Is it more consistent if `mp_for_each` invokes functor with `mp_identity<T>()`?

[oliora]

Currently, mp_for_each invokes f with T(). This adds a default-constructible requirement for type T which is not always feasible and also not very consistent with other parts of mp11 that never instantiate user-provided types. Should not mp_for_each invoke passed functor with mp_identity<T>() instead?

[pdimov]

mp_identity is often what you want, but not always; when the list contains integral constants, you don't want them wrapped. I went back and forth on adding the mp_identity a few times and at the end decided not to. We could add mp_for_each_id<L> that does mp_for_each<mp_transform<mp_identity, L>> even though I don't mind spelling that out.

[oliora]

mp_for_each_identity<L> or .._id sounds good as well.

[HDembinski]

Just passing by with a comment: I slightly lean against adding mp_for_each_id because it is trivial to write, but I am in favour of explicitly pointing out the construction mp_for_each<mp_transform<mp_identity, L>> in an example in the documentation of mp_for_each.

[oliora]

I think, extending the documentation but no code changes is a good trade-off.