In some methods of com.arjuna.ats.internal.jta.transaction.arjunacore.subordinate.TransactionImple, code of the form

int res = subAct.doFoo();
switch(res)
case ActionStatus.Bar:

is found. This is incorrect, as the methods on SubordinateAtomicAction return TwoPhaseOutome not ActionStatus. In the JTA the SubordinateAtomicAction returns ActionStatus which is probably where the confusion arises, so arguably it's actually the JTAX SubordinateAtomicTransaction which is at fault, not the TransactionImple.

Since the case statements will need reworking anyhow, it's probably worth looking that the way 'try to commit but actually have to rollback' is handled. In the JTA this throws RollbackException as the rollback is clean. In the JTAX it results in a HeuristicRollback due to the way the states and exceptions are mapped. This is perhaps less user friendly than it could be.