[英]Multiple inheritance and unique_ptr destruction
I have the classic (possible problematic) multiple inheritance diamond scheme. 我有经典(可能有问题的)多重继承钻石计划。
I want to have a std::vector
that can contain either C
or D
objects so I make it as std::vector<C>
which is D
's dad and it works fine. 我想有一个std::vector
,可以包含C
或D
对象,所以我把它作为std::vector<C>
这是D
的爸爸,它工作正常。
BUT when I use: std::vector<std::unique_ptr<C>>
then I have segmentation fault upon the destruction of the vector. 但是,当我使用: std::vector<std::unique_ptr<C>>
然后我在向量的破坏时有分段错误。
** glibc detected *** ./a.out: free(): invalid pointer: 0x0000000009948018***
Why is there a difference? 为什么会有区别? To me, even the first implementation is problematic. 对我来说,即使是第一次实施也是有问题的。
Code 码
#include <string>
#include <vector>
#include <memory>
class A
{
public:
A() = default;
};
class B : public virtual A
{
public:
B() = default;
};
class C : public virtual A
{
public:
C() = default;
};
class D : public B, public C
{
public:
D() = default;
};
int main()
{
{ // this crashes
std::vector<std::unique_ptr<C>> v;
std::unique_ptr<D> s1(new D());
v.push_back(std::move(s1));
std::unique_ptr<C> s2(new C());
v.push_back(std::move(s2));
}
{ // this is fine
std::vector<C> v;
D s1;
v.push_back(s1);
C s2;
v.push_back(s2);
}
return 0;
};
You should declare your destructors as virtual. 您应该将析构函数声明为虚拟。 Otherwise, if your class D
is deleted using a pointer to C
, then the ~C()
destructor will be called and essential parts of the cleanup will be missed. 否则,如果使用指向C
的指针删除类D
,则将调用~C()
析构函数,并且将错过清理的基本部分。
Also note that in your second part (not using the unique_ptr
) you do some object slicing. 另请注意,在第二部分(不使用unique_ptr
)中,您可以执行一些对象切片。 It means you are creating a copy of s1
of type D
into a new object of class C
, hence you can lose the extra information specific to D
. 这意味着您正在将类型D
的s1
副本创建到C
类的新对象中,因此您可能会丢失特定于D
的额外信息。
Here is the corrected code : 这是更正后的代码:
#include <string>
#include <vector>
#include <memory>
class A
{
public:
A() = default;
virtual ~A() {};
};
class B : public virtual A
{
public:
B() = default;
virtual ~B() {};
};
class C : public virtual A
{
public:
C() = default;
virtual ~C() {};
};
class D : public B, public C
{
public:
D() = default;
virtual ~D() {};
};
int main()
{
{ // this does not crashe anymore
std::vector<std::unique_ptr<C>> v;
std::unique_ptr<D> s1(new D());
v.push_back(std::move(s1));
std::unique_ptr<C> s2(new C());
v.push_back(std::move(s2));
}
{ // this is fine because you slice D into C, still that fine ?
std::vector<C> v;
D s1;
v.push_back(s1);
C s2;
v.push_back(s2);
}
return 0;
}
As stated in the comments, if you mark A destructor's as virtual, every derived class will also have a virtual destructor. 如评论中所述,如果将析构函数标记为虚拟,则每个派生类也将具有虚拟析构函数。 Writing it everywhere can make it more explicit. 到处写它可以使它更明确。 It is a matter of style. 这是一种风格问题。
Your "this is fine" example does slicing, the vector only contains instances of C
, that's why it 'works' but does not do what you expected. 你的“这很好”的例子是切片,向量只包含C
实例,这就是为什么它“有效”但却没有达到预期的效果。 The solution is as dkg points out is to use virtual dtors. 解决方案就像dkg指出的那样是使用虚拟dtors。
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.