http://buildbot.python.org/all/#/builders/58/builds/122

======================================================================
ERROR: tearDownClass (idlelib.idle_test.test_configdialog.KeysPageTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "D:\cygwin\home\db3l\buildarea\3.x.bolen-windows7\build\lib\idlelib\idle_test\test_configdialog.py", line 701, in tearDownClass
    del page.set_keys_type, page.load_keys_list
AttributeError: set_keys_type
======================================================================
FAIL: test_set_keys_type (idlelib.idle_test.test_configdialog.KeysPageTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "D:\cygwin\home\db3l\buildarea\3.x.bolen-windows7\build\lib\idlelib\idle_test\test_configdialog.py", line 859, in test_set_keys_type
    eq(d.custom_keyset_on.state(), ('selected',))
AssertionError: Tuples differ: ('selected', 'hover') != ('selected',)
First tuple contains 1 additional elements.
First extra element 1:
'hover'
- ('selected', 'hover')
?             --------
+ ('selected',)

The failure in tearDownClass is a side-effect of the failure in test_set_keys causing 'p.set_keys_type = Func' being skipped.  That could be prevented with 'try:<anything that could fail>finally: p.set_keys_type = Func'.

I am completely puzzled at the sudden failure on one machine.  configdialog and test_configdialog were lasted changed 11 days ago (10/27) by Serhey's patch to make font samples editable.  test_idle passed consistently everywhere, including that machine, until build 121, 16 hours ago, and then suddenly started failing, consistently, on that one machine (as far as I know).  It continues to pass on my Win 10 machine, freshly updated with Python rebuilt.  I also checked the git log for tkinter.__init__ and tkinter.ttk and they have not been changed either.

The failure message is also a surprise.  'custom_keyset_on' is a ttk Radiobutton.
https://docs.python.org/3/library/tkinter.ttk.html#widget-states
does not list 'hover' as a state.  However
https://www.tcl.tk/man/tcl8.6/TkCmd/ttk_widget.htm#M-state
does (so the doc needs fixing): "The mouse cursor is within the widget."

Perhaps 'hover' is a result of previous mouse cursor positioning from event_generates in the font tab test.  (Why the sudden change on one system would still be a puzzle.)  If this is the reason, then I only need to worry about this one state test.

Otherwise, I would have to think about either parking the cursor where it cannot interfere, or about minimizing the dialog when it does not need to be de-iconified for event_generate to work.

Serhiy, any thoughts on this?

In build 129, which finished perhaps 3 hours ago, test_idle passed again.  I think we should merge the fix anyway in case 'hover' appears again on some machine.

A similar try:except is needed elsewhere.

> I am completely puzzled at the sudden failure on one machine.

The "x86 Windows7 3.x" buildbot worker is very slow. So this bug is probably a race condition which is hard to trigger on a fast machine.

Terry: you proposed a fix, do you still want to write it? The buildbot didn't fail again since I reported the failure.

I want to add the try-except here and in the few other places where a method is temporarily masked.  One failure causing another strikes me a  test code bug.  I would like to hold off on adding 'hover', as it now seems to have been an idiosyncratic and temporary glitch on one buildbot.  I cannot see anything different about this particular button state check.  If 'hover' were needed here, so it would seem equally needed in several other places.

I didn't see the failure recently, I close the issue.