I am guessing that test_idle and hence this test have been run over 500 days X 20 runs/day (10000 times) since merged.  This is the first report I have seen, though failures that get retested would be masked.  My guess is a sporadic timing delay such that the assertion is tested before the widget is updated.  Serhiy, does this make sense?

I have reproduced the failure 8 times by running the attached file, derived from test_configdialog, in IDLE.  Reps required ranged from about 1000 to 30000, plus I stopped once at nearly 100000 with no failure.  About half appeared to be a direct result.  All appeared to be an up/down font test, the first of the pair.

Tomorrow, I intend to run this more, including from command line, and then insert update() or update_idletasks() to see if either dependably suppress failure.  I am curious what happens on Linux.