此搜索方法中IComparable和IComparable <T>之间的差异

我知道,有一个很大的区别IComparable,并IComparable在一般情况下,看到这,但在这个搜索方法就没有任何区别,还是会？

public static int Search(List a, T target) where T : IComparable
{
    for (int i = 0; i < a.Count; i++)
    {
        if (target.CompareTo(a[i]) == 0)
            return i;
    }
    return -1;
}

相比:

public static int Search(List a, T target) where T : IComparable
{
    ...
}

两者都会起作用并给出相同的结果吗？

两者都会起作用并给出相同的结果吗？

让我们看一下编译器为两种搜索方法发出的内容:

Search:
IL_0000:  ldc.i4.0    
IL_0001:  stloc.0     // i
IL_0002:  br.s        IL_0025
IL_0004:  ldarga.s    01 
IL_0006:  ldarg.0     
IL_0007:  ldloc.0     // i
IL_0008:  callvirt    06 00 00 0A 
IL_000D:  box         03 00 00 1B <---- Notice this!
IL_0012:  constrained. 03 00 00 1B 
IL_0018:  callvirt    System.IComparable.CompareTo
IL_001D:  brtrue.s    IL_0021
IL_001F:  ldloc.0     // i
IL_0020:  ret         
IL_0021:  ldloc.0     // i
IL_0022:  ldc.i4.1    
IL_0023:  add         
IL_0024:  stloc.0     // i
IL_0025:  ldloc.0     // i
IL_0026:  ldarg.0     
IL_0027:  callvirt    08 00 00 0A 
IL_002C:  blt.s       IL_0004
IL_002E:  ldc.i4.m1   
IL_002F:  ret         

SearchGeneric:
IL_0000:  ldc.i4.0    
IL_0001:  stloc.0     // i
IL_0002:  br.s        IL_0020
IL_0004:  ldarga.s    01 
IL_0006:  ldarg.0     
IL_0007:  ldloc.0     // i
IL_0008:  callvirt    06 00 00 0A 
IL_000D:  constrained. 03 00 00 1B 
IL_0013:  callvirt    09 00 00 0A 
IL_0018:  brtrue.s    IL_001C
IL_001A:  ldloc.0     // i
IL_001B:  ret         
IL_001C:  ldloc.0     // i
IL_001D:  ldc.i4.1    
IL_001E:  add         
IL_001F:  stloc.0     // i
IL_0020:  ldloc.0     // i
IL_0021:  ldarg.0     
IL_0022:  callvirt    08 00 00 0A 
IL_0027:  blt.s       IL_0004
IL_0029:  ldc.i4.m1   
IL_002A:  ret    

如果你仔细观察,你会发现主要的不同之处在于Search,在调用之前每个调用CompareTo都必须将值包装(这在box操作中是值得注意的),因为非泛型版本接受一个object类型.

让我们尝试使用值类型分析两者之间的性能差异.我将使用BenchmarkDotNet,这是一个小的(并且真正令人敬畏的)基准测试框架,它负责JITing,CPU预热等.

测试:

[BenchmarkTask(platform: BenchmarkPlatform.X86)]
[BenchmarkTask(platform: BenchmarkPlatform.X64)]
public class Test
{
    private readonly List list = Enumerable.Range(0, 1000000).ToList();

    [Benchmark]
    public void TestSearch()
    {
        Search(list, 999999);
    }

    [Benchmark]
    public void TestSearchGeneric()
    {
        SearchGeneric(list, 999999);
    }

    public static int Search(List a, T target) where T : IComparable
    {
        for (int i = 0; i < a.Count; i++)
        {
            if (target.CompareTo(a[i]) == 0)
                return i;
        }
        return -1;
    }
    public static int SearchGeneric(List a, T target) where T : IComparable
    {
        for (int i = 0; i < a.Count; i++)
        {
            if (target.CompareTo(a[i]) == 0)
                return i;
        }
        return -1;
    }
}

结果:

***** Competition: Finish  *****

BenchmarkDotNet=v0.7.8.0
OS=Microsoft Windows NT 6.1.7601 Service Pack 1
Processor=Intel(R) Core(TM) i7-4600U CPU @ 2.10GHz, ProcessorCount=4
HostCLR=MS.NET 4.0.30319.42000, Arch=32-bit
Type=Test  Mode=Throughput  Jit=HostJit  .NET=HostFramework


            Method | Platform |    AvrTime |    StdDev |   op/s |
------------------ |--------- |----------- |---------- |------- |
        TestSearch |      X64 | 35.8065 ms | 3.3439 ms |  27.93 |
 TestSearchGeneric |      X64 |  4.6427 ms | 0.3075 ms | 215.40 |
        TestSearch |      X86 | 26.4876 ms | 1.4776 ms |  37.75 |
 TestSearchGeneric |      X86 |  6.6500 ms | 0.1664 ms | 150.38 |

***** Competition: End *****

看到发生装箱操作的非泛型方法在x86上慢了4 倍,在x64上慢了8倍.这可能会对您的应用程序性能产生影响.

我通常不会使用非泛型IComparable,这主要是为了向后兼容泛型之前的日子.请注意,另一个不太重要的因素是通过泛型获得的类型安全性.