为什么 `peek` 与多态 Ptr return GHC.Prim.Any 一起使用时与绑定一起使用?

Why does `peek` with a polymorphic Ptr return GHC.Prim.Any when used with a bind?

使用 low-level GNU Science Library bindings Bindings.Gsl.RandomNumberGeneration,我 运行 进入 GHCi 中的这种奇怪类型行为,其中绑定将 return 类型从 peek 更改为 GHC.Prim.Any .我试图理解为什么我不能使用 c'rng_alloc,除非我保留指向 rng 的指针类型。例如:

λ> :t c'gsl_rng_alloc
c'gsl_rng_alloc :: Ptr C'gsl_rng_type -> IO (Ptr C'gsl_rng)
λ> :t p'gsl_rng_mt19937
p'gsl_rng_mt19937 :: Ptr (Ptr gsl_rng_type)
λ> :t peek p'gsl_rng_mt19937
peek p'gsl_rng_mt19937 :: IO (Ptr gsl_rng_type)
λ> x <- peek p'gsl_rng_mt19937
λ> :t x
x :: Ptr GHC.Prim.Any
λ> c'gsl_rng_alloc x

<interactive>:421:17:
    Couldn't match type ‘GHC.Prim.Any’ with ‘C'gsl_rng_type’
    Expected type: Ptr C'gsl_rng_type
      Actual type: Ptr GHC.Prim.Any
    In the first argument of ‘c'gsl_rng_alloc’, namely ‘x’
    In the expression: c'gsl_rng_alloc x
λ> 

试图明确指定 peek 的类型 return 也无济于事:

λ> x <- (peek p'gsl_rng_mt19937) :: IO (Ptr gsl_rng_type)
λ> :t x
x :: Ptr GHC.Prim.Any

对@user2407038 的评论进行一些扩展:

When you do x <- peek (ptr :: Ptr (Ptr a)) in the GHCi prompt, the type variable a must be instantiated to some concrete type. This is because the do notation x <- peek p means peek p >>= \x -> ..., where ... is what you type into GHCi afterwards. Since GHCi can't know the future, it has to "cheat" during typechecking.

回想一下 peek p >>= \x -> ...>>= 的右侧参数,即 lambda 抽象 \x -> ...is monomorphic in its argument。这就是 GHCi 必须为 x 分配单态类型的原因。

GHC.Prim.Any 是 GHC 在这种情况下使用的占位符类型,在这种情况下,需要将具体的单态类型分配给没有其他约束的对象。