为了便于理解，你可以参考这篇文章: https://www.cs.utexas.edu/~scottm/cs307/handouts/deepCopying.htm

浅拷贝:

深拷贝:

除了上述所有定义之外，还有一个最常用的深度复制，位于类的复制构造函数(或重载赋值操作符)中。

浅复制——>是当你没有提供复制构造函数时。在这里，只复制对象，但不复制类的所有成员。

深度复制——>是当您决定在类中实现复制构造函数或重载赋值，并允许复制类的所有成员时。

MyClass& MyClass(const MyClass& obj) // copy constructor for MyClass
{
          // write your code, to copy all the members and return the new object
}
MyClass& operator=(const MyClass& obj) // overloading assignment operator,
{
          // write your code, to copy all the members and return the new object
}

{想象两个对象:相同类型_t的A和B(相对于c++)，你正在考虑将A浅/深复制到B}

浅拷贝: 简单地将a的引用复制到b中，把它看作是a的地址的副本。 因此，A和B的地址将是相同的，即它们将指向相同的内存位置，即数据内容。

Deep copy: Simply makes a copy of all the members of A, allocates memory in a different location for B and then assigns the copied members to B to achieve deep copy. In this way, if A becomes non-existant B is still valid in the memory. The correct term to use would be cloning, where you know that they both are totally the same, but yet different (i.e. stored as two different entities in the memory space). You can also provide your clone wrapper where you can decide via inclusion/exclusion list which properties to select during deep copy. This is quite a common practice when you create APIs.

只有当你了解其中的利害关系时，你才可以选择做浅拷贝。当你在c++或C中有大量的指针要处理时，做一个对象的浅拷贝真的是一个坏主意。

EXAMPLE_OF_DEEP COPY_ An example is, when you are trying to do image processing and object recognition you need to mask "Irrelevant and Repetitive Motion" out of your processing areas. If you are using image pointers, then you might have the specification to save those mask images. NOW... if you do a shallow copy of the image, when the pointer references are KILLED from the stack, you lost the reference and its copy i.e. there will be a runtime error of access violation at some point. In this case, what you need is a deep copy of your image by CLONING it. In this way you can retrieve the masks in case you need them in the future.

EXAMPLE_OF_SHALLOW_COPY I am not extremely knowledgeable compared to the users in StackOverflow so feel free to delete this part and put a good example if you can clarify. But I really think it is not a good idea to do shallow copy if you know that your program is gonna run for an infinite period of time i.e. continuous "push-pop" operation over the stack with function calls. If you are demonstrating something to an amateur or novice person (e.g. C/C++ tutorial stuff) then it is probably okay. But if you are running an application such as surveillance and detection system, or Sonar Tracking System, you are not supposed to keep shallow copying your objects around because it will kill your program sooner or later.

复制array:

Array是一个类，这意味着它是引用类型，因此array1 = array2的结果 在引用同一个数组的两个变量中。

但是看看这个例子:

  static void Main()
    {
        int[] arr1 = new int[] { 1, 2, 3, 4, 5 }; 
        int[] arr2 = new int[] { 6, 7, 8, 9, 0 };

        Console.WriteLine(arr1[2] + " " + arr2[2]);
        arr2 = arr1;
        Console.WriteLine(arr1[2] + " " + arr2[2]); 
        arr2 = (int[])arr1.Clone();
        arr1[2] = 12;
        Console.WriteLine(arr1[2] + " " + arr2[2]);
    }

浅克隆意味着只复制克隆数组所表示的内存。

如果数组包含值类型对象，则复制值;

如果数组包含引用类型，则只复制引用-因此有两个数组，其成员引用相同的对象。

要创建复制引用类型的深度复制，必须循环遍历数组并手动克隆每个元素。

简而言之，这取决于什么指向什么。在浅拷贝中，对象B指向对象a在内存中的位置。在深度复制中，对象A的内存位置中的所有东西都被复制到对象B的内存位置。

这篇wiki文章有一个很好的图表。

http://en.wikipedia.org/wiki/Object_copy

为了避免混淆浅复制和简单地为list分配一个新变量名，再添加一点。

“假设我们有:

x = [
    [1,2,3],
    [4,5,6],
    ]

这个语句创建了3个列表:2个内部列表和一个外部列表。然后，外部列表的引用在名称x下可用

y = x

没有数据被复制。我们在内存的某个地方仍然有相同的3个列表。所有这一切所做的是使外部列表在名称y下可用，除了它之前的名称x

y = list(x)

or

y = x[:]

这将创建一个与x内容相同的新列表。列表x包含对两个内部列表的引用，因此新列表也将包含对这两个内部列表的引用。只复制了一个列表——外层列表。 现在内存中有4个列表，两个内部列表，一个外部列表，以及外部列表的副本。原始外部列表的名称为x，新的外部列表的名称为y。

内部列表没有被复制!此时，您可以从x或y访问和编辑内部列表!

如果你有一个二维(或更高)的列表，或者任何类型的嵌套数据结构，并且你想对所有内容进行完整复制，那么你想在复制模块中使用deepcopy()函数。你的解决方案也适用于2-D列表，迭代外部列表中的项目，并对每个项目进行复制，然后为所有内部副本构建一个新的外部列表。”

来源:https://www.reddit.com/r/learnpython/comments/1afldr/why_is_copying_a_list_so_damn_difficult_in_python/