valgrind 和 std::unique_ptr --- 是否误报？

Question

我正在尝试了解现代 C++ 中智能指针的使用，并且我编写了一个小而简单的程序来在

valgrind

中进行测试。问题在于下面的例子：

#include <iostream>
#include <memory>

class Base {
    private:
        virtual double meth_1( double x ) const = 0;
        virtual void meth_2( int y ) const = 0;
    protected:
        Base()
        {
            std::cout << "ctor of base for: " << this << std::endl;
        }
    public:
        virtual ~Base()
        {
            std::cout << "dtor of base for: " << this << std::endl;
        }
        double IMeth_1( double x ) const
        {
            return meth_1(x);
        }
        void IMeth_2( int y ) const
        {
            meth_2(y);
        }
};

class Derived_1 : public Base {
    private:
        double meth_1( double x ) const final
        {
            return x + 5.0;
        }
        void meth_2( int y ) const final
        {
            std::cout << (y + 5) << std::endl;
        }
    public:
        Derived_1() : Base()
        {
            std::cout << "ctor of Derived_1: " << this << std::endl;
        }
        ~Derived_1()
        {
            std::cout << "dtor of Derived_1: " << this << std::endl;
        }
};

class Derived_2 : public Base {
    private:
        double meth_1( double x ) const final
        {
            return x + 10.0;
        }
        void meth_2( int y ) const final
        {
            std::cout << (y + 10) << std::endl;
        }
    public:
        Derived_2() : Base()
        {
            std::cout << "ctor of Derived_2: " << this << std::endl;
        }
        ~Derived_2()
        {
            std::cout << "dtor of Derived_2: " << this << std::endl;
        }
};

void Fun( const Base& crBase )
{
    crBase.IMeth_2( 5 );
}

int main( int argc, char* argv[] ) {
    std::unique_ptr< Base > upBase;

    for ( std::size_t idx = 0ul; idx < 2ul; idx++ ) {
        upBase      = std::make_unique< Derived_1 >();
        std::cout   << upBase->IMeth_1( idx )   << std::endl;
        upBase->IMeth_2( idx );
        std::cout   << "----------"             << std::endl;
    }

    for ( std::size_t idx = 0ul; idx < 2ul; idx++ ) {
        upBase      = std::make_unique< Derived_2 >();
        std::cout   << upBase->IMeth_1( idx )   << std::endl;
        upBase->IMeth_2( idx );
        std::cout   << "----------"             << std::endl;
    }

    upBase = std::make_unique< Derived_1 >();
    Fun( *upBase );

    return 0;
}

使用

valgrind --leak-check=full --show-leak-kinds=all <prog_name>

运行时会出现内存泄漏：

==32350== HEAP SUMMARY:
==32350==     in use at exit: 72,704 bytes in 1 blocks
==32350==   total heap usage: 6 allocs, 5 frees, 72,744 bytes allocated
==32350== 
==32350== 72,704 bytes in 1 blocks are still reachable in loss record 1 of 1
==32350==    at 0x4C28C10: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==32350==    by 0x4EBE1EF: pool (eh_alloc.cc:117)
==32350==    by 0x4EBE1EF: __static_initialization_and_destruction_0 (eh_alloc.cc:244)
==32350==    by 0x4EBE1EF: _GLOBAL__sub_I_eh_alloc.cc (eh_alloc.cc:307)
==32350==    by 0x400F279: call_init.part.0 (in /usr/lib/ld-2.22.so)
==32350==    by 0x400F38A: _dl_init (in /usr/lib/ld-2.22.so)
==32350==    by 0x4000DB9: ??? (in /usr/lib/ld-2.22.so)
==32350== 
==32350== LEAK SUMMARY:
==32350==    definitely lost: 0 bytes in 0 blocks
==32350==    indirectly lost: 0 bytes in 0 blocks
==32350==      possibly lost: 0 bytes in 0 blocks
==32350==    still reachable: 72,704 bytes in 1 blocks
==32350==         suppressed: 0 bytes in 0 blocks
==32350== 
==32350== For counts of detected and suppressed errors, rerun with: -v
==32350== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

退出时仍在使用的 72,704 字节块是否误报，或者我是否滥用了智能指针？我假设我没有进行任何类型的切片，因为每次删除对象时都会调用基本 dtor。

很抱歉，如果这是一个愚蠢的问题，但我在 SO 中找不到任何 valgrind/误报/unique_ptr 相关的主题。此外，我不知道

unique_ptr

中可能创建了任何与

shared_ptr

中类似的附加块来跟踪该对象。

编辑： 不是 Valgrind 检测到的 Still Reachable Leak 的重复，因为在我的例子中，我没有使用线程（并且 valgrind

 已知会在 OpenMPI 环境中给出误报）。而且，在另一个问题中，通过对提供的代码进行适当修改就解决了该问题。尽管如此，对于什么叫做“真正的内存泄漏”还没有达成共识——也就是说，“退出时仍在使用”的“可到达块”是否应该被视为“内存泄漏”。

这不是 valgrind 错误。这是 libstdc++ 的特定功能，即 在 http://gcc.gnu.org/viewcvs/gcc?view=revision&revision=219988中引入。 如果您查看代码，您会发现 libstdc++-v3/libsupc++/eh_alloc.cc 中的类池没有析构函数，因为它是一个紧急内存池，旨在在应用程序的整个运行时期间保留。

即使是最小的程序也会显示问题：

Answer 1

 ~ % echo "int main () {}" | g++ -x c++ -
 ~ % valgrind --leak-check=full --show-leak-kinds=all ./a.out
==502== Memcheck, a memory error detector
==502== Copyright (C) 2002-2015, and GNU GPL'd, by Julian Seward et al.
==502== Using Valgrind-3.11.0 and LibVEX; rerun with -h for copyright info
==502== Command: ./a.out
==502== 
==502== 
==502== HEAP SUMMARY:
==502==     in use at exit: 72,704 bytes in 1 blocks
==502==   total heap usage: 1 allocs, 0 frees, 72,704 bytes allocated
==502== 
==502== 72,704 bytes in 1 blocks are still reachable in loss record 1 of 1
==502==    at 0x402CC6F: malloc (vg_replace_malloc.c:299)
==502==    by 0x40F420F: _GLOBAL__sub_I_eh_alloc.cc (in /usr/lib64/gcc/x86_64-pc-linux-gnu/5.2.1/libstdc++.so.6.0.21)
==502==    by 0x4010AA4: call_init.part.0 (dl-init.c:72)
==502==    by 0x4010D44: call_init (dl-init.c:30)
==502==    by 0x4010D44: _dl_init (dl-init.c:120)
==502==    by 0x4000C79: ??? (in /lib64/ld-2.22.90.so)

此问题已于 2016 年修复。请参阅以下 bugzilla 项目

为 libstdc++ 添加 freeres 钩子

通过@octoloid 的回答中的最小示例，我得到了

==551523== Memcheck, a memory error detector
==551523== Copyright (C) 2002-2024, and GNU GPL'd, by Julian Seward et al.
==551523== Using Valgrind-3.25.0.GIT and LibVEX; rerun with -h for copyright info
==551523== Command: ./a.out
==551523== 
==551523== 
==551523== HEAP SUMMARY:
==551523==     in use at exit: 0 bytes in 0 blocks
==551523==   total heap usage: 1 allocs, 1 frees, 72,704 bytes allocated
==551523== 
==551523== All heap blocks were freed -- no leaks are possible
==551523== 
==551523== For lists of detected and suppressed errors, rerun with: -s
==551523== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

Answer 2