I agree completely. I seriously doubt the there is any advantage in terms of (real-time) performance of using either the scalar-as-packed-integer-array, or the uninitialised buffer as packed array over using a hash.

I've always had problems with big-O notation. In part this comes down to the fact that O is measured in some theoretical unit that is completely divorced from any real-time unit of measurement.

It's perfectly possible to have two algorithms to do the same task. One which is defined as being O(1) and the other as O(N) where the former is much slower in realtime than the latter. I think this is why I tend to confuse myself when accesing algorithms. My natural instinct is to conclude that the fastest realtime implementation should have a 'lower' bigO rating than a slower implementation, but that simple isn't the case. For example, a brute force combinatorial solution to a problem written in C or assembler can outstrip an 'intelligent' algorithm written in perl, even though the former rates as O(n*m) and the latter O(n or even O(log n).

Regardless, if my reinterpretation of Abigail's challenge--a pure perl way to allocate an uninitialised buffer--was correct, and he didn't correct me, then I think that perl 5.8.x's 'memory files' meet the challenge.