Refactoring with C++ 11

Question

Given the new toolset provided by c++ lots of programmers, aiming at code simplification, expressiveness, efficiency, skim through their old code and make tweaks (some point

Answer 1

1. Prefer scoped enums to unscoped enums

   • In C++98 the enums, there is no scoped for enums like the following code snippet. The names of such enumerators belong to the scope containing enum, namely nothing else in that scope may have the same name.

     enum Color{ blue, green, yellow };
     bool blue = false;    // error: 'blue' redefinition
     

     However, in C++11, the scoped enums can fix this issue. scoped enum are declared var enum class.

     enum class Color{ blue, green, yellow };
     bool blue = false;     // fine, no other `blue` in scope
     Color cc = blue;       // error! no enumerator `blue` in this scope
     Color cc = Color::blue; // fine
     auto c = Color::blue;  // fine
     

   • The enumerators of scope enums are more strongly typed. But, the enumerators of unscoped enums implicitly convert to other types

     enum Color{ blue, green, yellow };
     std::vector<std::size_t> getVector(std::size_t x);
     Color c = blue;
     
     if (c < 10.1) {             // compare Color with double !! 
         auto vec = getVector(c); // could be fine !!
     }
     

     However, scoped enums will be failed in this case.

     enum class Color{ blue, green, yellow };
     std::vector<std::size_t> getVector(std::size_t x);
     Color c = Color::blue;
     
     if (c < 10.1) {             // error !
         auto vec = getVector(c); // error !!
     }
     

     Fix it through static_cast

     if (static_cast<double>(c) < 10.1) {
        auto vec = getVector(static_cast<std::size_t>(c));
     } 
     

   • unscoped enums may be forward-declared.

     enum Color;          // error!!
     enum class Color;    // fine
     

   • Both scoped and unscoped enums support specification of the underlying type. The default underlying type for scoped enums is int. Unscoped enums have no default underlying type.

2. Using Concurrency API

   • Prefer task-based to thread-based

     If you want to run a function doAsyncWork asynchronously, you have two basic choices. One is thread-based

     int doAsyncWork();
     std::thread t(doAsyncWork);
     

     The other is task-based.

     auto fut = std::async(doAsyncWork);
     

     Obviously, we can get the return value of doAsyncWork through task-based more easily than thread-based. With the task-based approach, it’s easy, because the future returned from std::async offers the get function. The get function is even more important if doAsyncWork emits an exception, because get provides access to that, too.

   • Thread-based calls for manual management of thread exhaustion, oversubscription, load balancing, and adaptation to new platforms. But Task-based via std::async with the default launch policy suffers from none of these drawbacks.

   Here are several links:

   Concurrency In C++

   C/C++ Programming Abstractions for Parallelism and Concurrency

Answer 2

1. Replacing rand

One of the big gains in C++11 has to be replacing the use of rand() with all the options available in the random header. Replacing rand() in many cases should be straight forward.

Stephan T. Lavavej probably made this point the strongest with his presentation rand() Considered Harmful. The examples show a uniform integer distribution from [0,10] using rand():

#include <cstdlib>
#include <iostream>
#include <ctime>

int main() 
{
    srand(time(0)) ;

    for (int n = 0; n < 10; ++n)
    {
            std::cout << (rand() / (RAND_MAX / (10 + 1) + 1)) << ", " ;
    }
    std::cout << std::endl ;
}

and using std::uniform_int_distrubution:

#include <iostream>
#include <random>

int main()
{
    std::random_device rd;

    std::mt19937 e2(rd());
    std::uniform_int_distribution<> dist(0, 10);

    for (int n = 0; n < 10; ++n) {
        std::cout << dist(e2) << ", " ;
    }
    std::cout << std::endl ;
}

Along with this should be moving from std::random_shuffle to std::shuffle which comes out of the effort to Deprecate rand and Friends. This was recently covered in the SO question Why are std::shuffle methods being deprecated in C++14?.

Note that the distributions are not guaranteed to be consistent across platforms.

2. Using std::to_string instead of std::ostringstream or sprintf

C++11 provides std::to_string which can be used to convert numerics to std::string it would produce the content as the equivalent std::sprintf. Most likely this would be used in place of either std::ostringstream or snprintf. This is more of a convenience, there is probably not much of performance difference and we can see from the Fast integer to string conversion in C++ article there are probably much faster alternatives available if performance is the main concern:

#include <iostream>
#include <sstream>
#include <string>

int main()
{
    std::ostringstream mystream;  
    mystream << 100 ;  
    std::string s = mystream.str();  

    std::cout << s << std::endl ;

    char buff[12] = {0};  
    sprintf(buff, "%d", 100);  
    std::string s2( buff ) ;
    std::cout << s2 << std::endl ;

    std::cout << std::to_string( 100 ) << std::endl ;
}

3. Using constexpr in place of template meta-programming

If you are dealing with literals there may be cases where using constexpr functions over template meta-programming may produce code that is more clear and possibly compiles faster. The article Want speed? Use constexpr meta-programming! provides an example of prime number determination using template meta-programming:

struct false_type 
{
  typedef false_type type;
  enum { value = 0 };
};

struct true_type 
{
  typedef true_type type;
  enum { value = 1 };
};

template<bool condition, class T, class U>
struct if_
{
  typedef U type;
};

template <class T, class U>
struct if_<true, T, U>
{
  typedef T type;
};

template<size_t N, size_t c> 
struct is_prime_impl
{ 
  typedef typename if_<(c*c > N),
                       true_type,
                       typename if_<(N % c == 0),
                                    false_type,
                                    is_prime_impl<N, c+1> >::type >::type type;
  enum { value = type::value };
};

template<size_t N> 
struct is_prime
{
  enum { value = is_prime_impl<N, 2>::type::value };
};

template <>
struct is_prime<0>
{
  enum { value = 0 };
};

template <>
struct is_prime<1>
{
  enum { value = 0 };
};

and using constexpr functions:

constexpr bool is_prime_recursive(size_t number, size_t c)
{
  return (c*c > number) ? true : 
           (number % c == 0) ? false : 
              is_prime_recursive(number, c+1);
}

constexpr bool is_prime_func(size_t number)
{
  return (number <= 1) ? false : is_prime_recursive(number, 2);
}

The constexpr version is much shorter, easier to understand and apparently performs much better than the template meta-programming implementation.

4. Using class member initialization to provide default values

As was recently covered in Has the new C++11 member initialization feature at declaration made initialization lists obsolete? class member initialization can be used to provide default values and can simplify cases where a class has multiple constructors.

Bjarne Stroustrup provides a good example in the C++11 FAQ, he says:

This saves a bit of typing, but the real benefits come in classes with multiple constructors. Often, all constructors use a common initializer for a member:

and provides an example of members which have a common initializer:

class A {
  public:
    A(): a(7), b(5), hash_algorithm("MD5"), s("Constructor run") {}
    A(int a_val) : a(a_val), b(5), hash_algorithm("MD5"), s("Constructor run") {}
    A(D d) : a(7), b(g(d)), hash_algorithm("MD5"), s("Constructor run") {}
    int a, b;
  private:
    HashingFunction hash_algorithm;  // Cryptographic hash to be applied to all A instances
    std::string s;                   // String indicating state in object lifecycle
};

and says:

The fact that hash_algorithm and s each has a single default is lost in the mess of code and could easily become a problem during maintenance. Instead, we can factor out the initialization of the data members:

class A {
  public:
    A(): a(7), b(5) {}
    A(int a_val) : a(a_val), b(5) {}
    A(D d) : a(7), b(g(d)) {}
    int a, b;
  private:
    HashingFunction hash_algorithm{"MD5"};  // Cryptographic hash to be applied to all A instances
    std::string s{"Constructor run"};       // String indicating state in object lifecycle
};

Note, that in C++11 a class using in class member initializers is no longer an aggregate although this restriction is removed in C++14.

5. Use fixed width integer types from cstdint instead of hand rolled typedefs

Since the C++11 standard uses C99 as a normative reference we get fixed width integer types, as well. For example:

int8_t
int16_t 
int32_t 
int64_t 
intptr_t

Although several of them an optional, for the exact width integer types the following from C99 section 7.18.1.1 applies:

These types are optional. However, if an implementation provides integer types with widths of 8, 16, 32, or 64 bits, no padding bits, and (for the signed types) that have a two’s complement representation, it shall define the corresponding typedef names.

Answer 3

Optimize simple mathematical functions with constexpr, especially if they are called inside inner loops. This would allow the compiler to calculate them at compilation saving you time

Example

constexpr int fibonacci(int i) {
    return i==0 ? 0 : (i==1 ? 1 : fibonacci(i-1) + fibonacci(i-2));
}

Another example is to use std::enable_if to limit the allowed template parameters types in a particular template function/class. This would make your code safer (in case you haven't use SFINAE to constrains the possible template arguments in your old code) when you implicit assume some property about the template types and it is just one extra line of code

example:

template
<
   typename T, 
   std::enable_if< std::is_abstract<T>::value == false, bool>::type = false // extra line
>
void f(T t) 
{ 
 // do something that depends on the fact that std::is_abstract<T>::value == false
}

Update 1: If you have a small array where the size is known at compile time and you you want to avoid the overhead of the heap allocation in std::vector (meaning: you want the array on the stack), you only choice in C++03 was to use c-style arrays. Change that to std::array. It is a simple change that provides you a lot of the functionally present in std::vector + stack allocation (much faster than heap allocation as I said before).

Answer 4

1. Changing std::map to std::unordered_map and std::set to std::unordered_set where ever order of container's elements is irrelevant, enhances significantly the performance.
2. Using std::map::at instead of using square bracket syntax insertion, when you want to avoid involuntary insertions.
3. Use alias templates when you want to typedef templates.
4. Use of initialization lists instead of for loops to initialize STL containers.
5. Replace fixed size C arrays with std::array.

Answer 5

I would add delegating constructors and in-class member initializers to the list.

Simplification By Using Delegating Constructors and In-Class Initialization

With C++03:

class A
{
  public:

    // The default constructor as well as the copy constructor need to 
    // initialize some of the members almost the same and call init() to
    // finish construction.
    A(double data) : id_(0), name_(), data_(data) {init();}
    A(A const& copy) : id_(0), name_(), data_(copy.data_) {init();}

    void init()
    {
       id_ = getNextID();
       name_ = getDefaultName();
    }

    int id_;
    string name_;
    double data_;
};

With C++11:

class A
{
  public:

    // With delegating constructor, the copy constructor can
    // reuse this constructor and avoid repetitive code.
    // In-line initialization takes care of initializing the members. 
    A(double data) : data_(data) {}

    A(A const& copy) : A(copy.data_) {}

    int id_ = getNextID();
    string name_ = getDefaultName();
    double data_;
};