Pessimistic locking and temporary nodes

[belaban]

FYI, I removed temporary nodes in TransactionEntry: they are not needed anymore.

To fix TransactionTest.testSimpleRollbackTransactions() (http://jira.jboss.com/jira/browse/JBCACHE-355), I modified DataNode.getOrCreateChildNode(): now when creating a new child node as a result of a put() on a non-existing child node, I simply create an undo operation (e.g. _remove()), which is added to the TransactionTable.
On rollback(), this will now remove temporary nodes

[bdueck]

Great change. A quick run on my benchmark seem to show a 15% performance improvement as a result of this for me... of course your mileage may vary.

Brian.

[belaban]

Good !

A further optimization could be to analyze the modifications and/or undo-ops *before* PREPARE/ROLLBACK, and to actually optimize them, e.g.
PUT(/a, null)
PUT(/a/b, null)
PUT(/a/b/c, null)

could be coerced into
PUT(/a/b/c)

Maybe we should create a JIRA issue for this...