推荐订阅源

Sometimes you want to write code that works for different primitive types, and as C# doesn't support generic type constraints on primitive type hence you can't avoid writing more code, but you may still one way or another minimise the code you have to write and mysteriously end up having to write something like below,

1  public override bool GetValue<T>(int row, int col, out T val)
2 {
3     CacheBlockDouble cache;
4     int r, c;
5     GetCacheAndPoint(row, col, out cache, out r, out c);
6     var dval = cache.Data[r, c];
7     val = (T) (object) dval;
8     return dval != NoDataValue;
9 }

The main concern is on line 7, where a seemingly boxing and unboxing is happening on primitive type dval (which is of double type here) and this is done like this
And we know that T is determined at compile time and is also double, so we hope that the compiler is smart enough to eliminate the unnecessary boxing and interpret that line as below at run time.

It looks like a simple task as everything can be easily evaluated at compile time. But we need proof before we can be sure.
The most reliable way is to examine the IL, however the following code is sufficient.

 1 using System;
 2 
 3 namespace boxing
 4 {
 5     class Program
 6     {
 7         public static T BoxingTest<T>(double v)
 8         {
 9             T result = (T)(object)v;
10             return result;
11         }
12 
13         public static double NonboxingTest(double v)
14         {
15             return v;
16         }
17 
18         static void Main(string[] args)
19         {
20             long countDown = 900000000;
21             const int buffersize = 4096;
22             var buffer = new double[buffersize];
23             var t1 = DateTime.UtcNow;
24             var i = 0;
25             for (; countDown>0; countDown--)
26             {
27                 var t = BoxingTest<double>(i);
28                 //var t = NonboxingTest(countDown);
29                 buffer[i] = t;
30                 i++;
31                 if (i == 4096)
32                 {
33                     i = 0;
34                 }
35             }
36             var t2 = DateTime.UtcNow;
37             Console.WriteLine("finished in {0} secs", (t2-t1).TotalSeconds);
38         }
39     }
40 }

If we compare them, the generic method takes same amount of time as the non-generic counterpart for any iterations. To this point, I'm quite convinced.

And one more thing which is a bit disappointing kind of adds to the confidence: we can't do such thing like below, which will cause a run time exception InvalidCastException: Specifid cast is not valid

1 var d = 1.0;
2 var f = (float)(object)d;

Not even this,

1 var f = 1.0f;
2 var d = (double)(object)f;