I reproduced the problem, and it seems to be related to less-than-complete test cases that weren't accessing the (lazy) user object before attempting to log out. I've trivially modified the patch to deal with that case (should never occur in real use), and feel that the modification makes the patch slightly safer overall.
I reproduced the problem, and it seems to be related to less-than-complete test cases that weren't accessing the (lazy) user object before attempting to log out. I've trivially modified the patch to deal with that case (should never occur in real use), and feel that the modification makes the patch slightly safer overall.
I'll be putting this in gerrit shortly.