We’ve got to change the code significantly to make this work. Note that all our efforts so far are not in vain, because we have good tests that we need to keep passing forever.

A simple way to make the spare test pass is this:

#include <stdexcept>
#include <array>

namespace Bowling
{
    class Game
    {
        std::array<int, 21> rolls;
        int currentRoll = 0;
    public:
        void Roll(int pins)
        {
            if (pins < 0 || pins > 10)
                throw std::out_of_range("'pins' is out of range");
            rolls[currentRoll++] = pins;
        }
        int Score() const
        {
            int score = 0;
            for(int i=0; i<21; ++i)
            {
                score += rolls[i];
                if (i%2 == 1)                        // it's a second roll
                    if (rolls[i-1] + rolls[i] == 10) // it's a spare!
                        score += rolls[i+1];         // add the bonus
            }
            return score;
        }
    };
}
#endif

So, we hang onto all the rolls in an std::array (of the maximum size, where we have a spare in the last frame), populated in the Roll method, but do all the scoring in the Score method. It works, but could use some refactoring. Whenever I see a comment, I take it as a hint that maybe I should extract an expressively named method. On the other hand, I don’t always extract a method if there isn’t duplication. Let’s hold off on that for a bit.

Now that line, if (i%2 == 1) will be wrong if we have a strike, but we don’t have a test for that yet. Even though the code is a little ugly, I’m going to hold off refactoring.

Let’s write some more spare tests, like nothing but spares. If you rolls 21 fives in a row, they’re all spares and the score will be 150.

Let’s write that test and then click Next