[Python] ConvertToSequenceAndInferSize leaks a reference to a list and the items when iterator raises an Exception

[raulcd]

Describe the bug, including details regarding any error messages, version, and platform.

While reviewing:

• GH-50829 [Python] Use more CPython Limited APIs in python/pyarrow #50409

I started investigating and I found that we could leak a reference and the items contained on the list when converting an iterator that raises an Exception at ConvertToSequenceAndInferSize.

I created a test that currently fails:

def test_failing_iterator_does_not_leak():
    # ConvertToSequenceAndInferSize builds an intermediate list.  If the
    # iterator raises, that list and everything already stored in it leaks.
    import gc

    # Create an arbitrary long int that is probably not cached by the interpreter
    value = 10**20

    def raising_iter():
        for _ in range(5):
            yield value
        raise ValueError("boom")

    gc.collect()
    original_refcount = sys.getrefcount(value)

    with pytest.raises(ValueError, match="boom"):
        pa.array(raising_iter(), size=100)

    gc.collect()
    assert sys.getrefcount(value) == original_refcount

The issue is in RETURN_IF_PYERROR(); error case where lst is not Py_DECREF:

arrow/python/pyarrow/src/arrow/python/python_to_arrow.cc

Lines 1260 to 1276 in 154962c

	PyObject* lst = PyList_New(n);
	RETURN_IF_PYERROR();
	for (i = 0; i < n; i++) {
	PyObject* item = PyIter_Next(iter);
	if (!item) {
	// either an error occurred or the iterator ended
	RETURN_IF_PYERROR();
	break;
	}
	PyList_SET_ITEM(lst, i, item);
	}
	// Shrink list if len(iterator) < size
	if (i < n && PyList_SetSlice(lst, i, n, NULL)) {
	Py_DECREF(lst);
	RETURN_IF_PYERROR();
	}
	*seq = lst;

The failing iterator case was detected here:

• ARROW-17253: [Python] Detect iterator exception instead of crashing #13764

Component(s)

Python

[raulcd]

Issue resolved by pull request 50594
#50594