With prior versions of PostSharp, you had to be very careful to make your software compatible with any obfuscator. You could not obfuscate aspect classes, you could not store reflection objects, and you could not add use a reflection object in RuntimeInitialize with obfuscated generic methods. Indeed, all these features caused PostSharp to store the name of the metadata objects at build time, and to deserialize it at run time. But since the obfuscator renamed metadata objects after build time, the deserialization failed. Briefly said, previous versions of PostSharp were incompatible with all obfuscators.
PostSharp 2,1 provides a mechanism through which assemblies can be “repaired” after they have been obfuscated. Assemblies are still fully obfuscated, but PostSharp provides a utility to repair the name-object table. The build process is the following:
1. The compiler produces an assembly.
2. PostSharp post-processes the assembly and embeds an name table in the output assembly.
3. The obfuscator renames metadata objects, therefore breaks the PostSharp name table.
4. The post-obfuscation step (implemented by PostSharp) repairs the name table based on the obfuscation map produced by the obfuscator, and mapping the old name to the new name.
Since the post-obfuscation utility works with the obfuscation map, we need one implementation of this utility for every obfuscation map format, therefore for every obfuscation tool. PostSharp 2.1 includes support for Dotfuscator, the leading obfuscator .NET.
Calling the post-obfuscator
Let’s see how it works practically. You can download the example and try yourself.
To call the post-obfuscation step, you have to create an MSBuild project file and import the targets AddIns\PostSharp.AddIn.PostObfuscation.targets.
<?xml version="1.0" encoding="utf-8"?>
<Project xmlns="http://schemas.microsoft.com/developer/msbuild/2003" DefaultTargets="Build">
<ItemGroup>
<!-- Required: Path of dependencies of obfuscated assemblies. -->
<ObfuscatedSearchPath Include="Dotfuscated" />
<!-- Required: Path of dependencies of unobfuscated assemblies. -->
<UnobfuscatedSearchPath Include="Unobfuscated"/>
<!-- Required: Obfuscation map file produced by Dotfuscator. -->
<MapFile Include="Dotfuscated\Map.xml"/>
<!-- Required: Obfuscated files -->
<ObfuscatedAssembly Include="Dotfuscated\*.dll"/>
<ObfuscatedAssembly Include="Dotfuscated\*.exe"/>
<!-- Optional -->
<ContentFiles Include="Unobfuscated\*.config"/>
</ItemGroup>
<PropertyGroup>
<!-- Required: output directory -->
<OutputPath>$(MSBuildProjectDirectory)\Remapped</OutputPath>
<!-- Required: selection of the remapping strategy -->
<Obfuscator>Dotfuscator</Obfuscator>
</PropertyGroup>
<Target Name="Build" DependsOnTargets="PostObfuscationRemap">
<!-- Optional: copy other files to the output directory-->
<Copy SourceFiles="@(ContentFiles)" DestinationFolder="$(OutputPath)"/>
</Target>
<Target Name="Rebuild" DependsOnTargets="Clean;Build">
</Target>
<Import Project="..\..\Build\bin\AddIns\PostSharp.AddIn.PostObfuscation.targets" />
</Project>
As you can see, this file defines some locations (required items and properties), and calls the PostObfuscationRemap target defined in PostSharp.AddIn.PostObfuscation.targets.
Safe code references
The post-obfuscator ensures the following references to code are repaired after obfuscation:
- References you are not aware of (used by PostSharp to provide the runtime features).
- Reflection objects (for instance MethodInfo, Type) serialized inside objects using the default serializer (BinaryFormatter). This feature is implemented thanks to serialization surrogates for reflection objects.
There is no magic. Other strings referencing to a metadata element, and stored somewhere else than in a reflection object serialized by the default serializer, will not be fixed. If you wrote your custom aspect serializer and want to take advantage of metadata reference serialization, you have to play with IMetadataEmitter/IMetadataDispenser.
In other words, if you want to write obfuscation-safe aspects, the only serialized metadata references should be reflection objects in an aspect field (or somewhere under an aspect field).
Stripping of events and properties
Dotfuscator allows users to remove events and properties from code. The reason is simple: they are useless at runtime for the CLR. They are pure compiler syntactic sugar… or rather were, before XAML. So what if your aspect stores a PropertyInfo or EventInfo?
PostSharp will be smart enough to provide you with a PropertyInfo or EventInfo. The only visible difference is that the Name property will return null. Other properties and methods will work as usually. Under the hood, PostSharp will give you an object of type ObfuscatedPropertyInfo or ObfuscatedEventInfo instead of RuntimePropertyInfo or RuntimeEventInfo, but your code should not see the difference.
Support for other obfuscators
As a first step, we provide support for Dotfuscator. If other vendors are interested to support PostSharp, we’ll gladly provide them with the specification of the binary format. If you want to contribute a post-obfuscation filter, then be pleased that the Dotfuscator-specific code is only 56 lines long (parsing of their XML obfuscation map). The post-obfuscation add-in is not obfuscated so feel free to use a decompiler :).
Summary
That’s it for obfuscation. A small, unexciting feature, but people who need to obfuscate their applications don’t have any excuse not to use PostSharp!
Happy PostSharping!
In my previous post, I introduced the new class ReflectionSearch and two of its methods: GetCustomAttributesOfType and GetCustomAttributesOnTarget. These methods give access to PostSharp’s internal repository of custom attributes.
Today, I would like to cover the second feature of ReflectionSearch: querying the relationships between elements of code. Three kinds of relationships are supported:
- type inheritance (classes and interfaces),
- usage in statements (a type, method, or field is used inside a method body),
- member type,
- any of the above, but as a type element (described below).
Navigating Type Inheritance
The system reflection API makes it easy to navigate from a child class to its parents, but the opposite is not true: if you want to retrieve all classes implementing a given interface in the assembly, you have to enumerate all types in this assembly, then enumerate all implemented interfaces, and do a recursion on the parent type. As was the case in the previous post with custom attributes, PostSharp indexes the type hierarchy for internal use. From PostSharp 2.1, ReflectionSearch exposes this feature to high-level aspects, and it comes with no performance cost.
The method ReflectionSearch.GetDerivedTypes does what it says, with one remark: by default, it returns only first-level derived types, i.e. it does not return derived types of derived types. If you want a deep search, you have to specify the option ReflectionSearchOptions.IncludeDerivedTypes.
Note that the method does not return an array of types, but of TypeInheritanceCodeReference. This object provides information about the base type. You may think the base type is always what you passed in the argument, but it isn’t always true. Consider the following code:
class GenericCollection<T> : ICollection<T> {}
class IntCollection : GenericCollection<int> {}
TypeInheritanceCodeReference[] derivedTypes =
ReflectionSearch.GetDerivedTypes( typeof(ICollection<>);
In this case, GetDerivedTypes would return two pairs: (GenericCollection<T>, ICollection<T>) and (IntCollection, ICollection<int>). In each case, the base type is a different instance of the generic type.
Navigating Usage
As a part of the post-compilation process, PostSharp scans all method bodies and builds a bi-directional map of using/used-by relationships, i.e. it determines which fields, methods or type are used in statements and expressions of a method body, and creates an index. This functionality is now available inside PostSharp.dll and is exposed on two methods:
- ReflectionSearch.GetDeclarationsUsedByMethod returns the set of fields, methods and types used by a given method.
- ReflectionSearch.GetMethodsUsingDeclaration returns all methods using a given field, method, or type.
As you can note, it is not possible to query references to properties and events. Indeed, properties and events are compiler “syntactic sugar” and are never referred to by instructions at MSIL level. If you want to query properties or events, you have to query each accessor separately (add, remove, get, set).
These methods return an array of MethodUsageCodeReference. This object contains both ends of the relationship, as well as a bit mask telling which instructions the declaration was used with – so you can distinguish between a get or a set operation on a field.
Navigating Member Types
The last thing you can do with ReflectionSearch is to get a list of all members of a given type. By member, I mean a field, a property, an event, a parameter, or a return-value parameter.
This feature is exposed on ReflectionsSearch.GetMembersOfType.
Note that PostSharp does not use this feature internally, so there is an additional performance cost in calling that method. The first time you invoke GetMembersOfType, the assembly will be scanned and indexed (using PostSharp’s internal APIes, not System.Reflection).
Type Elements
By default, all methods of ReflectionSearch match the exact type passed as an argument (or match derived types if ReflectionSearchOptions.IncludeDerivedTypes is specified). However, there are cases where you want a larger choice.
Consider the following code:
class Bar {}
class Foo : ICollection<Bar>
{
Bar[] array;
void ICollection<Bar> FooBar()
{
return new List<Bar>();
}
}
var derivedTypes = ReflectionSearch.GetDerivedTypes( typeof(Bar), options );
var members = ReflectionSearch.GetMembersOfType( typeof(Bar), options );
var usages = ReflectionSearch.GetMethodsUsingDeclaration( typeof(Bar), options );
With the default value of options, the three methods return an empty set. Indeed, there is no type derived from Bar, no field or return value of type Bar, and no instruction using Bar. However, if you include the option ReflectionSearchOptions.IncludeTypeElement, you will get:
- as derived type: Foo,
- as members: the field array and the return value of FooBar, and
- as usages: the method FooBar.
The IncludeTypeElement option asks PostSharp to index at a much deeper level. Pay attention: this comes at a performance cost. PostSharp does not use this feature internally, so the first time you use the option, it will have to scan all type signatures and build an index.
Summary
The new ReflectionSearch class allows you to make complex queries over System.Reflection. Most of the times, the features are already used internally by PostSharp, so they come at minimal performance cost.
The ReflectionSearch class is accessible at build time only. The primary use case is to write more powerful logic for IAspectProvider or MethodPointcut. The second use case is to validate your code. And this is for another blog post.
Happy PostSharping!
-gael
One of the breaking innovations of PostSharp 1.0 in 2005 was the ability to execute aspect code at build time. Whereas other AOP frameworks developed complex declarative syntaxes to express “pointcuts”, PostSharp just allowed you to use System.Reflection to select the target of aspects. LINQ made it even easier to express complex code queries.
Yet, some queries were difficult to achieve just with System.Reflection. For instance, it is fairly difficult and inefficient to get all properties annotated with the [DataMember] custom attribute: you have to enumerate all types (and do a recursion on nested types), get all properties of these types, and check the presence of this custom attribute on all properties. Wow. If we were querying a relational database, we would probably create an index on the Type column of the CustomAttribute table. Well, the analogy is not so stupid. Metadata in a .NET assembly is in fact a relational database, and there is a CustomAttribute table. What is more, PostSharp already builds these indexes internally. So why not making them available to user code? This was the principal idea behind this new feature: reflection search.
Reflection Search is simply this: a set of methods making it easier and more efficient to query metadata of the assembly being processed by PostSharp. Therefore, this API is available only at build time. It is not intended to “improve” System.Reflection for run-time use. That said, if you need to do some complex reflection query at runtime, it’s better to do it at build time using PostSharp, serialize the result and store it as a managed resource.
The API is exposed in the class PostSharp.Reflection.ReflectionSearch. Note that the feature is available in the Professional Edition only.
Enumerating Custom Attributes
Custom attributes of a type
The first set of methods you will find on the ReflectionSearch class allow you to enumerate custom attributes. GetCustomAttributesOnType retrieve all custom attributes of a given type in the current assembly (remember that this works only at build time, so the notion of “current assembly” refers to the assembly being processed by PostSharp). The second overload takes an additional parameter ReflectionSearchOptions. If you specify the value IncludeDerivedTypes, you will get all custom attributes of the specified type or any type derived from it.
There is no example in the system class library to illustrate a hierarchy of custom attributes, so let's create our own:
// A hierarchy of custom attributes.
class FruitAttribute : Attribute {}
class AppleAttribute : FruitAttribute {}
class OrangeAttribute : FruitAttribute {}
class VegetableAttribute : Attribute {}
class TomatoAttribute : VegetableAttribute {}
// Using the custom attributes
[Fruit]
class Team
{
[Apple]
Dude tom;
[Orange]
Dude jerry;
[Tomato]
Dude donald;
}
// Enumerating the custom attributes (build-time)
class MyAspect : AssemblyLevelAspect, IAspectProvider
{
public IEnumerable ProvideAspects(object target)
{
CustomAttributeInstances[] attributes = ReflectionSearch.GetCustomAttributesOfType(
typeof(FruitAttribute), ReflectionSearchOptions.IncludeDerivedTypes );
// TODO: Do something with that.
}
}
The method GetCustomAttributesType will return three instances of the custom attribute: (FruitAttribute, Team), (AppleAttribute ,Team.tom) and (OrangeAttribute, Team.jerry).
If you wonder why the method does not return a tomato: it is by decision of the U.S. Supreme Court, which, on May 10, 1893, declared that tomatoes are vegetables because – although they are classified as fruits in botany – they are usually served with dinner and not dessert. As far as I am concerned, I still don’t eat tomatoes for dessert, the only thing that changed is that we are now much more accustomed to bureaucracy (and today’s tomatoes have better resistance to pressure because the species were selected to accommodate mechanical picking, and long storage and transportation – so arguably we now eat more tomatoes than in the XIXth century, but they are probably less tasty).
Anyway.
Note that the method returns a CustomAttributeInstance, which includes information about the custom attribute and the element of code to which it is applied.
Multicast attributes
Keep in mind that ReflectionSearch reflects the internal repository of custom attributes. This repository does not only include custom attributes that you add to your code manually. It also includes all “pseudo” custom attributes added by different components of PostSharp, starting with multicast attributes (remember that aspect class typically derive from MulticastAttribute).
When you use multicast custom attributes in your code, PostSharp executes the multicasting algorithm to populate the internal repository. The repository then contains the concrete instances of the custom attribute.
Let’s take an example:
[MulticastAttributeUsage( MulticastTargets.Method )]
class FooAttribute : MulticastAttribute {}
[Foo]
class MyClass
{
void MyMethod() {}
}
In that case the repository contains an instance of FooAttribute on MyMethod but not on MyClass. The normal reflection API would give the opposite result (at least when it is run inside PostSharp or before PostSharp), because it is not aware of multicast semantics.
Since you can’t use normal reflection to retrieve multicast custom attributes, ReflectionSearch offers a second method: GetCustomAttributesOnTarget. As the name suggests, it returns the list of all custom attributes on a given element of code.
Summary
The high-level library PostSharp.dll now offers access to the internal repository of custom attributes. Since this repository, and all indexes, are used by PostSharp itself, using this new feature has minimal impact on build time. Now, you have a way to query custom attributes by type without killing performance. And you are finally able to query attributes that were created by PostSharp as the result of the multicasting algorithm.
But that’s not all ReflectionSearch has to offer. Next in line: browse relationships between elements of code. But this is for another day.
Happy PostSharping!
-gael
Yesterday we introduced members into our target class. Today we’re going to go the other way and introduce members from our target class into our aspect for consumption by the process of importing.
Importing Members
There are cases when you will need to bring in a field or method from a target class so that you have access to it from within the aspect. What we’re doing is making assumptions about the target. It’s important to document what the aspect is assuming about the target code. Another approach instead of importing, is to cast the instance to a interface (or type) known to be implemented on the target.
To import, we use the ImportMember attribute. ImportMember allows us to import fields, properties, methods and events. As an example, let’s look at some sample code.
public interface IIdentifiable
{
Guid ID { get; }
void PrintID();
}
public class ExampleClass : IIdentifiable
{
public Guid ID { get; private set; } //Satisfy the interface
public ExampleClass()
{
this.ID = Guid.NewGuid();
}
[ImportExampleAspect]
public void DoWork()
{
Console.WriteLine("Doing work");
}
public void PrintID()
{
Console.WriteLine(this.ID.ToString());
}
}
[Serializable]
public class ImportExampleAspect : OnMethodBoundaryAspect, IInstanceScopedAspect
{
[ImportMember("PrintID", IsRequired = true)]
public Action PrintID;
[ImportMember("ID")]
public Property IDFromClass;
public override void OnEntry(MethodExecutionArgs args)
{
Guid value = IDFromClass.Get();
PrintID();
Console.WriteLine("Value from IIdentifiable.ID is {0}", value.ToString());
}
public object CreateInstance(AdviceArgs adviceArgs)
{
return this.MemberwiseClone();
}
public void RuntimeInitializeInstance() {}
}
We have an example class that implements a simple interface and an aspect based on OnMethodBoundaryAspect. When you run this example code, the value of the ‘ID’ property will be written to the console.
178411e1-c79b-43fc-9aa0-ff309425abaa
Value from IIdentifiable.ID is 178411e1-c79b-43fc-9aa0-ff309425abaa
Doing work
Notice that our aspect implements the IInstanceScopedAspect interface. ImportMember cannot import static fields so we have to define our aspect as instance scoped. For more information on the life time and scope of aspects, refer to Day 9 and Day 10 of the PostSharp Principals series.
We expect the target to implement the interface which has a PrintID method defined. So we setup a public Action and we call it PrintID. We decorate our Action with the ImportMember attribute passing in the target’s member name which is ‘PrintID’.
Next we defined the IDFromClass public field of type Property<Guid>. This class has two properties: Get and Set. At runtime, the Get and Set properties will contain a delegate to the getter and setter of this property. In order to import an event, we need a public field of type Event<EventArgs>, which has two delegate properties: Add and Remove.
We do not have to use the same name as the target’s member name, so instead of ‘ID’ we are using ‘IDFromClass’.Inside of the OnEntry method we get the value of ID from our target class using the Get delegate provided by IDFromClass, make a call to the PrintID method and then write out the value to the console.
Note: ImportMember will work on members of any visibility, even private.
ImportMemberAttribute
There should be no surprise when I say that we can control the behavior of the import using the attribute’s parameters.
IsRequired
Specifies whether the member being imported is required or not. If true, a compiler error will occur if the member is not present in the target class. If false, the reference to the member will be null if it does not exist in the target.
Order
Order takes an enumeration from PostSharp.Aspects.Advices.ImportMemberOrder to determine when the importing will occur.
· AfterIntroductions (default) – Importing will occur only after the aspect has introduced its own members. This is useful if the imported member is virtual and we need to import the last override.BeforeIntroductions – Importing will occur before any members are introduced by the aspect. This gives you a chance to get a reference to the original member before any overriding occurs. This is similar to calling an overridden method by the “base” keyword.
· Default – The default is AfterIntroductions.
There can be any number of aspects on the same class that override the same method. Importing before the introductions allows you to call the next node in the chain of overwriting.
Better Example
Let’s have a look at a more in-depth example using both importing and introduction, the MakeDirtyOnChange aspect. When working with workspaces or MDI (multi-document interface) applications, many “documents” can be open at one time. There is a need to know when one or more of the “documents” have been modified so that you can inform the user visually and also to know which items need to be updated in the data store. Think about working with five C# files in Visual Studio. When a change is made to a file, you will see an asterisk next to the filename on the file’s tab. This is the indication that a change has been made and not yet saved.
Implementing the IsDirty pattern is similar to the INotifyPropertyChange, the code to wire up the change notification is redundant and tedious. Let’s check out our interface and test class before diving into our aspect
public interface IsDirty
{
bool IsDirty { get; }
event EventHandler WasMadeDirty;
void ResetDirtyState();
ReadOnlyCollection DirtyProperties { get; }
}
public class DirtyEventArgs : EventArgs
{
public string DirtyProperty { get; private set; }
public DirtyEventArgs(string dirtyProperty)
{
this.DirtyProperty = dirtyProperty;
}
}
[MakeDirtyOnChange]
public class Document
{
private Guid _docId = Guid.NewGuid();
public Guid DocId { get { return _docId; } }
public string Title { get; set; }
public string Author { get; set; }
public string Content { get; set; }
}
Our IDirty interface has a few requirements that we can and will use to determine if an item has changes. We specify a custom EventArgs class that allows us to provide details about the changes made to the item when invoking the WasMadeDirty event. Our Document class is a clean model that knows nothing about the IsDirty interface.
[Serializable]
[IntroduceInterface(typeof(IsDirty), OverrideAction = InterfaceOverrideAction.Ignore)]
public class MakeDirtyOnChange : InstanceLevelAspect, IsDirty
{
[OnLocationSetValueAdvice, MulticastPointcut(Targets=MulticastTargets.Property)]
public void OnValueChanged(LocationInterceptionArgs args)
{
MakeDirty(args.LocationName);
}
private bool _isDirty;
private List _dirtyProperties;
[ImportMember("SetDirty")]
public Action MakeDirty;
[IntroduceMember(IsVirtual=true, OverrideAction=MemberOverrideAction.Ignore)]
public void SetDirty(string property)
{
_isDirty = true;
if (WasMadeDirty != null)
{
WasMadeDirty.Invoke(this.Instance, new DirtyEventArgs(property));
}
}
#region IsDirty Members
[IntroduceMember(OverrideAction = MemberOverrideAction.Ignore)]
public bool IsDirty { get { return _isDirty; } }
[IntroduceMember(OverrideAction = MemberOverrideAction.Ignore)]
public ReadOnlyCollection DirtyProperties { get { return _dirtyProperties.AsReadOnly(); } }
[IntroduceMember(OverrideAction = MemberOverrideAction.Ignore)]
public event EventHandler WasMadeDirty;
[IntroduceMember(IsVirtual=true, OverrideAction=MemberOverrideAction.Ignore)]
public void ResetDirtyState()
{
_isDirty = false;
_dirtyProperties.Clear();
}
#endregion
public override void RuntimeInitializeInstance()
{
_isDirty = false;
_dirtyProperties = new List();
}
}
We decorate our aspect with the IntroduceInterface attribute, specifying the IsDirty interface. Our aspect implements the IsDirty interface to satisfy the requirements and then we introduce those members to the target.
We setup a location interception using OnLocationSetValueAdvice attribute and specify the target is MulticastPointcut.Property ([MARKER, Advice link]). When a property is changed, we’re going to invoke the MakeDirty method which we tell PostSharp to import from the target’s “SetDirty” method, if it has one. We’re using the defaults for ImportMember which means the import will happen after our members are introduced. Since we’re introducing our own SetDirty method, MakeDirty will contain our SetDirty implementation if the target class did not already have its own implementation.
Since our aspect derives from InstanceLevelAspect we override the RuntimeInitializeInstance method and use it to initialize our private members to their default states.
We can use the following code to try out the aspect
class Program
{
private static List _changedDocuments = new List();
private static List _openDocuments = new List();
static void Main(string[] args)
{
for (int i = 0; i < 5; i++)
{
Document doc = new Document();
Post.Cast(doc).WasMadeDirty
+= new EventHandler(doc_WasMadeDirty);
_openDocuments.Add(doc);
}
_openDocuments[0].Author = "Dustin Davis";
_openDocuments[0].Title = "PostSharp Principals - Day 1";
_openDocuments[2].Author = "Dustin Davis";
_openDocuments[2].Title = "PostSharp Principals - Day 3";
Console.ReadKey();
}
static void doc_WasMadeDirty(object sender, DirtyEventArgs e)
{
Document doc = (Document)sender;
if (_changedDocuments.Any(c => c.Equals(doc.DocId)))
{
return;
}
_changedDocuments.Add(doc.DocId);
Console.WriteLine("Document {0} was modified.", doc.DocId);
}
}
The code is pretty straight forward. We create five documents and then add them to our open documents collection. Finally we make changes to two documents. When we run the code, we see the following results
Document acbd247a-e742-499e-b27c-ee028e8e6789 was modified.
Document 9ded92cc-008f-48bc-b1b5-e1b0b967e42d was modified.
But wait, how are we handling the WasMadeDirty event? I’m glad you asked.
Post.Cast<>()
Post.Cast<>() allows us to cast an instance of a type to another type at design time. For example, our Document class doesn’t implement the IsDirty interface so we can’t access the IsDirty specific members unless we casting. We use the generic Cast<SourceType, TargetType>(SourceType Instance) method to give us back an instance of TargetType.
It’s basically nothing more than regular casting, but the difference is when you use Post.Cast<>() you receive compile-time errors if the cast cannot take place. The obvious benefit is that you know right away that the cast fails instead of at run time, potentially introducing bugs.
In the final result, the call to Post.Cast<>() is replaced with an actual cast.
Conclusion
Previous aspects we looked at have been pretty disconnected from the targets. Being able to introduce and import members gives us a connection and increased flexibility. Being able to automatically introduce interfaces and boiler plate code that is sometimes only consumed at run time frees us and keeps our code clean.
Today we continue our descent into the depths of the PostSharp framework to expose even greater powers to utilize in our projects.
Introduction
No, not the “Hello, my name is Dustin!” kind of introduction, but the “injection” type. What does that mean? PostSharp gives us the power to implement an interface on a class…at build time. We can also add (introduce) members to that class such as fields/properties, events and methods too. These members are injected at build time and are available at run time.
Why would you want to do this? As in most cases when applying aspect-oriented programming, you would use this to implement required interfaces that are little more than boilerplate code. One of the most popular examples of interface introduction is the NotifyPropertyChanged aspect which automatically introduces the INotifyPropertyChanged interface and required members. Anyone who has worked with WPF and the MVVM pattern would love to not have to write all of that scaffolding code just to get change notification. Since that aspect uses features we have not yet covered, we will not cover it today. If you’re feeling adventurous, you can check it out here.
Member Introduction
Member introduction allows us to add properties, events and methods to a class. Let’s start off by creating an aspect to introduce a property and a method.
[Serializable]
public class IntroduceAspect : InstanceLevelAspect
{
[IntroduceMember]
public int ID { get; set; }
[IntroduceMember]
public void SomeMethod(int param1)
{
Console.WriteLine("Inside of introduced method");
}
}
And now our target class
[IntroduceAspect]
public class TargetClass
{
}
You might be laughing at our test class, but don’t worry, our aspect will do the work for us. When we look at the compiled assembly with ILSpy, we see that instead of a blank class we have a few more members than we started with, including the members we wanted to introduce.
Amongst the aspect related code, we have our ID property and our SomeMethod method. Notice that the getter and setter of ID are delegated to our aspect and so does our method. This is important to keep in mind because when implementing members, they must be marked as public inside of the aspect (because our target class has to access them). However, if you happen to forget, PostSharp will remind you with a compiler error
But what happens if you don’t want the introduced members to be public in the target class? Have no fear, PostSharp thought of that too. Let’s have a look at the IntroduceMember attribute.
IntroduceMember attribute
By default, using IntroduceMember by itself will use public visibility and will cause compiler errors if a member with the same signature is already part of the class. We can control the behavior of how the member is implemented by changing the following parameters.
Visibility
By default, PostSharp will introduce the member to the target class with public visibility. We can specify one of the enumerations from PostSharp.Reflection.Visibility to control what visibility the member will have in the target class. Available values are
· Public (Default) – Is publically available.
· Family – Is available to the class and any derived classes. Same as protected.
· Assembly – Is publicly available within the assembly. Same as internal.
· FamilyOrAssembly – Is available to the class and any derived classes, but only within the assembly. Same as protected internal.
· FamilyAndAssembly – Protected types inside the assembly. There is no C# equivalent.
· Private – Only visible to the class.
OverrideAction
There is a chance that the target class already has a member with the same signature. By default, there will be a compiler error if this scenario is encountered. To change the behavior, we can provide one of the enumerations from PostSharp.Aspects.Advices.MemberOverrideAction.
· Default – Fails with a compiler error.
· Fail – Fails with a compiler error.
· Ignore – Continues on, without trying to introduce the member.
· OverrideOrFail – Tries to override the member with our own implementation. If the existing member is defined in a base class and is sealed or non-virtual, it will fail with a compiler error.
· OverrideOrIgnore – Tries to override the member with our own implementation. If the existing member is defined in a base class and is sealed or non-virtual, it will ignore the member introduction and continue on.
IsVirtual
If you would like to introduced member to be virtual (overridable in derived classes) then you can set IsVirtual to true. The member signature in the base class will be marked as virtual.
CopyCustomAttributesAttribute
Sometimes members need to be decorated with attributes. An example of this would be decorating members of a DataContract with DataMember. However, when introducing members from an aspect, any attributes applied to the member in the aspect will not be introduced along with the member in the target. We can use CopyCustomAttributes attribute in addition to the IntroduceMember attribute to introduce the attributes along with the member. Let’s look at an example.
[Serializable]
public class IntroduceAspect : TypeLevelAspect
{
[IntroduceMember]
[DataMember(IsRequired=true)]
public int ID { get; set; }
}
[IntroduceAspect]
[DataContract]
public class TargetClass
{
[DataMember]
public string FirstName { get; set; }
}
Our aspect is introducing a member, ID, which is decorated with DataMember. Let’s look at the result in ILSpy
The DataMember attribute is not present on ID. Let’s update the aspect to use CopyCustomAttributes.
[Serializable]
public class IntroduceAspect : TypeLevelAspect
{
[IntroduceMember, CopyCustomAttributes(typeof(DataMemberAttribute),
OverrideAction = CustomAttributeOverrideAction.MergeReplaceProperty)]
[DataMember(IsRequired=true)]
public int ID { get; set; }
}
In the constructor for CopyCustomAttributes we pass in the base type for the desired attribute and then we set the override action with a value from the CustomAttributeOverrideAction enumeration. When we look at the end result in ILSpy, we see that the attribute was introduced along with the member.
CustomAttributeOverrideAction
CustomAttributeOverrideAction is an enum that lets us tell PostSharp how to handle a situation when an attribute of the same type already exists on the target member.
· Default – Fails with a compile time error.
· Fail – Fails with a compile time error.
· Ignore – Ignores the attribute introduction and does not generate an error.
· Add – Adds the attribute as defined, even if it already exists on the target. This could cause duplicate attributes on the target.
· MergeAddProperty – Combines the existing attribute with the attribute being introduced. Any properties defined by the existing attribute will remain. No override will occur. Any properties defined by the introduced attribute will be added to the existing attribute.
· MergeReplaceProperty – Same as MergeAddProperty except that any properties defined by the existing attribute will overridden by the introduced attribute.
Interface Introduction
When introducing an interface via an aspect, the interface must be implemented on the aspect itself. The type will expose the interface at run time, but the aspect actually implements it. Let’s have a look at our interface:
public interface IFriendlyName
{
string Name { get; set; }
void PrintName();
}
And now our aspect:
[Serializable]
[IntroduceInterface(typeof(IFriendlyName))]
public class IntroduceAspect : InstanceLevelAspect, IFriendlyName
{
#region IFriendlyName Members
public string Name { get; set; }
public void PrintName()
{
Console.WriteLine(this.Name);
}
#endregion
}
Our test class remains the same, empty
[IntroduceAspect]
public class TestClass
{
}
When we look at the compiled result we see our interface has been implemented
Notice that we didn’t use the IntroduceMember attribute on the interface members. Also notice that the resulting implementations of the interface members are private. To make the interface members public we have to apply the IntroduceMember attribute to the members
[Serializable]
[IntroduceInterface(typeof(IFriendlyName))]
public class IntroduceAspect : InstanceLevelAspect, IFriendlyName
{
#region IFriendlyName Members
[IntroduceMember]
public string Name { get; set; }
[IntroduceMember]
public void PrintName()
{
Console.WriteLine(this.Name);
}
#endregion
}
And now the compiled result shows two implementations of our members
Looking at the PrintName method, the explicit interface implementation is private, but we’ve introduced a public version which the interface method calls.
IntroduceInterface attribute
To tell PostSharp that we want to introduce an interface, we decorate the aspect with the IntroduceInterface attribute. To tell PostSharp which interface to implement, we pass in a type using typeof(IFriendlyName). Just like the IntroduceMember attribute, there are parameters to control the behavior of the introduction.
· OverrideAction – Exactly the same as IntroduceMember. Determines what to do when the target already implements the interface. Default is to fail with a compiler error.
· IsProtected – If set to true, the interface is not directly implemented by the type. Instead, the type exposes the interface through the IProtectedInterface<T>. Since protected interfaces are considered obsolete, you should leave this as false (default).
· AncestorOverrideAction – Defines the behavior of the introduction when and ancestor of the interface is already applied to the target class. See example below. Available enumerations in the PostSharp.Aspects.Advices.InterfaceOverrideAction are Default (Fail), Fail and Ignore.
Extended Example
Let’s finish up with a bit more in-depth example using some of the behavior parameters.
public interface IIdentifiable
{
Guid ID { get; set; }
}
public interface IFriendlyName : IIdentifiable
{
string Name { get; set; }
void PrintName();
}
[IntroduceAspect]
public class TargetClass : IIdentifiable
{
#region IFriendlyNameBase Members
public Guid ID { get; set; }
#endregion
string Name { get; set; }
public void PrintName()
{
throw new NotImplementedException();
}
}
[Serializable]
[IntroduceInterface(typeof(IFriendlyName),
AncestorOverrideAction=InterfaceOverrideAction.Ignore)]
public class IntroduceAspect : InstanceLevelAspect, IFriendlyName
{
#region IFriendlyName Members
[IntroduceMember(OverrideAction=MemberOverrideAction.Ignore)]
public Guid ID { get; set; }
[IntroduceMember(OverrideAction=MemberOverrideAction.Ignore)]
public string Name { get; set; }
[IntroduceMember(OverrideAction=MemberOverrideAction.OverrideOrFail)]
public void PrintName()
{
Console.WriteLine(this.Name);
}
#endregion
}
We define two interfaces. IFriendlyName implements IIdentifiable. Our test class implements IIdentifiable and also has a PrintName method which throws an exception. Our aspect specifies the introduction of IFriendlyName and also implements the required interface members. We specify that we should ignore any implementation of an ancestor (IIdentifiable) of the introduced interface (IFriendlyName). We also specify that we want to ignore the member introduction on the two properties if they exist in the target class. We mark PrintName with the OverrideOrFail because we want to force our own implementation of the PrintName method. The end result looks like this
First take a look at the PrintName method. Instead of the original method body, which threw an exception, we see that there is a call to our aspect which invokes our implementation of that method.
Next we see that both interfaces are implemented, but we only have get/set methods for the Name property, not the ID property. This is because PostSharp ignored the implementation of IIdentifiable since it was already implemented on the target class. If we remove InterfaceOverride.Ignore from the IntroduceInterface attribute, we would get a compiler error.
If we removed the implementation of IIdentifiable from our test class, we would see get/set methods for ID in the compiled results.
Conclusion
Today we covered some good ground on introducing members and interfaces along with some of the nuances that you have to be aware of. Tomorrow we’ll continue with importing members and accessing introduced members at compile time.
I'm pleased to announce that the third and final Community Technical Preview of PostSharp 2.1 is now available for download on our website and as a NuGet package.
The new CTP is principally a stabilization release, with a large number of bug fixes (including all hot fixes from the 2.0 branch) and some corrections in the new API design. This CTP is more stable than the latest builds on the 2.0 branch, but this has of course to be validated by the community.
What’s New in PostSharp 2.1 CTP 3
- A streamlined experience for first-time users, including a redesigned PostSharp HQ application and new Learn PostSharp tool window in Visual Studio.
- PostSharp does not enter the evaluation mode automatically. Instead, a dialog box will ask you to explicitly enter the evaluation mode. This adds a little friction for those evaluating PostSharp but should help to clarify any licensing questions.
- Complete class reference documentation (the conceptual documentation is still not updated).
- Support for Windows Phone 7.1 Beta.
- Support for Silverlight and Windows Phone in the NuGet package.
What’s New in PostSharp 2.1 CTP 2
- Architecture constraints.
- Extended build-time reflection API.
- Support for obfuscation.
- Support for Silverlight 5.
- License server.
What's New in PostSharp 2.1 CTP 1
- Improvements in build-time performance.
What’s Next?
Now that the design is nearly finalized, we should finally blog about and document new features. The release candidate will be published as soon as the documentation is complete (the current release has been tested to release candidate-level quality).
Update Today
If your application is not going to production in the next few months, now is a great time to download PostSharp 2.1 and to upgrade your current application in a feature branch. Doing so will help us to validate the new version and move the final release date forward. Please report all issues on the support forum. Thanks!
Happy PostSharping!
-gael
When it comes to building complex aspects to solve a specific problem or implement a pattern, the base classes such as OnMethodBoundaryAspect and LocationInterceptionAspect aren’t always up-to the job. We covered the IAspectProvider yesterday, which allows us to dynamically tell PostSharp which aspects to apply to a target at compile time. Today we’re going to build complex aspects that encapsulate multiple transformations in a single aspect.
Advices and Pointcuts
Before we continue, we should cover some vocabulary. We’ve avoided the use of these terms until now to avoid any confusion.
· Advice – “An advice is anything that adds a behavior or a structural element to an element of code. For instance, introducing a method into a class, intercepting a property setter, or catching exceptions, are advices.”
· Pointcut – “A pointcut is a function returning a set of elements of code to which advices apply. For instance, a function returning the set of properties annotated with the custom attribute DataMember is a pointcut.”
When we override the OnEntry method when building an aspect based on OnMethodBoundaryAspect, we’re providing the advice to implement. By default the pointcut would be all methods in a class, if the class was decorated with our aspect based on the OnMethodBoundaryAspect base class.
PostSharp provides us with a set of attributes for declaring advice and pointcuts in any combination under a single aspect. Let’s have a look at an example to give a better picture
[Serializable]
public class ComplexAspect : TypeLevelAspect
{
private int MethodCounter = 0; //Shared between all advices
[OnMethodInvokeAdvice, MulticastPointcut(Targets =
MulticastTargets.Method, MemberName = "DoSomethingElse")]
public void OnInvoke(MethodInterceptionArgs args)
{
Console.WriteLine("Before method {0}. MethodCounter = {1}",
args.Method.Name, this.MethodCounter);
args.Proceed();
Console.WriteLine("After method {0}. MethodCounter = {1}",
args.Method.Name, this.MethodCounter);
}
[OnMethodEntryAdvice, MulticastPointcut(Targets = MulticastTargets.Method)]
public void OnEntry(MethodExecutionArgs args)
{
MethodCounter++;
Console.WriteLine("Entering {0}. MethodCounter = {1}",
args.Method.Name, this.MethodCounter);
}
[OnMethodExitAdvice(Master = "OnEntry")]
public void OnExit(MethodExecutionArgs args)
{
Console.WriteLine("Exiting {0}. MethodCounter = {1}",
args.Method.Name, this.MethodCounter);
}
[OnLocationSetValueAdvice, MulticastPointcut(Targets = MulticastTargets.Property)]
public void OnPropertySet(LocationInterceptionArgs args)
{
MethodCounter++;
Console.WriteLine("Setting property: {0} = {1}. MethodCounter = {2}",
args.LocationName, args.Value, this.MethodCounter);
}
}
Our aspect derives from TypeLevelAspect, not one of the base classes we’ve been using. We have four methods in our aspect and each method is decorated with an advice attribute along with a pointcut attribute.
The OnEntry method is decorated with the OnMethodEntryAdvice attribute which has the same semantics of overriding the OnEntry method in an OnMethodBoundaryAspect. The MulticastPointcut attribute is used and passed the MulticastTargets.Method flag to let PostSharp know that we want to apply this advice to methods in general.
Because we’re using TypeLevelAspect instead of OnMethodBoundaryAspect, we are able to share state between advices. When using OnMethodBoundaryAspect, an instance of our aspect is created for each target. For example, Method1 would have its own copy of our aspect and Method2 would have its own copy. Using TypeLevelAspect to implement our advices changes that behavior; we have a single instance of our aspect that is used for each target which means that the advices get to share state. We’re going to demonstrate this using the MethodCounter field to increment on each method entry and display its value throughout the other advices.
Notice the OnExit advice isn’t specifying a pointcut, but instead is passing in a value to the Master parameter on the OnMethodExitAdvice attribute. We’re defining the master advice method, which means we’re grouping the advices on the same “layer”. OnExit will be a slave method and will inherit the pointcut selectors from OnEntry since only master advice methods can define pointcuts. We only do this for advices of the same category of transformations. For example, you wouldn’t define OnEntry as the master advice method for OnPropertySet because they perform different transformations. We’ll cover grouping on another day.
The other methods in our aspect are the same, just using different advices. The OnInvoke method however has a different pointcut setup. We add an additional parameter, MemberName, giving it a value of “DoSomethingElse” which tells PostSharp to only apply the advice to methods matching “DoSomethingElse”.
Let’s run our example code and look at the result
class Program
{
static void Main(string[] args)
{
ExampleClass ec = new ExampleClass();
ec.DoSomething();
ec.DoSomethingElse();
ec.FirstName = "John";
ec.LastName = "Smith";
Console.ReadKey();
}
}
[ComplexAspect]
public class ExampleClass
{
public string FirstName { get; set; }
public string LastName { get; set; }
public void DoSomething()
{
Console.WriteLine("Doing something");
}
public void DoSomethingElse()
{
Console.WriteLine("Doing something else");
}
}
Entering DoSomething. MethodCounter = 1
Doing something
Exiting DoSomething. MethodCounter = 1
Before method DoSomethingElse. MethodCounter = 1
Entering DoSomethingElse. MethodCounter = 2
Doing something else
Exiting DoSomethingElse. MethodCounter = 2
After method DoSomethingElse. MethodCounter = 2
Entering set_FirstName. MethodCounter = 3
Setting property: FirstName = John. MethodCounter = 4
Exiting set_FirstName. MethodCounter = 4
Entering set_LastName. MethodCounter = 5
Setting property: LastName = Smith. MethodCounter = 6
Exiting set_LastName. MethodCounter = 6
In a single aspect, we have implemented advice and pointcuts that we normally would have written three different aspects using the base classes. In addition, we were able to share state between those advices. Examine the output of the MethodCounter. It’s incrementing as we continue along with the execution. If we had three individual aspects the provided equivalent advices, the output would look more like
Entering DoSomething. MethodCounter = 1
Doing something
Exiting DoSomething. MethodCounter = 1
Before method DoSomethingElse. MethodCounter = 0
Entering DoSomethingElse. MethodCounter = 1
Doing something else
Exiting DoSomethingElse. MethodCounter = 1
After method DoSomethingElse. MethodCounter = 0
Entering set_FirstName. MethodCounter = 1
Setting property: FirstName = John. MethodCounter = 1
Exiting set_FirstName. MethodCounter = 1
Entering set_LastName. MethodCounter = 1
Setting property: LastName = Smith. MethodCounter = 1
Exiting set_LastName. MethodCounter = 1
Advices
There are a few advice attributes that you can use. Each advice attribute has a corresponding simple aspect base class and behaves in the same way.
|
OnMethodEntryAdvice
OnMethodSuccessAdvice
OnMethodExceptionAdvice
OnMethodExitAdvice
|
These advices are the equivalent to the advices in the OnMethodBoundaryAspect base class
|
|
OnMethodInvokeAdvice
|
These advices are the equivalent to the advices in the MethodInterceptionAspect base class
|
|
OnLocationGetValueAdvice
OnLocationSetValueAdvice
|
These advices are the equivalent to the advices in the LocationInterceptionAspect base class
|
|
OnEventAddHandlerAdvice
OnEventRemoveHandlerAdvice
OnEventInvokeHandlerAdvice
|
These advices are the equivalent to the advices in the EventInterceptionAspect base class
|
|
IntroduceMember
|
Introduce a method, property or event to the target class.
|
|
IntroduceInterface
|
Introduce a method to the target class.
|
When applying advice to a method, the method must be public and have the same signature as the corresponding base class advice signature. For example, in order to apply OnLocationGetValueAdvice on a method, the method must be public and have a single LocationInterceptionArgs parameter with no return value (void).
Pointcuts
A pointcut has to be defined in order to tell PostSharp where to apply the advice. You can think of pointcuts as expressions that return a set of elements of code. These elements of code must be compatible with the type of advice (for instance, do not try to add an OnLocationGetValue advice to a field). Remember that you can only add advices to code that belong to the target of the aspect. So if the aspect has been applied to a type, you can only add advices to members of this type, or to the type itself.
|
MulticastPointcut
|
A declarative pointcut that works similarly to MulticastAttribute.
|
|
MethodPointcut
|
An imperative pointcut: the pointcut is a method that returns an enumeration of elements of code. The method can be implemented in C#, for instance using Linq.
|
|
SelfPointcut
|
A pointcut that evaluates to the target of the aspect.
|
MulticastPointcut
MemberName
MemberName takes an expression (static name, wildcard or regular expression) to specify targets.
Targets
Targets can be set to a combination of MulticastTargets flags. For example, MulticastTargets.Method | MulticastTargets.Property to specify the targets will be methods and properties.
Attributes
Attributes is how we define the scope and visibility of the intended targets and can be set to a combination of MulticastAttributes flags. For example, we can target members that are private and static by using MulticastAttributes.Private | MulticastAttributes.Static.
MethodPointcut
MethodPointcut allows us to pass in the name of a method that PostSharp can use to get a list of targets. Let’s look at an example
[Serializable]
public class ExampleAspect : TypeLevelAspect
{
public IEnumerable FindTargetMethods(Type target)
{
IEnumerable targets = target.GetMethods()
.Where(c => c.Name.Contains("Something"));
return targets;
}
[OnMethodEntryAdvice, MethodPointcut("FindTargetMethods")]
public void OnEntry(MethodExecutionArgs args)
{
Console.WriteLine("Entering method: " + args.Method.Name);
}
}
Our aspect has only one advice that we want to implement, OnMethodEntryAdvice. We use the MethodPointcut attribute and pass it “FindTargetMethods” which is the name of the method we’ve setup to determine which methods will be targets. The method that is going to return the targets has to have a specific signature. PostSharp documentation defines this signature as
IEnumerable SelectCodeElements(AspectTargetType target)
AdviceTargetType will be replaced with either object or a reflection type representing the targets of the advice. For example, MethodInfo when the advice targets are methods and PropertyInfo when the targets are Properties.
AspectTargetType will be replaced with either object or a reflection type corresponding to the targets of the aspect. For example, Type for AssemblyLevelAspect, TypeLevelAspect, InstanceLevelAspect and MethodInfo for MethodLevelAspect.
If the signature is not valid for the type of aspect then PostSharp will produce a compiler error. When we apply the aspect to our example class and run the application, we see that PostSharp has applied the advice to both of the class methods.
public class Program
{
static void Main(string[] args)
{
ExampleClass ec = new ExampleClass();
ec.DoSomething();
ec.DoSomething1();
Console.ReadKey();
}
}
[ExampleAspect]
public class ExampleClass
{
public int ID { get; set; }
public string First { get; set; }
public string Last { get; set; }
public void DoSomething()
{
this.First = "John";
this.Last = "Smith";
}
public void DoSomething1()
{
Console.WriteLine("Did something");
}
}
The output is
Entering method: DoSomething
Entering method: DoSomething1
Did something
SelfPointcut
There is a special attribute that you can use instead of MulticastPointcut, the SelfPointcut attribute. SelfPointcut tells PostSharp to select the exact aspect target. For example, if we have the following aspect
[Serializable]
public class ExampleAspect: MethodLevelAspect
{
[OnMethodEntryAdvice, SelfPointcut]
public void OnEntry(MethodExecutionArgs args)
{
Console.WriteLine("Entering " + args.Method.Name);
}
}
When applied to a method directly, the SelfPointcut will use that exact method as the target. If applied to a class, all methods in the class will get the advice because all of the methods would be the intended target. If the aspect was applied at the assembly level with specific targets configured, then PostSharp will use those exact targets. SelfPointcut is a way to defer the selection of pointcuts to a higher mechanism.
Benefits over base classes
Base classes encapsulate a single transformation, which means if you want to apply multiple transformations, you would need to build just as many aspects. For example, if you wanted to marshal a method invocation to a different thread and log exceptions that occur in that method, you would need to build two independent aspects and apply both to the target.
Sharing state between advices in multiple independent aspects requires convoluted mechanics. Building complex aspects using advices and pointcuts provides the benefit of sharing state between advices.
Conclusion
By now you should have a clear understanding of how to build complex aspects using the tools provided by PostSharp. The term “complex” shouldn’t be a deterrent because over the last two days we’ve seen just how easy it is to build aspects.
When it comes to building aspects, using the provided base classes such as OnMethodBoundaryAspect and LocationInterceptionAspect are quick and easy ways to implement a simple aspect. But they have their limitations: each aspect can only implement one transformation. But what if you need to encapsulate a design pattern made of several transformations? We’re going to look at two ways of building complex aspects: aspect providers today and advices tomorrow.
Base Aspect Classes
There are different base classes in which an aspect can derive from. For example, OnMethodBoundaryAspect is derived from MethodLevelAspect class because it deals with methods. These base classes are pre-configured for MulticastAttributeUsage and pre-implemented interfaces. These classes are just containers for behaviors, they do not implement any behavior themselves, but it’s important to choose the correct class when developing aspects.
· AssemblyLevelAspect – Base class for all aspects applied on assemblies
· TypeLevelAspect – Base class for all aspects applied on types
· MethodLevelAspect – Base class for all aspects applied on methods
· LocationLevelAspect – Base class for all aspects applied on fields, properties or parameters
· EventLevelAspect – Base class for all aspects applied on events
PostSharp does not rely on these classes, but on the interfaces they implement. You can create your own aspect classes by implementing the right interface – for instance IOnMethodBoundaryAspect.
IAspectProvider
When an aspect implements the IAspectProvider interface, it can provide additional aspects dynamically. Simply put, an aspect can tell PostSharp to apply other aspects to a target at compile time. Let’s take a look at an example
[Serializable]
public class Aspect1 : IOnMethodBoundaryAspect
{
public void OnEntry(MethodExecutionArgs args)
{
Console.WriteLine("Aspect1: OnEntry for " + args.Method.Name);
}
public void OnException(MethodExecutionArgs args) { }
public void OnExit(MethodExecutionArgs args) { }
public void OnSuccess(MethodExecutionArgs args) { }
public void RuntimeInitialize(MethodBase method) { }
}
[Serializable]
public class Aspect2 : IMethodInterceptionAspect
{
public void OnInvoke(MethodInterceptionArgs args)
{
Console.WriteLine("Aspect2: OnInvoke for " + args.Method.Name);
args.Proceed();
}
public void RuntimeInitialize(MethodBase method) { }
}
[Serializable]
public class ComplexAspect : MethodLevelAspect, IAspectProvider
{
private readonly Aspect1 _aspect1 = new Aspect1();
private readonly Aspect2 _aspect2 = new Aspect2();
#region IAspectProvider Members
public IEnumerable ProvideAspects(object targetElement)
{
MemberInfo nfo = (MemberInfo)targetElement;
yield return new AspectInstance(targetElement, _aspect1);
if (nfo.ReflectedType.IsPublic && !nfo.Name.Equals(".ctor"))
{
yield return new AspectInstance(targetElement, _aspect2);
}
}
#endregion
}
We have three aspects here. Aspect1 and Aspect2 are just simple aspects that write their status to the console. The ComplexAspect however has some work happening. ComplexAspect derives from MethodLevelAspect because we’re going to be dealing with methods and it implements the IAspectProvider interface because we want to dynamically determine which aspects to apply to each target method.
We start by defining two instances one for Aspect1 and another for Aspect2. We’ll use these inside of ProvideAspects method when telling PostSharp to apply them to a target.
Inside the ProvideAspect method we cast targetElement as a MemberInfo structure. If we were using a LocationLevelAspect class instead of MethodLevelAspect class, we would cast it as a LocationInfo instead of MemberInfo because we’re working on a different level.
Since the return type of ProvideAspects method is an IEnumerable, we can use a yield return to provide aspects for PostSharp to apply. We return the instance of Aspect1 for all targets and then we check to see if the target method is public and is not a constructor. If it meets the criteria then we return the instance of Aspect2.
public class Program
{
static void Main(string[] args)
{
ExampleClass ec = new ExampleClass();
ec.DoSomething();
Console.ReadKey();
}
}
[ComplexAspect]
public class ExampleClass
{
public void DoSomething()
{
Console.WriteLine("Did something");
}
}
Our test code is a simple class with the ComplexAspect applied to it and we’re just making a call to DoSomething. The output is
Aspect1: OnEntry for .ctor
Aspect1: OnEntry for DoSomething
Aspect2: OnInvoke for DoSomething
Did something
Aspect1 was applied to both the constructor and the DoSomething method while Aspect2 was only applied to the DoSomething method.
CustomAttributeIntroductionAspect
PostSharp provides us with a very nice aspect that we can use to apply custom attributes to targets. For example, we can apply attributes to a DTO for XmlSerialization. Let’s look at an example of that
[AddXmlIgnoreAttribute]
public class ExampleClass
{
[XmlElement]
public int ID { get; set; }
public string First { get; set; }
public string Last { get; set; }
}
[MulticastAttributeUsage(MulticastTargets.Field | MulticastTargets.Property,
TargetMemberAttributes = MulticastAttributes.Public | MulticastAttributes.Instance)]
public sealed class AddXmlIgnoreAttribute : LocationLevelAspect, IAspectProvider
{
private static readonly CustomAttributeIntroductionAspect
customAttributeIntroductionAspect =
new CustomAttributeIntroductionAspect(
new ObjectConstruction(typeof(XmlIgnoreAttribute)
.GetConstructor(Type.EmptyTypes)));
public IEnumerable ProvideAspects(object targetElement)
{
LocationInfo memberInfo = (LocationInfo)targetElement;
if (memberInfo.PropertyInfo.IsDefined(typeof(XmlElementAttribute), false) ||
memberInfo.PropertyInfo.IsDefined(typeof(XmlAttributeAttribute), false))
yield break;
yield return new AspectInstance(memberInfo.PropertyInfo,
customAttributeIntroductionAspect);
}
}
We define a simple DTO class with four public properties. The DTO class is decorated with the AddXmlIgnoreAttribute.
Our aspect is going to apply the [XmlIgnore] attribute to all members of the DTO. If a member is already marked for inclusion using the [XmlElement] or [XmlAttribute] attributes then it will not receive the [XmlIgnore] attribute.
We start by defining the multicast options for our aspect. We tell PostSharp to only apply the aspect to fields or properties that are public and not static. Our aspect derives from LocationLevelAspect so we receive information structures specific to fields, properties and parameters.
We declare a new instance of CustomAttributeIntroductionAspect and passing in the ObjectConstruction data for XmlIgnoreAttribute type using the default empty constructor for it. We’ll cover how CustomAttributeIntroductionAspect works internally at another time.
In the ProvideAspects method, we cast targetElement as LocationInfo so we can work with the reflection info for the target. We use the LocationInfo to check if the property has either of the xml attributes defined on it and if it does we don’t yield any results for that target. Otherwise, we pass back to PostSharp the instance of CustomAttributeIntroductionAspect for application on the target.
Using ILSpy to see the end result
Except for ID, all of the properties now have XmlIgnore attributes.
Conclusion
IAspectProvider is a great way to dynamically determine which aspects to apply to a target at compile time but because aspects are still independent of each other, it has its limitations. Tomorrow we’re going to look at another way to build complex aspects.
Previously we’ve covered interception for methods and “locations” (fields/properties). Today we’re going to finish up interception by looking at the EventInterceptionAspect.
Intercepting Events
Events in .NET are similar to automatic properties. They look like fields, but you can override their underlying methods, Get and Set. Events don’t have Get or Set methods, instead they have Add and Remove methods. PostSharp allows us to intercept these Add and Remove methods as well as the invocation of the event using the EventInterceptionAspect base class. Let’s have a look
[Serializable]
public class EventAspect : EventInterceptionAspect
{
public override void OnAddHandler(EventInterceptionArgs args)
{
args.ProceedAddHandler();
Console.WriteLine("Handler added");
}
public override void OnRemoveHandler(EventInterceptionArgs args)
{
args.ProceedRemoveHandler();
Console.WriteLine("Handler removed");
}
public override void OnInvokeHandler(EventInterceptionArgs args)
{
args.ProceedInvokeHandler();
Console.WriteLine("Handler invoked");
}
}
The aspect is very similar to the other aspects we’ve seen. We simply override the provided virtual methods for the actions we want to intercept - add, remove and invoke. Our test code is as follows
public class Program
{
static void Main(string[] args)
{
ExampleClass c = new ExampleClass();
c.SomeEvent += new EventHandler(c_SomeEvent);
c.DoSomething();
c.SomeEvent -= c_SomeEvent;
Console.ReadKey();
}
static void c_SomeEvent(object sender, EventArgs e)
{
Console.WriteLine("Hello Event!");
}
}
public class ExampleClass
{
[EventAspect]
public event EventHandler SomeEvent;
public void DoSomething()
{
if (SomeEvent != null)
{
SomeEvent.Invoke(this, EventArgs.Empty);
}
}
}
The output is as expected
Handler added
Hello Event!
Handler invoked
Handler removed
EventInterceptionAspect
This aspect is pretty straight forward. We’re given three points in which we can intercept Add, Remove and Invoke. As with the other interception aspects, if one or more is applied to a target, the next node in the chain will be invoked and may not be the target event.
EventInterceptionAspect.OnAddHandler
Instead of the Add semantic of the event, the OnAddHandler is invoked instead. This occurs when a new handler is attached to the event (C# +=).
EventInterceptionAspect.OnRemoveHandler
Instead of the Remove semantic of the event, the OnRemoveHandler is invoked instead. This occurs when a delegate is removed from the event (C# -=).
EventInterceptionAspect.OnInvokeHandler
When the event is fired, the OnInvokeHandler method is invoked for each delegate attached to the event. If several handlers have been registered to the event, this method is called once for every handler.
EventInterceptionArgs
Each of the methods that we implement will have an EventInterceptionArgs parameter that we can use to get information and take action.
EventInterceptionArgs.Handler
Handler is the delegate that is currently being added, removed or invoked
EventInterceptionArgs.Instance
Instance is a reference to the instance from which the invocation is occurring, usually the class that the event is a member of. The instance where the invocation will occur can be changed by setting Instance to another class instance.
EventInterceptionArgs.Arguments
Arguments provides access to the arguments passed in during invocation of the event. For example, args.Argument[0] would typically be the value of the “sender” parameter and args.Arguments[1] would be the EventArgs (or some derivation).
EventInterceptionArgs.Event
Event is an instance of System.Reflection.EventInfo containing the reflected information for the target event. For more information on EventInfo, see the MSDN reference.
EventInterceptionArgs.AddHandler
AddHandler is the representation of the Add semantic of the event. You can call this directly to add a handler. It is possible to add additional/specific delegates to the event using AddHandler.
EventInterceptionArgs.ProceedAddHandler
Continues with the original request of adding a delegate to the target event.
EventInterceptionArgs.RemoveHandler
RemoveHandler is the representation of the Remove semantic of the event. You can call this directly to remove a handler. It is possible to remove additional/specific delegates to the event using RemoveHandler.
EventInterceptionArgs.ProceedRemoveHandler
Continues with the original request of removing a delegate from the target event.
EventInterceptionArgs.InvokeHandler
InvokeHandler allows us to invoke the handler, but it allows us to do so by providing a different delegate and arguments. InvokeHandler returns an object which will contain the return value of the delegate (if the delegate is not void).
EventInterceptionArgs.ProceedInvokeHandler
Continues with the original invocation of the delegate with the specified arguments. After invocation, args.ReturnValue may contain a value if the delegate is not void.
EventInterceptionArgs.ReturnValue
ReturnValue will contain the value returned by the delegate is it is not void. The return value can be changed or manipulated by setting ReturnValue to a new value.
Making Events Asynchronous
One of the uses for event interception is to invoke the registered delegates asynchronously. Let’s check out what that aspect looks like
[Serializable]
public sealed class AsyncEventAttribute : EventInterceptionAspect
{
public override void OnInvokeHandler(EventInterceptionArgs args)
{
Task.Factory.StartNew(() => Invoke(args));
}
private static void Invoke(EventInterceptionArgs args)
{
try
{
args.ProceedInvokeHandler();
}
catch (Exception e)
{
args.ProceedRemoveHandler();
}
}
}
This is a very simple aspect. We are only implementing the OnInvokeHandler which has only one job, creating and starting a task. We use Task.Factory.StartNew() to create and immediately start the task asynchronously. We provide the StartNew method with an action which just makes a call to our Invoke method. Tasks are part of the Task Parallel Library which ships with .NET 4.0. If you are not familiar with the TPL or Tasks, please see the MSDN reference.
The Invoke method contains a try/catch structure. We make a call to args.ProceedInvokeHandler and if an exception occurs, we catch it and then remove that delegate from the event by calling args.ProceedRemoveHandler.
To test our aspect, we have modified our example from above.
public class Program
{
static void Main(string[] args)
{
ExampleClass c = new ExampleClass();
c.SomeEvent += new EventHandler(c_SomeEvent);
c.SomeEvent += new EventHandler(c_SomeEvent);
c.SomeEvent += new EventHandler(c_SomeEvent);
c.SomeEvent += new EventHandler(c_SomeEvent2);
c.SomeEvent += new EventHandler(c_SomeEvent2);
c.SomeEvent += new EventHandler(c_SomeEvent2);
c.DoSomething();
Console.ReadKey();
}
static void c_SomeEvent(object sender, EventArgs e)
{
Console.WriteLine("Hello Event! Task: " + Task.CurrentId);
}
static void c_SomeEvent2(object sender, EventArgs e)
{
Console.WriteLine("Hello Event! Task: " + +Task.CurrentId);
}
}
public class ExampleClass
{
[AsyncEventAttribute]
public event EventHandler SomeEvent;
public void DoSomething()
{
if (SomeEvent != null)
{
SomeEvent.Invoke(this, EventArgs.Empty);
}
}
}
We register a few handlers with the event and then we call the DoSomething method that is going to invoke the event. Our output shows that the registered handlers were invoked and that they have been invoked in their own task.
Hello Event! Task: 1
Hello Event! Task: 2
Hello Event! Task: 3
Hello Event! Task: 4
Hello Event! Task: 5
Hello Event! Task: 6
Under the Hood
Just for fun, let’s open up ILSpy and look at our example code, you’ll see that there is a lot of work going on.
If you look around, you’ll notice a few things. Even though we didn’t implement OnAddHandler or OnRemoveHandler, PostSharp has modified the event to use <SomeEvent>_Broker to do the adding and removing of handlers.
<SomeEvent>_Broker is a nested class that PostSharp has added and it derives from EventBroker, an abstract class that is used to realize the interception of the invocation.
Notice that the DoSomething method doesn’t call our aspect’s OnInvokeHandler method, nor does it call to the <SomeEvent>_Broker. It simply does the invocation of the event just as it was coded in Visual Studio. How is it that it can intercept the invocation then? When <SomeEvent>_Broker is instantiated in the <>z__InitializeAspects method it gets a reference to our instance. EventBroker is a black box that uses our instance to wrap around the event. We’ll leave it at that for now.
Conclusion
The further we dig, the more we see just how comprehensive PostSharp really is. We still have more to cover. The topics are getting more complex, but I’ll try to make sense of them.
Continuing from yesterday, we’re covering the lifetime and scope of aspects. Today we’ll look at the scope of aspects.
Scope
Aspects come in two scopes, static and instance. Static scoped aspects are created and initialized at compile time for consumption at runtime using a singleton pattern. Instance scoped aspects are a bit different. Instanced scoped aspects use a prototype pattern by creating and initializing the aspect at compile time, but at run time when a new instance of a target member’s declaring type is created, a new instance of the aspect is created and used.
Static scoped aspects have the same lifetime as the application while instance scoped aspects have the same lifetime as the instance of the type the aspect was applied to.
However, no matter which scope an aspect will have, PostSharp creates an instance of the aspect for each target in which the aspect has been applied. No doubt this is all confusing so let's see what's going on with a demo.
Aspect demo
To make sense of all of this, let's start with a basic aspect.
[Serializable]
public class TestAspect : LocationInterceptionAspect
{
private string InstID;
private string aspectID;
private string source;
private int getCount = 0;
public override void CompileTimeInitialize(LocationInfo targetLocation,
AspectInfo aspectInfo)
{
source = targetLocation.DeclaringType.Name + "." + targetLocation.Name
+ " (" + targetLocation.LocationType.Name + ")";
aspectID = Guid.NewGuid().ToString();
}
public override void RuntimeInitialize(LocationInfo locationInfo)
{
InstID = Guid.NewGuid().ToString();
}
public override void OnGetValue(LocationInterceptionArgs args)
{
getCount++;
Console.WriteLine(source + "\n\tInstance: " + this.InstID + "\n\tAspect: " +
aspectID + "\n\tCount: " + getCount);
args.ProceedGetValue();
}
}
In CompileTimeInitialize we setup our source which is a combination of declaring type, target name and then the target's type. Then we assign a guid to aspectID which we'll use when we evaluate the aspect's life. In RunTimeInitialize we do the same thing and create a new guid that we use for the instance id. Then in the OnGetValue method we just increment the count and print all of our data to the console before calling ProceedGetvalue.
Our example class is as follows
class TestClass
{
[TestAspect]
public int MyField1;
[TestAspect]
public int MyField2;
[TestAspect]
public static int MyField3;
public TestClass()
{
MyField1 = 1;
MyField2 = 2;
}
static TestClass()
{
MyField3 = 10;
}
}
And our test code
int val = 0;
for (int i = 1; i <= 2; i++)
{
Console.WriteLine("--- PASS {0} ---", i);
TestClass tc1 = new TestClass();
TestClass tc2 = new TestClass();
val = TestClass.MyField3;
Console.WriteLine();
val = tc1.MyField1;
val = tc1.MyField2;
Console.WriteLine();
val = tc2.MyField1;
val = tc2.MyField2;
Console.WriteLine();
}
What we're doing is creating two instances of our TestClass and then getting the value of Myfield1 and MyField2 from each instance which our aspect will intercept and provide us with information about which aspect is handling our request.
Aspect for each target
When we run our test code, we get the following output (I’ve put the two passes side by side for comparison)
--- PASS 1 ---
TestClass.MyField3 (Int32)
Instance: d6c15b92-3ba1-4aba-8325-1ca07d50979c
Aspect: b73e55cf-6568-4b1c-9465-8d3b77f3288c
Count: 1
TestClass.MyField1 (Int32)
Instance: 667d5f03-ac25-43ef-86dd-7aa88ccc46c7
Aspect: a6e612ed-7ab9-4f31-956d-6be2f3198352
Count: 1
TestClass.MyField2 (Int32)
Instance: a5efc352-31f0-430c-8153-d78e0e1453c4
Aspect: 8feb7451-5df2-48e5-b57d-3588456aca96
Count: 1
TestClass.MyField1 (Int32)
Instance: 667d5f03-ac25-43ef-86dd-7aa88ccc46c7
Aspect: a6e612ed-7ab9-4f31-956d-6be2f3198352
Count: 2
TestClass.MyField2 (Int32)
Instance: a5efc352-31f0-430c-8153-d78e0e1453c4
Aspect: 8feb7451-5df2-48e5-b57d-3588456aca96
Count: 2
|
--- PASS 2 ---
TestClass.MyField3 (Int32)
Instance: d6c15b92-3ba1-4aba-8325-1ca07d50979c
Aspect: b73e55cf-6568-4b1c-9465-8d3b77f3288c
Count: 2
TestClass.MyField1 (Int32)
Instance: 667d5f03-ac25-43ef-86dd-7aa88ccc46c7
Aspect: a6e612ed-7ab9-4f31-956d-6be2f3198352
Count: 3
TestClass.MyField2 (Int32)
Instance: a5efc352-31f0-430c-8153-d78e0e1453c4
Aspect: 8feb7451-5df2-48e5-b57d-3588456aca96
Count: 3
TestClass.MyField1 (Int32)
Instance: 667d5f03-ac25-43ef-86dd-7aa88ccc46c7
Aspect: a6e612ed-7ab9-4f31-956d-6be2f3198352
Count: 4
TestClass.MyField2 (Int32)
Instance: a5efc352-31f0-430c-8153-d78e0e1453c4
Aspect: 8feb7451-5df2-48e5-b57d-3588456aca96
Count: 4
|
Let's break it down. In Pass 1 we see 5 calls made to our aspect. The first call is to the static member MyField3 while the next 2 are instance calls to MyField1 and MyField3 on tc1 instance and the same for the next 2 calls, but for the tc2 instance.
If you compare the aspect ID's, there are actually only 3 different instances of our aspect, one for each of the properties we applied it to, MyField1, MyField2 and MyField3 even though we have two separate instances of TestClass.
Static Scoped
Now that we've identified that a new aspect is generated for each target, let's examine the Instance ID's. In Pass 1 we see there are 3 instance ID's. Examining Pass 2, we see those exact same instance ID's even though we created new instances of our class.
Notice how the counter is increasing. In Pass 1 the first call to tc1.MyField1 results in 1 while the call to tc2.MyField1 results in a 2. In Pass 2 we see that the trend continues. This is due to the static nature of the aspect instances.
Instance Scoped
Aspects are only instance scoped when implementing the IInstanceScopedAspect interface or inheriting from InstanceLevelAspect, so let's update our aspect.
[Serializable]
public class TestAspect : LocationInterceptionAspect, IInstanceScopedAspect
{
private string InstID;
private string aspectID;
private string source;
private int getCount = 0;
public override void CompileTimeInitialize(LocationInfo targetLocation,
AspectInfo aspectInfo)
{
source = targetLocation.DeclaringType.Name + "." + targetLocation.Name
+ " (" + targetLocation.LocationType.Name + ")";
aspectID = Guid.NewGuid().ToString();
}
public override void RuntimeInitialize(LocationInfo locationInfo)
{
InstID = Guid.NewGuid().ToString();
}
public override void OnGetValue(LocationInterceptionArgs args)
{
getCount++;
Console.WriteLine(source + "\n\tInstance: " + this.InstID
+ "\n\tAspect: " + aspectID + "\n\tCount: " + getCount);
args.ProceedGetValue();
}
#region IInstanceScopedAspect Members
public object CreateInstance(AdviceArgs adviceArgs)
{
return this.MemberwiseClone();
}
public void RuntimeInitializeInstance()
{
InstID = Guid.NewGuid().ToString();
}
#endregion
}
Now we implement IInstanceScopedAspect which requires CreateInstance() and RuntimeInitializeInstance(). CreateInstance is called to create a new instance of the aspect based on the current instance, thus using the current instance as a protoype. All we need to do is use the MemberwiseClone() and we're set.
RuntimeInitializeInstance is where we update our instance ID. If we didn't, we would only get the instance ID specified in the RuntimeInitialize which is only invoked once for each aspect when it's deserialized, not when a new instance is created.
Let's run our test code and see how things have changed. (I’ve put the two passes side by side for comparison)
--- PASS 1 ---
TestClass.MyField3 (Int32)
Instance: ea470da4-ed8d-49e1-be96-5ae16ac000a6
Aspect: b6aa3c5c-3868-40e8-af37-70234032d734
Count: 1
TestClass.MyField1 (Int32)
Instance: a0b96c14-0d81-463f-9374-6915b6def2a0
Aspect: a975c1ce-6839-454c-b054-14b185fd29a8
Count: 1
TestClass.MyField2 (Int32)
Instance: 3a6676c6-0325-4ee3-8dbb-fe4d6c93acdd
Aspect: dbd6865f-3b5b-48dd-a87e-ef8a89695e2f
Count: 1
TestClass.MyField1 (Int32)
Instance: 3ee8689a-a1a9-4dfc-9ca6-10bd50129a1d
Aspect: a975c1ce-6839-454c-b054-14b185fd29a8
Count: 1
TestClass.MyField2 (Int32)
Instance: 84f63d43-c91a-46ba-8bda-89ad16d0ba2c
Aspect: dbd6865f-3b5b-48dd-a87e-ef8a89695e2f
Count: 1
|
--- PASS 2 ---
TestClass.MyField3 (Int32)
Instance: ea470da4-ed8d-49e1-be96-5ae16ac000a6
Aspect: b6aa3c5c-3868-40e8-af37-70234032d734
Count: 2
TestClass.MyField1 (Int32)
Instance: 2ad257df-682e-4652-98fa-5466d2bb08aa
Aspect: a975c1ce-6839-454c-b054-14b185fd29a8
Count: 1
TestClass.MyField2 (Int32)
Instance: e39fe1e1-ca7a-4b26-90ac-8c81fc1aafe5
Aspect: dbd6865f-3b5b-48dd-a87e-ef8a89695e2f
Count: 1
TestClass.MyField1 (Int32)
Instance: cb40067e-2b69-4c5b-814e-4a7420a085a9
Aspect: a975c1ce-6839-454c-b054-14b185fd29a8
Count: 1
TestClass.MyField2 (Int32)
Instance: e4c21efc-92e6-41cc-bf34-babbddf4ea48
Aspect: dbd6865f-3b5b-48dd-a87e-ef8a89695e2f
Count: 1
|
First, have a look at the Aspect ID's. Notice that again, there are only 3 ID's as we only have 3 targets. They are used both in Pass 1 and Pass 2. This is the same as the static scoped example. The difference is the instance ID's. Except for TestClass.Myfield3, which is static, there are no repeating instance ID's. For each target, we get a new instance of the aspect when we instantiated a new instance of TestClass. This can be verified by examining the count. Since we only make one get call per TestClass instance, count is always 1 because it's scoped to the current instance of the declaring type, tc1 and tc2.
So what happened with TestClass.MyField3? Even though the aspect implements IInstanceScopedAspect, when applied to a static target, the aspect instance becomes static scoped. This only makes sense considering the nature of static members.
Conclusion
It's amazing how much flexibility PostSharp gives us, but as I've stated before, a solid understanding of how it works is key to producing quality results. This week we spent a lot of time under the hood looking at what PostSharp does when you click the build button.