[Pkg-mailman-hackers] mailman: fix for 412142 gives unicodify_archives traceback

[Lionel Elie Mamane]

On Wed, Apr 04, 2007 at 09:58:06AM +0800, Paul Wise wrote:

> The other day due to hardware issues, we migrated
> lists.indymedia.org data from mailman 2.1.7 to 2.1.9.

I presume you installed 2.1.9-7?

> Unfortunately, the unicodify_archives script gave a traceback:

:-(

Analysis: When loading the index to be converted, the said loading
fails for some reason or another (line 168 or 170 of HyperDatabase.py
raises an exception). This leads to the system resorting the index
that failed to load, exactly the operation that fails before the index
is unicodified... Actually, we don't need the sorted version of the
index (we'll resort it after unicodification), so one solution would
be to add a parallel loading path that never tries to sort ("leave
dirty" or something like that). But that is ugly and a maintenance
nightmare.

I'd like to know what exception is raised by lines 168-172 of
HyperArchive.py, to see if _that_ can be avoided rather than the ugly
solution.

Another (ugly, admin-specific) work-around is to change the default
python encoding to something that will always succeed (that is,
defines a total mapping from all strings (byte sequences) to unicode,
e.g. windows-1252). This may change the results of the
unicodification, though (because it tries the default encoding
first).

> I can provide the archive/database for the three public lists -
> cmi-brasil-contato has private archives. Please let me know which files
> are needed.

A tar of the /var/lib/mailman/archives/private/<listname>/ directories
would guarantee I have all the data to reproduce the problem.

> it might be nice to fix this case for etch, if there is time left.

I doubt this, but we can try anyway.

-- 
Lionel