http://blog.csdn.net/haoel/article/details/3081385
C++ 对象的内存布局(下)
陈皓
重复继承
下面我们再来看看,发生重复继承的情况。所谓重复继承,也就是某个基类被间接地重复继承了多次。
下图是一个继承图,我们重载了父类的f()函数。
其类继承的源代码如下所示。其中,每个类都有两个变量,一个是整形(4字节),一个是字符(1字节),而且还有自己的虚函数,自己overwrite父类的虚函数。如子类D中,f()覆盖了超类的函数, f1() 和f2() 覆盖了其父类的虚函数,Df()为自己的虚函数。
class B
{
public:
int ib;
char cb;
public:
B():ib(0),cb('B') {}
virtual void f() { cout << "B::f()" << endl;}
virtual void Bf() { cout << "B::Bf()" << endl;}
};
class B1 : public B
{
public:
int ib1;
char cb1;
public:
B1():ib1(11),cb1('1') {}
virtual void f() { cout << "B1::f()" << endl;}
virtual void f1() { cout << "B1::f1()" << endl;}
virtual void Bf1() { cout << "B1::Bf1()" << endl;}
};
class B2: public B
{
public:
int ib2;
char cb2;
public:
B2():ib2(12),cb2('2') {}
virtual void f() { cout << "B2::f()" << endl;}
virtual void f2() { cout << "B2::f2()" << endl;}
virtual void Bf2() { cout << "B2::Bf2()" << endl;}
};
class D : public B1, public B2
{
public:
int id;
char cd;
public:
D():id(100),cd('D') {}
virtual void f() { cout << "D::f()" << endl;}
virtual void f1() { cout << "D::f1()" << endl;}
virtual void f2() { cout << "D::f2()" << endl;}
virtual void Df() { cout << "D::Df()" << endl;}
};
我们用来存取子类内存布局的代码如下所示:(在VC++ 2003和G++ 3.4.4下)
typedef void(*Fun)(void);
int** pVtab = NULL;
Fun pFun = NULL;
D d;
pVtab = (int**)&d;
cout << "[0] D::B1::_vptr->" << endl;
pFun = (Fun)pVtab[0][0];
cout << " [0] "; pFun();
pFun = (Fun)pVtab[0][1];
cout << " [1] "; pFun();
pFun = (Fun)pVtab[0][2];
cout << " [2] "; pFun();
pFun = (Fun)pVtab[0][3];
cout << " [3] "; pFun();
pFun = (Fun)pVtab[0][4];
cout << " [4] "; pFun();
pFun = (Fun)pVtab[0][5];
cout << " [5] 0x" << pFun << endl;
cout << "[1] B::ib = " << (int)pVtab[1] << endl;
cout << "[2] B::cb = " << (char)pVtab[2] << endl;
cout << "[3] B1::ib1 = " << (int)pVtab[3] << endl;
cout << "[4] B1::cb1 = " << (char)pVtab[4] << endl;
cout << "[5] D::B2::_vptr->" << endl;
pFun = (Fun)pVtab[5][0];
cout << " [0] "; pFun();
pFun = (Fun)pVtab[5][1];
cout << " [1] "; pFun();
pFun = (Fun)pVtab[5][2];
cout << " [2] "; pFun();
pFun = (Fun)pVtab[5][3];
cout << " [3] "; pFun();
pFun = (Fun)pVtab[5][4];
cout << " [4] 0x" << pFun << endl;
cout << "[6] B::ib = " << (int)pVtab[6] << endl;
cout << "[7] B::cb = " << (char)pVtab[7] << endl;
cout << "[8] B2::ib2 = " << (int)pVtab[8] << endl;
cout << "[9] B2::cb2 = " << (char)pVtab[9] << endl;
cout << "[10] D::id = " << (int)pVtab[10] << endl;
cout << "[11] D::cd = " << (char)pVtab[11] << endl;
程序运行结果如下:
GCC 3.4.4 |
VC++ 2003 |
[0] D::B1::_vptr-> [0] D::f() [1] B::Bf() [2] D::f1() [3] B1::Bf1() [4] D::f2() [5] 0x1 [1] B::ib = 0 [2] B::cb = B [3] B1::ib1 = 11 [4] B1::cb1 = 1 [5] D::B2::_vptr-> [0] D::f() [1] B::Bf() [2] D::f2() [3] B2::Bf2() [4] 0x0 [6] B::ib = 0 [7] B::cb = B [8] B2::ib2 = 12 [9] B2::cb2 = 2 [10] D::id = 100 [11] D::cd = D |
[0] D::B1::_vptr-> [0] D::f() [1] B::Bf() [2] D::f1() [3] B1::Bf1() [4] D::Df() [5] 0x00000000 [1] B::ib = 0 [2] B::cb = B [3] B1::ib1 = 11 [4] B1::cb1 = 1 [5] D::B2::_vptr-> [0] D::f() [1] B::Bf() [2] D::f2() [3] B2::Bf2() [4] 0x00000000 [6] B::ib = 0 [7] B::cb = B [8] B2::ib2 = 12 [9] B2::cb2 = 2 [10] D::id = 100 [11] D::cd = D |
下面是对于子类实例中的虚函数表的图:
我们可以看见,最顶端的父类B其成员变量存在于B1和B2中,并被D给继承下去了。而在D中,其有B1和B2的实例,于是B的成员在D的实例中存在两份,一份是B1继承而来的,另一份是B2继承而来的。所以,如果我们使用以下语句,则会产生二义性编译错误:
D d;
d.ib = 0; //二义性错误
d.B1::ib = 1; //正确
d.B2::ib = 2; //正确
注意,上面例程中的最后两条语句存取的是两个变量。虽然我们消除了二义性的编译错误,但B类在D中还是有两个实例,这种继承造成了数据的重复,我们叫这种继承为重复继承。重复的基类数据成员可能并不是我们想要的。所以,C++引入了虚基类的概念。
钻石型多重虚拟继承
虚拟继承的出现就是为了解决重复继承中多个间接父类的问题的。钻石型的结构是其最经典的结构。也是我们在这里要讨论的结构:
上述的“重复继承”只需要把B1和B2继承B的语法中加上virtual 关键,就成了虚拟继承,其继承图如下所示:
上图和前面的“重复继承”中的类的内部数据和接口都是完全一样的,只是我们采用了虚拟继承:其省略后的源码如下所示:
class B {……};
class B1 : virtual public B{……};
class B2: virtual public B{……};
class D : public B1, public B2{ …… };
在查看D之前,我们先看一看单一虚拟继承的情况。下面是一段在VC++2003下的测试程序:(因为VC++和GCC的内存而局上有一些细节上的不同,所以这里只给出VC++的程序,GCC下的程序大家可以根据我给出的程序自己仿照着写一个去试一试):
int** pVtab = NULL;
Fun pFun = NULL;
B1 bb1;
pVtab = (int**)&bb1;
cout << "[0] B1::_vptr->" << endl;
pFun = (Fun)pVtab[0][0];
cout << " [0] ";
pFun(); //B1::f1();
cout << " [1] ";
pFun = (Fun)pVtab[0][1];
pFun(); //B1::bf1();
cout << " [2] ";
cout << pVtab[0][2] << endl;
cout << "[1] = 0x";
cout << (int*)*((int*)(&bb1)+1) <<endl; //B1::ib1
cout << "[2] B1::ib1 = ";
cout << (int)*((int*)(&bb1)+2) <<endl; //B1::ib1
cout << "[3] B1::cb1 = ";
cout << (char)*((int*)(&bb1)+3) << endl; //B1::cb1
cout << "[4] = 0x";
cout << (int*)*((int*)(&bb1)+4) << endl; //NULL
cout << "[5] B::_vptr->" << endl;
pFun = (Fun)pVtab[5][0];
cout << " [0] ";
pFun(); //B1::f();
pFun = (Fun)pVtab[5][1];
cout << " [1] ";
pFun(); //B::Bf();
cout << " [2] ";
cout << "0x" << (Fun)pVtab[5][2] << endl;
cout << "[6] B::ib = ";
cout << (int)*((int*)(&bb1)+6) <<endl; //B::ib
cout << "[7] B::cb = ";
其运行结果如下(我结出了GCC的和VC++2003的对比):
GCC 3.4.4 |
VC++ 2003 |
[0] B1::_vptr -> [0] : B1::f() [1] : B1::f1() [2] : B1::Bf1() [3] : 0 [1] B1::ib1 : 11 [2] B1::cb1 : 1 [3] B::_vptr -> [0] : B1::f() [1] : B::Bf() [2] : 0 [4] B::ib : 0 [5] B::cb : B [6] NULL : 0 |
[0] B1::_vptr-> [0] B1::f1() [1] B1::Bf1() [2] 0 [1] = 0x00454310 ç该地址取值后是-4 [2] B1::ib1 = 11 [3] B1::cb1 = 1 [4] = 0x00000000 [5] B::_vptr-> [0] B1::f() [1] B::Bf() [2] 0x00000000 [6] B::ib = 0 [7] B::cb = B |
这里,大家可以自己对比一下。关于细节上,我会在后面一并再说。
下面的测试程序是看子类D的内存布局,同样是VC++ 2003的(因为VC++和GCC的内存布局上有一些细节上的不同,而VC++的相对要清楚很多,所以这里只给出VC++的程序,GCC下的程序大家可以根据我给出的程序自己仿照着写一个去试一试):
D d;
pVtab = (int**)&d;
cout << "[0] D::B1::_vptr->" << endl;
pFun = (Fun)pVtab[0][0];
cout << " [0] "; pFun(); //D::f1();
pFun = (Fun)pVtab[0][1];
cout << " [1] "; pFun(); //B1::Bf1();
pFun = (Fun)pVtab[0][2];
cout << " [2] "; pFun(); //D::Df();
pFun = (Fun)pVtab[0][3];
cout << " [3] ";
cout << pFun << endl;
//cout << pVtab[4][2] << endl;
cout << "[1] = 0x";
cout << (int*)((&dd)+1) <<endl; //????
cout << "[2] B1::ib1 = ";
cout << *((int*)(&dd)+2) <<endl; //B1::ib1
cout << "[3] B1::cb1 = ";
cout << (char)*((int*)(&dd)+3) << endl; //B1::cb1
//---------------------
cout << "[4] D::B2::_vptr->" << endl;
pFun = (Fun)pVtab[4][0];
cout << " [0] "; pFun(); //D::f2();
pFun = (Fun)pVtab[4][1];
cout << " [1] "; pFun(); //B2::Bf2();
pFun = (Fun)pVtab[4][2];
cout << " [2] ";
cout << pFun << endl;
cout << "[5] = 0x";
cout << *((int*)(&dd)+5) << endl; // ???
cout << "[6] B2::ib2 = ";
cout << (int)*((int*)(&dd)+6) <<endl; //B2::ib2
cout << "[7] B2::cb2 = ";
cout << (char)*((int*)(&dd)+7) << endl; //B2::cb2
cout << "[8] D::id = ";
cout << *((int*)(&dd)+8) << endl; //D::id
cout << "[9] D::cd = ";
cout << (char)*((int*)(&dd)+9) << endl;//D::cd
cout << "[10] = 0x";
cout << (int*)*((int*)(&dd)+10) << endl;
//---------------------
cout << "[11] D::B::_vptr->" << endl;
pFun = (Fun)pVtab[11][0];
cout << " [0] "; pFun(); //D::f();
pFun = (Fun)pVtab[11][1];
cout << " [1] "; pFun(); //B::Bf();
pFun = (Fun)pVtab[11][2];
cout << " [2] ";
cout << pFun << endl;
cout << "[12] B::ib = ";
cout << *((int*)(&dd)+12) << endl; //B::ib
cout << "[13] B::cb = ";
cout << (char)*((int*)(&dd)+13) <<endl;//B::cb
下面给出运行后的结果(分VC++和GCC两部份)
GCC 3.4.4 |
VC++ 2003 |
[0] B1::_vptr -> [0] : D::f() [1] : D::f1() [2] : B1::Bf1() [3] : D::f2() [4] : D::Df() [5] : 1 [1] B1::ib1 : 11 [2] B1::cb1 : 1 [3] B2::_vptr -> [0] : D::f() [1] : D::f2() [2] : B2::Bf2() [3] : 0 [4] B2::ib2 : 12 [5] B2::cb2 : 2 [6] D::id : 100 [7] D::cd : D [8] B::_vptr -> [0] : D::f() [1] : B::Bf() [2] : 0 [9] B::ib : 0 [10] B::cb : B [11] NULL : 0 |
[0] D::B1::_vptr-> [0] D::f1() [1] B1::Bf1() [2] D::Df() [3] 00000000 [1] = 0x0013FDC4 ç 该地址取值后是-4 [2] B1::ib1 = 11 [3] B1::cb1 = 1 [4] D::B2::_vptr-> [0] D::f2() [1] B2::Bf2() [2] 00000000 [5] = 0x4539260 ç 该地址取值后是-4 [6] B2::ib2 = 12 [7] B2::cb2 = 2 [8] D::id = 100 [9] D::cd = D [10] = 0x00000000 [11] D::B::_vptr-> [0] D::f() [1] B::Bf() [2] 00000000 [12] B::ib = 0 [13] B::cb = B |
关于虚拟继承的运行结果我就不画图了(前面的作图已经让我产生了很严重的厌倦感,所以就偷个懒了,大家见谅了)
在上面的输出结果中,我用不同的颜色做了一些标明。我们可以看到如下的几点:
1)无论是GCC还是VC++,除了一些细节上的不同,其大体上的对象布局是一样的。也就是说,先是B1(黄色),然后是B2(绿色),接着是D(灰色),而B这个超类(青蓝色)的实例都放在最后的位置。
2)关于虚函数表,尤其是第一个虚表,GCC和VC++有很重大的不一样。但仔细看下来,还是VC++的虚表比较清晰和有逻辑性。
3)VC++和GCC都把B这个超类放到了最后,而VC++有一个NULL分隔符把B和B1和B2的布局分开。GCC则没有。
4)VC++中的内存布局有两个地址我有些不是很明白,在其中我用红色标出了。取其内容是-4。接道理来说,这个指针应该是指向B类实例的内存地址(这个做法就是为了保证重复的父类只有一个实例的技术)。但取值后却不是。这点我目前还并不太清楚,还向大家请教。
5)GCC的内存布局中在B1和B2中则没有指向B的指针。这点可以理解,编译器可以通过计算B1和B2的size而得出B的偏移量。
结束语
C++这门语言是一门比较复杂的语言,对于程序员来说,我们似乎永远摸不清楚这门语言背着我们在干了什么。需要熟悉这门语言,我们就必需要了解C++里面的那些东西,需要我们去了解他后面的内存对象。这样我们才能真正的了解C++,从而能够更好的使用C++这门最难的编程语言。
在文章束之前还是介绍一下自己吧。我从事软件研发有十个年头了,目前是软件开发技术主管,技术方面,主攻Unix/C/C++,比较喜欢网络上的技术,比如分布式计算,网格计算,P2P,Ajax等一切和互联网相关的东西。管理方面比较擅长于团队建设,技术趋势分析,项目管理。欢迎大家和我交流,我的MSN和Email是:haoel@hotmail.com