“Pattern matching” of algebraic type data constructors

后端 未结 5 1781
天涯浪人
天涯浪人 2021-02-07 15:07

Let\'s consider a data type with many constructors:

data T = Alpha Int | Beta Int | Gamma Int Int | Delta Int

I want to write a function to che

相关标签:
5条回答
  • 2021-02-07 15:41

    Look at the Data.Data module, the toConstr function in particular. Along with {-# LANGUAGE DeriveDataTypeable #-} that will get you a 1-line solution which works for any type which is an instance of Data.Data. You don't need to figure out all of SYB!

    If, for some reason (stuck with Hugs?), that is not an option, then here is a very ugly and very slow hack. It works only if your datatype is Showable (e.g. by using deriving (Show) - which means no function types inside, for example).

    constrT :: T -> String
    constrT = head . words . show
    sameK x y = constrT x == constrT y
    

    constrT gets the string representation of the outermost constructor of a T value by showing it, chopping it up into words and then getting the first. I give an explicit type signature so you're not tempted to use it on other types (and to evade the monomorphism restriction).

    Some notable disadvantages:

    • This breaks horribly when your type has infix constructors (such as data T2 = Eta Int | T2 :^: T2)
    • If some of your constructors have a shared prefix, this is going to get slower, as a larger part of the strings has to be compared.
    • Doesn't work on types with a custom show, such as many library types.

    That said, it is Haskell 98... but that's about the only nice thing I can say about it!

    0 讨论(0)
  • 2021-02-07 15:49

    Another possible way:

    sameK x y = f x == f y
      where f (Alpha _)   = 0
            f (Beta _)    = 1
            f (Gamma _ _) = 2
            -- runtime error when Delta value encountered
    

    A runtime error is not ideal, but better than silently giving the wrong answer.

    0 讨论(0)
  • 2021-02-07 15:51

    You can definitely use the generics to eliminate the boilerplate. Your code is a textbook example why I (and many others never use the _ wildcard at top level). While it is tedious to write out all the cases, it is less tedious than dealing with the bugs.

    In this happy example I would not only use Dave Hinton's solution but would slap an INLINE pragma on the auxiliary function f.

    0 讨论(0)
  • 2021-02-07 15:54

    You'll need to use a generics library like Scrap Your Boilerplate or uniplate to do this in general.

    If you don't want to be so heavy-handed, you can use Dave Hinton's solution, together with the empty record shortcut:

    ...
    where f (Alpha {}) = 0
          f (Beta {}) = 1
          f (Gamma {}) = 2
    

    So you don't have to know how many args each constructor has. But it obviously still leaves something to be desired.

    0 讨论(0)
  • 2021-02-07 15:56

    In some cases, "Scrap Your Boilerplate" library will help.

    http://www.haskell.org/haskellwiki/Scrap_your_boilerplate

    0 讨论(0)
提交回复
热议问题