Gmane
From: Brion Vibber <brion@...>
Subject: Re: New mail servers leave
Newsgroups: gmane.science.linguistics.wikipedia.technical
Date: 2007-01-10 17:40:28 GMT (1 year, 33 weeks, 20 hours and 29 minutes ago)

Mark Bergsma wrote:
> Mark Bergsma wrote:
> 
>> Ok, so the problem is that there is indeed a redirect from the old 
>> address to the new address, but some message article numbers seem to 
>> have changed by the rebuild of the archive - at least for this month.
>>
>> However if I look at LSS for December 2006, those addresses seem to 
>> work. Hopefully the same will hold for all or most of the older messages 
>> in the archive.
> 
> Ok maybe I didn't look that well - for previous months you don't seem to 
> get a 404, but the message IDs seem to refer to totally different 
> messages - which is not too helpful.

The short answer is that Mailman's archiving system sucks; we've had
problems sometimes in the past when rebuilding archives and it's pretty
annoying.

Since it assigns URLs on the basis of a count of messages from the
beginning of the list, a rebuild can change *every number after* some
particular weird message which gets processed differently by a new
version of the software.

An example I encountered a couple years ago was forwarded messages; the
embedded 'From' header got counted as a separate message in earlier
versions of Mailman from our olden days, but not by newer versions (or
something along those lines), breaking the numbering when the lists got
rebuilt after a couple years.

With a lot of tweaking and rebuilding, it is sometimes possible to
restore the old numbering by removing or inserting fake messages. :P

Not sure how worth it it is.

(What I'd *prefer* to see is a stable archiving system which generates
URLs based on internal properties of the message or an explicitly stored
ID number at receive time, so they aren't dependent on what else is in
the archive at rebuild time.)

-- brion vibber (brion @ pobox.com)