假设有两个数组arr1和arr2。

arr1 = arr2;   //shallow copy
arr1 = arr2.clone(); //deep copy

我从下面几句话中理解到。

浅拷贝将对象的值类型(int, float, bool)字段复制到目标对象中，对象的引用类型(字符串，类等)被复制为目标对象中的引用。在此目标引用类型将指向源对象的内存位置。

深度复制将对象的值和引用类型复制到目标对象的完整新副本中。这意味着值类型和引用类型都将被分配一个新的内存位置。

深拷贝

深度复制复制所有字段，并复制由字段指向的动态分配的内存。当对象与其引用的对象一起复制时，就会发生深度复制。

浅拷贝

浅复制是对象的逐位复制。创建一个新对象，该对象具有原始对象中值的精确副本。如果对象的任何字段是对其他对象的引用，则只复制引用地址，即只复制内存地址。

我想给出一个例子，而不是正式的定义。

var originalObject = { 
    a : 1, 
    b : 2, 
    c : 3,
};

这段代码显示了一个浅拷贝:

var copyObject1 = originalObject;

console.log(copyObject1.a);         // it will print 1 
console.log(originalObject.a);       // it will also print 1 
copyObject1.a = 4; 
console.log(copyObject1.a);           //now it will print 4 
console.log(originalObject.a);       // now it will also print 4

var copyObject2 = Object.assign({}, originalObject);

console.log(copyObject2.a);        // it will print 1 
console.log(originalObject.a);      // it will also print 1 
copyObject2.a = 4; 
console.log(copyObject2.a);        // now it will print 4 
console.log(originalObject.a);      // now it will print 1

这段代码显示了一个深度拷贝:

var copyObject2 = Object.assign({}, originalObject);

console.log(copyObject2.a);        // it will print 1 
console.log(originalObject.a);      // it will also print 1 
copyObject2.a = 4; 
console.log(copyObject2.a);        // now it will print 4 
console.log(originalObject.a);      // !! now it will print 1 !!

广度vs深度;从引用树的角度考虑，将对象作为根节点。

浅:

变量A和B指的是不同的内存区域，当B被分配给A时，这两个变量指的是相同的内存区域。后来对其中一个内容的修改立即反映在另一个的内容中，因为它们共享内容。

深:

变量A和B指的是不同的内存区域，当B被分配给A时，A所指向的内存区域中的值被复制到B所指向的内存区域。后来对其中一项内容的修改仍为A或B所独有;内容不共享。

{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.