从对集合数据去重到Distinct源码分析

今天在写代码的时候要对数据进行去重,正打算使用Distinct方法的时候,发现这个用了这么久的东西,竟然不知道它是怎么实现的,于是就有了这篇文章.

使用的.net core2.0

1.需求

假如我们有这样一个类

    public class Model
    {
        public int Code { get; set; }
        public int No { get; set; }
        public override string ToString()
        {
            return "No:" + No + ",Code:" + Code;
        }
    }

还有这样一组数据

        public static IEnumerable<Model> GetList()
        {
            return new List<Model>()
            {
                new Model(){No = 1,Code = 1},
                new Model(){No = 1,Code = 2},
                new Model(){No = 7,Code = 1},
                new Model(){No = 11,Code = 1},
                new Model(){No = 55,Code = 1},
                new Model(){No = 11,Code = 1},//重复
                new Model(){No = 6,Code = 7},
                new Model(){No = 1,Code = 1},
                new Model(){No = 6,Code = 7},//重复
            };
        }

我们要把集合中重复的数据去掉,对的就这么简单个需求,工作中可不会有这么简单的需求.

2.在刚学编程的时候我们可能这样写的

在很久以前一直使用这种简单粗暴的方法解决重复问题

        /// <summary>
        /// 双重循环去重
        /// </summary>
        /// <param name="list"></param>
        /// <returns></returns>
        public static IEnumerable<Model> MyDistinct(IEnumerable<Model> list)
        {
            var result = new List<Model>();
            foreach (var item in list)
            {
                //标记
                var flag = true;
                foreach (var item2 in result)
                {
                    //已经存在的标记为false
                    if (item2.Code == item.Code && item2.No == item.No)
                    {
                        flag = false;
                    }
                }

                if (flag)
                {
                    result.Add(item);
                }
            }

            return result;
        }

3.后来认识了Distinct

后来知道了Distinct去重,我们写法变成了这样

   /// <summary>
    /// 比较器
    /// </summary>
    public class ModelEquality : IEqualityComparer<Model>
    {
        public bool Equals(Model x, Model y)
        {
            return x.No == y.No && x.Code == y.Code;
        }

        public int GetHashCode(Model obj)
        {
            return obj.No.GetHashCode() + obj.Code.GetHashCode();
        }
    }
//这样就可以得到去重后的集合
GetList().Distinct(new ModelEquality());

4.探究Distinct源码

我们去github找一下源码,微软开源的仓库地址:https://github.com/dotnet/corefx

为了篇幅我删掉了一些不相关的一些代码

namespace System.Linq
{
    public static partial class Enumerable
    {
        public static IEnumerable<TSource> Distinct<TSource>(this IEnumerable<TSource> source) => Distinct(source, null);

        public static IEnumerable<TSource> Distinct<TSource>(this IEnumerable<TSource> source, IEqualityComparer<TSource> comparer)
        {
            if (source == null)
            {
                throw Error.ArgumentNull(nameof(source));
            }

            return new DistinctIterator<TSource>(source, comparer);
        }
        private sealed class DistinctIterator<TSource> : Iterator<TSource>, IIListProvider<TSource>
        {
            private readonly IEnumerable<TSource> _source;
            private readonly IEqualityComparer<TSource> _comparer;
            private Set<TSource> _set;
            private IEnumerator<TSource> _enumerator;

            public DistinctIterator(IEnumerable<TSource> source, IEqualityComparer<TSource> comparer)
            {
                _source = source;
                _comparer = comparer;
            }

            public override bool MoveNext()
            {
                switch (_state)
                {
                    case 1:
                        _enumerator = _source.GetEnumerator();
                        if (!_enumerator.MoveNext())
                        {
                            Dispose();
                            return false;
                        }

                        TSource element = _enumerator.Current;
                        _set = new Set<TSource>(_comparer);
                        _set.Add(element);
                        _current = element;
                        _state = 2;
                        return true;
                    case 2:
                        while (_enumerator.MoveNext())
                        {
                            element = _enumerator.Current;
                            if (_set.Add(element))
                            {
                                _current = element;
                                return true;
                            }
                        }
                        break;
                }

                Dispose();
                return false;
            }

            public override void Dispose()
            {
				//省略...
            }
        }
    }
}

Iterator<TSource>是一个抽象类实现了Iterator<TSource> : IEnumerable<TSource>, IEnumerator<TSource>

我们主要看DistinctIterator类中的代码,发现有这么一个私有成员Set<TSource> _set;,我们再看MoveNext方法中有这么一段

                            element = _enumerator.Current;
                            if (_set.Add(element))
                            {
                                _current = element;
                                return true;
                            }

到这里我似乎明白了什么,回忆下Set集合的特点"无序","不可重复",再看代码中只有对set Add成功才对_current赋值,return true.那么这个Set应该就是内部维护的一个集合,也就是我们要的去重后的数据,那么Set里的Add方法就是关键

同样去掉了一些没有用到的,加了注释

namespace System.Linq
{
    /// <summary>
    /// A lightweight hash set.
    ///一个 轻量级hash set
    /// </summary>
    /// <typeparam name="TElement">The type of the set's items.</typeparam>
    internal sealed class Set<TElement>
    {
        /// <summary>
        /// The comparer used to hash and compare items in the set.
        /// </summary>
        private readonly IEqualityComparer<TElement> _comparer;

        /// <summary>
        /// The hash buckets, which are used to index into the slots.
		/// hash环,每一个指向了下面Slot中的index
        /// </summary>
        private int[] _buckets;

        /// <summary>
        /// The slots, each of which store an item and its hash code.
		/// 数组的每一个储存了他们自身和自己的hash
        /// </summary>
        private Slot[] _slots;

        /// <summary>
        /// The number of items in this set.
        /// </summary>
        private int _count;

        /// <summary>
        /// Constructs a set that compares items with the specified comparer.
        /// </summary>
        /// <param name="comparer">
        /// The comparer. If this is <c>null</c>, it defaults to <see cref="EqualityComparer{TElement}.Default"/>.
        /// </param>
        public Set(IEqualityComparer<TElement> comparer)
        {
            _comparer = comparer ?? EqualityComparer<TElement>.Default;
			//初始化长度7
            _buckets = new int[7];
			//初始化长度7
            _slots = new Slot[7];
        }

        /// <summary>
        /// Attempts to add an item to this set.
		/// 我们要看的方法
        /// </summary>
        /// <param name="value">The item to add.</param>
        /// <returns>
        /// <c>true</c> if the item was not in the set; otherwise, <c>false</c>.
        /// </returns>
        public bool Add(TElement value)
        {
			//取的当前项的hash
            int hashCode = InternalGetHashCode(value);
			//重复的hashCode的话,  _buckets[hashCode % _buckets.Length] - 1的值就不会是-1
            //就会进入下面的if判断
            //
            for (int i = _buckets[hashCode % _buckets.Length] - 1; i >= 0; i = _slots[i]._next)
            {
				//如果存在重复就会直接返回false,没有的话i会变为_next所指向的hash相等的元素,减少了循环次数,类似链表
                if (_slots[i]._hashCode == hashCode && _comparer.Equals(_slots[i]._value, value))
                {
                    return false;
                }
            }
			//Slot数量满了后
            if (_count == _slots.Length)
            {
				//对数组进行扩容
                Resize();
            }
			//元素要添加进_slots的下标位置
            int index = _count;
			//对数量进行增加
            _count++;
			//对当前项的hash 取余
            int bucket = hashCode % _buckets.Length;
			//赋值
            _slots[index]._hashCode = hashCode;
            _slots[index]._value = value;
			//当hash第一次出现的时候值为-1,重复出现的时候为上一个出现重复bucket值存放在slots中的索引,-1是因为下一行+1了
            _slots[index]._next = _buckets[bucket] - 1;
			//指向当前元素索引+1 出现重复的bucket值则会覆盖旧的bucket位置的值
            _buckets[bucket] = index + 1;
            return true;
        }
        /// <summary>
        /// Expands the capacity of this set to double the current capacity, plus one.
		/// 对set扩容
        /// </summary>
        private void Resize()
        {
            int newSize = checked((_count * 2) + 1);
            int[] newBuckets = new int[newSize];
            Slot[] newSlots = new Slot[newSize];
            Array.Copy(_slots, 0, newSlots, 0, _count);
            for (int i = 0; i < _count; i++)
            {
                int bucket = newSlots[i]._hashCode % newSize;
                newSlots[i]._next = newBuckets[bucket] - 1;
                newBuckets[bucket] = i + 1;
            }

            _buckets = newBuckets;
            _slots = newSlots;
        }

        /// <summary>
        /// The number of items in this set.
        /// </summary>
        public int Count => _count;

        /// <summary>
        /// Gets the hash code of the provided value with its sign bit zeroed out, so that modulo has a positive result.
        /// </summary>
        /// <param name="value">The value to hash.</param>
        /// <returns>The lower 31 bits of the value's hash code.</returns>
        private int InternalGetHashCode(TElement value) => value == null ? 0 : _comparer.GetHashCode(value) & 0x7FFFFFFF;

        /// <summary>
        /// An entry in the hash set.
        /// </summary>
        private struct Slot
        {
            /// <summary>
            /// The hash code of the item.
			/// hash值
            /// </summary>
            internal int _hashCode;

            /// <summary>
            /// In the case of a hash collision, the index of the next slot to probe.
			/// 下一个用于检查的元素index
            /// </summary>
            internal int _next;

            /// <summary>
            /// The item held by this slot.
            /// </summary>
            internal TElement _value;
        }
    }
}

5.分析下去重的思路

图用自带画图画的,难看还请见谅.



我后面回放代码,一步一步调试可能会更容易理解.

1.假如我们第一个Model进行hash取余得到的为0,此时_buckets[0]为0,所以不会进入for循环条件,直接进行下面的赋值操作

_slots[0]=当前的元素 next=-1 hash=7
buckets[0]=1 指向当前元素索引+1

2.继续下一个Model进行hash取余,假如又为0,buckets[0]-1为0,满足循环条件,进入判断,取到_slots[0]的值,进行比较,发现相等的话则会直接返回.

3.继续上面的步骤,这次hash取余为3,没出现过,

_slots[1]=当前的元素 next=-1 hash=10
buckets[2]=2 指向当前元素索引+1

.........

4.这个时候又出现了一次hash取余为3,进入判断中,取到_slots[1]的值,进行比较发现不相等,next为-1不会有下一次循环,

_slots[3]=当前的元素 next=1 hash=10
buckets[2]=4 指向当前元素索引+1

注意此时next不是-1了,而是1,也就是上一个相同hash取余的元素在_slots中的位置,此时形成了一个链表.这样少了很多的比较次数.

5.这个时候又出现了一个hash取余为3的,进入判断中,取到_slots[3]的值,进行比较发现不相等,next为1,则再次与_slots[1]的元素进行比较,如果发现相等的舍弃,反之最后加入到set中

假如不相同,则:

_slots[4]=当前的元素 next=3 hash=10
buckets[2]=5 指向当前元素索引+1

6.结束

结束了,我们发现Distinct使用了hash进行去重,实现思路上感觉很值得我学习(我是写不出来的..).

Distinct很依赖于比较器的GetHashCode方法,如果随便返回一个固定值的话,会增加很大的开销.不要为了偷懒再返回一个固定int值了.

希望这篇文章可以对大家有帮助 有启发

代码地址:https://git.coding.net/changyell/DistinctDemo.git

本人是个菜鸟,文章如果有错误的地方,烦请大佬们指正,谢谢...