雁过请留痕...
代码改变世界

不可小瞧的GetHashCode函数

2013-03-04 17:27  xiashengwang  阅读(10194)  评论(1)  编辑  收藏  举报

要实现对象的相等比较，需要实现IEquatable<T>，或单独写一个类实现IEqualityComparer<T>接口。

像List<T>的Contains这样的函数，如果我们自己定义的对象不实现IEquatable<T>接口，这个函数会默认调用object的Equels来比较对象，得出非预期的结果。

先自定义一个类：

        public class DaichoKey 
        {
            public int ID { get; set; }
            public int SubID { get; set; }
        }


            List<DaichoKey> lst = new List<DaichoKey>() { 
            new DaichoKey(){ID = 1,SubID =2},
            new DaichoKey(){ID = 1,SubID = 3}
            };            
            var newItem = new DaichoKey() { ID = 1, SubID = 2 };
            bool isContains = lst.Contains(newItem);//false

 上面的代码调用Contains后得到false，我们预想1和2的对象都已经存在了，应该得到true才对呀。

要实现这个效果，需要实现IEquatable<T>接口。

        public class DaichoKey : IEquatable<DaichoKey>
        {
            public int ID { get; set; }
            public int SubID { get; set; }

            public bool Equals(DaichoKey other)
            {
                return this.ID == other.ID && this.SubID == other.SubID;
            }
        }

经过上面的改良，结果如我们预期了，但是还不够完善，微软建议我们重写object的Equels方法我GetHashCode方法，以保持语义的一致性，于是有了下面的代码：

        public class DaichoKey : IEquatable<DaichoKey>
        {
            public int ID { get; set; }
            public int SubID { get; set; }

            public bool Equals(DaichoKey other)
            {
                return this.ID == other.ID && this.SubID == other.SubID;
            }
            public override bool Equals(object obj)
            {
                if (obj == null) return base.Equals(obj);

                if (obj is DaichoKey)
                    return Equals(obj as DaichoKey);
                else
                    throw new InvalidCastException("the 'obj' Argument is not a DaichoKey object");
            }
            public override int GetHashCode()
            {
                return base.GetHashCode();//return object's hashcode
            }
        }

 上面的代码依然还有缺陷，没重写==和!=运算符，但这不是本文讨论的重点。绕了一大圈，终于来到了GetHashCode函数身上，貌似他对我们的Contains函数没有啥影响呀，不重写又何妨？我们再来试试List<T>的一个扩展函数Distinct： 

            List<DaichoKey> lst = new List<DaichoKey>() { 
            new DaichoKey(){ID = 1,SubID =2},
            new DaichoKey(){ID = 1,SubID = 3}
            };
            var newItem = new DaichoKey() { ID = 1, SubID = 2 };
            lst.Add(newItem);
            if (lst != null)
            {
                lst = lst.Distinct<DaichoKey>().ToList();
            }
            //result:
            //1 2
            //1 3
            //1 2

 悲剧发生了，数据1，2的重复数据没有被去掉呀，我们不是实现了IEquatable<T>接口接口吗。在园子上找到了一篇文章（c# 扩展方法奇思妙用基础篇八：Distinct 扩展），在回复中提到要将GetHashCode返回固定值，以强制调用IEquatable<T>的Equels方法。如下：

        public class DaichoKey : IEquatable<DaichoKey>
        {
            public int ID { get; set; }
            public int SubID { get; set; }

            public bool Equals(DaichoKey other)
            {
                return this.ID == other.ID && this.SubID == other.SubID;
            }
            public override bool Equals(object obj)
            {
                if (obj == null) return base.Equals(obj);

                if (obj is DaichoKey)
                    return Equals(obj as DaichoKey);
                else
                    throw new InvalidCastException("the 'obj' Argument is not a DaichoKey object");
            }
            public override int GetHashCode()
            {
                return 0;//base.GetHashCode();
            }
        }

 结果立马就对了，难道是这个Distinct函数在比较时，先比较的HashCode值？

带着这个疑问，反编译了下Distinct的代码，确实如我所猜测的那样。下面是源代码，有兴趣的同学，可以往下看看：

public static IEnumerable<TSource> Distinct<TSource>(this IEnumerable<TSource> source)
{
    if (source == null) throw Error.ArgumentNull("source");
    return DistinctIterator<TSource>(source, null);
}

 private static IEnumerable<TSource> DistinctIterator<TSource>(IEnumerable<TSource> source, IEqualityComparer<TSource> comparer)
{
    <DistinctIterator>d__81<TSource> d__ = new <DistinctIterator>d__81<TSource>(-2);
    d__.<>3__source = source;
    d__.<>3__comparer = comparer;
    return d__;
}

 private sealed class <DistinctIterator>d__81<TSource> : IEnumerable<TSource>, IEnumerable, IEnumerator<TSource>, IEnumerator, IDisposable
{
    // Fields
    private int <>1__state;
    private TSource <>2__current;
    public IEqualityComparer<TSource> <>3__comparer;
    public IEnumerable<TSource> <>3__source;
    public IEnumerator<TSource> <>7__wrap84;
    private int <>l__initialThreadId;
    public TSource <element>5__83;
    public Set<TSource> <set>5__82;
    public IEqualityComparer<TSource> comparer;
    public IEnumerable<TSource> source;

    // Methods
    [DebuggerHidden]
    public <DistinctIterator>d__81(int <>1__state);
    private void <>m__Finally85();
    private bool MoveNext();
    [DebuggerHidden]
    IEnumerator<TSource> IEnumerable<TSource>.GetEnumerator();
    [DebuggerHidden, TargetedPatchingOptOut("Performance critical to inline this type of method across NGen image boundaries")]
    IEnumerator IEnumerable.GetEnumerator();
    [DebuggerHidden]
    void IEnumerator.Reset();
    void IDisposable.Dispose();

    // Properties
    TSource IEnumerator<TSource>.Current { [DebuggerHidden] get; }
    object IEnumerator.Current { [DebuggerHidden] get; }
}

private sealed class <DistinctIterator>d__81<TSource> : IEnumerable<TSource>, IEnumerable, IEnumerator<TSource>, IEnumerator, IDisposable
{
    // Fields
    private int <>1__state;
    private TSource <>2__current;
    public IEqualityComparer<TSource> <>3__comparer;
    public IEnumerable<TSource> <>3__source;
    public IEnumerator<TSource> <>7__wrap84;
    private int <>l__initialThreadId;
    public TSource <element>5__83;
    public Set<TSource> <set>5__82;
    public IEqualityComparer<TSource> comparer;
    public IEnumerable<TSource> source;

    // Methods
    [DebuggerHidden]
    public <DistinctIterator>d__81(int <>1__state);
    private void <>m__Finally85();
    private bool MoveNext();
    [DebuggerHidden]
    IEnumerator<TSource> IEnumerable<TSource>.GetEnumerator();
    [DebuggerHidden, TargetedPatchingOptOut("Performance critical to inline this type of method across NGen image boundaries")]
    IEnumerator IEnumerable.GetEnumerator();
    [DebuggerHidden]
    void IEnumerator.Reset();
    void IDisposable.Dispose();

    // Properties
    TSource IEnumerator<TSource>.Current { [DebuggerHidden] get; }
    object IEnumerator.Current { [DebuggerHidden] get; }
}

private bool MoveNext()
{
    bool flag;
    try
    {
        switch (this.<>1__state)
        {
            case 0:
                this.<>1__state = -1;
                this.<set>5__82 = new Set<TSource>(this.comparer);
                this.<>7__wrap84 = this.source.GetEnumerator();
                this.<>1__state = 1;
                goto Label_0092;

            case 2:
                this.<>1__state = 1;
                goto Label_0092;

            default:
                goto Label_00A5;
        }
    Label_0050:
        this.<element>5__83 = this.<>7__wrap84.Current;
        if (this.<set>5__82.Add(this.<element>5__83))
        {
            this.<>2__current = this.<element>5__83;
            this.<>1__state = 2;
            return true;
        }
    Label_0092:
        if (this.<>7__wrap84.MoveNext()) goto Label_0050;
        this.<>m__Finally85();
    Label_00A5:
        flag = false;
    }
    fault
    {
        this.System.IDisposable.Dispose();
    }
    return flag;
}

internal class Set<TElement>
{
    // Fields
    private int[] buckets;
    private IEqualityComparer<TElement> comparer;
    private int count;
    private int freeList;
    private Slot<TElement>[] slots;

    // Methods
    [TargetedPatchingOptOut("Performance critical to inline this type of method across NGen image boundaries")]
    public Set();
    public Set(IEqualityComparer<TElement> comparer);
    public bool Add(TElement value);
    [TargetedPatchingOptOut("Performance critical to inline this type of method across NGen image boundaries")]
    public bool Contains(TElement value);
    private bool Find(TElement value, bool add);
    internal int InternalGetHashCode(TElement value);
    public bool Remove(TElement value);
    private void Resize();

    // Nested Types
    [StructLayout(LayoutKind.Sequential)]
    internal struct Slot
    {
        internal int hashCode;
        internal TElement value;
        internal int next;
    }
}
public bool Add(TElement value)
{
    return !this.Find(value, true);
}
 
public bool Contains(TElement value)
{
    return this.Find(value, false);
}

private bool Find(TElement value, bool add)
{
    int hashCode = this.InternalGetHashCode(value);
    for (int i = this.buckets[hashCode % this.buckets.Length] - 1; i >= 0; i = this.slots[i].next)
    {
        if (this.slots[i].hashCode == hashCode && this.comparer.Equals(this.slots[i].value, value)) return true;//就是这一句了
    }
    if (add)
    {
        int freeList;
        if (this.freeList >= 0)
        {
            freeList = this.freeList;
            this.freeList = this.slots[freeList].next;
        }
        else
        {
            if (this.count == this.slots.Length) this.Resize();
            freeList = this.count;
            this.count++;
        }
        int index = hashCode % this.buckets.Length;
        this.slots[freeList].hashCode = hashCode;
        this.slots[freeList].value = value;
        this.slots[freeList].next = this.buckets[index] - 1;
        this.buckets[index] = freeList + 1;
    }
    return false;
}

 在这段代码中可以看出，扩展函数Distinct在内部使用了一个Set<T>的类来帮助踢掉重复数据，而这个内部类使用的是hash表的方式存储数据，所以会调用到我们自定义类的GetHashCode函数，如果返回的hashcode值不等，它就不会再调用Equels方法进行比较了。

原因已经一目了然了，得出的结论就是：

1，重写Equles方法的时候，尽量重写GetHashCode函数，并且不要简单的调用object的GetHashCode函数，返回一个设计合理的hash值，以保证结果如我们的预期。上面的做法直接返回了0，虽然解决了问题，但明显不是每个对象的hash值都是0，做法欠妥。

2，List<T>的Contains，IndexOf方法，不会用到GetHashCode函数。

3，扩展函数Distinct，Except用到了GetHashCode函数，必须重写这个函数。其他还有哪些函数用到了GetHashCode函数，以后再做补充，使用时多加注意就是了。

4，如果对象要作为字典类（Dictionary）的主键，必须重写GetHashCode函数。

2014/07/08 补充

5，HashSet等容器的Add方法内部，也是先判断GetHashCode，如果GetHashCode值相等，进一步判断Equals方法是否相等来确定对象的相等性。

所以，Equals是相等的，那么GetHashCode也必须要保证相等。相反却不一定，GetHashCode相等，Equals方法可以不等。

6，改变影响GetHashCode返回值的字段值，会造成对象的HashCode值变化，如果对象已经存入了HashSet等容器中，将会是HashSet找不到这个对象，从而使得Remove等方法失败。

            Point a = new Point(1, 2);
            Point b = new Point(1, 2);

            HashSet<Point> hashSet = new HashSet<Point>();
            hashSet.Add(a);
            hashSet.Remove(b); //能删除a吗？答案是可以

//hashset的Count变为0，原因就是我们重新了Equals方法，a和
//b被认为相等的。

 

7，记录一个自定义值类型重写GetHashCode等方法的完整实现，作为参考。

 1     public struct Point
 2     {
 3         private int x;
 4         private int y;
 5         public Point(int x, int y)
 6         {
 7             this.x = x;
 8             this.y = y;
 9         }
10         public int X
11         {
12             get { return x; }
13         }
14         public int Y
15         {
16             get { return y; }
17         }
18 
19         public static bool operator ==(Point left,Point right)
20         {
21             if (object.ReferenceEquals(left, null))
22                 return object.ReferenceEquals(right, null);
23             return left.Equals(right);
24         }
25 
26         public static bool operator !=(Point left, Point right)
27         {
28             return !(left == right);
29         }
30 
31         public override bool Equals(object obj)
32         {
33             if (obj.GetType() != typeof(Point))
34                 return false;
35             Point other = (Point)obj;
36             return this.x == other.x && this.y == other.y;
37         }
38 
39         public override int GetHashCode()
40         {
41             return x.GetHashCode() ^ y.GetHashCode();
42         }
43     }
View Code