该Count()方法针对ICollection<T>类型进行了优化,因此GetEnumerator()/MoveNext()/Dispose()不使用模式。
list.Count();
被翻译成
((ICollection)list).Count;
而Any()必须建立一个枚举器。所以Count()方法更快。
IEnumerable这里是 4 个不同实例的基准。看起来MyEmpty像IEnumerable<T> MyEmpty<T>() { yield break; }
iterations : 100000000
Function Any() Count()
new List<int>() 4.310 2.252
Enumerable.Empty<int>() 3.623 6.975
new int[0] 3.960 7.036
MyEmpty<int>() 5.631 7.194
正如casperOne在评论中所说,Enumerable.Empty<int>() is ICollection<int>, 因为它是一个数组,并且数组不适合Count()扩展,因为转换ICollection<int>为不是微不足道的。
无论如何,对于自制的空IEnumerable,我们可以看到我们所期望的,即Count()比 慢Any(),因为测试是否IEnumerable为ICollection.
完整的基准:
class Program
{
public const long Iterations = (long)1e8;
static void Main()
{
var results = new Dictionary<string, Tuple<TimeSpan, TimeSpan>>();
results.Add("new List<int>()", Benchmark(new List<int>(), Iterations));
results.Add("Enumerable.Empty<int>()", Benchmark(Enumerable.Empty<int>(), Iterations));
results.Add("new int[0]", Benchmark(new int[0], Iterations));
results.Add("MyEmpty<int>()", Benchmark(MyEmpty<int>(), Iterations));
Console.WriteLine("Function".PadRight(30) + "Any()".PadRight(10) + "Count()");
foreach (var result in results)
{
Console.WriteLine("{0}{1}{2}", result.Key.PadRight(30), Math.Round(result.Value.Item1.TotalSeconds, 3).ToString().PadRight(10), Math.Round(result.Value.Item2.TotalSeconds, 3));
}
Console.ReadLine();
}
public static Tuple<TimeSpan, TimeSpan> Benchmark(IEnumerable<int> source, long iterations)
{
var anyWatch = new Stopwatch();
anyWatch.Start();
for (long i = 0; i < iterations; i++) source.Any();
anyWatch.Stop();
var countWatch = new Stopwatch();
countWatch.Start();
for (long i = 0; i < iterations; i++) source.Count();
countWatch.Stop();
return new Tuple<TimeSpan, TimeSpan>(anyWatch.Elapsed, countWatch.Elapsed);
}
public static IEnumerable<T> MyEmpty<T>() { yield break; }
}