您的位置:首页 > 其它

STL源码剖析 学习笔记 MiniSTL

2017-07-27 10:54 363 查看
https://github.com/joeyleeeeeee97

目录:

  第二章  空间适配器

  第三章  迭代器

  第四章  序列式容器(vector,list,deque,stack,heap,priority_queue,slist)

  第五章  关联式容器(树的算法 + RB_tree ,set,map,hashtable)

  第六章  算法

  第七章  仿函数

  第八章  适配器(adapet)

第二章  空间适配器

  具有次配置力的SGI空间适配器,STL allocator将对象的内存分配和初始化分离开,内存配置由alloc:allocate 负责,内存释放由deallocate 负责,对象构造操作 由 construct() 负责,对象析构操作由destroy() 负责。

  构造和析构的基本工具:construct 和destory

trivial constructor ::
In simple words a "trivial" special member function literally means a member function that does its job in a very straightforward manner. The "straightforward manner" means different thing for different kinds of special member functions.

For a default constructor and destructor being "trivial" means literally "do nothing at all". For copy-constructor and copy-assignment operator, being "trivial" means literally "be equivalent to simple raw memory copying" (like copy with
memcpy
).

If you define a constructor yourself, it is considered non-trivial, even if it doesn't do anything, so a trivial constructor must be implicitly defined by the compiler.

In order for a special member function to satisfy the above requirements, the class must have a very simplistic structure, it must not require any hidden initializations when an object is being created or destroyed, or any hidden additional internal manipulations when it is being copied.

For example, if class has virtual functions, it will require some extra hidden initializations when objects of this class are being created (initialize virtual method table and such), so the constructor for this class will not qualify as trivial.

For another example, if a class has virtual base classes, then each object of this class might contain hidden pointers that point to other parts of the very same object. Such a self-referential object cannot be copied by a simple raw memory copy routine (like
memcpy
). Extra manipulations will be necessary to properly re-initialize the hidden pointers in the copy. For this reason the copy constructor and copy-assignment operator for this class will not qualify as trivial.

For obvious reasons, this requirement is recursive: all subobjects of the class (bases and non-static members) must also have trivial constructors.

    通过type_traits 判断是不是POD类型然后使用相应的destroy

  

#pragma once
#ifndef _CONSTRUCT_H
#define _CONSTRUCT_H

#include<new.h>

template<class T1, class T2>
inline void construct(T1* p, const T2& value)
{
new(p) T1(value);
}

template<class T>
inline void destroy(T* ptr)
{
ptr->~T();
}

//以下是destroy()第二版本 接受两个迭代器此函数设法找出元素的数值型别
//进而利用__type_traits<>求取最适当的措施
template<class FowardIterator>
inline void __destroy(ForwardIterator first, ForwardIterator last, __true_type)
{

}

template<class FowardIteraot>
inline void __destroy(ForwardIterator first, ForwardIterator last, __false_type)
{
for (;; first < last; first++)
destroy(&*first);
}

template<class FowardIterator,class T>
inline void destroy(ForwardIterator first, ForwardIterator last, T*)
{
typedef typename __type_traits<T>::has_trivial_destructor trivial_destructor;
__destroy(first, last, trivial_destructor());
}
#endif


  空间的配置和释放 std::alloc

  SGI对此的设计哲学如下: 向system heap 要求空间;考虑多线程状态;考虑内存不足时候的应变措施;考虑过多“小型区块”可能造成的碎片问题

  双层配置器:第一层直接使用malloc 和free 第二级配置器在大于128bytes的请求下 转为一级配置器分配内存,在小于128bytes的内存请求下,内存池分配。

  

#ifndef _ALLOCATOR_H
#define _ALLOCATOR_H

#include<new>
#include<cstddef>
#include<climits>
#include<iostream>
#include<cstdlib>
#include"construct.h"
namespace ministl
{
template<class T>
inline T* _allocate(size_t n, T*)
{
return static_cast<T*>(alloc::allocate(n));
}

template<class T>
inline void _deallocate(T* buffer)
{
alloc::deallocate(static_cast<T*>(buffer), sizeof(T));
}

template<class T>
class allocator
{
public:
typedef T value_type;
typedef T* pointer;
typedef const T* const_pointer;
typedef T& reference;
typedef const T& const_reference;
typedef size_t size_type;
typedef ptrdiff_t difference_type;

template<class U>
struct rebind
{
typedef allocator<U> other;
};

pointer allocate(size_type n,const void* hint = 0)
{
return _allocate((difference_type)n, pointer(0));
}

void deallocate(pointer p, size_type n)
{
_deallocate(p);
}

pointer address(reference x)
{
return (pointer)&x;
}

const_pointer const_address(const reference x)
{
return (const_pointer)&x;
}

size_type max_size()const
{
return size_type(UINT_MAX / sizeof(T));
}
};
}
#endif


#pragma once
#ifndef _ALLOC_H__
#define _ALLOC_H__
#include <cstdlib>
/*
第二级适配器,当区块小于128bytes时,以内存池管理
*/
namespace ministl
{
class alloc
{
private:
enum {_ALIGN = 8};//小型区块
enum {_MAXBYTES = 128};
enum {_NFREELISTS = _MAXBYTES/_ALIGN};
enum { _NOBJS = 20 };
private:
//可以看作确定地址也可以看作指向下一节点
union obj
{
union obj *next;
char client[1];
};

static obj* freelist[_NFREELISTS];

private:
static char *start, *end;
static size_t heap_size;

private:
static size_t ROUND_UP(size_t bytes)
{
return ((bytes + _ALIGN - 1)&~(_ALIGN - 1));
}
static size_t FREELIST_INDEX(size_t bytes)
{
return (((bytes)+_ALIGN - 1) / _ALIGN - 1);
}
static void* refill(size_t n);
static char* chunk_alloc(size_t size, size_t &nobjs);
public:
static void* allocate(size_t bytes);
static void deallocate(void* ptr, size_t bytes);
static void *reallocate(void *ptr, size_t old_size, size_t new_size);
};

char *alloc::start = 0;
char *alloc::end = 0;
size_t alloc::heap_size = 0;
alloc::obj *alloc::freelist[_NFREELISTS] = { 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 };

void *alloc::allocate(size_t bytes)
{
if (bytes > _MAXBYTES)
{
return malloc(bytes);
}
obj *pos = freelist[FREELIST_INDEX(bytes)];
if (pos)
{
freelist[FREELIST_INDEX(bytes)] = pos->next;
return pos;
}
else
{
return refill(ROUND_UP(bytes));
}
}

void alloc::deallocate(void* ptr, size_t bytes)
{
if (bytes > _MAXBYTES)
{
free(ptr);
}
else
{
obj *tmp = (obj *)ptr;
tmp->next = freelist[FREELIST_INDEX(bytes)];
freelist[FREELIST_INDEX(bytes)] = tmp;
}
}

void *alloc::reallocate(void *ptr, size_t old_size, size_t new_size)
{
alloc::deallocate(ptr, old_size);
ptr = allocate(new_size);
return ptr;
}
//返回一个大小为n的对象,并且有时候会为适当的freelist增加节点
//假设n已经上调到8的倍数
void *alloc::refill(size_t bytes)
{
size_t nobjs = _NOBJS;
obj **myfreelist = NULL;
obj *result = NULL;
obj *current_obj = 0, *next_obj = 0;
char *chunk = chunk_alloc(bytes, nobjs);
if (nobjs == 1)
{
return chunk;
}
else
{
myfreelist = freelist+ FREELIST_INDEX(bytes);
//这一块返回给客户端
result = (obj*)chunk;
//下面导引freelist指向新配置的空间(内存池)
*myfreelist = next_obj = (obj*)(chunk + bytes);

for (int i = 1;; i++)
{
current_obj = next_obj;
next_obj = (obj*)((char*)next_obj + bytes);

if (nobjs - 1 == i)
{
current_obj->next = 0;
break;
}
else
{
current_obj->next = next_obj;
}
}
return result;
}
}
/*
从链表中分配nobjs个空间,每一块的大小为bytes.如果链表中至少有一个满足要求,那么分配尽可能多的块并且修改nobjs的值,如果数量不足就从内存中分配两倍的空间
*/
char *alloc::chunk_alloc(size_t bytes, size_t &nobjs)
{
char *result = 0;
size_t total_bytes = bytes*nobjs;
size_t bytes_left = end - start;
if (bytes_left >= total_bytes)
{
result = start;
start = start + total_bytes;
return result;
}
else if(bytes_left >= bytes)
{
nobjs = bytes_left / bytes;
total_bytes = nobjs*bytes;
result = start;
start = start + total_bytes;
return result;
}
else
{
size_t bytes_to_get = 2 * total_bytes + ROUND_UP(heap_size >> 4);
if (bytes_left > 0)
{
obj* volatile *myfreelist = freelist + FREELIST_INDEX(bytes_left);
((obj*)start)->next = *myfreelist;
*myfreelist = (obj*)start;
}
start = (char*)malloc(bytes_to_get);
if (!start)
{
obj **myfreelist = 0, *p = 0;
for (int i = 0; i < _MAXBYTES; i += _ALIGN)
{
myfreelist = freelist + FREELIST_INDEX(i);
p = *myfreelist;
if (!p)
{
*myfreelist = p->next;
start = (char*)p;
end = start + i;
return chunk_alloc(bytes, nobjs);
}
}
end = 0;
}
heap_size += bytes_to_get;
end = start + bytes_to_get;
return chunk_alloc(bytes, nobjs);
}
}

}
#endif


4.3 list

相比較vector的連續空間,list的好處在於每次插入或者刪除一個元素就配置或者釋放一個元素的空間,因此List對於空間的運用绝对的精准,list本身和list的结点是不同的结构需要分开设计。list不能用普通指针作为迭代器因为它的结点不能抱枕在存储空间中连续存在,list迭代器必须有能力指向List的节点并且正确的进行递增递减操作,list是一个双向迭代器。list有一个重要性质,插入和接和操作不会导致原有的迭代器失效。
内容来自用户分享和网络整理,不保证内容的准确性,如有侵权内容,可联系管理员处理 点击这里给我发消息
标签: