Partners: KDAB and Whole Tomato Software

18 February 2015

Non Static Data Members Initialization

My short summary for non static data members initialization from modern C++. A very useful feature. Should we use it or not?

Intro

Non-static data member initializers Paper N2756
Visual Studio Since VS 2013
GCC Since GCC 4.7
Intel Compiler Since version 14.0
Clang Since Clang 3.0

Previously you could only initialize static, integral and const members of a class. Now it is extended to support non static members that do not need to be const and may have any type.

Basic example

class SimpleType
{
private:
    int a { 1 };    // << wow!
    int b { 1 };    // << wow2!
    string name { "string" }; // wow3!

public:
    SimpleType() {
        cout << "SimpleType::ctor, {" 
                  << a << ", " 
                  << b << ", \"" 
                  << name << "\"}" 
                  << endl;
    }
    ~SimpleType() { 
        cout << "SimpleType::destructor" << endl; 
    }
};

If we create an object of type SimpleType:

SimpleType obj;

On the output we will get:

SimpleType::ctor, {1, 1, "string"}

All of member variables were properly initialized before our constructor was called. Note, that we did not initialize members in the constructor. Such approach is not only available for simple types like int, but also for a complicated type like std::string.

Why useful

  • Easier to write
  • You are sure that each member is properly initialized.
    • you cannot forget to initialize a member like when having a complicated constructor. Initialization and declaration are in one place - not separated.
  • Especially useful when we have several constructors.
    • Previously we would have to duplicate initialization code for members or write custom method like InitMembers() that would be called in constructors.
    • Now, you can do a default initialization and constructors will only do its specific jobs…

You can play with the exmple here

More details

Let’s now make some more advanced example:

SimpleType with a new constructor:

class SimpleType
{
private:
    int a { 1 };    // << wow!
    int b { 1 };    // << wow2!
    string name { "string" }; // wow3!

public:
    SimpleType() { /* old code... */ }
    SimpleType(int aa, int bb) 
        : a(aa), b(bb) // << custom init!
    {
    std::cout << "SimpleType::ctor(aa, bb), {"  
        << a << ", " 
        << b << ", \"" 
        << name << "\"}" 
        << std::endl;
    }
    ~SimpleType() { 
        cout << "SimpleType::destructor" << endl; 
    }
};

And AdvancedType:

class AdvancedType
{
private:
    SimpleType simple;

public:
    AdvancedType() {
        cout << "AdvancedType::ctor" << endl;
    }
    AdvancedType(int a) : simple(a, a) {
        cout << "AdvancedType::ctor(a)" << endl;
    }
    ~AdvancedType() { 
        cout << "AdvancedType::destructor" << endl; 
    }
};

So now, AdvancedType uses SimpleType as a member. And we have two constructors here.

If we write:

AdvancedType adv;

We will get:

SimpleType::ctor, {1, 1, "string"}
AdvancedType::ctor

SimpleType::ctor (default) was called before AdvancedType::ctor. Note that AdvancedType::ctor does nothing beside printing…

Then, if we write:

AdvancedType advObj2(10);

We will get:

SimpleType::ctor(aa, bb), {10, 10, "string"}
AdvancedType::ctor(a)

So this time, the second constructor of SimpleType was called.

Note: even if you have a default initialization for a member, you can easily overwrite it in a constructor. Only one initialization is performed.

As usual you can play with the code below:

Any negative sides?

The feature that we discuss, although looks nice and easy, has some drawbacks as well.
  • Performance: when you have performance critical data structures (for example a Vector3D class) you may want to have "empty" initialization code. You risk having uninitialized data members, but you will save several instructions.
  • Making class non-aggregate: I was not aware of this issue, but Shafik Yaghmour noted that in the comments below the article.
    • In C++11 spec did not allowed aggregate types to have such initialization, but in C++14 this requirement was removed.
    • Link to the StackOverflow question with details

Should you use it?

I do not think there are any serious drawbacks of using non static data members initialization. You should be aware of the negative sides (mentioned in the section above), but for something like 90% of cases it should be safe to use.

If your coding guideline contains a rule about initialization of every local variable in the code, then, in my opinion, non static data member initialization completes this approach.

BTW: If that puts any standard, this concept is not forbidden in Google C++ guide

Your turn

You can play with my basic code here: nonstatic_members_init.cpp

What do you think about Non static data member initialization?
Do you use it in your code?

Do you use non static data member initialization?
yes - at work & in private projects
yes - but only in private projects
no - this feature is evil!
Poll Maker

Comments

Thanks for all the comments on this site and also:!

Interested in new blog posts and bonus content? Sign up for my newsletter.

© 2017, Bartlomiej Filipek, Blogger platform
Any opinions expressed herein are in no way representative of those of my employers.
This site contains ads or referral links, which provide me with a commission. Thank you for your understanding.