Reference leak for object with non-trivial finalizer

[efimov-mikhail]

Bug report

Bug description:

If we put this code snippet to Lib/test/test_dict_user_leak.py

import unittest

class DictUser:
    def __init__(self, d: dict):
        self.d = d

    def __del__(self):
        self.d['a'] = DictUser(self.d)

class TestFinalizerLeak(unittest.TestCase):
    def test_dict_user(self):
        DictUser(dict())

if __name__ == '__main__':
    unittest.main()

then we can see leaks with the following command:
./python -m test -R 3:3 -v test_dict_user_leak

CPython versions tested on:

CPython main branch, 3.14, 3.13

Operating systems tested on:

Linux

[jamalhill1441-coder]

Lib/test/test_dict_user_leak.py

[colesbury]

I don't think this is a bug in Python. You are creating trash in in a finalizer that will only be collected on the next GC cycle.

[vstinner]

I suggest closing this issue as WONT FIX.

[efimov-mikhail]

Okay.

[efimov-mikhail]

Maybe we should document this little more?

[colesbury]

What would you say? I don't think this is the sort of thing people are likely to encounter in practice.

[efimov-mikhail]

I also don't think this is likely to encounter.
We can say something like this:

Some finalizers may produce memory leaks.
For example, if __del__ finalizer is creating new objects of the same class then it's impossible to destroy all objects of this class.

[vstinner]

I don't think that we should document such weird case.

[efimov-mikhail]

As a final suggestion, we can write this as a code comment or in the InternalDocs.
I agree that it's essential for CPython developers, not for regular Python users.