Sept. 1, 2020
9:17 a.m.
Having gotten to an implementation, you are correct, the dict iteration does not take the lion's share, or at least there are several other steps in my application which dwarf the dict traversal operations in any case. I don't think I in practice have a compelling case here, so all I'm left with is the vagueism that there must exist use cases where this would be the bottleneck, which is admittedly a cop-out. I mostly feel that the new dict internals are so convenient in enabling this that it would be a shame not to be able to have this performance improvement for the obscure case (which I can't think of) where creating a list copy would be undesirable for some reason other than the (as you point out) small memory increase.