homepage

The following commands show memory leaks introduced by bpo-30860:
$ ./python -m test -R 3:3 test_atexit -m test.test_atexit.SubinterpreterTest.test_callbacks_leak
$ ./python -m test -R 3:3 test_atexit -m test.test_atexit.SubinterpreterTest.test_callbacks_leak_refcycle
$ ./python -m test -R 3:3 test_threading -m test.test_threading.SubinterpThreadingTests.test_threads_join
$ ./python -m test -R 3:3 test_capi -m test.test_capi.SubinterpreterTest.test_subinterps

> The following commands show memory leaks introduced by bpo-30860:
Oh, I'm talking about the commit 2ebc5ce42a8a9e047e790aefbf9a94811569b2b6.

Could be the same as #31408.

The leak comes from the following lines of _PySys_EndInit():
---
 PyObject *warnoptions = get_warnoptions();
 if (warnoptions == NULL)
 return -1;
 SET_SYS_FROM_STRING_BORROW_INT_RESULT("warnoptions", warnoptions);
 PyObject *xoptions = get_xoptions();
 if (xoptions == NULL)
 return -1;
 SET_SYS_FROM_STRING_BORROW_INT_RESULT("_xoptions", xoptions);
---
It's not the first time that I have an issue with these attributes. The last reference weak caused by multiple interpreters was also related to this one if I recall correctly.
See bpo-30598 and my commit 865de27dd79571a4a5c7a7d22a07fb909c4a9f8e.

Extract of my msg295399: "The problem is that warnoptions is stored in a C global variable *and* in sys.warnoptions of each interpreter. The ownership of this variable is unclear."
Maybe we need a change similar to bpo-28411 (commit 86b7afdfeee77993fe896a2aa13b3f4f95973f16) which removed the "modules" field from Py_InterpreterState.

I'm looking into this.

Moving warnoptions (and xoptions) out of PyInterpreterState seems like a good idea to me for the same reasons as applied to sys.modules. The case for these two is even stronger since they are only used in sysmodule.c.

New changeset dae0276bb6bc7281d59fb0b8f1aab31634ee80dc by Eric Snow in branch 'master':
bpo-30860: Fix a refleak. (#3567)
https://github.com/python/cpython/commit/dae0276bb6bc7281d59fb0b8f1aab31634ee80dc