conceptC++

http://www.generic-programming.org/faq/?category=conceptcxx

Checking Concept Without Concepts in C++

By Anthony Williams, September 22, 2010

1 Comment

Get the benefits of C++0x but without the work

Anthony Williams is author of the book C++ Concurrency in Action and of the just::thread C++0x thread library. He can be contacted at anthony.ajw@gmail.com


"Concepts" was set to be one of the major selling points of the new C++0x standard, until it was removed from the draft in July 2009. The "Concepts" feature promised better compile-time checking of templates, and the ability to overload functions based on whether or not their parameters supported specific templates.

In this article I look at some techniques for obtaining some of the benefits of concepts, with the facilities we already have in C++. Source code that illustrates these techniques is listed at the end of this article.

Basic Tools

The basic tools we need for concept checking and concept-based overloading are static_assertenable_if, and a particular property of template instantiation dubbed SFINAE (short for "Substitution Failure Is Not An Error").static_assert is useful for the checking, enable_if for the overloading. SFINAE is the mechanism that makesenable_if work, and can be used to write additional checks for testing with static_assert or enable_if.

static_assert is a new C++0x language feature, and is available in the latest versions of some compilers, such as Microsoft Visual Studio 2010 and g++ 4.3 or later. It allows you to specify a boolean constant expression and an error message -- if the constant expression evaluates to False at compile time then the compilation fails and the specified error message is output. For example, this simple program compiled with gcc -std=c++0x:

    int main()
{
static_assert(false,"your message goes here");
}

yields this error message:

    test.cpp: In function 'int main()'
test.cpp:3: error: static assertion failed: "your message goes here"

The constant expression can of course depend on template parameters, and that's where the checking comes in: if you put a static_assert in a template then the compilation will fail if the specified condition is not True, and the error message will be a lot clearer than what you would get otherwise.

static_assert works really well with the standard type traits -- you can assert that a given type is convertible tobool, or is derived from a particular base class, or is a POD. Of course, you can also define your own traits for any characteristic that you can build a test for. Later in this article I'll show how you can build a test for the presence of particular member functions.

You could, for example enforce the constraint that a particular template parameter is a POD type, so you can copy instances with memcpy:

    template<typename T>
void copy(T const* source,T* dest,unsigned count)
{
static_assert(std::is_pod<T>::value,"T must be a POD");
memcpy(dest,source,count*sizeof(T));
}
is

If you try and use this copy function with a non-POD type such as std::string, then you will get a compilation error.

The Boost Concept Check Library provides an alternative to static_assert for checking for concept conformance. Just like static_assert, it generates compiler errors if the concept is not matched. However, it may provide an easier way of specifying the constraints than plain static_assert if concept-based overloading is not required.

Whereas static_assert is all about hard and fast requirements, enable_if is about choices. You use it to enable certain function overloads if and only if a given property is True. This enables you to specify different versions of an algorithm based on the properties of the template parameters.

For example, you could use enable_if to use memcpy when copying PODs, rather than copying each element individually:

    template<typename T>
typename std::enable_if<std::is_pod<T>::value,void>::type
copy(T const* source,T* dest,unsigned count)
{
memcpy(dest,source,count*sizeof(T));
} template<typename T>
typename std::enable_if<!std::is_pod<T>::value,void>::type
copy(T const* source,T* dest,unsigned count)
{
for(unsigned i=0;i<count;++i)
{
*dest++=*source++;
}
}

enable_if is quite simple in and of itself -- if the first template parameter (which must be a boolean constant expression) evaluates to true then the nested "type" member is a typedef to the second template parameter. If the the first parameter evaluates to False then there is no nested "type" member.

The SFINAE rules mean that if enable_if<some-expression,some-type>::type is used in the signature of a function template (as here), then that function overload is discarded if some-expression is False -- that overload is only enabled if some-expression is True.

SFINAEThe basic premise of SFINAE is that when the compiler is deducing the template parameters for a function template from a function call, if the deduced parameters would make the signature invalid then that function template is not considered for overload resolution rather than resulting in a compilation failure. There are limits, and some template instantiation errors will still cause compilation failure, but this is the basic principle that makes enable_if work.

For example, given the function template:

    template<typename T>
typename T::type foo(T t)
{}

If you try and call foo(3), then T is deduced to be int. The type int does not have a member called type. The instantiation is therefore invalid when substituting "int" as the template parameter. By the SFINAE rule, this is not an error: instead the overload is ignored. If there is another overload of foo that can match foo(3), then that will be chosen instead. Of course, if there are no other overloads or none of the others match, then you still get a compilation error.

SFINAE does not work if the required instantiation of the function template depends on the instantiation of another template and that instantiation fails; for example:

    template<typename T>
struct bar
{
typedef typename T::type type;
}; template<typename T>
typename bar<T>::type foo2(T t)
{
// ...
}

If you call foo2(3), then T is again deduced as int. However, bar<int> cannot be instantiated, since int does not have a member called type. This is a failure in the instantiation of bar<int> , not in the instantiation of foo2<int>, so is a real compiler error, and will not be ignored by SFINAE.


Availability of Features

Newer compilers (such as gcc 4.3 or later, and Microsoft Visual Studio 2010) are starting to provide C++0x features. static_assert is one of the most common C++0x language features added to compilers, but for those compilers that don't have static_assert, you can emulate it with BOOST_STATIC_ASSERT.

enable_if on the other hand is purely a library facility, and was part of the C++ Technical Report 1. As such it is even more widely available as part of the library supplied with compilers. For those compilers that don't provide their own version, it is also available as part of the Boost Library.

--A.W.

static_assert enable_if 模板编译期检查的更多相关文章

  1. 通过宏封装实现std::format编译期检查参数数量是否一致

    背景 std::format在传参数量少于格式串所需参数数量时,会抛出异常.而在大部分的应用场景下,参数数量不一致提供编译报错更加合适,可以促进我们更早发现问题并进行改正. 最终效果 // 测试输出接 ...

  2. 数值类型中JDk的编译期检查和编译期优化

    byte b1 = 5;//编译期检查,判断是否在byte范围内 byte b2 = 5+4;//编译期优化,相当于b2=9 byte b3 = 127;//编译通过,在byte范围内 byte b4 ...

  3. C++17尝鲜:编译期 if 语句

    Constexpr If(编译期 if 语句) 以 if constexpr 打头的 if 语句被称为 Constexpr If. Constexpr If 是C++17所引入的新的语法特性.它为C+ ...

  4. 简单的说一下:tarits技法就是一种模板元编程,起可以将本来处于运行期的事拉到编译期来做,增加了运行效率。 看以非模板元编程的例子,就是前面的那个例子:

    void adance(std::list<int>::iterator& iter, int d) { if(typeid(std::iterator_traits<std ...

  5. 读书笔记 effective c++ Item 41 理解隐式接口和编译期多态

    1. 显示接口和运行时多态 面向对象编程的世界围绕着显式接口和运行时多态.举个例子,考虑下面的类(无意义的类), class Widget { public: Widget(); virtual ~W ...

  6. c++ 编译期与运行期

    分享到 一键分享 QQ空间 新浪微博 百度云收藏 人人网 腾讯微博 百度相册 开心网 腾讯朋友 百度贴吧 豆瓣网 搜狐微博 百度新首页 QQ好友 和讯微博 更多... 百度分享 转自:http://h ...

  7. 《深入理解Java虚拟机》-----第10章 程序编译与代码优化-早期(编译期)优化

    概述 Java语言的“编译期”其实是一段“不确定”的操作过程,因为它可能是指一个前端编译器(其实叫“编译器的前端”更准确一些)把*.java文件转变成*.class文件的过程;也可能是指虚拟机的后端运 ...

  8. java编译期优化

    java语言的编译期其实是一段不确定的操作过程,因为它可以分为三类编译过程: 1.前端编译:把.java文件转变为.class文件 2.后端编译:把字节码转变为机器码 3.静态提前编译:直接把*.ja ...

  9. C++编译期多态与运行期多态

    前言 今日的C++不再是个单纯的"带类的C"语言,它已经发展成为一个多种次语言所组成的语言集合,其中泛型编程与基于它的STL是C++发展中最为出彩的那部分.在面向对象C++编程中, ...

随机推荐

  1. iPhone屏蔽IOS更新、iPhone系统更新的提示(免越狱,有效期更新至2021年)

    iPhone屏蔽IOS更新.iPhone系统更新的提示(免越狱,有效期更新至2021年) 1.在Safari浏览器中粘贴如下链接,按提示打开链接. 输入http://apt.dataage.pub 2 ...

  2. Spring框架第二天

    ## Spring框架第二天 ## ---------- **课程回顾:Spring框架第一天** 1. 概述 * IOC和AOP 2. 框架的IOC的入门 * 创建applicationContex ...

  3. unity UGUI UI跟随

    实现2dUI跟随游戏中角色的移动(应用于玩家名称,血条,称号) using UnityEngine; public class UI_Follow : MonoBehaviour { public C ...

  4. windows程序设计 MessageBox消息框

    MessageBox函数 int WINAPI MessageBoxW( HWND hWnd,//窗口句柄 LPCWSTR lpText,//消息框主体显示的字符串 LPCWSTR lpCaption ...

  5. 微信内置安卓x5浏览器请求超时自动重发问题处理小记

    X5内核  请求超时后会自动阻止请求返回并由代理服务器将原参数重新发送请求到服务层代码.但由于第一次请求已经请求到服务器,会导致出现重复下单.支付等重大问题. 该问题由于腾讯x5浏览器会自动阻止第一次 ...

  6. Substring (后缀数组 + 计数)

    题意:求出字符串中包含了某个字符的字符序列不一样的数量. 思路:其实主要的是找出每个被包含字符的数量,假设除了目标字符之外的所有字符都不一样,那么应该就很好求了,但是显然不可能,所以我们可以枚举每一个 ...

  7. 杨韬的Python/Jupyter学习笔记

    Python语法学习 https://zhuanlan.zhihu.com/p/24162430 Python 安装库 安装Jupyter Notebook 先安装Python cmd 进入K:\Ju ...

  8. ArrayList 除重

    看到一段简洁的 ArrayList 除重代码: protected final <T> List<T> removeDuplicates(List<T> list) ...

  9. 为什么一个java源文件中只能有一个public类

    问题:一个".java"源文件中是否可以包括多个类(不是内部类)?有什么限制? 答案:可以有多个类,但只能有一个public的类,并且public的类名必须与文件名相一致.一个文件 ...

  10. flask 操作数据时,db的要在app.config设置之后声明:如app.config['SQLALCHEMY_DATABASE_URI']

    flask 操作数据时,db的要在app.config设置之后声明:如app.config['SQLALCHEMY_DATABASE_URI'] 否则,运行程序时app.config里面做的设置就不会 ...