I am highly shocked (after looking deeper) that the change I am pushing here works. It works in the test suites, which is all I can verify atm. Scott will run the TCK for me to see how it affects those tests.

The problem is a cheick/egg condition. The change was to pass the "class/hierarchy default AccessType" into the PropertyContainer so that it knows how to build its list of persistent attributes. The problem is that the caller of PropertyContainer uses this list of persistent attributes in order to find the Id/EmbeddedId so that it can determine the "class/hierarchy default AccessType".