03 August 2018 code, csharp, lambda Oliver Guhr

Improving code

TL;DR;

Things I learned:

  • long one-liners are hard to read and understand
  • split up your code into small, easy to understand functions
  • less “plumping” (read infrastructure code) is the better
  • get indentation right
  • “Make it correct, make it clear, make it concise, make it fast. In that order.” Wes Dyer

Why should I bother?

Readable code is:

  • easier to debug
  • fast to fix
  • easier to maintain

The problem

Recently I wanted to implement an algorithm for a project we are doing. The goal was to create a so-called “Balanced Latin Square”, we used it to prevent ordering effects in user studies. You can find a little bit of background here and a nice description of the algorithm here.

It’s fairly simple, although it is not obvious how it works, just by looking at the code. The function takes an integer as an argument and returns a Balanced Latin Square. For example, a “4” would return this matrix of numbers:

1 2 4 3 
2 3 1 4 
3 4 2 1 
4 1 3 2 

And there is a little twist if your number is odd, then you need to reverse every row and append them to your result.

After I created the my implementation, I had an idea on how to simplify it. At least I thought its simpler ;)

First attempt - Loops

Based on the description and a Python version of that algorithm, I created a classical (read “imperative”) implementation.

So this is the C# Code:

public List<List<String>> BalancedLatinSquares(int n)
{
    var result = new List<List<String>>() { };
    for (int i = 0; i < n; i++)
    {
        var row = new List<String>();
        for (int j = 0; j < n; j++)
        {
            var cell = ((j % 2 == 1 ? j / 2 + 1 : n - j / 2) + i) % n;
            cell++; // start counting from 1
            row.Add(cell.ToString());
        }
        result.Add(row);
    }
    if (n % 2 == 1)
    {
        var reversedResult = result.Select(x => x.AsQueryable().Reverse().ToList()).ToList();                
        result.AddRange(reversedResult);
    }
    return result;
}

I also wrote some simple unit tests to ensure this works. But in the end, I really didn’t like this code. It contains two nested loops and a lot of plumbing code. There are four lines alone just to create the result object (list) and to add the values to it. Recently I looked into functional programming and since C# also has some functional inspired features, I tried to improve this code with some functional goodness :)

Second attempt - Lambda Expressions

public List<List<String>> BalancedLatinSquares(int n)
{
    var result = Enumerable.Range(0, n)
        .Select(i =>
                Enumerable.Range(0, n).Select(j => ((((j % 2 == 1 ? j / 2 + 1 : n - j / 2) + i) % n)+1).ToString()).ToList()
            )
        .ToList();     
    
    if (n % 2 == 1)
    {
        var reversedResult = result.Select(x => x.AsQueryable().Reverse().ToList()).ToList();
        result.AddRange(reversedResult);
        return result;
}

This is the result of my attempt to use some functional features. And hey, it is much shorter, therefore it must be better, right? Well, I posted a screenshot of both versions on Twitter and asked which one the people prefer. As it turned out, a lot of folks actually preferred the loop version. But why? Looking back at my code a saw two problems by looking at this line:

Enumerable.Range(0, n).Select(j => ((((j % 2 == 1 ? j / 2 + 1 : n - j / 2) + i) % n)+1).ToString()).ToList()

  • I squeezed a lot of code in this one liner. This makes it harder to read and therefore harder to understand.
  • Another issue is, that I omitted descriptive variable names since they are not needed anymore. Oh and I removed the only comment I wrote since this comment would not fit in the one line of code :)

So, shorter is not always better.

Third attempt - better Lambda Expressions

The smart folks on Twitter had some great ideas about how to improve my code.

The first step was to get rid of the unholy one-liner. You can - and should - always split up your code into smaller, meaningful code blocks. I pulled out the calculateCell function and out of that I also extracted a isEven function. The nice thing is, that the function names also working as a kind of documentation about whats going on.

By returning IEnumerable instead of lists, I was able to remove some .toList() calls. Also, I was able to shorten the code to create the reversedResult.

Another simple step to improve readability is to get line indentation right. Personally, I don’t care which indentation style people are using, as long as it’s used consistently.

public static IEnumerable<IEnumerable<int>> GenerateBalancedLatinSquares(int n)
{
    bool isEven (int i) => i % 2 == 0;        
    int calculateCell(int j, int i) =>((isEven(j) ? n - j / 2 : j / 2 + 1) + i) % n + 1;
    
    var result = Enumerable
                    .Range(0, n)
                    .Select(row =>
                        Enumerable
                            .Range(0, n)
                            .Select(col =>calculateCell(col,row))
                    );     
    
    if (isEven(n) != false)
    {
        var reversedResult = result.Select(x => x.Reverse());                
        result = result.Concat(reversedResult);
    }        
    return result;conditional
}

I think there is room for further improvement. For the calculateCell function I am using this ?: conditional operator, it allows you to write very compact code, on the other hand, it’s also harder to read. If you would replace this with an if statement you would need more lines of code, but also have more space to add comments. Functional languages like Scala, F#, and Haskel providing this neat match expression that could help here.

Extra: How does this algorithm look in other languages:

Python

def balanced_latin_squares(n):
    l = [[((j/2+1 if j%2 else n-j/2) + i) % n + 1 for j in range(n)] for i in range(n)]
    if n % 2:  # Repeat reversed for odd n
        l += [seq[::-1] for seq in l]
    return l

I took this sample from Paul Grau.

Haskell

Thank you Carsten