Hacker News Reader

Comment by silentbicycle | Hacker News Reader

Because it has typeclasses, which solve the same problems without creating the new problems that conventional objects do.

Why doesn't Java support pattern matching?

View on HN

Typeclasses don't solve the same problems as objects do. Typeclasses are a way of getting access to more operations in polymorphic code, which in functional languages is a compile time feature; they're also a way to implement overloading.

Objects are a way of implementing protocols without needing to know the details of the object that implements the protocol, even at runtime. Polymorphic code in OO systems is a runtime feature, not a compile time one; objects may be loaded dynamically and almost always are at some level in any large OO system, for configurability and testing if nothing else.

Since Haskell is statically typed, I think it's a feature, not a bug, that type parametricity is implemented at compile time.

Also, I think what you're describing is the difference between parametric polymorphism and ad-hoc polymorphism.

Because two features are different doesn't mean they don't solve the same problem. Here, most of the time, class polymorphism is used to solve problem best solved with parametric polymorphism (sometimes called genericness).

    class Base
      Base(int i_init)         -- constructor
      virtual int f(int param) --  a method
    
    class Inherit1 (extends Base)
      int f(int param) -- reimplementation 1
    
    class Inherit2 (extends Base)
      int f(int param)  -- reimplementation 2
    
    class Inherit3 … (ad nauseam)

We don't need class polymorphism with inheritance, here. We can do simpler:

    class Base
       Base(int i_init, int f_init(int)) { f =: f_init; … }
       int f(int)

Or even simpler:

    int f(int i_init, int f_init(int)) // C-like syntax
    f: int -> (int -> int) -> int      -- Haskell syntax

(And don't tell me that passing function as parameters is weird, or complicated. Functions are typically way simpler than "Objects".)

Typeclasses are a fine substitute for polymorphism in behavior (methods). But does Haskell throw out the baby with the bath water? Haskell seems to lack an ability to express a taxonomy of data models. Records are not even close to a substitute.

Why can't Haskell make it easier to express data structures? Something "turtles all the way down" introspective, like a MOF (meta-object facility, http://en.wikipedia.org/wiki/Meta-Object_Facility) would be nice.

Or have closures yet.

Reply on news.ycombinator.com