JBossAS transaction test regression

jhalliday Dec 11, 2008 7:57 AM

In pre-release testing of the new JBossTS 4.5, I see a regression for org.jboss.test.tm.test.MTTransactionManagerUnitTestCase.testCommitSameTxInTwoThreads

The test assumes that TransactioManager.commit() will barf if the transaction associated to the thread has already been committed. This is no longer the case.

The new JBossTS assumes that if you ask to commit the transaction and it's committed by the time the method returns then all is well. You'll get an warn in the logs but no exception.

I believe this is consistent with the JTA API specification and more user friendly, so I'm challenging the test rather than rolling back our modification.

Please let me know if you have any reason to believe this behaviour will break things other than the test and if so, on what grounds they rely on getting an exception.

1. Re: JBossAS transaction test regression

adrian.brock Jan 26, 2009 7:24 AM (in response to jhalliday)

I'm not really sure what this test is testing?
It was originally written by Dimitris so you might want to ping him.

Two different threads doing a commit on the same transaction
is pretty unusual (it won't happen normally in JBoss)
and will only happen if somebody is playing with the thread association.
Actions
2. Re: JBossAS transaction test regression

jason.greene Mar 3, 2009 10:21 AM (in response to jhalliday)

I ran into the same thing:

https://jira.jboss.org/jira/browse/JBAS-6561

It seems the association semantics are indeed wrong, so I am going to just "correct" the test.
Actions

Go to original post