYI is associated with a given
RFC. You can quote either. If you quote a paragraph number, then it
is wise to quote the RFC. If you quote it as a generallt useful
article to read, then quote the FYI number.
> Besides all that, WWW uses addresses -- article lodators, rather
> than article identifiers. A WWW client has no /usr/spool/news
> database to consult to get all its stuff. It can look at
> ftp sites, gopher hosts, etc.
This in only because we haven't got a good nameserver yet. The Udi
can become a name when a naming scheme exists.
> If URN's ever come to town and all the stuff on all those servers
> share a namespace like the usenet message-id namespace, then
> we may have a chance to play the game that way.
Yes. When we have an algorithm to dereference a URN.
> But in the mean time, there are several factors that motivate HTTP.
>
> I certainly agree HTTP should look a _lot_ like NNTP.
So do I! This why in the "Expires:" field for example, I tries to
use NNTP terminolgy but worry that we mustn't overload NNTP
vocabulary with different meanings and so pervert it. We must be
careful there is no semantic clash. For example, when a news article
Expires: then does it mean that the article itself vanishes or just
the cached copy? Some new articles really expire (the thingsthey
talk about happen and have gone) and others are important history but
still the caches throw them out.
> Dan
Tim