Isolated/worked around the common part of the log (at top) and the log is now pointing to another place in the code. So, this stack track doesn't help much to find what is firing the process which end up for kill the geometry window focus. 
The problem is really in the new code used to fix the old focus issue.
Some unregistered window is taking the focus. With the autosave enabled (which is the easy way to reproduce the issue) that window has the top_frame as parent, but it's not a registered window (not in wm_windows).

The problem is really in the new code used to fix the old focus issue.
Some unregistered window is taking the focus. With the autosave enabled (which is the easy way to reproduce the issue) that window has the top_frame as parent, but it's not a registered window (not in wm_windows).