声明一个数组类型

Question

While discussing another question on SO, I was told that the syntax

int[] numbers = new int[10];

Is better than

Array numbers = Array.CreateInstance(typeof(int), 10);

My question is, when is the first syntax useful and when is the second syntax useful? And why?

Answer 1

This creates a strongly typed monodimensional array "directly":

int[] array = new int[10];

Under the hood it uses the IL command newarr .

This one is more similar to using reflection to create an array (the type can be decided at runtime)

int[] array2 = (int[])Array.CreateInstance(typeof(int), 10);

The array created in the end is the same but the speed of creating it is much slower when using Array.CreateInstance . Note that with Array.CreateInstance you can dynamically select the type of the array (in the same way that through reflection you can create an instance of a type given the type at runtime), for example:

Type type = someCondition ? typeof(int) : typeof(string);
Array array2 = Array.CreateInstance(type, 10);

Another big difference: the "base" Array class is weakly typed (so its methods use object parameters, instead of int / string '/ something ). So:

int num = (int)array2.GetValue(1); // You have to cast the return value to int from object

---

Another reason to use

array[5] = 1;

instead of

array2.SetValue(5, 1);

is that the first method is optimized in the IL code as a direct access to a monodimensional array ( stelem and ldelem ). The same is true for GetValue .

---

The reason I'm using the term "monodimensional array":

In .NET there are two "types" of arrays: the monodimensional arrays and the "complex" arrays (they can be multidimensional, or with the first element not at the 0 index, or both). The second group is much slower. The stelem and ldelem work only with monodimensional arrays. For multidimensional/special arrays "hidden" methods are used to access them (the C# compiler changes the get and set so that these methods are called) (they are similar to the GetValue / SetValue of the Array class, see for example https://stackoverflow.com/a/597729/613130 )

Answer 2

On compile time, looks like there is no big difference between them. Just like arrays, Array.CreateInstance method also make reference-type elements are initialized to null . value-type elements are initialized to zero .

Only difference for the second one is; from MSDN ;

Unlike most classes, Array provides the CreateInstance method, instead of public constructors, to allow for late bound access.

Also as Stefano Altieri mentioned , first syntax requires array size in compile time , but the second one need the size in run-time . For example, you can build these codes successfully;

int[] numbers1 = new int[10];
Array numbers2 = Array.CreateInstance(typeof(int), -1);

But you can't build these;

int[] numbers1 = new int[];
Array numbers2 = Array.CreateInstance(typeof(int), -1);

Answer 3

The samples you wrote are equivalent. The point is that to use the first syntax (new int[size]) you need to know the array type at compile time.

The second syntax allow you to decide at runtime the type of the array items. This is far more powerful, but you just don't need this power always. Ie

Type myType = SelectTheBestType();
var numbers = Array.CreateInstance(myType, 10);

Answer 4

Soner Gönül 's answer is pretty good. A simple reason of your question is

If an array is passed as object or interface, the element type is unknown at compile time.

[ This answer ] might bring you the idea, and [ here ] is some code for example.

when is the first syntax useful and when is the second syntax useful?

The method in the link of example,

static IList CreateArray<T>(this T source);

the generic type parameter is T rather than T[] because of I cannot suppose that source necessarily been an array, but it still possibly . Thus I can only use Array.CreateInstance inside the method.

Most of time, if the element type is already known at compile time, it's better to use the syntax provided by the language.