使用ctypes在Python中调用C++动态库

使用ctypes在Python中调用C++动态库

入门操作

使用ctypes库可以直接调用C语言编写的动态库，而如果是调用C++编写的动态库，需要使用extern关键字对动态库的函数进行声明：

#include <iostream>
using namespace std;

extern "C" {
    void greet() {
        cout << "hello python" << endl;
    }
}

将上述的C++程序编译成动态链接库：

g++ hello.cpp -fPIC -shared -o hello.so

在Python代码中使用ctypes导入动态库，调用函数：

# -*- coding: utf-8 -*- #
from ctypes import CDLL

hello = CDLL('./hello.so')

if __name__ == '__main__':
    hello.greet()

运行上述Python程序：

xujijun@pc:~/codespace/python$ python3 hello.py
hello python

参数传递

编写一个整数加法函数

#include <iostream>
using namespace std;

extern "C" {
    int add(int a, int b) {
        return a + b;
    }
}

编译得到动态库，在Python代码中调用：

# -*- coding: utf-8 -*- #
from ctypes import CDLL

hello = CDLL('./hello.so')

if __name__ == '__main__':
    a = input('input num1: ')
    b = input('input num2: ')
    print('output: %d' % hello.add(int(a), int(b)))

运行上述代码，得到输出：

xujijun@pc:~/codespace/python$ python3 hello.py
input num1: 12
input num2: 34
output: 46

尝试传递字符串参数

#include <iostream>
#include <cstdio>
using namespace std;

extern "C" {
    void print_name(const char* name) {
        printf("%s\n", name);
    }
}

Python代码调用：

# -*- coding: utf-8 -*- #
from ctypes import CDLL

hello = CDLL('./hello.so')

if __name__ == '__main__':
    name = input('input name: ')
    hello.print_name(name.encode('utf-8'))  # 此处需要将Python中的字符串按照utf8编码成bytes

查看输出：

xujijun@pc:~/codespace/python$ python3 hello.py
input name: yanhewu
yanhewu

面向对象

使用C++编写动态链接库我们肯定会想到如何在Python中调用C++的类，由于ctypes只能调用C语言函数（C++中采用extern "C"声明的函数），我们需要对接口进行一定的处理：

C++代码示例

#include <iostream>
#include <cstdio>
#include <string>
using namespace std;

class Student {
private:
    string name;
public:
    Student(const char* n);
    void PrintName();
};

Student::Student(const char* n) {
    this->name.assign(n);
}

void Student::PrintName() {
    cout << "My name is " << this->name << endl;
}

extern "C" {
    Student* new_student(const char* name) {
        return new Student(name);
    }

    void print_student_name(Student* stu) {
        stu->PrintName();
    }
}

Python代码中调用：

# -*- coding: utf-8 -*- #
from ctypes import CDLL

hello = CDLL('./hello.so')

class Student(object):
    def __init__(self, name):
        self.stu = hello.new_student(name.encode('utf-8'))

    def print_name(self):
        hello.print_student_name(self.stu)

if __name__ == '__main__':
    name = input('input student name: ')
    s = Student(name)
    s.print_name()

输出：

xujijun@pc:~/codespace/python$ python3 hello.py
input student name: yanhewu
My name is yanhewu

内存泄漏？

上一部分我们我们尝试了如何使用ctypes调用带有类的C++动态库，这里我们不禁会想到一个问题，我们在动态库中使用new动态申请的内存是否会被Python的GC清理呢？这里我们完全可以猜想，C++动态库中的动态内存并不是使用Python中的内存申请机制申请的，Python不应该对这部分内存进行GC，如果真的是这样，C++的动态库就会出现内存泄漏的问题了。那事实是不是这样呢？我们可以使用内存检查工具Valgrind来检查上面的Python代码：

命令：

valgrind python3 hello.py

最终结果输出：

==17940== HEAP SUMMARY:
==17940==     in use at exit: 647,194 bytes in 631 blocks
==17940==   total heap usage: 8,914 allocs, 8,283 frees, 5,319,963 bytes allocated
==17940==
==17940== LEAK SUMMARY:
==17940==    definitely lost: 32 bytes in 1 blocks
==17940==    indirectly lost: 0 bytes in 0 blocks
==17940==      possibly lost: 4,008 bytes in 7 blocks
==17940==    still reachable: 643,154 bytes in 623 blocks
==17940==         suppressed: 0 bytes in 0 blocks
==17940== Rerun with --leak-check=full to see details of leaked memory
==17940==
==17940== For counts of detected and suppressed errors, rerun with: -v
==17940== Use --track-origins=yes to see where uninitialised values come from
==17940== ERROR SUMMARY: 795 errors from 86 contexts (suppressed: 0 from 0)

可以看到，definitely lost了32字节，确实出现了内存泄漏，但是是不是动态库的问题我们还要进一步验证:

C++代码加入析构函数定义：

#include <iostream>
#include <cstdio>
#include <string>
using namespace std;

class Student {
private:
    string name;
public:
    Student(const char* n);
    ~Student();
    void PrintName();
};

Student::Student(const char* n) {
    this->name.assign(n);
}

Student::~Student() {
    cout << "Student's destructor called" << endl;
}

void Student::PrintName() {
    cout << "My name is " << this->name << endl;
}

extern "C" {
    Student* new_student(const char* name) {
        return new Student(name);
    }

    void print_student_name(Student* stu) {
        stu->PrintName();
    }
}

运行相同的Python代码：

xujijun@pc:~/codespace/python$ python3 hello.py
input student name: yanhewu
My name is yanhewu

从输出可以看到，Student的析构函数并没有被调用。这里可以确定，Python的GC并没有对动态库中申请的内存进行处理，也确实不能进行处理（毕竟不是Python环境下申请的内存，在C++动态库中可能会先释放这部分内存，如果GC再次释放就会出现内存问题）。但是内存泄漏的问题还是需要解决的，可以参照以下做法：

C++代码中添加内存释放接口：

#include <iostream>
#include <cstdio>
#include <string>
using namespace std;

class Student {
private:
    string name;
public:
    Student(const char* n);
    ~Student();
    void PrintName();
};

Student::Student(const char* n) {
    this->name.assign(n);
}

Student::~Student() {
    cout << "Student's destructor called" << endl;
}

void Student::PrintName() {
    cout << "My name is " << this->name << endl;
}

extern "C" {
    Student* new_student(const char* name) {
        return new Student(name);
    }

    // 释放对象内存函数
    void del_student(Student* stu) {
        delete stu;
    }

    void print_student_name(Student* stu) {
        stu->PrintName();
    }
}

Python代码中，在Student类中调用内存释放函数：

# -*- coding: utf-8 -*- #
from ctypes import CDLL

hello = CDLL('./hello.so')

class Student(object):
    def __init__(self, name):
        self.stu = hello.new_student(name.encode('utf-8'))

    def __del__(self):
        # Python的对象在被GC时调用__del__函数
        hello.del_student(self.stu)

    def print_name(self):
        hello.print_student_name(self.stu)

if __name__ == '__main__':
    name = input('input student name: ')
    s = Student(name)
    s.print_name()

运行Python代码：

xujijun@pc:~/codespace/python$ python3 hello.py
input student name: yanhewu
My name is yanhewu
Student's destructor called

可以看到，C++动态库中的Student类的析构函数被调用了。再次使用Valgrind检查内存使用情况：

==23780== HEAP SUMMARY:
==23780==     in use at exit: 647,162 bytes in 630 blocks
==23780==   total heap usage: 8,910 allocs, 8,280 frees, 5,317,023 bytes allocated
==23780==
==23780== LEAK SUMMARY:
==23780==    definitely lost: 0 bytes in 0 blocks
==23780==    indirectly lost: 0 bytes in 0 blocks
==23780==      possibly lost: 4,008 bytes in 7 blocks
==23780==    still reachable: 643,154 bytes in 623 blocks
==23780==         suppressed: 0 bytes in 0 blocks
==23780== Rerun with --leak-check=full to see details of leaked memory
==23780==
==23780== For counts of detected and suppressed errors, rerun with: -v
==23780== Use --track-origins=yes to see where uninitialised values come from
==23780== ERROR SUMMARY: 793 errors from 87 contexts (suppressed: 0 from 0)

可以看到，definitely lost已经变为0，可以确定动态库申请的内存被成功释放。