php5 object copy, clone, shallow copy and deep copy

The origin of object copying
Why objects have the concept of "copying"? This is closely related to the value transfer method of objects in PHP5. Let us take a look at the following simple code

PHP code

* /** * * 电视机类 * */ * class Television * { * /** * * 屏幕高度 * */ * protected $_screenLength = 300; * * /** * * 屏幕宽度 * */ * protected $_screenHight = 200; * * /** * * 电视机外观颜色 * */ * protected $_color = 'black'; * * /** * * 返回电视外观颜色 * */ * public function getColor() * { * return $this->_color; * } * * /** * * 设置电视机外观颜色 * */ * public function setColor($color) * { * $this->_color = (string)$color; * return $this; * } * } * * $tv1 = new Television(); * $tv2 = $tv1;

This code defines a TV class Television, $tv1 is an instance of a TV, and then we assign the value of $tv1 to $t2 according to the ordinary variable assignment method. So now we have two TVs $tv1 and $tv2. Is this really the case? Let's test it out.

PHP code

* echo 'color of tv1 is: ' . $tv1->getColor();//tv1的颜色是black * echo '
'; * echo 'color of tv2 is: ' . $tv2->getColor();//tv2的颜色是black * echo '
'; * * //把tv2涂成白色 * $tv2->setColor('white'); * * echo 'color of tv2 is: ' . $tv2->getColor();//tv2的颜色是white * echo '
'; * echo 'color of tv1 is: ' . $tv1->getColor();//tv1的颜色是white

First we see that the colors of tv1 and tv2 are both black. Now we want tv2 to change its color, so we set its color to white. Let’s look at the color of tv2 again. It did become white, which seemed to meet our requirements, but it was not as smooth as expected. When we looked at the color of tv1, we found that tv1 also changed from black to black. white. We did not reset the color of tv1. Why did tv1 change black to white? This is because the assignment and value transfer of objects in PHP5 are all done by "reference". PHP5 uses Zend Engine II, objects are stored in independent structures Object Store, instead of being stored in Zval like other general variables (in PHP4, objects are stored in Zval like general variables). Store only the pointer of the object in Zval and not the content (value). When we copy an object or pass an object as a parameter to a function, we do not need to copy the data. Just keep the same object pointer and notify the Object that this particular object now points to by another zval Store. Since the object itself is located in Object Store, any changes we make to it will affect all zval structures holding pointers to the object - reflected in the program, any changes to the target object will affect the source object. .this makes PHP objects appear to be always passed by reference. So the above tv2 and tv1 actually point to the same TV instance, and the operations we do on tv1 or tv2 are actually for this same instance. So our "copy" failed. It seems that the direct variable assignment method cannot copy the object. For this reason PHP5 provides an operation specifically for copying objects, which is clone. This is where object copying comes in.


Use clone to copy objects
We now use PHP5’s clone language structure to copy objects. The code is as follows:
[size=+0]PHP code

* [size=+0]$tv1 = new Television(); * $tv2 = clone $tv1; * * echo 'color of tv1 is: ' . $tv1->getColor();//tv1的颜色是black * echo '
'; * echo 'color of tv2 is: ' . $tv2->getColor();//tv2的颜色是black * echo '
'; * * //把tv2换成涂成白色 * $tv2->setColor('white'); * * echo 'color of tv2 is: ' . $tv2->getColor();//tv2的颜色是white * echo '
'; * echo 'color of tv1 is: ' . $tv1->getColor();//tv1的颜色是black

In line 2 of this code, we Use the clone keyword to copy tv1. Now we have a real copy of tv1, tv2. We still follow the previous method to check whether the copy is successful. We can see that we changed the color of tv2 to white, and the color of tv1 is still black, so our copy operation is successful.



__clone magic method

Now we consider the situation that each TV should have its own number. This number should be unique like our ID number, so when we copy a TV, we don’t want this The number is also copied to avoid causing some trouble. One strategy we came up with is to clear the assigned TV numbers, and then reassign the numbers as needed.
Then the __clone magic method is specifically used to solve such problems. The __clone magic method will be triggered when the object is copied (that is, the clone operation). We modified the code of the TV class Television and added the number attribute and __clone method. The code is as follows.
PHP code

* /** * * 电视机类 * */ * class Television * { * * /** * * 电视机编号 * */ * protected $_identity = 0; * * /** * * 屏幕高度 * */ * protected $_screenLength = 300; * * /** * * 屏幕宽度 * */ * protected $_screenHight = 200; * * /** * * 电视机外观颜色 * */ * protected $_color = 'black'; * * /** * * 返回电视外观颜色 * */ * public function getColor() * { * return $this->_color; * } * * /** * * 设置电视机外观颜色 * */ * public function setColor($color) * { * $this->_color = (string)$color; * return $this; * } * * /** * * 返回电视机编号 * */ * public function getIdentity() * { * return $this->_identity; * } * * /** * * 设置电视机编号 * */ * public function setIdentity($id) * { * $this->_identity = (int)$id; * return $this; * } * * public function __clone() * { * $this->setIdentity(0); * } * }

Let’s copy such a TV object.

PHP code

* $tv1 = new Television(); * $tv1->setIdentity('111111'); * echo 'id of tv1 is ' . $tv1->getIdentity();//111111 * echo '
'; * * $tv2 = clone $tv1; * echo 'id of tv2 is ' . $tv2->getIdentity();//0

We produced a television tv1, And set its number to 111111, then we use clone to copy tv1 to tv2. At this time, the __clone magic method is triggered. This method will directly act on the copied object tv2. We called the setIdentity member in the __clone method. The method clears the _identity attribute of tv2 so that we can renumber it later. From this we can see that the __clone magic method allows us to do some additional operations very conveniently when cloning an object.

Fatal flaw of clone operation
Can clone really achieve the ideal copying effect? In some cases, you should find that the clone operation is not as perfect as we imagined. Let’s modify the above TV type and then do a test.
Every TV will come with a remote control, so we will have a remote control class. The remote control and the TV are an "aggregation" relationship (relative to the "combination" relationship, it is a weaker dependency relationship. Because generally TVs can be used normally even without a remote control), now our TV objects should all hold a reference to the remote control object. Let’s take a look at the code
PHP code

* /** * * 电视机类 * */ * class Television * { * * /** * * 电视机编号 * */ * protected $_identity = 0; * * /** * * 屏幕高度 * */ * protected $_screenLength = 300; * * /** * * 屏幕宽度 * */ * protected $_screenHight = 200; * * /** * * 电视机外观颜色 * */ * protected $_color = 'black'; * * /** * * 遥控器对象 * */ * protected $_control = null; * * /** * * 构造函数中加载遥控器对象 * */ * public function __construct() * { * $this->setControl(new Telecontrol()); * } * * /** * * 设置遥控器对象 * */ * public function setControl(Telecontrol $control) * { * $this->_control = $control; * return $this; * } * * /** * * 返回遥控器对象 * */ * public function getControl() * { * return $this->_control; * } * * /** * * 返回电视外观颜色 * */ * public function getColor() * { * return $this->_color; * } * * /** * * 设置电视机外观颜色 * */ * public function setColor($color) * { * $this->_color = (string)$color; * return $this; * } * * /** * * 返回电视机编号 * */ * public function getIdentity() * { * return $this->_identity; * } * * /** * * 设置电视机编号 * */ * public function setIdentity($id) * { * $this->_identity = (int)$id; * return $this; * } * * public function __clone() * { * $this->setIdentity(0); * } * } * * * /** * * 遥控器类 * */ * class Telecontrol * { * * }

Copy such a TV object and observe the remote control object of the TV.

PHP code

* $tv1 = new Television(); * $tv2 = clone $tv1; * * $contr1 = $tv1->getControl(); //获取tv1的遥控器contr1 * $contr2 = $tv2->getControl(); //获取tv2的遥控器contr2 * echo $tv1; //tv1的object id 为 #1 * echo '
'; * echo $contr1; //contr1的object id 为#2 * echo '
'; * echo $tv2; //tv2的object id 为 #3 * echo '
'; * echo $contr2; //contr2的object id 为#2

经过复制之后，我们查看对象id，通过clone操作从tv1复制出了tv2,tv1和tv2的对象id分别是 1和3，这表示tv1和tv2是引用两个不同的电视机对象，这符合clone操作的结果。然后我们分别获取了tv1的遥控器对象contr1和tv2的遥控器对象contr2，通过查看它们的对象 id我们发现contr1和contr2的对象id都是2，这表明它们是到同一个对象的引用，也就是说我们虽然从tv1复制出tv2,但是遥控器并没有被复制，每台电视机都应该配有一个遥控器，而这里tv2和tv1共用一个遥控器，这显然是不合常理的。

由此可见，clone操作有这么一个非常大的缺陷：使用clone操作复制对象时，当被复制的对象有对其它对象的引用的时候，引用的对象将不会被复制。然而这种情况又非常的普遍，现今 “合成/聚合复用”多被提倡用来代替“继承复用”，“合成”和“聚合”就是让一个对象拥有对另一个对象的引用，从而复用被引用对象的方法。我们在使用 clone的时候应该考虑到这样的情况。那么在clone对象的时候我们应该如何去解决这样的一个缺陷呢？可能你很快就想到了之前提到的__clone魔术方法，这确实是一种解决方案。

方案1:用__clone魔术方法弥补
前面我们已经介绍了__clone魔术方法的用法，我们可以在__clone方法中将被复制对象中其它对象的引用重新引用到一个新的对象。下面我们看看修改后的__clone()魔术方法：

[size=+0][size=+0]PHP代码

* [size=+0][size=+0]public function __clone() * { * $this->setIdentity(0); * //重新设置一个遥控器对象 * $this->setControl(new Telecontrol()); * }

第04行中我们为复制出来的电视机对象重新设置了一个遥控器，我们按照之前的方法查看对象的id可以发现，两台电视机的遥控器拥有不同的对象id，这样我们的问题就解决了。

但是这样的方式大概并不算太好，如果被复制对象中有多个到其它对象的引用，我们必须在__clone方法中逐个的重新设置，更糟糕的是如果被复制对象的类由第三方提供，我们无法修改代码，那复制操作基本就无法顺利完成了。
我们使用clone来复制对象，这种复制叫做“浅复制”：被复制对象的所有变量都含有与原来的对象相同的值，而所有的对其他对象的引用都仍然指向原来的对象。也就是说，浅复制仅仅复制所考虑的对象，而不复制它所引用的对象。相对于“浅复制”，当然也有一个“深复制”：被复制的对象的所有的变量都含有与原来的对象相同的值，除去那些引用其他对象的变量。也就是说，深复制把要复制的对象所引用的对象都复制了一遍。深复制需要决定深入到多少层，这是一个不容易确定的问题，此外可能会出现循环引用的问题，这些都必须小心处理。我们的方案2将是一个深复制的解决方案。

方案2:利用串行化做深复制
PHP有串行化(serialize)和反串行化(unserialize)函数，我们只需要用serialize()将一个对象写入一个流，然后从流中读回对象，那么对象就被复制了。在JAVA语言里面，这个过程叫做“冷藏”和“解冻”。下面我们将测试一下这个方法：
[size=+0][size=+0]PHP代码

* [size=+0][size=+0]$tv1 = new Television(); * $tv2 = unserialize(serialize($tv1));//序列化然后反序列化 * * $contr1 = $tv1->getControl(); //获取tv1的遥控器contr1 * $contr2 = $tv2->getControl(); //获取tv2的遥控器contr2 * * echo $tv1; //tv1的object id 为 #1 * echo '
'; * echo $contr1; //contr1的object id 为#2 * echo '
'; * echo $tv2; //tv2的object id 为 #4 * echo '
'; * echo $contr2; //contr2的object id 为#5

我们可以看到输出结果，tv1和tv2拥有了不同的遥控器。这比方案1要方便很多，序列化是一个递归的过程，我们不需要理会被对象内部引用了多少个对象以及引用了多少层对象，我们都可以彻底的复制。注意使用此方案时我们无法触发__clone魔术方法来完成一些附加操作，当然我们可以在深复制之后再进行一次clone操作来触发__clone魔术方法，只是会对效率些小的影响。另外此方案会触发被复制对象和所有被引用对象的__sleep和__wakeup魔术方法，所以这些情况都需要被考虑。



总结
不同的对象复制方式有着不同的效果，我们应该根据具体应用需求来考虑使用哪种方式以及如何改进复制方式。PHP5的面向对象特性和JAVA比较接近，相信我们可以从JAVA中借鉴很多宝贵的经验。