--- Note that this has not been extensively tested for reasonability,
--- but Knuth argues that repeated multiplication by the golden ratio
--- will minimize gaps in the hash space.
+-- Where hashInt32 works just as hashInt shown above.
+--
+-- Knuth argues that repeated multiplication by the golden ratio
+-- will minimize gaps in the hash space, and thus it's a good choice
+-- for combining together multiple keys to form one.
+--
+-- Here we know that individual characters c are often small, and this
+-- produces frequent collisions if we use ord c alone. A
+-- particular problem are the shorter low ASCII and ISO-8859-1
+-- character strings. We pre-multiply by a magic twiddle factor to
+-- obtain a good distribution. In fact, given the following test:
+--
+-- > testp :: Int32 -> Int
+-- > testp k = (n - ) . length . group . sort . map hs . take n $ ls
+-- > where ls = [] : [c : l | l <- ls, c <- ['\0'..'\xff']]
+-- > hs = foldl' f golden
+-- > f m c = fromIntegral (ord c) * k + hashInt32 m
+-- > n = 100000
+--
+-- We discover that testp magic = 0.
+