我已经阅读了维基百科上关于过程式编程和函数式编程的文章,但我还是有点困惑。有人能把它归结为核心吗?


当前回答

康拉德说:

因此,纯函数式程序总是为输入产生相同的值, 评价的顺序也不明确;这意味着不确定的值,比如 用户输入或随机值很难用纯函数式语言建模。

在一个纯函数式程序中求值的顺序可能很难(或者)解释(尤其是懒惰的人),甚至不重要,但我认为说它没有被很好地定义,听起来就像你根本无法判断你的程序是否会工作!

Perhaps a better explanation would be that control flow in functional programs is based on when the value of a function's arguments are needed. The Good Thing about this that in well written programs, state becomes explicit: each function lists its inputs as parameters instead of arbitrarily munging global state. So on some level, it is easier to reason about order of evaluation with respect to one function at a time. Each function can ignore the rest of the universe and focus on what it needs to do. When combined, functions are guaranteed to work the same[1] as they would in isolation.

... 像用户输入或随机值这样的不确定值很难纯粹地建模 函数式语言。

The solution to the input problem in purely functional programs is to embed an imperative language as a DSL using a sufficiently powerful abstraction. In imperative (or non-pure functional) languages this is not needed because you can "cheat" and pass state implicitly and order of evaluation is explicit (whether you like it or not). Because of this "cheating" and forced evaluation of all parameters to every function, in imperative languages 1) you lose the ability to create your own control flow mechanisms (without macros), 2) code isn't inherently thread safe and/or parallelizable by default, 3) and implementing something like undo (time travel) takes careful work (imperative programmer must store a recipe for getting the old value(s) back!), whereas pure functional programming buys you all these things—and a few more I may have forgotten—"for free".

我希望这听起来不像狂热,我只是想补充一些观点。命令式编程,特别是像c# 3.0这样的强大语言中的混合范式编程,仍然是完成工作的完全有效的方法,并且没有银弹。

[1]…除了内存使用方面(参考Haskell中的foldl和foldl')。

其他回答

进一步阐述康拉德的评论:

因此,纯函数式程序总是对输入产生相同的值,求值的顺序没有很好的定义;

因此,函数式代码通常更容易并行化。由于函数(通常)没有副作用,并且它们(通常)只是作用于它们的参数,因此许多并发问题都消失了。

当您需要能够证明您的代码是正确的时,也可以使用函数式编程。这在过程式编程中要困难得多(在函数式编程中不容易,但仍然容易)。

免责声明:我已经很多年没有使用函数式编程了,直到最近才开始重新研究它,所以我在这里可能不完全正确。:)

我在这里没有看到真正强调的一点是,现代函数语言(如Haskell)实际上更多地关注流控制的第一类函数,而不是显式递归。您不需要像上面那样在Haskell中递归地定义阶乘。我想是这样的

fac n = foldr (*) 1 [1..n]

是一个完美的惯用结构,在精神上更接近于使用循环,而不是使用显式递归。

过程性语言倾向于跟踪状态(使用变量),并倾向于按步骤序列执行。纯函数式语言不跟踪状态,使用不可变值,并倾向于作为一系列依赖项执行。在许多情况下,调用堆栈的状态所保存的信息与过程代码中存储在状态变量中的信息相同。

递归是函数式编程的一个经典例子。

@Creighton:

在Haskell中有一个叫做product的库函数:

prouduct list = foldr 1 (*) list

或者仅仅是:

product = foldr 1 (*)

惯用语的阶乘

fac n = foldr 1 (*)  [1..n]

很简单

fac n = product [1..n]

这里没有一个答案显示了惯用的函数式编程。递归阶乘的答案很适合在FP中表示递归,但大多数代码不是递归的,所以我不认为这个答案是完全具有代表性的。

假设你有一个字符串数组,每个字符串表示一个整数,比如“5”或“-200”。您希望根据内部测试用例检查这个输入字符串数组(使用整数比较)。两种解决方案如下所示

程序上的

arr_equal(a : [Int], b : [Str]) -> Bool {
    if(a.len != b.len) {
        return false;
    }

    bool ret = true;
    for( int i = 0; i < a.len /* Optimized with && ret*/; i++ ) {
        int a_int = a[i];
        int b_int = parseInt(b[i]);
        ret &= a_int == b_int;  
    }
    return ret;
}

功能

eq = i, j => i == j # This is usually a built-in
toInt = i => parseInt(i) # Of course, parseInt === toInt here, but this is for visualization

arr_equal(a : [Int], b : [Str]) -> Bool =
    zip(a, b.map(toInt)) # Combines into [Int, Int]
   .map(eq)
   .reduce(true, (i, j) => i && j) # Start with true, and continuously && it with each value

虽然纯函数式语言通常是研究语言(因为现实世界喜欢免费的副作用),但现实世界的过程式语言在适当的时候会使用更简单的函数式语法。

这通常是用Lodash这样的外部库实现的,或者是用Rust这样的新语言内置的。函数式编程的繁重工作是通过map、filter、reduce、currying、partial等函数/概念完成的,最后三个你可以查阅以进一步理解。

齿顶高

In order to be used in the wild, the compiler will normally have to work out how to convert the functional version into the procedural version internally, as function call overhead is too high. Recursive cases such as the factorial shown will use tricks such as tail call to remove O(n) memory usage. The fact that there are no side effects allows functional compilers to implement the && ret optimization even when the .reduce is done last. Using Lodash in JS, obviously does not allow for any optimization, so it is a hit to performance (Which isn't usually a concern with web development). Languages like Rust will optimize internally (And have functions such as try_fold to assist && ret optimization).