I'm seeing a case where a PySide object is incorrectly invalidated.  I don't know if you care about PySide and Qt4 any more, but it happens both there and PySide2 and Qt5.  I believe that what is happening is that if a PySide parent object goes away it invalidates its children, but it does this even if the PySide object does not own (did not create) the Qt object and will not destroy it.  So I think the fix might be that the invalidation only happens if the PySide object owns (created) the wrapped Qt object.

I've attached a test program that I wrote that demonstrates the bug.  The first assertion passes, but the second assertion fails.