Having looked at four different approaches to estimating, it’s time to take a brief but important aside: have you noticed just how prevalent the idea of heirarchy is in software? By hierarchy, I mean a general pattern of connectedness, where a parent has some children, who, in turn, may each have some further children, and so on.

In this section, I’m going to look at a whole bunch of examples, and then try to examine exactly why this is such a popular pattern. This will be a useful tool in our arsenal for estimation in the next section.

A Javascript Example

Here’s an example from the ChartJS website I was looking at, just the other day:

<canvas id="myChart" width="400" height="400"></canvas>
var ctx = document.getElementById('myChart').getContext('2d');
var myChart = new Chart(ctx, {
    type: 'bar',
    data: {
        labels: ['Red', 'Blue', 'Yellow', 'Green', 'Purple', 'Orange'],
        datasets: [{
            label: '# of Votes',
            data: [12, 19, 3, 5, 2, 3],
            backgroundColor: [
                'rgba(255, 99, 132, 0.2)',
                'rgba(54, 162, 235, 0.2)',
                'rgba(255, 206, 86, 0.2)',
                'rgba(75, 192, 192, 0.2)',
                'rgba(153, 102, 255, 0.2)',
                'rgba(255, 159, 64, 0.2)'
            borderColor: [
                'rgba(255, 99, 132, 1)',
                'rgba(54, 162, 235, 1)',
                'rgba(255, 206, 86, 1)',
                'rgba(75, 192, 192, 1)',
                'rgba(153, 102, 255, 1)',
                'rgba(255, 159, 64, 1)'
            borderWidth: 1

We’ve got…

  • A hierarchy of tags such as <canvas>, <script> and so on.
  • We have attributes on the tags, like width and height.
  • Within the <script> tag, we have a hierarchy of statements, where we assign the variablesctx and myChart in order.
  • A hierarchy of method calls on objects, such as: document.getElementById('myChart').getContext('2d')
  • A nested hierarchy of maps (delimited by { and }) and arrays (delimited by [ and ]), starting with { type: 'bar'...
  • Within those arrays, we have strings, consisting of characters, delimited by ' here, such as 'rgba(153, 102, 255, 1)'.
  • But actually, that’s another function call, which uses brackets to separate the name of the function rgba from it’s parameters.
  • And those parameters are separated by ,s…
  • And the Chart constructor is defined in a different package, which is a collection of files in a repository (maybe npm or jsdelivr.
  • Some hierarchy is indicated using indentation of the code. Some isn’t.

What is the take-away from this?

We heavily use syntax to indicate different types of hierarchies within software.

Just look at all the different ways it happens above: brackets, single-quotes, curly brackets, angle-brackets and tags, square brackets, double-quotes, commas and semi-colons.

Syntax Trees

Here is another piece of Javascript, a function to calculate a Fibonacci Sequence:

function fibonacci(num){
  var a = 1, b = 0, temp;

  while (num >= 0){
    temp = a;
    a = a + b;
    b = temp;

  return b;

When the Javascript parser goes to work, it builds an internal Abstract Syntax Tree from the code, looking something like this:

Abstract Syntax Tree, Rendered By [viswesh](https://viswesh.github.io/astVisualizer/)

Now, the syntax tree generated for Javascript is different to the way the code looks. In other languages, like Lisp, the syntax tree and the code structure are the same, and this is called homoiconicity.

Human Systems

We build hierarchies not just into our programming languages, but all over our societies. They seem fundamental to how we understand things. As well as Family Trees and Organisation Charts, we use hierarchies everywhere. For example, biologists often break down the complexity of the human body like this:

  • Organelles - such as Mitochondria, contained in…
  • Cells - such as blood cells, nerve cells, skin cells in the Human Body, inside…
  • Organs - like hearts livers, brains etc, held within…
  • Organ Systems - like the circulatory system, the immune system, the respiratory system, contained in…
  • Organisms - like you and me.

Wikipedia calls this a compositional containment hierarchy:

“The compositional hierarchy that every person encounters at every moment is the hierarchy of life. Every person can be reduced to organ systems, which are composed of organs, which are composed of tissues, which are composed of cells, which are composed of molecules, which are composed of atoms. In fact, the last two levels apply to all matter, at least at the macroscopic scale. Moreover, each of these levels inherit all the properties of their children. “ - Hierarchy, Wikipedia

An Essential Problem

Unfortunately, containment hierarchies break down when you look too closely.

You see that Javascript syntax tree? Unfortunately, we are passing things from one part of the hierarchy to another in the form of the variables (temp, num, a and b here) or named functions.

You see those veins in the Circulatory System? They connect with all of the bodily systems, as do nerves which are part of the Nervous System.

Where does one system end and another begin?

Although biological pressure seems to have led to a hierarchical organisation, it knows when to break it’s own rule.

That’s because on their own, hierarchies are too simple to express complexity. (For a graph-centric look at how we can measure complexity, please review Complexity Risk.)

Goto Considered Harmful

On the other hand, where would we be without hierarchy in our software code? It’s not impossible to imagine:

  • We could write code in a stack-less, goto-oriented way, but such programs are extremely hard to reason about, as discussed in E.W. Dijkstra’s seminal paper Goto Considered Harmful.
  • Turing Machines and the BrainFuck language both manage without any kind of hierarchy, and are Turing Complete, meaning that you can do any kind of computing in them. Although, they’re both very hard to reason about.

Classification Hierarchy

The other type of hierarchy we come across both in software and everywhere else in the human experience is the classification hierarchy. As an example of this, let’s consider planets. The definition of a planet is quite bogus, and has changed over time:

  • The Greeks coined asteres planetai to be the class of objects in the sky that moved separately from the rest of the body of stars. Possibly including moons, comets and asteroids. 1.
  • However, after the Copernican Revolution made the moon a satellite of earth, the defintion of planets seemed to be bodies orbiting the sun, and there were just 9 of them: Mercury, Mars, Earth, Venus, Saturn, Jupiter, Uranus, Neptune and Pluto.
  • In 2005, The Discovery of Eris, a body larger than Pluto orbiting in a trans-Neptunian orbit meant that potentially hundreds of objects deserved the term planet.
  • In response, Pluto was demoted to being a dwarf planet. In order to do this, the definition of planet was changed to include the clause that it had “cleared its neighbourhood” of most other orbiting bodies”. This excluded Kuiper-Belt objects such as Pluto, but is still problematic, as Alan Stern discusses below.

“I and many other planetary scientists — like the almost 400 that signed a petition against the IAU in 2006 — have a problem with the IAU definition because the implications of it are just nonsensical. Here’s why. The IAU’s “zone-clearing” criteria, when worked out mathematically, means that to qualify as a planet at larger and larger distances from the sun, a body has to have more and more mass than it would in a closer orbit. This is in part because the zones get larger (like distance cubed, or volume) as you go outward; it’s also in part because orbital speeds are slower further out, so zone-clearing takes longer.” - Alan Stern, Fighting for Pluto’s Planet Title

So the problem comes down to the fact that, on one hand, we want a nice classification of the eight or nine largest objects orbiting our sun, rather than a messy classification of hundreds.

Multiarchies and Typing

A second problem with classification hierarchies is that, unlike containment, you can classify things along many axes. For example, a cup might fit into the classifications “drinking receptacle”, “kitchenware” and “Star-Wars memorabilia” all at the same time.

Nevertheless, a lot of the power of Interfaces in programming languages comes from being able to do this.

Type Systems are invariably built on the concept of classification hierarchies. And this leads to a really interesting point: whenever we “reach out” of the containment hierarchy of a software program to call another piece of code (via calling a function or a variable) we fall back to using the classification hierarchy to determine whether that connection is a valid one.

In a strongly-typed language like Java, for example we might have this:

public class Numbers {
    public static void main(String[] args) {
        System.out.print("Square root of 4 is: " + Math.sqrt(4));

The compositional hierarchy you might draw like this:

But there are three places where we leave the compositional hierarchy to call static functions in other packages: Math.sqrt, System.out.println and + (string concatenation). In these cases, we rely on the classification hierarchy of the Java Type System to determine whether the call is acceptable:

  • Math.sqrt: takes a float, returns a float.
  • System.out.println: takes a String.
  • +: takes a String and something that can be converted to a String, returns a String.

Compositional hierarchies on a larger project:  methods, classes, packages, directories, projects

In Eclipse (my Java IDE) I can therefore view both these types of hierarchy. In the above screen-grab, you can see some more compositional hierarchy methods, classes, packages, directories and projects.

Classification hierarchy of the Resource class from Spring

Whereas in this screen grab, I can view the hierarchy of a class within Java (here the Resource class from Spring).

Although this is an Object-Oriented example, the same classification system exists within functional languages, too.



This has been a somewhat rambling introduction to two key types of hierarchy:

Containment hierarchies are used everywhere in software development: files in disks in servers, methods and functions in packages and namespaces etc. Good programming languages attempt to capture as much of the program’s complexity as possible within the containment hierarchy.

People understand containment hierarchies because they’re baked into (and invented by) our brains. When I look outside at a car, I can see that it is a containment hierarchy of windows, wheels and metal panels. When I think about my house, I think about different objects being contained within different rooms within a structure of bricks. But none of that exists: it’s all in my head. Everything is really just a bunch of atoms.

Classification hierarchies are also used everywhere in software development: strings, numbers, records, classes, types, schemas. A key ability for a programmer is often to be able to abstract from multiple areas and say “this is like this”.

Classification hierarchies are also baked into our brains: looking at my car, the windows and wheels, these are all classifications of objects that exist in my head. But everything is really just a bunch of atoms.

When we create programs (or set up databases) we are classifying things. Having a field for “Marital Status”, “Address” or “Planet / Not a Planet” is just something we’ve invented for the purposes of us processing data. We are copying the mental concepts we’ve developed for navigating the real world and pasting them into our code.

We should be cautious doing this, as we might well be wrong.

internal model picture, copying.

At this point, you’re probably wondering what any of this has to do with estimating in software development, so [let’s continue]…

Add Your Star On GitHub to receive an invite to the GitHub Risk-First team.

Rob Moffat
Rob Moffat Author of Risk-First Software Development. Developer. Working in the UK.