To properly test the unbind code (see comment), I think this should bind 3 scripts and unbind the middle one.

I would also like to test that the other two funcids still work in unbind.

There is a separate test, test_bind_events(), that tests that registered functions are called when the event occurs. Although it does not cover unbinding.

It is not so easy to make it working. For example, currently I cannot make it working in this test, an I do not know why. Maybe it does not work with keyboard events.

Well, I found how to do this. Will see whether it works on Windows.

The expanded test, not visible above, is what I proposed. When I just ran test_misc, whether by itself or with test_tkinter, whether directly or with regrtest, test_bind_events hangs with two blank tk windows on the screen. Hence the CI test timed out. When I close the first of the two windows, tests 'continue' after this error:

ERROR: test_bind_events (__main__.BindTest.test_bind_events)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "F:\dev\3x\Lib\test\test_tkinter\test_misc.py", line 709, in test_bind_events
    root.wait_visibility()  # needed on Windows
    ^^^^^^^^^^^^^^^^^^^^^^
  File "F:\dev\3x\Lib\tkinter\__init__.py", line 752, in wait_visibility
    self.tk.call('tkwait', 'visibility', window._w)
_tkinter.TclError: window "." was deleted before its visibility changed

This is due to the added "self.frame.wait_visibility()" in BindTest.setUp. With root deleted to let tests continue, the remaining BindTest functions fail with various messages.

DefaultRootTest.test_default_root also fails, with
"AttributeError: module 'tkinter' has no attribute '_default_root'.". A bad inter-testcase dependency? The following DefaultRootTest.test_getboolean and everything after passes.

With the the test_bind_events wait_visibility commented out, there is this error.

FAIL: test_bind_events (__main__.BindTest.test_bind_events)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "F:\dev\3x\Lib\test\test_tkinter\test_misc.py", line 729, in test_bind_events
    self.assertEqual(events, [
AssertionError: Lists differ: [] != [('frame', <tkinter.Frame object .!topleve[172 chars]me>)]

With the setUp wait_visibility instead commented out, there is this error:

>FAIL: test_unbind2 (__main__.BindTest.test_unbind2)
Traceback (most recent call last):
  File "F:\dev\3x\Lib\test\test_tkinter\test_misc.py", line 496, in test_unbind2
    self.assertEqual(events, ['a', 'b', 'c'])
AssertionError: Lists differ: [] != ['a', 'b', 'c']

With the setUp call instead moved into test_unbind2 either at the top or as f.wait_visibility() before the generate events call, there is no error.

Make clearer that this will not delete the Python function passed to the binding call, but only the binding-specific (tcl) wrapper (as the bind docstring promises).

What is "the binding-specific tcl wrapper function"? This term is not mentioned in the documentation and will confuse users.

It deletes the Tcl command and removes a reference to Python function. These details are not documented, but experienced users can discover this. I do not know what to write here.

Yes, it does not work with keyboard events.

I used different events for different tests to avoid possible order dependencies or chain reaction of failures after failing in one test. It was easy with keyboard events, it also minimizes risk of generating even by a user during running tests. But it may be not so easy to find a dozen different non-keyboard events (and we can add more tests in future). Is it necessary?

I withdraw the suggestion above. Either insert 'tcl' before 'command' or wait until someone reports being confused.

Done.

Without seeing what lines looks like for various scenarios, I cannot verify the code below. Hence would have to trust test. Hence would like testing of what seems like a harder scenario.

Usually it looks like:

if {"[140011831004144test1 %# %b %f %h %k %s %t %w %x %y %A %E %K %N %W %T %X %Y %D]" == "break"} break

if {"[140011830917168test2 %# %b %f %h %k %s %t %w %x %y %A %E %K %N %W %T %X %Y %D]" == "break"} break

if it only contains bindings added by bind(). It can also contain arbitrary Tcl commands, including multiline commands, if it is standard binding or bind() was called with a string instead of callable (AFAIK IDLE does such hacking). It does not matter, because it is unlikely that somebody add commands in the above form, using generated funcid.

I checked -- IDLE uses this in fix_x11_paste().

It's useful to manually register a string (to reuse functions multiple times or to directly call a Tk method), but that would entirely reasonably preclude using unbind() with that function. Should there be a test or something to verify it leaves other lines intact?

Right, you wrote that. This comment only serves to explain what I did not review in detail and why I requested the test change.

@TeamSpen210 If you manually register a string to directly call a Tk method, unbind() with funcid will not touch it, and unbind() without second argument will remove it. If you use funcid in other binding, it will stop working when unbind() deletes the command. If you change the line that calls funcid, unbind() may not find it or break the script after deleting the line. In any case, this feature is not documented, and doing so, you are at a risk. Of course, we will try to not break it, but our expectation of possible use cases can be wrong if you do something unusual.

I have no idea if it is legal to bind a multiple-script script, hence test should require that.

It is supported, but neither documented nor tested. I have not touched this, because it can mean doubling or tripling the number of tests.