C++ 设计篇之——PImpl 机制

源码仓库：https://github.com/yeshenyong/practice_cpp/tree/master/pimpl

什么是PImpl 机制

Pointer to implementation（PImpl ），通过将类的实现细节放在一个单独的类中，从其对象表示中删除它们，通过一个不透明的指针访问它们（cppreference 是这么说的）

通过一个私有的成员指针，将指针所指向的类的内部实现数据进行隐藏

class Demo {public:...private:DemoImp* imp_;}

为什么用PImpl 机制

个人拙见

  1. C++ 不像Java 后端型代码，能有行业定式的列目录名形成规范（controller、Dao等）
  2. 隐藏实现，降低耦合性和分离接口（隐藏类的具体实现）
  3. 通过编译期的封装（隐藏实现类的细节）

业界实现

  1. 优秀开源代码有实现

PImpl 实现

方法一

cook_cuisine.h

#pragma once#include #include #include //Pointer to impl ementationclass CookImpl;// 后厨class Cook {public:Cook(int, const std::vector<std::string>&);~Cook();std::vector<std::string> getMenu(); /* 获取菜单 */uint32_t getChefNum();/* 获取厨师数量 */private:CookImpl* impl_;};typedef std::shared_ptr<Cook> CookPtr;// 美妙的typedef 懒人工具

cook_cuisine.cc

#include "cook_cuisine.h"class CookImpl {public:CookImpl(uint32_t checf_num, const std::vector<std::string>& menu):checf_num_(checf_num), menu_(menu) {}std::vector<std::string> getMenu();uint32_t getChefNum();private:uint32_t checf_num_;std::vector<std::string> menu_;};std::vector<std::string> CookImpl::getMenu() {return menu_;}uint32_t CookImpl::getChefNum() {return checf_num_;}Cook::Cook(int chef_num, const std::vector<std::string>& menu) {impl_ = new CookImpl(chef_num, menu);}Cook::~Cook() {delete impl_;}std::vector<std::string> Cook::getMenu() {return impl_->getMenu();}uint32_t Cook::getChefNum() {return impl_->getChefNum();}

方法二

cook_cuisine.h

#pragma once#include #include #include #include "cook_cuisine_imp.h"// 后厨class Cook {public:Cook(int, const std::vector<std::string>&);~Cook();std::vector<std::string> getMenu(); /* 获取菜单 */uint32_t getChefNum();/* 获取厨师数量 */private:CookImplPtr impl_;};typedef std::shared_ptr<Cook> CookPtr;

cook_cuisine.cc

#include "cook_cuisine.h"Cook::Cook(int chef_num, const std::vector<std::string>& menu) {impl_.reset(new CookImpl(chef_num, menu));}Cook::~Cook() {}std::vector<std::string> Cook::getMenu() {return impl_->getMenu();}uint32_t Cook::getChefNum() {return impl_->getChefNum();}

cook_cuisine_imp.h

#pragma once#include #include #include class CookImpl {public:CookImpl(uint32_t checf_num, const std::vector<std::string>& menu):checf_num_(checf_num), menu_(menu) {}std::vector<std::string> getMenu();uint32_t getChefNum();private:uint32_t checf_num_;std::vector<std::string> menu_;};typedef std::shared_ptr<CookImpl> CookImplPtr;

cook_cusine_imp.cc

#include "cook_cuisine_imp.h"std::vector<std::string> CookImpl::getMenu() {return menu_;}uint32_t CookImpl::getChefNum() {return checf_num_;}

main.cc

#include "cook_cuisine.h"#include using namespace std;// Testing, 平时开发可千万别用这句int main() {int checf_num = 10;const std::vector<std::string> menus = { "Chicken", "Beef", "Noodle", "Milk" };CookPtr cook(new Cook(checf_num, menus));auto cook_menu = cook->getMenu();auto cook_checf_num = cook->getChefNum();cout << "======================Chinese Cook======================\n";cout << "============Checf: " << cook_checf_num << " people\n";cout << "==========Menu\n";for (size_t i = 0; i < cook_menu.size(); i++) {cout << "============" << i + 1 << " : " << cook_menu[i] << "\n";}return 0;}

CMakeLists.txt

mkdir buildcd buildcmake ..

PImpl 缺点

  1. 空间开销：每个类都需要额外的指针内存指向实现类
  2. 时间开销：每个类间接访问实现的时候多一个间接指针操作的开销
  3. 阅读开销：使用、阅读和调试上带来一些不便（不是啥问题）

总结

每种设计方法都有它的优点和缺点

PImpl 用一些内存空间和额外类的实现换取耦合性的下降，是可以接受的

但重点在：在性能/内存要求不敏感处，PImpl 技术才更优不错的发挥舞台

极端例子：

你不可能在斐波那契的实现中还加个PImpl 机制，多此一举