Free threading locking issues in listobject.c

[colesbury]

Bug report

In #127524, @mpage suggested adding an assert to check that the list object is locked when calling ensure_shared_on_resize. This uncovers a few more locking issues.

Additionally, we may return out of the critical section without unlocking here:

cpython/Objects/listobject.c

Lines 956 to 964 in c7dec02

	Py_BEGIN_CRITICAL_SECTION(a);
	Py_ssize_t n = PyList_GET_SIZE(a);
	PyObject *copy = list_slice_lock_held(a, 0, n);
	if (copy == NULL) {
	return -1;
	}
	ret = list_ass_slice_lock_held(a, ilow, ihigh, copy);
	Py_DECREF(copy);
	Py_END_CRITICAL_SECTION();

Linked PRs

• gh-127536: Add missing locks in listobject.c #127580
• [3.13] gh-127536: Add missing locks in listobject.c (GH-127580) #127613

[corona10]

Ah.. it's my fault, I will take a look at it

[colesbury]

Thanks @corona10. I started some work on this (but didn't finish) yesterday. Here's what I had in case it's helpful:

colesbury@49268ba

[corona10]

Ah in that case please go ahead

[picnixz]

@colesbury Are the build bot failures possibly related to that one or not? if not, I think we can close this one.

[colesbury]

I think the buildbot failures are unrelated. The other s390x 3.13 buildbots are passing. https://buildbot.python.org/#/builders/1469 is passing, but the runs slowed down (including on test_site) around the same time, but on the following commit, which makes me think some sort of network issue or problem with the machine.