int main(int argc, char* argv[])
{
    try
    {
        if (argc < 2)
        {
            std::cerr << "Usage: parallel_grep <string> <files...>\n";
            return 1;
        }

        // We use a fixed size pool of threads for reading the input files. The
        // number of threads is automatically determined based on the number of
        // CPUs available in the system.
        thread_pool pool;

        // To prevent the output from being garbled, we use a strand to synchronise
        // printing.
        strand<thread_pool::executor_type> output_strand(pool.get_executor());

        // Spawn a new coroutine for each file specified on the command line.
        std::string search_string = argv[1];
        for (int argn = 2; argn < argc; ++argn)
        {
            std::string input_file = argv[argn];
            spawn(pool,
                  [=](yield_context yield)
            {
                std::ifstream is(input_file.c_str());
                std::string line;
                std::size_t line_num = 0;
                while (std::getline(is, line))
                {
                    // If we find a match, send a message to the output.
                    if (line.find(search_string) != std::string::npos)
                    {
                        dispatch(output_strand,
                                 [=]
                        {
                            std::cout << input_file << ':' << line << std::endl;
                        });
                    }

                    // Every so often we yield control to another coroutine.
                    if (++line_num % 10 == 0)
                        post(yield);
                }
            });
        }

        // Join the thread pool to wait for all the spawned tasks to complete.
        pool.join();
    }
    catch (std::exception& e)
    {
        std::cerr << "Exception: " << e.what() << "\n";
    }

    return 0;
}
int main(int argc, char* argv[])
{
  try
  {
    if (argc < 2)
    {
      std::cerr << "Usage: parallel_grep <string> <files...>\n";
      return 1;
    }

    // We use a fixed size pool of threads for reading the input files. The
    // number of threads is automatically determined based on the number of
    // CPUs available in the system.
    thread_pool pool;

    // To prevent the output from being garbled, we use a strand to synchronise
    // printing.
    strand<thread_pool::executor_type> output_strand(pool.get_executor());

    // Spawn a new coroutine for each file specified on the command line.
    std::string search_string = argv[1];
    for (int argn = 2; argn < argc; ++argn)
    {
      std::string input_file = argv[argn];
      spawn(pool, boost::bind(&search_file,
            search_string, input_file, output_strand, _1));
    }

    // Join the thread pool to wait for all the spawned tasks to complete.
    pool.join();
  }
  catch (std::exception& e)
  {
    std::cerr << "Exception: " << e.what() << "\n";
  }

  return 0;
}