记一次 Rust 探险

2022-04-17T00:00:00+00:00
问题的起始</h2>
最近在看《The Book》，第 17 章的 17.3 节里面有这样一个细节：</p>
……</p>
请求审核博文来改变其状态</p>
接下来需要增加请求审核博文的功能，这应当将其状态由 Draft</code> 改为 PendingReview</code>。示例 17-15 展示了这段代码：</p> </blockquote>
pub</span> struct</span> Post</span> {</span></span>
    state</span>:</span> Option</span><</span>Box</span><</span>dyn</span> State</span>>>,</span></span>
    content</span>:</span> String</span>,</span></span>
}</span></span>
</span>
impl</span> Post</span> {</span></span>
   // --snip--</span></span>
    pub fn</span> new</span>()</span> -></span> Post</span> {</span></span>
        Post</span> {</span></span>
            state</span>:</span> Some</span>(</span>Box</span>::</span>new</span>(</span>Draft</span> {})),</span></span>
            content</span>:</span> String</span>::</span>new</span>(),</span></span>
        }</span></span>
    }</span></span>
</span>
    pub fn</span> add_text</span>(</span>&mut</span> self</span>, text</span>: &</span>str</span>) {</span></span>
        self</span>.</span>content</span>.</span>push_str</span>(text);</span></span>
    }</span></span>
</span>
    pub fn</span> content</span>(</span>&</span>self</span>)</span> -> &</span>str</span> {</span></span>
        ""</span></span>
    }</span></span>
</span>
    pub fn</span> request_review</span>(</span>&mut</span> self</span>) {</span></span>
        if</span> let</span> Some</span>(s)</span> =</span> self</span>.</span>state</span>.</span>take</span>() {</span></span>
            self</span>.</span>state </span>=</span> Some</span>(s</span>.</span>request_review</span>())</span></span>
        }</span></span>
    }</span></span>
}</span></span>
</span>
trait</span> State</span> {</span></span>
    fn</span> request_review</span>(</span>self</span>:</span> Box</span><</span>Self</span>>)</span> -></span> Box</span><</span>dyn</span> State</span>>;</span></span>
}</span></span>
</span>
struct</span> Draft</span> {}</span></span>
</span>
impl</span> State</span> for</span> Draft</span> {</span></span>
    fn</span> request_review</span>(</span>self</span>:</span> Box</span><</span>Self</span>>)</span> -></span> Box</span><</span>dyn</span> State</span>> {</span></span>
        Box</span>::</span>new</span>(</span>PendingReview</span> {})</span></span>
    }</span></span>
}</span></span>
</span>
struct</span> PendingReview</span> {}</span></span>
</span>
impl</span> State</span> for</span> PendingReview</span> {</span></span>
    fn</span> request_review</span>(</span>self</span>:</span> Box</span><</span>Self</span>>)</span> -></span> Box</span><</span>dyn</span> State</span>> {</span></span>
        self</span></span>
    }</span></span>
}</span></span></code></pre>

……</p>
我们需要将 state</code> 临时设置为 None</code> 来获取 state</code> 值，即老状态的所有权，而不是使用 self.state = self.state.request_review();</code> 这样的代码直接更新状态值。这确保了当 Post</code> 被转换为新状态后不能再使用老 state</code> 值。</strong></p>
……</p>
</blockquote>
这里让我非常困惑，我认为并不是这个原因，因此找了不少解释，以下是总结。</p>
Box 的性质</h2>
首先来看可变性。

在 Rust 中，一个变量是否可变，取决于是否用 mut</code> 修饰变量绑定。</p>
如果我们用 let var : T</code> 声明，那么 var</code> 是不可变的；而且 var</code> 内部所有的成员也都是不可变的；

如果我们用 let mut var : T</code> 声明，那么 var</code> 是可变的，相应的它的内部所有成员也都是可变的。

这样的话，如果有个结构体引用 &SomeStruct</code>，则 SomeStruct</code> 的所有字段都是不可变的。

但在实际开发中，确实存在需要结构体中的某个字段可变的情况。针对这种情况，Rust 的标准库中有个 std::cell</code> 模块，通过共享的可变容器允许以受控的方式进行可变性操作。</p>
在 std::cell</code> 模块中的 Cell</code> 和 RefCell</code> 是实现内部可变性的容器</strong>，在保持容器不被 drop</code> 的情况下可以修改其中的值；而 Box</code> 虽然一直都被拿来和前两个一起讨论，但 Box</code> 并没在 std::cell</code> 模块中，它对可变性其实没啥限定。

Box</code> 是一个指针</strong>，有所有权和生命周期，指向堆上的某个位置，和普通的指针不同的是，它独占</strong>了对数据的所有权；

Cell</code> 不是</strong>一个指针，Cell<T></code> 只是把数据 T 包装一下，告诉你它有“内部可变性”，数据还是那些数据。</p>
其次来看所有权规则。

Box<T></code> 不是 Copy</code> 类型，因为 Box</code> 拥有分配在堆上的缓冲区，非 Copy</code> 类型如果采用逐位复制，那么会使编译器无法分辨哪个值需要对被引用的原始资源</strong>负责。</p>
性质已经搞清，接下来我们来看具体问题。</p>
为什么直接赋值不行</h2>
我在 stackoverflow 上找到一个有关的问答，回答者把它换成了直接版本：</p>
pub</span> struct</span> Post</span> {</span></span>
    state</span>:</span> Box</span><</span>dyn</span> State</span>>,</span></span>
    content</span>:</span> String</span>,</span></span>
}</span></span>
</span>
trait</span> State</span> {</span></span>
    fn</span> request_review</span>(</span>self</span>:</span> Box</span><</span>Self</span>>)</span> -></span> Box</span><</span>dyn</span> State</span>>; </span></span>
}</span></span>
</span>
impl</span> Post</span> {</span></span>
    // ... </span></span>
    pub fn</span> request_review</span>(</span>&mut</span> self</span>) {</span></span>
        self</span>.</span>state </span>=</span> self</span>.</span>state</span>.</span>request_review</span>();</span></span>
    }</span></span>
    // ... </span></span>
}</span></span></code></pre>
编译器报错：</p>
self.state = self.state.request_review();</span></span>
             ^^^^^^ move occurs because `self.state` has type `std::boxed::Box<dyn State>`, which does not implement the `Copy` trait'.</span></span></code></pre>
原回答：

This is because calling State::request_review</code> will move Box<self></code>, which is allocated on heap, and Rust doesn’t allow you to just move values away from heap unless you implement Copy</code>, otherwise what’s left there? The book uses Option::take()</code> to move ownership out and leave None</code> on the place.</p>
翻译如下：

这是因为调用 State::request_review</code> 将 move</code> 分配在堆上的 Box<self></code>，而 Rust 不允许你将值从堆中 move</code>，除非你实现 Copy</code>，否则原值还剩下什么呢？The book 里面使用 Option::take()</code> 将所有权移出，并为原值保留了 None</code> 。</p>
其实已经讲得很清楚了，但我还要问一句：move</code> 发生在哪里？</p>
self.state = self.state.request_review();</span></span></code></pre>
这一句中的 self.state.request_review();</code> 使得结构体 Post</code> 的 state</code> 字段被 request_review()</code> 方法赋了新值，并取得了 state</code> 的所有权，此时 self.state</code> 的所有权转移进了 request_review()</code> 方法内，self.state</code> 变成未初始化状态，而 Rust 编译器禁止使用未经初始化的变量，因为这会产生未定义行为。

因此编译器报错，还指出了我们没有为 std::boxed::Box<dyn State></code> 实现 Copy</code> trait，如果实现了这个 trait，self.state</code> 在 self.state.request_review()</code> 之后依然能保持初始化状态，也就可以完成 self.state =</code> 这部分的赋值了。</p>
2023.2.16 追加</h2>
再读《The Book》，想到自己还写过这么一篇文章，回来看的时候发现有个地方没说透：</p>

Rust 不允许你将值从堆中 move</code>，除非你实现 Copy</code></p>
</blockquote>
这句话什么意思？还是要看报错信息</a>：</p>
   Compiling</span> playground v0</span>.</span>0</span>.</span>1</span> (</span>/</span>playground)</span></span>
error[</span>E0507</span>]</span>:</span> cannot</span> move</span> out of `</span>self</span>.</span>state` which is behind a mutable reference</span></span>
  --></span> src</span>/</span>lib</span>.</span>rs</span>:</span>12</span>:</span>22</span></span>
   |</span></span>
12</span> |</span>         self</span>.</span>state </span>=</span> self</span>.</span>state</span>.</span>request_review</span>();</span></span>
   |                      ^^^^^^^^^^ ----------------</span> `</span>self</span>.</span>state` moved due to this method call</span></span>
   |                      |</span></span>
   |</span>                      move</span> occurs because `</span>self</span>.</span>state` has</span> type</span> `</span>Box</span><</span>dyn</span> State</span>>`, which does not implement the `</span>Copy</span>` </span>trait</span></span>
   |</span></span>
note</span>:</span> this function takes ownership of the receiver `</span>self</span>`, which moves `</span>self</span>.</span>state`</span></span>
  --></span> src</span>/</span>lib</span>.</span>rs</span>:</span>7</span>:</span>23</span></span>
   |</span></span>
7</span>  |</span>     fn</span> request_review</span>(</span>self</span>:</span> Box</span><</span>Self</span>>)</span> -></span> Box</span><</span>dyn</span> State</span>>; </span></span>
   |                       ^^^^</span></span>
</span>
For</span> more information about this error,</span> try</span> `rustc</span> --</span>explain</span> E0507</span>`</span>.</span></span>
error</span>:</span> could not compile `playground` due to previous error</span></span></code></pre>
我们的操作会 cannot move out of *self.state* which is behind a mutable reference</code>，也就是说会在可变引用时 move</code> 变量（转移所有权）。

看一下函数签名： fn request_review(&mut self)</code>，果然使用的是 Post</code> 的可变引用（也就是 &mut self</code>），然后在 self.state.request_review()</code> 这一步时 move</code> 了 self.state</code>，如果 self.state</code> 没有实现 Copy</code> 这个 trait，那么根据 Rust 规则，使用可变引用时是不能 move</code> 的，一旦 move</code> 了，对象的内存地址改变了，那引用还有效吗？</p>

从借用规则角度看，可变引用是「独占引用」，也即在存在可变引用的时候是不能其他引用的，如果把值看成一种「可变引用」，那么很明显发生了冲突，之前的可变引用很可能就失效了。</li>
从所有权角度看，可以把可变引用看成持有一段时间的所有权的值的特殊状态，这段时间内原值是不拥有所有权的，所以原值也没有能力去 move</code>。</li>
实现了 Copy</code> trait 意味着不会真的转移所有权，而是会原封不动地复制一份出去，不会对原值有影响，因此引用就还会是有效状态。</li>
</ul>
说了这么多，让我看看书里的操作是什么：</p>
impl</span> Post</span> {</span></span>
    // --snip--</span></span>
    pub fn</span> request_review</span>(</span>&mut</span> self</span>) {</span></span>
        if</span> let</span> Some</span>(s)</span> =</span> self</span>.</span>state</span>.</span>take</span>() {</span></span>
            self</span>.</span>state </span>=</span> Some</span>(s</span>.</span>request_review</span>())</span></span>
        }</span></span>
    }</span></span>
}</span></span></code></pre>
人家的 state</code> 用 Option</code> 给包住了，而且取值和赋值是分开写的，没有写在一行里，从 take</code> 的签名：pub fn take(&mut self) -> Option<T></code> 中看出 take</code> 用的是可变借用而不是 self</code>，也就没有 move</code> 了，同时生命周期推断也能发挥作用，当然就规避掉了这些问题。</p>
问题解决！感谢这些在网上无私分享知识的人们！</p>
参考阅读</h2>

why do we need to call take() for Option<T> variable</a></li>
所有权和移动</a></li>
变量先声明</a></li>
聊聊 Rust 中的所有权机制</a></li>
Rust 中几个智能指针的异同与使用场景</a></li>
【Rust 每周一知】如何理解 Rust 中的可变与不可变？</a></li>
</ol>
whx's Blog - rust

记一次 Rust 探险
