[2026-03 LWG Motion 33] P3941R4 Scheduler Affinity

[jensmaurer]

Fixes NB US 232-366, US 233-365, US 234-364, US 235-363, US 236-362 (C++26 CD).

Fixes #8867

Also fixes cplusplus/papers#2594
Also fixes cplusplus/nbballot#937
Also fixes cplusplus/nbballot#938
Also fixes cplusplus/nbballot#939
Also fixes cplusplus/nbballot#940
Also fixes cplusplus/nbballot#941

[dietmarkuehl]

I believe one change was omitted: adding the get_start_scheduler object.

Another change was potentially garbled in the paper. The paper assumed P3826 (although revision 3) was applied (first editors note) and, thus, the reference to wasn't there. get-domain-early doesn't exist anymore based on P3826r5. If the corresponding change gets applied when P3826 get applied it doesn't have to done here.

While reviewing I noticed that I didn't mark at least one of the editor's note as such and, I think, some minor fixes were applied during editing: thank you very much!

[dietmarkuehl]

This is missing the definition of the inline object definition:

Suggested change

	inline constexpr get_start_scheduler_t @\libglobal{get_start_scheduler}@{};

[jensmaurer]

fixed

[dietmarkuehl]

The mark-up in the paper is certainly not right because it doesn't mention the transform_sender(get-domain-early(sender)): it should have had it and removed it! get-domain-early doesn't exist anymore. I believe the target expression is correct:

Suggested change

	transform_sender(@\exposid{get-domain-early}@(sndr), @\exposid{make-sender}@(affine_on, env<>(), sndr))
	@\exposid{make-sender}@(affine_on, env<>(), sndr)

[jensmaurer]

fixed

[jensmaurer]

@dietmarkuehl , there are still a number of mentions of "get-domain-early", so I don't know where we are with those.

[dietmarkuehl]

@dietmarkuehl , there are still a number of mentions of "get-domain-early", so I don't know where we are with those.

This paper isn't meant to fix these: I thought these should be fixed by P3826r5. Looking at the current PR for P3826r5 I see that currently only one place is addressed by the paper has a total of 12 places where "get-domain-early" gets removed. It seems the current [exec] wording has 14 places, though. Looking at the paper more closely: the two remaining places are removed via wording which isn't a diff: "Remove the prototypes of the get-domain-early and get-domain-late functions."

P3914R4 shouldn't reintroduce any uses of "get-domain-early" (and I don't think the wording does): there is no mention of "get-domain-early" in the changed text in the PR.