为什么这个Haskell表达式编译？

Question

Here are some definitions I wrote, to avoid mixing currencies

{-# LANGUAGE GeneralizedNewtypeDeriving #-}

data EUR
data USD 

newtype Amount a = Amount Double deriving (Fractional, Num, Show)

eur :: Double -> Amount EUR
eur = Amount

usd :: Double -> Amount USD
usd = Amount

• usd 34 + usd 3 type checks as expected
• usd 33 + eur 33 is a compilation error as expected
• I'm surprised but usd 33 + 3 is OK according to the compiler. Something I wanted to avoid, and don't understand. I suspect it is because Num instance, but then what is the difference with the second case?

Can you explain why usd 33 + 3 compiles and if it is possible to make the type-checker reject this expression.

Answer 1

Numbers in Haskell have lots of implicitness. Mentally, you ought to replace every number literal like 3 with fromInteger 3 . Since Amount uses GeneralizedNewtypeDeriving to be part of the Num typeclass, it inherits a fromInteger instance. So the compiler is doing this

usd 33 + 3
===                                      [implicit fromInteger & expand usd]
(Amount 33 :: Amount USD) + 
  fromInteger 3
===                                      [fromInteger :: Num a => a -> Amount a]
(Amount 33 :: Amount USD) + 
  (Amount 3 :: Amount a)
===                                      [unify a]
(Amount 33 :: Amount USD) + 
  (Amount 3 :: Amount USD)

Answer 2

When GHC derives the Num class, it provides an implementation for the fromInteger function. Integer literals like 3 actually have the type Num a => a .

ghci> :t (34)
(34) :: Num a => a

When the type checker sees that you are trying to add a value of type Amount USD to 3 , it determines 3 :: Amount USD , which is valid since it is a member of the Num typeclass.