[ZODB-Dev] BTree corrupted after conflict resolution

[Tim Peters]

[Christian Robottom Reis]
>>> Well, wouldn't that just be a roundabout way of making it explicit
>>> that nothing but simple types can be used as BTree keys (since no
>>> custom __cmp__ can be provided), or are you considering the
>>> zero-concurrency situation? :-)

[Dieter Maurer]
>> No. There is not problem to use class instances as keys,
>> provided they provide a persistent (complete) ordering
>> and they are not persistent.

[Christian]
> Okay. But then how do they provide complete ordering if not via
> __cmp__ or __rcmp__, given that they are instances?

By supplying __cmp__.  Your original "since no custom __cmp__ can be
provided" didn't make sense.  You can always slap a custom __cmp__ on a
class.  It apparently doesn't do any *good* to supply a custom __cmp__ on a
*persistent* class so far as conflict resolution is concerned, but Dieter is
explictly talking about non-persistent classes.  It's unclear whether your
original "simple types" did or didn't intend to include non-persistent
classes.  Dieter read your "simple types" as not including non-persistent
classes, in which case it's not true that nothing but simple types can be
used as BTree keys.  If you did intend to include non-persistent classes
with intelligently defined __cmp__ methods in "simple types", then it is
true that nothing but simple types can [etc].