3.2b2(~)

The following two test failures are seen with the newly released Cocoa-based ActiveState Tcl 8.5.9 on OS X 10.6.  No failures are reported when using the Apple Tcl/Tk 8.5 supplied with 10.6.

======================================================================
ERROR: test_tab_identifiers (tkinter.test.test_ttk.test_widgets.NotebookTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Library/Frameworks/Python.framework/Versions/3.2/lib/python3.2/tkinter/test/test_ttk/test_widgets.py", line 564, in test_tab_identifiers
    self.assertEqual(self.nb.tab('@5,5'), self.nb.tab('current'))
  File "/Library/Frameworks/Python.framework/Versions/3.2/lib/python3.2/tkinter/ttk.py", line 922, in tab
    return _val_or_dict(kw, self.tk.call, self._w, "tab", tab_id)
  File "/Library/Frameworks/Python.framework/Versions/3.2/lib/python3.2/tkinter/ttk.py", line 318, in _val_or_dict
    res = func(*(args + options))
_tkinter.TclError: tab '@5,5' not found

======================================================================
FAIL: test_traversal (tkinter.test.test_ttk.test_widgets.NotebookTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Library/Frameworks/Python.framework/Versions/3.2/lib/python3.2/tkinter/test/test_ttk/test_widgets.py", line 726, in test_traversal
    self.assertEqual(self.nb.select(), str(self.child1))
AssertionError: '.4654766160' != '.4654766352'
- .4654766160
?         ^^^
+ .4654766352
?         ^^^

Also multiple instances of the following apparently harmless message appear when using the ActiveState 8.5.9 Tcl:

setCanCycle: is deprecated.  Please use setCollectionBehavior instead

This is likely due to an upstream build problem from ActiveState.

http://permalink.gmane.org/gmane.comp.lang.tcl.mac/6524

Are you in a position to debug this a bit more?

I'll look into it.  No need to hold up beta2 for this.

After looking into these two test failures a bit more, it looks like both are due to regressions and/or bugs in the ActiveState Tk 8.5.9 Cocoa behavior versus the Apple Cocoa Tk 8.5.7, with which neither test fails.

The test_tab_identifiers failure may be a test case error where the test is making unwarranted assumptions about how the Notebook widget lays out its tabs.

The test_traversal failure may need to be modified to use a different key modifier for OS X (<alt> -> <option>) but there appears to be a deeper problem within Cocoa Tk on how the traversal is implemented (hint, option-char can run into the dead key problem documented in Issue10973).

http://docs.python.org/py3k/library/tkinter.ttk.html#tkinter.ttk.Notebook.enable_traversal

Neither of these failures appear to indicate any problems with tkinter itself.  At this stage in the 3.2 cycle, the failures can either be ignored or, if necessary, conditionally disabled for platform 'darwin'. This should be a release manager call.  Otherwise, the next step would be to isolate further and report as possible upstream bugs.

Also, the "setCanCycle: is deprecated" warnings have subsequently been shown to be avoidable by a change in the upstream Tcl/Tk build process and are expected to disappear in the next release of ActiveState Tcl/Tk 8.5.

I would prefer just letting the tests fail.  Disabled tests are usually forgotten, and never re-enabled.

The attached patch fixes the test failures with Tk 8.5.9, I haven't tested with other versions or other platforms.

The change to 'test_traversal' should be sane.

The change to 'test_tab_identifiers' is more phishy. I've done two changes there: (1) slightly different index to test for tab 0, as layout on OSX needs more space and (2) the code assumes that tab("idx") won't raise an exception, I've added a TclError catch to ensure that other locations get tested as well.

With these changes the tests pass, but I'd like a review from someone that knows about tkinter and ttk before committing.

New changeset de4d34a95603 by Ned Deily in branch 'default':
Issue #10736: Fix test_ttk test_widgets failures with Cocoa Tk 8.5.9
http://hg.python.org/cpython/rev/de4d34a95603

New changeset c2f61149764e by Ned Deily in branch 'default':
Issue #10736: Revert changeset de4d34a95603 as it causes failures with
http://hg.python.org/cpython/rev/c2f61149764e

The changes looked good to me and fixed the test failures with Cocoa Tk 8.5.9 and caused no regressions with the ActiveState Carbon Tk 8.4 so I committed them to get some exposure.  But then I recalled I had not tested with just the old Apple-supplied Tk 8.4 and, of course, the test fails there.  So reverting for now.

It appears the original (unmodified) test also fails with the old Apple-supplied Tk 8.4 (this is on OS X 10.5).  So I'll do some more testing and then probably re-apply this at some point after the current round of releases.

New changeset 4a0b929b5c3d by Ned Deily in branch '2.7':
Issue #10736: Fix test_ttk test_widgets failures with Cocoa Tk 8.5.9
http://hg.python.org/cpython/rev/4a0b929b5c3d

New changeset 570cdef34066 by Ned Deily in branch '3.2':
Issue #10736: Fix test_ttk test_widgets failures with Cocoa Tk 8.5.9
http://hg.python.org/cpython/rev/570cdef34066

New changeset b7c61000ceec by Ned Deily in branch 'default':
Issue #10736: Fix test_ttk test_widgets failures with Cocoa Tk 8.5.9
http://hg.python.org/cpython/rev/b7c61000ceec

No obvious buildbot problems so far so I'm going to close this as fixed.  Thanks for the patch, Ronald.