Thanks Hugo. I remember the discussion from when it came up originally. I even came up with a suggestion for fixing it that wasn't a million miles away from what got implemented, though I doubt my half-cocked idea was any influence :)

Given the OP was picking 4 at a time from a "very large hash", the inherent randomness (assuming he has it enabled (or can work out how to enable it!! Argggh!)), might be sufficiently random for his purpose.

It would have an advantage over the other solutions presented, in that it would avoid duplicates whilst ensuring full coverage. If that is a requirement.