anomalous subjects

[Tom Anderson]

From: "Matthias Andree" <matthias.andree at gmx.de>
> "Tom Anderson" <tanderso at oac-design.com> writes:
>
>>>> However, I don't know if this is coincidence or not, but just four
>>>> messages later, I had this error:
>>>>
>>>> bogofilter: (db) db_get_dbvalue( 'aperiodic' ), err: -30982,
>>>> DB_RUNRECOVERY: Fatal error, run database recovery
>>>> bogofilter: (db) db_close err: -30982, DB_RUNRECOVERY: Fatal error, run
>>>> database recovery
>
>> From: "Matthias Andree" <matthias.andree at gmx.de>
>>> Strange, haven't seen these in ages. Are you using NFS or something?
>>> Multiple wordlists?
>>
>> No on both counts.  It's a relatively standard, minimal Gentoo install, 
>> with
>> Sendmail, Procmail, Apache, and other common mail/web server software.
>> /home, /var, and / are reiserfs while /boot is ext3.  I noticed an oddity
>> with reiserfs in another thread... might this be related?
>
> No for reiserfs 3.6 in any 2.6.X release kernel and recent 2.4.X release
> kernels.

I'm using reiserfs 3.6.18 on kernel 2.6.9, so I guess that isn't what would 
have produced the error then.

>> BTW, since I upgraded my server, which included the change in filesystem
>> type along with an upgrade in bogofilter from 0.17.x to 0.92.8, I've 
>> noticed
>> a severe crippling of bogofilter's effectiveness.  Not sure why.  I hope
>> that ongoing training will fix it relatively quickly.
>
> Also strange. Have you kept a textual dump (bogoutil -d) of your
> database that you could compare (wdiff or diff) to a current dump?

It seems my current one was severely truncated (from 1M tokens to 91k 
tokens) at some point.  This may have nothing to do with bogofilter, as 
perhaps it became corrupted in another way, so there's no need to pursue 
this here.  I'm just going to continue training on error in order to rebuild 
my wordlist from my backup.

However, this may be of interest:  running "bogoutil -d | bogoutil -l", 
which was expected to compact the file, instead added 1.5M!  Is this normal? 
Has new information been added to the wordlist between 0.17.x and 0.92.8? 
Both versions pass db_verify.

Tom