list(set) should be atomic in the free-threaded build

[colesbury]

We have code that constructs a list from a set, while the set may be concurrently updated. For example:

cpython/Lib/multiprocessing/process.py

Lines 61 to 65 in 44f9a84

	def _cleanup():
	# check for processes which have finished
	for p in list(_children):
	if (child_popen := p._popen) and child_popen.poll() is not None:
	_children.discard(p)

This is a fairly common pattern currently, but in the free-threaded build, this may lead to RuntimeError: Set changed size during iteration. We should make it so that list(set) locks the set to avoid the error and so that the list contains a consistent copy of the set.

For additional context, see the following code from the nogil-3.12 fork:

• https://github.com/colesbury/nogil-3.12/blob/cedde4f5ec3759ad723c89d44738776f362df564/Objects/listobject.c#L1223-L1243
• https://github.com/colesbury/nogil-3.12/blob/cedde4f5ec3759ad723c89d44738776f362df564/Objects/listobject.c#L1372-L1381

cc @corona10

Linked PRs

• gh-116621: Set manual critical section for list.extend #116657
• gh-116621: Specialize list.extend for dict keys/values #116816
• gh-116621: Specialize list.extend for dict items #116888

[colesbury]

This is also important for test_multiprocessing_pool_circular_import in test_importlib.

[colesbury]

There are a few other cases in nogil-3.12 that we probably want to handle atomically as well, because I think we use the same pattern where the calling code assumes list(some_type) is atomic for certain built-in types:

• PyDictKeys_Check
• PyDictValues_Check
• PyDictItems_Check