Author Archive

ArchLinux 32

Most of my computers have x64 architecture, but one of them that I haven’t been using for a while is an old Asus Eee that runs on Intel Atom which is based on i686 architecture. When bringing it up to life, because I want to use it for some time now, I thought that it would be great update all packaged to latest version. I haven’t mention it yet but from the title anyone could deduce that this computer probably runs Arch Linux which is true. When I started updating packages I immediately run into problems because of messages similar to those:

error: GPGME error: No data
error: failed to update core (invalid or corrupted database (PGP signature))

After struggling for an hour I finally found two pages that solved my problems:

https://www.archlinux.org/news/the-end-of-i686-support/

https://archlinux32.org/

It seems that Arch Linux dropped support for i686 architecture but fortunately there’s still a community support.

Working on modular projects with Git

Last year in November there was a 3rd edition of the conference held in Wrocław called Code::Dive which was sponsored and organized by Nokia – company I work for. I’m proud to say that this time I was able to deliver one full time presentation on How to work on modular projects with Git using submodules and subtree and why you should rather use the latter. The presentation, among the others, is available on the conference’s YouTube channel.

I must say that this was the first time I gave a speech in public for such big audience. I give all the respect to other speakers since it’s not so easy. Nonetheless I’m pretty satisfied with the result.

There was also a second presentation I delivered with my colleague from Nokia – Łukasz Ziobroń. It was a quiz of knowledge about Modern C++.

Slides from all presentations can be found here (follow speaker’s name).

Slides from my presentation are here.

Range-based for loop vs std::transform

Some time ago, while doing some code review in work I came across the problem whether the loop in the code I was reviewing should be expressed with std::for_each, std::transform or range-based for loop. The initial code looked pretty much like this:

1
2
3
4
5
6
7
8
std::vector<Type> destination;
destination.reserve(source.size());
 
std::for_each(std::begin(source), std::end(source), 
              [&destination](auto const& p_value)
{
    destination.push_back(SomeStruct{p_value.a, p_value.b, p_value.c});
});

The first things that come to my mind when I see such code are “Why are you using std::for_each? Shouldn’t you use range-based for loop?” Or maybe there should be a std::transform?” Let’s consider those three things:

std::for_each

This is an algorithm from standard library that represents standard for loop that goes over a range defined by begin and end iterator and runs given functor or function. An example of usage is presented above. It utilizes lambda expression that invokes push_back on a std::vector to add elements.

range-based for

Given algorithm can be easily converted to this new form of loop (introduced in C++11).

1
2
3
4
5
6
7
std::vector<Type> destination;
destination.reserve(source.size());
 
for(auto const& p_value : source) 
{
    destination.push_back(SomeStruct{p_value.a, p_value.b, p_value.c});
};

This is the same algorithm but in much clearer and less noisy form.

std::transform

Another considered algorithm is std::transform that is supposed to be used (as the name says) to transform one structure to another. Let’s convert given code to utilize this algorithm.

1
2
3
4
5
6
7
8
std::vector<Type> destination;
destination.reserve(source.size());
 
std::transform(std::begin(source), std::end(source), std::back_inserter(destination),
              [](auto const& p_value)
{
    return SomeStruct{p_value.a, p_value.b, p_value.c};
});

It looks pretty much like std::for_each. The difference is in the name, the third argument and that the lambda returns new value instead of inserting in into the new container.

To be honest, when I have to choose between those versions I would choose std::transform because of its explicitness. This algorithm just says what is done. It gets one array and transforms it into another. std::for_each and range-based-for are just simple loops with no more special meaning than “run this function over this array”. From the last two if I had to choose one, then I would prefer range-based-for since its very clean form of loop over an array (no additional noise in form of lambda or std::begin/std::end).

But…

Since we are considering C++11 there’s one more thing to consider when inserting elements into std::vectoremplace_back. If we have a class with constructor and emplace_back then situation changes, because emplace_back allows us to create objects right inside the vector’s memory. In that case using std::transform doesn’t allow such operation because it requires copy (inside std::back_inserter).

Let’s take look at the code with all scenarios:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
#include <iostream>
#include <algorithm>
#include <vector>
 
struct TestStruct
{
    TestStruct(int a, std::string b)
        : m_a(a)
        , m_b(b)
    {
        std::cout << ">> Constructor: a=" << m_a << ", b=" << m_b << std::endl;
    }
 
    TestStruct(TestStruct const& src)
        : m_a(src.m_a)
        , m_b(src.m_b)
    {
        std::cout << ">> Copy Constructor: a=" << m_a << ", b=" << m_b << std::endl;
    }
 
    TestStruct(TestStruct && src)
        : m_a(src.m_a)
        , m_b(std::move(src.m_b))
    {
        std::cout << ">> Move Constructor: a=" << m_a << ", b=" << m_b << std::endl;
    }
 
    ~TestStruct()
    {
        std::cout << ">> Destructor: a=" << m_a << ", b=" << m_b << std::endl;
    }
 
private:
    int m_a;
    std::string m_b;
};
 
struct InputStruct
{
    int a;
    std::string b;
};
 
using namespace std;
 
int main(int argc, char *argv[])
{
    InputStruct input[] = {
        { 1, "one" },
        { 2, "two" },
    };
 
    std::cout << "std::for_each:" << std::endl;
 
    std::vector<TestStruct> va;
    va.reserve(std::end(input) - std::begin(input));
    std::for_each(std::begin(input), std::end(input),
                  [&va](auto const& p_in)
    {
        va.push_back(TestStruct{p_in.a, p_in.b});
    });
 
    std::cout << "\nrange-based-for:" << std::endl;
 
    std::vector<TestStruct> vb;
    va.reserve(std::end(input) - std::begin(input));
    for(auto const& p_in : input)
    {
        vb.push_back(TestStruct{p_in.a, p_in.b});
    }
 
    std::cout << "\nstd::transform:" << std::endl;
    std::vector<TestStruct> vc;
    vc.reserve(std::end(input) - std::begin(input));
    std::transform(std::begin(input), std::end(input), std::back_inserter(vc),
                  [](auto const& p_in)
    {
        return TestStruct{p_in.a, p_in.b};
    });
 
    std::cout << "\nrange-based-for + emplace_back:" << std::endl;
 
    std::vector<TestStruct> vd;
    vd.reserve(std::end(input) - std::begin(input));
    for(auto const& p_in : input)
    {
        vd.emplace_back(p_in.a, p_in.b);
    }
 
    std::cout << "\n\nDestruction of all items:" << std::endl;
}

The output is as follows:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
std::for_each:
>> Constructor: a=1, b=one
>> Move Constructor: a=1, b=one
>> Destructor: a=1, b=
>> Constructor: a=2, b=two
>> Move Constructor: a=2, b=two
>> Destructor: a=2, b=
 
range-based-for:
>> Constructor: a=1, b=one
>> Move Constructor: a=1, b=one
>> Destructor: a=1, b=
>> Constructor: a=2, b=two
>> Move Constructor: a=2, b=two
>> Destructor: a=2, b=
 
std::transform:
>> Constructor: a=1, b=one
>> Move Constructor: a=1, b=one
>> Destructor: a=1, b=
>> Constructor: a=2, b=two
>> Move Constructor: a=2, b=two
>> Destructor: a=2, b=
 
range-based-for + emplace_back:
>> Constructor: a=1, b=one
>> Constructor: a=2, b=two
 
 
Destruction of all items:
>> Destructor: a=1, b=one
>> Destructor: a=2, b=two
>> Destructor: a=1, b=one
>> Destructor: a=2, b=two
>> Destructor: a=1, b=one
>> Destructor: a=2, b=two
>> Destructor: a=1, b=one
>> Destructor: a=2, b=two

The output from the app reveals everything. The emplace_back version produces the smallest output, just as suspected, and all other loops does exactly the same thing.

What about POD?

Of course the previous version works smoothly only with classes that defines the constructor which can be used within emplace_back. When considering POD structs that doesn’t have one, the trick won’t work and won’t even compile. When adding POD object into vector it requires explicit use of the struct name when creating new objects that goes to the container.

In such case, in my opinion, the std::transform is the best way to express the intent.

Defined and not defined constructor in C++

C++ is a pretty interesting language because of different quirks that it has. One of such things is the fact that there is a difference between defined constructor, not defined constructor and deleted constructor, especially move constructor.

Consider following code:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
#include <iostream>
 
struct ExplicitMoveConstructor
{
    ExplicitMoveConstructor()
    {
        std::cout << "Default constructor" << std::endl;
    }
 
    ExplicitMoveConstructor(ExplicitMoveConstructor const& src)
    {
        std::cout << "Copy constructor" << std::endl;
    }
 
    ExplicitMoveConstructor(ExplicitMoveConstructor && src)
    {
        std::cout << "Move constructor" << std::endl;
    }
};
 
struct DeletedMoveConstructor
{
    DeletedMoveConstructor()
    {
        std::cout << "Default constructor" << std::endl;
    }
 
    DeletedMoveConstructor(DeletedMoveConstructor const& src)
    {
        std::cout << "Copy constructor" << std::endl;
    }
 
    DeletedMoveConstructor(DeletedMoveConstructor && src) = delete;
};
 
struct NotDefinedMoveConstructor
{
    NotDefinedMoveConstructor()
    {
        std::cout << "Default constructor" << std::endl;
    }
 
    NotDefinedMoveConstructor(NotDefinedMoveConstructor const& src)
    {
        std::cout << "Copy constructor" << std::endl;
    }
};
 
auto returnExplicitMoveConstructorClass()
{
    ExplicitMoveConstructor cm;
    return cm;
}
 
auto returnDeletedMoveConstructorClass()
{
    DeletedMoveConstructor cm;
    return cm;
}
 
auto returnNotDefinedMoveConstructorClass()
{
    NotDefinedMoveConstructor cm;
    return cm;
}
 
int main()
{
    auto first = returnExplicitMoveConstructorClass();
    auto second = returnDeletedMoveConstructorClass();
    auto third = returnNotDefinedMoveConstructorClass();
    return 0;
}

The presented code does not compile when trying to do so with following command:

clang++ constructor.cpp --std=c++14

The error code is as follows:

constructor.cpp:70:7: error: call to deleted constructor of 'DeletedMoveConstructor'
        auto second = returnDeletedMoveConstructorClass();
             ^        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
constructor.cpp:33:2: note: 'DeletedMoveConstructor' has been explicitly marked deleted here
        DeletedMoveConstructor(DeletedMoveConstructor && src) = delete;
        ^
1 error generated.

The error from the compiler says that the move constructor was called but it is marked as deleted. If compiled with g++ it’s pretty much the same.

Conclusions from this are two. The first one is that the deleted constructor is also a declared constructor so the compiler will use it as if it was defined by the user. The second one is that the value returned from function is returned by utilizing move semantic when move constructor is declared, even if it is deleted.

The explaination for this turns out to be quite simple. As written in this post on Stack Overflow, this is the intended behaviour of overload resolution. Since declared, the move constructor is the best match for value returned from function which is an r-value.

NAssist

This time I would like to publish my application that I wrote some time ago and I’ve been actively using. I call it NAssist. Purpose of this application is to provide an on demand window (accessed with global keyboard shortcut) that allows to enter some commands whose effects would be displayed inside it.

Basic application window

Basic application window

Application is written in Python 3 with PyQt5 and PyWin (for Windows) or python-xlib (for Linux). PyQt is used as an application framework while PyWin or python-xlib are needed for handling global hotkeys.

Usage

Global hotkeys are used to call application which is normally hidden in the background with icon in Notification Area or Tray (both on Windows and Linux). Application window can be called with predefined global hotkeys – CTRL + SHIFT + SPACE. On the other hand the window can be hidden with ESCAPE hotkey while focused.

Application in default mode appears only with one Command field which should be already focused by previously pressing mentioned hotkey combination (CTRL + SHIFT + SPACE). If it’s not, it can be made so by pressing them again. In this state application is ready to accept commands which are defined in __init__ function in mainwindow.py file. Those commands come from modules package.

Example of translation from English to Polish

Example of translation from English to Polish

Basic commands that come with the application are:

  • ap {text} – English -> Polish translator (google translate),
  • pa {text} – Polish -> English translator (google translate),
  • np {text} – German -> Polish translator (google translate),
  • pn {text} – Polish -> German translator (google translate),
  • sjp {word} – Polish language dictionary (sjp.pl).

To use one of those commands simply write them in Command field with arguments after space. The result of “ap I translate to Polish” is presented in the image on the right. As shown, the result appears right below the Command field, also the window is enlarged to make the result visible.

To quickly delete everything from Command field there is an additional hotkey – SHIFT + DELETE.

Writing own modules

Writing own modules is pretty straightforward. It requires a module to handle following events which are defined in BaseModule.py file:

  • onKeyPress(event) – key has been pressed inside application window, event represents PyQt event,
  • onTextUpdate(text) – argument text has been updated for given module (from command “ap text” only “text” will be passed,
  • onActivate() – current module has been activated,
  • onDeactivate() – current module has been deactivated (new module will be activated if command was changed, not removed),
  • onClose() – application is closing.

The basic commands are implemented using some more specialized base classes that are available for extension:

  • TranslatorModule – uses google translator to provide web page with results via “https://translate.google.com/#{}/{}/{}” pattern,
  • WebModule – is used to implement above module, can be used to add other webpages as commands.

Download and requirements

Application is already public on GitHub with GPL2 license.

Requirements for Windows:

Requirements for Linux:

  • Python 3
  • PyQt5 matching Python version
  • python3-xlib (pip install python3-xlib)

Machine Learning on Coursera

Recently I’ve finished Machine Learning course on Coursera and I think I can recommend it to anyone who wants to start doing something in this field. This area is not something new to me since my specialization on master’s degree was Intelligent Information Systems which covered this area pretty well. By doing this course I wanted to get another point of view on that. Now I can definitely say that prof. Andrew Ng does a good job explaining the material presented on the course. He starts with cost function, gradient descent which are the foundation of Machine Learning. Then he explains such topics as:

  • classification (with multiclass classifications),
  • logistic regression,
  • neural networks with backpropagation,
  • bias and variance,
  • SVM,
  • clustering,
  • anomaly detection,
  • photo OCR.

Everything is presented with real file examples. Students are required to write some code with the elements explained in each part in Matlab or Octave which are supported by scripts that simplify the way of submitting answers. If there is a one thing I could complain about it would be the out of date scripts that for newer version of Octaves require some patches which can be easily found on course’s wiki.

Variable Template in C++14

C++14 introduced new type of templates which are variable templates1. What it means is that it’s now possible to create template of a variable like this:

1
2
3
4
template<class Type>
constexpr Type PI = (Type)3.1415;
 
auto v = PI<double>;

This introduces a variable that can take different values for different template parameter. Main gain comes when such value is specialized like template:

1
2
3
4
template<>
constexpr int PI = 4;  // Because I can
 
auto v = PI<int>;

Of course in previous versions of standard this could be done in a slightly different way by using functions like this:

1
2
3
4
5
6
7
8
9
10
11
12
13
template<class Type>
constexpr Type PI()
{
   return (Type)3.1415;
}
 
template<>
constexpr int PI()
{
   return 4;  // Because I can
}
 
auto v = PI<double>();

The difference in usage is “()” which is used in a function call when using function template version. But there is also another difference which makes variable template unfortunately less useful than function template. Since C++11 it is possible to delete a specialization of function template but it seems it’s not possible to delete specialization of variable template:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
template<class Type>
constexpr Type PI()
{
   return (Type)3.1415;
}
 
template<>
constexpr int PI() = delete;
 
template<class Type>
constexpr Type ANSWER = (Type)42;
 
//template<>
//constexpr double ANSWER<double> = delete;   // not possible

But of course no one says that we cannot use both as a syntactic sugar:

1
2
template<class Type>
constexpr Type PI_VAR = PI<Type>();
  1. http://en.cppreference.com/w/cpp/language/variable_template []

C++ problems #1

Two days ago I was working on an old neural network project when some strange bug occured in my code. I was working on debug version of the project with no optimizations (-O0) and then I tried to switch them on to see what will be the gain in performance.

What happened is that it stopped working properly. With no optimizations it worked correctly, giving expected results but with -O2 flag it somehow stopped earlier than expected. After some debugging I’ve managed to find the place where the bug was hidden. This is the place:

1
2
3
4
size_t getOutputLayerSize() const
{
    m_forwardNetwork.getOutputLayerSize();
}

The bug here should be pretty obvious but when running through more code it may well hidden. The obvious thing is that there’s no return statement while the function is defined as returning an unsigned integer value. This code compiles because it may be a valid behaviour in some other cases like following:

1
2
3
4
size_t getOutputLayerSize() const
{
    throw std::runtime_error("");
}

But when nothing is thrown and there is no return statement this is an undefined behaviour and should be avoided.

Why it worked in my case with no optimizations? The m_forwardNetwork.getOutputLayerSize() function call returns an integer value that is saved in eax register. The same register is used in getOutputLayerSize() function to keep the return value which is normally saved by the return statement. With no return statement there is no action but since the value was written in m_forwardNetwork.getOutputLayerSize() it can be used by the caller of getOutputLayerSize() function so it happens to work correctly from caller POV. When optimizations are switched on, the line m_forwardNetwork.getOutputLayerSize() is removed by the compiler because it is not used so there is no function call which means eax register is not written to. In my case it happened that the overal function call return 0 value which wasn’t expected.

My fault was that I haven’t used any warning flags. By adding -Wall compiler starts to emit following warning:

warning: no return statement in function returning non-void [-Wreturn-type]

My advice is to always add -Wall to compilation flags so no error will be omitted. Another advice is to also add a -Werror flag which will force the compiler to fail the compilation if any warning occurs.

Deleting method in class hierarchy

C++11 has this new feature that allows programmer to remove:

  • automatically generated method or operator,
  • overload of free function or method (from implicit conversion)
  • specialization of template method.

This can be done by marking any function with = delete keyword and it works like this:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
void foo(unsigned a)
{}
void foo(int) = delete;                         // deleted function overload for implicit conversion to int
 
template<class T>
void bar(T a) 
{}
template<class T>
void bar<double>(double) = delete;              // deleted template function specialization for double type
 
struct Foo 
{
    Foo& operator=(Foo const&) = delete;        // deleted defaultly generated copying assignment operator
 
    void call(unsigned a) {}
    void call(int a) = delete;                  // deleted method overload for implicit conversion to int
};
 
int main()
{
    foo(1u);
    //foo(2);     // error: use of deleted function 'void foo(int)'
 
    bar(Foo());
    //bar(4.0);   // error: use of deleted function 'void bar(T) [with T = double]'
 
    Foo f;
    f.call(1u);
    //f.call(2);  // error: use of deleted function 'virtual void Foo::call(int)'
 
    Foo ff;
    //ff = f;     // error: use of deleted function 'Foo& Foo::operator=(const Foo&)'
}

It is pretty simple and useful. This allows to remove not wanted specialization from any interface, e.g. overloaded signed value if unsigned and only unsigned value is required. Let consider following class hierarchy:

1
2
3
4
5
6
7
8
struct Foo 
{
    virtual void call(unsigned a) {}
};
 
struct Bar : Foo
{
};

What if it is necessary to remove an overload of call method with int argument from interface of base class? Let see what will happen if different variants of call method are called from base class and derived class.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
struct Foo 
{
    void call(unsigned a) {}
    void call(int) = delete;
};
 
struct Bar : Foo
{
};
 
int main()
{
    Bar b;
    b.call(1u);
    //b.call(1);     // error: use of deleted function 'void Foo::call(int)'
    Foo & f = b;
    f.call(1u);
    //f.call(1);     // error: use of deleted function 'void Foo::call(int)'
}

It works fine. Bar class uses Foo class method declaration so it is not possible to call the call method from both Foo and Bar classes. But what if a programmer would like to have the call method virtual and overridden in derived class?

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
struct Foo 
{
    virtual void call(unsigned a) {}
    void call(int) = delete;
};
 
struct Bar : Foo
{
    void call(unsigned a) override {}
};
 
int main()
{
    Bar b;
    b.call(1u);
    b.call(1);       // works since it's not deleted in derived class and can be implicitely casted to unsigned
    Foo & f = b;
    f.call(1u);
    //f.call(1);     // error: use of deleted function 'void Foo::call(int)'
}

Now it is possible to call the call method from Bar class but not from Foo class. But what if ‘virtual’ is put before void call(int) = delete?

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
struct Foo 
{
    virtual void call(unsigned a) {}
    virtual void call(int) = delete;
};
 
struct Bar : Foo
{
    void call(unsigned a) override {}
};
 
int main()
{
    Bar b;
    b.call(1u);
    b.call(1);       // works since it's not deleted in derived class and can be implicitely casted to unsigned
    Foo & f = b;
    f.call(1u);
    //f.call(1);     // error: use of deleted function 'void Foo::call(int)'
}

It seems it does not matter if function is virtual or not. But there is another keyword that works with inheritence – final which means that virtual method marked with this keyword cannot be overriden in derived class.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
struct Foo 
{
    virtual void call(unsigned a) {}
    virtual void call(int) final = delete;
};
 
struct Bar : Foo
{
    void call(unsigned a) override {}
};
 
int main()
{
    Bar b;
    b.call(1u);
    b.call(1);       // works since it's not deleted in derived class and can be implicitely casted to unsigned
    Foo & f = b;
    f.call(1u);
    //f.call(1);     // error: use of deleted function 'void Foo::call(int)'
}

It does not matter either if method is final or not. So one and only option is just to declare the call method with int argument as deleted in derived class which should be a first choice of those considerations.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
struct Foo 
{
    virtual void call(unsigned a) {}
    virtual void call(int) final = delete;
};
 
struct Bar : Foo
{
    void call(unsigned a) override {}
    //void call(int) = delete;           // error: use of deleted function 'virtual void Bar::call(int)'
};
 
int main()
{
    Bar b;
    b.call(1u);
    b.call(1);       // works since it's not deleted in derived class and can be implicitely casted to unsigned
    Foo & f = b;
    f.call(1u);
    //f.call(1);     // error: use of deleted function 'void Foo::call(int)'
}

Ooops. Method void call(int) is marked as final in base class so it is not even possible to declare it as deleted in derived class. The one and only valid example that works as it supposed to from the beginning look like this:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
struct Foo 
{
    virtual void call(unsigned a) {}
    void call(int) = delete;
};
 
struct Bar : Foo
{
    void call(unsigned a) override {}    // If we want to override call()
    using Foo::call;                     // if we want to use Foo::call()
    void call(int) = delete;
};
 
int main()
{
    Bar b;
    b.call(1u);
    //b.call(1);     // error: use of deleted function 'void Bar::call(int)'
    Foo & f = b;
    f.call(1u);
    //f.call(1);     // error: use of deleted function 'void Foo::call(int)'
}

In conclusion it is not possible to declare the method overload as deleted in base class so that it will affect derived class if base method is marked as virtual and overriden in derived class. This concerns mostly the interfaces with methods that may accept some unsigned key value while there exists some signed key type, that should not be passed to considered methods.


All examples were compiled with GCC 5.2 with –std=c++14 flag.

Variadic arguments

Old C++ and C for a long time allow creating functions with unknown number of arguments. Most common example of such function is printf which prints text with formatted variables. It gets format as first argument and values to be printed in that format in other arguments. Following code shows its declaration:

int printf(const char * restrict format, ... );

Processing such argument list is done in runtime and it utilizes special macros or types from following list1:

  • va_start – enable access to variadic function arguments,
  • va_arg – accesses the next variadic function argument,
  • va_copy – makes a copy of the variadic function arguments,
  • va_end – ends traversal of the variadic function arguments,
  • va_list – holds the information needed by va_start, va_arg, va_end, and va_copy.

Following example presents simple usage of such function:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
#include <iostream>
#include <cstdarg>
 
double sum(int count, ...) 
{
    double sum = 0.0;
    va_list args;
    va_start(args, count);
    for (int i = 0; i < count; ++i) 
    {
        sum += va_arg(args, double);
    }
    va_end(args);
    return sum;
}
 
int main() 
{
    std::cout << sum(4, 25.0, 27.3, 26.9, 25.7);
}

Firstly va_start binds arguments passed to the function by elipsis (…) to variable args of type va_list. It takes instance of such type and number of arguments it want to access. Then all arguments are accessed subsequently by macro va_arg which takes list as an argument and returned type. At the end of processing arguments list should be unbound with macro va_end. To process same list of arguments more than once it can be copied to another list right after processing is started with macro va_copy (see2 for more).

Of course presented code is just an example. Real implementation of such function should utilize std::accumulate. Also this code shows some pitfalls of variadic arguments – number of arguments must be passed as one of firsts arguments and may not match the exact number of arguments passed to the function, also types can be different than expected in method.

  1. http://en.cppreference.com/w/cpp/language/variadic_arguments []
  2. http://en.cppreference.com/w/cpp/language/variadic_arguments []