Re: Bug#1117897: pygobject FTBFS with Python 3.14: TestReferences AssertionErrors

Version: 3.54.3-1

On Sun, 12 Oct 2025 at 18:03:19 +0100, Simon McVittie wrote:
>On Sun, 12 Oct 2025 at 11:24:48 +0200, Stefano Rivera wrote:
>>pygobject fails to build with Python 3.14 as an available version.
>...
>>I think it has been fixed in a new upstream version, but that needs to
>>be checked.

Confirmed fixed in experimental (but *not* in unstable).

>Please note that we are not yet ready to upgrade pygobject to its 
>latest upstream versions

Still true.

>if the python3.14 transition is to happen any time soon, it is 
>likely to be necessary to backport the fixes for test regressions

I tried backporting the obvious fix for expected refcounts in tests 
(0972c8b49040c08416648eca5ec9aac7a340666f), but this is not sufficient. 
Two tests still fail with 3.14.

One is in tests/test_generictreemodel.py, which is testing pygobject 2 
compatibility code that was deleted in the current upstream release (so 
there will be no upstream fix to backport for this):

______________________ TestReferences.test_iteration_refs ______________________

self = <tests.test_generictreemodel.TestReferences testMethod=test_iteration_refs>

     def test_iteration_refs(self):
...
         for node in nodes:
             if hasattr(sys, "getrefcount"):
>               self.assertEqual(sys.getrefcount(node), 5)
E               AssertionError: 4 != 5

tests/test_generictreemodel.py:258: AssertionError

The other is that in tests/test_properties.py, a different value 
is observed for the *GLib* refcount:

_____________ TestCPropsAccessor.test_held_object_ref_count_setter _____________

self = <tests.test_properties.TestCPropsAccessor testMethod=test_held_object_ref_count_setter>

     def test_held_object_ref_count_setter(self):
...
>       self.assertEqual(held.__grefcount__, 2)
E       AssertionError: 3 != 2

     smcv