Great catch. I have set up my system so that I can bisect the problem. So far, I know that the current code in mainline fails, thus the problem has not been fixed in the 4.5-git merge on the way toward 4.6-rc1. It appears that commit 8c6d408 is good, but that one is 8796 commits later than v4.5, thus I must have screwed something up. I have just restarted the bisection between v4.5 and v4.4, but it is now late for me and I will need to finish later this morning.