[ZODB-Dev] BTree corrupted after conflict resolution

[Jeremy Hylton]

On Wed, 2004-03-03 at 19:34, John Belmonte wrote:
> Dieter Maurer wrote:
> >   During conflict resolution the state does not contain
> >   the persistent subobjects themselves but only references.
> > 
> >   Therefore, your "__cmp__" cannot be used to compare
> >   the persistent subobjects. Almost surely, the default
> >   "__cmp__" is used which is, of course, wrong and
> >   can lead to corruption.
> 
> It seems you are right.  If I change the key class to a second-class 
> object, the problem goes away.

I'm glad we understand it, now.  (Thanks, Dieter.)

> > Almost surely, we must add a further restriction for BTrees:
> > the keys must not be persistent objects.
> 
> I think that would be an unreasonable restriction.  If the server 
> context can access the _p_conflictResolution method of a class, why not 
> __cmp__, etc.?

_p_conflictResolution should really be a staticmethod, but it was
invented before Python had staticmethods.  That is, the method is passed
three object states but it isn't supposed to access self.  Furthermore,
it only sees a single object state, it does not have access to the state
of any other first-class persistent objects.

Perhaps we should define an __cmp__() method that raises an error on the
stub objects that conflict resolution provides.

Jeremy