The problem is this:
instance Num a => Num [a] where ...
test = deep_fmap (+1) [[[ 1, 2, 3 :: Int ]]]
What (+1) should be used? (+1) :: Int -> Int (+1) :: [Int] -> [Int] (+1) :: [[Int]] -> [[Int]] (+1) :: [[[Int]]] -> [[[Int]]] They could all be type-correct, so the snippet is ambiguous. Monotypes are required to avoid the ambiguity. As to why you might want an instance of this form, they are very nice for representing (possibly infinite) power series[1] of the form c0 + c1 * x + c2 * x^2 + c3 * x^3 + ... As to your other question, in my experience it seems that yes, type equality constraints can replace the awkward "TypeCast" classes of oleg-ery. -- ryan [1] Mcilroy: Functional Pearl: Power Series, Power Serious. http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.38.9450 Code here: http://journals.cambridge.org/fulltext_content/supplementary/JFP/online/jfpv...