I'll dig into this further tomorrow. But just by quickly looking at the test case you mention, and according to what I saw TreeCache doing today, that test may pass since it only seems to deal with non-versioning stuff within a transaction. But that might not be the case when running for real, since you'll have mixed non-versioning (query cache) calls and versioning (regular versioned data cache) calls within the *same transaction*. But as it seems, as pointed out before, TreeCache's transaction workspace is set to explicit user versioning as soon as you pass in a DataVersion via an Option (e.g. using write, writeLoad, ...). Hence subsequent non-versioning calls seem to be ought versioning calls, which, according to what I saw, fails. So your test may fail once you add a regular versioning call in between there. As for your reply to point 2, yes, then that is likely to be the reason, since once source==null, you effectively pass in an Option with dataversion=null (=non-versioning), which fails in the aforementioned situation, i.e. when mixed in the same transaction with a versioning call. View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=3957157#... Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&a...