Mailing List Archive

Debian's Perl Update broke Bayes ?
Dear list,

yesterday I've installed via apt-get the perl updates from
security.debian.org. After that update bayes_db seems to
be broken (talks s.th. about untieing etc.).
Anybody on the list who makes the same expirience ?
I've deleted bayes_* and updated SA from 2.61. to 2.63.
Now bayes works fine again ...

Michael
Re: Debian's Perl Update broke Bayes ? [ In reply to ]
Muenz, Michael wrote:
> yesterday I've installed via apt-get the perl updates from
> security.debian.org. After that update bayes_db seems to
> be broken (talks s.th. about untieing etc.).
> Anybody on the list who makes the same expirience ?
> I've deleted bayes_* and updated SA from 2.61. to 2.63.
> Now bayes works fine again ...

No. I have updated many machines and all of those work fine for me.
Perhaps you took the perl version changing update from 'unstable'
instead of the security update 'stable/updates'. The new perl version
change from 5.6 to 5.8 uses a different db format. Updating to the
new perl will have the exact behavior you describe because the
underlying db format has changed.

Bob
Re: Debian's Perl Update broke Bayes ? [ In reply to ]
I experienced the same problem after we applied the perl security
updates for debian stable. Something bad happened with the lock file
and the database was corrupted. I moved the bayes_* files aside and
retrained the bayes database with known spam and ham. It worked fine
afterwards.

I did not have to upgrade spamassassin to fix this. (I do need to
upgrade, though ... )

Here is the debian security annuouncement:
http://www.debian.org/security/2004/dsa-431

-Eric Rz.

On Fri, Feb 06, 2004 at 03:27:50PM +0100, Muenz, Michael wrote:
> Dear list,
>
> yesterday I've installed via apt-get the perl updates from
> security.debian.org. After that update bayes_db seems to
> be broken (talks s.th. about untieing etc.).
> Anybody on the list who makes the same expirience ?
> I've deleted bayes_* and updated SA from 2.61. to 2.63.
> Now bayes works fine again ...
>
> Michael