Continue in Range Based for Loop

The range-based for loop changed in C++17 to allow the begin and end expressions to be of different types. And in C++20, an init-statement is introduced for initializing the variables in the loop-scope.

Overview

The range-based for loop (or range-for in short), along with auto, is one of the most significant features added in the C++11 standard. These are some of the typical usages of range-based for loop:

          // Iterate over STL container std::vector<int> v{1,2,3,4}; for(const auto& i : v)   std::cout << i << "\n";  // Over std::string std::string s{"1234"}; for(auto c : s)   std::cout << c << "\n";  // over an array int a[4]{1,2,3,4}; for(auto& i : a)   std::cout << i << "\n";  // over a brace-init-list (std::initializer_list) for(auto& i : {1,2,3,4})   std::cout << i << "\n";                  

This article, inspired by the cppreference page, explains the internal functioning of the range-for loop. We will start by outlining how C++11/C++14 range-for works, and then briefly describe the changes made to it in C++17 and C++20 in later sections.

Range-for in C++11/C++14

The range-based for loop has the following format:

          for (range_declaration : range_expression) {   /*loop body*/  }                  

In C++11/C++14, the above format results in a code similar to the following:

          /* modified code from cppreference */ {     auto&& range = range_expression ;    for (auto b = beginExpr, e = endExpr; b != e; ++b) {      range_declaration = *b;      /*loop body*/   }  }                  

These are the focal points regarding the above implementation:

The Universal Reference:

The range is a universal reference (because it is declared as auto&&). A universal reference, also known as a forwarding reference, can bind to either an lvalue or an rvalue expression. This suggests the range_expression can be anything including but not limited to - a variable, a const reference, or a function call that returns a temporary.

The Nesting Block:

The entire implementation is nested within a block ({}) statement. If the range_expression returns a temporary, the temporary's lifetime is extended until the end of the loop by the enclosing block.

Types of beginExpr and endExpr:

The beginExpr and the endExpr are of the same type, and they are resolved depending upon the range_expression as follows:

1. If the range_expression is an array of N elements, the beginExpr is range and the endExpr is range+N.

2. If the range_expression is a class with members begin and end, the beginExpr is range.begin() and endExpr is range.end(). All the STL containers (e.g., std::vector and std::map ) have begin and end methods that return the iterators. Note that, if the begin and end members are not functions returning an iterator (or a pointer), this results in a compilation error.

3. If none of the above, the beginExpr is begin(range) and endExpr is end(range). Note that here, the begin and end are unqualified function names (e.g., begin instead of std::begin), and they are resolved using Argument Dependent Lookup (ADL). In a nutshell, ADL means, the compiler performs the lookup for an unqualified function name in its arguments' namespaces also.

A Custom range-for Iterable

Let's take an example of a custom range-for iterable type to get everything together. Consider a minimal null-terminated custom string class, FixedString, that can only store a fixed number of chars. The FixedString class also has an inner type Iterator and - begin() and end() - methods so it can be used in a range-for loop:

          template<size_t S> class FixedString { public:  FixedString() = default;  FixedString(const char* str) {   if(str)    ::strncpy(str_, str, S);  }   const char* c_str() const { return str_; }  size_t count() const { return ::strlen(str_); }  const char& operator[](size_t i) const { return str_[i]; }    // default memberwise copies   // Minimum required for range-for loop  template<typename T>  struct Iterator {   T* p;   T& operator*() { return *p; }   bool operator != (const Iterator& rhs) {     return p != rhs.p;   }   void operator ++() { ++p; }  };    // auto return requires C++14  auto begin() const { // const version    return Iterator<const char>{str_};  }  auto end() const { // const version   return Iterator<const char>{str_+count()};  }  private:  char str_[S+1]{}; // '\0' everywhere };                  

Although a FixedString object can be assigned a new value, its elements cannot be modified. This helps keep things straightforward for our purpose here as we don't have to worry about defining the non-const begin and end methods. Note that instead of member functions begin/end, we could define free begin() and end() function templates in the same namespace to make FixedString iterable in the range-for loop:

          template<size_t Size> const char* begin(const FixedString<Size>& fs) {  return fs.c_str(); }  template<size_t Size> const char* end(const FixedString<Size>& fs) {  return fs.c_str() + fs.count(); }                  

We have chosen the member functions way because that helps us explain the C++17 changes in the next section. The FixedString can be used in a range-based loop as follows:

          FixedString<12> fs("hello world"); // print all chars for(auto& c : fs)     std::cout << c << "\n";                  

The range-based for loop has gone over some changes since C++11/C++14. The first change was made in C++17 to allow a range_expression's end to be of a different type than its begin. The second and most recent change, which is from the C++20 standard, adds an optional init-statement for initializing the variables in the loop-scope. We talk about these changes in the next two sections.

The C++17 Version

Until C++14, the beginExpr and endExpr had to be of the same type, hence constraining the end-ness of the range_expression. Observing the range-for structure shows that the endExpr only needs to be equality comparable to beginExpr. The C++17 standard lifts this restriction on endExpr to be of the same type as beginExpr. Thus, since C++17, the endExpr can also be a sentinel integer value( e.g., a null byte) or even a predicate. The C++17 composition of the range-for loop is:

          {   auto&& range = range_expression ;   auto  b = beginExpr ;   auto  e = endExpr ;   for ( ; b != e; ++b) {    range_declaration = *b;    /*loop body*/   } }                  

So how can we take advantage of this evolution for FixedString? We can modify the FixedString to have an end() method that returns a sentinel null char (\0), instead of an end() method that returns an iterator. The FixedString::Iterator can be made comparable to a sentinel char value instead of its own type:

          template<size_t S> class FixedString { public:    .......  template<typename T>  struct Iterator {   T* p;   T& operator*() { return *p; }   // compare with the sentinel byte   bool operator != (char rhs) {       return *p != rhs;   }   void operator ++() { ++p; }  };   auto begin() const { return Iterator<const char>{str_}; }  /*  end method.       Returns sentinel byte. */  auto end() const { return '\0'; }  .... };                  

Please check out "An Iterable's End May Have a Different Type Than Its Begin" for more details on this C++17 change.

The C++20 Version

The range-for loop since C++20 has the following format:

          for (init-statement(optional) range_declaration : range_expression)  {     /* loop body */ }                  

To understand the motivation behind adding init-statement, let's consider a range_expression that returns a temporary:

          std::string foo() {  return "This is a test string"; }  for(auto& c : foo())  // OK with temporary      std::cout << c << "\n";                  

That loop works fine because the lifetime of the temporary is extended until the end of the loop. However, if the range_expression is changed to have a temporary within it, the results are undefined:

          class A { public:  A(const char* s):str(s){}  std::string& foo() { return str; }  private:  std::string str;  };  for(auto& c : A("Hello World").foo())  //!! Undefined behavior     std::cout << c << "\n";                  

A few more examples of a temporary within a range_expression that can be very tough to spot:

          std::vector<std::vector<int>> foo(); for(auto i : foo().front()) //The parent vector will be destroyed!!    std::cout << i;          std::shared_ptr<std::string> foo();  for(auto c : *foo()) //The shared_ptr will be destroyed!!    std::cout << c;                  

All the above cases can be resolved by introducing a variable before the loop to remove the intermediate temporary, e.g.:

          A a("Hello World");  for(auto& c : a.foo())     std::cout << c << "\n";                  

Or, we can use the C++20's init-statement as an alternative to create a variable in the loop-scope:

          for(A a("Hello World"); auto& c : a.foo())     std::cout << c << "\n";                  

Clearly, the C++20 init-statement offers an elegant way to initialize a local scope variable for a range_expression.

Further Reading

Range-based for loop: cppreference

Universal References in C++11: Scott Meyers

Argument-dependent lookup: cppreference

How the new range-based for loop in C++17 helps Ranges TS?: stackoverflow

Range-based for statements with initializer: open-std

Why use non-member begin and end functions in C++11? - stackoverflow

jurgensseir1936.blogspot.com

Source: https://www.nextptr.com/tutorial/ta1208652092/how-cplusplus-rangebased-for-loop-works

Share this post