Back when I was doing this C and/or C++ was required (in general and for the job).
Today, I don't know what I'd do (besides ask them to do it with pointers anyway, pseudo-code is fine for this problem). Someone from a Javaschool might work out well, but they'll be lost in some harder debugging situations, so I guess I'd try to make sure they'll ask for help when it gets that messy and make sure I had some uber-programmers to support them (such as myself :-).
I have a friend whose test is to sort a linked list using any sorting algorithm (it can be bubble sort) and any language (excluding trivial calls to library functions). It's fairly hard in C, trivial in ML or Haskell, and this is the point of the test: if you're able to use a high-level language, this is a benefit.
Errr, one of the points of my test is "do you grok pointers?" Which is an hard or impossible thing for many people and is needed in some subset of programming jobs.
But I'm sure many fewer today, note that I started out with punched card FORTRAN "IV" on an IBM 1130 and then did a lot of UNIX on PDP-11s. Fortunately I was able to play with LISP starting a couple of years after that first experience.
Today the world is almost entirely different, e.g. it's hard to buy a processor that had less L1 cache than the max address space of those machines' macroarchitecture.
ADDED: Or as I like to say, echoing someone I forget:
Today, I don't know what I'd do (besides ask them to do it with pointers anyway, pseudo-code is fine for this problem). Someone from a Javaschool might work out well, but they'll be lost in some harder debugging situations, so I guess I'd try to make sure they'll ask for help when it gets that messy and make sure I had some uber-programmers to support them (such as myself :-).