On Feb 12, 2014, at 1:49 AM, Wolfgang Schnerring <w...@gocept.com> wrote:
> On 11 Feb 2014 06:54:22 Michael Bayer <mike...@zzzcomputing.com> wrote: >> On Feb 11, 2014, at 3:44 AM, Wolfgang Schnerring <w...@gocept.com> wrote: >>> parent = session.query(Parent).first() >>> self.assertEqual(1, len(parent.children)) >>> session.begin_nested() >>> session.delete(parent.children[0]) >>> self.assertEqual(0, len(parent.children)) >>> >>> >>> My point is, the last assertion fails, which I find both surprising and >>> inconvenient. ;) I'd be grateful for any insights you have about this. >> >> I’m pretty sure I mentioned this was what it seemed like you were >> describing. this is the “delete() on an object doesn’t remove it from all >> collections in which it is contained”. it’s not related to savepoints and >> you can read about how to work with this behavior here: >> http://docs.sqlalchemy.org/en/rel_0_9/orm/session.html#session-deleting-from-collections > > I know that flush does not trigger expiry. ;) I was wondering whether > savepoints > qualified as being a stronger boundary than flush and thus might be worthy of > triggering expiry. But I guess that answers my question then: the current > behaviour *is* intentional, and if I want expire_all then I'll just have to > call it myself (which is fine, I guess). well I’d look into using events if you’d like every begin_nested() to issue an expiry. The after_transaction_create should be a good event to use: http://docs.sqlalchemy.org/en/rel_0_9/orm/events.html#sqlalchemy.orm.events.SessionEvents.after_transaction_create - check if the given SessionTransaction is “nested” by seeing if it has a non-None ._parent attribute.
signature.asc
Description: Message signed with OpenPGP using GPGMail
