On Sep 30, 2009, at 9:05 AM, Luis Fernando Planella Gonzalez wrote:
Hi.
I've reported http://issues.apache.org/jira/browse/OPENEJB-1027 in
may, which
is modifying a bit the heuristics to resolve a data source by
matching the app
name (web app context in our case) if the data source has not been
explicitly
specified. This was previously discussed with David Blevings, and he
suggested
that way.
Just to explain the use case: We're migrating cyclos (http://www.cyclos.org
),
an open source struts / hibernate application to gwt / ejb3. Many
people host
several instances of the very same application in the same tomcat,
and to keep
supporting that it would be necessary to unpack, change every
persistence.xml
from each and repack each instance again. With this feature
implemented, we
can just assume there will be a data source with the same name of the
application context and we're done, and our users will be happy :)
We can try patching ourselves, but we need some guidance on both
fundamental
questions:
* Where is this heuristic implemented?
* How to get the current application name?
The logic is in:
http://svn.apache.org/repos/asf/openejb/trunk/openejb3/container/openejb-core/src/main/java/org/apache/openejb/config/AutoConfig.java
There's a pretty good test case for the persistence unit side of that
logic here:
http://svn.apache.org/repos/asf/openejb/trunk/openejb3/container/openejb-core/src/test/java/org/apache/openejb/config/AutoConfigPersistenceUnitsTest.java
The javadoc was incorrect in some places, so I did a sweep through it
and fixed it all for you.
I recommend starting with a copy of the AutoConfigPersistenceUnitsTest
and expand it to always have two apps and spend a good amount of time
covering all the scenarios (what happens when there are more apps
than data sources, what if there are no data sources that match,
etc.) The logic in AutoConfig is a bit of a beast.
Even if you aren't able to get the logic in, having a really good test
case that covers more than just the "happy path" would be extremely
valuable. For functionality like this I spend at least 60% of the
time on the test case, so even having that would be a major help
towards getting this feature in. The test case also kind of like a
requirements doc, which is a cool thing when collaborating.
-David
