Skip to content

Latest commit

 

History

History
 
 

p20

Go Quiz: 从Go面试题看函数命名返回值的注意事项

题目

Redhat的首席工程师、Prometheus开源项目Maintainer Bartłomiej Płotka 在Twitter上出了一道Go编程题,结果超过80%的人都回答错了。

题目如下所示,回答下面这段程序的输出结果。

// named_return.go
package main

import "fmt"

func aaa() (done func(), err error) {
	return func() { print("aaa: done") }, nil
}

func bbb() (done func(), _ error) {
	done, err := aaa()
	return func() { print("bbb: surprise!"); done() }, err
}

func main() {
	done, _ := bbb()
	done()
}
  • A: bbb: surprise!
  • B: bbb: surprise!aaa: done
  • C: 编译报错
  • D: 递归栈溢出

大家可以先思考下这段代码的输出结果是什么。

解析

在函数bbb最后执行return语句,会对返回值变量done进行赋值,

done := func() { print("bbb: surprise!"); done() }

注意:闭包func() { print("bbb: surprise!"); done() }里的done并不会被替换成done, err := aaa()里的done的值。

因此函数bbb执行完之后,返回值之一的done实际上成为了一个递归函数,先是打印"bbb: surprise!",然后再调用自己,这样就会陷入无限递归,直到栈溢出。因此本题的答案是D

那为什么函数bbb最后return的闭包func() { print("bbb: surprise!"); done() }里的done并不会被替换成done, err := aaa()里的done的值呢?如果替换了,那本题的答案就是B了。

这个时候就要搬出一句老话了:

This is a feature, not a bug

我们可以看下面这个更为简单的例子,来帮助我们理解:

// named_return1.go
package main

import "fmt"

func test() (done func()) {
	return func() { fmt.Println("test"); done() }
}

func main() {
	done := test()
	// 下面的函数调用会进入死循环,不断打印test
	done()
}

正如上面代码里的注释说明,这段程序同样会进入无限递归直到栈溢出。

如果函数test最后return的闭包func() { fmt.Println("test"); done() }里的done是被提前解析了的话,因为done是一个函数类型的变量,done的零值是nil,那闭包里的done的值就会是nil,执行nil函数是会引发panic的。

但实际上Go设计是允许上面的代码正常执行的(通过这种方式可以返回一个递归函数),因此函数test最后return的闭包里的done的值并不会提前解析,test函数执行完之后,实际上产生了下面的效果,返回的是一个递归函数,和本文开始的题目一样。

done := func() { fmt.Println("test"); done() }

因此也会进入无限递归,直到栈溢出。

总结

这个题目其实很tricky,在实际编程中,要避免对命名返回值采用这种写法,非常容易出错。

想了解国外Go开发者对这个题目的讨论详情可以参考Go Named Return Parameters Discussion

另外题目作者也给了如下所示的解释,原文地址可以参考详细解释

package main

func aaa() (done func(), err error) {
	return func() { print("aaa: done") }, nil
}

func bbb() (done func(), _ error) {
	// NOTE(bwplotka): Here is the problem. We already defined special "return argument" variable called "done".
	// By using `:=` and not `=` we define a totally new variable with the same name in
	// new, local function scope.
	done, err := aaa()

	// NOTE(bwplotka): In this closure (anonymous function), we might think we use `done` from the local scope,
	// but we don't! This is because Go "return" as a side effect ASSIGNS returned values to
	// our special "return arguments". If they are named, this means that after return we can refer
	// to those values with those names during any execution after the main body of function finishes
	// (e.g in defer or closures we created).
	//
	// What is happening here is that no matter what we do in the local "done" variable, the special "return named"
	// variable `done` will get assigned with whatever was returned. Which in bbb case is this closure with
	// "bbb:surprise" print. This means that anyone who runs this closure AFTER `return` did the assignment
	// will start infinite recursive execution.
	//
	// Note that it's a feature, not a bug. We use this often to capture
	// errors (e.g https://github.com/efficientgo/tools/blob/main/core/pkg/errcapture/doc.go)
	//
	// Go compiler actually detects that `done` variable defined above is NOT USED. But we also have `err`
	// variable which is actually used. This makes compiler to satisfy that unused variable check,
	// which is wrong in this context..
	return func() { print("bbb: surprise!"); done() }, err
}

func main() {
	done, _ := bbb()
	done()
}

不过这个解释是有瑕疵的,主要是这句描述:

By using := and not = we define a totally new variable with the same name in new, local function scope.

对于done, err := aaa(),返回变量done并不是一个新的变量,而是和函数bbb的返回变量done是同一个变量。

这里有一个小插曲:本人把这个瑕疵反馈给了原作者,原作者同意了我的意见,删除了这块解释

最新版的英文解释如下,原文地址可以参考修正版解释

package main

func aaa() (done func()) {
	return func() { print("aaa: done") }
}

func bbb() (done func()) {
	done = aaa()

	// NOTE(bwplotka): In this closure (anonymous function), we might think we use `done` value assigned to aaa(),
	// but we don't! This is because Go "return" as a side effect ASSIGNS returned values to
	// our special "return arguments". If they are named, this means that after return we can refer
	// to those values with those names during any execution after the main body of function finishes
	// (e.g in defer or closures we created).
	//
	// What is happening here is that no matter what we do with our "done" variable, the special "return named"
	// variable `done` will get assigned with whatever was returned when the function ends.
	// Which in bbb case is this closure with "bbb:surprise" print. This means that anyone who runs
	// this closure AFTER `return` did the assignment, will start infinite recursive execution.
	//
	// Note that it's a feature, not a bug. We use this often to capture
	// errors (e.g https://github.com/efficientgo/tools/blob/main/core/pkg/errcapture/doc.go)
	return func() { print("bbb: surprise!"); done() }
}

func main() {
	done := bbb()
	done()
}

思考题

下面这段代码同样使用了命名返回值,大家可以看看这个道题的输出结果是什么。可以给微信公众号发送消息nrv获取答案。

package main

func bar() (r int) {
	defer func() {
		r += 4
		if recover() != nil {
			r += 8
		}
	}()
	
	var f func()
	defer f()
	f = func() {
		r += 2
	}

	return 1
}

func main() {
	println(bar())
}

开源地址

文章和示例代码开源在GitHub: Go语言初级、中级和高级教程

公众号:coding进阶。关注公众号可以获取最新Go面试题和技术栈。

个人网站:Jincheng's Blog

知乎:无忌

References