Swapping nibbles range style

[bearophile]

Marco Leise:

>   return s.chunks(2).map!(
>     c => [cast(char)c.dropOne.front, cast(char)c.front]
>   ).join;
>
> So I would use the second version offered by bearophile:
>
>   auto result = new char[s.length];
>   foreach (immutable i, ref c; result)
>     c = s[i + (i & 1 ? -1 : 1)];
>   return result;
>
> I can easily reason about it from looking at it. It makes one
> allocation "new char[s.length]" and does no UTF-8 decoding.

But the first version with dropOne and two fronts is for me 
simper to see as correct, while in the second code I need a bit 
of thinking to prove there are no out of bounds accesses.

Perhaps for/foreach are becoming a code smell, they are becoming 
an optimization over the more common and safer functional code 
(or an alternative solution where the functional code is a 
problem).

To avoid the conditional (I have not benchmarked them):

c = s[i + (2 * !(i & 1) - 1)];

Bye,
bearophile