gh-135552: Add tests that check if weakref for tp_subclasses cleared after finalization

[sergey-miryanov]

Tests for #136189

• Issue: Segmentation fault, possibly due to a GC issue (tp_subclasses) #135552

[sergey-miryanov]

I believe this PR should skip news

[nascheme]

This looks okay to me. Some suggested improvements. The new test might fit better into test_weakref.py but that's minor.

I don't think it's required to run these in a separate Python process. That makes the test run slower and it's not supposed to crash. Generally a separate process is used for the tests that exercise the interpreter shutdown logic and we are not testing that. So you could just do:

    def test_clearing_weakrefs_in_gc(self):
        # This test checks that:
        # 1. weakrefs for types with callbacks are cleared before the
        # finalizer is called
        # 2. weakrefs for types without callbacks are cleared after the
        # finalizer is called
        # 3. other weakrefs cleared before the finalizer is called
        errors = []
        class Class:
            def __init__(self):
                self._self = self
                self._1 = weakref.ref(Class, lambda x: None)
                self._2 = weakref.ref(Class)
                self._3 = weakref.ref(self)

            def __del__(self):
                if self._1() is not None:
                    errors.append("Type weakref with callback is not None as expected")
                if self._2() is not Class:
                    errors.append("Type weakref is not Class as expected")
                if self._3() is not None:
                    errors.append("Instance weakref is not None as expected")
        Class()
        gc.collect()
        self.assertEqual(errors, [])

I find the variable naming a bit esoteric. I would have just called them wr1, wr2 and wr3.

[sergey-miryanov]

@nascheme Thanks for review! Fixed.

[sergey-miryanov]

@kumaraditya303 Thanks for review!

[sergey-miryanov]

Thanks everyone!