On Feb 14, 2008, at 9:43 AM, Jim Fulton wrote:

On the other hand, it seems
clear that cross-storage references make the system as a whole
(software + people) less reliable.

I don't agree in general.

Uh. I thought we were talking about reported POSKeyErrors that would not exist if there were no such thing as cross-storage references. I don't understand
what there is to disagree with here.  Are you saying that cross-storage
references reduce the risk of other application-breaking conditions?


Perhaps I misunderstand, but low-level cross-site references seem like "secret" weakrefs. Maybe we'd be better off if they were explicit, higher- level objects,
used and dereferenced intentionally where required.


I don't object to this approach. I'll note that we had an approach like this before for Zope (mounts) and it had plenty of disadvantages of it's own, including injecting Zope-specific code into ZODB.

If someone can make something like this work without modifying ZODB, I won't object.

Right.  There is no reason for this approach to involve ZODB at all.
If broken "references" must be tolerated, I'm suggesting that they should be dealt
with in application/Zope-level code, not in ZODB.


I think it would be easier to make the current multi-database approach a bit more explicit. I'll probably do that.

Explicit is good.



_______________________________________________
For more information about ZODB, see the ZODB Wiki:
http://www.zope.org/Wikis/ZODB/

ZODB-Dev mailing list  -  ZODB-Dev@zope.org
http://mail.zope.org/mailman/listinfo/zodb-dev

Reply via email to