Domain-Driven Design

Software development is a creative process. You do not stand on an assembly line, but are increasingly exposed to new challenges. Each software differs from each other, and every time you start with a new program you start from scratch. You have to find out, what your client wants to achieve and after that you want to implement this. Though in general you have great freedom for your development misunderstandings between the customer and you as a developer can rise. There is hardly anyone who never has been working a project in which the requirements were documented and the specifications were written but still the client was not satisfied at the end or in worst case the software hasn't solved his problems at all.

If you develop software for customers, you must first understand their problems, so you can offer your customer a tailored solution. This problem area is called the application domain within the terms of domain-driven design. By knowing the application domain of customers, understanding their problems and having them clearly in mind, you are in a position to adequately implement this domain in software. Extbase supports you in supplying you with the required technical infrastructure.

Domain-Driven Design is a development paradigm that does not include technical concepts. Instead, by using different techniques, extbase tries to to direct the creativity of the software development to structured directions and thus channel it and make it more effective.

In domain-driven design your understanding of the relevant problems and their environment is the focus: a problem which the customer wishes to get resolved by your program. In close cooperation with the customer the problem is collaboratively explored. During this process, which proceeds iteratively (i.e. stepwise), you will develop a model together with the customer that represents the problem adequately. This model is the base of the generated program. By focusing on the model you are not distracted with the actual problem of the customer during prototyping - you can focus on the problem domain.

Note

The involvement of customers in the working phase is absolutely essential because only they know the problem well enough.

Domain-Driven Design is a very pragmatic approach. Yet at the beginning of a software project you are writing code to explore the problem and its dimensions that shall be solved. Often you will need a few prototypes and iterations, until you reach the final model. Take this time!

Extbase offers you a variety support of Domain-Driven Design. You do not have to care for e.g. storing data in the database. If the domain contains complex invariants (e.g. rules that must not be violated) you can implement these elegantly using so-called validators. In addition, you have fluid as a powerful templating engine for efficient output of data available. So you can focus on the problem domain, rather than investing much time in the output of the data and user interaction.

We now hope that your appetite is aroused! Below we show you some of the core concepts of the Domain-driven design. These include certain approaches such as the elaboration of a language used by all (the so-called Ubiquitous Language) or dealing with associations. Besides these we also show you the technical elements such as entities, value objects and repositories.

Develop a common language

In software projects, people are involved in different roles: the customer is an expert in his business area and has a problem he wants to get solved by a program. As he is very well acquainted with the application domain of this program, we refer to him as a so-called domain expert.

You are usually in the role of the developer: You are familiar with many technical possibilities but you do usually not know the business of the customer. Therefore, misunderstandings are inevitable, because the developer uses a very different language from the domain expert.

In Domain-driven design, important core terms that are needed for the characterization and solution of the problem should be determined. These will be compiled in the form a glossary to ensure always that the domain expert and the developer understand each other correctly.

This so-called ubiquitous language does not apply only during communication: it should be related in the source code (e.g. class or method names) as well. This enables experts to capture the problems within the domain on the one hand and domain expert to take decisions on the basis of the source code on the other hand on whether the business logic has been implemented correctly.

To model the domain

During your meeting with the domain expert you will start to create a model and after that you will refine it. Usually during the dialog a UML-Diagram will arise that reflects the features, functions and relationships of the relevant components of the problem area. These objects are called domain objects because of its focus on the problem domain of the user.

Domain-driven design provides some building blocks that will help you in creating a good domain model: Entities, Value Objects and Services. First, let's concentrate on the first two building blocks: Does the domain object has its own identity, which is received over time, even if the object goes through different states? Then the object is an entity. Or is it an attribute that describes other things in detail? Then the object is of type value object. Hereafter we will go into more detail on the distinction between these two types.

There are also concepts that couldn't assigned directly to be entities or value objects - this happens every time you speak of activities during the modeling phase. For this purpose the concept of services is introduced.

Entities

Entities are objects that possess a unique identity. For example, a user has an username as an identity, a product has a product number, and a student has a student number. Therefore, all examples mentioned are entities. The identity of the object remains the same over time, even if the properties of the object changes.

The identity of an object is defined by an immutable property or a combination of several immutable properties of the object. At least one property must be identity-determining. Normally extbase takes care about automatically and therefore assigns a unique value to a (hidden) identity property. You can also select this by yourself and select which combination of properties should be identity-determining.

Note

Extbase uses as an automatically generated identity property as identifier, which is generated by the underlying database (the so-called Unique identifier, UID). The word unique clearly means in this context really "unique within a database table". This will, however, possibly change in a future version of Extbase to ensure a global uniqueness.

Depending on the application, it may be useful to establish their own identity-determining properties that are important within your domain. How can I identify a human clearly? The passport number and country - taken together- perhaps are such a clear identification, but is often less practical: Who wants to give his passport number if he logs into a Web site? For an internet forum, it's practical, for example, to determine the identity of the forum member with his mail address. But if you write an e-government application, the identity card number would considered instead.

It is important that the identity-determining properties of an object are set when it is created and after that never changes: If you make these properties changeable, different identities could be converted together - but this is undesirable in most cases. Because you have to promote the new identity to all objects, which know the object under his old identity. Otherwise you will loose all connections from other objects to this object. In general, the change of identity-determining properties should be prohibited by exclusive read access to these properties.

If you have written databased applications, you may have noticed that you have often used entities subconsciously, by having provided the database tables with indexes and primary keys (in TYPO3, for example, the UID will always added as an identifier). Therefore, you may wonder why we now need a different type of object at all. In the next section we answer this question.

Value Objects

PHP offers several build-in value-types, such as integer, float or string. Often you will notice that you need domain specific values-types​​, such as colors or tags. These are represented by so-called Value Objects.

Value Objects are objects that are determined by the values ​​of their properties. Take a graphic program as an example: Somewhere it must be defined, what is meant by a "color". A color is determined only by its value, such as the three color components for red, green and blue in RGB mode. If two objects have the same RGB color values, so they are effectively the same and must not continue be distinguished.

Within value objects all properties are identity-determining. If all of the properties of two value objects have the same value, then these two objects are the same. Nevertheless, value objects are often longer more than just simple data structures for primitive data types. They can contain potentially much complex domain logic. Our color-object could - for example - use methods for converting color to other color spaces like CMYK or HSV and could include color profiles for this.

Since all properties are identity-determining and as it is not allowed to change these properties after the creation of the object, value objects are immutable. They are generated completely initialized and after that they can't change their value. You can only create a new value object and eliminate the old.

Note

Although you could provide methods for changing the internal state of a value object, it is not allowed to change the state at any time. Take the example of the color object, which could contain a new method "makeBrighter()". This method must change the color value and give back a new color object with these changed values. It must not change the existing object.

By this simple semantics value objects can easily generated, cloned, transmitted to other computers or transfered to other objects. This not only the implementation is easier, but it is also clearly communicated that these objects are just simple values​​.

In an application that optimizes the delivery of letters for the post office, addresses can be associated with other characteristics such as the name of the postman who delivered the mail. This name, however, belongs not to the identity of the object and can change over time (e.g. if a postman retired) - a clear reference to the usage of an entity.

So you see: you can not always say clearly whether objects are entities or value objects - it depends entirely on the application and the application domain.

Note

The distinction between entities and value objects will be perhaps difficult for you at the beginning and appear as an unnecessary expense. Extbase treats the two object types very different in the background. The administration of value-objects is more efficient than those of entities. The additional expenses for the administration and monitoring of the uniqueness omitted here, for example, completely.

Associations

You should never leave the implementation out of sight during the modeling. So let us talk briefly about a very complex field of implementation: associations between objects.

Domain objects are related to each other. Such relationships are called in the language of the domain with the following phrases: A "consists of" B, C 'has' D, E "processed" F, or G 'belongs to' I. These relationships are referred in abstract domain model as associations.

At a university professors and students are in relation of each other: the professor lectures and students are enrolled for classes. To reflect this relationship in our domain model, we add an association as an abstraction of the real world. Practically, this means that an professor object contains a list of pointers to the student objects, who sit with him in the lecture.

Particularly complicated to implement are many-to-many associations here, as shown in the above example, (a professor taught many students and a student is taught by various professors) - and moreover, if these associations are bidirectional. This means that the association can point from a professor to his students, but also in the other direction.

If you use many-to-many associations during the design, consider to simplify and restructure them. It is nature that you use a great number of bidirectional many-to-many associations especially at the beginning of the modeling. In the refinement of associations you can find help with the following questions:

  1. Is the many-to-many association important in the application domain?
  2. Can the association be made ​​unilaterally, as there is a main direction in which the objects are queried?
  3. Could the association be specified in greater detail, e.g. by qualifying the individual elements more closely?
  4. Is the association for the core functionality needed at all?

So remember to use very simple associations, as it is easier to implement them and they are better understandable.

Aggregates

If you build a complex domain model, you have to deal with many classes offered at the same hierarchical level. Often it is given that certain objects are part of a larger object. If you want to model an application for a garage, so maybe you have to model not only the car but the engine and the wheels too, because these are of particular importance for the garage. Intuitively, we look at the wheels and the engine of a car as part of the car, so this understanding should be visible in the model as well. We call such a relationship between closely related parts and the whole "aggregates". You see this domain model in Figure 2-1.

../_images/figure-2-1.png

Figure 2-1: The domain model of an auto repair shop. Objects outside an aggregate may only reference on the aggregate root.

An aggregate possesses a root object, the so-called "aggregate root". This is responsible for the integrity of their sub-objects. Objects outside of the aggregate must only refer to the aggregate root, but never to parts of it, because otherwise the units couldn't ensure the integrity of the root objects. To the outside, the units have only an external identity: the aggregate root. As aggregate roots need an identity, which is used for referencing, their type has to be an entity.

Transferred to the car analogy it is like this: The service station may not maintain a permanent reference to the engine, but needs to remember a permanent reference to the car (e.g. by the vehicle number as the external identity). If you require a reference to engine for your work - you can reach it via the car.

Through this rules of referencing the domain will be structured further, which reduces the complexity of the application and makes it manageable.

So far we have shown how real world objects can be mapped into entities and value objects objects. However, there are concepts in the world that do not fit into this scheme. To reflect this, we introduce services.

Services

In practice there are actions while modeling an application, which could not directly assigned to certain domain objects. In object-oriented programming, you are trying to force entities or value objects to these actions, although it does not really belongs in there. To circumvent this problem, there exist so-called services. These are containers for actions that may belong to the domain of application but can't be assigned to any particular object.

A service should be stateless, ie should not use or manipulate internal states. A service should be used, without knowing its internal state to be known be taken into account. A service often receives entities or value objects as input and performs complex operations on them.

Lifecycle of objects

In the real world objects have a certain life cycle. A car is built, then it changes during its lifetime (the mileage increases, brakes are replaced, wear, ...), and at some point the car is scrapped.

Since we model a domain with domain-driven design, which has a counterpart in the real world, the life cycle of objects in our program are very similar to the objects in the real world. Objects are created at a time, then they are active and can be changed, and eventually they will be deleted. This is shown in Figure 2-2.

../_images/figure-2-2.png

Figure 2-2: The life cycle of an object in the real world

We can not always keep all existing objects instantiated in memory of our program of course - our program would be unusable slow and memory-hungry (not to mention the case where the power fails and then the objects are gone).

Chapter 2.2 / page 34

So we need a way to maintain only needed objects in the memory. The active state actually consists of several sub-states, which are shown in Figure 2-3.

../_images/figure-2-3.png

Figure 2-3: The life cycle of an object in extbase is more complex, because the object can be stored in the data storage (like a database).

When an object is newly created, it is transient, i.e. at the end of the current request PHP will remove the object from the memory: It will be deleted. To change object to be permanent, i.e. over multiple requests, it must be converted from a transient object to a persistent object. Repositories are responsible for this. These allow the permanent storage and retrieval of objects based on certain criteria. But how to use repositories now in practice? While you add an object to a repository, it will be persisted. Now the repository is responsible for the object. It automatically cares about the storing of an object at the end of a request.

You can also get back an object reference from the repository when you need it - the repository will reconstitute the object in this case automatically from the database.

It is important that the logical object still exists when it is stored in the database. Just for performance reasons it is not held in the memory. It is very important to distinguish between the creation of an object and the reconstitution of the object from the database. Imagine therefore that the objects continue to exist in the database, only in a different form of representation.

Note

The constructor of an object is only called when creating the object. If the object is reconstituted from the database, the constructor is not called because the object still exists logical.

You can retransform a persistent object in a transient object by explicitly removing it from the repository. This means that the repository for this object has no responsibility anymore. At the end of a request the object is then deleted.

Extbase does the work for you in terms of persistence of objects in the database as much as possible. You are not in touch with database layer directly, as Extbase cares about the whole life cycle of the objects.

Now - as you have learned many things about the life cycle of objects - we want to substantiate two pieces of the life cycle: the creation of objects and the reconstitution of objects.

Create objects with the help of factories

Now that you know the life cycle of objects more accurately, we will deal initially with the creation of objects. You are allowed to produce only self-consistent aggregates. In the car example from earlier, this means that, in preparing the car, the wheels and the engine must be created and immediately too, because the car object is in an inconsistent state otherwise.

In simple initializations, it is recommended that you use the constructor of the aggregate root for these purposes. If a complex object networks are built with many cross connections, then you should move that functionality into its own factory. This is a class that assembles complex objects and gives them back finished.

The following is an example of the initialization of the cars in the constructor of the aggregate root:

<?php
class Car
{
   protected $engine;
   protected $wheels;

   public function __construct()
   {
      $this->engine = new Engine();
      $this->wheels[0] = new Wheel();
      $this->wheels[1] = new Wheel();
      $this->wheels[2] = new Wheel();
      $this->wheels[3] = new Wheel();
   }
}

For simplicity we have omitted the base classes and the full class names to show you the essence: As the constructor is executed at creating an object, it is always built a consistent object.

Note

In TYPO3 you can not generate classes with the new operator, but with GeneralUtility::makeInstance (className). In the example above, we wanted to concentrate on the essentials, so we have used new there.

Reconstitute objects with repositories

You can imagine a repository like a library: go to the circulation desk and ask for a specific book (based on certain criteria such as the title or author). If the book is available, the librarian will get it and gives it to you. You do not have to know on which shelf the book is or whether it perhaps has to be delivered from another library for you. Now you can read the book and find perhaps a typo and correct it with a pencil. After expiry of the loan period, you have to give the book back again, and then the next person will borrow it - of course with your corrections are still exists in the book.

But how will new books get into the library? You can, for example, donate books that you have read, to the library. Here, a librarian will write the title of the book, the author and some keywords in the central library database , so the book can be found and borrowed by other users. Conversely, a book will be sorted out if it is old and broken. Of course, the entry in the library database have to be deleted, so the book can no longer be found.

With a repository, it behaves like a library.With a repository you can find objects of a certain type. If you send a query like findByTitle('Domain-Driven Design') to the BookRepository for example, you get all objects back where the title is "Domain-driven Design". If you now change a book-object (for example, by correcting a typo in the table of contents), these changes are saved automatically, and the next search operation, will return the revised object.

So how can you make a repository responsible for an object? For this the repository has the method add($object). If you want to commit a new object to the BookRepository for example, you can create it using $book = new Book('Extbase and Fluid'). A new book titled "Extbase and Fluid" can added to the BookRepository with add($book). Similarly, you can remove an object from a repository, by calling the method remove($object). Now the object is not findable in the repository anymore and therefore it is deleted from the database.

For each aggregate root exactly one repository has to exists, which is responsible for that object type and his sub objects. By using this repository, you can then locate the desired aggregate root object by different criteria. Conversely, this means: In extbase you define an object type as aggregate root object by creating a repository for this type.

We have now explained how the domain of the application can be efficiently packed into a software model. Therefore, we have explained "Domain-Driven Design" as a "toolbox" of techniques, which are supported by extbase. But an growing application consists not only of the model: presentation logic is important too. With an effective separation of model and presentation logic we will continue in the following section.