[2026-03 LWG Motion 16] P3948R1 constant_wrapper is the only tool needed

[eisenwave]

Fixes NB FR-019-210 and FR-021-218 (C++26 CD).

Fixes #8850

Also fixes cplusplus/papers#2598
Also fixes https://github.com/cplusplus/nbballot/issues/779
Also fixes https://github.com/cplusplus/nbballot/issues/788

[eisenwave]

@jwakely @mattkretz we have goofed HARD. The paper deletes constant_arg_t but not the deduction guides that function_ref has for it.

https://eel.is/c++draft/func.wrap.ref.deduct

(will deal with overfull hboxes once home btw)

[jwakely]

@jwakely @mattkretz we have goofed HARD. The paper deletes constant_arg_t but not the deduction guides that function_ref has for it.

Can we just delete that (which is fine as an editorial change, because the draft is inconsistent if it refers to a type which no longer exists), or does it need to be changed to use constant_wrapper?

[eisenwave]

I believe the purpose of the deduction guides is to make things possible like

void f();
// CTAD
std::function_ref ref = std::nontype<f>;

Or now equivalently with std::cw<f>. If this no longer works as a result of deleting the deduction guides, that seems like an inadvertent design change.

I suppose the only editorial way to get the draft into a working state is to delete the deduction guides editorially and to reintroduce them by LWG issue. Any LWG resolutions won't be merged into the DIS anyway, right? We need to ship these motions.

[zhihaoy]

Herb reminded us that accepting motion can also introduce editorial change, so I suggest just finishing the renaming here.

[Mick235711]

Personally, I support treating this as an editorial wording defect in the proposal, as the intention is obviously to replace all usage of constant_arg mechanically. In other similar renaming proposals, if the author left over a few instances in the wording, they are also editorially applied without LWG issues as long as the intention is clear when polling the motions.

[Bronek]

@Mick235711 @zhihaoy I think in this case the intention was clear, so I would agree that this change should just replace constant_arg with constant_wrapper, ensuring that this CTAD works std::function_ref ref = std::cw<foo>; as long as that used to work std::function_ref ref = std::constant_arg<foo>; , for some nullary foo; and treat this as a necessary editorial.

EDIT: The intention was unfortunately not captured as clearly as I would wish, but my memory is rather clear that we wanted to "replace all the uses of constant_arg with constant_wrapper and remove constant_arg" and this is also what the paper being accepted says in the subject ("constant_wrapper is the only tool needed", emphasis mine)

[mattkretz]

1. Sorry!
2. I did change the deduction guide in my libstdc++ patch, so I really should have noticed by looking through the patch. 🤦
3. I'll provide wording that matches a tested implementation and then we can decide whether we call that editorial or not. I agree that the intent of the change is very clear (replace constant_arg_t by constant_wrapper) and that the removal of constant_arg_t forces a (editorial) change to the deduction guide.

[mattkretz]

This would be the necessary change for CTAD:
mattkretz@92a7ba1
I implemented it and it passes all (CTAD) tests in libstdc++.

[Bronek]

This would be the necessary change for CTAD: mattkretz@92a7ba1 I implemented it and it passes all (CTAD) tests in libstdc++.

This looks like what we voted for in LEWG, i.e. design intent.

[Bronek]

@jwakely @mattkretz we have goofed HARD. The paper deletes constant_arg_t but not the deduction guides that function_ref has for it.

Can we just delete that (which is fine as an editorial change, because the draft is inconsistent if it refers to a type which no longer exists), or does it need to be changed to use constant_wrapper?

I would argue that this needs to be changed to constant_wrapper like @mattkretz has demonstrated above, but it is your decision whether do it immediately or use a DR instead. Just a note that change to constant_wrapper reflects the design intent as approved by LEWG, which is why I personally think DR would be suboptimal.

[eisenwave]

I've incorporated Matthias' commit into the PR.

[jwakely]

I agree with the comments here that what LEWG approved did not include "and disable CTAD" so preserving support for CTAD is part of the intended change.

Thanks for the fixes to make the wording match that approved design.

[tkoeppe]

Thanks, everyone!