gh-118965: Share NotImplemented in subinterpreters

[sobolevn]

I still use Py_NewRef despite the fact that NotImplemented is immortal:

static void
notimplemented_dealloc(PyObject *notimplemented)
{
    /* This should never get called, but we also don't want to SEGV if
     * we accidentally decref NotImplemented out of existence. Instead,
     * since Notimplemented is an immortal object, re-set the reference count.
     */
    _Py_SetImmortal(notimplemented);
}

• Issue: Support sharing NotImplemented singletons in subinterpreters #118965

[nineteendo]

cpython/Lib/test/test_interpreters/test_api.py

Lines 1128 to 1130 in 176b42a

	not_shareables = [
	# singletons
	NotImplemented,

Should be moved to

cpython/Lib/test/test_interpreters/test_api.py

Lines 1099 to 1101 in 176b42a

	shareables = [
	# singletons
	None,

By the way, is there a reason ... is not shareable?

[sobolevn]

@nineteendo thanks.

By the way, is there a reason ... is not shareable?

I don't think so. I plan to address it in the next PR.

[ericsnowcurrently]

FWIW, there's no urgency here. The other singletons have a clear value in being shareable. However, NotImplemented and Ellipsis do not. There's nothing wrong with making them shareable. However, there isn't much clear advantage either, AFAIK.

[erlend-aasland]

Should we generalize this with a macro? (Future refactoring)

[ZeroIntensity]

I don't think so. Last I heard, Eric's plan was to eventually get a dedicated type slot for crossinterpreter registry things, so this will (hopefully) be obsolete sooner than later.

[ZeroIntensity]

Trying to work through the subinterpreter backlog a bit :)

[ZeroIntensity]

I don't think so. Last I heard, Eric's plan was to eventually get a dedicated type slot for crossinterpreter registry things, so this will (hopefully) be obsolete sooner than later.

[ZeroIntensity]

Merging from main seemed to blow up CI. I think something changed with XI registries after this was created.