[Spambayes] Corrupt database

[Skip Montanaro]

    Tim> So people running a high-volume filter daemon on a server-class
    Tim> machine would be better off with a pickled dict (they don't care
    Tim> about startup or shutdown time, have plenty of RAM, and scoring
    Tim> speed matters in high-volume applications).  

The one exception I can think of to this is if the training and scoring
operations are in separate processes.  For example, you might have scoring
done by a sendmail or postfix (or Exchange :-) plugin, while training is
handled by the user with a separate program.  In that case, you probably
want to use a real database.

To put this in context of the current SB apps, the Outlook plugin, POP3
proxy and IMAP filter should probably default to using pickled dicts, while
sb_filter should default to using a database file.  I don't know that this
is the case right now.  I think the choice of classifier type is
application-independent.  Maybe we should change that.

Skip