[Python-Dev] 2.3.1 is (almost) a go

[Skip Montanaro]

    >> Many are deserting what at times appears to be a sinking bsddb ship
    >> in favor of pickles...  For long-running SpamBayes apps like
    >> pop3proxy and the Outlook plugin, pickles make sense if shared access
    >> to the scoring database is not needed.

    Anthony> :-(

    Anthony> I haven't been following the spambayes lists too closely. Are
    Anthony> there concrete problems with bsddb that are cropping up, or
    Anthony> just a general wariness of it?

It's a combination of things.  First and foremost, using the pickle storage
is much faster assuming your database is not too huge and you have a
long-running spambayes app.  Second, people still use pre-2.3 Windows
versions of Python to run their spambayes apps.  By default, these people
get dumbdbm by default.  That's a disaster waiting to happen because of
storage size and speed concerns.  There were also bugs in dumbdbm (which
have since been fixed).

    Anthony> If there _is_ a problem with bsddb, it needs to be
    Anthony> addressed. Too many things depend on it.

I think it's more that bsddb is an innocent bystander in a drive-by
shooting, though some folks have been seeing DB_RUNRECOVERY (sp?) errors.
The anydbm-style access to Berkeley databases doesn't provide the necessary
locking for multiple process (or multiple thread?) access.

I agree if there are problems they should be fixed.  One step in the right
direction would probably be to create a default DBEnv for use by the
bsddb.*open() compatibility calls.  Also, the docs should mention threading
issues for people using the more featureful API so they don't unwittingly
shoot themselves in the foot.  (Maybe DBEnv=None should not be allowed in
open calls if threads are enabled?)

Skip