Different behaviour of Int and Integer?

Question

The followng snippet contains a solution for exercise 3 on page 69 (write a function mean to calculate the mean of a list).

While writing some QuickCheck tests to verify whether the its results are more or less sane, I found that on my system (ghc 6.12.3, Haskell Platform 2010.2.0.0 on 32-but Ubuntu 10.4) the tests work for Integer inputs, but not for Int ones. Any idea on why?

import Test.QuickCheck

-- From text and previous exercises
data List a = Cons a (List a)
            | Nil
              deriving (Show)

fromList        :: [a] -> List a
fromList []     = Nil
fromList (x:xs) = Cons x (fromList xs)

listLength             :: List a -> Int
listLength Nil         = 0
listLength (Cons x xs) = 1 + listLength xs

-- Function ``mean`` is the aim of this exercise
mean             :: (Integral a) => List a -> Double
mean Nil         = 0
mean (Cons x xs) = (fromIntegral x + n * mean xs) / (n + 1)
    where n = fromIntegral (listLength xs)

-- To overcome rounding issues
almostEqual     :: Double -> Double -> Bool
almostEqual x y = (abs (x - y)) < 0.000001

-- QuickCheck tests for ``mean``
prop_like_arith_mean :: (Integral a) => [a] -> Property
prop_like_arith_mean xs = not (null xs) ==>
                          almostEqual
                          (mean (fromList xs))
                          (fromIntegral (sum xs) / fromIntegral (length xs))

prop_sum :: (Integral a) => [a] -> Bool
prop_sum xs = almostEqual
              (fromIntegral (length xs) * mean (fromList xs))
              (fromIntegral (sum xs))

-- This passes:
check_mean_ok =
    quickCheck (prop_like_arith_mean :: [Integer] -> Property) >>
    quickCheck (prop_sum :: [Integer] -> Bool)

-- This fails:
check_mean_fail =
    quickCheck (prop_like_arith_mean :: [Int] -> Property) >>
    quickCheck (prop_sum :: [Int] -> Bool)

main = check_mean_ok >>
       check_mean_fail

Answer 1

Int is based on the underlying system's int implementation, and will probably the same lower and upper limits as the underlying system (but at least a range of [ -2^29, 2^29 - 1]. Integer has arbitrary precision. Therefore you might be seeing an overflow or underflow when you use Int.