Trying to get rid of https://issues.jboss.org/browse/JBAS-8937 turned to be very "interesting".
Dunno how we avoided this bug for so long.
What happens in that case is the following.
(1) RARDeployer adds ServiceMetaData attachment to jboss-local-jdbc.rar deployment unit.
(2) Then AppNamingDeployer adds BeanMetaData attachment to same deployment unit.
(3) From (2) a new component deployment unit is created
(4) From (1) a new component deployment unit is created
(5) ServiceDeployer goes over all component deployment units
* for (4) it finds the right component ServiceMetaData attachment
* for (3) it finds the ServiceMetaData in its parent ( == (1)), which is of course the same as (4)
==> duplicate service registry
Atm I actually don't see how this ever worked,
must have been some really lucky deployers order.
I guess same could happen if the order of deployers that handle SMD and BMD would be turned.
e.g. instead of duplicate SMD, we would get duplicate BMD
Any idea how to properly fix this?