The Real Answer to the FizzBuzz Interview Question

A popular interview question is coding FizzBuzz. Now I’ve had similar experiences to what is shown at Coding Horror — I had a description for a trivial piece of code that would offer insight into the depth of engineering skill, with my careful posing of the requirements.

I had a sheet with this written down, so I would always say it in the same carefully-crafted way. I lost that some years ago, but here is the problem from memory:

Write a class whose constructor takes two integral arguments, and has members that return their sum and their difference.

Trivial enough? Not even a “problem” with logic to be considered; just a routine microtask that anyone with any fluency can express without having to think about it.

What I expected was to see how well the applicant processed the statement of requirements: making assumptions, or asking for clarification. I figure an experienced engineer would ask what was meant by integral, or use a typedef to abstract the decision of which integer type to use and specify it at only one point. More fluent coders would write it as a template.

I was very surprised to find that most applicants who got past the screening process thus far had trouble writing a C++ class at all! The best insight I had was from someone who explained that he usually uses the IDE and doesn’t know the details of syntax on how to write a class definition. Because of the reliance on what Visual Studio calls wizards, we dubbed these people Wizards’ Apprentices with a reference to the segment from Disney’s Fantasia. The story is simple: the young apprentice tries to use his master’s power but cannot control it; what we would recognize as a buggy program.

What Are You Showing Them?

Now suppose that you are fluent in C++ (or whichever language you are using) and would have no trouble jotting down a piece of code that does what you want.

I see many solutions to FizzBuzz, as well as many other problems, that are written in a way that is completely unlike what we see in “real” code. Someone who can effortlessly produce this might be in “scripting mode” which is used for single-use small programs — written quickly, sloppy, and highly specialized. It does not illustrate good software engineering practices, and the interviewer can’t tell if the applicant always (only) writes like that, or is capable of working on projects like they have: large bodies of code, need for maintainability, testability; follows good practices, etc.

So, I thought about FizzBuzz as if it were a small utility feature that was ordered for use in a large project.

Goodness

Certainly, a short piece of code can be written using up-to-date style and best practices as applicable for software development at the scale that is involved in real projects at work.

But I also thought about how much “good engineering” I can illustrate without making the result not-so-trivial. Here is what I came up with:

1. Separate the logic from I/O.
   Most solutions to simple problems like this mix console I/O with the logic, which is completely unlike real programs in multiple ways. It should not directly solicit input from the user as if he was a text file, or print output in the middle of the work; the “solution” should feature a function that takes parameters and produces results to return.
   As with real projects, this can be called by a unit-testing main program as well as find a home in the application that needs that feature.
2. One thing that bugs me about the typical implementation of FizzBuzz is the explicit repetition of code, checking each factor and often the combination as well. Don’t duplicate code that varies only by a data value! Instead, store the values and loop over them.
3. Be mindful of future enhancements.
   Code always gets more complex over time. It is an art forged by long experience to balance future maintenance with extra work now. Don’t program in features that are not needed — rather, anticipate what will be changed or added later and consider that in the overall design, and allow for such things when it does not add complexity to the task at hand. (I can probably write a whole essay on this topic. Message me if you have examples or general thoughts.)

Results

I was very successful in the point 2 makes the code simpler and shorter. Point 3 takes the form of putting the configuration at the very top of the file, and showing how it can be extended; and the extensibility is free due to the architecture noted in point 2.

There are the needed #includes at the top of the file, and then the interesting part starts off with:

constexpr std::pair<int, const char*> fbchart[] {
    { 3, "Fizz" }, { 5, "Buzz" },  // standard FizzBuzz
    { 7, "Boom" }  // easily generalized and extended by adding to this chart!
};

When I wrote this, Visual Studio’s C++17 compiler is still in per-release, and part of the exercise for me was to use new features and see where style needs to be updated.

“constexpr is the new static const.” Naturally, this tabular data will be declared as constexpr.

Now, why did I use const char* rather than an object type? This was a deliberate choice. First of all, the use of this table (described later) is flexible in what it can take, so I don’t require a std::string here even though that’s what will be used in the constructed object. There is no need to store a std::string object here, which would make another copy of the lexical string literals, and string does not have a constexpr constructor.

The usual worries about primitive C-style strings do not apply since this is constant data. After all, do you worry about writing cout<<"hello"; and demand that you store that in a string object before passing it to the function?

A hot-off-the-press alternative would be std::string_view. But there is simply no need here. I chose not to use gsl::zstring, since it would be the only use of the Guidelines Support Library in the program, and there is no question that C-strings are being used since they are right there, and only right there. This is not a function parameter that needs documenting.

Likewise for the use of the plain array, rather than std::array. Arrays are not as poor as they used to be: with the preference for free functions begin, end, size, etc. what does the object wrapper do that the plain doesn’t? Only the regular value semantics of passing and assigning — and I’m not doing that.

string fbencode (int n)
{
    string retval;
    for (auto& [div, codeword] : fbchart) {
        if ((n%div)==0)  retval += codeword;
    }
    if (retval.empty())  retval= std::to_string (n);
    return retval;
}

Here is the meat: a simple stateless function that accepts the number as a parameter and produces the results to return as a string.

The loop has a one-line body, and it is automatically iterated over the data table shown earlier. That is why adding another item just works. This code is smaller and simpler than the cascade of tests you normally see.

// and to drive it: 
int main() {
    using boost::irange;
    for (auto n : irange(1,101)) {
        cout << n << ": " << fbencode(n) << '\n';
    }
}

In the main, I avoided a legacy for-loop by using Boost.Range 2.0. That is tried and true for production work. I’ve not tried to get the latest Visual Studio compiler drop to swallow Range-v3 yet.

Discussion

Not only does this show fluency in C++ with up-to-date skills, it is well crafted and provides a number of discussion points. As summarized along with the code samples, I can explain why I did things a certain way, alternatives considered, what I might do under different circumstances, etc.

It is good to understand the constructs used and choose them purposefully, not apply things in a cargo-cult or voodoo programming situation. On either end of the interview process, understand that simple coding problems can give far more insight than simply finding out whether the applicant can “write code” at all.