re: initial message

[Konstantin Riabitsev <icon linux duke edu>]

On Tue, 2003-09-09 at 12:04, Mark Finlay wrote:
> > 1. That's not good. We haven't run into that yet, but will shortly, at
> > least in a few "shared-user" situations. Any response from hp on how
> > that situation is supposed to work?
> 
> It was on havoc's 2.4 list of things that should be fixed, but it never
> got fixed.

Well, there is a "GCONF_LOCAL_LOCKS=1" option that is a workaround
proposed by Havoc, but we've been able to achieve instability during our
abuse tests. I've re-run these tests recently, since we've had a bout of
user complaints about gconf errors, and I've been more successful, but
perhaps I've just been lucky.

GCONF_LOCAL_LOCKS does help the situation somewhat but it still causes
problems both in terms of possible preferences corruption and stability.
I'd much rather see a mode of operation that does not use a global lock
-- i.e. filesystem locking just on the %gconf.xml file currently being
modified. This would make much more sense and play much, much better
with roaming user configurations.

Regards,
-- 
Konstantin Riabitsev <icon linux duke edu>
Linux DUKE