[NNTP] Article number wording

[Russ Allbery]

Thomas Gschwind <tom at infosys.tuwien.ac.at> writes:

> I perfectly agree that this should be specified.  At the same time,
> since we are aware of the fact that article numbers will eventually
> overflow, I think a solution SHOULD be pointed out in the RFC.
> Otherwise the RFC renders itself obsolete before we have a better
> solution and everybody will come up with a different solution.  This I
> would consider a bad thing.

I continue to disagree with the perceived severity of this problem.

> I think the best solution would be to recommend news servers
> implementors to limit their article numbers to unsigned 32-bit numbers
> for the next 2(?) years and at the same time, we should recommend news
> client implementors to update their clients to handle unsigned 64-bit
> article numbers.  This gives them 2 years time which to update their
> clients which should be sufficient.

Flag days are strongly discouraged in IETF standards, mostly because they
just don't work very well.

> I don't think using an extension will provide an adequate solution since
> again old clients would be blocked from reading news unless they
> implement the extension (which they don't) or unless the server provides
> two different numberings (in which case the extension is obviously
> unnecessary).  Please correct me, if I am wrong.

They would only be blocked from reading groups that had overflowed 2^32-1,
not all groups.

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>