(Sorry for the force push, but this one was a very tricky and subtle git mistake of mine and I prefer not confusing y'all)

A couple notes on the solution:

I added a private parameter _classcell to collections.namedtuple() and its wrappers in typing in order to deliver it to the targeted type.__new__.
It's the easiest approach here.

I also pop __classcell__ from the original class namespace, as it would otherwise leak to the end class (metaclasses should clean up the __classcell__ attribute).

I expect it is not added to typeshed, as it is only intended for internal use.

I expect _repr added to typeshed though and offer a PR for it upon acceptance of this solution.

If you disagree with the addition of _repr to named tuples, we can split this PR. Just let me know.

If you disagree with the addition of _repr to named tuples, we can split this PR. Just let me know.

It would make sense to me to split this PR into two. I'm not necessarily saying I disagree with the _repr addition, but it seems less clearly a bugfix than the __classcell__ thing. Small, incremental PRs are also easier for us to review :-)

@AlexWaygood Thanks, you're right. I should have split that in the first place. I'll work through all the possible CI problems and then proceed to split it.

This looks super-bad!

Closing this one and opening two new PRs, as concluded in #129344 (comment).