Since I've stalled on the AOP classloader testing until Kabir gets back from holiday:
I've added support for "All or Nothing" deployment heuristics.
This is something that has never been in the deployers before
(either old or new).
The basic idea is that if any subdeployment or component throws an
error during deployment, the whole deployment is undeployed with the
deployment in error marked as ERROR.
Actually, the way it is currently implemented, the top level deployment is also
marked as ERROR, but I don't think this should cause problems?
With these changes it should also be easy to add other heuristics
like "Russian Doll" and ordering of child subdeployments.
Unlike this JIRA task:
I think the ordering of children should be left to the structure deployers to decide
and the "Russian Doll" should be specified by each deployer,
i.e. some want to start from the top level (like the classloader)
and others like the mbean/pojo deployers can do Russion Doll,
i.e. deploy components from subdeployments first.
I've completed the other heuristics on that feature request
except for those at the front end which depend upon
what the Scanner/ProfileService does rather than what the deployers do.