software design

This book may change the way you think about software design. Of all the software books I purchased over the last year, this book ranks as one of the most important (another is Lean Software Development by Mary and Tom Poppendieck). The book presents an elegant pattern language to help the modeler understand and design the domain in which he is crafting software. The pattern language addresses both the human factor of modeling (i.e., effective communication with all stakeholders) and the technique of modeling (i.e., the application of object concepts).

This well-organized book succeeds on many levels:

* It helps prepare the modeler to work with his domain by suggesting he must develop the right mindset and communicate with the people who understand the domain.

* Emphasizing the need to keep the domain model as pure as possible, it explains the absolute fundamentals of a domain model, including the all-important notions of entities, value objects, services, and modules. It presents the object-lifecycle, including the mechanisms of factory and repository (and their advantages).

* It provides techniques for object discovery and refinement and model flexibility.

* It provides ideas for organizing cooperative teams.

* It is highly compatible with agile processes, such as eXtreme Programming (XP). In fact, the author himself has practiced XP.

This book should be the second book a novice modeler should purchase. A seasoned modeler should buy this book immediately.

If you have even been involved in a software project (a) as a developer and did not know what the end product is going to be used for or how it will be used or (b) as an architect who spent countless hours with your stakeholders and domain experts trying to figure out how to go about architecting your application, then you should read this book. Read it again after you have read it for the first time. This book is packed with pointers, information, tips, how-tos, "down to earth" practical samples, and even conversational examples that one could have while gathering requirements. Evans in his book fills a wide gap that we all tend to come across while designing software applications.

There are many software engineering processes out there, and each one tries to tackle the complexities of designing software applications for a given domain in its own way. Evans recognizes the tools and the processes that are popular in the industry, UML, Agile, and focuses on some aspects of the software engineering process that we tend to miss. He starts the book by talking about the importance of creating and having a Ubiquitous Language. There is a similar concept in the RUP, but not emphasizes as much - or at all. Evans goes into a great detail on why, from the inception of a project, it is important to have a common language and gives many pointers on what makes up the Ubiquitous Language for each project:

"Use the model as the backbone of a language. Commit the team to exercising that language relentlessly within the team and the in the code. Use the same language in diagrams, writing, and especially speech."

Parts II-IV of the book put domain-driven design in perspective, and show the reader thru examples and patterns, architectural patterns, design patterns and process patterns, the importance of having a consistent model that maps to the domain and how to go about achieving such model. In an essence, "Model-Driven Design discards the dichotomy of analysis model and design to search out a single model that serves both purposes".

Part II of the book, introduces the building blocks of a Model-Driven Design. This section, as with the others, takes popular patterns from the Gamma, Flower, or others and applies them to the topic at hand - Model-Driven Design. In that aspect, the reader can easily follow the text and relate to topic at hand. Evans uses the ever-popular Model-View-Controller (MVC) design pattern to get things going in part II. He then goes off to explain why the layered architecture approach is an important aspect of a Domain-Driven Design and how it would makes things simpler:

"[Layered Architecture] allows a model to evolve to be rich enough and clear enough to capture essential business knowledge and put it to work."

The author then goes into great detail in explaining the elements that express a model:
1) Entities: An object that is tracked thru different states or even across different implementations.
2) Value Objects: An attribute that describes the state of something else.
3) Services: Aspects of domain that are expressed as actions or operations, rather than objects.
4) Packages: Organize the objects and services.

What do you want to do after you have designed such elements? The creation and life cycle management of objects are discussed next in this book. Three patterns, mostly from the Gamma book, are used to manage the life cycle of objects:
1) Aggregates.
2) Factories.
3) Repositories.

Aggregates represent the hierarchy of objects or services and their interactions. Factories and Repositories operate of Aggregates and encapsulate the complexity of specific life cycle transitions.

Part III of the book talks about the things developers and architects need to do to achieve a Supple Design. Refactoring over and over represents the topic in this section:

"Each refinement of code and model gives developers a clearer view"

The author talks about a breakthrough point during the design that the "designers see the light" and both the domain experts and the designers, after many iterations, have finally come to this higher level of understanding of the domain and the value of refactoring exponentially increases after that.

Part IV of this book talks about a very important topic that we all have struggled with one time or another: the ability of the model and the modeling process to scale up to very complicated domains. It is great that we can model a small domain, but one goes about modeling an enterprise, which is most likely, too complex to model as a single unit? Low-coupling and high cohesion still applies here, but the goal is to not loose anything during the integration process. The author goes in to a great detail in this part to emphasize that even in large circumstances such as modeling an enterprise, every decision must have a direct impact on system development. Three different themes are covered in this section in order to assist with modeling of large units:
1) Context: the model has to be logically consistent throughout, without contradictory or overlapping definitions. For this theme, the author introduces the concept of a Bonded Context- a way that relationship to other context are defined a overlapping is then avoided.
2) Distillation: Reducing the clutter and focusing attention appropriately.
3) Large-scale Structure. The concept of Responsibility layers are introduced

In summary, Evans did a great job in writing this book, and filling it with useful ways of designing and architecting software applications that target a domain, which in most cases we do not know much about.

I read this book in its draft form on a cross-country flight and was just blown away by it, enough so that I bought a bound version to make it easier to carry around and reread.

I suppose what blew me away was that Evans crystallized and laid out quite clearly about a dozen ideas which were existing at the edge of my consciousness, but which I could not clearly verbalize.

It fits quite nicely between the patterns books and the process books, but it's not a cookbook and it's not strictly a method. It's a must read for the multitude of Java/C#/C++ developers who continue to write procedural code while claiming they're OO developers because they're using an OO language and they've read Design Patterns.