Domain-Driven Design in PHP
Table of Contents
- Foreword by Matthias Noback
- About the Authors
- Value Objects
Appendix: Hexagonal Architecture with PHP
- First Approach
- Repositories and the Persistence Edge
- Decoupling Business and Persistence
- Migrating our Persistence to Redis
- Decouple Business and Web Framework
- Rating an idea using the API
- Console app rating
- Testing Rating an Idea UseCase
- Testing Infrastructure
- Arggg, So Many Dependencies!
- Domain Services and Notification Hexagon Edge
- Let’s Recap
- Hexagonal Architecture
- Key Points
- What’s Next?
Foreword by Matthias Noback
I must admit that when I first heard of the Domain-Driven Design in PHP initiative, I was a bit worried. The danger was twofold: first of all, when glancing over the table of contents, the subject matter looked like it was a rehash of content that was already available in several other Domain-Driven Design books. Second, writing a book on Domain-Driven Design targeted specifically toward the PHP community seemed needlessly narrowing, particularly as Domain-Driven Design itself is not language specific. As such, this might inhibit PHP developers from looking past the boundaries of their own community, especially when considering that there’s a lot going on beyond the scope of PHP. In fact, even Domain-Driven Design is one of those things, as it didn’t originate in the PHP community.
After reading the book, I’m happy to inform you that my worries have been invalidated!
With regard to my first concern: of course there is some overlap with previously published Domain-Driven Design books. Yet the authors have restrained themselves. The theoretical parts are exactly what you need to be able to understand what’s going on in the code samples. Besides, if you never read another Domain-Driven Design book, this one gives you what you need to start applying some Domain-Driven Design principles and patterns in your code, as it’s practical by nature.
My second concern — about the PHP aspect of this book — has been addressed very well. It turns out there are a lot of things to say about Domain-Driven Design in a PHP world. This book is specifically targeted at an audience consisting of PHP developers. The code samples resemble real-world PHP projects, and the authors use a programming style we know from projects using Symfony or Silex. For persisting Domain objects, Doctrine ORM — which is the de facto standard data mapper for PHP — is used.
This book also fulfills a need I’ve often seen in the PHP community: the need for concrete examples. It’s not always easy for authors to come up with proper illustrations of how to apply certain ideas that have a low risk of being misinterpreted or abused in real-world projects. And in Domain-Driven Design, which is philosophical by nature, this is even more challenging.
In the case of this book, the authors haven’t been afraid to show many useful examples, along with some interesting alternative solutions. They aren’t just handwaving at solutions either; they take the time to provide detailed explanations — such as when they talk about saving snapshots for Aggregates with a large number of Domain Events, or when they discuss integrating Bounded Contexts using RabbitMQ. I can’t recall having previously seen an implementation of these things in a book or article on Domain-Driven Design.
For me personally, Domain-Driven Design is one the most interesting subjects in software development today. There is so much to discover, and there are many subjects related to it: Agile software development, TDD, and BDD, but also living documentation, visualization, and knowledge crunching techniques. Once you start looking into all of this, you’ll realize that Domain-Driven Design is an area of expertise worth investigating, as it enables you to add much more to your own worth as a software developer.
So, I guess what I want to say is this: dive into this book, learn from it, and then pick up another book (see the list of references at the end of this book for suggestions of future reading). Continuous learning is a fundamental part of keeping up to date in the software industry, so don’t stop here.
Oh, and by the way: if you get a chance to go to Barcelona, make sure you take part in one of the many PHP or Symfony events. The community is big, friendly, and full of interesting ideas. You’ll find the authors of this book there too. They are all invested in the local PHP community and are happy to share their insights and experiences with you!
Author of A Year with Symfony
In 2014, after two years of reading about and working with Domain-Driven Design, Carlos and Christian, friends and workmates, traveled to Berlin to participate in Vaughn Vernon’s Implementing Domain-Driven Design Workshop. The training was fantastic, and all the concepts that were swirling around in their minds prior to the trip suddenly became very real. However, they were the only two PHP developers in a room full of Java and .NET developers.
Around the same time, php[tek], an annual PHP conference, opened its call for papers, and Carlos sent one about Hexagonal Architecture. His talk was rejected, but Eli White — of musketeers.me and php[architect] fame — got in touch with him a month later wondering if he was interested in writing an article about Hexagonal Architecture for the magazine php[architect]. So in June 2014, Hexagonal Architecture with PHP was published. That article, which you’ll find in the appendix, was the origin of this book.
In late 2014, Carlos and Christian talked about extending the article and sharing all their knowledge of and experience in applying Domain-Driven Design in production. They were very excited about the idea behind the book: helping the PHP community delve into Domain-Driven Design from a practical approach. At that time, concepts such as Rich Domain Models and framework-agnostic applications weren’t so common in the PHP community. So in December 2014, the first commit to the GitHub book repository was pushed.
Around the same time, in a parallel universe, Keyvan co-founded Funddy, a crowdfunding platform for the masses built on top of the concepts and building blocks of Domain-Driven Design. Domain-Driven Design proved itself effective in the exploratory process and modeling of building an early-stage startup like Funddy. It also helped handle the complexity of the company, with its constantly changing environment and requirements. And after connecting with Carlos and Christian and discussing the book, Keyvan proudly signed on as the third writer.
Together, we’ve written the book we wanted to have when we started with Domain-Driven Design. It’s full of examples, production-ready code, shortcuts, and our recommendations based on our experiences of what worked and what didn’t for our respective teams. We arrived at Domain-Driven Design via its building blocks — Tactical Patterns — which is why this book is mainly about them. Reading it will help you learn them, write them, and implement them. You’ll also discover how to integrate Bounded Contexts using synchronous and asynchronous approaches, which will open your world to strategic design — though the latter is a road you’ll have to discover on your own.
This book is heavily inspired by Implementing Domain-Driven Design by Vaughn Vernon (aka the Red Book), and Domain-Driven Design: Tackling Complexity in the Heart of Software by Eric Evans (aka the Blue Book). You should buy both books. You should read them carefully. You should love them.
Who Should Read This Book
If you’re a PHP Developer, Architect, or Tech Lead, we highly recommend this book. It will help you become a better professional. It will give you a new overview of and approach to the applications you’re developing. If you’re a Junior profile, getting into Value Objects, Entities, Repositories, and Domain Events is important in order to model any Domain you’ll face in the future. For an average profile, understanding the benefits of Hexagonal Architecture and the boundaries between your framework and your Application is key for writing code that’s easier to maintain in the real world (framework migrations, testing, etc.). More advanced readers will have fun both exploring how to use Domain Events in order to integrate Applications and delving deeper into Aggregate design.
Although Domain-Driven Design is not about technology, you still need it to make HTTP requests to access your Domain. Throughout the book, we recommend using specific PHP frameworks and libraries, such as Symfony, Silex, and Doctrine. For some examples, we also use specific technologies, such as MySQL, RabbitMQ, Redis, and Elasticsearch. However, most important are the behind-the-scenes concepts — concepts that are applicable regardless of the technology used to implement them.
Additionally, the book is loaded with tons of details and examples, such as how to properly design and implement all the building blocks of Domain-Driven Design — including Value Objects, Entities, Services, Domain Events, Aggregates, Factories, Repositories, and Application Services — with PHP. It explains what the role of the main PHP libraries and frameworks used in Domain-Driven Design are. The book also teaches how to apply Hexagonal Architecture within your application, regardless of whether you use an open source framework or your own one. Finally, it shows how to integrate Bounded Contexts using REST frameworks and messaging mechanisms. If you’re interested in any of these subjects, this book is for you.
DDD and PHP Community
In 2016, Carlos and Christian went to the first official Domain-Driven Design conference, DDD Europe. They were really happy to see some PHP open source leaders, such as Marco Pivetta (Doctrine) and Sebastian Bergmann (PHPUnit), attending the conference.
Domain-Driven Design arrived in the PHP community two years prior to that conference. However, there’s still a lack of documentation and real code examples. Why? We think not many people have worked with this kind of approach in production yet — even people in other more established communities such as Java. Maybe this is because their project complexity is low, or maybe it’s because they don’t know how to do it. Whatever the reason, this book is written for the community. One of our goals is to teach you how you can write an application that solves your Domain issues without being coupled to specific frameworks or technologies.
Summary of Chapters
The book is arranged with each chapter exploring a separate tactical building block of Domain-Driven Design. It also includes an introduction to Domain-Driven Design, information on how to integrate different Bounded Contexts or applications, and an appendix.
Chapter 1: Getting Started with Domain-Driven Design
What is Domain-Driven Design about? What role does it play in complex systems? Is it worth learning about and exploring? What are the main concepts a developer needs to know when jumping into it?
Chapter 2: Architectural Styles
Bounded Contexts can be implemented in different ways and using different approaches. However, two styles are getting more popular, and they are Hexagonal Architecture and CQRS + ES. In this chapter, we’ll see these two main Architectural Styles, understand what their main strengths are, and discover when to use them.
Chapter 3: Value Objects
Value Objects are the basic pieces for rich modeling. We’ll learn what their properties are and what makes them so important. We’ll figure out how to persist them using Doctrine and custom ORMs. We’ll show how to properly validate and unit test them. And finally, we’ll see what a test case of testing immutability looks like.
Chapter 4: Entities
Entities are Domain-Driven Design building blocks that are uniquely identified and mutable. We’ll see how to create and validate them and how to properly map them using a custom ORM and Doctrine. We’ll also assess whether or not annotations are the best mapping approach for Entities and look at the different strategies for generating an Identity.
Chapter 5: Domain Services
In this chapter, you’ll learn about what a Domain Service is and when to use it. We’ll review what Anemic Domain Models and Rich Domain Models are. Lastly, we’ll deal with Infrastructure issues when writing Domain Services.
Chapter 6: Domain Events
Domain Events are a great Inversion of Control (IoC) mechanism. In Domain-Driven Design, they’re important for communicating different Bounded Contexts asynchronously, improving your Application performance using eventual consistency, and decoupling your Application from its Infrastructure.
Chapter 7: Modules
With so many tactical building blocks, it’s a bit difficult to know where to place them in code, especially if you’re dealing with a framework like Symfony. We’ll review how PHP namespaces can be used for implementing Modules. We’ll also discover different hierarchies of folders for organizing Domain Model code, Application Code, and Infrastructure Code.
Chapter 8: Aggregates
Aggregates are probably the most difficult part of tactical Domain-Driven Design. We’ll look at the key concepts when dealing with them and discover how to design them. We’ll also propose a practical scenario where two Aggregates become one when adding a business rule, and we’ll demonstrate how the rest of the objects must be refactored.
Chapter 9: Factories
Factory Methods and objects help us keep business invariants, which is why they’re so important in Domain-Driven Design. Here, we’ll also explore the relationship between Factories and Aggregates.
Chapter 10: Repositories
Repositories are key for retrieving and adding Entities and Aggregates to collections. We’ll review the different types of Repositories and learn how to implement them using Doctrine, custom ORMs, and Redis.
Chapter 11: Application
An Application is the thin layer that connects outside clients to your Domain. In this chapter, we’ll show you how to write your Application Services so that they’re easy to test and keep thin. We’ll also review how to prepare request objects, define dependencies, and return results.
Chapter 12: Integrating Bounded Contexts
We’ll explore the different tactical approaches to communicate Bounded Contexts and see real implementations. REST is our suggestion for synchronous communication, and messaging with RabbitMQ is our suggestion for asynchronous communication.
Appendix: Hexagonal Architecture with PHP
Here is where you’ll find the original article written by Carlos and published by php[architect] in June 2014.
Code and Examples
The authors have created an organization at GitHub called Domain-Driven Design in PHP, which is where all the code examples from this book, additional snippets, and some complete sample projects are available. For example, you can find Last Wishes, a simple Domain-Driven Design-style application showing different examples explained in this book. Additionally, you’ll find our CQRS Blog, along with Gamify, a Bounded Context that adds gamification capabilities to Last Wishes.
Finally, if you find any issue or fix or have a suggestion or comment while reading this book, you can create an issue in the DDD in PHP Book Issues repository. We fix them as they come in. If you’re interested, we also urge you to watch our projects and provide feedback.
First of all, we would like to thank all our friends and family. Without their support, writing this book would have been an even more difficult task. Thanks for accommodating our schedules and taking care of our children in order to free up time for us to focus on writing. You’re wonderful, and part of this book is also yours.
We are three Spaniards who wrote a book in English, so if you’d guess our English is far from perfect, you’d be correct. Luckily for us, Edd Mann has supported us with the language since the beginning. He’s not just a great collaborator but also a great friend, and we owe him a huge thanks. The final review was done by the professional copy editor Natalye Childress. She has done a great work rewriting our words to make them understandable. Thank you so much. Our book is easier and more enjoyable to read.
A group of PHP developers in Barcelona defends what we call el camino del rigor, or the path of rigor. It existed before the craftsmanship movement, and it means to struggle with everything stacked against us in order to build exceptional things in an exceptional way. Two particular developers and friends from that group are Albert Casademont and Ricard Clau, both of whom are extraordinary people committed to the community. Thank you so much for helping with the revision process. Your contributions have been incredibly valuable.
We would like to thank every developer who has worked with us in the companies where we’ve applied Domain-Driven Design. We know you’ve been struggling when learning and applying these concepts. Some of you weren’t so open-minded at the beginning, but after using the basic building blocks for a while, you became evangelists. Thanks for your faith.
Our book was for sale from the moment we put the first chapters on Leanpub. Early adopters who bought the book in its beginning stages gave us the much needed love and support to get this done. Thanks for the motivation to keep going.
Thanks also to Matthias Noback for his foreword and feedback on the book. The end result is better because of his contributions.
A special mention to Vaughn Vernon — not just because his work was an incredible source of information and inspiration for us, but also because he helped us find a good publisher, gave us valuable advice, and shared ideas with us. Thanks so much for your help.
Last but not least, we’d like to express our gratitude to all the people who have reported issues, made suggestions, and otherwise contributed to our GitHub repository. To all of you, thank you. You’ve helped us make this book better. More importantly, you’ve helped the community grow and helped other developers be better developers. As Robert C. Martin wrote in his book, Clean Code: A Handbook of Agile Software Craftsmanship, “You are reading this book for two reasons. First, you are a programmer. Second, you want to be a better programmer. Good. We need better programmers.” So thanks to Jordi Abad, Jonathan Wondrusch, César Rodríguez, Yannick Voyer, Victor Guardiola, Oriol González, Henry Snoek, Tom Jowitt, Sascha Schimke, Sven Herrmann, Daniel Abad, Luis Rovirosa, Luis Cordova, Raúl Ramos, Juan Maturana, Nil Portugués, Nikolai Zujev, Fernando Pradas, Raúl Araya, Neal Brooks, Hubert Béague, Aleksander Rekść, Sebastian Machuca, Nicolas Oelgart, Sebastiaan Stok, Vladimir Hraban, Vladas Dirzys, Max Gulturyan, Ivan Đurđevac, Marc Verney, Matthias Gutjahr, Dennis König, Jordi Puig, David Fernández, Vladimir Shadyan and Marc Aube.
About the Authors
- Twitter: @buenosvinos
- Web: https://carlosbuenosvinos.com
- GitHub: https://github.com/carlosbuenosvinos
Christian is a passionate Software Developer, Software Journeyman, and Craftsman Apprentice. He’s an Extreme Programmer soul with more than 10 years of experience in web development. He’s also a Zend PHP 5.3 Certified Engineer, a Zend Framework Certified Engineer, and a SensioLabs Certified Symfony Developer. He has worked as a freelancer, as well as at Privalia, Emagister, Atrapalo, and Enalquiler as a Software Architect.
Keyvan is a polyglot Software Developer who loves Software fundamentals, the Craftsmanship movement, Extreme Programming, SOLID principles, Clean Code, Design Patterns, and Testing. He’s also a sporadic Functional Programmer. He understands technology as a medium for providing value. He has worked on countless projects as a freelancer, on video streaming (Youzee), and on an online marketplace (MyBuilder) â€” all in addition to founding a crowdfunding company (Funddy). Currently, Keyvan is working in FinTech as a Lead Developer at TransferWise London.
Value Objects are a fundamental building block of Domain-Driven Design, and they’re used to model concepts of your Ubiquitous Language in code. A Value Object is not just a thing in your Domain — it measures, quantifies, or describes something. Value Objects can be seen as small, simple objects — such as money or a date range — whose equality is not based on identity, but instead on the content held.
For example, a product price could be modeled using a Value Object. In this case, it’s not representing a thing, but rather a value that allows us to measure how much money a product is worth. The memory footprint for these objects is trivial to determine (calculated by their constituent parts) and there’s very little overhead. As a result, new instance creation is favored over reference reuse, even when being used to represent the same value. Equality is then checked based on the comparability of the fields of both instances.
Ward Cunningham defines a Value Object as:
a measure or description of something. Examples of value objects are things like numbers, dates, monies and strings. Usually, they are small objects which are used quite widely. Their identity is based on their state rather than on their object identity. This way, you can have multiple copies of the same conceptual value object. Every $5 note has its own identity (thanks to its serial number), but the cash economy relies on every $5 note having the same value as every other $5 note.
Martin Fowler defines a Value Object as:
a small object such as a Money or date range object. Their key property is that they follow value semantics rather than reference semantics. You can usually tell them because their notion of equality isn’t based on identity, instead two value objects are equal if all their fields are equal. Although all fields are equal, you don’t need to compare all fields if a subset is unique - for example currency codes for currency objects are enough to test equality. A general heuristic is that value objects should be entirely immutable. If you want to change a value object you should replace the object with a new one and not be allowed to update the values of the value object itself - updatable value objects lead to aliasing problems.
Examples of Value Objects are numbers, text strings, dates, times, a person’s full name (composed of first name, middle name, last name, and title), currencies, colors, phone numbers, and postal addresses.
Value Object vs. Entity
Consider the following examples from Wikipedia, in order to better understand the difference between Value Objects and Entities.
When people exchange dollar bills, they generally do not distinguish between each unique bill; they only are concerned about the face value of the dollar bill. In this context, dollar bills are value objects. However, the Federal Reserve may be concerned about each unique bill; in this context each bill would be an entity.
Most airlines distinguish each seat uniquely on every flight. Each seat is an entity in this context. However, Southwest Airlines, EasyJet and Ryanair do not distinguish between every seat; all seats are the same. In this context, a seat is actually a value object.
Currency and Money Example
Money Value Objects are probably the most used examples for explaining Value Objects, thanks to the Money pattern. This design pattern provides a solution for modeling a problem that avoids a floating-point rounding issue, which in turn allows for deterministic calculations to be performed.
In the real world, a currency describes monetary units in the same way that meters and yards describe distance units. Each currency is represented with a three-letter uppercase ISO code:
One of the main goals of Value Objects is also the holy grail of Object-Oriented design: encapsulation. By following this pattern, you’ll end up with a dedicated location to put all the validation, comparison logic, and behavior for a given concept.
Money is used to measure a specific amount of currency. It’s modeled using an amount and a
Amount, in the case of the
Money pattern, is implemented using an integer representation of the currency’s least-valuable fraction — e.g. in the case of USD or EUR, cents.
As a bonus, you might also notice that we’re using self encapsulation to set the ISO code, which centralizes changes in the Value Object itself:
Now that you know the formal definition of Value Objects, let’s dive deeper into some of the powerful features they offer.
While modeling an Ubiquitous Language concept in code, you should always favor Value Objects over Entities. Value Objects are easier to create, test, use, and maintain.
Keeping this in mind, you can determine whether the concept in question can be modeled as a Value Object if:
- It measures, quantifies, or describes a thing in the Domain.
- It can be kept immutable.
- It models a conceptual whole by composing related attributes as an integral unit.
- It can be compared with others through value equality.
- It is completely replaceable when the measurement or description changes.
- It supplies its collaborators with side-effect-free behavior.
Measures, Quantifies, or Describes
As discussed before, a Value Object should not be considered just a thing in your Domain. As a value, it measures, quantifies, or describes a concept in the Domain.
In our example, the
Currency object describes what type of money it is. The
Money object measures or quantifies units of a given
This is one of the most important aspects to grasp. Object values shouldn’t be able to be altered over their lifetime. Because of this immutability, Value Objects are easy to reason and test and are free of undesired/unexpected side effects.
As such, Value Objects should be created through their constructors. In order to build one, you usually pass the required primitive types or other Value Objects through this constructor. Value Objects are always in a valid state; that’s why we create them in a single atomic step. Empty constructors with multiple setters and getters move the creation responsibility to the client, resulting in the Anemic Domain Model, which is considered an anti-pattern.
It’s also good to point out that it’s not recommended to hold references to Entities in your Value Objects. Entities are mutable, and holding references to them could lead to undesirable side effects occurring in the Value Object.
In languages with method overloading, such as Java, you can create multiple constructors with the same name. Each of these constructors are provided with different options to build the same type of resulting object. In PHP, we’re able to provide a similar capability by way of factory methods. These specific factory methods are also known as semantic constructors. The main goal of
fromMoney is to provide more contextual meaning than the plain constructor. More radical approaches propose to make the
_construct method private and build every instance using a semantic constructor.
Money object, we could add some useful factory methods like the following:
By using the
self keyword, we don’t couple the code with the class name. As such, a change to the class name or namespace won’t affect these factory methods. This small implementation detail helps when refactoring the code at a later date.
Due to this immutability, we must consider how to handle mutable actions that are commonplace in a stateful context. If we require a state change, we now have to return a brand new Value Object representation with this change.
If we want to increase the amount of, for example, a
Money Value Object, we’re required to instead return a new
Money instance with the desired modifications. Fortunately, it’s relativity simple to abide by this rule, as shown in the example below:
Money object returned by
increaseAmountBy is different from the
Money client object that received the method call. This can be observed in the example comparability checks below:
So why not just implement something similar to the following example, avoiding the need for a new Value Object class altogether?
This approach has some noticeable flaws, if say, for example, you want to validate the ISO. It doesn’t really make sense for the
Product to be responsible for the currency’s ISO validation (thus violating the Single Responsibility Principle). This is highlighted even more so if you want to reuse the accompanying logic in other parts of your Domain (to abide by the DRY principle).
With these factors in mind, this use case is a perfect candidate for being abstracted out into a Value Object. Using this abstraction not only gives you the opportunity to group related properties together, but it also allows you to create higher-order concepts and a more concrete Ubiquitous Language.
As discussed at the beginning of the chapter, two Value Objects are equal if the content they measure, quantify, or describe is the same.
For example, imagine two
Money objects representing 1 USD. Can we consider them equal? In the “real world,” are two bills of 1 USD valued the same? Of course they are. Directing our attention back to the code, the Value Objects in question refer to separate instances of
Money. However, they both represent the same value, which makes them equal.
In regards to PHP, it’s commonplace to compare two Value Objects using the
== operator. Examining the PHP Documentation definition of the operator highlights an interesting behavior:
When using the comparison operator (==), object variables are compared in a simple manner, namely: Two object instances are equal if they have the same attributes and values, and are instances of the same class.
This behavior works in agreement with our formal definition of a Value Object. However, as an exact class match predicate is present, you should be wary when handling subtyped Value Objects.
Keeping this in mind, the even stricter
=== operator doesn’t help us, unfortunately:
When using the identity operator (===), object variables are identical if and only if they refer to the same instance of the same class.
The following example should help confirm these subtle differences:
A solution is to implement a conventional
equals method in each Value Object. This method is tasked with checking the type and equality of its composite attributes. Abstract data type comparability is easy to implement using PHP’s built-in type hinting. You can also use the
get_class() function to aid in the comparability check if necessary. The language, however, is unable to decipher what equality truly means in your Domain concept, meaning it’s your responsibility to provide the answer.
In order to compare
Currency objects, we just need to confirm that both their associated ISO codes are the same. The
=== operator does the job pretty well in this case:
Money objects use
Currency objects, the
equals method needs to perform this comparability check, along with comparing the amounts:
Product Entity that contains a
Money Value Object used to quantify its price. Additionally, consider two
Product Entities with an identical price — for example 100 USD. This scenario could be modeled using two individual
Money objects or two references pointing to a single Value Object.
Sharing the same Value Object can be risky; if one is altered, both will reflect the change. This behavior can be considered an unexpected side effect. For example, if Carlos was hired on February 20, and we know that Christian was also hired on the same day, we may set Christian’s hire date to be the same instance as Carlos’s. If Carlos then changes the month of his hire date to May, Christian’s hire date changes too. Whether it’s correct or not, it’s not what people expect.
Due to the problems highlighted in this example, when holding a reference to a Value Object, it’s recommended to replace the object as a whole rather than modifying its value:
This kind of behavior is similar to how basic types such as strings work in PHP. Consider the function
strtolower. It returns a new string rather than modifying the original one. No reference is used; instead, a new value is returned.
If we want to include some additional behavior — like an
add method — in our
Money class, it feels natural to check that the input fits any preconditions and maintains any invariance. In our case, we only wish to add monies with the same currency:
If the two currencies don’t match, an exception is raised. Otherwise, the amounts are added. However, this code has some undesirable pitfalls. Now imagine we have a mysterious method call to
otherMethod in our code:
Everything is fine until, for some reason, we start seeing unexpected results when we’re returning or finished with
$aMoney no longer contains 100 USD. What happened? And what happens if
otherMethod internally uses our previously defined
add method? Maybe you’re unaware that
add mutates the state of the
Money instance. This is what we call a side effect. You must avoid generating side effects. You must not mutate your arguments. If you do, the developer using your objects may experience strange behaviors. They’ll complain, and they’ll be correct.
So how can we fix this? Simple — by making sure that the Value Object remains immutable, we avoid this kind of unexpected problem. An easy solution could be returning a new instance for every potentially mutable operation, which the
add method does:
With this simple change, immutability is guaranteed. Each time two instances of
Money are added together, a new resulting instance is returned. Other classes can perform any number of changes without affecting the original copy. Code free of side effects is easy to understand, easy to test, and less error prone.
Consider the following code snippet:
$b are different variables stored in different memory locations, when compared, they’re the same. They hold the same value, so we consider them equal. You can change the value of
20 at any time that you want, making the new value
20 and eliminating the
10. You can replace integer values as much as you want without consideration of the previous value because you’re not modifying it; you’re just replacing it. If you apply any operation — such as addition (i.e.
$b) — to these variables, you get another new value that can be assigned to another variable or a previously defined one. When you pass
$a to another function, except when explicitly passed by reference, you’re passing a value. It doesn’t matter if
$a gets modified within that function, because in your current code, you’ll still have the original copy. Value Objects behave as basic types.
Testing Value Objects
Value Objects are tested in the same way normal objects are. However, the immutability and side-effect-free behavior must be tested too. A solution is to create a copy of the Value Object you’re testing before performing any modifications. Assert both are equal using the implemented equality check. Perform the actions you want to test and assert the results. Finally, assert that the original object and copy are still equal.
Let’s put this into practice and test the side-effect-free implementation of our
add method in the
Persisting Value Objects
Value Objects are not persisted on their own; they’re typically persisted within an Aggregate. Value Objects shouldn’t be persisted as complete records, though that’s an option in some cases. Instead, it’s best to use Embedded Value or Serialize LOB patterns. Both patterns can be used when persisting your objects with an open source ORM such as Doctrine, or with a bespoke ORM. As Value Objects are small, Embedded Value is usually the best choice because it provides an easy way to query Entities by any of the attributes the Value Object has. However, if querying by those fields isn’t important to you, serialize strategies can be very easy to implement.
Consider the following
Product Entity with string id, name, and price (
Money Value Object) attributes. We’ve intentionally decided to simplify this example, with the id being a string and not a Value Object:
Assuming you have a Repository for persisting
Product Entities, an implementation to create and persist a new
Product could look like this:
Now let’s look at both the ad hoc ORM and the Doctrine implementations that could be used to persist a
Product Entity containing Value Objects. We’ll highlight the application of the Embedded Value and Serialized LOB patterns, along with the differences between persisting a single Value Object and a collection of them.
Persisting Single Value Objects
Many different options are available for persisting a single Value Object. These range from using Serialize LOB or Embedded Value as mapping strategies, to using an ad hoc ORM or an open source alternative, such as Doctrine. We consider an ad hoc ORM to be a custom-built ORM that your company may have developed in order to persist Entities in a database. In our scenario, the ad hoc ORM code is going to be implemented using the DBAL library. According to the official documentation, “The Doctrine database abstraction & access layer (DBAL) offers a lightweight and thin runtime layer around a PDO-like API and a lot of additional, horizontal features like database schema introspection and manipulation through an OO API.”
Embedded Value with an Ad Hoc ORM
If we’re dealing with an ad hoc ORM using the Embedded Value pattern, we need to create a field in the Entity table for each attribute in the Value Object. In this case, two extra columns are needed when persisting a
Product Entity — one for the amount of the Value Object, and one for its currency ISO code:
For persisting the Entity in the database, our Repository has to map each of the fields of the Entity and the ones from the
Money Value Object. If you’re using an ad hoc ORM Repository based on DBAL – let’s call it
DbalProductRepository – you must take care of creating the
INSERT statement, binding the parameters, and executing the statement:
After executing this snippet of code to create a
Product Entity and persist it into the database, each column is filled with the desired information:
As you can see, you can map your Value Objects and query parameters in an ad hoc manner in order to persist your Value Objects. However, everything is not as easy as it seems. Let’s try to fetch the persisted
Product with its associated
Money Value Object. A common approach would be to execute a
SELECT statement and return a new Entity:
There are some benefits to this approach. First, you can easily read, step by step, how the persistence and subsequent creations occur. Second, you can perform queries based on any of the attributes of the Value Object. Finally, the space required to persist the Entity is just what is required — no more and no less.
However, using the ad hoc ORM approach has its drawbacks. As explained in the Domain Events chapter, Entities (in Aggregate form) should fire an Event in the constructor if your Domain is interested in the Aggregate’s creation. If you use the
new operator, you’ll be firing the Event as many times as the Aggregate is fetched from the database.
This is one of the reasons why Doctrine uses internal proxies and
unserialize methods to reconstitute an object with its attributes in a specific state without using its constructor. An Entity should only be created with the
new operator once in its lifetime:
If your intention is still to roll out your own ORM, be ready to solve some fundamental problems such as Events, different constructors, Value Objects, lazy load relations, etc. That’s why we recommend giving Doctrine a try for Domain-Driven Design applications.
Besides, in this instance, you need to create a
DbalProduct Entity that extends from the
Product Entity and is able to reconstitute the Entity from the database without using the
new operator, instead using a static factory method.
Embedded Value (Embeddables) with Doctrine >= 2.5.*
The latest stable Doctrine release is currently version 2.5 and it comes with support for mapping Value Objects, thereby removing the need to do this yourself as in Doctrine 2.4. Since December 2015, Doctrine also has support for nested embeddables. The support is not 100 percent, but it’s high enough to give it a try. In case it doesn’t work for your scenario, take a look at the next section. For official documentation, check the Doctrine Embeddables reference. This option, if implemented correctly, is definitely the one we recommend most. It would be the simplest, most elegant solution, that also provides search support in your DQL queries.
Currency classes have already been shown, the only thing remaining is to show the Doctrine mapping files:
In the product mapping, we’re defining
price as an instance variable that will hold a
Money instance. At the same time,
Money is designed with an amount and a
Finally, it’s time to show the Doctrine mapping for our
Currency Value Object:
As you can see, the above code has a standard embeddable definition with just one string field that holds the ISO code. This approach is the easiest way to use embeddables and is much more effective. By default, Doctrine names your columns by prefixing them using the Value Object name. You can change this behavior to meet your needs by changing the
column-prefix attribute in the XML notation.
Embedded Value with Doctrine <= 2.4.*
If you’re still stuck in Doctrine 2.4, you may wonder what an acceptable solution for using Embedded Values with Doctrine < 2.5 is. We need to now surrogate all the Value Object attributes in the
Product Entity, which means creating new artificial attributes that will hold the information of the Value Object. With this in place, we can map all those new attributes using Doctrine. Let’s see what impact this has on the
As you can see, there are two new attributes: one for the amount, and another for the ISO code of the currency. We’ve also updated the
setPrice method in order to keep attribute consistency when setting it. On top of this, we updated the
price getter in order to return the
Money Value Object built from the new fields. Let’s see how the corresponding XML Doctrine mapping should be changed:
If we wanted to push these two attributes outside of the Domain, this could be achieved through the use of an Abstract Factory. First, we need to create a new Entity,
DoctrineProduct, in our Infrastructure folder. This Entity will extend from
Product Entity. All surrogate fields will be placed in the new class, and methods such as
setPrice should be reimplemented. We’ll map Doctrine to use the new
DoctrineProduct as opposed to the
Now we’re able to fetch Entities from the database, but what about creating a new
Product? At some point, we’re required to call
new Product, but because we need to deal with
DoctrineProduct and we don’t want our Application Services to know about Infrastructure details, we’ll need to use Factories to create
Product Entities. So, in every instance where Entity creation occurs with
new, you’ll instead call
There could be many additional classes required to avoid placing the surrogate attributes in the original Entity. As such, it’s our recommendation to surrogate all the Value Objects to the same Entity, though this admittedly leads to a less pure solution.
Serialized LOB and Ad Hoc ORM
If the addition of searching capabilities to the Value Objects attributes is not important, there’s another pattern that can be considered: the Serialized LOB. This pattern works by serializing the whole Value Object into a string format that can easily be persisted and fetched. The most significant difference between this solution and the embedded alternative is that in the latter option, the persistence footprint requirements are reduced to a single column:
In order to persist
Product Entities using this approach, a change in the
DbalProductRepository is required. The
Money Value Object must be serialized into a string before persisting the final Entity:
Let’s see how our
Product is now represented in the database. The table column
price is a
TEXT type column that contains a serialization of a
Money object representing 9.99 USD:
This approach does the job. However, it’s not recommended due to problems occurring when refactoring classes in your code. Could you imagine the problems if we decided to rename our
Money class? Could you imagine the changes that would be required in our database representation when moving the
Money class from one namespace to another? Another tradeoff, as explained before, is the lack of querying capabilities. It doesn’t matter whether you use Doctrine or not; writing a query to get the products cheaper than, say, 200 USD is almost impossible while using a serialization strategy.
The querying issue can only be solved by using Embedded Values. However, the serialization refactoring problems can be fixed using a specialized library for handling serialization processes.
Improved Serialization with JMS Serializer
serialize/unserialize native PHP strategies have a problem when dealing with class and namespace refactoring. One alternative is to use your own serialization mechanism — for example, concatenating the amount, a one character separator such as “|,” and the currency ISO code. However, there’s another favored approach: using an open source serializer library such as JMS Serializer. Let’s see an example of applying it when serializing a
In order to unserialize the object, the process is straightforward:
With this example, you can refactor your
Money class without having to update your database. JMS Serializer can be used in many different scenarios — for example, when working with REST APIs. An important feature is the ability to specify which attributes of an object should be omitted in the serialization process — for example, a password.
Serialized LOB with Doctrine
In Doctrine, there are different ways of serializing objects in order to eventually persist them.
Doctrine Object Mapping Type
Doctrine has support for the Serialize LOB pattern. There are plenty of predefined mapping types you can use in order to match Entity attributes with database columns or even tables. One of those mappings is the
object type, which maps an SQL CLOB to a PHP object using
According to the Doctrine DBAL 2 Documentation,
maps and converts object data based on PHP serialization. If you need to store an exact representation of your object data, you should consider using this type as it uses serialization to represent an exact copy of your object as string in the database. Values retrieved from the database are always converted to PHP’s object type using unserialization or null if no data is present.
This type will always be mapped to the database vendor’s
texttype internally as there is no way of storing a PHP object representation natively in the database. Furthermore this type requires a SQL column comment hint so that it can be reverse engineered from the database. Doctrine cannot correctly map back this type correctly using vendors that do not support column comments, and will instead fall back to the text type instead.
Because the built-in text type of PostgreSQL does not support NULL bytes, the object type will result in unserialization errors. A workaround to this problem is to serialize()/unserialize() and base64_encode()/base64_decode() PHP objects and store them into a text field manually.
Let’s look at a possible XML mapping for the
Product Entity by using the
The key addition is
type="object", which tells Doctrine that we’re going to be using an
object mapping. Let’s see how we could create and persist a
Product Entity using Doctrine:
Let’s check that if we now fetch our
Product Entity from the database, it’s returned in an expected state:
Last but not least, the Doctrine DBAL 2 Documentation states that:
Object types are compared by reference, not by value. Doctrine updates this value if the reference changes and therefore behaves as if these objects are immutable value objects.
This approach suffers from the same refactoring issues as the ad hoc ORM did. The
object mapping type is internally using
serialize/unserialize. What about instead using our own serialization?
Doctrine Custom Types
Another option is to handle the Value Object persistence using a Doctrine Custom Type. A Custom Type adds a new mapping type to Doctrine — one that describes a custom transformation between an Entity field and the database representation, in order to persist the former.
As the Doctrine DBAL 2 Documentation explains:
Just redefining how database types are mapped to all the existing Doctrine types is not at all that useful. You can define your own Doctrine Mapping Types by extending
Doctrine\DBAL\Types\Type. You are required to implement 4 different methods to get this working.
object type, the serialization step includes information, such as the class, that makes it quite difficult to safely refactor our code. Let’s try to improve on this solution. Think about a custom serialization process that could solve the problem. One such way could be to persist the
Money Value Object as a string in the database encoded in
Using Doctrine, you’re required to register all Custom Types. It’s common to use an
EntityManagerFactory that centralizes this
EntityManager creation. Alternatively, you could perform this step by bootstrapping your application:
Now we need to specify in the mapping that we want to use our Custom Type:
Let’s check the database to see how the price was persisted using this approach:
This approach is an improvement on the one before in terms of future refactoring. However, searching capabilities remain limited due to the format of the column. With the Doctrine Custom types, you can improve the situation a little, but it’s still not the best option for building your DQL queries. See Doctrine Custom Mapping Types for more information.
Persisting a Collection of Value Objects
Imagine that we’d now like to add a collection of prices to be persisted to our
Product Entity. These prices could represent the different prices the product has borne throughout its lifetime or the product price in different currencies. This could be named
HistoricalPrice, as shown below:
HistoricalProduct extends from
Product, so it inherits the same behavior, plus the price collection functionality.
As in the previous sections, serialization is a plausible approach if you don’t care about querying capabilities. However, Embedded Values are a possibility if we know exactly how many prices we want to persist. But what happens if we want to persist an undetermined collection of historical prices?
Collection Serialized into a Single Column
Serializing a collection of Value Objects into a single column is most likely the easiest solution. Everything that was previously explained in the section about persisting a single Value Object applies in this situation. With Doctrine, you can use an Object or Custom Type — with some additional considerations to bear in mind: Value Objects should be small in size, but if you wish to persist a large collection, be sure to consider the maximum column length and the maximum row width that your database engine can handle.
Collection Backed by a Join Table
In case you want to persist and query an Entity by its related Value Objects, you have the choice to persist the Value Objects as Entities. In terms of the Domain, those objects would still be Value Objects, but we’ll need to give them an id and set them up with a one-to-many/one-to-one relation with the owner, a real Entity. To summarize, your ORM handles the collection of Value Objects as Entities, but in your Domain, they’re still treated as Value Objects.
The main idea behind the Join Table strategy is to create a table that connects the owner Entity and its Value Objects. Let’s see a database representation:
historical_products table will look the same as
products. Remember that
Product Entity in order to easily show how to deal with persisting a collection. A new
prices table is now required in order to persist all the different
Money Value Objects that a
Product Entity can handle:
Finally, a table that relates
prices is needed:
Collection Backed by a Join Table with Doctrine
Doctrine requires that all database Entities have a unique identity. Because we want to persist
Money Value Objects, we need to then add an artificial identity so Doctrine can handle its persistence. There are two options: including the surrogate identity in the
Money Value Object, or placing it in an extended class.
The issue with the first option is that the new identity is only required due to the Database persistence layer. This identity is not part of the Domain.
An issue with the second option is the amount of alterations required in order to avoid this so-called boundary leak. With a class extension, creating
new instances of the
Money Value Object class from any Domain Object isn’t recommended, as it would break the Inversion Principle. The solution is to again create a
Money Factory that would need to be passed into Application Services and any other Domain Objects.
In this scenario, we recommend using the first option. Let’s review the changes required to implement it in the
Money Value Object:
As seen above, two new attributes have been added. The first one,
surrogateId, is not used by our Domain, but it’s required for the persistence Infrastructure to persist this Value Object as an Entity in our Database. The second one,
surrogateCurrencyIsoCode, holds the ISO code for the currency. Using these new attributes, it’s really easy to map our Value Object with Doctrine.
Money mapping is quite straightforward:
Using Doctrine, the
HistoricalProduct Entity would have following mapping:
Collection Backed by a Join Table with an Ad Hoc ORM
It’s possible to do the same with an ad hoc ORM, where Cascade INSERTS and JOIN queries are required. It’s important to be careful about how the removal of Value Objects is handled, in order to not leave orphan
Money Value Objects.
Collection Backed by a Database Entity
Database Entity is the same solution as Join Table, with the addition of the Value Object that’s only managed by the owner Entity. In the current scenario, consider that the
Money Value Object is only used by the
HistoricalProduct Entity; a Join Table would be overly complex. So the same result could be achieved using a one-to-many database relation.
What about NoSQL mechanisms such as Redis, MongoDB, or CouchDB? Unfortunately, you can’t escape these problems. In order to persist an Aggregate using Redis, you need to serialize it into a string before setting the value. If you use PHP
serialize/unserialize methods, you’ll face namespace or class name refactoring issues again. If you choose to serialize in a custom way (JSON, custom string, etc.), you’ll be required to again rebuild the Value Object during Redis retrieval.
PostgreSQL JSONB and MySQL JSON Type
If our database engine would allow us to not only use the Serialized LOB strategy but also search based on its value, we would have the best of both approaches. Well, good news: now you can do this. As of PostgreSQL version 9.4, support for JSONB has been added. Value Objects can be persisted as JSON serializations and subsequently queried within this JSON serialization.
- Automatic validation of JSON documents stored in JSON columns. Invalid documents produce an error.
- Optimized storage format. JSON documents stored in JSON columns are converted to an internal format that permits quick read access to document elements. When the server later must read a JSON value stored in this binary format, the value need not be parsed from a text representation. The binary format is structured to enable the server to look up subobjects or nested values directly by key or array index without reading all values before or after them in the document.
If Relational Databases add support for document and nested document searches with high performance and with all the benefits of an ACID (Atomicity, Consistency, Isolation, Durability) philosophy, it could save a lot of complexity in many projects.
Another interesting detail of modeling your Domain concepts using Value Objects is regarding its security benefits. Consider an application within the context of selling flight tickets. If you deal with International Air Transport Association airport codes, also known as IATA codes, you can decide to use a string or model the concept using a Value Object. If you choose to go with the string, think about all the places where you’ll be checking that the string is a valid IATA code. What’s the chance you forget somewhere important? On the other hand, think about trying to instantiate
new IATA("BCN'; DROP TABLE users;--"). If you centralize the guards in the constructor and then pass an IATA Value Object into your model, avoiding SQL Injections or similar attacks gets easier.
Using Value Objects for modeling concepts in your Domain that measure, quantify, or describe is highly recommended. As shown, Value Objects are easy to create, maintain, and test. In order to handle persistence within a Domain-Driven Design application, using an ORM is a must. However, in order to persist Value Objects using Doctrine, the preferred way is using embeddables. In case you’re stuck in version 2.4, there are two options: adding the Value Object fields directly into your Entity and mapping them (less elegant, but easier), or extending your Entities (far more elegant, but more complex).
Beck, Kent. Test-Driven Development: By Example. Addison-Wesley Professional, 2002.
Brandolini, Alberto. Introducing EventStorming. Leanpub, 2016.
Evans, Eric. Domain-Driven Design Reference: Definitions and Pattern Summaries. Dog Ear Publishing, 2014.
Evans, Eric. Domain-Driven Design: Tackling Complexity in the Heart of Software. Addison-Wesley Professional, 2003.
Fowler, Martin. Patterns of Enterprise Application Architecture. Addison-Wesley Professional, 2002.
Hohpe, Gregor, and Bobby Woolf. Enterprise Integration Patterns: Designing, Building, and Deploying Messaging Solutions. Addison-Wesley Professional, 2012.
Martin, Robert C. Agile Software Development, Principles, Patterns, and Practices. Pearson, 2002.
Martin, Robert C. Clean Code: A Handbook of Agile Software Craftsmanship. Prentice Hall, 2008.
Meszaros, Gerard. xUnit Test Patterns: Refactoring Test Code. Addison-Wesley Professional, 2007.
Newman, Sam. Building Microservices. O’Reilly Media, 2015.
Nilsson, Jimmy. Applying Domain-Driven Design and Patterns: With Examples in C# and .NET. Addison-Wesley Professional, 2006.
Sadalage, Pramod J., and Martin Fowler. NoSQL Distilled: A Brief Guide to the Emerging World of Polyglot Persistence. Addison-Wesley Professional, 2012.
Vernon, Vaughn. Domain-Driven Design Distilled. Addison-Wesley Professional, 2016.
Vernon, Vaughn. Implementing Domain-Driven Design. Addison-Wesley Professional, 2013.
Appendix: Hexagonal Architecture with PHP
The following article was posted in php|architect magazine in June 2014 by Carlos Buenosvinos.
With the rise of Domain-Driven Design (DDD), architectures promoting domain centric designs are becoming more popular. This is the case with Hexagonal Architecture, also known as Ports and Adapters, that seems to have being rediscovered just now by PHP developers. Invented in 2005 by Alistair Cockburn, one of the Agile Manifesto authors, the Hexagonal Architecture allows an application to be equally driven by users, programs, automated tests or batch scripts, and to be developed and tested in isolation from its eventual run-time devices and databases. This results into agnostic infrastructure web applications that are easier to test, write and maintain. Let’s see how to apply it using real PHP examples.
Your company is building a brainstorming system called Idy. Users add and rate ideas so the most interesting ones can be implemented in a company. It is Monday morning, another sprint is starting and you are reviewing some user stories with your team and your Product Owner. “As a not logged in user, I want to rate an idea and the author should be notified by email”, that’s a really important one, isn’t it?
As a good developer, you decide to divide and conquer the user story, so you’ll start with the first part, “I want to rate an idea”. After that, you will face “the author should be notified by email”. That sounds like a plan.
In terms of business rules, rating an idea is as easy as finding the idea by its identifier in the ideas repository, where all the ideas live, add the rating, recalculate the average and save the idea back. If the idea does not exist or the repository is not available we should throw an exception so we can show an error message, redirect the user or do whatever the business asks us for.
In order to execute this UseCase, we just need the idea identifier and the rating from the user. Two integers that would come from the user request.
Your company web application is dealing with a Zend Framework 1 legacy application. As most of companies, probably some parts of your app may be newer, more SOLID, and others may just be a big ball of mud. However, you know that it does not matter at all which framework you are using, it is all about writing clean code that makes maintenance a low cost task for your company.
You’re trying to apply some Agile principles you remember from your last conference, how it was, yeah, I remember “make it work, make it right, make it fast”. After some time working you get something like Listing 1.
I know what readers are thinking: “Who is going to access data directly from the controller? This is a 90’s example!”, ok, ok, you’re right. If you are already using a framework, it is likely that you are also using an ORM. Maybe done by yourself or any of the existing ones such as Doctrine, Eloquent, ZendDB, etc. If this is the case, you are one step further from those who have some Database connection object but don’t count your chickens before they’re hatched.
For newbies, Listing 1 code just works. However, if you take a closer look at the Controller, you’ll see more than business rules, you’ll also see how your web framework routes a request into your business rules, references to the database or how to connect to it. So close, you see references to your infrastructure.
Infrastructure is the detail that makes your business rules work. Obviously, we need some way to get to them (API, web, console apps, etc.) and effectively we need some physical place to store our ideas (memory, database, NoSQL, etc.). However, we should be able to exchange any of these pieces with another that behaves in the same way but with different implementations. What about starting with the Database access?
Zend_DB_Adapter connections (or straight MySQL commands if that’s your case) are asking to be promoted to some sort of object that encapsulates fetching and persisting Idea objects. They are begging for being a Repository.
Repositories and the Persistence Edge
Whether there is a change in the business rules or in the infrastructure, we must edit the same piece of code. Believe me, in CS, you don’t want many people touching the same piece of code for different reasons. Try to make your functions do one and just one thing so it is less probable having people messing around with the same piece of code. You can learn more about this by having a look at the Single Responsibility Principle (SRP). For more information about this principle: http://www.objectmentor.com/resources/articles/srp.pdf
Listing 1 is clearly this case. If we want to move to Redis or add the author notification feature, you’ll have to update the
rateAction method. Chances to affect aspects of the
rateAction not related with the one updating are high. Listing 1 code is fragile. If it is common in your team to hear “If it works, don’t touch it”, SRP is missing.
So, we must decouple our code and encapsulate the responsibility for dealing with fetching and persisting ideas into another object. The best way, as explained before, is using a Repository. Challenged accepted! Let’s see the results in Listing 2.
The result is nicer. The
rateAction of the
IdeaController is more understandable. When read, it talks about business rules.
IdeaRepository is a business concept. When talking with business guys, they understand what an
IdeaRepository is: A place where I put Ideas and get them.
A Repository “mediates between the domain and data mapping layers using a collection-like interface for accessing domain objects.” as found in Martin Fowler’s pattern catalog.
If you are already using an ORM such as Doctrine, your current repositories extend from an
EntityRepository. If you need to get one of those repositories, you ask Doctrine
EntityManager to do the job. The resulting code would be almost the same, with an extra access to the
EntityManager in the controller action to get the
At this point, we can see in the landscape one of the edges of our hexagon, the persistence edge. However, this side is not well drawn, there is still some relationship between what an
IdeaRepository is and how it is implemented.
In order to make an effective separation between our application boundary and the infrastructure boundary we need an additional step. We need to explicitly decouple behavior from implementation using some sort of interface.
Decoupling Business and Persistence
Have you ever experienced the situation when you start talking to your Product Owner, Business Analyst or Project Manager about your issues with the Database? Can you remember their faces when explaining how to persist and fetch an object? They had no idea what you were talking about.
The truth is that they don’t care, but that’s ok. If you decide to store the ideas in a MySQL server, Redis or SQLite it is your problem, not theirs. Remember, from a business standpoint, your infrastructure is a detail. Business rules are not going to change whether you use Symfony or Zend Framework, MySQL or PostgreSQL, REST or SOAP, etc.
That’s why it is important to decouple our IdeaRepository from its implementation. The easiest way is to use a proper interface. How can we achieve that? Let’s take a look at Listing 3.
Easy, isn’t it? We have extracted the
IdeaRepository behavior into an interface, renamed the IdeaRepository into
MySQLIdeaRepository and updated the
rateAction to use our
MySQLIdeaRepository. But what’s the benefit?
We can now exchange the repository used in the controller with any implementing the same interface. So, let’s try a different implementation.
Migrating our Persistence to Redis
During the sprint and after talking to some mates, you realize that using a NoSQL strategy could improve the performance of your feature. Redis is one of your best friends. Go for it and show me your Listing 4.
Easy again. You’ve created a
RedisIdeaRepository that implements
IdeaRepository interface and we have decided to use Predis as a connection manager. Code looks smaller, easier and faster. But what about the controller? It remains the same, we have just changed which repository to use, but it was just one line of code.
As an exercise for the reader, try to create the IdeaRepository for SQLite, a file or an in-memory implementation using arrays. Extra points if you think about how ORM Repositories fit with Domain Repositories and how ORM @annotations affect this architecture.
Decouple Business and Web Framework
We have already seen how easy it can be to changing from one persistence strategy to another. However, the persistence is not the only edge from our Hexagon. What about how the user interacts with the application?
Your CTO has set up in the roadmap that your team is moving to Symfony2, so when developing new features in you current ZF1 application, we would like to make the incoming migration easier. That’s tricky, show me your Listing 5.
Let’s review the changes. Our controller is not having any business rules at all. We have pushed all the logic inside a new object called
RateIdeaUseCase that encapsulates it. This object is also known as Controller, Interactor or Application Service.
The magic is done by the
execute method. All the dependencies such as the
RedisIdeaRepository are passed as an argument to the constructor. All the references to an
IdeaRepository inside our UseCase are pointing to the interface instead of any concrete implementation.
That’s really cool. If you take a look inside
RateIdeaUseCase, there is nothing talking about MySQL or Zend Framework. No references, no instances, no annotations, nothing. It is like your infrastructure does not mind. It just talks about business logic.
Additionally, we have also tuned the Exceptions we throw. Business processes also have exceptions. NotAvailableRepository and IdeaDoesNotExist are two of them. Based on the one being thrown we can react in different ways in the framework boundary.
Sometimes, the number of parameters that a UseCase receives can be too many. In order to organize them, it is quite common to build a UseCase request using a Data Transfer Object (DTO) to pass them together. Let’s see how you could solve this in Listing 6.
The main changes here are introducing two new objects, a Request and a Response. They are not mandatory, maybe a UseCase has no request or response. Another important detail is how you build this request. In this case, we are building it getting the parameters from ZF request object.
Ok, but wait, what’s the real benefit? it is easier to change from one framework to other, or execute our UseCase from another delivery mechanism. Let’s see this point.
Rating an idea using the API
During the day, your Product Owner comes to you and says: “by the way, a user should be able to rate an idea using our mobile app. I think we will need to update the API, could you do it for this sprint?”. Here’s the PO again. “No problem!”. Business is impressed with your commitment.
As Robert C. Martin says: “The Web is a delivery mechanism […] Your system architecture should be as ignorant as possible about how it is to be delivered. You should be able to deliver it as a console app, a web app, or even a web service app, without undue complication or any change to the fundamental architecture”.
Your current API is built using Silex, the PHP micro-framework based on the Symfony2 Components. Let’s go for it in Listing 7.
Is there anything familiar to you? Can you identify some code that you have seen before? I’ll give you a clue.
“Man! I remember those 3 lines of code. They look exactly the same as the web application”. That’s right, because the UseCase encapsulates the business rules you need to prepare the request, get the response and act accordingly.
We are providing our users with another way for rating an idea; another delivery mechanism.
The main difference is where we created the
RateIdeaRequest from. In the first example, it was from a ZF request and now it is from a Silex request using the parameters matched in the route.
Console app rating
Sometimes, a UseCase is going to be executed from a Cron job or the command line. As examples, batch processing or some testing command lines to accelerate the development.
While testing this feature using the web or the API, you realize that it would be nice to have a command line to do it, so you don’t have to go through the browser.
If you are using shell scripts files, I suggest you to check the Symfony Console component. What would the code look like?
Again those 3 lines of code. As before, the UseCase and its business logic remain untouched, we are just providing a new delivery mechanism. Congratulations, you’ve discovered the user side hexagon edge.
There is still a lot to do. As you may have heard, a real craftsman does TDD. We have already started our story so we must be ok with just testing after.
Testing Rating an Idea UseCase
Michael Feathers introduced a definition of legacy code as code without tests. You don’t want your code to be legacy just born, do you?
In order to unit test this UseCase object, you decide to start with the easiest part, what happens if the repository is not available? How can we generate such behavior? Do we stop our Redis server while running the unit tests? No. We need to have an object that has such behavior. Let’s use a mock object in Listing 9.
NotAvailableRepository has the behavior that we need and we can use it with
RateIdeaUseCase because it implements
Next case to test is what happens if the idea is not in the repository. Listing 10 shows the code.
Here, we use the same strategy but with an
EmptyIdeaRepository. It also implements the same interface but the implementation always returns
null regardless which identifier the
find method receives.
Why are we testing these cases?, remember Kent Beck’s words: “Test everything that could possibly break”.
Let’s carry on with the rest of the feature. We need to check a special case that is related with having a read available repository where we cannot write to. Solution can be found in Listing 11.
Ok, now the key part of the feature is still remaining. We have different ways of testing this, we can write our own mock or use a mocking framework such as Mockery or Prophecy. Let’s choose the first one. Another interesting exercise would be to write this example and the previous ones using one of these frameworks.
Bam! 100% Coverage for the UseCase. Maybe, next time we can do it using TDD so the test will come first. However, testing this feature was really easy because of the way decoupling is promoted in this architecture.
Maybe you are wondering about this:
We need a way to guarantee that the update method has been called during the UseCase execution. This does the trick. This test double object is called a spy, mocks cousin.
When to use mocks? As a general rule, use mocks when crossing boundaries. In this case, we need mocks because we are crossing from the domain to the persistence boundary.
What about testing the infrastructure?
If you want to achieve 100% coverage for your whole application you will also have to test your infrastructure. Before doing that, you need to know that those unit tests will be more coupled to your implementation than the business ones. That means that the probability to be broken with implementation details changes is higher. So it is a trade-off you will have to consider.
So, if you want to continue, we need to do some modifications. We need to decouple even more. Let’s see the code in Listing 13.
If we want to 100% unit test
RedisIdeaRepository we need to be able to pass the
Predis\Client as a parameter to the repository without specifying TypeHinting so we can pass a mock to force the code flow necessary to cover all the cases.
This forces us to update the Controller to build the Redis connection, pass it to the repository and pass the result to the UseCase.
Now, it is all about creating mocks, test cases and having fun doing asserts.
Arggg, So Many Dependencies!
Is it normal that I have to create so many dependencies by hand? No. It is common to use a Dependency Injection component or a Service Container with such capabilities. Again, Symfony comes to the rescue, however, you can also check PHP-DI 4 http://php-di.org/.
Let’s see the resulting code in Listing 14 after applying Symfony Service Container component to our application.
The controller has been modified to have access to the container, that’s why it is inheriting from a new base controller
ContainerAwareController that has a
get method to retrieve each of the services contained.
In Listing 15, you can also find the XML file used to configure the Service Container. It is really easy to understand but if you need more information, take a look to the Symfony Service Container Component site in http://symfony.com/doc/current/book/service_container.html
Domain Services and Notification Hexagon Edge
Are we forgetting something? “the author should be notified by email”, yeah! That’s true. Let’s see in Listing 16 how we have updated the UseCase for doing the job.
As you realize, we have added a new parameter for passing
AuthorNotifier Service that will send the email to the author. This is the port in the “Ports and Adapters” naming. We have also updated the business rules in the
Repositories are not the only objects that may access your infrastructure and should be decoupled using interfaces or abstract classes. Domain Services can too. When there is a behavior not clearly owned by just one Entity in your domain, you should create a Domain Service. A typical pattern is to write an abstract Domain Service that has some concrete implementation and some other abstract methods that the adapter will implement.
As an exercise, define the implementation details for the AuthorNotifier abstract service. Options are SwiftMailer or just plain
In order to have a clean architecture that helps you create easy to write and test applications, we can use Hexagonal Architecture. To achieve that, we encapsulate user story business rules inside a UseCase or Interactor object. We build the UseCase request from our framework request, instantiate the UseCase and all its dependencies and then execute it. We get the response and act accordingly based on it. If our framework has a Dependency Injection component you can use it to simplify the code.
The same UseCase objects can be used from different delivery mechanisms in order to allow users access the features from different clients (web, API, console, etc.)
For testing, play with mocks that behave like all the interfaces defined so special cases or error flows can also be covered. Enjoy the good job done.
In almost all the blogs and books you will find drawings about concentric circles representing different areas of software. As Robert C. Martin explains in his “Clean Architecture” post, the outer circle is where your infrastructure resides. The inner circle is where your Entities live. The overriding rule that makes this architecture work is The Dependency Rule. This rule says that source code dependencies can only point inwards. Nothing in an inner circle can know anything at all about something in an outer circle.
Use this approach if 100% unit test code coverage is important to your application. Also, if you want to be able to switch your storage strategy, web framework or any other type of third-party code. The architecture is especially useful for long-lasting applications that need to keep up with changing requirements.
If you are interested in learning more about Hexagonal Architecture and other near concepts you should review the related URLs provided at the beginning of the article, take a look at CQRS and Event Sourcing. Also, don’t forget to subscribe to google groups and RSS about DDD such as http://dddinphp.org and follow on Twitter people like @VaughnVernon, and @ericevans0.