发现自己已经有很长一段时间写代码没什么进步了,随便读读FCL的源码,看看之前一直用的方法是如何实现的,也顺便提高下自己。FCL很是庞大,很难下口,于是用最笨的办法,先看常见的命名空间,逐个展开。战五渣的水平,必定有很多理解上的错误,欢迎斧正,不胜感激。
System.Collections 命名空间中的集合包含(如列表、队列、位数组、哈希表和字典)的集合。
本篇目录:
集合的接口
在刚开始写程序的时候的时候经常会写一些接口,哪怕这个接口只被用到了一次,也要抽象一个接口出来,这样显得牛X一些。到后来,接口几乎从日常的代码中消失,能简单就简单。看了FCL的源码,发现,这些接口很有必要,而且抽象的恰到好处,不经拍案叫绝。
很难想象,如果没有这些接口,庞大的FCL将如何构建,如何约束那些类。每个集合的操作方法类似,名称各不相同,对于使用者来说,也绝对是件很苦逼的事情。接口 是一种规范,实现了某一个接口,便具备了改接口的功能。所以了解某一个集合的性质和功能,首先需要了解它实现了哪些接口。
集合中常见的接口有IEnumerable,IEnumerator,ICollection,IComparer,IDictionary,IDictionaryEnumerator,ListDictionaryInternal,IEnumerator,IHashCodeProvider,IList,IStructuralComparable等。彻底晕菜了!新建 Class Diagram ,将几个主要接口拖入后,结构便很清晰了。
所有的集合都是继承了IEnumerable,逐个分析每个接口的实现。IEnumerable的源码,其中 PureAttribute来表示自己是很纯的,协定的东西。DispId 属性被用来指定一个OLE 的自动化 DISPID,COM交互时会使用。查了下使用的地方在 ComAwareEventInfo.cs 中,请自行查阅。 IEnumerator GetEnumerator(); 使用了组合模式,正是因为这个方法,所有的集合才可以使用Foreach方法。在后面研究集合的时候会详细的看下IEnumerator的实现。
IEnumerable源码
public interface IEnumerable { // Interfaces are not serializable // Returns an IEnumerator for this enumerable Object. The enumerator provides // a simple way to access all the contents of a collection. [Pure] [DispId(-4)] IEnumerator GetEnumerator(); }
IEnumerator源码
public interface IEnumerator { bool MoveNext();//索引位置向后移 Object Current {//当前对象 get; } void Reset();//重置索引到第一个位置 }
接口ICollection 继承了IEnumerable,定义了集合基本的元素,大小(count)枚举器(继承自IEnumerable 的GetEnumerator),同步方法(使用 IsSynchronized,SyncRoot),这里涉及一个锁的问题,如果对这个集合元素锁定后不可读与写,那么锁定这个集合的本身,如果锁定这个集合,不可写,可以读,那么锁定这个集合的SyncRoot。为什么使用Synchronized 方法返回的类是线程安全的呢,来看下具体的实现方式吧,以ArrayList为例
var al= ArrayList.Synchronized(new ArrayList());
public static ArrayList Synchronized(ArrayList list) { if (list==null) throw new ArgumentNullException("list"); Contract.Ensures(Contract.Result<ArrayList>() != null); Contract.EndContractBlock(); return new SyncArrayList(list); }
具体的实现在SyncArrayList中下面是部分源码:
private class SyncArrayList : ArrayList { private ArrayList _list; private Object _root; internal SyncArrayList(ArrayList list) : base( false ) { _list = list; _root = list.SyncRoot; } public override int Capacity { get { lock(_root) { return _list.Capacity; } } [SuppressMessage("Microsoft.Contracts", "CC1055")] // Skip extra error checking to avoid *potential* AppCompat problems. set { lock(_root) { _list.Capacity = value; } } } public override int Count { get { lock(_root) { return _list.Count; } } } public override bool IsReadOnly { get { return _list.IsReadOnly; } } public override bool IsFixedSize { get { return _list.IsFixedSize; } } public override bool IsSynchronized { get { return true; } } public override Object this[int index] { get { lock(_root) { return _list[index]; } } set { lock(_root) { _list[index] = value; } } } public override Object SyncRoot { get { return _root; } } public override int Add(Object value) { lock(_root) { return _list.Add(value); } } ... } }
实现的方式也很简单,所有的数据操作全部上锁,所以就线程安全了
学习数据结构的时候,线性表有几种操作,初始化,清空,获取某一个位置元素,判断元素是否存在,插入,删除,获取长度。很多元素是可以标准化的,IList就是干这个的。
IList的具体实现:
public interface IList : ICollection { //索引器 Object this[int index] { get; set; } //插入 int Add(Object value); //判断是否包含 bool Contains(Object value); //清空 void Clear(); //判断是否为只读 只读集合在创建之后不允许添加、移除或修改元素。 bool IsReadOnly { get; } //是否是固定大小 bool IsFixedSize { get; } //取元素索引值 int IndexOf(Object value); //指定索引位置插入元素 void Insert(int index, Object value); //删除某个元素 void Remove(Object value); //删除指定索引的元素 void RemoveAt(int index); }
字典集合的抽象接口为IDictionary,字典类型的操作有,获取keys,获取values,添加,删除,清空等。实现源码如下:
public interface IDictionary : ICollection { // Interfaces are not serializable // The Item property provides methods to read and edit entries // in the Dictionary. Object this[Object key] { get; set; } // Returns a collections of the keys in this dictionary. ICollection Keys { get; } // Returns a collections of the values in this dictionary. ICollection Values { get; } // Returns whether this dictionary contains a particular key. // bool Contains(Object key); // Adds a key-value pair to the dictionary. void Add(Object key, Object value); // Removes all pairs from the dictionary. void Clear(); bool IsReadOnly { get; } bool IsFixedSize { get; } // Returns an IDictionaryEnumerator for this dictionary. new IDictionaryEnumerator GetEnumerator(); // Removes a particular key from the dictionary. // void Remove(Object key); }
ArrayList
ArrayList动态容量的实现
ArrayList 为动态数组,动态的添加和减少线性表的长度,不用担心长度不够而抛异常。首先我们来探究下,这个动态的长度是如何实现的。查看arraylist.cs文件。
public virtual int Add(Object value) { Contract.Ensures(Contract.Result<int>() >= 0); if (_size == _items.Length) EnsureCapacity(_size + 1); _items[_size] = value; _version++; return _size++; }
ArrayList添加元素方法, Contract.Ensures(Contract.Result<int>() >= 0);这句可以忽略,契约式编程,可以自己搜索。顺腾摸瓜,进入EnsureCapacity函数:
private void EnsureCapacity(int min) { if (_items.Length < min) { int newCapacity = _items.Length == 0? _defaultCapacity: _items.Length * 2;//定义一个新的容量,如果当然容量是0,就用默认的,否则就当前容量*2 // Allow the list to grow to maximum possible capacity (~2G elements) before encountering overflow. // Note that this check works even when _items.Length overflowed thanks to the (uint) cast if ((uint)newCapacity > Array.MaxArrayLength) newCapacity = Array.MaxArrayLength;//如果新的容量比Array数组的最大值还大,那么就赋最大的值 if (newCapacity < min) newCapacity = min;//如果新的容量比传入的最小值要小,那么赋最小值 Capacity = newCapacity;//容量等于新的容量 } }
学过数据结构我们都知道,线性表增加容量,肯定要移动元素的。这个地方没看到,那么在找Capacity的Set方法。如下:
public virtual int Capacity { get { Contract.Ensures(Contract.Result<int>() >= Count); return _items.Length; } set { if (value < _size) { throw new ArgumentOutOfRangeException("value", Environment.GetResourceString("ArgumentOutOfRange_SmallCapacity")); } Contract.Ensures(Capacity >= 0); Contract.EndContractBlock(); // We don't want to update the version number when we change the capacity. // Some existing applications have dependency on this. if (value != _items.Length) { if (value > 0) { Object[] newItems = new Object[value]; if (_size > 0) { Array.Copy(_items, 0, newItems, 0, _size);//ArrayList的长度发生改变时,就要来一次迁移 } _items = newItems; } else { _items = new Object[_defaultCapacity]; } } } }
private const int _defaultCapacity = 4;默认的容量是4.不管容量是变大还是变小,都要移动元素,性能肯定是会有到影响的。
ArrayList 方法底层实现探究
继续跟踪Array.Copy 来看下具体是怎么实现的,最后跟踪的代码:
[System.Security.SecurityCritical] // auto-generated [ReliabilityContract(Consistency.MayCorruptInstance, Cer.MayFail)] [ResourceExposure(ResourceScope.None)] [MethodImplAttribute(MethodImplOptions.InternalCall)] internal static extern void Copy(Array sourceArray, int sourceIndex, Array destinationArray, int destinationIndex, int length, bool reliable);
[MethodImplAttribute(MethodImplOptions.InternalCall)]由这个属性猜出来,这个是CLR内部实现的,没法看。好奇心有强怎么办,好吧,拿出sscli(.net 2.0 的clr源码),在ecall.cpp里面看到这个
FCFuncElement("Copy", SystemNative::ArrayCopy)//
仔细看,能看出个大概。经过阅读FCL源码会发现,几乎所有的集合Copy,CopyTo 方法,最终都是调用Array.Copy,Array.Copy最终调用的是下面这个CLR中的方法
CLR中的源码如下:
FCIMPL6(void, SystemNative::ArrayCopy, ArrayBase* m_pSrc, INT32 m_iSrcIndex, ArrayBase* m_pDst, INT32 m_iDstIndex, INT32 m_iLength, CLR_BOOL reliable) { BYTE *src; BYTE *dst; int size; struct _gc { BASEARRAYREF pSrc; BASEARRAYREF pDst; } gc; gc.pSrc = (BASEARRAYREF)m_pSrc; gc.pDst = (BASEARRAYREF)m_pDst; // // creating a HelperMethodFrame is quite expensive, // so we want to delay this for the most common case which doesn't trigger a GC. // FCThrow is needed to throw an exception without a HelperMethodFrame // WRAPPER_CONTRACT; STATIC_CONTRACT_SO_TOLERANT; // cannot pass null for source or destination if (gc.pSrc == NULL || gc.pDst == NULL) { FCThrowArgumentNullVoid(gc.pSrc==NULL ? L"source" : L"dest"); } // source and destination must be arrays _ASSERTE(gc.pSrc->GetMethodTable()->IsArray()); _ASSERTE(gc.pDst->GetMethodTable()->IsArray()); g_IBCLogger.LogEEClassAndMethodTableAccess(gc.pSrc->GetArrayClass()); // Equal method tables should imply equal rank _ASSERTE(!(gc.pSrc->GetMethodTable() == gc.pDst->GetMethodTable() && gc.pSrc->GetRank() != gc.pDst->GetRank())); // Which enables us to avoid touching the EEClass in simple cases if (gc.pSrc->GetMethodTable() != gc.pDst->GetMethodTable() && gc.pSrc->GetRank() != gc.pDst->GetRank()) { FCThrowResVoid(kRankException, L"Rank_MustMatch"); } // Variant is dead. _ASSERTE(gc.pSrc->GetMethodTable() != COMVariant::s_pVariantClass); _ASSERTE(gc.pDst->GetMethodTable() != COMVariant::s_pVariantClass); int srcLB = gc.pSrc->GetLowerBoundsPtr()[0]; int destLB = gc.pDst->GetLowerBoundsPtr()[0]; // array bounds checking const unsigned int srcLen = gc.pSrc->GetNumComponents(); const unsigned int destLen = gc.pDst->GetNumComponents(); if (m_iLength < 0) FCThrowArgumentOutOfRangeVoid(L"length", L"ArgumentOutOfRange_NeedNonNegNum"); if (m_iSrcIndex < srcLB || (m_iSrcIndex - srcLB < 0)) FCThrowArgumentOutOfRangeVoid(L"srcIndex", L"ArgumentOutOfRange_ArrayLB"); if (m_iDstIndex < destLB || (m_iDstIndex - destLB < 0)) FCThrowArgumentOutOfRangeVoid(L"dstIndex", L"ArgumentOutOfRange_ArrayLB"); if ((DWORD)(m_iSrcIndex - srcLB + m_iLength) > srcLen) FCThrowResVoid(kArgumentException, L"Arg_LongerThanSrcArray"); if ((DWORD)(m_iDstIndex - destLB + m_iLength) > destLen) FCThrowResVoid(kArgumentException, L"Arg_LongerThanDestArray"); int r = 0; // Small perf optimization - we copy from one portion of an array back to // itself a lot when resizing collections, etc. The cost of doing the type // checking is significant for copying small numbers of bytes (~half of the time // for copying 1 byte within one array from element 0 to element 1). if (gc.pSrc == gc.pDst) r = AssignWillWork; else r = CanAssignArrayTypeNoGC(gc.pSrc, gc.pDst); if (r == AssignWrongType) { FCThrowResVoid(kArrayTypeMismatchException, L"ArrayTypeMismatch_CantAssignType"); } if (r == AssignWillWork) { src = (BYTE*)gc.pSrc->GetDataPtr(); dst = (BYTE*)gc.pDst->GetDataPtr(); size = gc.pSrc->GetMethodTable()->GetComponentSize(); g_IBCLogger.LogMethodTableAccess(gc.pSrc->GetMethodTable()); m_memmove(dst + ((m_iDstIndex - destLB) * size), src + ((m_iSrcIndex - srcLB) * size), m_iLength * size); if (gc.pDst->GetMethodTable()->ContainsPointers()) { GCHeap::GetGCHeap()->SetCardsAfterBulkCopy( (Object**) (dst + (m_iDstIndex * size)), m_iLength * size); } FC_GC_POLL(); return; } else if (reliable) { FCThrowResVoid(kArrayTypeMismatchException, L"ArrayTypeMismatch_ConstrainedCopy"); } BOOL castEachElement = false; BOOL boxEachElement = false; BOOL unboxEachElement = false; BOOL primitiveWiden = false; HELPER_METHOD_FRAME_BEGIN_PROTECT(gc); if (r == AssignDontKnow) { r = CanAssignArrayType(gc.pSrc, gc.pDst); } CONSISTENCY_CHECK(r != AssignDontKnow); switch (r) { case AssignWrongType: COMPlusThrow(kArrayTypeMismatchException, L"ArrayTypeMismatch_CantAssignType"); break; case AssignMustCast: castEachElement = true; break; case AssignWillWork: break; case AssignBoxValueClassOrPrimitive: boxEachElement = true; break; case AssignUnboxValueClassAndCast: castEachElement = true; unboxEachElement = true; break; case AssignPrimitiveWiden: primitiveWiden = true; break; default: _ASSERTE(!"Fell through switch in Array.Copy!"); } // If we were called from Array.ConstrainedCopy, ensure that the array copy // is guaranteed to succeed. _ASSERTE(!reliable || r == AssignWillWork); if (m_iLength > 0) { // Casting and boxing are mutually exclusive. But casting and unboxing may // coincide -- they are handled in the UnboxEachElement service. _ASSERTE(!boxEachElement || !castEachElement); if (r == AssignWillWork) { src = (BYTE*)gc.pSrc->GetDataPtr(); dst = (BYTE*)gc.pDst->GetDataPtr(); size = gc.pSrc->GetMethodTable()->GetComponentSize(); g_IBCLogger.LogMethodTableAccess(gc.pSrc->GetMethodTable()); m_memmove(dst + ((m_iDstIndex - destLB) * size), src + ((m_iSrcIndex - srcLB) * size), m_iLength * size); if (gc.pDst->GetMethodTable()->ContainsPointers()) { GCHeap::GetGCHeap()->SetCardsAfterBulkCopy( (Object**) (dst + (m_iDstIndex * size)), m_iLength * size); } } else if (unboxEachElement) { UnBoxEachElement(gc.pSrc, m_iSrcIndex - srcLB, gc.pDst, m_iDstIndex - destLB, m_iLength, castEachElement); } else if (boxEachElement) { BoxEachElement(gc.pSrc, m_iSrcIndex - srcLB, gc.pDst, m_iDstIndex - destLB, m_iLength); } else if (castEachElement) { _ASSERTE(!unboxEachElement); // handled above CastCheckEachElement(gc.pSrc, m_iSrcIndex - srcLB, gc.pDst, m_iDstIndex - destLB, m_iLength); } else if (primitiveWiden) { PrimitiveWiden(gc.pSrc, m_iSrcIndex - srcLB, gc.pDst, m_iDstIndex - destLB, m_iLength); } } HELPER_METHOD_FRAME_END(); } FCIMPLEND
IndexOf 的具体实现如下,为了更容易阅读,我把里面的前置判断去掉了。可以看到一个简单的函数,但是极为严谨,再回想自己写的代码,弱爆了。加了一点注释。
public static int IndexOf(Array array, Object value, int startIndex, int count) { int lb = array.GetLowerBound(0); // Try calling a quick native method to handle primitive types. int retVal; bool r = TrySZIndexOf(array, startIndex, count, value, out retVal); if (r) return retVal; Object[] objArray = array as Object[]; //转换为object 数组 int endIndex = startIndex + count; if (objArray != null) { //转换之后不为null if (value == null) { //如果传入的值为null,则查找objArray里面为null的,返回改值的位置 for (int i = startIndex; i < endIndex; i++) { if (objArray[i] == null) return i; } } else {//如果传入的值不为null,则逐个查找,找到后返回该值的位置 for (int i = startIndex; i < endIndex; i++) { Object obj = objArray[i]; //这个地方为什么要单独出来呢,直接这样 if (objArray[i]!= null && objArray[i].Equals(value)) ,之前自己的代码全是这么写的。 // 仔细推敲下,在if()里面只要取一次就好,性能应该可以稍微快一点。 if (obj != null && obj.Equals(value)) return i; } } } else { for (int i = startIndex; i < endIndex; i++) { Object obj = array.GetValue(i);//这个地方应该和上面类似 if( obj == null) { if(value == null) return i; } else { if( obj.Equals(value)) return i; } } } // Return one less than the lower bound of the array. This way, // for arrays with a lower bound of -1 we will not return -1 when the // item was not found. And for SZArrays (the vast majority), -1 still // works for them. return lb-1; }
里面的调用,可以在sscli中找到具体的实现
bool r = TrySZIndexOf(array, startIndex, count, value, out retVal);
最终的c++代码,很简单不是么
static int IndexOf(KIND array[], UINT32 index, UINT32 count, KIND value) { LEAF_CONTRACT; _ASSERTE(array != NULL && index >= 0 && count >= 0); for(UINT32 i=index; i<index+count; i++) if (array[i] == value) return i; return -1; }
LastIndexOf方法和IndexOf方法类似,不详细看了。看下Add和Insert方法的实现:
public virtual int Add(Object value) { if (_size == _items.Length) EnsureCapacity(_size + 1);//首先确保容量要够,不够会自动成倍添加,上面说过 _items[_size] = value; _version++; return _size++; } public virtual void Insert(int index, Object value) { if (index < 0 || index > _size) throw new ArgumentOutOfRangeException("index", Environment.GetResourceString("ArgumentOutOfRange_ArrayListInsert")); if (_size == _items.Length) EnsureCapacity(_size + 1); if (index < _size) {//如果插入的index比当前的长度要小,那么index之后的元素要后移 Array.Copy(_items, index, _items, index + 1, _size - index); //调用内部方法,上面已给出 } _items[index] = value; _size++; _version++; }
有源码可以看出,Add方法是直接添加到线性表的表尾,Insert方法是直接插入到指定位置,制定位置之后的元素要依次后移。显然Add方法的效率要高一些。用于添加的方法还有 AddRange和InsertRange,顾名思义就是插入一个范围数据即插入集合。那这两个方法有什么异同呢?先看AddRange吧
public virtual void AddRange(ICollection c) { InsertRange(_size, c); }
AddRange的方法是直接调用的InsertRange,从末尾插入一个集合。InsertRange源码如下:
public virtual void InsertRange(int index, ICollection c) { int count = c.Count; if (count > 0) { EnsureCapacity(_size + count); // shift existing items if (index < _size) {//依次向后移位,腾出位置 Array.Copy(_items, index, _items, index + count, _size - index); } Object[] itemsToInsert = new Object[count];//新建一个object数组 c.CopyTo(itemsToInsert, 0);//将新加的集合拷贝到新建的object数组中 itemsToInsert.CopyTo(_items, index);//再讲这个新建的数组拷贝到源列表中 _size += count; _version++; } }
下面看下ArrayList的排序方法的实现Sort:
public virtual void Sort() { Sort(0, Count, Comparer.Default); } //fcl里的最终的实现 public static void Sort(Array keys, Array items, int index, int length, IComparer comparer) { if (length > 1) { //如果是默认的,那么调用内部方法,下面会详细给出 if (comparer == Comparer.Default || comparer == null) { bool r = TrySZSort(keys, items, index, index + length - 1); if (r) return; } Object[] objKeys = keys as Object[]; Object[] objItems = null; if (objKeys != null) objItems = items as Object[]; if (objKeys != null && (items==null || objItems != null)) { SorterObjectArray sorter = new SorterObjectArray(objKeys, objItems, comparer); sorter.QuickSort(index, index + length - 1); } else { SorterGenericArray sorter = new SorterGenericArray(keys, items, comparer); sorter.QuickSort(index, index + length - 1); } } }
可以自己实现一个比较器IComparer,也可以使用默认的比较器。如果使用的是默认的比较器,那么将会调用clr底层的快速排序方法,下面是从sscli中查到的C++源码:
static void QuickSort(KIND keys[], KIND items[], int left, int right) {//KIND 在头部给出了定义 template <class KIND> WRAPPER_CONTRACT; // Make sure left != right in your own code. _ASSERTE(keys != NULL && left < right); do { int i = left; int j = right; KIND x = keys[i + ((j - i) >> 1)]; do { while (keys[i] < x) i++; while (x < keys[j]) j--; _ASSERTE(i>=left && j<=right); if (i > j) break; if (i < j) { KIND key = keys[i]; keys[i] = keys[j]; keys[j] = key; if (items != NULL) { KIND item = items[i]; items[i] = items[j]; items[j] = item; } } i++; j--; } while (i <= j); if (j - left <= right - i) { if (left < j) QuickSort(keys, items, left, j); left = i; } else { if (i < right) QuickSort(keys, items, i, right); right = j; } } while (left < right); }
和Sort方法类似,BinarySearch(二分查找)方法也可以使用自定义的比较器看下BinarySearch的具体实现:
public static int BinarySearch(Array array, int index, int length, Object value, IComparer comparer) { //去掉一些前置判断 if (comparer == null) comparer = Comparer.Default; if (comparer == Comparer.Default) { int retval; bool r = TrySZBinarySearch(array, index, length, value, out retval); if (r) return retval; } int lo = index; int hi = index + length - 1; Object[] objArray = array as Object[]; if(objArray != null) { while (lo <= hi) { // i might overflow if lo and hi are both large positive numbers. int i = GetMedian(lo, hi);//取中位数 int c; try { c = comparer.Compare(objArray[i], value);//比较这个中间值是否是要查找的值,c=0找到 c为负数在右边,c为正数在左边 } catch (Exception e) { throw new InvalidOperationException(Environment.GetResourceString("InvalidOperation_IComparerFailed"), e); } if (c == 0) return i; //找到,返回下标 if (c < 0) { lo = i + 1; } else { hi = i - 1; } } } else { while (lo <= hi) { int i = GetMedian(lo, hi); int c; try { c = comparer.Compare(array.GetValue(i), value); } catch (Exception e) { throw new InvalidOperationException(Environment.GetResourceString("InvalidOperation_IComparerFailed"), e); } if (c == 0) return i; if (c < 0) { lo = i + 1; } else { hi = i - 1; } } } return ~lo; }
看下这个TrySZBinarySearch在clr中的具体实现吧,同样很易懂。
static int BinarySearchBitwiseEquals(KIND array[], int index, int length, KIND value) { WRAPPER_CONTRACT; _ASSERTE(array != NULL && length >= 0 && index >= 0); int lo = index; int hi = index + length - 1; // Note: if length == 0, hi will be Int32.MinValue, and our comparison // here between 0 & -1 will prevent us from breaking anything. while (lo <= hi) { int i = lo + ((hi - lo) >> 1); if (array[i] < value) { lo = i + 1; } else if (array[i] > value){ hi = i - 1; } else { return i; } } return ~lo; }
ArrayList的其他方法,也极为易懂和类似,不在罗列。
Queue
队列是特殊的线性表,先进先出的结构。从源码中可以看出,FCL中的Queue是一种循环队列。先看Queue的属性
private Object[] _array; //存储的数据 private int _head; // 对一个有效元素 private int _tail; // 最后一个有效元素 private int _size; // 元素数量 private int _growFactor; // 增长因素 100 == 1.0, 130 == 1.3, 200 == 2.0,取值范围 1.0到10.0之间 private int _version; [NonSerialized] private Object _syncRoot; private const int _MinimumGrow = 4; //最小增长量 private const int _ShrinkThreshold = 32;//这个地方极为扯淡,定义了没用,下面直接写死32
初始化:
public Queue() : this(32, (float)2.0) { } // Creates a queue with room for capacity objects. The default grow factor // is used. // public Queue(int capacity) : this(capacity, (float)2.0) { } // Creates a queue with room for capacity objects. When full, the new // capacity is set to the old capacity * growFactor. // public Queue(int capacity, float growFactor) { if (capacity < 0) throw new ArgumentOutOfRangeException("capacity", Environment.GetResourceString("ArgumentOutOfRange_NeedNonNegNum")); if (!(growFactor >= 1.0 && growFactor <= 10.0)) throw new ArgumentOutOfRangeException("growFactor", Environment.GetResourceString("ArgumentOutOfRange_QueueGrowFactor", 1, 10)); Contract.EndContractBlock(); _array = new Object[capacity]; _head = 0; _tail = 0; _size = 0; _growFactor = (int)(growFactor * 100); }
入队操作:
public virtual void Enqueue(Object obj) { if (_size == _array.Length) {//如果队满,则重新分配容量 int newcapacity = (int)((long)_array.Length * (long)_growFactor / 100); if (newcapacity < _array.Length + _MinimumGrow) {//如果新分配的容量小于当前容量加上最小增长量,那么把当前容量加最小增长量分配给新分配的容量 newcapacity = _array.Length + _MinimumGrow; } SetCapacity(newcapacity);//重新设置容量 } _array[_tail] = obj; _tail = (_tail + 1) % _array.Length;//如果_taill+1<_array.Length 那么 _tail=_tail+1;否则,_tail=0;表示队列已满。 //循环队列的写法 _size++; _version++; }
private void SetCapacity(int capacity) { Object[] newarray = new Object[capacity]; if (_size > 0) { if (_head < _tail) {//环形队列头部项索引在尾部索引前面 Array.Copy(_array, _head, newarray, 0, _size); } else {//环形队列头部项索引在尾部索引后面 Array.Copy(_array, _head, newarray, 0, _array.Length - _head);//copy _head 到 Length部分 Array.Copy(_array, 0, newarray, _array.Length - _head, _tail);//copy 0 到_tail 部分 } } _array = newarray; _head = 0; _tail = (_size == capacity) ? 0 : _size; _version++; }
出队操作Dequeue(),出队,并从队列中删除;Peek()方法,取队列的第一位元素,不从队列中删除。
public virtual Object Dequeue() { if (Count == 0) throw new InvalidOperationException(Environment.GetResourceString("InvalidOperation_EmptyQueue")); Contract.EndContractBlock(); Object removed = _array[_head]; //取元素 _array[_head] = null; //删除元素 _head = (_head + 1) % _array.Length;//循环队列,头部索引移到下一位 _size--; _version++; return removed; }
public virtual Object Peek() { if (Count == 0) throw new InvalidOperationException(Environment.GetResourceString("InvalidOperation_EmptyQueue")); Contract.EndContractBlock(); return _array[_head]; }
可以看出队列的入队和出队的时间复杂度是O(1);但是入列的时候如果需要重置容量,那么时间复杂度会变为O(n)
队列中的其他操作:
//是否包含 public virtual bool Contains(Object obj) { int index = _head; int count = _size; while (count-- > 0) { if (obj == null) { if (_array[index] == null) return true; } else if (_array[index] != null && _array[index].Equals(obj)) { return true; } index = (index + 1) % _array.Length;//index向后移 } return false; } //取某个元素 internal Object GetElement(int i){ return _array[(_head + i) % _array.Length] }
和ArrayList类似,Queue也有线程安全的实现,Queue.Synchronized(),返回一个 同步Queue。实现的方式就是在队列的操作上加锁。
Stack
栈,先进后出,像弹夹。属性定义:
private Object[] _array; // Storage for stack elements private int _size; // Number of items in the stack. private int _version; // Used to keep enumerator in [....] w/ collection. private Object _syncRoot; private const int _defaultCapacity = 10;
可以看出栈的结构属性更为简单,默认的容量是10;初始化操作:
public Stack() { _array = new Object[_defaultCapacity]; _size = 0; _version = 0; } // Create a stack with a specific initial capacity. The initial capacity // must be a non-negative number. public Stack(int initialCapacity) { if (initialCapacity < 0) throw new ArgumentOutOfRangeException("initialCapacity", Environment.GetResourceString("ArgumentOutOfRange_NeedNonNegNum")); Contract.EndContractBlock(); if (initialCapacity < _defaultCapacity) initialCapacity = _defaultCapacity; // Simplify doubling logic in Push. _array = new Object[initialCapacity]; _size = 0; _version = 0; }
入栈操作:
public virtual void Push(Object obj) { //Contract.Ensures(Count == Contract.OldValue(Count) + 1); if (_size == _array.Length) {//如果容量满了,那么成2倍增加 Object[] newArray = new Object[2*_array.Length]; Array.Copy(_array, 0, newArray, 0, _size); _array = newArray; } _array[_size++] = obj; _version++; }
出栈操作:
public virtual Object Pop() { _version++; Object obj = _array[--_size];//取元素并将长度减1 _array[_size] = null; // 删除元素 return obj; }
和Queue类似,Stack也有Peek操作,实现方式类似,同样也有Synchronized方法,实现方式也是类似的。
SortedList
属性:
private Object[] keys;//键数组 private Object[] values;//值数组 private int _size; private int version; private IComparer comparer; private KeyList keyList;//建集合 继承IList 内部使用 SortList 与keys关联 private ValueList valueList;//值集合 继承IList 内部使用 SortList 与 values关联 [NonSerialized] private Object _syncRoot; private const int _defaultCapacity = 16;//默认容量 16 private static Object[] emptyArray = EmptyArray<Object>.Value;//空数组 等效 New Object[0];
SortedList的容量也是动态的
private void EnsureCapacity(int min) { int newCapacity = keys.Length == 0? 16: keys.Length * 2;//2倍增上 // Allow the list to grow to maximum possible capacity (~2G elements) before encountering overflow. // Note that this check works even when _items.Length overflowed thanks to the (uint) cast if ((uint)newCapacity > Array.MaxArrayLength) newCapacity = Array.MaxArrayLength; if (newCapacity < min) newCapacity = min; Capacity = newCapacity; } public virtual int Capacity { get { return keys.Length; } set { if (value < Count) { throw new ArgumentOutOfRangeException("value", Environment.GetResourceString("ArgumentOutOfRange_SmallCapacity")); } Contract.EndContractBlock(); if (value != keys.Length) { if (value > 0) {//如果容量发生改变,新建键数组和值指数组,并将源键值数组复制进去。如果新的容量不大于0则将键数组和值数组置空 Object[] newKeys = new Object[value]; Object[] newValues = new Object[value]; if (_size > 0) { Array.Copy(keys, 0, newKeys, 0, _size); Array.Copy(values, 0, newValues, 0, _size); } keys = newKeys; values = newValues; } else { // size can only be zero here. Contract.Assert( _size == 0, "Size is not zero"); keys = emptyArray; values = emptyArray; } } } }
SortedList 的IndexOfKey方法,可以发现内部使用二分法查找,具体C++代码上面ArrayList中已经给出。
public virtual int IndexOfKey(Object key) { if (key == null) throw new ArgumentNullException("key", Environment.GetResourceString("ArgumentNull_Key")); Contract.EndContractBlock(); int ret = Array.BinarySearch(keys, 0, _size, key, comparer); return ret >=0 ? ret : -1; }
SortedList的IndexOfValue方法内部也是调用的IndexOfKey这个方法,如下:
public virtual int IndexOfValue(Object value) { return Array.IndexOf(values, value, 0, _size); }
插入操作如下,很易懂。可以看出,Add一个元素的时候
private void Insert(int index, Object key, Object value) { if (_size == keys.Length) EnsureCapacity(_size + 1); if (index < _size) { Array.Copy(keys, index, keys, index + 1, _size - index); Array.Copy(values, index, values, index + 1, _size - index); } keys[index] = key; values[index] = value; _size++; version++; } public virtual void Add(Object key, Object value) { if (key == null) throw new ArgumentNullException("key", Environment.GetResourceString("ArgumentNull_Key")); Contract.EndContractBlock(); int i = Array.BinarySearch(keys, 0, _size, key, comparer);//这个地方是二分法查找。comparer是排序器,可以自己实现。默认的是按照key来排序。每次新添加元素都会重新排序。 if (i >= 0) throw new ArgumentException(Environment.GetResourceString("Argument_AddingDuplicate__", GetKey(i), key)); Insert(~i, key, value);//这个地方比较绕,Array.BinarySearch 如果找不到,返回的是index取反,结果为-1,在对这个-1取反,结果为0.在0这个位置插入
}
删除操作,同样也很易懂,代码如下:
public virtual void RemoveAt(int index) { if (index < 0 || index >= Count) throw new ArgumentOutOfRangeException("index", Environment.GetResourceString("ArgumentOutOfRange_Index")); Contract.EndContractBlock(); _size--; if (index < _size) { Array.Copy(keys, index + 1, keys, index, _size - index); Array.Copy(values, index + 1, values, index, _size - index); } keys[_size] = null; values[_size] = null; version++; } public virtual void Remove(Object key) { int i = IndexOfKey(key); if (i >= 0) RemoveAt(i); }
同样的,SortedList也有线程同步的方法 SortedList.Synchronized() 实现方式和ArrayList,Queue,Stack 并无二致。
至此,告一段落。几个常用的集合,自己有了更为深刻的理解。阅读优秀的代码是一种享受,阅读渣渣的代码,是虐心!