DB_INIT_LOCK deadlocks?

[Matthias Andree]

John <xd890cc2b41c31d74 at f4n.org> writes:

> configure runs fine with patch:
> checking if Berkeley DB supports shared environments... yes

Good. The patch just opens an environment with the subsystems bogofilter
uses (that's why I dropped DB_INIT_LOCK) and closes it again, if
something like that locks up, something's broken.

> I also tried the old DB_INIT_LOCK-version with different versions of
> Berkeley DB compiled from source, but it only added to the confusion:
> 4.1.25, 4.2.52 (!) and 4.4.20 configured OK, but 4.3.29 deadlocked.

Can you report this to the maintainer/packager of the OpenBSD db-4.3.29
package? They should investigate and perhaps contact SleepyCat about this.

> Unfortunately, I don't have time to investigate this properly at the
> moment, so I'll just hope the weirdness is isolated in the locking
> code not used by bogofilter.

It is.

> (Oh, and by the way, mail to the list appears to generate bounces
> for paulk at girlscoutsht.org and one from Mailer-Daemon at fc.vdu.lt
> concerning "beachhead,VDU FCIS". It's great how everyone respects the
> return-path.)

Yup, its concept is so young, not even 25 years, so it's impossible for
everyone to know... I killed both subscriptions since I had the same,
and we'd killed one of the second kind in a Swedish domain a few days
ago (they reported the problem upstream). The former site was using a
b0rked qmail¹ setup, probably a site-specific issue.

I wish I could filter all return messages through Mailman before passing
them on to list subscribers, that way, we could identify and unsubscribe
offending sites (a) more quickly, (b) without letting end users see such
troubles. It might also be some incentive for sites to use standards
conformant software (the idea is: if their users - who have the money -
complain that they're unsubscribed from lists because the site runs
broken software, it might be fixed faster).

-- 
Matthias Andree