在.NET中,GetHashCode方法在整个.NET基类库的许多地方都使用。正确执行它对于在集合中或确定相等时快速查找项目尤为重要。
对于如何为自定义类实现GetHashCode,是否有标准算法或最佳实践,以便不会降低性能?
在.NET中,GetHashCode方法在整个.NET基类库的许多地方都使用。正确执行它对于在集合中或确定相等时快速查找项目尤为重要。
对于如何为自定义类实现GetHashCode,是否有标准算法或最佳实践,以便不会降低性能?
当前回答
直到最近,我的回答都很接近乔恩·斯基特的回答。然而,我最近开始了一个使用两个哈希表的幂的项目,即内部表大小为8、16、32等的哈希表。
而且非常糟糕。因此,经过一点实验和研究后,我开始用以下方法重新散列我的散列:
public static int ReHash(int source)
{
unchecked
{
ulong c = 0xDEADBEEFDEADBEEF + (ulong)source;
ulong d = 0xE2ADBEEFDEADBEEF ^ c;
ulong a = d += c = c << 15 | c >> -15;
ulong b = a += d = d << 52 | d >> -52;
c ^= b += a = a << 26 | a >> -26;
d ^= c += b = b << 51 | b >> -51;
a ^= d += c = c << 28 | c >> -28;
b ^= a += d = d << 9 | d >> -9;
c ^= b += a = a << 47 | a >> -47;
d ^= c += b << 54 | b >> -54;
a ^= d += c << 32 | c >> 32;
a += d << 25 | d >> -25;
return (int)(a >> 1);
}
}
然后我的两个哈希表的能力就不再糟糕了。
但这让我很不安,因为上面的方法不应该奏效。或者更准确地说,除非原始的GetHashCode()以非常特殊的方式很差,否则它不应该工作。
重新混合哈希代码并不能改善一个好的哈希代码,因为唯一可能的效果是我们引入了更多的冲突。
重新混合哈希代码并不能改善糟糕的哈希代码,因为唯一可能的效果是我们将值53上的大量冲突更改为值183487291的大量冲突。
重新混合哈希代码只能改进哈希代码,该哈希代码至少在避免整个范围内的绝对冲突(232个可能值)方面做得相当好,但在为哈希表中的实际使用而进行模化时,在避免冲突方面做得很差。虽然二次幂表的简单模使这一点更加明显,但它对更常见的素数表也有负面影响,这并不是那么明显(重新散列的额外工作将超过好处,但好处仍然存在)。
编辑:我还使用了开放寻址,这也会增加对冲突的敏感度,也许比二的幂更敏感。
好吧,这令人不安的是,.NET(或这里的研究)中的string.GetHashCode()实现可以通过这种方式改进多少(由于较少的冲突,测试运行速度大约快20-30倍),更令人不安我自己的哈希代码可以改进多少(远远不止于此)。
我过去编写的所有GetHashCode()实现,实际上都是这个网站上答案的基础,比我想象的要糟糕得多。很多时候,它对于很多用途来说“足够好”,但我想要更好的东西。
所以我把这个项目放在一边(反正它是一个宠物项目),开始研究如何在.NET中快速生成一个好的、分布良好的哈希代码。
最后,我决定将SpookyHash移植到.NET。实际上,上面的代码是使用SpookyHash从32位输入生成32位输出的快速路径版本。
现在,SpookyHash不是一个好的快速记忆代码。我的端口就更少了,因为我手动内联了很多端口以提高速度*。但这就是代码重用的目的。
然后我把这个项目放在一边,因为正如最初的项目产生了如何产生更好的哈希代码的问题,所以这个项目产生了怎样产生更好的.NET memcpy的问题。
然后我回来了,并生成了大量重载,以便将几乎所有的原生类型(十进制†除外)轻松地输入到哈希代码中。
它速度很快,鲍勃·詹金斯(Bob Jenkins)值得称赞,因为我移植的原始代码速度更快,尤其是在64位机器上,算法经过了优化。
完整的代码可以在https://bitbucket.org/JonHanna/spookilysharp/src但是考虑到上面的代码是它的简化版本。
然而,由于它现在已经写好了,因此可以更容易地使用它:
public override int GetHashCode()
{
var hash = new SpookyHash();
hash.Update(field1);
hash.Update(field2);
hash.Update(field3);
return hash.Final().GetHashCode();
}
它还需要种子值,因此,如果您需要处理不受信任的输入,并希望防止哈希DoS攻击,您可以根据正常运行时间或类似情况设置种子,并使攻击者无法预测结果:
private static long hashSeed0 = Environment.TickCount;
private static long hashSeed1 = DateTime.Now.Ticks;
public override int GetHashCode()
{
//produce different hashes ever time this application is restarted
//but remain consistent in each run, so attackers have a harder time
//DoSing the hash tables.
var hash = new SpookyHash(hashSeed0, hashSeed1);
hash.Update(field1);
hash.Update(field2);
hash.Update(field3);
return hash.Final().GetHashCode();
}
*这方面的一个大惊喜是,手动内联返回(x<<n)|(x>>-n)的旋转方法改进了性能。我本可以确定抖动会为我内联,但评测显示的情况并非如此。
†十进制虽然来自C#,但从.NET角度看不是本机。它的问题是,它自己的GetHashCode()将精度视为重要,而它自己的Equals()则没有。两者都是有效的选择,但不是那样混合。在实现自己的版本时,您需要选择执行一个或另一个,但我不知道您想要哪个。
‡通过比较。如果在字符串上使用,64位的SpookyHash要比32位的string.GetHashCode()快得多,这比64位的string.GetHashCode()要快得多。
其他回答
在Equals()比较多个字段的大多数情况下,GetHash()对一个字段或多个字段进行散列并不重要。您只需确保计算哈希值非常便宜(请不要分配)和快速(没有繁重的计算,当然也没有数据库连接),并提供良好的分布。
重型起吊应是Equals()方法的一部分;哈希应该是一个非常便宜的操作,以便能够对尽可能少的项目调用Equal()。
最后一个提示:不要依赖GetHashCode()在多个应用程序运行中保持稳定。许多.Net类型不能保证它们的哈希代码在重新启动后保持不变,因此只能对内存中的数据结构使用GetHashCode()的值。
这是我的简单方法。我使用的是经典的生成器模式。它是类型安全的(无装箱/拆箱),并且与.NET 2.0兼容(无扩展方法等)。
它的用法如下:
public override int GetHashCode()
{
HashBuilder b = new HashBuilder();
b.AddItems(this.member1, this.member2, this.member3);
return b.Result;
}
这里是实际的生成器类:
internal class HashBuilder
{
private const int Prime1 = 17;
private const int Prime2 = 23;
private int result = Prime1;
public HashBuilder()
{
}
public HashBuilder(int startHash)
{
this.result = startHash;
}
public int Result
{
get
{
return this.result;
}
}
public void AddItem<T>(T item)
{
unchecked
{
this.result = this.result * Prime2 + item.GetHashCode();
}
}
public void AddItems<T1, T2>(T1 item1, T2 item2)
{
this.AddItem(item1);
this.AddItem(item2);
}
public void AddItems<T1, T2, T3>(T1 item1, T2 item2, T3 item3)
{
this.AddItem(item1);
this.AddItem(item2);
this.AddItem(item3);
}
public void AddItems<T1, T2, T3, T4>(T1 item1, T2 item2, T3 item3,
T4 item4)
{
this.AddItem(item1);
this.AddItem(item2);
this.AddItem(item3);
this.AddItem(item4);
}
public void AddItems<T1, T2, T3, T4, T5>(T1 item1, T2 item2, T3 item3,
T4 item4, T5 item5)
{
this.AddItem(item1);
this.AddItem(item2);
this.AddItem(item3);
this.AddItem(item4);
this.AddItem(item5);
}
public void AddItems<T>(params T[] items)
{
foreach (T item in items)
{
this.AddItem(item);
}
}
}
这是一个实现Josh Bloch实现的静态助手类;并且提供了显式重载来“防止”装箱,并且还专门为长原语实现哈希。
您可以传递与equals实现匹配的字符串比较。
因为Hash输出始终是int,所以您可以只链接Hash调用。
using System;
using System.Collections;
using System.Collections.Generic;
using System.Reflection;
using System.Runtime.CompilerServices;
namespace Sc.Util.System
{
/// <summary>
/// Static methods that allow easy implementation of hashCode. Example usage:
/// <code>
/// public override int GetHashCode()
/// => HashCodeHelper.Seed
/// .Hash(primitiveField)
/// .Hsh(objectField)
/// .Hash(iEnumerableField);
/// </code>
/// </summary>
public static class HashCodeHelper
{
/// <summary>
/// An initial value for a hashCode, to which is added contributions from fields.
/// Using a non-zero value decreases collisions of hashCode values.
/// </summary>
public const int Seed = 23;
private const int oddPrimeNumber = 37;
/// <summary>
/// Rotates the seed against a prime number.
/// </summary>
/// <param name="aSeed">The hash's first term.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static int rotateFirstTerm(int aSeed)
{
unchecked {
return HashCodeHelper.oddPrimeNumber * aSeed;
}
}
/// <summary>
/// Contributes a boolean to the developing HashCode seed.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="aBoolean">The value to contribute.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int Hash(this int aSeed, bool aBoolean)
{
unchecked {
return HashCodeHelper.rotateFirstTerm(aSeed)
+ (aBoolean
? 1
: 0);
}
}
/// <summary>
/// Contributes a char to the developing HashCode seed.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="aChar">The value to contribute.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int Hash(this int aSeed, char aChar)
{
unchecked {
return HashCodeHelper.rotateFirstTerm(aSeed)
+ aChar;
}
}
/// <summary>
/// Contributes an int to the developing HashCode seed.
/// Note that byte and short are handled by this method, through implicit conversion.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="aInt">The value to contribute.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int Hash(this int aSeed, int aInt)
{
unchecked {
return HashCodeHelper.rotateFirstTerm(aSeed)
+ aInt;
}
}
/// <summary>
/// Contributes a long to the developing HashCode seed.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="aLong">The value to contribute.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int Hash(this int aSeed, long aLong)
{
unchecked {
return HashCodeHelper.rotateFirstTerm(aSeed)
+ (int)(aLong ^ (aLong >> 32));
}
}
/// <summary>
/// Contributes a float to the developing HashCode seed.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="aFloat">The value to contribute.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int Hash(this int aSeed, float aFloat)
{
unchecked {
return HashCodeHelper.rotateFirstTerm(aSeed)
+ Convert.ToInt32(aFloat);
}
}
/// <summary>
/// Contributes a double to the developing HashCode seed.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="aDouble">The value to contribute.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int Hash(this int aSeed, double aDouble)
=> aSeed.Hash(Convert.ToInt64(aDouble));
/// <summary>
/// Contributes a string to the developing HashCode seed.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="aString">The value to contribute.</param>
/// <param name="stringComparison">Optional comparison that creates the hash.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int Hash(
this int aSeed,
string aString,
StringComparison stringComparison = StringComparison.Ordinal)
{
if (aString == null)
return aSeed.Hash(0);
switch (stringComparison) {
case StringComparison.CurrentCulture :
return StringComparer.CurrentCulture.GetHashCode(aString);
case StringComparison.CurrentCultureIgnoreCase :
return StringComparer.CurrentCultureIgnoreCase.GetHashCode(aString);
case StringComparison.InvariantCulture :
return StringComparer.InvariantCulture.GetHashCode(aString);
case StringComparison.InvariantCultureIgnoreCase :
return StringComparer.InvariantCultureIgnoreCase.GetHashCode(aString);
case StringComparison.OrdinalIgnoreCase :
return StringComparer.OrdinalIgnoreCase.GetHashCode(aString);
default :
return StringComparer.Ordinal.GetHashCode(aString);
}
}
/// <summary>
/// Contributes a possibly-null array to the developing HashCode seed.
/// Each element may be a primitive, a reference, or a possibly-null array.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="aArray">CAN be null.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int Hash(this int aSeed, IEnumerable aArray)
{
if (aArray == null)
return aSeed.Hash(0);
int countPlusOne = 1; // So it differs from null
foreach (object item in aArray) {
++countPlusOne;
if (item is IEnumerable arrayItem) {
if (!object.ReferenceEquals(aArray, arrayItem))
aSeed = aSeed.Hash(arrayItem); // recursive call!
} else
aSeed = aSeed.Hash(item);
}
return aSeed.Hash(countPlusOne);
}
/// <summary>
/// Contributes a possibly-null array to the developing HashCode seed.
/// You must provide the hash function for each element.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="aArray">CAN be null.</param>
/// <param name="hashElement">Required: yields the hash for each element
/// in <paramref name="aArray"/>.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int Hash<T>(this int aSeed, IEnumerable<T> aArray, Func<T, int> hashElement)
{
if (aArray == null)
return aSeed.Hash(0);
int countPlusOne = 1; // So it differs from null
foreach (T item in aArray) {
++countPlusOne;
aSeed = aSeed.Hash(hashElement(item));
}
return aSeed.Hash(countPlusOne);
}
/// <summary>
/// Contributes a possibly-null object to the developing HashCode seed.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="aObject">CAN be null.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int Hash(this int aSeed, object aObject)
{
switch (aObject) {
case null :
return aSeed.Hash(0);
case bool b :
return aSeed.Hash(b);
case char c :
return aSeed.Hash(c);
case int i :
return aSeed.Hash(i);
case long l :
return aSeed.Hash(l);
case float f :
return aSeed.Hash(f);
case double d :
return aSeed.Hash(d);
case string s :
return aSeed.Hash(s);
case IEnumerable iEnumerable :
return aSeed.Hash(iEnumerable);
}
return aSeed.Hash(aObject.GetHashCode());
}
/// <summary>
/// This utility method uses reflection to iterate all specified properties that are readable
/// on the given object, excluding any property names given in the params arguments, and
/// generates a hashcode.
/// </summary>
/// <param name="aSeed">The developing hash code, or the seed: if you have no seed, use
/// the <see cref="Seed"/>.</param>
/// <param name="aObject">CAN be null.</param>
/// <param name="propertySelector"><see cref="BindingFlags"/> to select the properties to hash.</param>
/// <param name="ignorePropertyNames">Optional.</param>
/// <returns>A hash from the properties contributed to <c>aSeed</c>.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int HashAllProperties(
this int aSeed,
object aObject,
BindingFlags propertySelector
= BindingFlags.Instance
| BindingFlags.Public
| BindingFlags.GetProperty,
params string[] ignorePropertyNames)
{
if (aObject == null)
return aSeed.Hash(0);
if ((ignorePropertyNames != null)
&& (ignorePropertyNames.Length != 0)) {
foreach (PropertyInfo propertyInfo in aObject.GetType()
.GetProperties(propertySelector)) {
if (!propertyInfo.CanRead
|| (Array.IndexOf(ignorePropertyNames, propertyInfo.Name) >= 0))
continue;
aSeed = aSeed.Hash(propertyInfo.GetValue(aObject));
}
} else {
foreach (PropertyInfo propertyInfo in aObject.GetType()
.GetProperties(propertySelector)) {
if (propertyInfo.CanRead)
aSeed = aSeed.Hash(propertyInfo.GetValue(aObject));
}
}
return aSeed;
}
/// <summary>
/// NOTICE: this method is provided to contribute a <see cref="KeyValuePair{TKey,TValue}"/> to
/// the developing HashCode seed; by hashing the key and the value independently. HOWEVER,
/// this method has a different name since it will not be automatically invoked by
/// <see cref="Hash(int,object)"/>, <see cref="Hash(int,IEnumerable)"/>,
/// or <see cref="HashAllProperties"/> --- you MUST NOT mix this method with those unless
/// you are sure that no KeyValuePair instances will be passed to those methods; or otherwise
/// the generated hash code will not be consistent. This method itself ALSO will not invoke
/// this method on the Key or Value here if that itself is a KeyValuePair.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="keyValuePair">The value to contribute.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int HashKeyAndValue<TKey, TValue>(this int aSeed, KeyValuePair<TKey, TValue> keyValuePair)
=> aSeed.Hash(keyValuePair.Key)
.Hash(keyValuePair.Value);
/// <summary>
/// NOTICE: this method is provided to contribute a collection of <see cref="KeyValuePair{TKey,TValue}"/>
/// to the developing HashCode seed; by hashing the key and the value independently. HOWEVER,
/// this method has a different name since it will not be automatically invoked by
/// <see cref="Hash(int,object)"/>, <see cref="Hash(int,IEnumerable)"/>,
/// or <see cref="HashAllProperties"/> --- you MUST NOT mix this method with those unless
/// you are sure that no KeyValuePair instances will be passed to those methods; or otherwise
/// the generated hash code will not be consistent. This method itself ALSO will not invoke
/// this method on a Key or Value here if that itself is a KeyValuePair or an Enumerable of
/// KeyValuePair.
/// </summary>
/// <param name="aSeed">The developing HashCode value or seed.</param>
/// <param name="keyValuePairs">The values to contribute.</param>
/// <returns>The new hash code.</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int HashKeysAndValues<TKey, TValue>(
this int aSeed,
IEnumerable<KeyValuePair<TKey, TValue>> keyValuePairs)
{
if (keyValuePairs == null)
return aSeed.Hash(null);
foreach (KeyValuePair<TKey, TValue> keyValuePair in keyValuePairs) {
aSeed = aSeed.HashKeyAndValue(keyValuePair);
}
return aSeed;
}
}
}
这是Jon Skeet发布的上述算法的另一个流畅实现,但不包括分配或装箱操作:
public static class Hash
{
public const int Base = 17;
public static int HashObject(this int hash, object obj)
{
unchecked { return hash * 23 + (obj == null ? 0 : obj.GetHashCode()); }
}
public static int HashValue<T>(this int hash, T value)
where T : struct
{
unchecked { return hash * 23 + value.GetHashCode(); }
}
}
用法:
public class MyType<T>
{
public string Name { get; set; }
public string Description { get; set; }
public int Value { get; set; }
public IEnumerable<T> Children { get; set; }
public override int GetHashCode()
{
return Hash.Base
.HashObject(this.Name)
.HashObject(this.Description)
.HashValue(this.Value)
.HashObject(this.Children);
}
}
由于泛型类型约束,编译器将确保不使用类调用HashValue。但是没有编译器支持HashObject,因为添加泛型参数也会添加装箱操作。
我通常会使用Josh Bloch精彩的有效Java中给出的实现。它很快,创建了一个很好的哈希,不太可能导致冲突。选择两个不同的素数,例如17和23,并执行以下操作:
public override int GetHashCode()
{
unchecked // Overflow is fine, just wrap
{
int hash = 17;
// Suitable nullity checks etc, of course :)
hash = hash * 23 + field1.GetHashCode();
hash = hash * 23 + field2.GetHashCode();
hash = hash * 23 + field3.GetHashCode();
return hash;
}
}
正如评论中所指出的,你可能会发现最好选择一个大素数来乘。很显然486187639很好。。。虽然我见过的大多数小数字的例子都倾向于使用素数,但至少有一些类似的算法经常使用非素数。例如,在后面的FNV示例中,我使用的数字显然很好,但初始值不是质数。(不过乘法常数是质数。我不知道这有多重要。)
这比XORing散列码的常见做法要好,主要原因有两个。假设我们有一个具有两个int字段的类型:
XorHash(x, x) == XorHash(y, y) == 0 for all x, y
XorHash(x, y) == XorHash(y, x) for all x, y
顺便说一下,早期的算法是C#编译器当前用于匿名类型的算法。
这个页面提供了很多选项。我认为,在大多数情况下,上述内容“足够好”,而且非常容易记住并正确理解。FNV替代方案同样简单,但使用不同的常数和XOR代替ADD作为组合操作。它看起来像下面的代码,但正常的FNV算法对单个字节进行操作,因此这需要进行修改,以每个字节执行一次迭代,而不是每个32位哈希值。FNV也设计用于可变长度的数据,而我们在这里使用它的方式总是用于相同数量的字段值。对这个答案的评论表明,这里的代码实际上并不像上面的添加方法那样有效(在测试的示例案例中)。
// Note: Not quite FNV!
public override int GetHashCode()
{
unchecked // Overflow is fine, just wrap
{
int hash = (int) 2166136261;
// Suitable nullity checks etc, of course :)
hash = (hash * 16777619) ^ field1.GetHashCode();
hash = (hash * 16777619) ^ field2.GetHashCode();
hash = (hash * 16777619) ^ field3.GetHashCode();
return hash;
}
}
请注意,需要注意的一点是,理想情况下,您应该防止在将其添加到依赖于哈希代码的集合后,对等式敏感(因此对哈希代码敏感)的状态发生变化。
根据文件:
可以为不可变引用类型重写GetHashCode。通常,对于可变引用类型,只有在以下情况下才应重写GetHashCode:您可以从不可变的字段计算哈希代码;或当可变对象包含在依赖其哈希代码的集合中时,可以确保该对象的哈希代码不会更改。
FNV文章的链接已断开,但这是互联网档案馆的一份副本:永恒的困惑-哈希的艺术