From: Brad Templeton (brad@templetons.com)
Date: Mon Apr 30 2001 - 00:01:51 CDT
On Sun, Apr 29, 2001 at 09:53:37PM -0700, Russ Allbery wrote:
> The correct way to support extensibility is not to make up as many
> different solutions as we can think of and throw them all into a standard
> in the hope that when one of those unforseen problems comes up it will
> happen to luckily coincide with one of those solutions.
You will never perfectly predict what future tools will need, so you have
no choice but to overdesign. However, you do it in a way that's as simple
as you can -- but no simpler.
>
> Show me a useful local header. Show me someone who wanted this feature
> and couldn't find any other, better way of solving their problem. Once we
Well, when I was at ClariNet, we used local headers extensively. That's
because our process created lots of useful inforamtion for use by our
editors, who read the news using a newsreader, and could then pipe it into
other tools etc.
So we created rafts of headers that started with C-. And we programmed
our NNTP feeding tools to strip C- headers from any article leaving our
network, but not to strip it within the network.
It seemed to me this had general value. Injectors can put lots of audit
trail info into local headers for convenient access by anybody on the
local system if that is the goal. Of course, the Xref header turned out
to be very handy for various news tools, and eventually even got networked
for use in mirrored spools with identical article numbers. Who knows
what else people might want to include locally only.
The only thing the spec needs to say is to either strip 'em when you feed
them out, or more simply, strip then when the article comes in where the
site is not your site.
The other thing local tells you is that if there is a signature system, you
never include local headers in the signature, unless you have your own
local signature, which would of course not need to follow the spec.
> Xref doesn't count as a local header, since Xref also exists to support
> slaving and therefore can't always be deleted when it leaves the news
> server.
That's why the definition of "local" would include a domain name (with
possibly subtree) so tools can do exactly that.