Tk Source Code

One example of this, pointed out by Nicolas Bats is that if one selects the
"About" menu item, causing a standard About Panel to open, then when the About
Panel is closed no window receives focus.  The correct behavior is that the
focus should move to the window still on the screen which is in the stacking
order.

I have fixed this in the aqua_key_windows branch by adding code in the NSApp
method which handles notifications that a window has obtained or resigned
key status.

Now [0602f5b81637cd74|backported] to 9.0 as well.

Thanks!

OK, it works again and tests still pass.

This is really a classic case.  Now that the tests pass it is no longer the
case that closing the About window transfers focus to the window which
had focus before it was opened.  So it looks like another round will be
needed.

You're welcome!

Many thanks for taking this!

I have now dealt with the issues in the focus tests.  The assignment
of the key window now takes into account which interpreter is associated
with the window, and only transfers focus to a window with the same
interpreter (i.e. with the same "Tk application").  On Aqua there is only
one NSApplication, but that one NSApplication can host multiple
interpreters, each of which is considered a separate Tk application.

The focus command takes into account which Tk application a window belongs
to. This now works the way that the tests expect.

The hang was caused by the focusClear proc.  There is no way to arrange
that a Tk application has no focused window other than to create another
Tk application and give it focus.  So that is what focusClear does.  To
avoid test failures on Aqua because focusClear did not work reliably to
clear the focus (i.e. it would fail sporadically on the CI runner) I had
added a while loop that runs until the focus command returns an empty
string.  But with the first round of aqua_key_events that command was
hanging because the focus would get reassigned to the highest window
in the same NSApplication, which was in the original Tk application,
as soon as the first Tk application had no focused window.  This would
prevent the focusClear loop from terminating.

Now, with the interpreter checks, it is allowed for the main Tk
application to have no focused windows, and it is no longer necessary
to have a loop in the focusClear proc.  Moreover all focus tests pass.

In fact I have checked that merging main with aqua_key_windows produces
a Tk for which all tests pass on the CI runner.  I did that by cloning
the merge into a github repository in my own account for which I can
trigger a CI run when I want.

I have scheduled a CI run for aqua_key_windows in the tcltk/tk github
repository which will run tonight. When that succeeds I will push the
merged main to fossil and we can wait another day to check that all
tests still pass.

Yes, I know that.  I did not add that to fix the focus tests.  I am
working on fixing the focus tests now.

I removed the changes which broke the focus test from main.  They are
all back in the aqua_key_windows branch.

Checking the TK_ALREADY_DEAD flag doesn't help :-(

[https://github.com/tcltk/tk/actions/runs/13851459698/job/38759337333]

Thanks Jan.  Sorry about that.  I reopened this ticket and renamed the last
commit back into the bugfix branch.

focus.test is hanging now.

See: https://github.com/tcltk/tk/actions/runs/13828797808/job/38702606780

That's why I didn't cherry-pick it to the core-9-0-branch yet.