with

Unity

Over the years software systems have evolutionarily become more and more
complex. One of the techniques for dealing with this inherent complexity
of software systems is dependency injection a design pattern that
allows the removal of hard-coded dependencies and makes it possible to
assemble a service by changing dependencies easily, whether at run-time
or compile-time. It promotes code reuse and loosely-coupled design which
leads to more easily maintainable and flexible code.

In addition, the guide includes the Tales from the Trenches a collection of
case studies that offer a different perspective through the eyes of developers
working on the real world projects and sharing their experiences. These
chapters make clear the range of scenarios in which you can use Unity, and
also highlight its ease of use and flexibility.
Whether you are a seasoned developer or just starting your development
journey, we hope this guide will be worth your time studying it. We hope you
discover that Unity container adds significant benefits to your applications
and helps you to achieve the goals of maintainability, testability, flexibility,
and extensibility in your own projects. Happy coding!
Im thrilled to see this book published. For the first time, theres one
place you can look for both the concepts of DI and how to apply those
concepts using the Unity container.
Read the book, embrace the concepts, and enjoy the world of loosely
coupled, highly cohesive software that DI makes so easy to build!

Save time and reduce risk on your

software development projects by
incorporating patterns & practices,
Microsofts applied engineering
guidance that includes both production
quality source code and documentation.
The guidance is designed to help
software development teams:
Make critical design and technology
selection decisions by highlighting
the appropriate solution architectures,
technologies, and Microsoft products
for common scenarios

Dominic Betts
Grigori Melnik
Fernando Simonazzi
Mani Subramanian

Understand the most important

concepts needed for success by
explaining the relevant patterns and
prescribing the important practices
Get started with a proven code base
by providing thoroughly tested
software and source that embodies
Microsofts recommendations

Foreword by Chris Tavares

The patterns & practices team consists

of experienced architects, developers,
writers, and testers. We work openly
with the developer community and
industry experts, on every project, to
ensure that some of the best minds in
the industry have contributed to and
reviewed the guidance as it is being
developed.
We also love our role as the bridge
between the real world needs of our
customers and the wide range of
products and technologies that
Microsoft provides.

Chris Tavares
Microsoft Senior Software Development Engineer and co-creator
of Unity

For more information explore:

microsoft.com/practices
msdn.com/unity
Software Architecture and
Software Development

Dependency Injection
with Unity

Unity

The guide contains plenty of trade-off discussions and tips and tricks for
managing your application cross-cutting concerns and making the most
out of both dependency injection and Unity. These are accompanied by a
real world example that will help you master the techniques. Keep in mind
that Unity can be used in a wide range of application types such as desktop,
web, services, and cloud. We encourage you to experiment with the sample
code and think beyond the scenarios discussed in the guide.

Proven practices for predictable results

with

The guide you are holding in your hands is a primer on using dependency
injection with Unity a lightweight extensible dependency injection
container built by the Microsoft patterns & practices team. It covers
various styles of dependency injection and also additional capabilities
of Unity container, such as object lifetime management, interception,
and registration by convention. It also discusses the advanced topics of
enhancing Unity with your custom extensions.

patterns & practices

Dependency Injection

Dependency Injection

Dependency Injection with Unity

Dependency Injection
with Unity

Dominic Betts
Grigori Melnik
Fernando Simonazzi
Mani Subramanian
Foreword by Chris Tavares

ISBN: 978-1-62114-028-3
This document is provided as-is. Information and views expressed in
this document, including URL and other Internet Web site references,
may change without notice.
Some examples depicted herein are provided for illustration only and are
fictitious. No real association or connection is intended or should be
inferred.
This document does not provide you with any legal rights to any
intellectual property in any Microsoft product. You may copy and use
this document for your internal, reference purposes.
2013 Microsoft. All rights reserved.
Microsoft, Visual Basic, Visual Studio, Windows, and Windows Server are
trademarks of the Microsoft group of companies. All other trademarks
are property of their respective owners.

Contents

Foreword ix
Preface xi
About This Guide
xi
Who This Book Is For
xii
What Do You Need to Get Started?
xii
Whos Who
xii
Motivations 1
Maintainability 1
Chapter 1: Introduction
1
Testability 2
Flexibility and Extensibility
2
Late Binding
2
Parallel Development
3
Crosscutting Concerns
3
Loose Coupling
3
A Simple Example
3
When Should You Use a Loosely Coupled Design?
7
Principles of Object-Oriented Design
8
Single Responsibility Principle
8
The Open/Closed Principle
8
The Liskov Substitution Principle
8
Interface Segregation Principle
9
Dependency Inversion Principle
9
Summary 10
More Information
10

vi

Chapter 2: Dependency Injection

11
Introduction 11
Factories, Service Locators, and Dependency Injection
11
Factory Patterns
11
The Factory Method Pattern
11
Simple Factory Pattern
13
Abstract Factory Pattern
14
Service Locator Pattern
14
Dependency Injection
15
Object Composition
17
Object Lifetime
17
Types of Injection
18
Property Setter Injection
18
Method Call Injection
19
When You Shouldnt Use Dependency Injection
19
Summary 20
More Information
20
Chapter 3: Dependency Injection with Unity
21
Introduction 21
The Dependency Injection Lifecycle: Register,
Resolve, Dispose
21
Register 22
Resolve 22
Dispose 23
Registering and Resolving in your Code
23
Adding Unity to Your Application
23
A Real-World Example
24
Type Registrations in the Example
26
Instance Registration
26
Simple Type Registration
27
Constructor Injection
27
Registering Open Generics
28
Parameter Overrides
29
Resolving Types in the Example
29
Simple Resolve
29
Resolving in an MVC Application
30
Using the Per Request Lifetime Manager
in MVC and WebAPI Application
30
Resolving with Run Time Information
31
Registration 32
Named Type Registrations
32
Design-Time Configuration
33
Registration by Convention
34
Registration by Convention and Generic Types
38
Using Child Containers
38
Viewing Registration Information
39

vii

Resolving 41
Resolving in an ASP.NET Web Application
41
Resolving in a WCF Service
43
Using the UnityServiceHost Class with a
Self-hosted Service
46
Using the UnityServiceHost Class with
Service Hosted in IIS or WAS
47
Automatic Factories
47
Deferred Resolution
50
Lifetime Management
51
Hierarchical Lifetime Management
52
Per Resolve Lifetime Management
53
Externally Controlled Lifetime Management
55
Per Request Lifetime Management
55
Per Thread Lifetime Management
56
Dependency Injection and Unit Testing
56
Summary 58
More Information
58

Chapter 4: Interception
59
Introduction 59
Crosscutting Concerns
59
The Decorator Pattern
60
Using Unity to Wire Up Decorator Chains
63
Aspect Oriented Programming
63
Interception 64
Instance Interception
64
Type Interception
65
Summary 66
More Information
66
Chapter 5: Interception using Unity
67
Introduction 67
Crosscutting Concerns and Enterprise Library
67
Interceptors in Unity
67
Configuring the Unity Container to Support Interception
68
Defining an Interceptor
68
Registering an Interceptor
71
Using an Interceptor
72
Alternative Interception Techniques
73
Instance Interception/Type Interception
73
Using the TransparentProxyInterceptor Type
73
Using the VirtualMethodInterceptor Type
74
Using a Behavior to Add an Interface to an Existing Class
76
Interception Without the Unity Container
77
Design Time Configuration
78
Policy Injection
79
Policy Injection and Attributes
83
Policy Injection and the Enterprise Library Blocks
85
A Real World Example
87
Summary 89
More Information
89

viii

Chapter 6: Extending Unity

91
Introduction 91
Creating Custom Lifetime Managers
91
Lifetime Managers and Resolved Objects
92
Extending the SynchronizedLifetimeManager Type
92
Extending the LifetimeManager Type
95
Extending the Unity Container
96
Without the Event Broker Container Extension
96
With the Event Broker Extension
97
Implementing the Simple Event Broker
98
Implementing the Container Extension
100
Discovering the Publishers and Subscribers
101
Wiring Up the Publishers and Subscribers
103
Summary 104
More Information
104
Summary 105
Tales from the Trenches: Using Unity

107

Tales from the Trenches: One Users Story Customizing

Unity 111
More Information
115
Tales from the Trenches: Using Unity in a Windows
Store app
116
AdventureWorks Shopper
116
References 117
More Information
117
Appendix A: Unity and Windows Store apps
The UnityServiceLocator Class
Unity Design-time Configuration
Unity Interception
More Information

119
119
119
119
119

Index 121

Foreword

The set of ideas that later congealed into the Unity container were originally conceived while I was working on
the Web Client Software Factory project. Microsofts patterns & practices team had been using the concept
of dependency injection for several years at that point, most famously in the Composite Application Block
(CAB). It was also core to the configuration story in Enterprise Library 2.0, and was again central when we
started tackling composite applications for the web (a library that became known as CWAB).
Our goal had always been to promote the concepts of Dependency Injection as a way to build loosely coupled
systems. However, the way p&p approached DI at the time was different then how we think about it now. Instead of a single reusable container it was felt that the DI implementation should be specialized to the system
in which it was being used. We used a library called ObjectBuilder, which was described as a framework to
build DI containers. This would in theory let us write a container per project that did exactly what we wanted.
A lofty aspiration, but in practice it didnt work out so well. ObjectBuilder was a highly decoupled, abstract set
of parts that had to be manually assembled. Combined with a lack of documentation it took a lot of time to
understand what needed to go where and how to put it together into something useful. That turned into time
spent writing, debugging, and optimizing the DI container instead of working on our actual project requirements.
It got even more fun when somebody wanted to use CAB (which used one DI container based on one version
of ObjectBuilder) and Enterprise Library (with a separate container based on a different version of ObjectBuilder) in the same project. Integration was very difficult; just dealing with referencing two different versions
of ObjectBuilder in the same project was a challenge. Also the one-off containers led to one-off extensibility
and integration interfaces: what worked in Enterprise Library was useless in CAB and vice versa.
It finally came to a head when wed just spent yet another week near the end of the Web Client Software
Factory project fixing a bunch of bugs in CWAB: bugs that looked very similar to ones wed fixed before in CAB.
Wouldnt it be nice, we asked, if we could just have one container implementation and just use it instead of
writing them over and over again?
From this frustration grew Unity. The Enterprise Library 4.0 team put the Dependency Injection Application
Block (as Unity used to be known originally) on the product backlog. Our goals for the project were straightforward. First, introduce and promote the concepts of dependency injection to our community, unencumbered
by a lot of low-level implementation details. Second, have a core container with an easy to use API that we,
other teams at Microsoft, or anyone whose organization was uncomfortable using the available open source
projects (for whatever reason) could just use. Third, have a variety of extensibility mechanisms so that new
features could be added by anyone without having to rip open the core code.
In my opinion Unity has succeeded on all these goals. Im particularly proud of how we affected the .NET developer community. Unity quickly became one of the most commonly used DI containers in the .NET ecosystem. More importantly, other DI container usage has increased as well. Unity introduced DI to a new set of
people who would have otherwise never heard of it. Some of them later moved on to other containers that
better suited their needs. Thats not a loss for Unity: theyre using the concepts, and thats the important part.

ix

Theres not a whole lot of evangelism published for DI containers anymore. In my opinion, this is because DI is
no longer an expert technique: its now part of the mainstream. When even frameworks from Microsoft (ASP.
NET MVC and WebAPI in particular) come with support for DI built in, you know that a concept has reached
the core audience. I think Unity had a very large role in making this happen.
Im thrilled to see this book published. For the first time, theres one place you can look for both the concepts
of DI and how to apply those concepts using the Unity container. And theres coverage of the extensibility
story, something I always wanted to write but never seemed to get started. I dont need to feel guilty about
that anymore!
Read the book, embrace the concepts, and enjoy the world of loosely coupled, highly cohesive software that
DI makes so easy to build!
Chris Tavares
Redmond, WA, USA
April 2013

Preface

About This Guide

This guide is one of the resources available with the Unity 3 release to help you to learn about Unity, learn about
some of the problems and issues that Unity can help you to address, and get started using Unity in your applications. Unity is primarily a dependency injection container and so the guide also contains an introduction to dependency injection that you can read in isolation even if you dont plan to use Unity, although we hope you will.
The chapters are designed to be read in order, each one building on the previous one, and alternating chapters
that cover some of the conceptual background material with chapters that address the specifics of using Unity
in your own applications. If youre already familiar with concepts such as dependency injection and interception,
you can probably focus on Chapter 3, Dependency Injection with Unity, Chapter 5, Interception with Unity,
and Chapter 6, Extending Unity.
The first two chapters introduce the conceptual background and explain what dependency injection is, what
are its benefits and drawbacks, and when you should consider using it. Chapter 3 then applies this theoretical
knowledge to the use of the Unity container and provides examples and guidance on how to use it in a variety
of scenarios.
Chapter 4 describes interception as a technique to dynamically insert code that provides support for crosscutting concerns into your application.
Chapter 5 discusses advanced topics of interception and policy injection, along with alternatives, and offers
some suggestions about when you should use it.
The final chapter introduces some of the ways that you can extend Unity such as creating container extensions
or creating custom lifetime managers.
This guide also includes several case studies, called Tales from the Trenches, of developers adapting and customizing Unity. Additional case studies may be available online (http://msdn.com/unity), so make sure to check them
out. If you want to share your story with the developer community at large, send it to ourstory@microsoft.com.
All of the chapters include references to additional resources such as books, blog posts, and papers that will
provide additional detail if you want to explore some of the topics in greater depth. For your convenience, there
is a bibliography online that contains all the links so that these resources are just a click away. You can find the
bibliography at: http://aka.ms/unitybiblio
The majority of the code samples in the chapters come from a collection of sample applications that you can
download and play with.
This guide does not include detailed information about every Unity feature or every class in the Unity assemblies. For that information, you should look at the Unity Reference Documentation and the Unity API Documentation.

xi

xii

Who This Book Is For

This book is intended for any architect, developer, or information technology (IT) professional who designs,
builds, or operates applications and services and who wants to learn how to realize the benefits of using the
Unity dependency injection container in his or her applications. You should be familiar with the Microsoft .NET
Framework, and Microsoft Visual Studio to derive full benefit from reading this guide.

What Do You Need to Get Started?

The system requirements and prerequisites for using Unity are:
Supported architectures: x86 and x64.
Operating systems: Microsoft Windows 8, Microsoft Windows 7, Windows Server 2008 R2, Windows

Server 2012.

Supported .NET Frameworks: Microsoft .NET Framework 4.5, .NET for Windows Store Apps (previously

known as WinRT).

Rich development environment: Microsoft Visual Studio 2012, Professional, Ultimate, or Express editions.

You can use the NuGet package manager in Visual Studio to install the Unity assemblies in your projects.

Whos Who
The guide includes discussions and examples that relate to the use of Unity in a variety of scenarios and types
of application. A panel of experts provides a commentary throughout the book, offering a range of viewpoints
from developers with various levels of skill, an architect, and an IT professional. The following table lists the
various experts who appear throughout the guide.

Markus is a software developer who is new to Unity. He is analytical, detail-oriented,

and methodical. Hes focused on the task at hand, which is building a great LOB
application. He knows that hes the person whos ultimately responsible for the code.
I want to get started using Unity quickly, so I want it to be simple to incorporate
into my code and be easy to configure with plenty of sensible defaults.

Beth is a developer who used Unity some time ago but abandoned it for her more recent
projects.
Im happy using libraries and frameworks but I dont want to get tied into dependencies that I dont need.
I want to be able to use just the components I need for the task in hand.

xiii

Jana is a software architect. She plans the overall structure of an application. Her
perspective is both practical and strategic. In other words, she considers not only what
technical approaches are needed today, but also what direction a company needs to
consider for the future. Jana has worked on many projects that have used Unity as well
as other dependency injection containers. Jana is comfortable assembling a solution
using multiple libraries and frameworks.
Its not easy to balance the needs of the company, the users, the IT organization, the developers, and
the technical platforms we rely on while trying to ensure component independence.

Carlos is an experienced software developer and Unity expert. As a true professional, he

is well aware of the common crosscutting concerns that developers face when building
line-of-business (LOB) applications for the enterprise. His team is used to relying on
Unity and he is happy to see continuity in Unity releases. Quality, support, and ease of
migration are his primary concerns.
Our existing LOB applications use Unity for dependency management and interception. This provides a
level of uniformity across all our systems that make them easier to support and maintain. We want to be
able to migrate our existing applications to the new version with a minimum of effort.

Poe is an IT professional whos an expert in deploying and managing LOB applications. Poe
has a keen interest in practical solutions; after all, hes the one who gets paged at 3:00 AM
when theres a problem. Poe wants to be able to tweak application configuration without
recompiling or even redeploying them in order to troubleshoot.
I want a consistent approach to configuration for all our applications both on-premises and in the cloud.

Introduction

Before you learn about dependency injection and Unity, you need to understand
why you should use them. And in order to understand why you should use them,
you should understand what types of problems dependency injection and Unity
are designed to help you address. This introductory chapter will not say much
about Unity, or indeed say much about dependency injection, but it will provide
some necessary background information that will help you to appreciate the
benefits of dependency injection as a technique and why Unity does things the
way it does.
The next chapter, Chapter 2, Dependency Injection, will show you how dependency injection can help you meet the requirements outlined in this chapter, and the
following chapter, Chapter 3, Dependency Injection with Unity, shows how Unity
helps you to implement the dependency injection approach in your applications.

Motivations
When you design and develop software systems, there are many requirements
to take into account. Some will be specific to the system in question and some
will be more general in purpose. You can categorize some requirements as functional requirements, and some as non-functional requirements (or quality attributes). The full set of requirements will vary for every different system. The set
of requirements outlined below are common requirements, especially for lineof-business (LOB) software systems with relatively long anticipated lifetimes.
They are not all necessarily going to be important for every system you develop,
but you can be sure that some of them will be on the list of requirements for
many of the projects you work on.

This chapter introduces

a lot of requirements and
principles. Dont assume
that they are all relevant all
of the time. However, most
enterprise systems have
some of the requirements,
and the principles all point
towards good design and
coding practices.

Maintainability
As systems become larger, and as the expected lifetimes of systems get longer,
maintaining those systems becomes more and more of a challenge. Very often, the
original team members who developed the system are no longer available, or no
longer remember the details of the system. Documentation may be out of date or
even lost. At the same time, the business may be demanding swift action to meet
some pressing new business need. Maintainability is the quality of a software
system that determines how easily and how efficiently you can update it. You may
need to update a system if a defect is discovered that must be fixed (in other
words, performing corrective maintenance), if some change in the operating environment requires you to make a change in the system, or if you need to add new
features to the system to meet a business requirement (perfective maintenance).
Maintainable systems enhance the agility of the organization and reduce costs.

It is very hard to make

existing systems more
maintainable. It is much
better to design for
maintainability from the very
start.
1

ch a pter one

Therefore, you should include maintainability as one of your design goals, along
with others such as reliability, security, and scalability.

Testability
A testable system is one that enables you to effectively test individual parts of
the system. Designing and writing effective tests can be just as challenging as
designing and writing testable application code, especially as systems become
larger and more complex. Methodologies such as test-driven development (TDD)
require you to write a unit test before writing any code to implement a new
feature and the goal of such a design technique is to improve the quality of your
application. Such design techniques also help to extend the coverage of your unit
tests, reduce the likelihood of regressions, and make refactoring easier. However,
as part of your testing processes you should also incorporate other types of tests
such as acceptance tests, integration tests, performance tests, and stress tests.

Using test doubles is a

great way to ensure that
you can continuously run
your unit tests during the
development process.
However, you must still fully
test your application in a real
environment.

Running tests can also cost money and be time consuming because of the requirement to test in a realistic environment. For example, for some types of
testing on a cloud-based application you need to deploy the application to the
cloud environment and run the tests in the cloud. If you use TDD, it may be
impractical to run all the tests in the cloud all of the time because of the time it
takes to deploy your application, even to a local emulator. In this type of scenario, you may decide to use test doubles (simple stubs or verifiable mocks) that
replace the real components in the cloud environment with test implementations in order to enable you to run your suite of unit tests in isolation during the
standard TDD development cycle.
Testability should be another of the design goals for your system along with
maintainability and agility: a testable system is typically more maintainable, and
vice versa.

Flexibility and Extensibility

For a great discussion on the

use of test doubles, see the
point/counterpoint debate
by Steve Freeman, Nat Pryce
and Joshua Kerievsky in IEEE
Software (Volume: 24, Issue:
3), May/June 2007, pp.80-83.
Not all systems have a
requirement for late binding.
It is typically required to
support a specific feature
of the application such as
customization using a plugin architecture.

Flexibility and extensibility are also often on the list of desirable attributes of
enterprise applications. Given that business requirements often change, both
during the development of an application and after it is running in production,
you should try to design the application to make it flexible so that it can be
adapted to work in different ways and extensible so that you can add new features. For example, you may need to convert your application from running
on-premises to running in the cloud.

Late Binding
In some application scenarios, you may have a requirement to support late binding. Late binding is useful if you require the ability to replace part of your system
without recompiling. For example, your application might support multiple relational databases with a separate module for each supported database type.
You can use declarative configuration to tell the application to use a specific
module at runtime. Another scenario where late binding can be useful is to enable users of the system to provide their own customization through a plug-in.
Again, you can instruct the system to use a specific customization by using a
configuration setting or a convention where the system scans a particular location on the file system for modules to use.

Introduction

Parallel Development
When you are developing large scale (or even small and medium scale) systems,
it is not practical to have the entire development team working simultaneously
on the same feature or component. In reality, you will assign different features
and components to smaller groups to work on in parallel. Although this approach enables you to reduce the overall duration of the project, it does introduce additional complexities: you need to manage multiple groups and to ensure
that you can integrate the parts of the application developed by different groups
to work correctly together.

Crosscutting Concerns
Enterprise applications typically need to address a range of crosscutting concerns
such as validation, exception handling, and logging. You may need these features
in many different areas of the application and you will want to implement them
in a standard, consistent way to improve the maintainability of the system. Ideally, you want a mechanism that will enable you to efficiently and transparently
add behaviors to your objects at either design time or run time without requiring
you make changes to your existing classes. Often, you need the ability to configure these features at runtime and in some cases, add features to address a new
crosscutting concern to an existing application.

Loose Coupling
You can address many of the requirements listed in the previous sections by
ensuring that your design results in an application that loosely couples the many
parts that make up the application. Loose coupling, as opposed to tight coupling,
means reducing the number of dependencies between the components that
make up your system. This makes it easier and safer to make changes in one area
of the system because each part of the system is largely independent of the
other.

A Simple Example

It can be a significant
challenge to ensure that
classes and components
developed independently do
work together.

For a large enterprise

system, its important to be
able to manage crosscutting
concerns such as logging and
validation in a consistent
manner. I often need to
change the logging level
on a specific component at
run time to troubleshoot an
issue without restarting the
system.

The following example illustrates tight coupling where the ManagementController class depends directly on the TenantStore class. These classes might
be in different Visual Studio projects.
public class TenantStore
{
...
public Tenant GetTenant(string tenant)
{
...
}
public IEnumerable<string> GetTenantNames()
{
...
}
}

Loose coupling should be a

general design goal for your
enterprise applications.

ch a pter one

public class ManagementController

{
private readonly TenantStore tenantStore;
public ManagementController()
{
tenantStore = new TenantStore(...);
}
public ActionResult Index()
{
var model = new TenantPageViewData<IEnumerable<string>>
(this.tenantStore.GetTenantNames())
{
Title = "Subscribers"
};
return this.View(model);
}
public ActionResult Detail(string tenant)
{
var contentModel = this.tenantStore.GetTenant(tenant);
var model = new TenantPageViewData<Tenant>(contentModel)
{
Title = string.Format("{0} details", contentModel.Name)
};
return this.View(model);
}
...
}

The ManagementController and TenantStore classes are used in various forms throughout this guide.
Although the ManagementController class is an ASP.NET MVC controller, you dont need to know about
MVC to follow along. However, these examples are intended to look like the kinds of classes you would
encounter in a real-world system, especially the examples in Chapter 3.
In this example, the TenantStore class implements a repository that handles access to an underlying data store
such as a relational database, and the ManagementController is an MVC controller class that requests data
from the repository. Note that the ManagementController class must either instantiate a TenantStore object
or obtain a reference to a TenantStore object from somewhere else before it can invoke the GetTenant and
GetTenantNames methods. The ManagementController class depends on the specific, concrete TenantStore
class.
If you refer back to the list of common desirable requirements for enterprise applications at the start of this
chapter, you can evaluate how well the approach outlined in the previous code sample helps you to meet them.
Although this simple example shows only a single client class of the TenantStore class, in practice there

may be many client classes in your application that use the TenantStore class. If you assume that each
client class is responsible for instantiating or locating a TenantStore object at runtime, then all of those
classes are tied to a particular constructor or initialization method in that TenantStore class, and may all
need to be changed if the implementation of the TenantStore class changes. This potentially makes
maintenance of the TenantStore class more complex, more error prone, and more time consuming.

Introduction

In order to run unit tests on the Index and Detail methods in the ManagementController class, you need

to instantiate a TenantStore object and make sure that the underlying data store contains the appropriate
test data for the test. This complicates the testing process, and depending on the data store you are using,
may make running the test more time consuming because you must create and populate the data store
with the correct data. It also makes the tests much more brittle.
It is possible to change the implementation of the TenantStore class to use a different data store, for
example Windows Azure table storage instead of SQL Server. However, it might require some changes to
the client classes that use TenantStore instances if it was necessary for them to provide some initialization
data such as connection strings.
You cannot use late binding with this approach because the client classes are compiled to use the TenantStore class directly.
If you need to add support for a crosscutting concern such as logging to multiple store classes, including
the TenantStore class, you would need to modify and configure each of your store classes independently.
The following code sample shows a small change, the constructor in the client ManagementController class
now receives an object that implements the ITenantStore interface and the TenantStore class provides an
implementation of the same interface.
public interface ITenantStore
{
void Initialize();
Tenant GetTenant(string tenant);
IEnumerable<string> GetTenantNames();
void SaveTenant(Tenant tenant);
void UploadLogo(string tenant, byte[] logo);
}
public class TenantStore : ITenantStore
{
...
public TenantStore()
{
...
}
...
}
public class ManagementController : Controller
{
private readonly ITenantStore tenantStore;
public ManagementController(ITenantStore tenantStore)
{
this.tenantStore = tenantStore;
}

ch a pter one

public ActionResult Index()

{
...
}
public ActionResult Detail(string tenant)
{
...
}
...
}

This change has a direct impact on how easily you can meet the list of requirements.
It is now clear that the ManagementController class, and any other clients of the TenantStore class are

no longer responsible for instantiating TenantStore objects, although the example code shown doesnt
show which class or component is responsible for instantiating them. From the perspective of maintenance, this responsibility could now belong to a single class rather than many.
Its now also clear what dependencies the controller has from its constructor arguments instead of being
buried inside of the controller method implementations.
To test some behaviors of a client class such as the ManagementController class, you can now provide a
lightweight implementation of the ITenantStore interface that returns some sample data. This is instead
of creating a TenantStore object that queries the underlying data store for sample data.
Introducing the ITenantStore interface makes it easier to replace the store implementation without
requiring changes in the client classes because all they expect is an object that implements the interface.
If the interface is in a separate project to the implementation, then the projects that contain the client
classes only need to hold a reference to the project that contains the interface definition.
It is now also possible that the class responsible for instantiating the store classes could provide additional
services to the application. It could control the lifetime of the ITenantStore instances that it creates, for
example creating a new object every time the client ManagementController class needs an instance, or
maintaining a single instance that it passes as a reference whenever a client class needs it.
It is now possible to use late binding because the client classes only reference the ITenantStore interface
type. The application can create an object that implements the interface at runtime, perhaps based on a
configuration setting, and pass that object to the client classes. For example, the application might create
either a SQLTenantStore instance or a BlobTenantStore instance depending on a setting in the web.
config file, and pass that to the constructor in the ManagementController class.
If the interface definition is agreed, two teams could work in parallel on the store class and the controller
class.
The class that is responsible for creating the store class instances could now add support for the crosscutting concerns before passing the store instance on to the clients, such as by using the decorator pattern to
pass in an object that implements the crosscutting concerns. You dont need to change either the client
classes or the store class to add support for crosscutting concerns such as logging or exception handling.

Introduction

The approach shown in the second code sample is an example of a loosely

coupled design that uses interfaces. If we can remove a direct dependency between classes, it reduces the level of coupling and helps to increase the maintainability, testability, flexibility, and extensibility of the solution.
What the second code sample doesnt show is how dependency injection and
the Unity container fit into the picture, although you can probably guess that
they will be responsible for creating instances and passing them to client classes.
Chapter 2 describes the role of dependency injection as a technique to support
loosely coupled designs, and Chapter 3 describes how Unity helps you to implement dependency injection in your applications.

When Should You Use a Loosely Coupled Design?

Loose coupling doesnt

necessarily imply
dependency injection,
although the two often do
go together.

Before we move on to dependency injection and Unity, you should start to

understand where in your application you should consider introducing loose
coupling, programming to interfaces, and reducing dependencies between
classes. The first requirement we described in the previous section was maintainability, and this often gives a good indication of when and where to consider reducing the coupling in the application. Typically, the larger and more
complex the application, the more difficult it becomes to maintain, and so the
more likely these techniques will be helpful. This is true regardless of the type
of application: it could be a desktop application, a web application, or a cloud
application.
At first sight, this perhaps seems counterintuitive. The second example shown
above introduced an interface that wasnt in the first example, it also requires the
bits we havent shown yet that are responsible for instantiating and managing
objects on behalf of the client classes. With a small example, these techniques
appear to add to the complexity of the solution, but as the application becomes
larger and more complex, this overhead becomes less and less significant.
The previous example also illustrates another general point about where it is
appropriate to use these techniques. Most likely, the ManagementController
class exists in the user interface layer in the application, and the TenantStore
class is part of the data access layer. It is a common approach to design an application so that in the future it is possible to replace one tier without disturbing
the others. For example, replacing or adding a new UI to the application (such
as creating an app for a mobile platform in addition to a traditional web UI)
without changing the data tier or replacing the underlying storage mechanism
and without changing the UI tier. Building the application using tiers helps to
decouple parts of the application from each other. You should try to identify
the parts of an application that are likely to change in the future and then decouple them from the rest of the application in order to minimize and localize
the impact of those changes.

Small examples of
loosely coupled design,
programming to interfaces,
and dependency injection
often appear to complicate
the solution. You should
remember that these
techniques are intended
to help you simplify and
manage large and complex
applications with many
classes and dependencies.
Of course small applications
can often grow into large
and complex applications.

ch a pter one

The list of requirements in the previous section also includes crosscutting concerns that you might need to
apply across a range of classes in your application in a consistent manner. Examples include the concerns addressed by the application blocks in Enterprise Library (http://msdn.microsoft.com/entlib) such as logging, exception handling, validation, and transient fault handling. Here you need to identify those classes where you might
need to address these crosscutting concerns, so that responsibility for adding these features to these classes
resides outside of the classes themselves. This helps you to manage these features consistently in the application
and introduces a clear separation of concerns.

Principles of Object-Oriented Design

Finally, before moving on to dependency injection and Unity, we want to relate the five SOLID principles of
object-oriented programming and design to the discussion so far. SOLID is an acronym that refers to the following principles:
Single responsibility principle
Open/close principle
Liskov substitution principle
Interface segregation principle
Dependency inversion principle

The following sections describe each of these principles and their relationship to loose coupling and the requirements listed at the start of this chapter.

Single Responsibility Principle

The single responsibility principle states that a class should have one, and only one, reason to change. For more
information, see the article Principles of Object Oriented Design by Robert C. Martin1.
In the first simple example shown in this chapter, the ManagementController class had two responsibilities: to
act as a controller in the UI and to instantiate and manage the lifetime of TenantStore objects. In the second
example, the responsibility for instantiating and managing TenantStore objects lies with another class or component in the system.

The Open/Closed Principle

The open/closed principle states that software entities (classes, modules, functions, and so on) should be open
for extension, but closed for modification (Meyer, Bertrand (1988). Object-Oriented Software Construction.)
Although you might modify the code in a class to fix a defect, you should extend a class if you want to add any
new behavior to it. This helps to keep the code maintainable and testable because existing behavior should not
change, and any new behavior exists in new classes. The requirement to be able to add support for crosscutting
concerns to your application can best be met by following the open/closed principle. For example, when you
add logging to a set of classes in your application, you shouldnt make changes to the implementation of your
existing classes.

The Liskov Substitution Principle

The Liskov substitution principle in object-oriented programming states that in a computer program, if S is a
subtype of T, then objects of type T may be replaced with objects of type S without altering any of the desirable properties, such as correctness, of that program.

1 The Principles of OOD, Robert C Martin, http://www.butunclebob.com/ArticleS.UncleBob.PrinciplesOfOod.

Introduction

In the second code sample shown in this chapter, the ManagementController class should continue to work
as expected if you pass any implementation of the ITenantStore interface to it. This example uses an interface
type as the type to pass to the constructor of the ManagementController class, but you could equally well use
an abstract type.

Interface Segregation Principle

The interface segregation principle is a software development principle intended to make software more maintainable. The interface segregation principle encourages loose coupling and therefore makes a system easier to
refactor, change, and redeploy. The principle states that interfaces that are very large should be split into
smaller and more specific ones so that client classes only need to know about the methods that they use: no
client class should be forced to depend on methods it does not use.
In the definition of the ITenantStore interface shown earlier in this chapter, if you determined that not all client
classes use the UploadLogo method you should consider splitting this into a separate interface as shown in the
following code sample:
public interface ITenantStore
{
void Initialize();
Tenant GetTenant(string tenant);
IEnumerable<string> GetTenantNames();
void SaveTenant(Tenant tenant);
}
public interface ITenantStoreLogo
{
void UploadLogo(string tenant, byte[] logo);
}
public class TenantStore : ITenantStore, ITenantStoreLogo
{
...
public TenantStore()
{
...
}
...
}

Dependency Inversion Principle

The dependency inversion principle states that:
High-level modules should not depend on low-level modules. Both should depend on abstractions.
Abstractions should not depend upon details. Details should depend upon abstractions.

The two code samples in this chapter illustrate how to apply this principle. In the first sample, the high-level
ManagementController class depends on the low-level TenantStore class. This typically limits the options for
re-using the high-level class in another context.
In the second code sample, the ManagementController class now has a dependency on the ITenantStore
abstraction, as does the TenantStore class.

10

ch a pter one

Summary
In this chapter, you have seen how you can address some of the common requirements in enterprise applications
such as maintainability and testability by adopting a loosely coupled design for your application. You saw a very
simple illustration of this in the code samples that show two different ways that you can implement the dependency between the ManagementController and TenantStore classes. You also saw how the SOLID principles
of object-oriented programming relate to the same concerns.
However, the discussion in this chapter left open the question of how to instantiate and manage TenantStore
objects if the ManagementController is no longer responsible for this task. The next chapter will show how
dependency injection relates to this specific question and how adopting a dependency injection approach can
help you meet the requirements and adhere to the principles outlined in this chapter.

More Information
All links in this book are accessible from the books online bibliography available at: http://aka.ms/unitybiblio

Dependency Injection

Introduction
Chapter 1 outlines how you can address some of the most common requirements in enterprise applications by
adopting a loosely coupled design to minimize the dependencies between the different parts of your application. However, if a class does not directly instantiate the other objects that it needs, some other class or component must take on this responsibility. In this chapter, youll see some alternative patterns that you can use to
manage how objects are instantiated in your application before focusing specifically on dependency injection
as the mechanism to use in enterprise applications.

Factories, Service Locators, and Dependency Injection

Factories, service locators, and dependency injection are all approaches you can take to move the responsibility for instantiating and managing objects on behalf of other client objects. In this section, youll see how you
can use them with the same example you saw in the previous chapter. Youll also see the pros and cons of the
different approaches and see why dependency injection can be particularly useful in enterprise applications.

Factory Patterns
There are three common factory patterns. The Factory Method and Abstract Factory patterns from Design
Patterns: Elements of Reusable Object-Oriented Software by Gamma, Erich, Richard Helm, Ralph Johnson, and
John Vlissides. Addison Wesley Professional, 1994., and the Simple Factory pattern.
The Factory Method Pattern
The following code samples show how you could apply the factory method pattern to the example shown in
the previous chapter. The first code sample shows how you could use a factory method to return an instance
of the TenantStore class to use in the ManagementController class. In this example, the CreateTenantStore
method is the factory method that creates the TenantStore instance and the Index method uses this instance
as part of its logic.
public class ManagementController : Controller
{
protected ITenantStore tenantStore;
public ManagementController()
{
this.tenantStore = CreateTenantStore();
}

11

12

ch a pter t wo

protected virtual ITenantStore CreateTenantStore()

{
var storageAccount = AppConfiguration
.GetStorageAccount("DataConnectionString");
var tenantBlobContainer = new EntitiesContainer<Tenant>
(storageAccount, "Tenants");
var logosBlobContainer = new FilesContainer
(storageAccount, "Logos", "image/jpeg");
return new TenantStore(tenantBlobContainer, logosBlobContainer);
}
public ActionResult Index()
{
var model = new TenantPageViewData<IEnumerable<string>>
(this.tenantStore.GetTenantNames())
{
Title = "Subscribers"
};
return this.View(model);
}
...
}

Using this approach does not remove the dependencies the ManagementController has on the TenantStore class, nor the FilesContainer and EntitiesContainer classes. However, it is now possible to replace the underlying storage
mechanism without changing the existing ManagementController class as the
following code sample shows.

The factory method pattern

enables you to modify the
behavior of a class without
modifying the class itself by
using inheritance.

public class SQLManagementController : ManagementController

{
protected override ITenantStore CreateTenantStore()
{
var storageAccount = ApplicationConfiguration
.GetStorageAccount("DataConnectionString");
var tenantSQLTable = ...
var logosSQLTable = ....
return new SQLTenantStore(tenantSQLTable, logosSQLTable);
}
...
}

The application can use the SQLManagementController class to use a SQL-based

store without you needing to make any changes to the original ManagementController class. This approach results in a flexible and extensible design and implements the open/closed principle described in the previous chapter. However, it
does not result in a maintainable solution because all the client classes that use the
TenantStore class are still responsible for instantiating TenantStore instances correctly and consistently.

Dependency Injection

It is also still difficult to test the ManagementController class because it depends on the TenantStore type, which in turn is tied to specific storage types
(FilesContainer and EntitiesContainer). One approach to testing would be to
create a MockManagementController type that derives from ManagementController and that uses a mock storage implementation to return test data: in
other words you must create two mock types to manage the testing.
In this example, there is an additional complication because of the way that
ASP.NET MVC locates controllers and views based on a naming convention:
you must also update the MVC routes to ensure that MVC uses the new
SQLManagementController class.
Simple Factory Pattern
While the factory method pattern does not remove the dependencies from the
high-level client class, such as the ManagementController class, on the lowlevel class, you can achieve this with the simple factory pattern. In this example,
you can see that a new factory class named TenantStoreFactory is now responsible for creating the TenantStore instance on behalf of the ManagementController class.
public class ManagementController : Controller
{
private readonly ITenantStore tenantStore;
public ManagementController()
{
var tenantStoreFactory = new TenantStoreFactory();
this.tenantStore = tenantStoreFactory.CreateTenantStore();
}
public ActionResult Index()
{
var model = new TenantPageViewData<IEnumerable<string>>
(this.tenantStore.GetTenantNames())
{
Title = "Subscribers"
};
return this.View(model);
}
...
}

This approach removes much of the complexity from the high-level ManagementController class, although in this example the ManagementController class is still
responsible for selecting the specific type of tenant store to use. You could easily
move this logic into the factory class that could read a configuration setting to
determine whether to create a BlobTenantStore instance or a SQLTenantStoreInstance. Making the factory class responsible for selecting the specific type to
create makes it easier to apply a consistent approach throughout the application.

The simple factory

pattern removes the
direct dependency of the
ManagementController
class on a specific store
implementation. Instead of
including the code needed
to build a TenantStore
instance directly, the
controller class now relies
on the TenantStoreFactory
class to create the instance
on its behalf.

13

14

ch a pter t wo

Abstract Factory Pattern

The abstract factory

pattern is useful if you have
a requirement to create
families of related objects in
a consistent way.

One of the problems that can arise from using the simple factory pattern in a
large application is that it can be difficult to maintain consistency. For example,
the application may include multiple store classes such as SurveyStore, LogoStore, and ReportStore classes in addition to the TenantStore class youve
seen in the examples so far. You may have a requirement to use a particular type
of storage for all of the stores. Therefore, you could implement a BlobStoreFactory abstract factory class that can create multiple blob-based stores, and
a SQLStoreFactory abstract factory class that can create multiple SQL based
stores.
The abstract factory pattern is described in Design Patterns: Elements of Reusable Object-Oriented Software by Gamma, et al.

Service Locator Pattern

Using a service locator provides another variation to this general approach of
using another class to create objects on your behalf. You can think of a service
locator as a registry that you can look up an instance of an object or service that
another class in your application created and registered with the service locator.
The service locator might support querying for objects by a string key or by
interface type. Often, in contrast to the factory patterns where the factory
creates the object but gives responsibility for managing its lifetime to the client
class, the service locator is responsible for managing the lifetime of the object
and simply returns a reference to the client. Also, factories are typically responsible for creating instances of specific types or families of types as in the case of
the abstract factory pattern, while a service locator may be capable of returning
a reference to an object of any type in the application.
The section Object Lifetime later in this chapter discusses object lifetimes
in more detail.
Any classes that retrieve object references or service references from the service
locator will have a dependency on the service locator itself.
For a description of the service locator pattern, see the section Using a Service
Locator in the article Inversion of Control Containers and the Dependency Injection pattern by Martin Fowler.
For a discussion of why the service locator may be considered an anti-pattern,
see the blog post Service Locator is an Anti-Pattern by Mark Seeman.

When using a service

locator, every class will
have a dependency on your
service locator. This is not
the case with dependency
injection.

For a shared interface for service location that application and framework developers can reference, see the Common Service Locator library. The library provides an abstraction over dependency injection containers and service locators.
Using the library allows an application to indirectly access the capabilities
without relying on hard references.

Dependency Injection

15

Dependency Injection
A common feature of the all the factory patterns and the service locator pattern, is that it is still the high-level
client objects responsibility to resolve its own dependencies by requesting the specific instances of the types
that it needs. They each adopt a pull model of varying degrees of sophistication, assigning various responsibilities to the factory or service locator. The pull model also means that the high-level client class has a dependency on the class that is responsible for creating or locating the object it wants to use. This also means that
the dependencies of the high-level client classes are hidden inside of those classes rather specified in a single
location, making them harder to test.
Figure 1 shows the dependencies in the simple factory pattern where the factory instantiates a TenantStore
object on behalf of the ManagementController class.

ManagementController
class

TenantStoreFactory
class

TenantStore class

ITenantStore
interface

Figure 1
Dependencies in the factory pattern

Dependency injection takes the opposite approach, adopting a push model in place of the pull model. Inversion
of Control is a term thats often used to describe this push model and dependency injection is one specific
implementation of the inversion of control technique.
Martin Fowler states: With service locator the application class asks for it explicitly by a message to the locator. With injection there is no explicit request, the service appears in the application classhence the inversion
of control. (Inversion of Control Containers and the Dependency Injection pattern.)
With dependency injection, another class is responsible for injecting (pushing) the dependencies into the highlevel client classes, such as the ManagementController class, at runtime. The following code sample shows
what the high-level ManagementController class looks like if you decide to use dependency injection.

16

ch a pter t wo

public class ManagementController : Controller

{
private readonly ITenantStore tenantStore;
public ManagementController(ITenantStore tenantStore)
{
this.tenantStore = tenantStore;
}
public ActionResult Index()
{
var model = new TenantPageViewData<IEnumerable<string>>(this.tenantStore.GetTenantNames())
{
Title = "Subscribers"
};
return this.View(model);
}
...
}

As you can see in this sample, the ManagementController constructor receives an ITenantStore instance as a
parameter, injected by some other class. The only dependency in the ManagementContoller class is on the
interface type. This is better because it doesnt have any knowledge of the class or component that is responsible for instantiating the ITenantStore object.
In Figure 2, the class that is responsible for instantiating the TenantStore object and inserting it into the
ManagementController class is called the DependencyInjectionContainer class.

ManagementController
class

DependencyInjectionContainer
class

TenantStore class

ITenantStore
interface

Figure 2
Dependencies when using dependency injection

Chapter 3, Dependency Injection with Unity, will describe in more detail what happens in the
DependencyInjectionContainer class.

Dependency Injection

17

The key difference between the Figure 1 and Figure 2 is the direction of the
dependency from the ManagementController class. In Figure 2, the only dependency the ManagementController class has is on the ITenantStore interface.

Object Composition
So far in this chapter, you have seen how dependency injection can simplify classes such as the ManagementController
class and minimize the number of dependencies between
classes in your application. The previous chapter explained
some of the benefits of this approach, such as maintainability and testability, and showed how this approach relates to
the SOLID principles of object-oriented programming. You
will now see how this might work in practice: in particular,
how and where you might use dependency injection in your
own applications.

In Figure 2, the DependencyInjectionContainer

class may manage the dependencies of
multiple high level client classes such as the
ManagementController class on multiple service
classes such as the TenantStore class.
You can use either a dependency injection
container or implement dependency injection
manually using factories. As youll see in the next
chapter, using a container is easier and provides
additional capabilities such as lifetime management,
interception, and registration by convention.

If you adopt the dependency injection approach, you will

have many classes in your application that require some other class or component to pass the necessary dependencies into their constructors or methods as
parameters or as property values before you can use them. This implies that your
application requires a class or component that is responsible for instantiating all
the required objects and passing them into the correct constructors, methods,
and properties: your application must know how to compose its object graph
before it can perform any work. This must happen very early in the applications
lifecycle: for example, in the Main method of a console application, in the
Global.asax in a web application, in a roles OnStart method in a Windows
Azure application, or in the initialization code for a test method.

Typically, you should

place all the code tells the
application how to build
its object graph in a single
location; this is known as the
Composition Root pattern.
This makes it much easier
to maintain and update the
application.

Object Lifetime
You should determine when to create the objects in your application based on
criteria such as which object is responsible for managing the state, is the object
shared, and how long the object will live for. Creating an object always takes a
finite amount of time that is determined by the objects size and complexity, and
once you have created an object, it occupies some of your systems memory.
In the example, youve seen in this chapter, there is a single ManagementController client class that uses an implementation of the ITenantStore interface. In a real application, there may be many other client classes that all need
ITenantStore instances. Depending on the specific requirements and structure
of your application, you might want each client class to have its own ITenantStore object, or have all the client classes share the same ITenantStore instance,
or for different groups of client classes each have their own ITenantStore instance.

Whichever way you create

an object, there is always
a trade-off between
performance and resource
utilization when you decide
where to instantiate it.

18

ch a pter t wo

If every client object has its own ITenantStore instance, then the ITenantStore
instance can be garbage collected along with the client object. If multiple client
objects share an ITenantStore instance, then the class or component that instantiates the shared ITenantStore object must responsible for tidying it up
when all the clients are finished with it.

Types of Injection
Typically, when you instantiate an object you invoke a class constructor and pass
any values that the object needs as parameters to the constructor. In the example that you saw earlier in this chapter, the constructor in the ManagementController class expects to receive an object that implements the ITenantStore
interface. This is an example of constructor injection and is the type of injection
you will use most often. There are other types of injection such as property
setter injection and method call injection, but they are less commonly used.

Property Setter Injection

As an alternative or in addition to passing a parameter to a constructor, you may
want to set a property value when you instantiate an object in your application.
The following code sample shows part of a class named AzureTable in an application that uses property injection to set the value of the ReadWriteStrategy
property when it instantiates AzureTable object.
public class AzureTable<T> : ...
{
public AzureTable(StorageAccount account)
: this(account, typeof(T).Name)
{
}
...
public IAzureTableRWStrategy ReadWriteStrategy
{ get; set; }
...
}

You should use property

setter injection when the
dependency is optional.
However dont use property
setter injection as a
technique to avoid polluting
your constructor with
multiple dependencies; too
many dependencies might
be an indicator of poor
design because it is placing
too much responsibility in
a single class. See the single
responsibility principle
discussed in Chapter 1.

Notice that the constructors are not responsible for setting the read/write
strategy and that the type of the ReadWriteStrategy property is an interface
type. You can use property setter injection to provide an instance of the IAzureTableRWStrategy type when your dependency injection container constructs
an instance of AzureTable<T>.
You should only use property setter injection if the class has a usable default
value for the property. While you cannot forget to call a constructor, you can
forget to set a property such as the ReadWriteStrategy property in the example above.
However, dependencies are rarely optional when you are building a LOB
application. If you do have an optional dependency, consider using
constructor injection and injecting an empty implementation (the Null
Object Pattern.)

Dependency Injection

19

Method Call Injection

In a similar way to using property setter injection, you might want to invoke a
method when the application instantiates an object to perform some initialization that is not convenient to perform in a constructor. The following code
sample shows part of a class named MessageQueue in an application that uses
method injection to initialize the object.
public class MessageQueue<T> : ...
{
...
public MessageQueue(StorageAccount account)
: this(account, typeof(T).Name.ToLowerInvariant())
{
}
public MessageQueue(StorageAccount account, string queueName)
{
...
}
public void Initialize(TimeSpan visibilityTimeout,
IRetryPolicyFactory retryPolicyFactory)
{
...
}
...
}

In this example, the Initialize method has one concrete parameter type and one
interface parameter type. You can use method injection to provide an instance
of the IRetryPolicyFactory type when your dependency injection container
constructs an instance of MessageQueue<T>.
Method call injection is useful when you want to provide some additional information about the context that the object is being used in that cant be passed
in as a constructor parameter.

When You Shouldnt Use Dependency Injection

Dependency injection is not a silver bullet. There are reasons for not using it in
your application, some of which are summarized in this section.
Dependency injection can be overkill in a small application, introducing

additional complexity and requirements that are not appropriate or useful.

In a large application, it can make it harder to understand the code and

what is going on because things happen in other places that you cant
immediately see, and yet they can fundamentally affect the bit of code you
are trying to read. There are also the practical difficulties of browsing code
like trying to find out what a typical implementation of the ITenantStore
interface actually does. This is particularly relevant to junior developers and
developers who are new to the code base or new to dependency injection.

Both property setter and

method injection may be
useful when you need to
support legacy code that
uses properties and methods
to configure instances.

20

ch a pter t wo

You need to carefully consider if and how to introduce dependency injec-

Programming languages
shape the way we think and
the way we code. For a good
exploration of the topic of
dependency injection when
the functional programming
model is applied, see the
article Dependency Injection
Without the Gymnastics by
Tony Morris.
According to Mark Seeman,
using dependency injection
can be dangerous for
your career because it
may increase your overall
knowledge of good API
design. Once you learn how
proper loosely coupled code
can look like, it may turn out
that you will have to decline
lots of job offers because
you would otherwise have
to work with tightly coupled
legacy apps.
What are the downsides to
using Dependency Injection?
On StackOverflow.

tion into a legacy application that was not built with inversion of control
in mind. Dependency injection promotes a specific style of layering and
decoupling in a system that may pose challenges if you try to adapt an
existing application, especially with an inexperienced team.
Dependency injection is far less important in functional as opposed to
object-oriented programming. Functional programming is becoming a more
common approach when testability, fault recovery, and parallelism are key
requirements.
Type registration and resolving do incur a runtime penalty: very negligible
for resolving, but more so for registration. However, the registration should
only happen once.

Summary
In this chapter, youve seen how dependency injection differs from patterns such
as the factory patterns and the service locator pattern by adopting a push model,
whereby some other class or component is responsible for instantiating the dependencies and injecting them into your objects constructor, properties, or
methods. This other class or component is now responsible for composing the
application by building the complete object graph, and in some cases it will also
be responsible for managing the lifetime of the objects that it creates. In the next
chapter, youll see how you can use the Unity container to manage the instantiation of dependent objects and their lifetime.

More Information
All links in this book are accessible from the books online bibliography
available at: http://aka.ms/unitybiblio

Dependency Injection
with Unity

Introduction
In previous chapters, you saw some of the reasons to use dependency injection
and learned how dependency injection differs from other approaches to decoupling your application. In this chapter youll see how you can use the Unity dependency injection container to easily add a dependency injection framework
to your applications. On the way, youll see some examples that illustrate how
you might use Unity in a real-world application.

The Dependency Injection Lifecycle: Register,

Resolve, Dispose
In the previous chapter, you saw how the ManagementController class has a
constructor that expects to be injected with an object of type ITenantStore.
The application must know at run time which implementation of the ITenantStore interface it should instantiate before it can go ahead and instantiate a
ManagementController object. There are two things happening here: something in the application is making a decision about how to instantiate an object
that implements the ITenantStore interface, and then something in the application is instantiating both that object and the ManagementController object.
We will refer to the first task as registration and the second as resolution. At
some point in the future, the application will finish using the ManagementController object and it will become available for garbage collection. At this
point, it may also make sense for the garbage collector to dispose of the ITenantStore instance if other client classes do not share the same instance.

Typically, you perform the

registration of the types
that require dependency
injection in a single method
in your application; you
should invoke this method
early in your applications
lifecycle to ensure that the
application is aware of all of
the dependencies between
its classes. Unity also
supports configuring the
container declaratively from
a configuration file.

The Unity container can manage this register, resolve, dispose cycle making it
easy to use dependency injection in your applications. The following sections
illustrate this cycle using a simple example. Later in this chapter you will see a
more sophisticated real-world sample and learn about some alternative approaches.

You should always try to write container-agnostic code (except

for the one place at the root of the application where you
configure the container) in order to decouple your application
from the specific dependency injection container you are using.

21

22

ch a pter thr ee

Register
Using the Unity container, you can register a set of mappings that determine
what concrete type you require when a constructor (or property or method)
identifies the type to be injected by an interface type or base class type. As a
reminder, here is a copy of the constructor in the ManagementController class
showing that it requires an injection of an object that implements the ITenantStore interface.
public ManagementController(ITenantStore tenantStore)
{
this.tenantStore = tenantStore;
}

The following code sample shows how you could create a new Unity container
and then register the concrete type to use when a ManagementController instance requires an ITenantStore instance.
var container = new UnityContainer();
container.RegisterType<ITenantStore, TenantStore>();

The RegisterType method shown here tells the container to instantiate a

TenantStore object when it instantiates an object that requires an injection of
an ITenantStore instance through a constructor, or method, or property. This
example represents one of the simplest types of mapping that you can define
using the Unity container. As you continue through this chapter, you will see
other ways to register types and instances in the Unity container, that handle
more complex scenarios and that provide greater flexibility.
Youll see later that with
Unity you can also register
a class type directly without
a mapping from an interface
type.

Resolve
The usage of the RegisterType method shown in the previous section defines
the mapping between the interface type used in the client class and the concrete
type that you want to use in the application. To instantiate the ManagementController and TenantStore objects, you must invoke the Resolve method.
var controller = container.Resolve<ManagementController>();

Note that in this example, you do not instantiate the ManagementController

object directly, rather you ask the Unity container to do it for you so that the
container can resolve any dependencies. In this simple example, the dependency
to resolve is for an ITenantStore object. Behind the scenes, the Unity container
first constructs a TenantStore object and then passes it to the constructor of
the ManagementController class.

Dependency Injection with Unit y

23

Dispose
In the simple example shown in the previous two sections on registering and
resolving types, the application stores a reference to the ManagementController object in the controller variable and the Unity container creates a
new TenantStore instance to inject whenever you call the Resolve method.
When the controller variable goes out of scope and becomes eligible for garbage
collection, the TenantStore object will also be eligible for garbage collection.

Registering and Resolving in your Code

One of the original motivations, discussed in Chapter 1, for a loosely coupled
design and dependency injection was maintainability. One of the ways that dependency injection can help you to create more maintainable solutions is by describing, in a single location, how to compose your application from all of its
constituent classes and components. From the perspective of Unity, this is the
type registration information. Therefore, it makes sense to group all of the type
registrations together in a single method that you invoke very early on in your
applications lifecycle; usually, directly in the applications entry point. For example, in a web application, you could invoke the method that performs all of the
registrations from within the Application_Start method in the global.asax.cs or
global.asax.vb file, in a desktop application you invoke it from the Main method.

By default, the Unity

container doesnt hold a
reference to the objects
it creates: to change this
default behavior you need to
use one of the Unity lifetime
managers.

Typically, you can call the Resolve method when you need an instance of a
particular type in your application. The section Lifetime Management later in
this chapter discusses the options for controlling the lifetime of objects resolved
from the container: for example, do you want the container return a new instance each time you resolve a particular type, or should the container maintain
a reference to the instance.
You should perform all the registrations in a single location in your code or in a configuration file. This makes it easy to manage
the dependencies in your application. In a highly modular application, each module might be responsible for its own registration
and manage its own container.
Using a configuration file for registrations can be a brittle and error prone solution. It can also lead to the illusion that this
configuration can be changed without proper testing. Consider which settings, if any, need to be configurable after your
solution is deployed.

Adding Unity to Your Application

As a developer, before you can write any code that uses Unity, you must configure your Visual Studio project with all of the necessary assemblies, references,
and other resources that youll need. For information about how you can use
NuGet to prepare your Visual Studio project to work with Unity, see the topic
Adding Unity to Your Application.
NuGet makes it very easy
for you to configure your
project with all of the
prerequisites for using Unity.

24

ch a pter thr ee

A Real-World Example

Its useful to adopt a

standard name for the
class that contains your
type registration code;
for example ContainerBootstrapper.

The following example is taken from a web role implemented using ASP.NET MVC.
You may find it useful to open the sample application, DIwithUnitySample, that
accompanies this guide in Visual Studio while you read this section. At first sight
the contents of this RegisterTypes method (in the ContainerBootstrapper class
in the Surveys project) might seem to be somewhat complex; the next section will
discuss the various type registrations in detail, and the following section will describe how the application uses these registrations to resolve the types it needs at
runtime. This example also illustrates how you should perform all of the type
registration in a single method in your application.
publicstaticvoidRegisterTypes(IUnityContainercontainer)
{
varstorageAccountType=typeof(StorageAccount);
varretryPolicyFactoryType=typeof(IRetryPolicyFactory);
//Instanceregistration
StorageAccountaccount=
ApplicationConfiguration.GetStorageAccount("DataConnectionString");
container.RegisterInstance(account);
//Registerfactories
container
.RegisterInstance<IRetryPolicyFactory>(
newConfiguredRetryPolicyFactory())
.RegisterType<ISurveyAnswerContainerFactory,
SurveyAnswerContainerFactory>(
newContainerControlledLifetimeManager());
//Registertabletypes
container
.RegisterType<IDataTable<SurveyRow>,DataTable<SurveyRow>>(
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,Constants.SurveysTableName))
...
//Registermessagequeuetype,usetypeofwithopengenerics
container
.RegisterType(
typeof(IMessageQueue<>),
typeof(MessageQueue<>),
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,typeof(String)));
...
//Registerstoretypes
container
.RegisterType<ISurveyStore,SurveyStore>()
.RegisterType<ITenantStore,TenantStore>()
.RegisterType<ISurveyAnswerStore,SurveyAnswerStore>(
newInjectionFactory((c,t,s)=>newSurveyAnswerStore(

Dependency Injection with Unit y

25

container.Resolve<ITenantStore>(),
container.Resolve<ISurveyAnswerContainerFactory>(),
container.Resolve<IMessageQueue<SurveyAnswerStoredMessage>>(
newParameterOverride(
"queueName",Constants.StandardAnswerQueueName)),
container.Resolve<IMessageQueue<SurveyAnswerStoredMessage>>(
newParameterOverride(
"queueName",Constants.PremiumAnswerQueueName)),
container.Resolve<IBlobContainer<List<String>>>())));
}

To see the complete ContainerBootstrapper class, you can open the DIwithUnitySample sample application
that accompanies this guidance.
Figure 1 illustrates the object graph that the container will generate if a client resolves the ISurveyAnswerStore
type from the container with the type registrations shown in the previous code sample.
SurveyAnswerStore :
ISurveyAnswerStore

EntitiesBlobContainer<List<String>> :
IBlobContainer<List<string>>

TenantStore :
ITenantStore

EntitiesBlobContainer<Tenant>:
IBlobContainer<Tenant>

FilesBlobContainer :
IBlobContainer<byte[]>

MessageQueue<SurveyAnswerStoredMessage> :
IMessageQueue<SurveyAnswerStoredMessage>
(Standard messages)

SurveyAnswerContainerFactory :
ISurveyAnswerContainerFactory

MessageQueue<SurveyAnswerStoredMessage> :
IMessageQueue<SurveyAnswerStoredMessage>
(Premium messages)

RetryPolicyFactory :
IRetryPolicyFactory

CloudStorageAccount

Figure 1
Resolving the ISurveyAnswerStore type

UnityContainer :
IUnityContainer

26

ch a pter thr ee

Figure 1 illustrates the object graph that the container creates when you resolve
the ISurveyAnswerStore type from the registrations shown in the previous
code listing. There are some important points to note from Figure1.
The container injects the SurveyAnswerStore with five objects that the

container itself resolves: a TenantStore object, a SurveyAnswerContainerFactory object, an EntitiesBlobContainer object, and two MessageQueue
objects. Note that an explicit factory delegate is used to determine what
must be injected to create the store.
The container also resolves additional objects such as an EntitiesBlobContainer object and a FilesBlobContainer object to inject into the
TenantStore instance.
Many of the objects instantiated by the container share the same instances
of the RetryPolicyFactory and CloudStorageAccount objects which are
registered using the RegisterInstance method. Instance registration is
discussed in more detail later in this chapter.
The container injects the SurveyAnswerContainerFactory instance with
an instance of the Unity container. Note that as a general rule, this is not a
recommended practice.
The following sections discuss all of these points (and more) in detail. Figure 1
is intended to give an idea of what you can achieve with dependency injection
in your applications.

Type Registrations in the Example

The previous code listing gives examples of many of the different types of registration that you can perform with the Unity container. This section examines
each part of the registration individually.
Instance Registration
The simplest type of registration is instance registration where the container is
responsible for maintaining a reference to a singleton instance of a type. For
example:
Bear in mind, that if your
registrations start to
become too complicated
or fragile, you are probably
doing it wrong.

StorageAccountaccount=
ApplicationConfiguration.GetStorageAccount("DataConnectionString");
container.RegisterInstance(account);

Here, instead of registering a mapping for a type to resolved later, the application creates a CloudStorageAccount object and registers the instance with the
container. This means that the CloudStorageAccount object is created at registration time, and that only a single instance of the object exists in the container. This single instance is shared by many of the other objects that the container instantiates. Figure 1 shows that many of the objects that the container
creates when a client resolves the ISurveyAnswerStore type share this
CloudStorageAccount object instance.

Dependency Injection with Unit y

You can also use the ContainerControlledLifetimeManager class with the

RegisterType method to create a singleton instance where the container maintains a reference to the object. The section Lifetime Management later in this
chapter covers this in more detail.
Simple Type Registration
The most common type registration you will see maps an interface type to a
concrete type. For example:
container.RegisterType<ISurveyStore,SurveyStore>();

Later, you can resolve the ISurveyStore type as shown in the following example, and the container will inject any of the required dependencies into the
SurveyStore object that it creates.
varsurveyStore=container.Resolve<ISurveyStore>();

Constructor Injection
The following code sample shows the constructor for the DataTable class that
takes three parameters.
publicDataTable(StorageAccountaccount,
IRetryPolicyFactoryretryPolicyFactory,stringtableName)
{
...
}

The registrations for the DataTable types in the container includes an InjectionConstructor that defines how the container should resolve the parameter types.
The container passes to the constructor references to the registered StorageAccount and RetryPolicyFactory instances, and a string that specifies the name
of the table to use.
container
.RegisterType<IDataTable<SurveyRow>,DataTable<SurveyRow>>(
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,Constants.SurveysTableName))
.RegisterType<IDataTable<QuestionRow>,DataTable<QuestionRow>>(
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,Constants.QuestionsTableName));

If the SurveyStore class

has multiple constructors
with the same number
of parameters you
can use either the
InjectionConstructor
attribute, the API,
or the configuration
file to disambiguate
between the different
SurveyStore constructors.
However, although
InjectionConstructor
attributes are easy to use,
they do couple your code to
the container.
In most cases, components
should have a single
constructor and the
constructor defines the
dependencies of that
component.

27

28

ch a pter thr ee

The sample application uses a similar approach to register the blob container
types it uses:
container
.RegisterType<IBlobContainer<List<string>>,
EntitiesBlobContainer<List<string>>>(
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,Constants.SurveyAnswersListsBlobName))
.RegisterType<IBlobContainer<Tenant>,
EntitiesBlobContainer<Tenant>>(
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,Constants.TenantsBlobName))
.RegisterType<IBlobContainer<byte[]>,
FilesBlobContainer>(
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,Constants.LogosBlobName,"image/jpeg"))
.RegisterType<IBlobContainer<SurveyAnswer>,
EntitiesBlobContainer<SurveyAnswer>>(
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,typeof(string)));

Unity supports property and method injection in addition to the constructor

injection shown in this example. If you use property injection then, as with any
property, you should ensure that any properties have a useful default value. It is
easy to forget to set a property.
Registering Open Generics

Both the storage account

and retry policy factory
are singletons. The storage
account is registered using
the RegisterInstance
method, the retry policy
factory is registered using
the ContainerControlledLifetimeManager class that
youll learn more about later
in this chapter.

The example code uses a slightly different approach to register the message
queue types: it uses an overload of the RegisterTypes method that takes types
as standard parameters instead of using type parameters.
container
.RegisterType(
typeof(IMessageQueue<>),
typeof(MessageQueue<>),
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,typeof(string)));

This approach enables you to resolve the message queue type with any type
parameter. The example uses the SurveyAnswerStoredMessage type:
container.Resolve<IMessageQueue<SurveyAnswerStoredMessage>>(...);

Dependency Injection with Unit y

29

Parameter Overrides
The ContainerBootstrapper class contains several examples where one of the InjectionConstructor constructor parameters is typeof(string). For example, in the message queue registration and in the blob container for
survey answers registration:
container.RegisterType(
typeof(IMessageQueue<>),
typeof(MessageQueue<>),
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,typeof(string)));
...
container.RegisterType<IBlobContainer<SurveyAnswer>,
EntitiesBlobContainer<SurveyAnswer>>(
newInjectionConstructor(storageAccountType,
retryPolicyFactoryType,typeof(string)));

The container does not include a registration that it can use to resolve this type. Therefore, when you resolve
either the IMessageQueue<> or IBlobContainer<SurveyAnswer> types, you must provide a value for this
parameter otherwise the resolution will fail. This provides a convenient method to pass parameter values that
arent known at registration time to instances created by the container using the ParameterOverride type.

Resolving Types in the Example

The example solution performs the type registration described in the previous section in three locations: in a
standalone application that initializes the storage, in the web applications start-up phase, and in a factory class.
Simple Resolve
The usage in this simple standalone application is straightforward: it calls the RegisterTypes method to perform
all the registration, resolves a number of objects, and then invokes the Initialize method on each of them to
perform some initialization work before the application terminates. The following code sample shows this.
staticvoidMain(string[]args)
{
using(varcontainer=newUnityContainer())
{
ContainerBootstrapper.RegisterTypes(container);
container.Resolve<ISurveyStore>().Initialize();
container.Resolve<ISurveyAnswerStore>().Initialize();
container.Resolve<ITenantStore>().Initialize();
Console.WriteLine("Done");
Console.ReadLine();
}
}

In this example, after the Initialization methods have run, the container is disposed.

30

ch a pter thr ee

Resolving in an MVC Application

The usage in the MVC application is more sophisticated: the application configures a container that the application will use at start-up, and then resolves the various types as and when it needs them. Remember that this is
an ASP.NET MVC application; therefore, the container must be able to inject the MVC controller classes with
the various store and queue objects that they need. The Unity bootstrapper for ASP.NET MVC NuGet package (search for Unity3 in the NuGet package manager) simplifies this by adding libraries and source code to the
project in Visual Studio. The following code sample shows the RegisterTypes method in the UnityConfig class
that the NuGet package added to the project; you can choose to load the Unity configuration from your configuration file or add the registrations directly.
publicstaticvoidRegisterTypes(IUnityContainercontainer)
{
// NOTE: To load from web.config uncomment the line below...
// container.LoadConfiguration();
// TODO: Register your types here
// container.RegisterType<IProductRepository, ProductRepository>();
}

The Unity bootstrapper for ASP.NET MVC provides a UnityDependencyResolver class that resolves controllers from the container. If you need to configure injection for the controller classes then you need to add the
registration manually or add injection attributes to the controller classes
The following code sample shows part of the ManagementController class that custom factory class can resolve along with its dependency on the ITenantStore type from the registration information.
publicclassManagementController:Controller
{
privatereadonlyITenantStoretenantStore;
publicManagementController(ITenantStoretenantStore)
{
this.tenantStore=tenantStore;
}
...
}

Using the Per Request Lifetime Manager in MVC and WebAPI Application
The previous example showed how to use the Unity bootstrapper for ASP.NET MVC NuGet package to
handle registering and resolving controllers in an MVC application. The package also includes a PerRequestLifetime manager that you can use in an MVC application. This lifetime manager enables you to create instances of registered types that behave like singletons within the scope of an HTTP request.
If you are working with an ASP.NET Web API project, there is a Unity bootstrapper for ASP.NET WebApi
NuGet package that offers equivalent features (search for Unity3 in the NuGet package manager). You can use
both the Unity bootstrapper for ASP.NET WebApi and Unity bootstrapper for ASP.NET MVC packages in
the same project and they will share a single container configuration class.
There are third-party solutions available that offer similar support for ASP.NET WCF applications.

Dependency Injection with Unit y

Resolving with Run Time Information

You dont always know the values that you need to construct a dependency at
design time. In the example shown below, a user provides the name of the blob
container that the application must create at run time. In this example, type
resolution occurs in a factory class that determines the value of a constructor
parameter at registration time. The following code sample shows this factory
class.
publicclassSurveyAnswerContainerFactory:ISurveyAnswerContainerFactory
{
privatereadonlyIUnityContainerunityContainer;
publicSurveyAnswerContainerFactory(IUnityContainerunityContainer)
{
this.unityContainer=unityContainer;
}
publicIBlobContainer<SurveyAnswer>Create(stringtenant,stringsurveySlug)
{
varblobContainerName=string.Format(
CultureInfo.InvariantCulture,
"surveyanswers-{0}-{1}",
tenant.ToLowerInvariant(),
surveySlug.ToLowerInvariant());
returnthis.unityContainer.Resolve<IBlobContainer<SurveyAnswer>>(
newParameterOverride("blobContainerName",blobContainerName));
}
}

In this example, the Resolve method uses a parameter override to provide a value
for the blobContainerName parameter to the constructor of the EntitiesBlobContainer class that is registered in the container instance injected into the
SurveyAnswerContainerFactory object. Figure 1 shows how the SurveyAnswerContainerFactory object is injected with a Unity container instance when it is
resolved.
You saw previously how the application registered the IBlobContainer<SurveyAnswer> type using a string parameter in the injection constructor. Without the
parameter override, this registration would fail because the container cannot
resolve the string type.
You can also see the parameter overrides in use in the ContainerBootstrapper
class as part of the registration of the survey answer store. In this example, the
parameter overrides provide the name of the message queues to create. The
parameter overrides are supplied at resolve time when the registered InjectionFactory executes.

Because the application

supplies the name of the
blob container to create at
run time, the factory class
uses a parameter override
to supply this value to the
Resolve method.

31

32

ch a pter thr ee

container
.RegisterType<ISurveyAnswerStore,SurveyAnswerStore>(
newInjectionFactory((c,t,s)=>newSurveyAnswerStore(
container.Resolve<ITenantStore>(),
container.Resolve<ISurveyAnswerContainerFactory>(),
container.Resolve<IMessageQueue<SurveyAnswerStoredMessage>>(
newParameterOverride(
"queueName",Constants.StandardAnswerQueueName)),
container.Resolve<IMessageQueue<SurveyAnswerStoredMessage>>(
newParameterOverride(
"queueName",Constants.PremiumAnswerQueueName)),
container.Resolve<IBlobContainer<List<string>>>())));

Registration
In this section, youll learn more about how you can register types with the Unity container and the advantages and disadvantages of the different approaches. All of the examples youve seen so far have registered
types with the Unity container programmatically by using methods such as RegisterType and RegisterInstance.
Programmatically configuring a Unity container at runtime is convenient, and if you keep all of your registration
code together, it makes it easy to change those registrations when you modify the application. However, it does
mean that you must recompile the application if you need to change your registrations. In some scenarios, you
may want a mechanism that enables you to change the registrations in a configuration file and cause your application to use a different set of concrete classes at run time.

Named Type Registrations

Previously, you saw an example of how you can use parameter overrides to provide the name of the message
queue when you are resolving the IMessageQueue type. An alternative approach for this scenario is to use
named type registrations. In this case the message queue registration looks like the following where the two
alternative registrations are named Standard and Premium:
container
.RegisterType<IMessageQueue<SurveyAnswerStoredMessage>,
MessageQueue<SurveyAnswerStoredMessage>>(
"Standard",
newInjectionConstructor(storageAccountType,retryPolicyFactoryType,
Constants.StandardAnswerQueueName))
.RegisterType<IMessageQueue<SurveyAnswerStoredMessage>,
MessageQueue<SurveyAnswerStoredMessage>>(
"Premium",
newInjectionConstructor(storageAccountType,retryPolicyFactoryType,
Constants.PremiumAnswerQueueName));

Dependency Injection with Unit y

With this registration, you can now resolve the message queue parameters to
the SurveyAnswerStore constructor as follows using the named registrations:
container
.RegisterType<ISurveyAnswerStore,SurveyAnswerStore>(
newInjectionFactory((c,t,s)=>newSurveyAnswerStore(
container.Resolve<ITenantStore>(),
container.Resolve<ISurveyAnswerContainerFactory>(),
container.Resolve<IMessageQueue<SurveyAnswerStoredMessage>>(
"Standard"),
container.Resolve<IMessageQueue<SurveyAnswerStoredMessage>>(
"Premium"),
container.Resolve<IBlobContainer<List<string>>>())));

Design-Time Configuration
Unity enables you to load a collection of registrations from a configuration file
into a container. For example, you could add the following sections to your app.
config or web.config file to register mapping from the ITenantStore interface to
the TenantStore class.
You cannot use design-time configuration for Unity containers in Windows
Store apps. For more information, see Appendix A Unity and Windows
Store apps.
<configuration>
<configSections>
<section name="unity" type="Microsoft.Practices.Unity.Configuration.
UnityConfigurationSection, Microsoft.Practices.Unity.Configuration" />
</configSections>
<unity xmlns="http://schemas.microsoft.com/practices/2010/unity">
<namespace name="Tailspin.Web.Survey.Shared.Stores" />
<container>
<register type="ITenantStore" mapTo="TenantStore" />
</container>
</unity>
</configuration>

For more information about the structure of this configuration file, see the
topic Design Time Configuration.
To load the registration details from the configuration file, you can use the
following code. The LoadConfiguration extension method is defined in the
Microsoft.Practices.Unity.Configuration namespace.
IUnityContainer container = new UnityContainer();
container.LoadConfiguration();

If your Unity configuration includes any sensitive information, you should encrypt it. For more information, see Encrypting Configuration Information Using
Protected Configuration on MSDN.

Defining the registrations

in a configuration file
means that its possible
to make changes without
recompiling the application.
This can be useful for
customized deployments
and troubleshooting. You
can also adopt a hybrid
approach.

33

34

ch a pter thr ee

Registration by Convention

In many cases, you dont

need to register a type to
resolve it because Unitys
auto-wiring will figure it
out for you. You do need
to explicitly configure
mappings though, and for
this the registration by
convention feature may
prove useful.

This feature (also known as auto-registration) is intended to minimize the

amount of type registration code that you need to write. You may find it useful
to open the sample application, OtherUnitySamples, that accompanies this
guide in Visual Studio while you read this section. Rather than specify each type
mapping individually, you can direct the Unity container to scan a collection of
assemblies and then automatically register multiple mappings based on a set of
rules. If you only have a handful of simple registrations, it doesnt make sense to
use this feature, but if you have many types to register it will save you a considerable amount of effort.
This feature is only supported when you are configuring Unity
programmatically. You cant specify registration by convention in the
configuration file.
The registration by convention feature can scan a collection of assemblies, and
then create a set of mappings in a Unity container for some or all of the types it
discovers in the assemblies. Additionally, you can specify the lifetime managers
to use (these are described in more detail later in this chapter) and the details of
any injection parameters. You can easily modify the set of rules included out-ofthe-box to implement more sophisticated scenarios.
The following samples illustrate how you can use the registration by convention feature to create mappings in the Unity container using the RegisterTypes
method. These examples are based on the code in the OtherUnitySamples
Visual Studio solution included with this guidance. The RegisterTypes method
has the following parameters:

Parameter

Description

Types

This parameter is an enumerable collection of types that you want to register with the container.
These are the types that you want to register directly or create mappings to. You can create this
collection by providing a list of types directly or by using one of the methods of the built-in
AllClasses helper class: for example, the method FromLoadedAssemblies loads all of the
available types from the currently loaded assemblies.
You can use LINQ to filter this enumeration.

getFromTypes

This optional parameter identifies the types you want to map from in the container. The built-in
WithMappings helper class provides several options for this mapping strategy: for example, the
MatchingInterface property creates mappings where there are interfaces and implementations
that follow the naming convention ITenant and Tenant.

getName

This optional parameter enables you to control whether to create default registrations or named
registrations for the types. The built-in helper class WithName, enables you to choose between
using default registrations or named registrations that use the type name.

getLifeTimeManager

This optional parameter enables you to select from the built-in lifetime managers.

getInjectionMembers

This optional parameter enables you to provide definitions for any injection members for the
types that you are registering.

overwriteExistingMappings

This optional parameter enables you to control how the method behaves if it detects an attempt
to overwrite an existing mapping in the Unity container. By default, the RegisterTypes method
throws an exception if it detects such an attempt. If this parameter is true, the method silently
overwrites an existing mapping with a new one based on the values of the other parameters.

Dependency Injection with Unit y

35

The first example shows how you can create registrations for all the types that
implement an interface where the ITenant/Tenant naming convention is in use.
varcontainer=newUnityContainer();
container.RegisterTypes(
AllClasses.FromLoadedAssemblies(),
WithMappings.MatchingInterface,
WithName.Default);
}

This creates a set of transient registrations that map interfaces to types in the
container.
The following example creates named registrations and uses the ContainerControlledLifetimeManager type to ensure that resolving from the container
results in singletons.
varcontainer=newUnityContainer();
container.RegisterTypes(
AllClasses.FromLoadedAssemblies(),
WithMappings.MatchingInterface,
WithName.TypeName,
WithLifetime.ContainerControlled);

Because this example is using the lifetime manager, it registers all loaded types
in addition to mapping any interfaces to their matching types.
Lifetime managers are discussed later in this chapter in the section Lifetime
Management.
In some scenarios, you may need to combine registration by convention with
explicit registration. You can use registration by convention to perform the basic
registration of multiple types, and then explicitly add information for specific
types. The following example adds an InjectionConstructor to the type registration for the TenantStore class that was one of the types registered by calling
the RegisterTypes method.
varcontainer=newUnityContainer();
container.RegisterTypes(
AllClasses.FromLoadedAssemblies(),
WithMappings.MatchingInterface,
WithName.Default,
WithLifetime.ContainerControlled);
//Providesomeadditionalinformationforthisregistration
container.RegisterType<TenantStore>(newInjectionConstructor("Adatum"));

The examples youve seen so far use the FromLoadedAssemblies method to

provide a list of assemblies to scan for types; you may want to filter this list so
that you only register a subset of the types from these assemblies. The following
sample shows how to create a filter based on the namespace containing the
type. In this example, only types in the OtherUnitySamples namespace are
registered in the container.

Registration by convention
is intended to simplify
registering types with the
Unity container when you
have a large number of types
that must be registered with
similar settings.

36

ch a pter thr ee

varcontainer=newUnityContainer();
container.RegisterTypes(
AllClasses.FromLoadedAssemblies().Where(
t=>t.Namespace=="OtherUnitySamples"),
WithMappings.MatchingInterface,
WithName.Default,
WithLifetime.ContainerControlled);

The next example illustrates the use of the getInjectionMembers parameter:

this enables you specify types that should be injected when the registered type
is resolved from the container. Note that any types to be injected will be injected into all the types registered in the container by the call to the RegisterTypes method. The following example assumes that all of the types registered
have a constructor with a string parameter: any attempt to resolve a type without such a constructor parameter will result in an exception being thrown from
the container.
varcontainer=newUnityContainer();
container.RegisterTypes(
AllClasses.FromLoadedAssemblies(),
WithMappings.MatchingInterface,
getInjectionMembers:t=>newInjectionMember[]
{
newInjectionConstructor("Adatum")
});

A more practical use of the getInjectionMembers method is to use it to configure interception for all of the registered types (for more information about interception in Unity, see Chapter 5, Interception with Unity). In the following example, the registration by convention injects all of the registered types with the
custom LoggingInterceptionBehavior type using virtual method interception.
varcontainer=newUnityContainer();

You can keep the list of

classes in the OtherUnitySamples namespace
to use in other calls to the
RegisterTypes method to
avoid the need to re-scan
the assemblies.

container.AddNewExtension<Interception>();
container.RegisterTypes(
AllClasses.FromLoadedAssemblies().Where(
t=>t.Namespace=="OtherUnitySamples"),
WithMappings.MatchingInterface,
getInjectionMembers:t=>newInjectionMember[]
{
newInterceptor<VirtualMethodInterceptor>(),
newInterceptionBehavior<LoggingInterceptionBehavior>()
});

Note how this example uses a filter to ensure that only types in the OtherUnitySamples namespace are registered in the container. Without this filter, the
container will try to inject the interceptor into all the types from the loaded
assemblies: this includes the LoggingInterceptionBehavior type itself and this
results in a stack overflow.

Dependency Injection with Unit y

37

The block provides helper classes such as the AllClasses class that you can use
as parameters to the RegisterTypes method. You can create your own helper
classes, or use a lambda such the one used for the getInjectionMembers parameter in the previous example, to customize the behavior of registration by convention to your own requirements.
The following artificial example shows how the overwriteExistingMappings
parameter prevents the RegisterTypes method from throwing an exception if
you attempt to overwrite an existing mapping. The result is that the container
contains a mapping of the ITenantStore interface type to the TenantStore2
type.
varcontainer=newUnityContainer();
container.RegisterTypes(
new[]{typeof(TenantStore),typeof(TenantStore2)},
t=>new[]{typeof(ITenantStore)},
overwriteExistingMappings:true);

The mapping for the last type overwrites the previous mapping in the container.
You can extend the abstract RegistrationConvention class to define a reusable
convention that you can pass to the RegisterTypes method as a single parameter.
The following sample shows a how you can extend the RegistrationConvention
class.
class SampleConvention : RegistrationConvention
{
public override Func<Type, IEnumerable<Type>> GetFromTypes()
{
return t => t.GetTypeInfo().ImplementedInterfaces;
}
public override Func<Type, IEnumerable<InjectionMember>> GetInjectionMembers()
{
return null;
}
public override Func<Type, LifetimeManager> GetLifetimeManager()
{
return t => new ContainerControlledLifetimeManager();
}
public override Func<Type, string> GetName()
{
return t => t.Name;
}
public override IEnumerable<Type> GetTypes()
{
yield return typeof(TenantStore);
yield return typeof(SurveyStore);
}
}

Be cautious when you

use registration by
convention and consider the
implications of registering all
of the types discovered by
the RegisterTypes method
using the same options.

38

ch a pter thr ee

Registration by Convention and Generic Types

You can use registration by convention to register generic types. The following interface and class definitions
use a generic type and you can use the WithMappings.FromMatchingInterface helper method to create a
mapping between these types in the container.
publicinterfaceICustomerStore<T>
{
...
}
publicclassCustomerStore<T>:ICustomerStore<T>
{
...
}

This registers the open generic types but enables you to resolve using a closed generic. For example, you could
resolve instances using the following code.
var blobCustomerStore = container.Resolve<ICustomerStore<BlobStorage>>();
var tableCustomerStore = container.Resolve<ICustomerStore<TableStorage>>();

Its also possible to combine using registration by convention to register mappings for open generic types with
a specific registration a closed generic type as shown in the following sample. The closed type registration will
always take priority.
container.RegisterTypes(
//Registersopengenerics
AllClasses.FromLoadedAssemblies(),
WithMappings.FromMatchingInterface,
WithName.Default);
//Addaregistrationforaclosedgeneric type
container.RegisterType<ICustomerStore<TableStorage>,
CustomerStore<TableStorage>>();

Using Child Containers

Although you can use named registrations to define different mappings for the same type, an alternative approach is to use a child container. The following code sample illustrates how to use a child container to resolve
the IMessageQueue<SurveyAnswerStoredMessage> type with a different connection string.
var storageAccountType = typeof(StorageAccount);
var storageAccount = ApplicationConfiguration
.GetStorageAccount("DataConnectionString");
container.RegisterInstance(storageAccount);
container
.RegisterType<IMessageQueue<SurveyAnswerStoredMessage>,
MessageQueue<SurveyAnswerStoredMessage>>(
"Standard",
new InjectionConstructor(storageAccountType,
"StandardAnswerQueue"),
retryPolicyFactoryProperty);

Dependency Injection with Unit y

39

var childContainer = container.CreateChildContainer();

var alternateAccount = ApplicationConfiguration
.GetStorageAccount("AlternateDataConnectionString");
childContainer.RegisterInstance(alternateAccount);
childContainer
.RegisterType<IMessageQueue<SurveyAnswerStoredMessage>,
MessageQueue<SurveyAnswerStoredMessage>>(
"Standard",
new InjectionConstructor(storageAccountType,
"StandardAnswerQueue"),
retryPolicyFactoryProperty);

You can now resolve the type IMessageQueue<SurveyAnswerStoredMessage> from either the original parent
container or the child container. Depending on which container you use, the MessageQueue instance is injected with a different set of account details.
The advantage of this approach over using different named registrations, is that if you attempt to resolve a type
from the child container and that type is not registered in the child container, then Unity will automatically fall
back to try and resolve the type from the parent container.
You can also use child containers to manage the lifetime of objects. This use of child containers is discussed
later in this chapter.

Viewing Registration Information

You can access the registration data in the container programmatically if you want to view details of the registration information. The following code sample shows a basic approach to viewing the registrations in a container.
Console.WriteLine("Containerhas{0}Registrations:",
container.Registrations.Count());
foreach(ContainerRegistrationitemincontainer.Registrations)
{
Console.WriteLine(item.GetMappingAsString());
}

This example uses an extension method called GetMappingAsString to display formatted output. The output
using the registrations shown at the start of this chapter looks like the following:
Output
Container has 14 Registrations:
+ IUnityContainer '[default]' Container
+ StorageAccount '[default]' ContainerControlled
+ IRetryPolicyFactory '[default]' ContainerControlled
+ IDataTable`1<SurveyRow> -> DataTable`1<SurveyRow> '[default]' Transient
+ IDataTable`1<QuestionRow> -> DataTable`1<QuestionRow> '[default]' Transient
+ IMessageQueue`1<SurveyAnswerStoredMessage> ->
MessageQueue`1<SurveyAnswerStoredMessage> 'Standard' Transient
+ IMessageQueue`1<SurveyAnswerStoredMessage> ->
MessageQueue`1<SurveyAnswerStoredMessage> 'Premium' Transient

40

ch a pter thr ee

Having a utility that can

display your registrations
can help troubleshoot issues.
This is especially useful if
you define the registrations
in a configuration
file. However, getting
registration information is
expensive and you should
only do it to troubleshoot
issues.

+ IBlobContainer`1<List`1<String>> -> EntitiesBlobContainer`1<List`1<String>>

'[default]' Transient
+ IBlobContainer`1<Tenant> -> EntitiesBlobContainer`1<Tenant> '[default]'
Transient
+ IBlobContainer`1<Byte[]> -> FilesBlobContainer '[default]' Transient
+ ISurveyStore -> SurveyStore '[default]' Transient
+ ITenantStore -> TenantStore '[default]' Transient
+ ISurveyAnswerStore -> SurveyAnswerStore '[default]' Transient
+ ISurveyAnswerContainerFactory -> SurveyAnswerContainerFactory '[default]'
Transient

The following code sample shows the extension method that creates the formatted output.
staticclassContainerRegistrationsExtension
{
publicstaticstringGetMappingAsString(
thisContainerRegistrationregistration)
{
stringregName,regType,mapTo,lifetime;
varr=registration.RegisteredType;
regType=r.Name+GetGenericArgumentsList(r);
varm=registration.MappedToType;
mapTo=m.Name+GetGenericArgumentsList(m);
regName=registration.Name??"[default]";
lifetime=registration.LifetimeManagerType.Name;
if(mapTo!=regType)
{
mapTo="->"+mapTo;
}
else
{
mapTo=string.Empty;
}
lifetime=lifetime.Substring(
0,lifetime.Length-"LifetimeManager".Length);
returnstring.Format(
"+{0}{1}'{2}'{3}",regType,mapTo,regName,lifetime);
}
privatestaticstringGetGenericArgumentsList(Typetype)
{
if(type.GetGenericArguments().Length==0)returnstring.Empty;
stringarglist=string.Empty;
boolfirst=true;

Dependency Injection with Unit y

41

foreach(Typetintype.GetGenericArguments())
{
arglist+=first?t.Name:","+t.Name;
first=false;
if(t.GetGenericArguments().Length>0)
{
arglist+=GetGenericArgumentsList(t);
}
}
return"<"+arglist+">";
}
}

Resolving
You have already seen some sample code that shows some simple cases of resolving types from the Unity
container. For example:
var surveyStore = container.Resolve<ISurveyStore>();
container.Resolve<IMessageQueue<SurveyAnswerStoredMessage>>("Premium");

The second of these two examples shows how to resolve a named type registration from the container.
You can override the registration information in the container when you resolve a type. The following code
sample uses a dependency override to specify the type of controller object to create:
protected override IController GetControllerInstance(
RequestContext requestContext, Type controllerType)
{
return this.container.Resolve(controllerType,
new DependencyOverride<RequestContext>(requestContext)) as IController;
}

This example assumes that there is no registration in the container for the controller type that is passed to the
GetControllerInstance method, so the dependency override defines a registration as the type is resolved. This
enables the container to resolve any other dependencies in the controller class. You can use a dependency
override to override an existing registration in addition to providing registration information that isnt registered
with the container.
The SurveyAnswerContainerFactory class uses a parameter override to specify the value of a constructor
parameter that the application cannot know until run time.

Resolving in an ASP.NET Web Application

In the example shown earlier in this chapter, you saw how to integrate Unity into an MVC application, so that
you can use Unity to resolve any dependencies in your MVC controller classes by creating a custom MVC
controller factory.
Standard ASP.NET web applications face a similar problem: how do you resolve any dependencies in your web
page classes when you have no control over how and when ASP.NET instantiates your page objects. The
aExpense reference implementation demonstrates how you can address this issue.

42

ch a pter thr ee

The following code sample shows part of a page class in the aExpense web application.
public partial class Default : Page
{
[Dependency]
public IExpenseRepository Repository { get; set; }
protected void Page_Init(object sender, EventArgs e)
{
this.ViewStateUserKey = this.User.Identity.Name;
}
protected void Page_Load(object sender, EventArgs e)
{
var expenses = Repository.GetExpensesByUser(this.User.Identity.Name);
this.MyExpensesGridView.DataSource = expenses;
this.DataBind();
}
...
}

This example shows a property, of type IExpenseRepository decorated with the Dependency attribute, and
some standard page life-cycle methods, one of which uses the Repository property. The Dependency attribute
marks the property for property setter injection by the Unity container.
The following code sample shows the registration of the IExpenseRepository type.
public static void Configure(IUnityContainer container)
{
container
.RegisterInstance<IExpenseRepository>(new ExpenseRepository())
.RegisterType<IProfileStore, SimulatedLdapProfileStore>()
.RegisterType<IUserRepository, UserRepository>(
new ContainerControlledLifetimeManager());
...
}

The following code sample, from the Global.asax.cs file, shows how the web application performs the type
registration in the Application_Start method, and uses the BuildUp method in the Application_PreRequestHandlerExecute method to perform the type resolution.
public class Global : System.Web.HttpApplication
{
protected void Application_Start(object sender, EventArgs e)
{
IUnityContainer container = Application.GetContainer();
ContainerBootstrapper.Configure(container);
}
...

Dependency Injection with Unit y

protected void Application_PreRequestHandlerExecute(object sender, EventArgs e)

{
var handler = HttpContext.Current.Handler as System.Web.UI.Page;
if (handler != null)
{
var container = Application.GetContainer();
if (container != null)
{
container.BuildUp(handler.GetType(), handler);
}
}
}
}

The BuildUp method passes an existing object, in this case the ASP.NET page
object, through the container so that the container can inject any dependencies
into the object.

Resolving in a WCF Service

If you want to use Unity to automatically resolve types in a WCF service, you
need to modify the way that WCF instantiates the service so that Unity can
inject any dependencies. The example in this section is based on a simple WCF
calculator sample, and the following example code shows the interface and part
of the service class. You may find it useful to open the sample application,
UnityWCFSample, that accompanies this guide in Visual Studio while you
read this section. The example uses constructor injection, but you could just as
easily use method or property setter injection if required.
[ServiceContract(Namespace = "http://Microsoft.ServiceModel.Samples")]
public interface ICalculator
{
[OperationContract]
double Add(double n1, double n2);
[OperationContract]
double Subtract(double n1, double n2);
[OperationContract]
double Multiply(double n1, double n2);
[OperationContract]
double Divide(double n1, double n2);
}
public class CalculatorService : ICalculator
{
private ICalculationEngine calculatorEngine;
public CalculatorService(ICalculationEngine calculatorEngine)
{
this.calculatorEngine = calculatorEngine;
}

Using the BuildUp method,

you can only perform
property and method
injection. You cannot
perform constructor
injection because the object
has already been created.

43

44

ch a pter thr ee

public double Add(double n1, double n2)

{
double result = calculatorEngine.Add(n1, n2);
Console.WriteLine("Received Add({0},{1})", n1, n2);
Console.WriteLine("Return: {0}", result);
return result;
}
...
}

To modify WCF to use Unity to instantiate the service, you must provide a custom ServiceHost class that can
pass a Unity container instance into the WCF infrastructure as shown in the following example.
public class UnityServiceHost : ServiceHost
{
public UnityServiceHost(IUnityContainer container,
Type serviceType, params Uri[] baseAddresses)
: base(serviceType, baseAddresses)
{
if (container == null)
{
throw new ArgumentNullException("container");
}
foreach (var cd in this.ImplementedContracts.Values)
{
cd.Behaviors.Add(new UnityInstanceProvider(container));
}
}

The following code sample shows the UnityInstanceProvider class that resolves the service type (the
CalculatorService type in this example) from the Unity container.
public class UnityInstanceProvider
: IInstanceProvider, IContractBehavior
{
private readonly IUnityContainer container;
public UnityInstanceProvider(IUnityContainer container)
{
if (container == null)
{
throw new ArgumentNullException("container");
}
this.container = container;
}

Dependency Injection with Unit y

45

#region IInstanceProvider Members

public object GetInstance(InstanceContext instanceContext, Message message)
{
return this.GetInstance(instanceContext);
}
public object GetInstance(InstanceContext instanceContext)
{
return this.container.Resolve(instanceContext.Host.Description.ServiceType);
}
public void ReleaseInstance(InstanceContext instanceContext, object instance)
{
}
#endregion
#region IContractBehavior Members
public void AddBindingParameters(
ContractDescription contractDescription,
ServiceEndpoint endpoint,
BindingParameterCollection bindingParameters)
{
}
public void ApplyClientBehavior(
ContractDescription contractDescription,
ServiceEndpoint endpoint, ClientRuntime clientRuntime)
{
}
public void ApplyDispatchBehavior(
ContractDescription contractDescription,
ServiceEndpoint endpoint,
DispatchRuntime dispatchRuntime)
{
dispatchRuntime.InstanceProvider = this;
}
public void Validate(
ContractDescription contractDescription,
ServiceEndpoint endpoint)
{
}
#endregion
}

Now that you have defined the new UnityServiceHost class that uses Unity to instantiate your WCF service,
you must create an instance of the UnityServiceHost class in your run time environment. How you do this for
a self-hosted service is different from how you do it for a service hosted in IIS or WAS.

46

ch a pter thr ee

Using the UnityServiceHost Class with a Self-hosted Service

If you are self-hosting the service, you can instantiate the UnityServiceHost class directly in your hosting application and pass it a Unity container as shown in the following code sample.
class Program
{
static void Main(string[] args)
{
// Register types with Unity
using (IUnityContainer container = new UnityContainer())
{
RegisterTypes(container);
// Step 1 Create a URI to serve as the base address.
Uri baseAddress = new Uri("http://localhost:8000/GettingStarted/");
// Step 2 Create a ServiceHost instance
ServiceHost selfHost = new UnityServiceHost(container,
typeof(CalculatorService.CalculatorService), baseAddress);
try
{
// Step 3 Add a service endpoint.
selfHost.AddServiceEndpoint(typeof(ICalculator), new WSHttpBinding(),
"CalculatorService");
// Step 4 Enable metadata exchange.
var smb = new ServiceMetadataBehavior();
smb.HttpGetEnabled = true;
selfHost.Description.Behaviors.Add(smb);
// Step 5 Start the service.
selfHost.Open();
Console.WriteLine("The service is ready.");
Console.WriteLine("Press <ENTER> to terminate service.");
Console.WriteLine();
Console.ReadLine();
// Close the ServiceHostBase to shutdown the service.
selfHost.Close();
}
catch (CommunicationException ce)
{
Console.WriteLine("An exception occurred: {0}", ce.Message);
selfHost.Abort();
}
}
}
private static void RegisterTypes(IUnityContainer container)
{
container.RegisterType<ICalculationEngine, SimpleEngine>();
}
}

Dependency Injection with Unit y

47

Using the UnityServiceHost Class with Service Hosted in IIS or WAS

If you are hosting your WCF service in IIS or WAS it is a little more complex because you can no longer directly create a service host and by default, IIS will create a ServiceHost and not a UnityServiceHost instance.
To get around this problem, you must create a service host factory as shown in the following code sample.
class UnityServiceHostFactory : ServiceHostFactory
{
private readonly IUnityContainer container;
public UnityServiceHostFactory()
{
container = new UnityContainer();
RegisterTypes(container);
}
protected override ServiceHost CreateServiceHost(
Type serviceType, Uri[] baseAddresses)
{
return new UnityServiceHost(this.container,
serviceType, baseAddresses);
}
private void RegisterTypes(IUnityContainer container)
{
container.RegisterType<ICalculationEngine, SimpleEngine>();
}
}

This factory class creates a Unity container instance and passes it in to the constructor of the new UnityServiceHost class.
The final step is to instruct IIS or WAS to use the factory to create a service host. You can do this in the .svc
file for the service as shown in the following example.
<%@ ServiceHost Language="C#" Debug="True"
Service="CalculatorService.CalculatorService"
Factory="CalculatorService.UnityServiceHostFactory" %>

Automatic Factories
Sometimes, your application does not know all the details of the objects to construct until run time. For example, a class called SurveyAnswerStore uses one of two queues, depending on whether the tenant is a premium or standard tenant. A simple approach is to use Unity to resolve both queue types as shown in the following sample.
class SurveyAnswerStore : IsurveyAnswerStore
{
...
public class SurveyAnswerStore : IsurveyAnswerStore
{

48

ch a pter thr ee

...
private readonly IMessageQueue<SurveyAnswerStoredMessage>
standardSurveyAnswerStoredQueue;
private readonly IMessageQueue<SurveyAnswerStoredMessage>
premiumSurveyAnswerStoredQueue;
public SurveyAnswerStore(
ITenantStore tenantStore,
ISurveyAnswerContainerFactory surveyAnswerContainerFactory,
IMessageQueue<SurveyAnswerStoredMessage> standardSurveyAnswerStoredQueue,
IMessageQueue<SurveyAnswerStoredMessage> premiumSurveyAnswerStoredQueue,
IBlobContainer<List<string>> surveyAnswerIdsListContainer)
{
...
this.standardSurveyAnswerStoredQueue = standardSurveyAnswerStoredQueue;
this.premiumSurveyAnswerStoredQueue = premiumSurveyAnswerStoredQueue;
}
public void SaveSurveyAnswer(SurveyAnswer surveyAnswer)
{
var tenant = this.tenantStore.GetTenant(surveyAnswer.Tenant);
...
(SubscriptionKind.Premium.Equals(tenant.SubscriptionKind)
? this.premiumSurveyAnswerStoredQueue
: this.standardSurveyAnswerStoredQueue)
.AddMessage(new SurveyAnswerStoredMessage
{
...
});
}
}
...
}

In this example, when the container resolves the SurveyAnswerStore type it will inject two IMessageQueue
<SurveyAnswerStoredMessage> instances. If you know that only one of these instances will be used, you
might consider optimizing the solution to create only the instance you need.
One approach is to write a factory class that will instantiate the correct instance, and then take a dependency
on the factory. The following code sample shows this approach.
class SurveyAnswerStore : IsurveyAnswerStore
{
...
private readonly ISurveyAnswerQueueFactory surveyAnswerQueueFactory;
public SurveyAnswerStore(
ITenantStore tenantStore,
ISurveyAnswerContainerFactory surveyAnswerContainerFactory,
ISurveyAnswerQueueFactory surveyAnswerQueueFactory,
IBlobContainer<List<string>> surveyAnswerIdsListContainer)
{

Dependency Injection with Unit y

49

...
this.surveyAnswerQueueFactory = surveyAnswerQueueFactory;
}
public void SaveSurveyAnswer(SurveyAnswer surveyAnswer)
{
var tenant = this.tenantStore.GetTenant(surveyAnswer.Tenant);
...
((tenant.SubscriptionKind == "Premium")
? this.surveyAnswerQueueFactory.GetPremiumQueue()
: this.surveyAnswerQueueFactory.GetStandardQueue())
.AddMessage(new SurveyAnswerStoredMessage
{
...
});
}
...
}

For this approach to work, in addition to writing the factory class, you must register the factory class with the
container so that the container can inject it when it resolves the SurveyAnswerStore type.
A further refinement is to use Unitys automatic factory approach. Using this approach you do not need to write
and register a factory class, Unity creates a lightweight factory and registers it on your behalf. The following
code sample shows this approach.
class SurveyAnswerStore : IsurveyAnswerStore
{
...
private readonly Func<IMessageQueue<SurveyAnswerStoredMessage>>
standardSurveyAnswerQueueFactory;
private readonly Func<IMessageQueue<SurveyAnswerStoredMessage>>
premiumSurveyAnswerQueueFactory;
public SurveyAnswerStore(
ITenantStore tenantStore,
ISurveyAnswerContainerFactory surveyAnswerContainerFactory,
[Dependency("Standard")]Func<IMessageQueue<SurveyAnswerStoredMessage>>
standardSurveyAnswerQueueFactory,
[Dependency("Premium")]Func<IMessageQueue<SurveyAnswerStoredMessage>>
premiumSurveyAnswerQueueFactory,
IBlobContainer<List<string>> surveyAnswerIdsListContainer)
{
...
this.standardSurveyAnswerQueueFactory = standardSurveyAnswerQueueFactory;
this.premiumSurveyAnswerQueueFactory = premiumSurveyAnswerQueueFactory;
}

50

ch a pter thr ee

public void SaveSurveyAnswer(SurveyAnswer surveyAnswer)

{
var tenant = this.tenantStore.GetTenant(surveyAnswer.Tenant);
...
((tenant.SubscriptionKind == "Premium")
? premiumSurveyAnswerQueueFactory()
: standardSurveyAnswerQueueFactory())
.AddMessage(new SurveyAnswerStoredMessage
{
...
});
}
...
}

In this example, the dependencies of the SurveyAnswerStore class are on

values of the form Func<T>. This enables the container to generate delegates
that perform the type resolution when they are invoked: in the sample code,
the delegates are called premiumSurveyAnswerQueueFactory and standardSurveyAnswerQueueFactory.

You dont need to change

the registrations in the
container to make this
approach work.

One drawback of this specific example is that because the two queues use named
registrations in the container, you must use the Dependency attribute to specify
which named registration to resolve. This means that the SurveyAnswerStore
class has a dependency on Unity.

Deferred Resolution
Sometimes, you may want to resolve an object from the container, but defer the
creation of the object until you need to use it. You can achieve this with Unity
by using the Lazy<T> type from the .NET Framework; this type provides support
for the lazy initialization of objects.
To use this approach with Unity, you can register the type you want to use in
the standard way, and then use the Lazy<T> type when you resolve it. The following code sample shows this approach.
Lazy<T> doesnt work very
well with value types, and it
is better to avoid in this case.
You should use the Lazy<T>
type very cautiously.

// Register the type

container.RegisterType<MySampleObject>(newInjectionConstructor("default"));
// Resolve using Lazy<T>
vardefaultLazy=container.Resolve<Lazy<MySampleObject>>();
// Use the resolved object
varmySampleObject=defaultLazy.Value;

This example is adapted from the sample application, OtherUnitySamples,

included with this guidance.
You can use lazy resolution with the Unity lifetime managers. The following
example, again adapted from the sample application illustrates this with the
ContainerManagedLifetime class.

Dependency Injection with Unit y

// Register the type with a lifetime manager

container.RegisterType<MySampleObject>(
"other",newContainerControlledLifetimeManager(),
newInjectionConstructor("other"));
// Resolve the lazy type
vardefaultLazy1=container.Resolve<Lazy<MySampleObject>>("other");
// Resolve the lazy type a second time
vardefaultLazy2=container.Resolve<Lazy<MySampleObject>>("other");
// defaultLazy1 == defaultLazy2 is false
// defaultLazy1.Value == defaultLazy2.Value is true

For more information about Lazy<T>, see the topic Lazy<T> Class on MSDN.
You can also use the Resolve method to resolve registered types by using
Func<T> in a similar way.

Lifetime Management
When you resolve an object that you registered using the RegisterType method,
the container instantiates a new object when you call the Resolve method: the
container does not hold a reference to the object. When you create a new instance using the RegisterInstance method, the container manages the object
and holds a reference to it for the lifetime of the container.
Lifetime Managers manage the lifetimes of objects instantiated by the container.
The default lifetime manager for the RegisterType method is the TransientLifetimeManager and the default lifetime manager for the RegisterInstance
method is the ContainerControlledLifetimeManager. If you want the container to create or return a singleton instance of a type when you call the
Resolve method, you can use the ContainerControlledLifetimeManager type
when you register your type or instance. The following example shows how you
could tell the container to create a singleton instance of the TenantStore.
container.RegisterType<ITenantStore, TenantStore>(
new ContainerControlledLifetimeManager());

The first time that you resolve the ITenantStore type the container creates a
new TenantStore object and keeps a reference to it. On subsequent times when
you resolve the ITenantStore type, the container returns a reference to the
TenantStore object that it created previously. Some lifetime managers, such as
the ContainerControlledLifetimeManager, are used to dispose the created
objects when the container is disposed.
Unity includes five other lifetime managers, described in the following sections,
that you can use to address specific scenarios in your applications.

Lifetime managers enable

you to control for how
long the objects created
by the container should
live in your application.
You can override the
default lifetime managers
that the RegisterType and
RegisterInstance methods
use.

51

52

ch a pter thr ee

Hierarchical Lifetime Management

This type of lifetime management is useful if you have a hierarchy of containers. Earlier in this chapter, you saw
how to use child containers to manage alternative mappings for the same type. You can also use child containers to manage the lifetime of resolved objects. Figure 2 illustrates a scenario where you have created two child
containers and registered a type using the ContainerControlledLifetimeManager type to create a singleton.

Parent container

Child container #1

Client object

Child container #2

Resolved TenantStore
object
(Singleton)

Figure 2
Container hierarchy with ContainerControlledLifetimeManager lifetime manager

If the client object executes the following code that creates the containers, performs the registrations, and then
resolves the types, the three variables (tenant1, tenant2, and tenant3) all refer to the same instance managed
by the containers.
IUnityContainer container = new UnityContainer();
container.RegisterType<ITenantStore, TenantStore>(
new ContainerControlledLifetimeManager());
IUnityContainer child1 = container.CreateChildContainer();
IUnityContainer child2 = container.CreateChildContainer();
var tenant1 = child1.Resolve<ITenantStore>();
var tenant2 = child2.Resolve<ITenantStore>();
var tenant3 = container.Resolve<ITenantStore>();

However, if you use the HierarchicalLifetimeManager type, the container resolves the object as shown in Figure 3.

Dependency Injection with Unit y

Parent container

Child container #1

Child container #2

Client object
Resolved TenantStore
object A
(Singleton)

Resolved TenantStore
object C
(Singleton)

Resolved TenantStore
object B
(Singleton)

Figure 3
Container hierarchy with HierarchicalLifetimeManager lifetime manager

If the client executes the following code, the three variables (tenant1, tenant2,
and tenant3) each refer to different TenantStore instances.
IUnityContainer container = new UnityContainer();
container.RegisterType<ITenantStore, TenantStore>(
new HierarchicalLifetimeManager());
IUnityContainer child1 = container.CreateChildContainer();
IUnityContainer child2 = container.CreateChildContainer();
var tenant1 = child1.Resolve<ITenantStore>();
var tenant2 = child2.Resolve<ITenantStore>();
var tenant3 = container.Resolve<ITenantStore>();

Although you register the type with the parent container, each child container
now resolves its own instance. Each child container manages its own singleton
instance of the TenantStore type; therefore, if you resolve the same type from
container #1 a second time, the container returns a reference to the instance it
created previously.

Per Resolve Lifetime Management

Figure 4 shows part of the dependency tree for an application: the SurveysController type depends on the SurveyStore and SurveyAnswerStore types,
both the SurveyStore and SurveyAnswerStore types depend on the TenantStore type.

This approach is useful in

web applications where
you want to register your
types once, but then resolve
separate instances for each
client session. This assumes
that it is possible, with your
design, to map one child
container to each session.

53

54

ch a pter thr ee

SurveysController type

SurveyStore type

TenantStore type

SurveyAnswerStore
type

TenantStore type

Figure 4
Sample dependency tree

If you use the default TransientLifetimeManager class when you register the SurveysController type, then
when you resolve the SurveysController type, the container builds the object graph shown in Figure 5.

SurveysController
instance

SurveyStore
instance

TenantStore
instance

SurveyAnswerStore
instance

TenantStore
instance

Figure 5
Object graph generated using TransientLifetimeManager lifetime manager

Dependency Injection with Unit y

55

However, if you use the PerResolveLifetimeManager class in place of the TransientLifetimeManager class,
then the container builds the object graph shown in Figure 6. With the PerResolveLifetimeManager class, the
container reuses any instances it resolves during a call to the Resolve method in any other types it resolves
during the same call.

SurveysController
instance

SurveyStore
instance

TenantStore
instance

SurveyAnswerStore
instance

Figure 6
Object graph generated using the PerResolveLifetimeManager class

Externally Controlled Lifetime Management

If you resolve a type that was registered using the ContainerControlledLifetimeManager class, the container
creates a singleton instance and holds a strong reference to it: this means that the instance lives at least as long
as the container. However, if you use the ExternallyControlledLifetimeManager class, when you resolve the
type, the container creates a singleton instance but holds only a weak reference to it. In this case, you can directly manage the lifetime of the object: because of the weak reference, you can dispose of the object when
you no longer need it. This enables you to inject objects that are not owned by the container; for example you
might need to inject objects whose lifetime is managed by ASP.NET into instances created by the container.
Also, the ExternallyControlledLifetimeManager class does not dispose the instances it holds references to
when the container is disposed.

Per Request Lifetime Management

This lifetime manager is only available for use in Web applications when youve added the Unity bootstrapper
for ASP.NET MVC NuGet package to your project. The PerRequestLifetimeManager class enables the container to create new instances of registered types for each HTTP request in an ASP.NET MVC application or
an ASP.NET Web API application. Each call to Resolve a type within the context of a single HTTP request will
return the same instance: in effect, the Unity container creates singletons for registered types for the duration
of the HTTP request.

56

ch a pter thr ee

Think carefully about

the implications for state
management in your
application if you plan to use
the PerRequestLifetimeManager lifetime manager
class.

Although the PerRequestLifetimeManager class works correctly and can help

you to work with stateful or thread-unsafe dependencies within the scope of
an HTTP request, it is generally not a good idea to use it if you can avoid it.
Using this lifetime manager can lead to bad practices or hard to find bugs in the
end users application code when used incorrectly. The dependencies you register with the Unity container should be stateless, and if you have a requirement
to share common state between several objects during the lifetime of an HTTP
request, then you can have a stateless service that explicitly stores and retrieves
this state using the System.Web.HttpContext.Items collection of the System.
Web.HttpContext.Current object.

Per Thread Lifetime Management

The final lifetime manger included with Unity enables you to resolve instances
on a per thread basis. All calls to the Resolve method from the same thread return the same instance.
For more information about Unitys lifetime managers, see the topic Understanding Lifetime Managers.

Dependency Injection and Unit Testing

In Chapter 1, one of the motivations for adopting a loosely coupled design was
that it facilitates unit testing. In the example used in this chapter, one of the
types registered with the container is the TenantStore class. The following code
sample shows an outline of this class.
public class TenantStore : ItenantStore
{
...
public TenantStore(IBlobContainer<Tenant> tenantBlobContainer,
IBlobContainer<byte[]> logosBlobContainer)
{
...
}
public Tenant GetTenant(string tenant)
{
...
}
public IEnumerable<string> GetTenantNames()
{
...
}
public void SaveTenant(Tenant tenant)
{
...
}

Dependency Injection with Unit y

57

public void UploadLogo(string tenant, byte[] logo)

{
...
}
}

This class has dependencies on the IBlobContainer<Tenant> and IBlobContainer<byte[]> types which the
container resolves when it instantiates a TenantStore object. However, to test this class in a unit test, you dont
want to have to create these blob containers: now its easy to replace them with mocks for the purpose of the
tests. The following code sample shows some example tests.
IUnityContainer container = new UnityContainer();
[TestMethod]
public void GetTenantReturnsTenantFromBlobStorage()
{
var mockTenantBlobContainer = new Mock<IBlobContainer<Tenant>>();
var store = new TenantStore(mockTenantBlobContainer.Object, null);
var tenant = new Tenant();
mockTenantBlobContainer.Setup(c => c.Get("tenant")).Returns(tenant);
var actualTenant = store.GetTenant("tenant");
Assert.AreSame(tenant, actualTenant);
}
[TestMethod]
public void UploadLogoGetsTenantToUpdateFromContainer()
{
var mockLogosBlobContainer = new Mock<IBlobContainer<byte[]>>();
var mockTenantContainer = new Mock<IBlobContainer<Tenant>>();
var store = new TenantStore(
mockTenantContainer.Object, mockLogosBlobContainer.Object);
mockTenantContainer.Setup(c => c.Get("tenant"))
.Returns(new Tenant() { Name = "tenant" }).Verifiable();
mockLogosBlobContainer.Setup(
c => c.GetUri(It.IsAny<string>())).Returns(new Uri("http://bloburi"));
store.UploadLogo("tenant", new byte[1]);
mockTenantContainer.Verify();
}

These two example tests provide mock objects that implement the IBlobContainer<Tenant> and IBlobContainer<byte[]> interfaces when they create the TenantStore instances to test.
These examples use the Moq mocking library to create the mock objects. For more information,
see http://code.google.com/p/moq/. Moq is also available as a NuGet package.

58

ch a pter thr ee

Summary
In this chapter, you saw how to use the Unity container to add support for dependency injection to a real-world
application and how you can use a Unity container to register types, resolve types at runtime, and manage the
lifetime of the resolved objects. In addition to seeing how the Unity container made it possible to build the
applications object graph at startup, you also saw how this approach facilitated designing and running unit
tests.

More Information
All links in this book are accessible from the books online bibliography available at: http://aka.ms/unitybiblio

Interception

Introduction
In Chapter 1, you learned about some of the motivations for adopting a loosely coupled design, and in the
following chapters, you saw how dependency injection (DI) and the Unity container can help you to realize
these benefits in your own applications. One of the motivations that Chapter 1 described for adopting a
loosely coupled design was crosscutting concerns. Although DI enables you to inject dependencies when you
instantiate the objects that your application will use at run time and helps you to ensure that the objects that
Unity instantiates on your behalf address your crosscutting concerns, you will still want to try and follow the
single responsibility and open/closed principles in your design. For example, the classes that implement your
business behaviors should not also be responsible for logging or validation, and you should not need to modify
your existing classes when you add support for a new crosscutting concern to your application. Interception
will help you to separate out the logic that handles crosscutting concerns from your LOB classes.

Crosscutting Concerns
Crosscutting concerns are concerns that affect many areas of your application. For example, in a LOB application, you may have a requirement to write information to a log file from many different areas in your application.
The term crosscutting concerns, refers to the fact that these types of concern typically cut across your application and do not align with the modules, inheritance hierarchies, and namespaces that help you to structure your
application in ways that align with your applications business domain. Common crosscutting concerns for LOB
applications include:
Logging. Writing diagnostic messages to a log for troubleshooting, tracing, or auditing purposes.
Validation. Checking that the input from users or other systems complies with a set of rules.
Exception handling. Using a common approach to exception handling in the application.
Transient fault handling. Using a common approach to identifying transient faults and retrying opera-

tions.

Authentication and authorization. Identifying the caller and determining if that caller should be allowed

to perform an operation.

Caching. Caching frequently used objects and resources to improve performance.

Performance monitoring. Collecting performance data in order to measure SLA compliance.
Encryption. Using a common service to encrypt and decrypt messages within the application.
Mapping. Providing a mapping or translation service for data as it moves between classes or components.
Compression. Providing a service to compress data as it moves between classes or components.

59

60

ch a pter four

Its possible that many different classes and components within your application
will need to implement some of these behaviors (Unity often uses the term
behaviors to refer to the logic that implements crosscutting concerns in your
code). However, implementing support for these crosscutting concerns in a LOB
application introduces a number of challenges such as how you can:
Maintain consistency. You want to be sure that all the classes and compo-

nents that need to implement one of these behaviors do so in a consistent

way. Also, if you need to modify the way that your application supports
one of these crosscutting concerns, then you want to be sure that the
change is applied everywhere.
Create maintainable code. The single responsibility principle helps to make
your code more maintainable. A class that implements a piece of business
functionality should not also be responsible for implementing a crosscutting concern such as logging.
Avoid duplicate code. You dont want to have the same code duplicated in
multiple locations within your application.
Before examining how interception and Unitys implementation of interception
can help you to address crosscutting concerns in your applications, its worth
examining some alternative approaches. The decorator pattern offers an approach that doesnt require a container and that you can implement yourself
without any dependencies on special frameworks or class libraries. Aspect oriented programming (AOP) is another approach that adopts a different paradigm
for addressing crosscutting concerns.

The Decorator Pattern

A common approach to implementing behaviors to address crosscutting concerns in your application is to use the decorator pattern. The basis of the decorator pattern is to create wrappers that handle the crosscutting concerns around
your existing objects. In previous chapters, you saw the TenantStore class that
is responsible for retrieving tenant information from, and saving tenant information to a data store. You will now see how you can use the decorator pattern to
add logic to handle crosscutting concerns such as logging and caching without
modifying or extending the existing TenantStore class.
If you extended the
TenantStore class to
add support for logging
you would be adding an
additional responsibility to
an existing class, breaking
the single responsibility
principle.

The following code sample shows the existing TenantStore class and ITenantStore interface.
public interface ITenantStore
{
void Initialize();
Tenant GetTenant(string tenant);
IEnumerable<string> GetTenantNames();
void SaveTenant(Tenant tenant);
void UploadLogo(string tenant, byte[] logo);
}
public class TenantStore : ITenantStore

Interception

61

{
...
public TenantStore(IBlobContainer<Tenant> tenantBlobContainer,
IBlobContainer<byte[]> logosBlobContainer)
{
...
}
...
}

The following code sample shows a decorator class that adds logging functionality to the existing TenantStore
class:
class LoggingTenantStore : ITenantStore
{
private readonly ITenantStore tenantStore;
private readonly ILogger logger;
public LoggingTenantStore(ITenantStore tenantstore, ILogger logger)
{
this.tenantStore = tenantstore;
this.logger = logger;
}
public void Initialize()
{
tenantStore.Initialize();
}
public Tenant GetTenant(string tenant)
{
return tenantStore.GetTenant(tenant);
}
public IEnumerable<string> GetTenantNames()
{
return tenantStore.GetTenantNames();
}
public void SaveTenant(Tenant tenant)
{
tenantStore.SaveTenant(tenant);
logger.WriteLogMessage("Saved tenant" + tenant.Name);
}
public void UploadLogo(string tenant, byte[] logo)
{
tenantStore.UploadLogo(tenant, logo);
logger.WriteLogMessage("Uploaded logo for " + tenant);
}
}

62

ch a pter four

Note how this decorator class also implements the ITenantStore interface and accepts an ITenantStore instance as a parameter to the constructor. In each method body, it invokes the original method before it performs
any necessary work related to logging. You could also reverse this order and perform the work that relates to
the crosscutting concern before you invoke the original method. You could also perform work related to the
crosscutting concern both before and after invoking the original method.
If you had another decorator class called CachingTenantStore that added caching behavior you could create
an ITenantStore instance that also handles logging and caching using the following code.
var basicTenantStore = new TenantStore(tenantContainer, blobContainer);
var loggingTenantStore = new LoggingTenantStore(basicTenantStore, logger);
var cachingAndLoggingTenantStore = new CachingTenantStore(loggingTenantStore, cache);

If you invoke the UploadLogo method on the cachingAndLoggingTenantStore object, then you will first invoke the UploadLogo method in the CachingTenantStore class that will in turn invoke the UploadLogo
method in the LoggingTenantStore class, which will in turn invoke the original UploadLogo method in the
TenantStore class before returning through the sequence of decorators.
Figure 1 shows the relationships between the various objects at run time after you have instantiated the
classes. You can see how each UploadLogo method performs its crosscutting concern functionality before it
invokes the next decorator in the chain, until it gets to the end of the chain and invokes the original UploadLogo method on the TenantStore instance.

LoggingTenantStore : ITenantStore

ITenantStore tenantStore

ITenantStore tenantStore

UploadLogo
-Perform logging

UploadLogo
-Perform caching

Figure 1
The decorator pattern at run time

UploadLogo

CachingTenantStore : ITenantStore

UploadLogo

Client object

TenantStore :
ITenantStore

Interception

63

Using Unity to Wire Up Decorator Chains

Instead of wiring the decorators up manually, you could use the Unity container to do it for you. The following
set of registrations will result in the same chain of decorators shown in figure 1 when you resolve the default
ITenantStore type.
container.RegisterType<ILogger, Logger>();
container.RegisterType<ICacheManager, SimpleCache>();
container.RegisterType<ITenantStore, TenantStore>("BasicStore");
container.RegisterType<ITenantStore, LoggingTenantStore>(
"LoggingStore",
new InjectionConstructor(
new ResolvedParameter<ITenantStore>("BasicStore"),
new ResolvedParameter<ILogger>()));
// Default registration
container.RegisterType<ITenantStore, CachingTenantStore>(
new InjectionConstructor(
new ResolvedParameter<ITenantStore>("LoggingStore"),
new ResolvedParameter<ICacheManager>()));

Aspect Oriented Programming

Aspect Oriented Programming (AOP) is closely related to the issue of handling crosscutting concerns in your
application. In AOP, you have some classes in your application that handle the core business logic and other
classes that handle aspects or crosscutting concerns. In the example shown in the previous section, the TenantStore class is responsible for some of the business logic in your LOB application, and the classes that implement
the ILogger and ICacheManager interfaces are responsible for handling the aspects (or crosscutting concerns)
in the application.
The previous example shows how you can use the decorator pattern to explicitly wire-up the classes responsible for the crosscutting concerns. While this approach works, it does require you write the decorator classes
(CachingTenantStore and LoggingTenantStore) in addition to wiring everything together, either explicitly or
using DI.
AOP is a mechanism that is intended to simplify how you wire-up the classes that are responsible for the
crosscutting concerns to the business classes: there should be no need to write the decorator classes and you
should be able to easily attach the aspects that handle the crosscutting concerns to your standard classes. The
way that AOP frameworks are implemented is often technology dependent because it requires a dynamic
mechanism to wire-up the aspect classes to the business classes after they have all been compiled.
For an example of an AOP framework for C#, see What is PostSharp? AspectJ is a widely used AOP framework
for Java.

64

ch a pter four

Interception
Interception is one approach to implementing the dynamic wire up that is necessary for AOP. You can use interception as a mechanism in its own right to insert
code dynamically without necessarily adopting AOP, but interception is often
used as the underlying mechanism in AOP approaches.

The interception approach

that Unity adopts is not,
strictly speaking, an AOP
approach, although it does
have many similarities with
true AOP approaches. For
example, Unity interception
only supports preprocessing
and post-processing
behaviors around method
calls and does not insert
code into the methods of
the target object. Also,
Unity interception does not
support interception for
class constructors.

Interception works by dynamically inserting code (typically code that is responsible for crosscutting concerns) between the calling code and the target object.
You can insert code before a method call so that it has access to the parameters
being passed, or afterwards so that it has access to the return value or unhandled
exceptions. This inserted code typically implements what are known as behaviors (behaviors typically implement support for crosscutting concerns), and you
can insert multiple behaviors into a pipeline between the client object and the
target object in a similar way to using a chain of decorators in the decorator
pattern to add support for multiple crosscutting concerns. The key difference
between interception and the decorator pattern is that the interception framework dynamically creates decorator classes such as LoggingTenantStore and
CachingTenantStore at run time. This makes it much easier to add behaviors
that provide support for crosscutting concerns or aspects because you no longer
need to manually create decorator classes for every business class that needs to
support the behaviors. Now, you can use a configuration mechanism to associate the classes that implement the behaviors with the business classes that need
the behaviors.
For a discussion of AOP and interception with Unity, see the article AspectOriented Programming, Interception and Unity 2.0 by Dino Esposito on MSDN.
There are many ways that you can implement interception, but the two approaches that Unity supports are known as instance interception and type interception. The next two sections describe these two different approaches.

Instance Interception

With instance interception, Unity dynamically creates a proxy object that it inserts between the client and the target object. The proxy object is responsible
for passing the calls made by the client to the target object through the behaviors that are responsible for the crosscutting concerns.
Figure 2 shows how when you use instance interception, the Unity container
instantiates the target TenantStore object, instantiates the pipeline of behaviors that implement the crosscutting concerns, and generates and instantiates a
TenantStore proxy object.

Interception

Resolve(ITenantStore)

65

Unity
container
Instantiate

Instantiate

Generate and instantiate

UploadLogo
Client object

Behavior pipeline
TenantStore
proxy object

Logging
Interception
Behavior

Caching
Interception
Behavior

TenantStore
object

Figure 2
An example of instance interception

Instance interception is the more commonly used of the two interception techniques supported by Unity and
you can use it if the target object either implements the MarshalByRefObject abstract class or implements a
public interface that defines the methods that you need to intercept.
You can use Unity instance interception to intercept objects created both by the Unity container and outside
of the container, and you can use instance interception to intercept both virtual and non-virtual methods.
However, you cannot cast the dynamically created proxy type to the type of the target object.

Type Interception

With type interception, Unity dynamically creates a new type that derives from the type of the target object
and that includes the behaviors that handle the crosscutting concerns. The Unity container instantiates objects
of the derived type at run time.
Figure 3 shows how with type interception, the Unity container generates and instantiates an object of a type
derived from the target TenantStore type that encapsulates the behavior pipeline in the overridden method calls.
Resolve(ITenantStore)

Unity
container

Generate and instantiate

Derived TenantStore object

Behavior pipeline
Client object

UploadLogo

Figure 3
An example of type interception

Logging
Interception
Behavior

Caching
Interception
Behavior

66

ch a pter four

You can only use Unity type interception to intercept virtual methods. However, because the new type derives
from the original target type, you can use it wherever you used the original target type.
For more information about the differences between these approaches, see Unity Interception Techniques.

Summary
In this chapter, you learned how interception enables you to add support for crosscutting concerns to your
application by intercepting calls to your business objects at run time. Interception uses dynamically created
classes to implement the decorator pattern at run time. The next chapter describes in detail how you can implement interception using the Unity container.

More Information
All links in this book are accessible from the books online bibliography available at: http://aka.ms/unitybiblio

Interception using Unity

Introduction
Chapter 4 describes interception as a technique that you can use to dynamically insert code that provides support for crosscutting concerns into your application without explicitly using the decorator pattern in your code. In this
chapter, youll learn how you can use Unity to implement interception, and the
various types of interception that Unity supports. The chapter starts by describing a common scenario for using interception and illustrates how you can implement it using Unity interception. The chapter then explores a number of alternative approaches that you could adopt, including the use of policy injection and
attributes.
You cannot use Unity Interception in Windows Store Applications. For more
information, see Appendix A Unity and Windows Store apps.

Crosscutting Concerns and Enterprise Library

This chapter begins by describing how you can use Unity interception to implement your own custom code to add support for crosscutting concerns into your
application. However, the blocks in Enterprise Library provide rich support for
crosscutting concerns, and therefore its not surprising that instead of your own
custom code you can use the Enterprise Library blocks with Unity interception;
the chapter also describes this approach.

Interceptors in Unity
This section describes the basic steps that youll need to complete in order to
use interception in your application. To illustrate some of the advantages of the
interception approach, this chapter uses the same example crosscutting concerns, logging and caching, that the discussion of the decorator pattern in
Chapter 4 uses. You may find it useful to refer back to Chapter 4, Interception,
for some of the details that relate to the logging and caching functionality.

You can use policy injection

and call handlers to
integrate the functionality
provided by the Enterprise
Library blocks with Unity
interception.

This section uses logging as an example of a crosscutting concern and shows

how you can implement it using interception. Later in this chapter, you ll see
how you can use the Enterprise Library logging block in place of your own
implementation.

67

68

ch a pter fi v e

Configuring the Unity Container to Support Interception

By default, a Unity container does not support interception; to add support for
interception you must add the Unity interception extension as shown in the
following code sample. You can add the Unity Interception Extension libraries
to your project using NuGet.
using Microsoft.Practices.Unity.InterceptionExtension;
...
IUnityContainer container = new UnityContainer();
container.AddNewExtension<Interception>();

If you dont need to use

Unity interception, you dont
need to install the libraries
through NuGet: interception
is an extension to the core
Unity installation.

For more information about how to add the interception extension to a Unity
container instance and about how you can use a configuration file instead of
code, see the topic Configuring a Container for Interception.

Defining an Interceptor
In Chapter 4, the two example crosscutting concerns in the decorator example
were logging and caching. The example showed implementations of these behaviors in classes that implemented the ILogger and ICacheManager interfaces.
To implement these behaviors as interceptors, you need classes that implement
the IInterceptionBehavior interface. The following example shows how you
can implement the logging behavior.
using Microsoft.Practices.Unity.InterceptionExtension;
class LoggingInterceptionBehavior : IInterceptionBehavior
{
public IMethodReturn Invoke(IMethodInvocation input,
GetNextInterceptionBehaviorDelegate getNext)
{
// Before invoking the method on the original target.
WriteLog(String.Format(
"Invoking method {0} at {1}",
input.MethodBase, DateTime.Now.ToLongTimeString()));
// Invoke the next behavior in the chain.
var result = getNext()(input, getNext);
// After invoking the method on the original target.
if (result.Exception != null)
{
WriteLog(String.Format(
"Method {0} threw exception {1} at {2}",
input.MethodBase, result.Exception.Message,
DateTime.Now.ToLongTimeString()));
}

Interception using Unit y

69

else
{
WriteLog(String.Format(
"Method {0} returned {1} at {2}",
input.MethodBase, result.ReturnValue,
DateTime.Now.ToLongTimeString()));
}
return result;
}
public IEnumerable<Type> GetRequiredInterfaces()
{
return Type.EmptyTypes;
}
public bool WillExecute
{
get { return true; }
}
private void WriteLog(string message)
{
...
}
}

The IInterceptionBehavior interface defines three methods: WillExecute, GetRequiredInterfaces, and Invoke. In many scenarios, you can use the default implementations of the WillExecute and GetRequiredInterfaces methods shown
in this example. The WillExecute method enables you to optimize your chain of
behaviors by specifying whether this behavior should execute; in this example
the method always returns true so the behavior always executes. The GetRequiredInterfaces method enables you to specify the interface types that you
want to associate with the behavior. In this example, the interceptor registration
will specify the interface type, and therefore the GetRequiredInterfaces method returns Type.EmptyTypes. For more information about how to use these two
methods, see the topic Behaviors for Interception.
The Invoke method takes two parameters: input contains information about
the call from the client that includes the method name and parameter values,
getNext is a delegate that enables you to call the next behavior in the pipeline,
or the target object if this is the last behavior in the pipeline. In this example,
you can see how the behavior captures the name of the method invoked by the
client and records details of the method invocation in the log before it invokes
the next behavior in the pipeline, or target object itself. This behavior then examines the result of the call to the next behavior in the pipeline and writes a log
message if an exception was thrown in the call, or details of the result if the call
succeeded. Finally, it returns the result of the call back to the previous behavior
in the pipeline (or the client object if this behavior was first in the pipeline).

In the Invoke method, you

can apply pre and post
processing to the call to the
target object by placing your
code before and after the
call to the getNext delegate.

70

ch a pter fi v e

The following code example shows a slightly more complex behavior that provides support for caching.
using Microsoft.Practices.Unity.InterceptionExtension;
class CachingInterceptionBehavior : IInterceptionBehavior
{
public IMethodReturn Invoke(IMethodInvocation input,
GetNextInterceptionBehaviorDelegate getNext)
{
// Before invoking the method on the original target.
if (input.MethodBase.Name == "GetTenant")
{
var tenantName = input.Arguments["tenant"].ToString();
if (IsInCache(tenantName))
{
return input.CreateMethodReturn(
FetchFromCache(tenantName));
}
}
IMethodReturn result = getNext()(input, getNext);
// After invoking the method on the original target.
if (input.MethodBase.Name == "SaveTenant")
{
AddToCache(input.Arguments["tenant"]);
}
return result;
}
public IEnumerable<Type> GetRequiredInterfaces()
{
return Type.EmptyTypes;
}
public bool WillExecute
{
get { return true; }
}
private bool IsInCache(string key) {...}
private object FetchFromCache(string key) {...}
private void AddToCache(object item) {...}
}

Interception using Unit y

71

This behavior filters for calls to a method called GetTenant and then attempts
to retrieve the named tenant from the cache. If it finds the tenant in the cache,
it does not need to invoke the target object to get the tenant, and instead uses
the CreateMethodReturn to return the tenant from the cache to the previous
behavior in the pipeline.
The behavior also filters for a method called SaveTenant after it has invoked the
method on the next behavior in the pipeline (or the target object) and adds to
the cache a copy of the tenant object saved by the target object.
This example embeds a filter that determines when the interception behavior
should be applied. Later in this chapter, you ll see how you replace this with
a policy that you can define in a configuration file or with attributes in your
code.

Behaviors do not always

have to pass on a call to the
next behavior in the pipeline.
They can generate the return
value themselves or throw
an exception.

Registering an Interceptor
Although it is possible to use behaviors such as the CachingInterceptionBehavior and LoggingInterceptionBehavior without the Unity container, the
following code sample shows how you can use the container to register the
interception behaviors to the TenantStore type so that the container sets up
the interception whenever it resolves the type.
using Microsoft.Practices.Unity.InterceptionExtension;
...
container.AddNewExtension<Interception>();
container.RegisterType<ITenantStore, TenantStore>(
new Interceptor<InterfaceInterceptor>(),
new InterceptionBehavior<LoggingInterceptionBehavior>(),
new InterceptionBehavior<CachingInterceptionBehavior>());

The first parameter to the RegisterType method, an Interceptor<InterfaceInterceptor> object, specifies the type of interception to use. In this example, the
next two parameters register the two interception behaviors with the TenantStore type.
The InterfaceInterceptor type defines interception based on a proxy object.
You can also use the TransparentProxyInterceptor and VirtualMethodInterceptor types that are described later in this chapter.
The order of the interception behavior parameters determines the order of
these behaviors in the pipeline. In this example, the order is important because
the caching interception behavior does not pass on the request from the client
to the next behavior if it finds the item in the cache. If you reversed the order
of these two interception behaviors, you wouldnt get any log messages if the
requested item was found in the cache.
Placing all your registration
code in one place means
that you can manage all the
interception behaviors in
your application from one
location.

72

ch a pter fi v e

Using an Interceptor
The final step is to use the interceptor at run time. The following code sample shows how you can resolve and
use a TenantStore object that has the logging and caching behaviors attached to it.
var tenantStore = container.Resolve<ITenantStore>();
tenantStore.SaveTenant(tenant);

The type of the tenantStore variable is not TenantStore, it is a new dynamically created proxy type that implements the ITenantStore interface. This proxy type includes the methods, properties, and events defined in the
ITenantStore interface.
Figure 1 illustrates the scenario implemented by the two behaviors and type registration youve seen in the
previous code samples.
Resolve(ITenantStore)

Unity
container

1
2

Client
object

GetTenant(adatum)
9

3
TenantStore
proxy object
8

Behavior pipeline
Logging
Interception
Behavior

4
7

Caching
Interception
Behavior

5
6

TenantStore
object

Figure 1
The behavior pipeline

The numbers in the following list correspond to the numbers in Figure 1.

The interceptors get access

to the parameters passed to
the original call, the value
returned from the original
call, and any exceptions
thrown by the original call.

1. The client object calls the Resolve method on the Unity container to
obtain an object that implements the ITenantStore interface. The container uses the ITenantStore registration information to instantiate a
TenantStore object, create a pipeline with the interception behaviors,
and dynamically generate a proxy TenantStore object.
2. The client invokes the GetTenant method on the TenantStore proxy
object, passing a single string parameter identifying the tenant to fetch
from the store.
3. The TenantStore proxy object, calls the Invoke method in the first
interception behavior. This logging interception behavior logs details of
the call.
4. The first interception behavior, calls the Invoke method in the next
behavior in the pipeline. If the caching behavior finds the tenant in the
cache, jump to step 7.
5. The last interception behavior, calls the GetTenant method on the
TenantStore object.

Interception using Unit y

73

6. The TenantStore object returns a tenant to the last interception behavior in the pipeline.
7. If the caching interception behavior found the tenant in the cache it returns the cached tenant, otherwise it returns the tenant from the TenantStore object and caches it.
8. The logging interception behavior logs details of the result of the call that it made to the next behavior
in the pipeline and returns the result back to the TenantStore proxy object.
9. The TenantStore proxy object returns the tenant object to the client object.

Alternative Interception Techniques

The example shown earlier in this chapter illustrated the most common way to use interception in Unity. This
section discusses some variations on the basic approach and suggests some scenarios where you might wish to
use them.

Instance Interception/Type Interception

The approach used in the example earlier in this chapter is an example of instance interception where Unity
dynamically generates a proxy object and enables interception of methods defined in an interface, in this example the ITenantStore interface. As a reminder, here is the code that registers the interception.
container.RegisterType<ITenantStore, TenantStore>(
new Interceptor<InterfaceInterceptor>(),
new InterceptionBehavior<LoggingInterceptionBehavior>(),
new InterceptionBehavior<CachingInterceptionBehavior>());

In this example, the Interceptor<InterfaceInterceptor> parameter specifies you are using a type of instance
interception known as interface interception, and the InterceptionBehavior parameters define the two behaviors to insert into the behavior pipeline.
Unity interception includes two other interceptor types: TransparentProxyInterceptor and VirtualMethodInterceptor.
The following table summarizes the available interceptor types:
Interception Type

Interceptor classes

Instance

InterfaceInterceptor
TransparentProxyInterceptor

Type

VirtualMethodInterceptor

For more information about the different interceptor types, see the topic Unity Interception Techniques.
Using the TransparentProxyInterceptor Type
The InterfaceInterceptor type enables you to intercept methods on only one interface; in the example above,
you can intercept the methods on the ITenantStore interface. If the TenantStore class also implements an
interface such as the ITenantLogoStore interface, and you want to intercept methods defined on that interface
in addition to methods defined on the ITenantStore interface then you should use the TransparentProxyInterceptor type as shown in the following code sample.

74

ch a pter fi v e

container.RegisterType<ITenantStore, TenantStore>(
new Interceptor<TransparentProxyInterceptor>(),
new InterceptionBehavior<LoggingInterceptionBehavior>(),
new InterceptionBehavior<CachingInterceptionBehavior>());
var tenantStore = container.Resolve<ITenantStore>();
// From the ITenantStore interface.
tenantStore.SaveTenant(tenant);
// From the ITenantLogoStore interface.
((ITenantLogoStore)tenantStore).SaveLogo("adatum", logo);

The drawback to using the TransparentProxyInterceptor type in place of the

InterfaceInterceptor type is that it is significantly slower at run time. You can
also use this approach if the TenantStore class doesnt implement any interfaces but does extend the MarshalByRef abstract base class.
Using the VirtualMethodInterceptor Type

Although the TransparentProxyInterceptor type

appears to be more flexible
than the InterfaceInterceptor type, it is not
nearly as performant.

The third interceptor type is the VirtualMethodInterceptor type. If you use the
InterfaceInterceptor or TransparentProxyInterceptor type, then at run time
Unity dynamically creates a proxy object. However, the proxy object is not type
compatible with the target object. In the previous code sample, the tenantStore
object is not an instance of, or derived from, the TenantStore class.
In the example shown below, which uses the VirtualMethodInterceptor interception type, the type of the tenantStore object derives from the TenantStore
type so you can use it wherever you can use a TenantStore instance. However,
you cannot use the VirtualMethodInterceptor interceptor on existing objects,
you can only configure type interception when you are creating the target object. For example, you cannot use type interception on a WCF service proxy
object that is created for you by a channel factory. For the logging and caching
interception behaviors to work when you invoke the SaveTenant method in the
following example, the SaveTenant method in the TenantStore class must be
virtual, and the TenantStore class must be public.
container.RegisterType<ITenantStore, TenantStore>(
new Interceptor<VirtualMethodInterceptor>(),
new InterceptionBehavior<LoggingInterceptionBehavior>(),
new InterceptionBehavior<CachingInterceptionBehavior>());
var tenantStore = container.Resolve<ITenantStore>();
tenantStore.SaveTenant(tenant);

If you use virtual method interception, the container generates a new type
that directly extends the type of the target object. The container inserts the
behavior pipeline by overriding the virtual methods in the base type.
Calls made with the VirtualMethodInterceptor are
much faster than those made
with the TransparentProxyInterceptor.

Another advantage of virtual method interception is that you can intercept internal calls in the class that happen when one method in a class invokes another
method in the same class. This is not possible with instance interception.

Interception using Unit y

75

Figure 2 shows the objects that are involved with interception behaviors in this example.
Resolve(ITenantStore)

Unity
container

1
Derived TenantStore object
2
Client object

GetTenant(adatum)
9

Behavior pipeline
3
8

Logging
Interception
Behavior

4
7

Caching
Interception
Behavior

5
6

Figure 2
Interception using the virtual method interceptor

The numbers in the following list correspond to the numbers in Figure 2.

1. The client object calls the Resolve method on the Unity container to obtain an object that implements
the ITenantStore interface. The container uses the ITenantStore registration information to instantiate
an object from a class derived from the TenantStore type that includes a pipeline with the interception
behaviors.
2. The client invokes the GetTenant method on the TenantStore derived object, passing a single string
parameter identifying the tenant to fetch from the store.
3. The TenantStore derived object, calls the Invoke method in the first interception behavior. This logging
interception behavior logs details of the call.
4. The first interception behavior, calls the Invoke method in the next behavior in the pipeline. If the
caching behavior finds the tenant in the 5.
5. The last interception behavior, calls the GetTenant method in the TenantStore base class.
6. The TenantStore base class returns a tenant to the last interception behavior in the pipeline.
7. If the caching interception behavior found the tenant in the cache it returns the cached tenant, otherwise it returns the tenant from the TenantStore object.
8. The logging interception behavior logs details of the result of the call that it made to the next behavior in
the pipeline and returns the result back to the TenantStore derived object.
9. The TenantStore derived object returns the tenant object to the client object.
For more information about the different interceptor types, see Unity Interception Techniques.

76

ch a pter fi v e

Using a Behavior to Add an Interface to an Existing Class

Sometimes it is useful to be able to add an interface implementation to an existing class without changing that
class. For example, the following TenantStore class only implements the ITenantStore interface.
public class TenantStore : ITenantStore
{
...
}

However, you may have a requirement for TenantStore instances to implement some custom logging methods
such as WriteLogMessage defined in the ILogger interface. This requirement may be in addition to that of
using interception to write log messages when you invoke methods, such as GetTenant or SaveTenant, on the
TenantStore class. Using a behavior, you can make it possible to write code such as that shown in the following
example without modifying the original TenantStore class.
((ILogger)tenantStore).WriteLogMessage(Message: Write to the log directly...);

When you register the TenantStore type with the container, you can specify that it supports additional interfaces, such as the ILogger interface, as shown in the following example:
container.RegisterType<ITenantStore, TenantStore>(
new Interceptor<InterfaceInterceptor>(),
new InterceptionBehavior<LoggingInterceptionBehavior>(),
new InterceptionBehavior<CachingInterceptionBehavior>(),
new AdditionalInterface<ILogger>());

Then, in a behavior, you can intercept any calls made to methods defined on that interface. The following code
sample shows part of the LoggingInterceptionBehavior class that filters for calls to methods on the ILogger
interface and handles them.
class LoggingInterceptionBehavior : IInterceptionBehavior
{
public IMethodReturn Invoke(IMethodInvocation input,
GetNextInterceptionBehaviorDelegate getNext)
{
var methodReturn = CheckForILogger(input);
if (methodReturn != null)
{
return methodReturn;
}
...
}

Interception using Unit y

77

...
private IMethodReturn CheckForILogger(IMethodInvocation input)
{
if (input.MethodBase.DeclaringType == typeof(ILogger)
&& input.MethodBase.Name == "WriteLogMessage")
{
WriteLog(input.Arguments["message"].ToString());
return input.CreateMethodReturn(null);
}
return null;
}
}

Note how in this example the behavior intercepts the call to the WriteLogMessage interface and does not
forward it on to the target object. This is because the target object does not implement the ILogger interface
and does not have a WriteLogMessage method.

Interception Without the Unity Container

The examples youve seen so far show how to setup interception as part of the type registration in the Unity
container. If you are not using a Unity container for DI or want to use interception without using a container,
you can use the Intercept class. The following two examples show how you can setup interception using the
Unity container, and using the Intercept class. Both examples attach the same interception pipeline to a TenantStore object.
// Example 1. Using a container.
// Configure the container for interception.
container = new UnityContainer();
container.AddNewExtension<Interception>();
// Register the TenantStore type for interception.
container.RegisterType<ITenantStore, TenantStore>(
new Interceptor<InterfaceInterceptor>(),
new InterceptionBehavior<LoggingInterceptionBehavior>(),
new InterceptionBehavior<CachingInterceptionBehavior>());
// Obtain a proxy object with an interception pipeline.
var tenantStore = container.Resolve<ITenantStore>();

// Example 2. Using the Intercept class.

ITenantStore tenantStore = Intercept.ThroughProxy<ITenantStore>(
new TenantStore(tenantContainer, blobContainer),
new InterfaceInterceptor(),
new IInterceptionBehavior[] {
new LoggingInterceptionBehavior(), new CachingInterceptionBehavior()
});

78

ch a pter fi v e

It is sometimes convenient to use stand-alone interception and the Intercept

class if you want to attach an intercept pipeline to an existing object as shown
in the following example.
TenantStore tenantStore =
new TenantStore(tenantContainer, blobContainer);
...
// Attach an interception pipeline.
ITenantStore proxyTenantStore = Intercept.ThroughProxy<ITenantStore>(
tenantStore,
new InterfaceInterceptor(),
new IInterceptionBehavior [] {
new LoggingInterceptionBehavior(), new CachingInterceptionBehavior()
});

You can use the ThroughProxy methods of the Intercept class to set up instance interception that uses a proxy object , and the NewInstance methods to
set up type interception that uses a derived object. You can only attach an interception pipeline to an existing object if you use the ThroughProxy methods;
the NewInstance methods always create a new instance of the target object.
For more information, see Stand-alone Unity Interception.
You dont need to use a
Unity container if you want
to use interception in your
application.

Design Time Configuration

All of the examples so far in this chapter have configured interception programmatically, either in a Unity container or by using the Intercept class. However, in
the same way that you can define your type registrations for the container in a
configuration file, you can also define your interception behavior pipeline.
Youve seen the following registration code for the TenantStore class several
times already in this chapter.
// Configure the container for interception.
container = new UnityContainer();
container.AddNewExtension<Interception>();
// Register the TenantStore type for interception.
container.RegisterType<ITenantStore, TenantStore>(
new Interceptor<InterfaceInterceptor>(),
new InterceptionBehavior<LoggingInterceptionBehavior>(),
new InterceptionBehavior<CachingInterceptionBehavior>());
// Obtain a proxy object with an interception pipeline.
var tenantStore = container.Resolve<ITenantStore>();

Instead of this programmatic configuration, you can add the following configuration information to your configuration file.

Interception using Unit y

<unity xmlns="http://schemas.microsoft.com/practices/2010/unity">
<sectionExtension type=
"Microsoft.Practices.Unity.InterceptionExtension
.Configuration.InterceptionConfigurationExtension,
Microsoft.Practices.Unity.Interception.Configuration" />
<namespace name="Tailspin.Web.Survey.Shared.Stores" />
<namespace name="Tailspin.Utilities.InterceptionBehaviors" />
<assembly name="Tailspin.Web.Survey.Shared />
<assembly name="Tailspin.Utilities />
<container>
<register type="ITenantStore" mapTo="TenantStore">
<interceptor type="InterfaceInterceptor"/>
<interceptionBehavior type="LoggingInterceptionBehavior" />
<interceptionBehavior type="CachingInterceptionBehavior" />
</register>
</container>
</unity>

Note that this snippet from the configuration file includes a sectionExtension
element to enable you to use the interception specific elements. There are also
namespace and assembly elements to enable Unity to find your interception
behavior classes in addition to the TenantStore class and ITenantStore interface.
The following code sample shows how you can load this configuration information and then resolve a TenantStore instance with an interception pipeline that
includes the logging and caching behaviors.
IUnityContainer container = new UnityContainer();
container.LoadConfiguration();
// Obtain a proxy object with an interception pipeline.
var tenantStore = container.Resolve<ITenantStore>();
...
tenantStore.UploadLogo(tenant, logo);

Policy Injection
The following code sample shows how you insert two behaviors into the pipeline for a TenantStore instance created by the Unity container.
container.RegisterType<ITenantStore, TenantStore>(
new Interceptor<InterfaceInterceptor>(),
new InterceptionBehavior<LoggingInterceptionBehavior>(),
new InterceptionBehavior<CachingInterceptionBehavior>());

Defining the interception

behaviors in the
configuration file makes
it possible to change their
configuration without
recompiling the code.

79

80

ch a pter fi v e

One of the drawbacks of this approach to implementing support for crosscutting

concerns is that you must configure the interceptors for every class that needs
them. In practice, you may have additional store classes in your application that
all require caching support, and many more that require logging behavior.

You may be able to use

registration by convention
to configure interceptors for
multiple registered types.

The following code sample shows an alternative approach based on policies for
adding interception behaviors to objects in your application. Note that definitions of the LoggingCallHandler and CachingCallHandler classes are given
later in this section.
container.RegisterType<ITenantStore, TenantStore>(
new InterceptionBehavior<PolicyInjectionBehavior>(),
new Interceptor<InterfaceInterceptor>());
container.RegisterType<ISurveyStore, SurveyStore>(
new InterceptionBehavior<PolicyInjectionBehavior>(),
new Interceptor<InterfaceInterceptor>());
var first = new InjectionProperty("Order", 1);
var second = new InjectionProperty("Order", 2);
container.Configure<Interception>()
.AddPolicy("logging")
.AddMatchingRule<AssemblyMatchingRule>(
new InjectionConstructor(
new InjectionParameter("Tailspin.Web.Survey.Shared")))
.AddCallHandler<LoggingCallHandler>(
new ContainerControlledLifetimeManager(),
new InjectionConstructor(),
first);
container.Configure<Interception>()
.AddPolicy("caching")
.AddMatchingRule<MemberNameMatchingRule>(
new InjectionConstructor(new [] {"Get*", "Save*"}, true)))
.AddMatchingRule<NamespaceMatchingRule>(
new InjectionConstructor(
"Tailspin.Web.Survey.Shared.Stores", true)))
.AddCallHandler<CachingCallHandler>(
new ContainerControlledLifetimeManager(),
new InjectionConstructor(),
second);

You must remember to

configure policy injection
behavior for each type that
you want to use it on as well
as defining the policies.

This example registers two store types with the container using a PolicyInjectionBehavior behavior type. This behavior makes use of policy definitions to insert
handlers into a pipeline when a client calls an object instantiated by the container.
The example shows two policy definitions: one for the LoggingCallHandler interception handler and one for the CachingCallHandler interception handler. Each
policy has a name to identify it and one or more matching rules to determine when
to apply it.

Interception using Unit y

The Policy Injection Application Block has built-in matching rules based on the
following:
Assembly name
Namespace
Type
Tag attribute
Custom attribute
Member name
Method signature
Parameter type
Property
Return type

For more information about these matching rules and how to use them, see the
topic Policy Injection Matching Rules.
You can also define your own, custom matching rule types. For more
information, see the topic Creating Policy Injection Matching Rules.
The policy for logging has a single matching rule based on the name of the assembly that contains the class definition of the target object: in this example, if
the TenantStore and SurveyStore classes are located in the Tailspin.Web.
Survey.Shared assembly, then the handler will log all method calls to those
objects.
The policy for caching uses two matching rules that are ANDed together: in this
example, the caching handler handles all calls to methods that start either with
Get or Save (the true parameter makes it a case sensitive test), and that are in
the Tailspin.Web.Survey.Shared.Stores namespace.
The first and second parameters control the order that the container invokes
the handlers if the matching rules match multiple handlers to a single instance.
In this example, you want to be sure that the container invokes the logging
handler before the caching handler in order to ensure that all calls are logged.
Remember, in the example used in this chapter, the caching handler does not
pass on the call if it locates the item in the cache.
Both policies use the ContainerControlledLifetimeManager type to ensure
that the handlers are singleton objects in the container.
The handler classes are very similar to the behavior interception classes you saw
earlier in this chapter. The following code samples shows the logging handler
and the cache handler used in the examples shown in this section.
class LoggingCallHandler : ICallHandler
{
public IMethodReturn Invoke(IMethodInvocation input,
GetNextHandlerDelegate getNext)
{
// Before invoking the method on the original target
WriteLog(String.Format("Invoking method {0} at {1}",
input.MethodBase, DateTime.Now.ToLongTimeString()));

You can use the policy

injection matching rules
to avoid the need embed
filters in an injection
behavior class.

81

82

ch a pter fi v e

// Invoke the next handler in the chain

var result = getNext().Invoke(input, getNext);
// After invoking the method on the original target
if (result.Exception != null)
{
WriteLog(String.Format("Method {0} threw exception {1} at {2}",
input.MethodBase, result.Exception.Message,
DateTime.Now.ToLongTimeString()));
}
else
{
WriteLog(String.Format("Method {0} returned {1} at {2}",
input.MethodBase, result.ReturnValue,
DateTime.Now.ToLongTimeString()));
}
return result;
}
public int Order
{
get;
set;
}
private void WriteLog(string message)
{
...
}

public class CachingCallHandler : ICallHandler

{
public IMethodReturn Invoke(IMethodInvocation input,
GetNextHandlerDelegate getNext)
{
//Before invoking the method on the original target
if (input.MethodBase.Name == "GetTenant")
{
var tenantName = input.Arguments["tenant"].ToString();
if (IsInCache(tenantName))
{
return input.CreateMethodReturn(FetchFromCache(tenantName));
}
}
IMethodReturn result = getNext()(input, getNext);

Interception using Unit y

//After invoking the method on the original target

if (input.MethodBase.Name == "SaveTenant")
{
AddToCache(input.Arguments["tenant"]);
}
return result;
}
public int Order
{
get;
set;
}
private bool IsInCache(string key)
{
...
}
private object FetchFromCache(string key)
{
...
}
private void AddToCache(object item)
{
...
}
}

In many cases, policy injection handler classes may be simpler than interception
behavior classes that address the same crosscutting concerns. This is because
you can use the policy matching rules to control when the container invokes the
handler classes whereas very often with interception behavior classes, such as
the CachingInterceptionBehavior class shown earlier in the chapter, you need
to implement a filter within Invoke method to determine whether the behavior
should execute in a particular circumstance.
The Order property, which is set when you configure the policy, controls the
order in which the container executes the handlers when a policy rule matches
more than one handler.

Policy Injection and Attributes

Using policies with matching rules such as those shown in the previous example
means that you can apply and manage the policies when you configure your
container. You can define the policies either in code, as shown in the example,
or in a configuration file in a similar way to that described in the section Design
Time Configuration earlier in this chapter.

You may need to consider

the order of the call
handlers carefully,
especially if some of them
dont always pass the call
on to the next handler.

83

84

ch a pter fi v e

However, an alternative approach is to use attributes in your business classes to

indicate whether the container should invoke a particular call handler. In this
case, the person writing the business class is responsible for making sure that all
the crosscutting concerns are addressed.
In this scenario, you can register the types that the container should inject with
policies as shown in the following code sample.
Typically, if you define
policies for your call
handlers, they are all
defined in a single class or
configuration file where
they are easy to manage.
Using attributes means that
information about which
crosscutting concerns are
associated with particular
classes and methods is
stored in those classes
which is a less maintainable
approach.

container.RegisterType<ITenantStore, TenantStore>(
new InterceptionBehavior<PolicyInjectionBehavior>(),
new Interceptor<InterfaceInterceptor>());
container.RegisterType<ISurveyStore, SurveyStore>(
new InterceptionBehavior<PolicyInjectionBehavior>(),
new Interceptor<InterfaceInterceptor>());

There is no need to configure any policies, but you do need to define your attributes. The following code sample shows the LoggingCallHandlerAttribute
class. The CachingCallHandlerAttribute class is almost identical.
using Microsoft.Practices.Unity;
using Microsoft.Practices.Unity.InterceptionExtension;
class LoggingCallHandlerAttribute : HandlerAttribute
{
private readonly int order;
public LoggingCallHandlerAttribute(int order)
{
this.order = order;
}
public override ICallHandler CreateHandler(IUnityContainer container)
{
return new LoggingCallHandler() { Order = order };
}
}

The CreateHandler method creates an instance of the LoggingCallHandler

class that you saw previously and sets the value of the Order property.
You can now use the attributes to decorate your business classes as shown in
the following example.
class MyTenantStoreWithAttributes : ITenantStore, ITenantLogoStore
{
...
[LoggingCallHandler(1)]
public void Initialize()
{
...
}

Interception using Unit y

85

[LoggingCallHandler(1)]
[CachingCallHandler(2)]
public Tenant GetTenant(string tenant)
{
...
}
[LoggingCallHandler(1)]
public IEnumerable<string> GetTenantNames()
{
...
}
[LoggingCallHandler(1)]
[CachingCallHandler(2)]
public void SaveTenant(Tenant tenant)
{
...
}
[LoggingCallHandler(1)]
public virtual void UploadLogo(string tenant, byte[] logo)
{
...
}
}

In this example, the LoggingCallHandler call handler logs all the method calls,
and the GetTenant and SaveTenant methods also have a caching behavior applied to them. The example uses the Order property of the handlers to place
the logging call handler before the caching call handler in the pipeline.

Policy Injection and the Enterprise Library Blocks

In the previous examples, you implemented the caching and logging behaviors
in your own call handler classes: CachingCallHandler and LoggingCallHandler.
The blocks in the Enterprise Library, such as the Logging Application Block and
the Validation Application Block, provide some pre-written call handlers that
you can use in your own applications. This enables you to use the Enterprise
Library blocks to address crosscutting concerns in your application using policy
injection.
For example, you could replace your LoggingCallHandler with LogCallHandler
from the Enterprise Library Policy Injection Application Block as shown in the
following code sample. However, Enterprise Library 6 does not include a caching
handler.

Using the Enterprise Library

blocks to address your crosscutting concerns is often
easier than implementing the
behaviors yourself.

86

ch a pter fi v e

ConfigureLogger();
container.RegisterType<ITenantStore, TenantStore>(
new InterceptionBehavior<PolicyInjectionBehavior>(),
new Interceptor<InterfaceInterceptor>());
var second = new InjectionProperty("Order", 2);
container.Configure<Interception>()
.AddPolicy("logging")
.AddMatchingRule<AssemblyMatchingRule>(
new InjectionConstructor(
new InjectionParameter("Tailspin.Web.Survey.Shared")))
.AddCallHandler<LogCallHandler>(
new ContainerControlledLifetimeManager(),
new InjectionConstructor(
9001, true, true,
"start", "finish", true, false, true, 10, 1));
container.Configure<Interception>()
.AddPolicy("caching")
.AddMatchingRule<MemberNameMatchingRule>(
new InjectionConstructor(new[] { "Get*", "Save*" }, true))
.AddMatchingRule<NamespaceMatchingRule>(
new InjectionConstructor(
"Tailspin.Web.Survey.Shared.Stores", true))
.AddCallHandler<CachingCallHandler>(
new ContainerControlledLifetimeManager(),
new InjectionConstructor(),
second);

The LogCallHandler class from the Logging Application Block is a call handler
to use with the policy injection framework.
The LogCallHandler constructor parameters enable you to configure the logging behavior and specify the order of the handler in relation to other call handlers. In this example, the container will invoke the LogCallHandler call handler
before the user defined CachingCallHandler call handler.
Its important that you
configure the Logging
Application Block before
you configure the policy
injection. The sample code
in the PIABSamples solution
that accompanies this
guide includes a method
ConfigureLogger to do this.

If you want to control the behavior of the call handlers by using attributes in
your business classes instead of through policies, you can enable this approach
as shown in the following code sample.
container.RegisterType<ITenantStore, TenantStore>(
new InterceptionBehavior<PolicyInjectionBehavior>(),
new Interceptor<TransparentProxyInterceptor>());

Interception using Unit y

87

You can then use one of the attributes, such as the LogCallHandler attribute, defined in the Enterprise Library
blocks as shown in the following example.
[LogCallHandler(AfterMessage = "After call",
BeforeMessage = "Before call",
Categories = new string[] { "General" },
EventId = 9002,
IncludeCallStack = false,
IncludeCallTime = true,
IncludeParameters = true,
LogAfterCall = true,
LogBeforeCall = true,
Priority = 10,
Severity = System.Diagnostics.TraceEventType.Information,
Order = 1)]
[CachingCallHandler(2)]
public Tenant GetTenant(string tenant)
{
...
}

You can customize the information contained in the log message using attribute parameters, and control the
order of the handlers in the pipeline using the Order parameter.
An alternative approach to configuring the Enterprise Library call handlers is through the PolicyInjection
static faade. The sample code in the PIABSamples solution that accompanies this guide includes an example
of this approach.

A Real World Example

This example is from the aExpense reference implementation that accompanies the Enterprise Library guidance.
To install the sample, download the Enterprise Library Reference Implementation. The aExpense application is an
ASP.NET application that makes extensive use of the Enterprise Library blocks, including Unity and Policy Injection. This example is taken from the aExpense implementation that use Enterprise Library 6.
The goal is to use the Semantic Logging Application Block to log calls to a specific method in the data access
layer of the application. The following code sample shows part of the ExpenseRepository class where the
SaveExpense method has a Tag attribute that identifies a policy.
public class ExpenseRepository : IExpenseRepository
{
...
[Tag("SaveExpensePolicyRule")]
public virtual void SaveExpense(Model.Expense expense)
{
...
}
...
}

88

ch a pter fi v e

Note that the ExpenseRepository model class implements the IExpenseRepository interface: this means that
you can use the VirtualMethodInterceptor interceptor type. The following code sample shows the registration
of the ExpenseRepository type that includes adding the PolicyInjectionBehavior.
container
.RegisterType<IExpenseRepository, ExpenseRepository>(
new Interceptor<VirtualMethodInterceptor>(),
new InterceptionBehavior<PolicyInjectionBehavior>());

The following code sample shows the SemanticLogCallHandler call handler class that performs the logging.
public class SemanticLogCallHandler : ICallHandler
{
public SemanticLogCallHandler(NameValueCollection attributes)
{
}
public IMethodReturn Invoke(IMethodInvocation input,
GetNextHandlerDelegate getNext)
{
if (getNext == null) throw new ArgumentNullException("getNext");
AExpenseEvents.Log.LogCallHandlerPreInvoke(
input.MethodBase.DeclaringType.FullName, input.MethodBase.Name);
var sw = Stopwatch.StartNew();
IMethodReturn result = getNext()(input, getNext);
AExpenseEvents.Log.LogCallHandlerPostInvoke(
input.MethodBase.DeclaringType.FullName, input.MethodBase.Name,
sw.ElapsedMilliseconds);
return result;
}
public int Order { get; set; }
}

This call handler uses the Semantic Logging Application Block to log events before and after the call, and to
record how long the call took to complete.
The application configures the Policy Injection Application Block declaratively. The following snippet from the
Web.config file shows how the Tag attribute is associated with the SemanticLogCallHandler type.

Interception using Unit y

89

<policyInjection>
<policies>
<add name="ExpenseTracingPolicy">
<matchingRules>
<add name="TagRule" match="SaveExpensePolicyRule" ignoreCase="false"
type="Microsoft.Practices.EnterpriseLibrary.PolicyInjection
.MatchingRules.TagAttributeMatchingRule, ..."/>
</matchingRules>
<handlers>
<add type="AExpense.Instrumentation.SemanticLogCallHandler, AExpense,
Version=1.0.0.0, Culture=neutral, PublicKeyToken=null"
name="SemanticLogging Call Handler" />
</handlers>
</add>
</policies>
</policyInjection>

Summary
In this chapter, you learned how to use Unity and the Unity container to add support for crosscutting concerns
in your application by using interception. Interception enables you to implement support for crosscutting
concerns while continuing to write code that follows the single responsibility principle and the open/closed
principle. The chapter also described how you can use the policy injection technique that allows you define
policies to manage how and where in your application you address the crosscutting concerns. As an alternative
to defining policies that are typically managed in a single class in your project or in a configuration file, you can
also use attributes that enable developers to decorate their code to specify how they want to address the
crosscutting concerns.

More Information
All links in this book are accessible from the books online bibliography available at: http://aka.ms/unitybiblio

Extending Unity

Introduction
Although Unity is very flexible, there are scenarios where you might need to extend Unity in order to meet
some specific requirements. Unity is extensible in many different ways, some of which you have already seen
in the previous chapters of this guide, other extensibility options are covered in this chapter.
In Chapter 5, Interception with Unity, you saw how you can customize and extend the ways that Unity supports interception by:
Creating custom behaviors. Interception in Unity uses interception behaviors in a pipeline to implement

your crosscutting concerns. You implement your custom behaviors by implementing the IInterceptionBehavior interface.
Creating custom policy injection matching rules. If you use policy injection to insert behaviors into the
interception pipeline, you can add your own custom matching rules by implementing the IMatchingRule
interface.
Creating custom policy injection call handlers. Again, if you use policy injection, you can create custom
handler classes that enable you to define custom behaviors in the policy injection pipeline. You do this by
implementing the ICallHandler interface.
Creating custom handler attributes. Associated with custom call handlers, you can also define custom
attributes to decorate methods in your business classes to control the way that policy injection works in
your application.
In this chapter, youll see how you can create custom lifetime managers and extend the Unity container. Chapter 3, Dependency Injection with Unity, describes the built-in lifetime managers such as the TransientLifetimeManager, the ContainerControlledLifetimeManager, and the HierarchicalLifetimeManager and how
they manage the lifetime of the objects instantiated by the Unity container. In this chapter, youll see an example of how Unity implements a simple caching lifetime manager as a guide to creating your own custom
lifetime managers.
By creating a Unity container extension, you can customize what Unity does when you register and resolve
types. In this chapter, youll see an example of how you can add functionality to the container that automatically wires up event handlers when you resolve objects from the container.

Creating Custom Lifetime Managers

Chapter 3, Dependency Injection with Unity, describes how you can use Unitys built-in lifetime managers to
manage the lifetime of the objects that Unity creates when you resolve types registered with the container. For
example, if you use the default TransientLifetimeManager lifetime manager, the container does not hold a
reference to the resolved object and such objects should not, in general, hold on to any resources that must be
disposed deterministically. If, on the other hand, you use the ContainerControlledLifetimeManager lifetime
manager, the container itself tracks the object and is responsible for determining when the object it creates
becomes eligible for garbage collection.
91

92

ch a pter si x

In addition to the seven built-in lifetime managers (ContainerControlledLifetimeManager, TransientLifetimeManager, PerResolveLifetimeManager, PerThreadLifetimeManager, ExternallyControlledLifetimeManager,
PerRequestLifetimeManager and HierarchicalLifetimeManager), you can also create your own custom lifetime managers by extending either the abstract LifetimeManager class or the abstract SynchronizedLifetimeManager class in the Microsoft.Practices.Unity namespace.
You should extend the LifetimeManager class when your custom lifetime manager does not need to concern
itself with concurrency issues: the built-in PerThreadLifetimeManager class extends the LifetimeManager
class because it will not encounter any concurrency issues when it stores a resolved object in thread local
storage. However, the built-in ContainerControlledLifetimeManager class could encounter concurrency issues when it stores a reference to a newly created object: therefore, it extends the SynchronizedLifetimeManager class.
You may find it useful to open the sample application, CustomLifetimeManagers, that accompanies this guide
in Visual Studio while you read this section.

Lifetime Managers and Resolved Objects

Before examining an example of a custom lifetime manager, you should understand the relationship between
lifetime managers and the objects resolved by the container. Given the following code sample, the container
creates an instance of the ContainerControlledLifetimeManager type during the call to the RegisterType
method.
IUnityContainer container = new UnityContainer();
container.RegisterType<Tenant>(new ContainerControlledLifetimeManager());

When you execute the following line of code, the container creates a new Tenant instance that is referenced
by the ContainerControlledLifetimeManager instance.
var tenant = container.Resolve<Tenant>();

If you call the Resolve method a second time, the container returns a reference to the existing Tenant instance
that is referenced by the ContainerControlledLifetimeManager instance.
Both the Tenant instance and the ContainerControlledLifetimeManager instance will remain in memory
until the container itself is garbage collected by the runtime.

Extending the SynchronizedLifetimeManager Type

The example in this section shows how to create a custom lifetime manager that uses a cache to store the
objects resolved by the container. It extends the SynchronizedLifetimeManager class because potentially two
(or more) clients could try to store an instance of a type in the cache at the same time.
The sample shows the basics of creating a custom lifetime manager. However, it does not show how to create a
fully featured lifetime manager. For example, the code shown here does not support registering generic types.
This example defines a very simple interface that abstracts the logic of dealing with the cache from the lifetime
manager. The following code sample shows this interface.
public interface IStorage
{
object GetObject(string key);
void StoreObject(string key, object value);
void RemoveObject(string key);
}

E xtending Unit y

93

The following code sample shows the implementation of the CachedLifetimeManager class that provides
overrides of three methods from the SynchronizedLifetimeManager class.
public class CachedLifetimeManager
: SynchronizedLifetimeManager, IDisposable
{
private IStorage storage;
private string key;
public CachedLifetimeManager(IStorage storage)
{
this.storage = storage;
this.key = Guid.NewGuid().ToString();
}
public string Key
{
get { return key; }
}
protected override object SynchronizedGetValue()
{
return storage.GetObject(key);
}
protected override void SynchronizedSetValue(object newValue)
{
storage.StoreObject(key, newValue);
}
public override void RemoveValue()
{
Dispose();
}
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
protected void Dispose(bool disposing)
{
if (disposing && storage != null)
{
storage.RemoveObject(key);
storage = null;
}
}
}

94

ch a pter si x

The SynchronizedGetValue method could return a null if the object does not exist in the cache; in this scenario, the container instantiates a new object and then calls SynchronizedSetValue to add it to the cache.
Nothing in Unity calls the RemoveValue method: it is shown here for completeness.
When Unity creates an instance of the SynchronizedLifetimeManager class, the constructor generates a
GUID to use as the cache key for whatever object this instance SynchronizedLifetimeManager class is responsible for managing.
This example also provides a simple implementation of the IDisposable interface so that when the Unity
container is disposed, the custom lifetime manager has the opportunity to perform any clean up; in this case,
removing the resolved object from the cache.
The following code sample shows how you can use this custom lifetime manager. The SimpleMemoryCache
class implements the IStorage interface.
class Program
{
static void Main(string[] args)
{
IUnityContainer container = new UnityContainer();
var cache = new SimpleMemoryCache();
container.RegisterType<Tenant>(new CachedLifetimeManager(cache));
var tenant = container.Resolve<Tenant>();
var tenant2 = container.Resolve<Tenant>();
}
}

The sequence diagram shown in Figure 1 summarizes how the container uses the custom lifetime manager.

E xtending Unit y

95

UnityContainer
RegisterType<Tenant>

Constructor

Resolve<Tenant>

Resolve for
the first time

CachedLifetimeManager

SynchronizedGetValue
null
Constructor
SynchronizedSetValue

Tenant

Tenant

Resolve<Tenant>
SynchronizedGetValue
Resolve on
subsequent times

Tenant

Tenant

Figure 1
Container Interactions with a Custom Lifetime Manager

The sequence diagram shows how the call to the RegisterType method results in a new CachedLifetimeManager instance. The first time you call the Resolve method, the CachedLifetimeManager instance does
not find an instance of the requested type in the cache, so the Unity container creates an instance of the
requested type and then calls SynchronizedSetValue to store the instance in the cache. Subsequent calls to
the Resolve method result in the object being returned from the cache.

Extending the LifetimeManager Type

Extending the LifetimeManager class is almost the same as extending the SynchronizedLifetimeManager
class. Instead of overriding the SynchronizedGetValue and SynchronizedSetValue methods, you override the
GetValue and SetValue methods.

96

ch a pter si x

Extending the Unity Container

Unity enables you to create container extensions that add functionality to a container. Unity uses this extension
mechanism to implement some of its own functionality such as interception: to start using interception, the
first thing you must do is to add the container extension as shown in the following code sample.
using Microsoft.Practices.Unity.InterceptionExtension;
...
IUnityContainer container = new UnityContainer();
container.AddNewExtension<Interception>();

This section will show you an example of a container extension that can automatically wire up event handlers
in objects that you resolve from the container. However, before seeing how to implement the container extension, you should understand what this extension does. This will make it easier for you to follow the details of
the extension implementation later in this chapter.

Without the Event Broker Container Extension

The example uses a pair of very simple classes to illustrate the functionality of the extension. The two classes
are the Publisher and Subscriber classes shown in the following code sample. These two classes use some
basic event declarations.
class Publisher
{
public event EventHandler RaiseCustomEvent;
public void DoSomething()
{
OnRaiseCustomEvent();
}
protected virtual void OnRaiseCustomEvent()
{
EventHandler handler = RaiseCustomEvent;
if (handler != null)
{
handler(this, EventArgs.Empty);
}
}
}

class Subscriber
{
private string id;
public Subscriber(string ID, Publisher pub)
{
id = ID;
pub.RaiseCustomEvent += HandleCustomEvent;
}

public void HandleCustomEvent(object sender, EventArgs e)

{
Console.WriteLine(
"Subscriber {0} received this message at: {1}", id, DateTime.Now);
}

E xtending Unit y

97

If you dont have the event broker extension and you want to use Unity to instantiate a single Publisher instance wired to two Subscriber instances, then you have several options. For example, you could register and
resolve the types as shown in the following code sample:
IUnityContainer container = new UnityContainer();
container
.RegisterType<Publisher>(
new ContainerControlledLifetimeManager())
.RegisterType<Subscriber>(
new InjectionConstructor("default", typeof(Publisher)));
var pub = container.Resolve<Publisher>();
var sub1 = container.Resolve<Subscriber>(
new ParameterOverride("ID", "sub1"));
var sub2 = container.Resolve<Subscriber>(
new ParameterOverride("ID", "sub2"));
// Call the method that raises the event.
pub.DoSomething();

In this example, its important to use the ContainerControlledLifetimeManager lifetime manager when you
register the Publisher class: this ensures that the container creates a single Publisher instance that the Resolve
method returns and that the container passes to the Subscriber class constructor when resolving the Subscriber
type. Resolving the two Subscriber instances uses a parameter override to pass different IDs to the two instances so that you can identify the output from each instance.

With the Event Broker Extension

With the event broker extension, its possible to simplify the implementation of the Subscriber class as shown
in the following code sample.
You may find it useful to open the sample application, EventBroker, that accompanies this guide in Visual
Studio while you read this section.
class Publisher
{
[Publishes("CustomEvent")]
public event EventHandler RaiseCustomEvent;
public void DoSomething()
{
OnRaiseCustomEvent();
}
protected virtual void OnRaiseCustomEvent()
{
EventHandler handler = RaiseCustomEvent;
if (handler != null)
{
handler(this, EventArgs.Empty);
}
}
}

98

ch a pter si x

class Subscriber
{
private string id;
public Subscriber(string ID)
{
id = ID;
}
[SubscribesTo("CustomEvent")]
public void HandleCustomEvent(object sender, EventArgs e)
{
Console.WriteLine(
"Subscriber {0} received this message at: {1}", id, DateTime.Now);
}
}

In these new versions of the two classes, the event in the Publisher class is decorated with the Publishes attribute, and the handler method in the Subscriber
class is decorated with the SubscribesTo attribute. In addition, the constructor in
the Subscriber class is much simpler in that it no longer receives a reference to
the Publisher class and no longer hooks up the event handler.
The registration code is now also simpler.
You have effectively
decoupled the Subscriber
class from the Publisher
class. There is no longer a
reference to the Publisher
type anywhere in the
Subscriber class.

IUnityContainer container = new UnityContainer();

container
.AddNewExtension<SimpleEventBrokerExtension>()
.RegisterType<Publisher>()
.RegisterType<Subscriber>(new InjectionConstructor("default"));
var sub1 = container.Resolve<Subscriber>(
new ParameterOverride("ID", "sub1"));
var sub2 = container.Resolve<Subscriber>(
new ParameterOverride("ID", "sub2"));
// Call the method
pub.DoSomething();

The AddNewExtension method registers the container extension, which is

custom class that extends the abstract UnityContainerExtension class. Registering the Publisher type now uses the default TransientLifetimeManager
lifetime manager, and the Subscriber registration only needs to define how the
ID value is passed to the constructor.

Implementing the Simple Event Broker

Now that youve seen what the event broker extension does, using attributes to
define how the container should wire up event handlers and simplifying the
registration of the two classes, its time to see how to implement the extension.

E xtending Unit y

99

An EventBroker class tracks the event subscriber classes that subscribe to events in event publisher classes. The
following code sample shows an outline of this class.
public class EventBroker
{
...
public IEnumerable<string> RegisteredEvents
{
get { ... }
}
public void RegisterPublisher(string publishedEventName,
object publisher, string eventName)
{ ... }
public void UnregisterPublisher(string publishedEventName,
object publisher, string eventName)
{ ... }
public void RegisterSubscriber(string publishedEventName,
EventHandler subscriber)
{ ... }
public void UnregisterSubscriber(string publishedEventName,
EventHandler subscriber)
{ ... }
public IEnumerable<object> GetPublishersFor(string publishedEvent)
{ ... }
public IEnumerable<EventHandler> GetSubscribersFor(string publishedEvent)
{ ... }
private PublishedEvent GetEvent(string eventName)
{ ... }
}

The EventBroker class uses the PublishedEvent class shown in the following code sample.
public class PublishedEvent
{
private List<object> publishers;
private List<EventHandler> subscribers;
...
public IEnumerable<object> Publishers
{
get { ... }
}
public IEnumerable<EventHandler> Subscribers
{
get { ... }
}

100

ch a pter si x

public void AddPublisher(object publisher, string eventName)

{ ... }
public void RemovePublisher(object publisher, string eventName)
{ ... }
public void AddSubscriber(EventHandler subscriber)
{ ... }
public void RemoveSubscriber(EventHandler subscriber)
{ ... }
private void OnPublisherFiring(object sender, EventArgs e)
{
foreach(EventHandler subscriber in subscribers)
{
subscriber(sender, e);
}
}
}

The SimpleEventBroker extension must create and populate an EventBroker instance when you register and
resolve types from the container that use the Publishes and SubscribesTo attributes.

Implementing the Container Extension

You may find it useful to open the sample application, EventBroker, that accompanies this guide in Visual
Studio while you read this section.
The first step to implement a container extension is to extend the abstract UnityContainerExtension class and
to override the Initialize method as shown in the following code sample.
public class SimpleEventBrokerExtension :
UnityContainerExtension, ISimpleEventBrokerConfiguration
{
private readonly EventBroker broker = new EventBroker();
protected override void Initialize()
{
Context.Container.RegisterInstance(broker);
Context.Strategies.AddNew<EventBrokerReflectionStrategy>(
UnityBuildStage.PreCreation);
Context.Strategies.AddNew<EventBrokerWireupStrategy>(
UnityBuildStage.Initialization);
}
public EventBroker Broker
{
get { return broker; }
}
}

E xtending Unit y

101

The Intialize method first registers an instance of the EventBroker class shown in the previous section with the
container. The Initialize method then adds two new strategies to the container: an EventBrokerReflectionStrategy and an EventBrokerWireUpStrategy.
You can add strategies to the container that take effect at different stages in the containers activities as it
builds up an object instance. The EventBrokerReflectionStrategy strategy takes effect at the PreCreation
stage: during this stage, the container is using reflection to discover the constructors, properties, and methods
of the type currently being resolved. The EventBrokerWireUpStrategy strategy takes effect at the Initialization stage: during this stage, the container performs property and method injection on the object currently
being instantiated by the container.
The following table summarizes the different stages where you can add your custom strategies to the container.
Stage

Description

Setup

The first stage. By default, nothing happens here.

TypeMapping

The second stage. Type mapping takes place here.

Lifetime

The third stage. The container checks for a lifetime manager.

PreCreation

The fourth stage. The container uses reflection to discover the constructors, properties, and methods
of the type being resolved.

Creation

The fifth stage. The container creates the instance of the type being resolved.

Initialization

The sixth stage. The container performs any property and method injection on the instance it has just
created.

PostInitialization

The last stage. By default, nothing happens here.

If the container discovers a lifetime manager during the Lifetime phase that indicates that an instance
already exists, then the container skips the remaining phases. You can see this happening in Figure 1, the
sequence diagram that shows how the custom lifetime manager works.
The strategy classes that you add to the list of strategies at each stage all extend the BuilderStrategy class. In
these strategy classes, you typically override the PreBuildUp method to modify the way that the container
builds the object it is resolving, although you can also override the PostBuildUp method to modify the object
after the container has completed its work and the PreTearDown and PostTearDown methods if you need to
modify the way the container removes the object.
Discovering the Publishers and Subscribers
In the event broker example, during the PreCreation stage, EventBrokerReflectionStrategy strategy scans the
type the container will create for events decorated with the Publishes attribute and methods decorated with
the SubscribesTo attribute and stores this information in an EventBrokerInfoPolicy object as shown in the
following code sample.

102

ch a pter si x

public class EventBrokerReflectionStrategy : BuilderStrategy

{
public override void PreBuildUp(IBuilderContext context)
{
if (context.Policies.Get<IEventBrokerInfoPolicy>
(context.BuildKey) == null)
{
EventBrokerInfoPolicy policy = new EventBrokerInfoPolicy();
context.Policies.Set<IEventBrokerInfoPolicy>(policy, context.BuildKey);
AddPublicationsToPolicy(context.BuildKey, policy);
AddSubscriptionsToPolicy(context.BuildKey, policy);
}
}
private void AddPublicationsToPolicy(NamedTypeBuildKey buildKey,
EventBrokerInfoPolicy policy)
{
Type t = buildKey.Type;
foreach(EventInfo eventInfo in t.GetEvents())
{
PublishesAttribute[] attrs =
(PublishesAttribute[])eventInfo.GetCustomAttributes(
typeof(PublishesAttribute), true);
if(attrs.Length > 0)
{
policy.AddPublication(attrs[0].EventName, eventInfo.Name);
}
}
}
private void AddSubscriptionsToPolicy(NamedTypeBuildKey buildKey,
EventBrokerInfoPolicy policy)
{
foreach (MethodInfo method in buildKey.Type.GetMethods())
{
SubscribesToAttribute[] attrs =
(SubscribesToAttribute[])
method.GetCustomAttributes(typeof(SubscribesToAttribute), true);
if(attrs.Length > 0)
{
policy.AddSubscription(attrs[0].EventName, method);
}
}
}
}

E xtending Unit y

103

Wiring Up the Publishers and Subscribers

During the Initialization stage, the EventBrokerWireUpStrategy strategy retrieves the policy information
stored by the EventBrokerReflectionStrategy strategy during the PreCreation stage and populates the EventBroker object with information about the events and publishers that the subscribers subscribe to. The following code sample shows how the EventBrokerWireUpStrategy strategy implements this final step in the event
broker extension.
public class EventBrokerWireupStrategy : BuilderStrategy
{
public override void PreBuildUp(IBuilderContext context)
{
if (context.Existing != null)
{
IEventBrokerInfoPolicy policy =
context.Policies.Get<IEventBrokerInfoPolicy>(context.BuildKey);
if(policy != null)
{
EventBroker broker = GetBroker(context);
foreach(PublicationInfo pub in policy.Publications)
{
broker.RegisterPublisher(pub.PublishedEventName,
context.Existing, pub.EventName);
}
foreach(SubscriptionInfo sub in policy.Subscriptions)
{
broker.RegisterSubscriber(sub.PublishedEventName,
(EventHandler)Delegate.CreateDelegate(
typeof(EventHandler),
context.Existing,
sub.Subscriber));
}
}
}
}
private EventBroker GetBroker(IBuilderContext context)
{
var broker = context.NewBuildUp<EventBroker>();
if (broker == null)
{
throw new InvalidOperationException("No event broker available");
}
return broker;
}
}

104

ch a pter si x

Summary
This chapter described two ways that you can extend Unity. First, it described
how to create custom lifetime managers that enable you to modify the way that
Unity stores and tracks the instances of registered types created by the container. Second, it described how you can create container extensions that enable
you to add custom behavior into the build-up process that occurs when the
container resolves a type and creates a new instance.
The Unity source code
is a great reference for
implementations of both
lifetime managers and
container extensions
because Unity uses both
of these mechanisms to
implement some of its
standard functionality.

More Information
All links in this book are accessible from the books online bibliography
available at: http://aka.ms/unitybiblio

Summary

The goal of this guide was to introduce you to dependency injection, interception, and the Unity application
block from Enterprise Library. Dependency injection and interception are widely used techniques that help
developers write maintainable, testable, flexible, and extensible code, especially for large, enterprise applications.
Unity is not the only tool you can use to add dependency injection and interception to your applications, but
if you have read this guide it should be clear that Unity is easy to use and easy to extend. You can use Unity in
a wide range of application types such as desktop, web, WCF, and cloud.
Chapters 2 and 3 of this guide focused on the topic of dependency injection, while Chapters 4 and 5 focused
on interception. Chapter 6 described a number of ways that you can extend Unity itself if you find that out-ofthe-box it doesnt support a specific scenario.
The remaining parts of this guide, Tales from the Trenches, offer a different perspective: these chapters offer
brief case studies that describe real-world use of Unity. They make clear the range of scenarios in which you
can use Unity, and also highlight its ease of use and flexibility.
We hope you find Unity adds significant benefits to your applications and helps you to achieve those goals of
maintainability, testability, flexibility, and extensibility in your own projects.

105

Tales from the Trenches

Using Unity

My first exposure to Unity was a number of years ago. It was consistently presented to me as something we should
use to help with testing by other developers. It seemed like a good idea in practice, but it was never something
that was really pushed. In fact, it seemed like most developers would go out of their way to avoid using it due to
the perceived difficulty of setting it up. Eventually, I stopped looking for places to inject it. No pun intended.
But after serving on the Advisory Board for Unity 3, I was determined to once again look for opportunity in
which to use it responsibly.
While working on bringing up a collection of OData-enabled ASP.NET Web API services querying a DatabaseFirst Entity Framework implementation, it struck me as fantastic place to use it for disposing my DbContexts.
When executing a LINQ-to-entities statement when using an OData enabled method, a standard using statement, like-so, will not work as the DbContext will be closed before the OData query attempts to apply itself.
publicclassProductsController:OdataController
{
[HttpGet]
[Queryable]
publicIQueryable<ProductRestObject>Get()
{

varmanager=newProductsManager();

}
}

returnmanager.GetProducts();

publicclassProductsManager
{
publicIQueryable<ProductRestObject>GetProducts()
{

using(varcontext=newProductsContext())
{

returncontext.Ef_Products

.Select(product=>

newProductRestObject()
{

Id=product.ProductId,
Name=product.ProductName,
Price=product.ProductPrice
});
}
}
}

Case study provided by Luke Sigler and edited by Julin Domnguez.

107

108

Ta les from the Tr enches

Simply removing the using statement alone was unwise so I decided to have the ProductsManager class handle
disposing of the context manually.
publicclassProductsManager:Idisposable
{
publicProductsContextCurrentProductsContext{get;set;}
publicProductsManager()
{

CurrentProductsContext=newProductsContext();
}
publicIQueryable<ProductRestObject>GetProducts()
{

returnCurrentProductsContext.Ef_Products

.Select(product=>

newProductRestObject()

Id=product.ProductId,
Name=product.ProductName,
Price=product.ProductPrice
});
}
publicvoidDispose()
{

CurrentProductsContext.Dispose();
}
}
publicclassProductsController:OdataController
{
publicProductsManagerCurrentProductsManager{get;set;}
[HttpGet]
[Queryable]
publicIQueryable<ProductRestObject>Get()
{

CurrentProductsManager=newProductsManager();
returnCurrentProductsManager.GetProducts();
}
protectedoverridevoidDispose(booldisposing)
{

CurrentProductsManager.Dispose();
}
}

Using Unit y

109

But that was just plain sloppy. This is when it occurred to me that Unity would work great here. After getting
the Unity WebApi Bootstrapper package from NuGet, I changed it up a bit.
publicinterfaceIProductsDataSource:Idisposable
{
IQueryable<ProductRestObject>GetProducts();
}
publicclassProductsManager:IproductsDataSource
{
publicProductsContextCurrentProductsContext{get;set;}
publicProductsManager()
{

CurrentProductsContext=newProductsContext();
}
publicIQueryable<ProductRestObject>GetProducts()
{

returnCurrentProductsContext.Ef_Products

.Select(product=>

newProductRestObject()

Id=product.ProductId,
Name=product.ProductName,
Price=product.ProductPrice
});
}
publicvoidDispose()
{

CurrentProductsContext.Dispose();
}
}
publicclassProductsController:OdataController
{
publicIProductsDataSourceCurrentProductsDataSource{get;set;}
publicProductsController(IProductsDataSourceproductsDataSource)
{

CurrentProductsDataSource=productsDataSource;
}
[HttpGet]
[Queryable]
publicIQueryable<ProductRestObject>Get()
{

returnCurrentProductsDataSource.GetProducts();
}
protectedoverridevoidDispose(booldisposing)
{

CurrentProductsDataSource.Dispose();
base.Dispose(disposing);
}
}

110

Ta les from the Tr enches

I then added the following line to the RegisterTypes method in the UnityConfig class generated when I
brought in the package.
container.RegisterType<IProductsDataSource,ProductsManager>();

It all worked great, but I didnt love the fact that I would have to override the Dispose method to dispose of items
that were being resolved by Unity. I knew I would have to use a LifetimeManager.
I then made the following changes:
In the Start method of my UnityWebApiActivator class, I deleted the preexisting DependencyResolver instantiation:
varresolver=newMicrosoft.Practices.Unity.WebApi.
UnityDependencyResolver(UnityConfig.GetConfiguredContainer());

And I uncommented out the provided UnityHierarchicalDependencyResolver instantiation:

varresolver=newUnityHierarchicalDependencyResolver(UnityConfig.GetConfiguredContainer());

I then changed the RegisterTypes method in my UnityConfig file to the following:

container.RegisterType<IProductsDataSource,ProductsManager>(newHierarchicalLifetimeManager());
container.RegisterType<ProductsController>(
newInjectionConstructor(
newResolvedParameter<IProductsDataSource>()));

I was then able to remove the overridden Dispose method in the ProductsController.
And that was it.

Tales from the Trenches

One Users Story

Customizing Unity
My use of Unity started with attending the first Unity workshop hosted by the p&p group post 1.0 release.
Admittedly, those were the early days of Dependency Injection and Inversion of Control (IoC) and I was completely off base as to my understanding of the concepts. However, as members of the Unity team began asking
people in the room for their use cases, I began to understand how Unity was not only a well-designed IoC
container, but extremely flexible and lightweight in its implementation.
My first stab at customizing Unity was to have an ILogger interface in our application that automatically set
its logger name to the class it was being injected into. I looked at the extension interface; it looked simple
enough, just have a class implement the extension base class and youre good to go right? Well its a little more
complicated than that.
First, lets take a look at what we wanted the logger to do. Most loggers require something to identify where a
log message is coming from such as a category, a source, or a name. More often than not, a fully qualified class
name is used to enable a user to then filter by namespace. Since the value of the category is static, after you have
injected the logger, it becomes redundant to include it on every log line or to resolve it on every call; therefore,
factoring it into dependency injection makes life easier for the developer. An example of this would be as follows:
classMyClass
{
privatereadonlyILogger_logger;
publicMyClass(ILoggerlogger)
{
_logger=logger;
}
publicstringSayHello(stringname)
{
//Loggernameherewillbe'MyApplication.MyClass'
_logger.LogDebug("SayingHelloto{0}",name);
returnstring.Format("Hello{0}",name);
}
}

The first critical thing to understand is that Unity includes a sequence of build stages that are similar to a road
map: each stage in this pipeline facilitates Unity in building the object that needs to be resolved. Some pipeline
stages run each time an object is constructed, others only run the first time the container needs to build an
object. Because of this, you need to plan how your customization fits in to this build pipeline, and which
stages you need to modify to create the instance you need. There was some trial and error but in the end, it
was fairly easy to modify the correct stages for the custom logger.
Case study provided by Dan Piessens.
111

112

Ta les from the Tr enches

The next step is to think about how your extension will store or discover the additional data it needs; most of
this revolves around the build context. The build context, which is similar to a SQL query plan, provides information to the pipeline on subsequent builds of the object and indicates what information the pipeline can be
cached and what it needs to determine on each build. The Unity team made this straightforward with the build
context so its not nearly as error prone as with some other IoC containers.
For this extension, I created the LoggerNameResolverPolicy. This policy is ultimately what constructs and injects
the logger into the target class. It also stores the logger name, which in this case is the parent class name. Now,
you might ask why we called it the parent? One if the tricky points to grasp is that the build plan is actually
for the logger, not the class that you are injecting the logger into. The policy class looks like this:
publicsealedclassLoggerNameResolverPolicy:IDependencyResolverPolicy
{
privatereadonlyType_type;
privatereadonlystring_name;
privatereadonlystring_parentClassName;
publicLoggerNameResolverPolicy(Typetype,stringname,
stringparentClassName)
{
_type=type;
_name=name;
_parentClassName=parentClassName;
}
publicobjectResolve(IBuilderContextcontext)
{
varlifetimeManager=newContainerControlledLifetimeManager();
lifetimeManager.SetValue(_parentClassName);
varloggerNameKey=
newNamedTypeBuildKey(typeof(string),"loggerName");
//Createthebuildcontextforthelogger
varnewKey=newNamedTypeBuildKey(_type,_name);
//Registertheitemasatransientpolicy
context.Policies.Set<IBuildKeyMappingPolicy>(
newBuildKeyMappingPolicy(loggerNameKey),loggerNameKey);
context.Policies.Set<ILifetimePolicy>(
lifetimeManager,loggerNameKey);
context.Lifetime.Add(lifetimeManager);
try
{
returncontext.NewBuildUp(newKey);
}
finally
{
context.Lifetime.Remove(lifetimeManager);
context.Policies.Clear<IBuildKeyMappingPolicy>(loggerNameKey);
context.Policies.Clear<ILifetimePolicy>(loggerNameKey);
}
}
}

One User's Story Customizing Unit y

113

I could refactor this implementation of the resolver policy class to use the parameter override support in Unity
2.0 and 3.0, but this code worked with Unity 1.0 and still works today. I wanted to show how even though
Unity has evolved, older plugins such as this continue to work. If youre wondering what the code is doing, it is
using the concept of a transient policy in Unity. The policy only exists while the container is building the
object and is then it is disposed. The interesting part is how the parent policy is cached with the object so
whether you need one instance or a thousand, the container only creates the policy once for the class.
The next step was to capture the logger name whenever a property or constructor parameter asks for it. This
was fairly simple as Unity has specific policies that run whenever it resolves a property or constructor argument
(methods too but we dont use method resolution). These are represented by interfaces IPropertySelectorPolicy and IConstructorSelectorPolicy. There are also abstract implementations that are generally very useful.
In our case however, we had the need to create resolver implementations that used type metadata in several
places and wanted to abstract out that portions of the property or constructor selection policy with our own
that passed in the buildContext variable and the constructing type. The following code example shows how we
overrode the default constructor selector policy to suit our needs. The primary modification is in the CreateResolver method, passing in the build context to look for our own policies and using the build key to capture
the type being created.
publicclassContainerConstructorSelectorPolicy:IConstructorSelectorPolicy
{
publicvirtualSelectedConstructorSelectConstructor(
IBuilderContextcontext,IPolicyListresolverPolicyDestination)
{
//Sameasdefaultimplementation...
}
privatestaticSelectedConstructorCreateSelectedConstructor(
IBuilderContextcontext,ConstructorInfoctor,
IPolicyListresolverPolicyDestination)
{
varresult=newSelectedConstructor(ctor);
foreach(varparaminctor.GetParameters())
{
varkey=Guid.NewGuid().ToString();
varpolicy=CreateResolver(context,param);
resolverPolicyDestination.Set(policy,key);
DependencyResolverTrackerPolicy.TrackKey(context.PersistentPolicies,
context.BuildKey,key);
result.AddParameterKey(key);
}
returnresult;
}
privatestaticIDependencyResolverPolicyCreateResolver(
IBuilderContextcontext,ParameterInfoparameterInfo)
{
varpolicy=context.Policies
.Get<IParameterDependencyResolver>(
newNamedTypeBuildKey(parameterInfo.ParameterType));
IDependencyResolverPolicyresolverPolicy=null;
if(policy!=null)

114

Ta les from the Tr enches

{
resolverPolicy=policy.CreateResolver(context.BuildKey.Type,
parameterInfo);
}
returnresolverPolicy??newFixedTypeResolverPolicy(
parameterInfo.ParameterType);
}
privatestaticConstructorInfoFindInjectionConstructor(TypetypeToConstruct)
{
//Sameasdefaultimplementation...
}
privatestaticConstructorInfoFindLongestConstructor(TypetypeToConstruct)
{
//Sameasdefaultimplementation...
}
}

Once we had these main build policies in place, we needed to implement our new interfaces IPropertyDependencyResolver and IParameterDependencyResolver. The signatures are almost identical, except that
one passes in parameter arguments and the other passes in constructor arguments. For simplicity, Ill only
show the constructor resolver, but the pattern is the same for a parameter resolver or even a method resolver if you choose to support them.
publicclassLoggerNameConstuctorParameterPolicy:IParameterDependencyResolver
{
publicIDependencyResolverPolicyCreateResolver(
TypecurrentType,ParameterInfoparam)
{
returnnewLoggerNameResolverPolicy(param.ParameterType,null,
currentType.FullName);
}
}

The final module wire up is easy; in fact, most of the items dont need additional configuration. We had a few
errors on the first try, but the debugging experience made those errors very descriptive. The fact that Unity is
open source also helped greatly, there were a few times when I peeked at the default implementation code to
get hints about the right way to construct things. The Unity guides that ship with this release will also prove to
be a big help with customization. The code to create the extension itself simply involved registering the policies
into the parameter and constructor discovery stages and indicating that this should occur for any ILogger item.
It also included a section to override the default constructor selector and property selector policies with our
custom ones. In the end we moved this code to a separate extension so it could be better reused. The final
registration code looked like this:

One User's Story Customizing Unit y

115

publicclassLoggerNameExtension:UnityContainerExtension
{
protectedoverridevoidInitialize()
{
// Override base Unity policies.
Context.Policies.ClearDefault<IConstructorSelectorPolicy>();
Context.Policies.SetDefault<IConstructorSelectorPolicy>(
new ContainerConstructorSelectorPolicy());
Context.Policies.SetDefault<IPropertySelectorPolicy>(
new ContainerPropertySelectorPolicy());
// Set logging specific policies
varbuildKey=newNamedTypeBuildKey(typeof(ILogger));
Context.Policies.Set<IParameterDependencyResolver>(
newLoggerNameConstuctorParameterPolicy(),buildKey);
Context.Policies.Set<IPropertyDependencyResolver>(
newLoggerNamePropertyPolicy(),buildKey);
}
}

Emboldened by our initial success, my team and I went on to create several more Unity extensions; some added
additional dependency resolution attribute options, others effectively created specialized factory instances of
objects. Over the past few years, many of the extensions were retired as Unity itself expanded to support
factory methods, alternate registration attributes, and most recently Lazy<T> item support and registration by
convention. The only real customization needed today is for scenarios such as the logger, where metadata is
pulled from the code that calls it.

More Information
All links in this book are accessible from the books online bibliography available at: http://aka.ms/unitybiblio

Tales from the Trenches

Using Unity in a
Windows Store app
AdventureWorks Shopper
The AdventureWorks Shopper reference implementation is a Windows Store business app, which uses Prism
for the Windows Runtime to accelerate app development. It provides guidance to developers who want to
create a Windows Store business app using C#, Extensible Application Markup Language (XAML), the Windows
Runtime, and modern development practices.
The reference implementation shows how to:
Implement pages, controls, touch, navigation, settings, suspend/resume, search, tiles, and tile notifications.
Implement the Model-View-ViewModel (MVVM) pattern.
Validate user input for correctness.
Manage application data.
Test your app and tune its performance.

Prism for the Windows Runtime accelerates app development by providing core services commonly required
by a Windows Store app. These core services include providing support for bootstrapping MVVM apps, state
management, validation of user input, navigation, event aggregation, data binding, commands, Flyouts, settings,
and search.
While Prism for the Windows Runtime does not require you to use a dependency injection container, for AdventureWorks Shopper we chose to use the Unity dependency injection container. AdventureWorks Shopper
uses the MVVM pattern, and in the context of a Windows Store app that uses the MVVM pattern, there are
specific advantages to using Unity:
It can be used to register and resolve view models and views.
It can be used to register services, and inject them into view models.
It can be used to create view models and inject the views.

In AdventureWorks Shopper we used Unity to manage the instantiation of view model and infrastructure
service classes. Only one class holds a reference to the Unity dependency injection container. This class instantiates the UnityContainter object and registers instances and types with the container.
The main reason for using Unity is that it gave us the ability to decouple our concrete types from the code that
depends on those types. During an objects creation, the container injects any dependencies that the object
requires into it. If those dependencies have not yet been created, the container creates and resolves them first.
Using Unity has resulted in several advantages for AdventureWorks Shopper:
It removed the need for a component to locate its dependencies and manage their lifetime.
It allowed mapping of implemented dependencies without affecting the components.

Case study provided by David Britch.

116

Using Unit y in a Windows Store a pp

117

It facilitated testability by allowing dependencies to be mocked.

It increased maintainability by allowing new components to be easily added to the system.

Overall, using Unity in a Windows Store business app has resulted in a much cleaner architecture and has helped
to further our separation of concerns.

References
For further information about the AdventureWorks Shopper reference implementation, and Prism for the
Windows Runtime, see:
Developing a business app for the Windows Store using C#: AdventureWorks Shopper
Prism for the Windows Runtime reference
Prism StoreApps library
Prism PubSubEvents library

More Information
All links in this book are accessible from the books online bibliography available at: http://aka.ms/unitybiblio

Appendix A

Unity and
Windows Store apps
When you create applications for Windows 8, those applications are known as Windows Store apps. You can
develop Windows Store apps using C#, Visual Basic, JavaScript, and C++.
You can use Unity when you develop Windows Store apps using C# and Visual Basic. However, there are some
limitations in the functionality of the version of Unity that targets the .NET for Windows Store apps version of
the .NET Framework; this appendix describes those limitations.

The UnityServiceLocator Class

You cannot use the Microsoft.Practices.Unity.UnityServiceLocator class with the .NET for Windows Store
apps version of the .NET Framework. This class depends on the Common Service Locator library, which is not
available for Windows Store apps.

Unity Design-time Configuration

You cannot configure your Unity container using a configuration file such as app.config. Windows Store apps
must configure Unity containers programmatically.
This is because the Unity.Configuration assembly is not compatible with Windows Store apps.

Unity Interception
Unity Interception is not available to Windows Store apps.
This is because the Unity.Interception and Unity.Interception.Configuration assemblies are not compatible with
Windows Store apps.

More Information
All links in this book are accessible from the books online bibliography available at: http://aka.ms/unitybiblio

119

Index

CachingCallHandlerAttribute class, 84-85

about this guide, xi

child containers, 38-39

abstract factory pattern, 14

community support, xi

AdventureWorks Shopper, 116-117

constructor injection, 27-28

alternative interception techniques, 73-79

ContainerControlledLifetimeManager class, 55

AOP See Aspect Oriented Programming (AOP)

ContainerControlledLifetimeManager lifetime manager, 91

appendix, 119

ContainerControlledLifetimeManager type, 35, 52

Application_Start method, 42-43

ContainerManagedLifetime class, 50-51

Aspect Oriented Programming (AOP), 60, 63-64

containers

ASP.NET web applications resolution, 41-42

extension, 100-101

audience, xii

interactions with a custom lifetime manager, 95

auto-registration, 34-37

corrective maintenance, 1

automatic factories, 47-50

CreateResolver method, 113-114

CreateTenantStore method, 11-12

crosscutting concerns

behaviors, 60

interception, 59-60

to add an interface to an existing class, 76

pipeline, 72

overview, 3
custom lifetime manager creation, 91-95

Beth See developer role

build plan, 112

BuildUp method, 42-43

DbContexts, 107

C
CachedLifetimeManager class, 93-94

decorator chains, 63
decorator pattern, 60-62
deferred resolution, 50-51

CachingCallHandler class, 80-83, 85-87

121

122

dependency injection, 11-20

abstract factory pattern, 14
dependency illustration, 16
factory method pattern, 11-13
factory pattern, 15
ManagementController class, 11-17
object composition, 17
object lifetime, 17-18
overview, 1
service locator pattern, 14
simple factory pattern, 13, 15

externally controlled lifetime management, 55

ExternallyControlledLifetimeManager class, 55

F
factory class, 48-50
factory method pattern, 11-13
factory patterndependencies in, 15
flexibility, 2

G
getInjectionMembers parameter, 36
guide, xi

and unit testing, 56-57

when not to use, 19-20

See also injection; Unity dependency injection

hierarchical lifetime management, 52-53

dependency inversion principle, 9

DependencyInjectionContainer class, 16-17
design-time configuration, 33
developer role (Beth), xii
DI See dependency injection
dispose task, 21, 23

HierarchicalLifetimeManager type, 52-53

I
IConstructorSelectorPolicy interface, 113
IExpenseRepository type, 42
IInterceptionBehavior interface, 68-71
IIS, 47

ILogger interface, 76, 111-115

event broker containers

injection

extensions lacking, 96-97

AzureTable class, 18

extensions present, 97-98

MessageQueue class, 19

EventBroker class, 99-100

method call injection, 19

EventBrokerReflectionStrategy strategy, 101-102

property setter injection, 18

EventBrokerWireUpStrategy strategy, 103

types of, 18-19

examples

See also dependency injection; Unity dependency injection

dependency injection with Unity, 24-32

InjectionConstructor, 29, 35

type registrations, 26-29

InjectionConstructor attributes, 27

type resolutions, 29-32

instance interception, 64-65

ExpenseRepository class, 87-88

instance interception/type interception, 73

extensibility, 2

instance registration, 26-27

index

interception, 59-66

policy injection and attributes, 83-85

Aspect Oriented Programming (AOP), 63

Policy Injection Application Block, 81

crosscutting concerns, 59-60

registering an interceptor, 71

decorator pattern, 60-62

SemanticLogCallHandler type, 88-89

described, 64

TenantStore class, 78-79

instance interception, 64-65

TransparentProxyInterceptor type, 73-74

interception, 65-66

Unity Interception Extension libraries, 68

ITenantStore interface, 60-62

using an interceptor, 72-73

ITenantStore type, 63

virtual method interceptor, 75

TenantStore class, 60-62

VirtualMethodInterceptor type, 74-75

UploadLogo method, 62

interceptors, 67-73

wiring up decorator chains with Unity, 63

defining, 68-71

without the Unity container, 77-78

using, 72-73

See also interception using Unity; interceptors

See also interception

interception using Unity, 67-89

interface segregation principle, 9

alternative interception techniques, 73-79

InterfaceInterceptor type, 73

behavior pipeline, 72

introduction, 1-10

behavior to add an interface to an existing class, 76

IParameterDependencyResolver interface, 114

CachingCallHandler class, 80-83, 85-87

IPropertyDependencyResolver interface, 114

CachingCallHandlerAttribute class, 84-85

IPropertySelectorPolicy interface, 113

configuring the Unity container to support

interception, 68

IStorage interface, 94-95

defining an interceptor, 68-71

described, 67
example, 87-88

ISurveyAnswerStore type, 25-26

IT professional role (Poe), xiii
ITenantStore interface, 60-62
ITenantStore type, 63

ExpenseRepository class, 87-88

IInterceptionBehavior interface, 68-71

ILogger interface, 76

Jana See software architect role (Jana)

instance interception/type interception, 73

interceptors, 67-73
InterfaceInterceptor type, 73
LogCallHandler class, 86-87
LoggingCallHandler class, 80-83, 85-87
LoggingCallHandlerAttribute class, 84-85
LoggingInterceptionBehavior class, 76-77
policy injection, 79-83

L
late binding, 2
Lazy<T> type, 50-51
lifecycle, 21-23
lifetime management, 51-56
and resolved objects, 92
LifetimeManager class, 95

123

124

Liskov substitution principle, 8-9

per request lifetime management, 55-56

LoadConfiguration method, 33

per resolve lifetime management, 53-55

LogCallHandler class, 86-87

per thread lifetime management, 56

LoggerNameResolverPolicy, 112

perfective maintenance, 1

loggers, 111

PerRequestLifetime manager, 30

LoggingCallHandler class, 80-83, 85-87

PerRequestLifetimeManager class, 55-56

LoggingCallHandlerAttribute class, 84-85

PerResolveLifetimeManager class, 55

LoggingInterceptionBehavior class, 76-77

Poe See IT professional role (Poe)

loose coupling, 3-7

policy injection, 79-83

See also tight coupling

and attributes, 83-85

Policy Injection Application Block, 81

prerequisites, xii

maintainability, 1-2
ManagementController class, 3-6, 11-17, 22
Markus See software developer role (Markus)
MessageQueue class, 19
method call injection, 19

ProductsManager class, 108

property setter injection, 18
PublishedEvent class, 99-100
Publisher classes, 96-98
publishers and subscribers discovery, 101-102

motivations, 1
MVC application resolution, 30

R
RegisterTypes method, 24-25, 34

registration, 39-41

named type registrations, 32-33

by convention, 34-37

NuGet, 23, 30, 55, 68

by convention and generic types, 38

information viewing, 39-41

interceptor, 71

object composition, 17

and resolution in the code, 23

object lifetime, 17-18

object-oriented programming and design principles, 8
open/closed principle, 8
open generics registration, 28
overview
dependency injection, 1
Unity, xi

P
parallel development, 3
parameter overrides, 29

registration task, 21-22

RegistrationConvention class, 37
requirements, xii
resolution task, 21-22
roles
described, xii-xiii
see also developer role (Beth); IT professional role (Poe);
software developer role (Markus); software architect
role (Jana); Unity expert (Carlos)
run time information, 31-32

index

LoggerNameResolverPolicy, 112

self-hosting, 46

loggers, 111

SemanticLogCallHandler type, 88-89

ProductsManager class, 108

service locator pattern, 14

transient policy, 113

ServiceHost class, 44

Unity WebApi Bootstrapper package, 109

simple event broker, 98-100

target audience, xii

simple factory pattern, 13, 15

TDD See test-driven development (TDD)

simple types

team, ix

registering, 27

TenantStore class, 56-57, 60-62, 78-79

resolving, 29

tight coupling overview, 3-8

SimpleMemoryCache class, 94-95

test doubles, 2

single responsibility principle, 8

test-driven development (TDD), 2

software architect role (Jana), xiii

testability, 2

software developer role (Markus), xii

tight coupling, 3-8

SOLID acronym, 8

See also loose coupling

strategies, 101

transient policy, 113

Subscriber classes, 96-98

TransientLifetimeManager class, 54-55

summary, 105

TransientLifetimeManager lifetime manager, 91

support, xi

TransparentProxyInterceptor type, 73-74

SurveyAnswerStore class, 47-48

type interception, 65-66

SurveyAnswerStoredMessage type, 28

type registration, 32-41

SynchronizedLifetimeManager class, 92-95

type resolution, 41

system requirements, xii

in a WCF service, 43-45

tales from the trenches, 107-117

Unity

build plans, 112

adding to the application, 23

CreateResolver method, 113-114

configuring container to support interception, 68

customizing Unity, 111-115

customizing, 111-115

DbContexts, 107

design-time configuration, 119

IConstructorSelectorPolicy interface, 113

expert role (Carlos), xiii

ILogger interface, 111-115

history of, ix-x

IParameterDependencyResolver interface, 114

overview, xi

IPropertyDependencyResolver interface, 114

See also interception; interception using Unity

IPropertySelectorPolicy interface, 113

Unity container, 96-103

125

126

Unity dependency injection, 21-58

MVC application resolution, 30

adding Unity to the application, 23

named type registrations, 32-33

Application_Start method, 42-43

open generics registering, 28

ASP.NET web applications resolution, 41-42

parameter overrides, 29

automatic factories, 47-50

per request lifetime management, 55-56

behaviors, 60

per resolve lifetime management, 53-55

BuildUp method, 42-43

per thread lifetime management, 56

child containers, 38-39

PerRequestLifetime manager, 30

constructor injection, 27-28

PerRequestLifetimeManager class, 55-56

ContainerControlledLifetimeManager class, 55

PerResolveLifetimeManager class, 55

ContainerControlledLifetimeManager type, 35, 52

registering and resolving in the code, 23

ContainerManagedLifetime class, 50-51

RegisterTypes method, 24-25, 34

deferred resolution, 50-51

registration by convention, 34-37

dependency injection and unit testing, 56-57

registration by convention and generic types, 38

design-time configuration, 33

registration information viewing, 39-41

dispose task, 21, 23

registration task, 21-22

example of, 24-32

RegistrationConvention class, 37

example type registrations, 26-29

resolution task, 21-22

example type resolutions, 29-32

resolving types, 41

externally controlled lifetime management, 55

run time information, 31-32

ExternallyControlledLifetimeManager class, 55

self-hosting, 46

factory class, 48-50

ServiceHost class, 44

getInjectionMembers parameter, 36

simple type registration, 27

hierarchical lifetime management, 52-53

simple type resolution, 29

HierarchicalLifetimeManager type, 52-53

SurveyAnswerStore class, 47-48

IExpenseRepository type, 42

SurveyAnswerStoredMessage type, 28-29

InjectionConstructor, 27, 35

TenantStore class, 56-57

InjectionConstructor attributes, 27

TransientLifetimeManager class, 54-55

instance interception, 65

type registration, 32-41

instance registration, 26-27

type resolution in a WCF service, 43-45

ISurveyAnswerStore type, 25-26

UnityInstanceProvider class, 44-45

Lazy<T> type, 50-51

UnityServiceHost class, 46, 47

lifecycle, 21-23

WCF service in IIS or WAS, 47

lifetime management, 51-56

See also dependency injection; injection; interception using

Unity

LoadConfiguration method, 33

index

Unity expert (Carlos), xiii, 27

Unity extension, 91-104

virtual method interceptor, 75

CachedLifetimeManager class, 93-94

container extension, 100-101
container interactions with a custom lifetime manager, 95
ContainerControlledLifetimeManager lifetime
manager, 91

VirtualMethodInterceptor type, 74-75

W
WAS, 47
WCF service in IIS or WAS, 47

custom lifetime manager creation, 91-95

whos who, xii-xiii

discovering the publishers and subscribers, 101-102

Windows Store business app, 116-117, 119

event broker container extensions lacking, 96-97

event broker container extensions present, 97-98
EventBroker class, 99-100
EventBrokerReflectionStrategy strategy, 101-102
EventBrokerWireUpStrategy strategy, 103
extending the Unity container, 96-103
IStorage interface, 94-95
lifetime managers and resolved objects, 92
LifetimeManager class, 95
PublishedEvent class, 99-100
Publisher classes, 96-98
simple event broker, 98-100
SimpleMemoryCache class, 94-95
strategies, 101
Subscriber classes, 96-98
SynchronizedLifetimeManager class, 92-95
TransientLifetimeManager lifetime manager, 91
UnityContainerExtension class, 100-101
wiring up the publishers and subscribers, 103
Unity Interception, 119
Unity Interception Extension libraries, 68
Unity WebApi Bootstrapper package, 109
UnityContainerExtension class, 100-101
UnityInstanceProvider class, 44-45
UnityServiceHost class, 46, 47
UnityServiceLocator class, 119
UploadLogo method, 62

127