当我接触的F#编程越多,我用到递归的可能性就越大,也正是因为这样,我时常会遇到堆栈溢出的问题,要想避免堆栈溢出问题,Continuation Style Program(CSP)是唯一的方法。以下我们列出普通的递归和CSP的版本代码进行对比,在这里,关键的一点,该方法不会返回,因此它不会在调用的堆栈上创建元素,同时,由于会延迟计算Continuation方法,它不需要被保存在栈元素中:

1 module FunctionReturnModule = 2 let l = [1..1000000] 3 let rec sum l = 4 match l with 5 | [] -> 0 6 | h::t -> h + sum t 7 sum l

1 module CPSModule = 2 let l = [1..1000000] 3 let rec sum l cont = 4 match l with 5 | [] -> cont 0 6 | h::t -> 7 let afterSum v = 8 cont (h+v) 9 sum t afterSum 10 sum l id
好吧,接下来的问题是如何从普通递归的方法得到CSP的递归版本呢?以下是我遵循的步骤,记住:其中一些中间代码并不能通过编译。
首先看看我们原始的递归代码:

1 module FunctionReturnModule = 2 let l = [1..1000000] 3 let rec sum l = 4 match l with 5 | [] -> 0 6 | h::t -> 7 let r = sum t 8 h + r 9 sum l
第一步:

1 module FunctionReturnModule = 2 let l = [1..1000000] 3 let rec sum l cont = 4 match l with 5 | [] -> 0 6 | h::t -> 7 let r = sum t 8 cont (h + r) 9 sum l
第二步:处理递归函数中的sum,将cont移动到afterSum中,afterSum方法获得到参数v并将它传递给cont(h+v):

1 module CPSModule = 2 let l = [1..1000000] 3 let rec sum l cont = 4 match l with 5 | [] -> cont 0 6 | h::t -> 7 let afterSum v = 8 cont (h+v) 9 sum t afterSum 10 sum l id
那么,接下来让我们使用相同的方法来遍历树,下面先列出树的定义:

1 type NodeType = int 2 type BinaryTree = 3 | Nil 4 | Node of NodeType * BinaryTree * BinaryTree
最终的结果如下:

1 module TreeModule = 2 let rec sum tree = 3 match tree with 4 | Nil -> 0 5 | Node(v, l, r) -> 6 let sumL = sum l 7 let sumR = sum r 8 v + sumL + sumR 9 sum deepTree 10 module TreeCSPModule = 11 let rec sum tree cont = 12 match tree with 13 | Nil -> cont 0 14 | Node(v, l, r) -> 15 let afterLeft lValue = 16 let afterRight rValue = 17 cont (v+lValue+rValue) 18 sum r afterRight 19 sum l afterLeft 20 sum deepTree id
开始使用相同的步骤将它转换成CSP方式:
首先切入Continuation函数:

1 module TreeModule = 2 let rec sum tree cont = 3 match tree with 4 | Nil -> 0 5 | Node(v, l, r) -> 6 let sumL = sum l 7 let sumR = sum r 8 cont (v + sumL + sumR) 9 sum deepTree
第一步:处理sumR,将cont方法移动到afterRight中并将它传给sum r:

1 module TreeModule = 2 let rec sum tree cont = 3 match tree with 4 | Nil -> 0 5 | Node(v, l, r) -> 6 let sumL = sum l 7 // let sumR = sum r 8 let afterRight rValue = 9 cont (v + sumL + rValue) 10 sum r afterRight 11 sum deepTree
第二步:处理sumL:

1 module TreeModule = 2 let rec sum tree cont = 3 match tree with 4 | Nil -> 0 5 | Node(v, l, r) -> 6 //let sumL = sum l 7 let afterLeft lValue = 8 let afterRight rValue = 9 cont (v + lValue + rValue) 10 sum r afterRight 11 sum l afterLeft 12 sum deepTree
结束了,接下来让我们用下面的代码进行测试吧:

1 let tree n = 2 let mutable subTree = Node(1, Nil, Nil) 3 for i=0 to n do 4 subTree <- Node(1, subTree, Nil) 5 subTree 6 let deepTree = tree 1000000
注:本文为译文,原文来自:http://apollo13cn.blogspot.com/2012/10/f-continuation-style-programming.html