复制KeyedCollection的最快方法

Question

我正在开发一个C#对象复制构造函数，其中一部分涉及到将KeyedCollection的内容复制到新的KeyedCollection中。这是我目前已经实现的：

class MyKeyedCollection : KeyedCollection<uint, DataObject>
{
    protected override uint GetKeyForItem( DataObject do )
    {
        return do.Key;
    }
}

class MyObject
{
    private MyKeyedCollection kc;

    // Copy constructor
    public MyObject( MyObject that )
    {
        this.kc = new MyKeyedCollection();
        foreach ( DataObject do in that.kc )
        {
            this.kc.Add( do );
        }
    }
}

这做了正确的事情--按照预期复制集合。问题是它也有点慢。我猜问题在于每个.Add(do)都需要对现有数据进行唯一性检查，即使我知道它来自保证唯一性的来源。

怎样才能让这个复制构造函数尽可能快呢？

Answer 1

好吧，用一些不安全的代码来解决这个问题怎么样？只是为了好玩？

警告！这是为windows OS和32位编写的，但没有理由不能将此技术修改为适用于64位或其他操作系统。最后，我在3.5框架上测试了它。我认为它可以在2.0和3.0上工作，但我没有测试。如果Redmond在修订或补丁之间更改实例变量的数量、类型或顺序，那么这将不起作用。

但这太快了！

它侵入KeyedCollection、它的底层List<>和Dictionary<>，并复制所有内部数据和属性。这是一个技巧，因为要做到这一点，你必须访问私有内部变量。我基本上为KeyedCollection、List和Dictionary做了一些结构，它们是这些类的私有变量，顺序是正确的。我只需将这些结构指向类所在的位置，voila...you就可以处理私有变量！！我使用RedGate反射器来查看所有代码都在做什么，这样我就可以知道要复制什么。然后，只需复制一些值类型，并在几个地方使用Array.Copy。

结果是CopyKeyedCollection<，>，CopyDict<>和CopyList<>。你会得到一个可以快速复制Dictionary<>的函数和一个可以免费快速复制List<>的函数！

在解决所有问题时，我注意到一件事，KeyedCollection包含一个列表和一个字典，它们都指向相同的数据！一开始我认为这很浪费，但是评论者指出KeyedCollection是专门为同时需要有序列表和字典的情况而设计的。

无论如何，我是一个汇编/c程序员，被迫使用vb一段时间，所以我不怕这样做。我是C#的新手，所以请告诉我我是否违反了任何规则，或者您是否认为这很酷。

顺便说一句，我研究了垃圾收集，这应该可以很好地与GC一起工作。我认为如果我添加一些代码来修复我们复制所花费的ms的一些内存，这将是谨慎的。你们来告诉我。如果有人请求em，我会添加一些注释。

using System;
using System.Collections.Generic;
using System.Text;
using System.Runtime.InteropServices;
using System.Collections.ObjectModel;
using System.Reflection;

namespace CopyCollection {

  class CFoo {
    public int Key;
    public string Name;
  }

  class MyKeyedCollection : KeyedCollection<int, CFoo> {
    public MyKeyedCollection() : base(null, 10) { }
    protected override int GetKeyForItem(CFoo foo) {
      return foo.Key;
    }
  }

  class MyObject {
    public MyKeyedCollection kc;

    // Copy constructor
    public MyObject(MyObject that) {
      this.kc = new MyKeyedCollection();
      if (that != null) {
        CollectionTools.CopyKeyedCollection<int, CFoo>(that.kc, this.kc);
      }
    }
  }

  class Program {

    static void Main(string[] args) {

      MyObject mobj1 = new MyObject(null);
      for (int i = 0; i < 7; ++i)
        mobj1.kc.Add(new CFoo() { Key = i, Name = i.ToString() });
      // Copy mobj1
      MyObject mobj2 = new MyObject(mobj1);
      // add a bunch more items to mobj2
      for (int i = 8; i < 712324; ++i)
        mobj2.kc.Add(new CFoo() { Key = i, Name = i.ToString() });
      // copy mobj2
      MyObject mobj3 = new MyObject(mobj2);
      // put a breakpoint after here, and look at mobj's and see that it worked!
      // you can delete stuff out of mobj1 or mobj2 and see the items still in mobj3,
    }
  }

  public static class CollectionTools {

    public unsafe static KeyedCollection<TKey, TValue> CopyKeyedCollection<TKey, TValue>(
     KeyedCollection<TKey, TValue> src, 
     KeyedCollection<TKey, TValue> dst) {

      object osrc = src;
      // pointer to a structure that is a template for the instance variables 
      // of KeyedCollection<TKey, TValue>
      TKeyedCollection* psrc = (TKeyedCollection*)(*((int*)&psrc + 1));  
      object odst = dst;
      TKeyedCollection* pdst = (TKeyedCollection*)(*((int*)&pdst + 1));
      object srcObj = null;
      object dstObj = null;
      int* i = (int*)&i;  // helps me find the stack

      i[2] = (int)psrc->_01_items;
      dstObj = CopyList<TValue>(srcObj as List<TValue>);
      pdst->_01_items = (uint)i[1];

      // there is no dictionary if the # items < threshold
      if (psrc->_04_dict != 0) {
        i[2] = (int)psrc->_04_dict;
        dstObj = CopyDict<TKey, TValue>(srcObj as Dictionary<TKey, TValue>);
        pdst->_04_dict = (uint)i[1];
      }

      pdst->_03_comparer = psrc->_03_comparer;
      pdst->_05_keyCount = psrc->_05_keyCount;
      pdst->_06_threshold = psrc->_06_threshold;
      return dst;
    }

    public unsafe static List<TValue> CopyList<TValue>(
     List<TValue> src) {

      object osrc = src;
      // pointer to a structure that is a template for 
      // the instance variables of List<>
      TList* psrc = (TList*)(*((int*)&psrc + 1));  
      object srcArray = null;
      object dstArray = null;
      int* i = (int*)&i;  // helps me find things on stack

      i[2] = (int)psrc->_01_items;
      int capacity = (srcArray as Array).Length;
      List<TValue> dst = new List<TValue>(capacity);
      TList* pdst = (TList*)(*((int*)&pdst + 1));
      i[1] = (int)pdst->_01_items;
      Array.Copy(srcArray as Array, dstArray as Array, capacity);

      pdst->_03_size = psrc->_03_size;

      return dst;
    }

    public unsafe static Dictionary<TKey, TValue> CopyDict<TKey, TValue>(
     Dictionary<TKey, TValue> src) {

      object osrc = src;
      // pointer to a structure that is a template for the instance 
      // variables of Dictionary<TKey, TValue>
      TDictionary* psrc = (TDictionary*)(*((int*)&psrc + 1)); 
      object srcArray = null;
      object dstArray = null;
      int* i = (int*)&i;  // helps me find the stack

      i[2] = (int)psrc->_01_buckets;
      int capacity = (srcArray as Array).Length;
      Dictionary<TKey, TValue> dst = new Dictionary<TKey, TValue>(capacity);
      TDictionary* pdst = (TDictionary*)(*((int*)&pdst + 1));
      i[1] = (int)pdst->_01_buckets;
      Array.Copy(srcArray as Array, dstArray as Array, capacity);

      i[2] = (int)psrc->_02_entries;
      i[1] = (int)pdst->_02_entries;
      Array.Copy(srcArray as Array, dstArray as Array, capacity);

      pdst->_03_comparer = psrc->_03_comparer;
      pdst->_04_m_siInfo = psrc->_04_m_siInfo;
      pdst->_08_count = psrc->_08_count;
      pdst->_10_freeList = psrc->_10_freeList;
      pdst->_11_freeCount = psrc->_11_freeCount;

      return dst;
    }

    // these are the structs that map to the private variables in the classes
    // i use uint for classes, since they are just pointers
    // statics and constants are not in the instance data.
    // I used the memory dump of visual studio to get these mapped right.
    // everything with a * I copy.  I Used RedGate reflector to look through all
    // the code to decide what needed to be copied.
    struct TKeyedCollection {
      public uint _00_MethodInfo;                  // pointer to cool type info
      // Collection
      public uint _01_items;                       // * IList<T>
      public uint _02_syncRoot;                    //   object
      // KeyedCollection
      public uint _03_comparer;                    //   IEqualityComparer<TKey> 
      public uint _04_dict;                        // * Dictionary<TKey, TItem> 
      public int _05_keyCount;                     // *
      public int _06_threshold;                    // *
      // const int defaultThreshold = 0;
    }

    struct TList {
      public uint _00_MethodInfo;                   //
      public uint _01_items;                        // * T[] 
      public uint _02_syncRoot;                     //   object
      public int _03_size;                          // *
      public int _04_version;                       //
    }

    struct TDictionary {
      // Fields
      public uint _00_MethodInfo;                   //
      public uint _01_buckets;                     // * int[] 
      public uint _02_entries;                     // * Entry<TKey, TValue>[] 
      public uint _03_comparer;                    //   IEqualityComparer<TKey> 
      public uint _04_m_siInfo;                    //   SerializationInfo
      public uint _05__syncRoot;                   //   object 
      public uint _06_keys;                        //   KeyCollection<TKey, TValue> 
      public uint _07_values;                      //   ValueCollection<TKey, TValue> 
      public int _08_count;                        // *
      public int _09_version;
      public int _10_freeList;                     // * 
      public int _11_freeCount;                    // *
    }

  }


}

Answer 2

我刚刚运行了一个测试，将10,000,000个条目和添加到不同的集合中，KeyedCollection花费的时间大约是列表的7倍，但仅比Dictionary对象长约50%。考虑到KeyedCollection是这两者的组合，Add的性能是完全合理的，并且它运行的重复键检查显然不会占用和太多时间。您可能想要在您的KeyedCollection上运行类似的测试，如果速度明显变慢，您可以开始寻找其他地方(检查您的MyObject.Key获取方法，以确保您不会因此而获得开销)。

旧的回应

您是否尝试过：

this.kc = that.kc.MemberwiseClone() as MyKeyedCollection;

有关MemberwiseClone here的更多信息。

Answer 3

您可以尝试序列化对象，然后将其反序列化为新对象-我不知道这是否会获得任何性能，但它可能会。