Andriy,
I am continuing on ContextCommit... When I switch
DataContext.commitChanges() to ContextCommit locally, I am down to 2
failed unit tests, not bad at all ;). Most important - now BatchQueries
use the same logging mechanism as the rest of Cayenne, so they are much
easier to debug.
But suddenly I started having a bunch integrity problems with regression
tests that didn't happen before. From looking at my changes, I couldn't
figure out what may have caused this, there was no changes to the order
of the queries what so ever.
So I approached this from another end. I turned on SQL debugging of
regression test batches, and discovered that there is only one kind of
queries that fail: DELETE queries for the rows that reference themselves
via an FK (and I think also rows that reference rows that reference
themselves). From the RDBMS standpoint, having such row is totally
acceptable I guess, though I don't think I ever had them in real life.
Not sure if there are any other failures though...
Ok, so via a wayback machine of CVS, I switched to ContextCommit, r1.11
(the one that worked just yesterday) - the same problem.... So I wonder
if any changes done in other parts of Cayenne may have affected this?
Any insights?
Andrus
Andrus Adamchik wrote:
> Andriy,
>
> I started looking at ContextCommit integration with the main commit
> process. First thing I had to do is to refactor ContextCommit, creating
> a helper class DataNodeCommitHelper to store some of the intermediate
> data structures.
>
> After this change regression tests seem to pass just fine. Let me know
> if you find any problems though. If we have to revert the changes, just
> switch back to ContextCommit, CVS revision 1.11.
>
> Andrus
This archive was generated by hypermail 2.0.0 : Wed Mar 05 2003 - 02:06:09 EST