Apparent bug in evict

Question

Hi,
Table 2 (the state transitions table) in the spec says that evict on a persistent-dirty instance
leaves its state unchanged. It does not say that an exception should be thrown as Kodo does. The
spec in section 12.5.1 implies that eviction may be delayed. In other words, the behavior that I
would expect, if I am correctly interpreting the murky depths of the spec here, is that evicting a
persistent-dirty instance has no immediate effect, but after the transaction completes, the instance
is then evicted. Evict marks it for eviction and this mark overrides the behavior indicated by the
transaction's RetainValues property.
If I were deciding implementation behavior, I would make all evictions inside a transaction put a
mark, even if they immediately evict the object, so that when the transaction completes, all evicted
objects are, in fact, evicted at transaction completion. So a PNT or PC object that is evicted and
reused within a transaction would get evicted again when the transaction completed. At transaction
completion, the eviction mark would be removed. As you can tell, I kind of like expected behavior
to be simply described and uniformly implemented. It makes an application programmer's life a lot
easier.
David Ezzio

Answer 1

David-
I believe you are correct. I have made a bug report about this:
http://bugzilla.solarmetric.com:8080/show_bug.cgi?id=124
David Ezzio <[email protected]> wrote:
Hi,
Table 2 (the state transitions table) in the spec says that evict on a persistent-dirty instance
leaves its state unchanged. It does not say that an exception should be thrown as Kodo does. The
spec in section 12.5.1 implies that eviction may be delayed. In other words, the behavior that I
would expect, if I am correctly interpreting the murky depths of the spec here, is that evicting a
persistent-dirty instance has no immediate effect, but after the transaction completes, the instance
is then evicted. Evict marks it for eviction and this mark overrides the behavior indicated by the
transaction's RetainValues property.
If I were deciding implementation behavior, I would make all evictions inside a transaction put a
mark, even if they immediately evict the object, so that when the transaction completes, all evicted
objects are, in fact, evicted at transaction completion. So a PNT or PC object that is evicted and
reused within a transaction would get evicted again when the transaction completed. At transaction
completion, the eviction mark would be removed. As you can tell, I kind of like expected behavior
to be simply described and uniformly implemented. It makes an application programmer's life a lot
easier.
David Ezzio--
Marc Prud'hommeaux [email protected]
SolarMetric Inc. http://www.solarmetric.com
Kodo Java Data Objects Full featured JDO: eliminate the SQL from your code