C++20: Pythonic with the Ranges Library

This is a cross-post from www.ModernesCpp.com.

Today, I start an experiment. I want to implement beloved functions in Python in C++ using the ranges library. I'm curious about how it goes.

You maybe know it. I'm also a Python trainer since 2004. Python has awesome functions and often Python is for me the threshold of how comfortable a programming language could be. Today, I want to implement the Python functions range and filter.

• range creates a list "containing an arithmetic progression of integers" (Pythons built-in help).
• filter applies a predicate to a sequence and returns those elements for which the predicate returns true. 

A sequence is a term in Python which stands for something iterable such as a list ([1, 2, 3]), a tuple ((1, 2, 3)), or a string ("123"). Instead of a list, I use a std::vector in C++. The functions filter stand for the functional style in Python.

Before I start with the range function, I have to make a few remarks.

1. I use in my examples the range-v3 library from Eric Niebler, which is the basis for the C++20 ranges. I showed in my previous post C++20: The Ranges Library, how to translate the ranges-v3 to the C++20 syntax.
2. The Python code is often shorter than the C++ Code for two reasons. First, I don't store the Python lists in a variable and second, I don't display the result.
3. I don't like religious wars about programming languages. The middle-ages are long gone. I will not react to these comments.

Let's start with the range function. The range function is a kind of building-block for creating integers.

range

In the following examples, I first show the python expression commented out and then the corresponding C++ call.

// range.cpp

#include <iostream>
#include <range/v3/all.hpp>
#include <vector>

std::vector<int> range(int begin, int end, int stepsize = 1) {
    std::vector<int> result{};
    if (begin < end) {                                     // (5)
        auto boundary = [end](int i){ return i < end; };
        for (int i: ranges::views::iota(begin) | ranges::views::stride(stepsize) 
                                               | ranges::views::take_while(boundary)) {
            result.push_back(i);
        }
    }
    else {                                                 // (6)
        begin++;
        end++;
        stepsize *= -1;
        auto boundary = [begin](int i){ return i < begin; };
        for (int i: ranges::views::iota(end) | ranges::views::take_while(boundary) 
                                             | ranges::views::reverse 
                                             | ranges::views::stride(stepsize)) {
            result.push_back(i);
        }
    }
    return result;
}
        
int main() {
    
    std::cout << std::endl;

    // range(1, 50)                                       // (1)
    auto res = range(1, 50);
    for (auto i: res) std::cout << i << " ";
    
    std::cout << "\n\n";
    
    // range(1, 50, 5)                                    // (2)
    res = range(1, 50, 5);
    for (auto i: res) std::cout << i << " ";
    
    std::cout << "\n\n";
    
    // range(50, 10, -1)                                  // (3)
    res = range(50, 10, -1);
    for (auto i: res) std::cout << i << " ";
    
    std::cout << "\n\n";
    
    // range(50, 10, -5)                                  // (4)
    res = range(50, 10, -5);
    for (auto i: res) std::cout << i << " ";
    
    std::cout << "\n\n";
    
}

 The calls in the lines (1) - (4) should be quite easy to read when you look at the output.

The first two arguments of the range call stand for the beginning and end of the created integers. The beginning is included but not the end. step size as the third parameter is per default 1. When the interval [begin, end[ is decreasing, the step size should be negative. If not, you get an empty list or an empty std::vector<int>.

I cheat a little in my range implementation. I use the function ranges::views::stride that is not part of C++20. stride(n) returns the n-th element of the given range. If you know an elegant implementation based on C++20, please let me know.

The if condition (begin < end) of the range function in line (1) should be quite easy to read. Create all numbers starting with begin (ranges::views::iota(begin)), take each n-th element (ranges::views::stride(stepsize), and do it as long as the boundary condition holds (ranges::views::take_while(boundary). Finally, push the integers on the std::vector<int>.

In the else case (line 2), I use a little trick. I create the numbers [end++, begin++[, take them until the boundary condition is meet, reverse them (ranges::views::reverse), and take each n-th element.

I implement the eager version for filter and map (next post) in my examples. With Python 3 filter and map are lazy. filter and map return in this case generators. To get the eager behavior of Python 2 put a list around the filter and map calls in Python 3.

filter(lambda i: (i % 2) == 1 , range(1, 10))       # Python 2   

list(filter(lambda i: (i % 2) == 1, range(1, 10))) # Python 3

 Both calls produce the same list: [1, 3, 5, 7, 9].

I continue with the function filter because it is easier to implement such as the map function.

filter

// filter.cpp

#include "range.hpp"                          // (1)

#include <fstream>
#include <iostream>
#include <range/v3/all.hpp>
#include <sstream>
#include <string>
#include <vector>
#include <utility>

template <typename Func, typename Seq>       // (2)
auto filter(Func func, Seq seq) {
    
    typedef typename Seq::value_type value_type;

    std::vector<value_type> result{};
    for (auto i : seq | ranges::views::filter(func)) result.push_back(i);
    
    return result;
}


int main() {
    
    std::cout << std::endl;
    
    // filter(lambda i: (i % 3) == 0 , range(20, 50))     // (3)
    auto res = filter([](int i){ return (i % 3) == 0; }, range(20, 50) );
    for (auto v: res) std::cout << v << " ";
    
                                                          // (4) 
    // filter(lambda word: word[0].isupper(), ["Only", "for", "testing", "purpose"]) 
    std::vector<std::string> myStrings{"Only", "for", "testing", "purpose"};
    auto res2 = filter([](const std::string& s){ return static_cast<bool>(std::isupper(s[0])); }, myStrings);
     
    std::cout << "\n\n";
     
    for (auto word: res2) std::cout << word << std::endl;
    
    std::cout << std::endl;
                                                        
                                                          // (5)
    // len(filter(lambda line: line[0] == "#", open("/etc/services").readlines()))
    std::ifstream file("/etc/services", std::ios::in);
    std::vector lines;
    std::string line;
    while(std::getline(file, line)){
        lines.push_back(line);
    }
    std::vector<std::string> commentLines = filter([](const std::string& s){  return s[0] == '#'; }, lines);
    
    std::cout << "Comment lines: " << commentLines.size() << "\n\n";
                                          
}

 Before I explain the program let me show you the output.

This time, I include the range implementation from before. The filter function (line 2) should be easy to read. I just apply the callable func to each element of the sequence and materialize the elements in the std::vector. line (3) creates all numbers i from 20 to 50 for which hold (i % 3) == 0. Only the strings that start with an uppercase letter can pass the filter in line (4). Line (5) counts, how many lines in the file "/etc/services" are comments. Comments are lines which start with the '#' character.

If you ignore the different ways to implement lambdas in Python and in C++, the filter calls are quite similar.

What's next?

map was way more complicated to implement than filter. First, map may change the type of the input sequence. Second, my implementation of map triggered a GCC bug report. Afterward, I combine the functions map and filter in a function and I get ... . Read the details in my next post.