This article is an excerpt from my book about Data-Oriented Programming.

More excerpts are available on my blog.


The principle in a nutshell

Principle #2: Model the data part of the entities of your application using generic data structures (mostly maps and arrays).

Remarks on Principle #2

  • It’s optional to specify or not the shape of the data of an entity.

  • FP Languages that are statically typed (e.g. Haskell and Ocaml) are not compliant with this principle.

  • The most common data structures are maps (a.k.a dictionaries) and arrays. Other data structures: sets, lists and queues.

  • Principle #2 doesn’t deal with the mutability or the immutability of the data. This is the theme of Principle #3: Data is immutable.

Illustration of Principle #2

According to Principle #1: Separate code from data, we have to separate code and data. The theme of Principle #2 is about the programming constructs that we should use to model our data.

In DO, we model our data with generic data structures (like maps and arrays) instead of specific classes. Most of the data entities that appear in a typical application could be modeled with maps and arrays.

Let’s look at the same simplistic example as the one used to illustrate Principle #1: the data that represents and author.

An author is a data entity with a firstName, a lastName and a number of books he/she wrote.

We break this principle when we use classes to represent an author, like this:

class AuthorData {
  constructor(firstName, lastName, books) {
    this.firstName = firstName;
    this.lastName = lastName;
    this.books = books;
  }
}

We are compliant with this principle when we use a map (which is a generic data structure) to represent an author:

function createAuthorData(firstName, lastName, books) {
 var data = new Object;
 data.firstName = firstName;
 data.lastName = lastName;
 data.books = books;
return data;
}

In a language like JavaScript, a map could be instantiated also via literals, which is a bit more convenient:

function createAuthorData(firstName, lastName, books) {
   return {firstName: firstName, lastName: lastName, books: books};
}

Benefits of Principle #2

When we use generic data structures to represent our data, our programs benefit from:

  • Leverage functions that are not limited to our specific use case

  • Flexible data model

Leverage functions that are not limited to our specific use case

There is a famous quote by Alan Perlis that summarizes this benefit very well:

It is better to have 100 functions operate on one data structure than 10 functions on 10 data structures.
— Alan Perlis

When we use generic data structures to represent entities, we have the privilege to manipulate the entities with the rich set of functions available on maps natively in our programming language in addition to the ones provided by third party libraries.

For instance, JavaScript natively provides some basic functions on objects and third party libraries like lodash extend the functionality with even more functions.

As an example, when an author is represented as a map, we can serialize it into JSON for free, using JSON.stringify which is part of JavaScript:

var data = createAuthorData("Isaac", "Asimov", 500);
JSON.stringify(data);

And if we want to serialize the author data without the number of books, we can use lodash’s pick function to create an object with a subset of keys:

var data = createAuthorData("Isaac", "Asimov", 500);
var dataWithoutBooks = _.pick(data, ['firstName', 'lastName']);
JSON.stringify(dataWithoutBooks);

When you adhere to Principle #2, all this wealth of functionalities is available to manipulate all your entities.

Flexible data model

When we use generic data structures, our data model is flexible in the sense that our data is not forced to adhere to a specific shape. We are free to create data with no predefined shape. And we are free to modify the shape of our data.

In classical OO, each piece of data is instantiated via a class. As a consequence, even when a slightly different data shape is needed, we have to define a new class.

Take for example a class AuthorData that represent an author entity that made of 3 fields: firstName, lastName and books. Suppose that you want to add a field fullName with the full name of the author. In OO, you would have to define a new class AuthorDataWithFullName.

However in DO, you are free to add (or remove) fields to a map "on the fly":

var data = createAuthorData("Isaac", "Asimov", 500);
data.fullName = "Isaac Asimov";
data

Price for Principle #2

There are no free meals. Applying Principle #2 comes at a price.

The price we have to pay when we mode entities with generic data structures is that:

  • Performance hit

  • Data shape needs to be documented manually

  • No compile time check that the data is valid

Price #1: Performance hit

When we use specific classes to instantiate data, retrieving the value of a class member is super fast. The reason is that the compiler knows upfront how the data is going to look like and it can do all kinds of optimizations.

However, when we use generic data structures to store data, the data structure is optimized for the general case, like retrieving an arbitrary key from a map.

Retrieving an arbitrary key from a map is slower than retrieving the value of a class member.

Similarly setting an arbitrary key to a value is slower that setting the valued of a class member.

Usually, this performance hit is not significant, but it is something to keep in mind.

Price #2: Data shape needs to be documented manually

When an object is instantiated from a class, the information of the data shape is in the class definition. It is helpful for developers and for IDEs (think about auto-completion features).

When we use generic data structures to store data, the shape of the data needs to be documented manually.

Even when we are disciplined enough and we document our code, it may occur that we modify slightly the shape of an entity and we forget to update the documentation.

In that case, we have to explore the code in order to figure out what is the real shape of our data.

Price #3: No compile time check that the data is valid

Take a look again at the fullName function that we created during our exploration of Principle #1:


function fullName(data) {
   return data.firstName + " " + data.lastName;
}

When we pass to fullName a piece of data that doesn’t conform to the shape fullName expects, an error occurs at runtime. For example, we could mistype the field that stores the first name (fistName instead of firstName):

fullName({fistName: "Issac", lastName: "Asimov"})

When data is instantiated only via classes, this type of error is caught at compile time.

Wrapping up

DO guides us to use generic data structures to model our entities.

When we adhere to this principle, it allows us to manipulate the entities with generic functions (provided by the language and by third party libraries) and it keeps our data model flexible.

This flexibility causes a (small) performance hit and forces us to document manually the shape of our data as we cannot rely on the compiler to statically validate it.

Continue your exploration of Data Oriented Programming principles and move to Principle #3: Data is immutable.

This article is an excerpt from my book about Data-Oriented Programming.

More excerpts are available on my blog.