Haskell - 查找未声明的变量

时间:2017-10-05 15:09:43

标签: haskell functional-programming

AST:

data AST = Nr Int | Sum AST AST | Mul AST AST | Min AST | If AST AST AST | 
Let String AST AST | Var String deriving (Eq, Show)

嗨!我需要一些帮助在输入中找到未声明的变量。我的问题是,我不能像我这样在评估员中做到这一点:

eval :: Env -> AST -> Int
eval env (Nr nr) = nr
eval env (Sum xs xss) = eval env xs + eval env xss
eval env (Mul xs xss) = eval env xs * eval env xss
eval env (Min xs ) = - eval env xs
eval env (If x xs xss) = if (eval env x == 0)
                then eval env xs
                else eval env xss
eval env (Let s xs xss) = eval ((s, (eval env xs)) : env) xss 
eval env (Var x) = case lookup x env of
    Just n -> n
    Nothing -> error ("Variable " ++ x ++ " is undeclared!")

如果有任何未声明的变量,我需要在解析时给出包含所有未声明变量列表的适当错误,或者在评估之前对我的AST进行后处理。而且我不知道从哪里开始。这是解析表达式的一个示例:

parse "let X = + 1 2 in * X + 2 - X"
    Let "X" (Sum (Nr 1) (Nr 2)) (Mul (Var "X") (Sum (Nr 2) (Min (Var "X"))))

1 个答案:

答案 0 :(得分:6)

让我们从类型开始:

  

如果有任何未声明的变量,我需要在解析时给出包含所有未声明变量列表的相应错误。

函数eval如何为您提供未声明的变量列表,或Int(如果没有未声明的变量)。

type Identifier = String

eval :: Env -> AST -> Either [Identifier] Int

我们需要将原始数字包装在Right中:

eval env (Nr nr) = Right nr

Var案例中声明的变量相同, 而未声明的变量包含在列表和Left

eval env (Var x) = case lookup x env of
    Just n -> Right n
    Nothing -> Left [x]

对于Min案例,我们不能再拒绝递归调用了,因为 没有为Either [Identifier] Int定义否定。

我们可以模式匹配,看看我们得到了什么:

eval env (Min xs ) = case eval env xs of
  Left err -> Left err
  Right x  -> Right (-x)

但这非常罗嗦,与使用fmap仿函数实例中的Either e完全相同:

eval env (Min xs ) = fmap negate (eval env xs)

对于Sum,我们可以在两个参数上进行模式匹配:

eval env (Sum xs xss) = case (eval env xs, eval env xss) of
  (Left err, Left err') -> Left (err ++ err')
  (Left err, Right _)   -> Left err
  (Right _, Left err')  -> Left err'
  (Right a, Right b)    -> Right (a + b)

注意如果两个子系统都包含未声明的变量,我们将它们连接起来以获取Sum下的未声明变量列表。

这与我们其他构造函数所需的技巧相同。但是,我不想每次都要输入一个巨大的case语句。为了一点点补充,这需要做很多工作! IfLet将有8个案例!

所以让我们为我们做一个帮助函数:

apply :: Either [Identifier] (a -> b) -> Either [Identifier] a -> Either [Identifier] b
apply (Left err) (Left err') = Left (err ++ err')
apply (Left err) (Right _)   = Left err
apply (Right _)  (Left err') = Left err'
apply (Right f)  (Right a)   = Right (f a)

现在为SumMulIf定义案例要容易得多:

eval env (Sum xs xss) = fmap (+) (eval env xs) `apply` eval env xss
eval env (Mul xs xss) = fmap (*) (eval env xs) `apply` eval env xss
eval env (If x xs xss) = fmap jnz (eval env x) `apply` eval env xs `apply` eval env xss
  where jnz i a a' = if i == 0 then a else a'

Let略有不同:

eval env (Let s xs xss) = fmap second v `apply` eval env' xss
  where val = eval env xs
        env' = (s,val) : env
        getRight (Right a) = a
        getRight (Left _) = 0
        second _ a = a

请注意我们如何欺骗"当第一个术语包含未声明的变量时,通过为第二个术语的环境提供虚假值。由于我们不会使用任何Int值,所以在这种情况下第二项可能会产生,这是可以的。

一旦你在Haskell中进一步了解,你可能会注意到apply看起来很像<*>中的Applicative。我们之所以没有使用它,是Either e Applicative个实例无法按照我们想要的方式工作的原因。它不是聚合错误,而是在它碰到第一个错误时退出:

>>> Left ["foo"] `apply` Left ["bar", "baz"]
Left ["foo", "bar", "baz"]
>>> Left ["foo"] <*> Left ["bar", "baz"]
Left ["foo"]

但是,Validation type from the either package有一个适用于这种方式的应用实例,所以如果你愿意,你可以使用它:

>>> Failure ["foo"] <*> Failure ["bar", "baz"]
Failure ["foo", "bar", "baz"]

可能会使Let案例变得不那么严重的一种方法是将eval的返回类型从Either [Identifier] Int更改为([Identifier], [(Identifier, Int)] -> Int) - 让它返回所有列表表达式中的自由变量,以及在给定这些变量的绑定的情况下评估表达式的方法。

如果我们给该类型命名:

data Result a = Result { freeVariables :: [Identifier], eval :: [(Identifier,Int)] -> a }

我们可以为它定义FunctorApplicative个实例:

instance Functor Result where
  fmap f (Result is g) = Result is (f . g)
instance Applicative Result where
  pure a = Result [] (const a)
  Result is ff <*> js fa = Result (is ++ js) (ff <*> js)

使用它们来轻松定义一个函数来解析自由变量和一个eval表达式:

parse :: AST -> Result Int
parse (Nr nr) = pure nr
parse (Sum xs xss) = (+) <$> parse xs <*> parse xss
parse (Mul xs xss) = (*) <$> parse xs <*> parse xss
parse (Min xs ) = negate <$> parse xs
parse (If x xs xss) = jnz <$> parse x <*> parse xs <*> parse xss
  where jnz a b c = if a == 0 then b else c
parse (Let s xs xss) = Result ks h
  where Result is f = parse xs
        Result js g = parse xss
        ks = is ++ delete s js
        h env = g ((s,f env):env)
parse (Var x) = Result [x] $ \env -> case lookup x env of
  Just n -> n
  Nothing -> error ("Variable " ++ x ++ " is undeclared!")