Should I use typeclasses or not?

I have some difficulties to understand when use and when not use typeclass in my code. I mean create my own, and not use already defined typeclasses, of course. By example (very stupid example), should I do:

data Cars = Brakes | Wheels | Engine
data Computers = Processor | RAM | HardDrive  

class Repairable a where
    is_reparaible :: a -> Bool

instance Repairable Cars where
    is_repairable (Brakes) = True
    is_repairable (Wheels) = False
    is_repairable (Engine) = False

instance Repairable Computers where
    is_repairable (Processor) = False
    is_repairable (RAM)       = False
    is_repairable (HardDrive) = True

checkState :: (Reparaible a) => a -> ... 
checkState a = ...

(Obviously, this is an stupid, incomplete example).

But this is a lot for a little use, no? Why I shouldn't do something simple and only defining functions without defining new data types and typeclasses (with their instances).

This example is too simple, but in facts I often see somethings like that (new data types+typeclasses+instances) when I browse Haskell code on github instead of only defining functions.

So, when I should create new data types, typeclasses etc and when should I use functions?

Thanks.


Why I shouldn't do something simple and only defining functions without defining new data types and typeclasses (with their instances).

Why indeed? You could just define:

checkState :: (a -> Bool) -> (a -> b) -> (a -> b) -> a -> b
checkState is_repairable repairs destroy a
    = if (is_repairable a) then repairs a else destroy a

People misuse type classes all the time. It doesn't mean that it's idiomatic.

To answer your more general question, here are some rules of thumb for when to use type classes and when not to use them:

Use type classes if:

  • There is only one correct behavior per given type

  • The type class has associated equations (ie "laws") that all instances must satisfy

  • Don't use type classes if:

  • You are trying to just namespace things. That's what modules and namespaces are for.

  • A person using your type class cannot reason about how it will behave without looking at the source code of the instances

  • You find that the extensions you have to turn on are getting out of control


  • You can often use a data type instead of a type-class, eg

    data Repairable a = Repairable 
       { getRepairable :: a
       , isRepairable :: Bool
       , canBeRepairedWith :: [Tool] -> Bool  -- just to give an example of a function
       } 
    

    Of course you need to pass this value explicitly, but this can be a good thing if you have multiple choices (eg think of Sum and Product as possible Monoid s for numbers). Except that you have more or less the same expressiveness as for a type-class.

    链接地址: http://www.djcxy.com/p/43466.html

    上一篇: Haskell中的Comonad类型类是什么?

    下一篇: 我应该使用typeclasses或不?