OK, found it: the issue in EditorTest is that the SimpleBar that the setUserName is applied to and the one in the SimpleFoo's barField are not the same instances!
The instances are initialized in response to the finishTestAndReset call from RequestBatcherTest, which explains the differing behavior when this test is commented out (or moved to run later): in this case, everything is initialized from the call to findSimpleFooById at the beginning of EditorTest#test. The real issue is actually that finisTestAndReset resets SimpleFoo before resetting SimpleBar, so when SimpleFoo asks for SimpleBar.getSingleton() it gets the "previous instance", and then the SimpleBars are reinitialized. When everything is initialized in EditorTest#test() through findSimpleFooById, the SimpleFoo's barField and SimpleBar are the same instances. It might be that they differ in subsequent tests but that just does not seem to be an issue then. Inverting the order of the reset calls in finishAndResetTest fixes the issue. I also simplified the changes to AbstractRequestContext (adding a "diff'ing" state to the AbstractRequestContext/RequestState that's then used in the various AutoBean Categories to a) avoid auto-editing proxies (what was done by temporarily locking the context in patchsets 1 to 6) b) change the equals() behavior for EntityProxies (fixing issue 5952) http://gwt-code-reviews.appspot.com/1601806/ -- http://groups.google.com/group/Google-Web-Toolkit-Contributors
