We Keep Coding

Can I write tests for custom attributes without defining x^n classes?

I'm building a class library that includes several custom attributes that apply to properties. Then I have methods that do certain things based on the placement of the attributes.
Now I want to build some unit tests, but how to I make the tests without creating something on the order of x^(number of attributes) classes just for testing purposes? Can I leverage metadata classes or something?
Basically I'd love for there to be a way for me to apply attributes to properties at runtime (i.e. inside the "Arrange" part of my test method), but I'm pretty sure that's impossible.
Edit
This is the reflection code I'm using to test attributes, since apparently how I'm reading them may make a difference:
bool skip = false, iip = false;
string lt = null;
SerializeAsOptions sa = SerializeAsOptions.Ids;
object[] attrs = prop.GetCustomAttributes(true);
foreach (object attr in attrs)
{
Type attrType = attr.GetType();
if (typeof(JsonIgnoreAttribute).IsAssignableFrom(attrType))
{
skip = true;
continue;
}
if (typeof(IncludeInPayload).IsAssignableFrom(attrType))
iip = ((IncludeInPayload)attr).Include;
if (typeof(SerializeAs).IsAssignableFrom(attrType))
sa = ((SerializeAs)attr).How;
if (typeof(LinkTemplate).IsAssignableFrom(attrType))
lt = ((LinkTemplate)attr).LinkTemplate;
}
if (skip) continue;

I'm adding another answer, because since you now provided some code, the old one is too broad. It's now (mostly) obvious that:
you control the attribute-reading code
you are reading the code via reflection (PropertyInfo.GetCustomAttributes)
So. Since you are using Reflection, TypeDescriptors will not help. You'd need to:
either read the attrs differently so TypeDescr can be used
dynamically generate assemblies at runtime to generate classes with properties on the fly during tests
It can be very interesting/entertaining, but it can also turn into nice amount of work. But, since you control both sides of the code, none of these two is actually needed.
First, let's trim the code to significant parts:
somemethod(PropertyInfo prop)
{
// ...
object[] attrs = prop.GetCustomAttributes(true); // read ATTRs from PROP
foreach (object attr in attrs) // scan the PROP's ATTRs
{
// check attr type, do something
}
// ...
}
The core of your problem is not:
adding/removing attributes during Arrange/Teardown part
but
forcing the loop over PROP's ATTRs to see attributes that your test specifies
Looking at the problem like this, the answer is almost obvious: your loop has to abstract from the "Read attributes" part.
object[] attributeReader(PropertyInfo prop)
{
return prop.GetCustomAttributes(true);
}
somemethod(PropertyInfo prop)
{
// ...
object[] attrs = attributeReader(prop); // read ATTRs from PROP
foreach (object attr in attrs) // scan the PROP's ATTRs
{
// check attr type, do something
}
// ...
}
Now, your processing code is independent of the way the attributes are read. Sure, in the example above that way is hardcoded. But it does not have to be. Depending on how you want/like to organize your tests, you can use many ways to replace the attributeReader method with other mechanisms.
For example, just add 'virtual' to the attributeReader and use inheritance to create a class that will enable AttributeFaking:
// original class:
virtual object[] attributeReader(PropertyInfo prop)
{
return prop.GetCustomAttributes(true);
}
// derived class:
object[] AttributesOverrides {get;set;}
override object[] attributeReader(PropertyInfo prop)
{
if(prop.Name = "ShoeSize") return AttributesOverrides; // return what I say!
return base.attributeReader(prop);
}
// your test setup
var t = ... // that DERIVED object
t.AttributesOverrides = new [] { ... } ; // attributes to use
For example, use delegates/lambdas, no inheritace
// note the attributeReader is now a field, not function
Func<PropertyInfo, object[]> attributeReader = defaultAttributeReader;
static object[] defaultAttributeReader(PropertyInfo prop)
{
return prop.GetCustomAttributes(true);
}
// and your test setup
var t = ... // that ORIGNAL object
t.attributeReader = customReaderForTheTest; // change the reader on the fly
// that's the reader-function to use in THIS TEST setup
static object[] customReaderForTheTest(PropertyInfo prop)
{
if(prop.Name = "ShoeSize") return null; // crash when I say so! muhaHAHAhaa!
return prop.GetCustomAttributes(true);
}
Both of those two examples end up with one class that is enables faking the attributes in some way, but that's not the only ways to do that. You can use IoC to inject the correct attributeReader. You can do that in any way you like - you just need to abstract from reading part and leave it 'open'.

It is not possible to really apply the attribute at runtime to an existing class, but there are at least two ways you could do something similar to it - it depends on how exactly are you reading those attributes later.
The options focus on the 'really' and 'existing class' part:
1) don't do that, just fake adding them
2) apply them on a class that does not exist yet! ;)
First option is a CustomTypeDescriptor. In its implementations, you will be able to dynamically answer to any queries about Attributes for some class that uses it (-> see virtual GetAttributes method).
This leads to first way:
Create AttributableTestObject that i.e. inherits from your ClassCompatibleWithThatAttribute etc
Create something like DynamicallyAttributedClass : CustomTypeProvider that exposes a static property similar to IEnumerable<Attribute>
override the GetAttributes and return whatever was provided by that static property
on your AttributableTestObject class set a TypeDecriptorProvider attribute pointing to provider (you've got to implement it, again) that returns DynamicallyAttributedClass
Now, using that static property you can change what the GetAttributes returns, and therefore you can dynamically change the setof attributes that are visible through typedescriptor.
And here's the catch: Not all engines/observers/readers/(..) actually care about TypeDescriptors. Some simply read the metadata right away from the Reflection. Reflection will not check the typedescriptors. It will simply return an information that the AttributableTestObject class has a TypeDecriptorProvider property. But whe nusing used the ComponentModel mechanisms, the custom list of attribues will be visible.
That reminds me that the reading mechanisms simply sit at TypeDescriptor class, see its methods. There's also AddAttribute, so maybe you can even get what you want without actually implementing the stuff I said above - Try using AddAttribute and then GetAttribute on your target Type and check the results. It may "just work". But again, it will not fool the reflection.
So, there's a second, "more hardcore" approach - dynamic classes. With System.Reflection.Emit you can dynamically generate an assembly that will contain whatever IL code you wish, and then load the assembly to the current runtime.
This gives you a unique option to simply generate a new fresh class(es) at runtime, with any contents and any attributes you like. You can use inheritance, so you can inherit from any ClassCompatibleWithThatAttributes. It's not that easy to do manually, but there are some libraries that make using the IL emitter easier.
Note that the generated types will be generated at runtime. This means that at compile-time you will not have them checked, you must generate them perfectly or face some really rarely seen Exceptions. Also, note that as the types are not known at compile-time, you cannot simply new the objects. You will have to create the objects basing on their freshly-generated Type through i.e. Activator.CreateInstance(Type).
Also, even though you can generate as much new classes as you want - if you overdo it, you probably will eventually hit some CLR limit, or at leat OutOfMemory, since the generated and loaded assemblies actually occupy space. You can overcome it with AppDomains - you can create some extra AppDomains and generate and load the assemblies inside them, and then, finally, you can unload that extra domain to release all memory and also unload any assemblies that were loaded there. But, I suppose you will not generate that much types to really need that. I'm writing about it just-in-case. Dont worry until you hit the OutOfMemory or similar.

Finding where a method is called for a convention test

I'm trying to write a convention test that specifies that a method should only be called in some contexts - specifically I have a static Empty getter that I only want to allow used in test methods, vis methods decorated with TestAttribute.
I know that I should also mark the getter as obsolete, use another method etc, but I also want a convention test around this so it doesn't break in the future.
I am guessing I want to use static analysis through reflection in my convention test. How would I go about performing this kind of analysis?

Yes, Roslyn can help with this sort of thing. An example of what this might look like as a standalone analysis would be something like:
var solution = Solution.Load(pathToSolution);
foreach (var project in solution.Projects)
{
var type = project.GetCompilation().GetTypeByMetadataName(typeNameContainingMethod);
var method = type.GetMembers("Empty").Single();
var references = method.FindAllReferences(solution);
foreach (var referencedSymbol in references)
{
foreach (var referenceLocation in references)
{
CheckIfCallIsAllowed(referenceLocation);
}
}
}
You might also look at the Creating a Code Issue walkthrough and the Code Issue template that comes with the Roslyn CTP for another approach to doing this at edit time, instead of in a test.

Set or change Attribute's properties or fields at runtime in C#. Possible?

I believe there is no human way to change any attribute or field inside an Attribute apart from doing it in the constructor. That is, short of redesigning and recompiling Visual Studio yourself. There is already a similar question posted here:
Change Attribute's parameter at runtime
but I believe the peculiarities of my problem are different enough to require a new post.
I use an enumeration to keep track of the different columns of a DataTable. I use attributes in each enumeration element to indicate the underlying type and the description -in case the .ToString() would give an "ugly" result due to the rigid set of characters that are allowed to name an enumeration element, such as "Tomato_Field" when you want "Tomato Field", and the like. This allows me to place all the related information in the same object, which is, I believe, what it should be. This way I can later create all the columns with a simple and clean foreach that cycles through the elements of the enumeration and extracts the metedata (description and type) to create each column.
Now, some of the columns are autocalculated, which means that during their creation -via DataTable Identifier.Columns.Add.(NameOfColumn,underlyingType,optional: autocalculatedString)- I need to specify a string that determines how it should be calculated. That string must use the names of other columns, which might be in the Description Attribute. The approach that looks logical is to use another attribute that holds the string, which should be built using the names of the other columns, requiring access to the metadata. Now that seems impossible in the constructor: you are forced to provide a constant string. You can't use a method or anything.
This problem could be solved if there were a way to change a property inside the attribute (lets call it AutocalculatedStringAttribute) at runtime. If you access the metadata you can retrieve the string you used at the constructor of the Attribute, and you can of course change that string. However, if you later access the metadata again that change is ignored, I believe the constructor is called every time the metadata is accessed at runtime, thus ignoring any changes.
There are, of course, dirty ways to achive what I am trying to do, but my question is specifically if there is a way to properly use attributes for this. Short of resorting to CodeDOM to recompile the whole assembly with the constructor of the AutocalculatedStringAttribute changed, a certain overkill.

Right, the metadata that's used to initialize the attribute is immutable. But you can add properties and methods to an attribute class that can run code and return relevant info after the attribute object is constructed. The data they rely on doesn't have to be stored in metadata, it can be persisted anywhere.
Of course, such code wouldn't have to be part of the attribute class implementation, it could just as well be part of the code that instantiates the attribute. Which is where it belongs.

It isn't entirely clear to me what code is consuming this attribute, and it matters...
You cannot change an attribute that is burned into the code - you can query it with reflection, but that is about it. However, in many cases you can still do interesting things - I don't know if they apply to your scenario, though:
you can subclass many attributes like [Description], [DisplayName], etc - and while you pass in a constant string (typically a key) to the .ctor, it can return (through regular C#) more flexible values - perhaps looking up the description from a resx to implement i18n
if the caller respects System.ComponentModel, you can attach attributes at runtime to types etc very easily - but much harder on individual properties, especially in the case of DataTable etc (since that has a custom descriptor model via DataView)
you can wrap things and provide your own model via ICustomTypeDescriptor / TypeDescriptionProvider / PropertyDescriptor - lots of work, but provides access to set your own attributes, or return a description (etc) outside of attributes
I don't know how much of this is suitable for your environment (perhaps show some code of what you have and what you want), but it highlights that (re the question title) yes: there are things you can do to tweak how attributes are perceived at runtime.

I wanted to post this as a comment but since I wanted to include some code I couldn't, given the 600 characters limit. This is the cleanest solution I have managed to find, although it does not include all the info to create the columns on the enum, which is my goal. I have translated every field to make it easier to follow. I am not showing some code which has an obvious use (in particular the implementations of the other custom attributes and their static methods to retrieve the metadata, assume that it works).
This gets the job done, but I would ideally like to include the information stored in the strings "instancesXExpString " and "totalInstancesString" in the Autocalculated attribute, which currently only marks the columns that have such a string. This is what I have been unable to do and what, I believe, cannot be easily accomplished via subclassing -although it is an ingenious approach, I must say.
Thanks for the two prompt replies, btw.
And without any further ado, lets get to the code:
// Form in which the DataGridView, its underlying DataTable and hence the enumeration are:
public partial class MainMenu : Form {
(...)
DataTable dt_expTable;
//Enum that should have all the info on its own... but does not:
public enum e_columns {
[TypeAttribute(typeof(int))]
Experiments = 0,
[TypeAttribute(typeof(decimal))]
Probability,
[DescriptionAttribute("Samples / Exp.")]
[TypeAttribute(typeof(int))]
SamplesXExperiment,
[DescriptionAttribute("Instances / Sample")]
[TypeAttribute(typeof(int))]
InstancesXSample,
[DescriptionAttribute("Instances / Exp.")]
[TypeAttribute(typeof(int))]
[Autocalculated()]
InstancesXExp,
[DescriptionAttribute("Total Instances")]
[TypeAttribute(typeof(long))]
[Autocalculated()]
Total_Instances
};
//These are the two strings
string instancesXExpString = "[" + DescriptionAttribute.obtain(e_columns.SamplesXExperiment) + "] * [" + DescriptionAttribute.obtain(e_columns.InstancesXMuestra) + "]";
string totalInstancesString = "[" + DescriptionAttribute.obtain(e_columns.InstancesXExp) + "] * [" + DescriptionAttribute.obtain(e_columns.Experiments) + "]";
public MainMenu() {
InitializeComponent();
(...)
}
private void MainMenu_Load(object sender, EventArgs e) {
(...)
// This is the neat foreach I refered to:
foreach (e_columns en in Enum.GetValues(typeof(e_columnas))) {
addColumnDT(en);
}
}
private void addColumnDT(Enum en) {
//*This is a custom static method for a custom attrib. that simply retrieves the description string or
//the standard .ToString() if there is no such attribute.*/
string s_columnName = DescriptionAttribute.obtain(en);
bool b_typeExists;
string s_calculusString;
Type TypeAttribute = TypeAttribute.obtain(en, out b_typeExists);
if (!b_typeExists) throw (new ArgumentNullException("Type has not been defined for one of the columns."));
if (isCalculatedColumn(DescriptionAttribute.obtain(en))) {
s_calculusString = calcString(en);
dt_expTable.Columns.Add(s_columnName, TypeAttribute, s_calculusString);
} else {
dt_expTable.Columns.Add(s_columnName, TypeAttribute);
}
}
private string calcString(Enum en) {
if (en.ToString() == e_columns.InstancessXExp.ToString()) {
return instancesXExpString;
} else if (en.ToString() == e_columns.Total_Samples.ToString()) {
return totalInstancesString;
} else throw (new ArgumentException("There is a column with the autocalculated attribute whose calculus string has not been considered."));
}
(...)
}
I hope this piece of code clarifies the situation and what I am trying to do.

Allow or disallow method run in .NET

I need to organize some simple security in a class depends on value of the enum.
All that I can figure out is using attribute on a method and then run check then if it fails throw an exception.
Sample:
[ModulePermission(PermissonFlags.Create)]
public void CreateNew()
{
CheckPermission();
System.Windows.Forms.MessageBox.Show("Created!");
}
protected void CheckPermission()
{
var method = new System.Diagnostics.StackTrace().GetFrame(1).GetMethod();
if (!flags.HasFlag(method.GetCustomAttributes(true).Cast<ModulePermissionAttribute>().First().Flags))
{
throw new ApplicationException("Access denied");
}
}
is there more elegant or simple way to do this, like just to trigger an event when method run?

Why not just use standard Code Access Security instead of reimplementing the attribute handling and stack walking?
I think that if you read through the linked documentation, you'll see that what you have is nowhere close to what is needed to achieve actual security. Thankfully, this hard problem has already been solved...

Not with an enum, but with strings - voila (enforced by the runtime, even in full-trust):
public static class PermissionFlags {
public const string Create = "Create";
}
[PrincipalPermission(SecurityAction.Demand, Role = PermissionFlags.Create)]
public void CreateNew() {
System.Windows.Forms.MessageBox.Show("Created!");
}
All you need to do now is to represent the user as a principal. This is done for you in ASP.NET, and there is a winform plugin (in VS2008 etc) to use ASP.NET for membership. It can be configured for vanilla winforms and WCF, too; at the most basic level, GenericPrincipal / GenericIdentity:
// during login...
string[] roles = { PermissionFlags.Create /* etc */ };
Thread.CurrentPrincipal = new GenericPrincipal(
new GenericIdentity("Fred"), // user
roles);
But you can write your own principal / identity models easily enough (deferred / cached access checks, for example).

You might want to look at doing this with something like PostSharp, which will give you a framework for applying the attributes so that you don't have to run the check in your method. This may, however, increase the complexity depending on how the currently active flags are accessed. You'd probably need some class to cache the current permissions for the current user.

You could take a look at Aspect Oriented Programming.
Check out Postsharp for instance, which will enable you to 'weave' some additional logic at compile time in the methods that you've decorated with your ModulePermission attribute.
By doing so, you will not have to call the 'CheckPermission' method anymore inside that 'secured' method, since that logic can be weaved by Postsharp.
(A while ago, I've been playing around with Postsharp: http://fgheysels.blogspot.com/2008/08/locking-system-with-aspect-oriented.html )

Can attributes be added dynamically in C#?

Is it possible to add attributes at runtime or to change the value of an attribute at runtime?

This really depends on what exactly you're trying to accomplish.
The System.ComponentModel.TypeDescriptor stuff can be used to add attributes to types, properties and object instances, and it has the limitation that you have to use it to retrieve those properties as well. If you're writing the code that consumes those attributes, and you can live within those limitations, then I'd definitely suggest it.
As far as I know, the PropertyGrid control and the visual studio design surface are the only things in the BCL that consume the TypeDescriptor stuff. In fact, that's how they do about half the things they really need to do.

Attributes are static metadata. Assemblies, modules, types, members, parameters, and return values aren't first-class objects in C# (e.g., the System.Type class is merely a reflected representation of a type). You can get an instance of an attribute for a type and change the properties if they're writable but that won't affect the attribute as it is applied to the type.

You can't. One workaround might be to generate a derived class at runtime and adding the attribute, although this is probably bit of an overkill.

Well, just to be different, I found an article that references using Reflection.Emit to do so.
Here's the link: http://www.codeproject.com/KB/cs/dotnetattributes.aspx , you will also want to look into some of the comments at the bottom of the article, because possible approaches are discussed.

No, it's not.
Attributes are meta-data and stored in binary-form in the compiled assembly (that's also why you can only use simple types in them).

I don't believe so. Even if I'm wrong, the best you can hope for is adding them to an entire Type, never an instance of a Type.

If you need something to be able to added dynamically, c# attributes aren't the way. Look into storing the data in xml. I recently did a project that i started w/ attributes, but eventually moved to serialization w/ xml.

Why do you need to? Attributes give extra information for reflection, but if you externally know which properties you want you don't need them.
You could store meta data externally relatively easily in a database or resource file.

Like mentionned in a comment below by Deczaloth, I think that metadata is fixed at compile time. I achieve it by creating a dynamic object where I override GetType() or use GetCustomType() and writing my own type. Using this then you could...
I tried very hard with System.ComponentModel.TypeDescriptor without success. That does not means it can't work but I would like to see code for that.
In counter part, I wanted to change some Attribute values.
I did 2 functions which work fine for that purpose.
// ************************************************************************
public static void SetObjectPropertyDescription(this Type typeOfObject, string propertyName, string description)
{
PropertyDescriptor pd = TypeDescriptor.GetProperties(typeOfObject)[propertyName];
var att = pd.Attributes[typeof(DescriptionAttribute)] as DescriptionAttribute;
if (att != null)
{
var fieldDescription = att.GetType().GetField("description", BindingFlags.NonPublic | BindingFlags.Instance);
if (fieldDescription != null)
{
fieldDescription.SetValue(att, description);
}
}
}
// ************************************************************************
public static void SetPropertyAttributReadOnly(this Type typeOfObject, string propertyName, bool isReadOnly)
{
PropertyDescriptor pd = TypeDescriptor.GetProperties(typeOfObject)[propertyName];
var att = pd.Attributes[typeof(ReadOnlyAttribute)] as ReadOnlyAttribute;
if (att != null)
{
var fieldDescription = att.GetType().GetField("isReadOnly", BindingFlags.NonPublic | BindingFlags.Instance);
if (fieldDescription != null)
{
fieldDescription.SetValue(att, isReadOnly);
}
}
}

When faced with this situation, yet another solution might be questioning you code design and search for a more object-oriented way. For me, struggling with unpleasant reflection work arounds is the last resort. And my first reaction to this situation would be re-designing the code. Think of the following code, which tries to solve the problem that you have to add an attribute to a third-party class you are using.
class Employee {} // This one is third-party.
And you have code like
var specialEmployee = new Employee();
// Here you need an employee with a special behaviour and want to add an attribute to the employee but you cannot.
The solution might be extracting a class inheriting from the Employee class and decorating it with your attribute:
[SpecialAttribute]
class SpecialEmployee : Employee
{
}
When you create an instance of this new class
var specialEmployee = new SpecialEmployee();
you can distinguish this specialEmployee object from other employee objects. If appropriate, you may want to make this SpecialEmployee a private nested class.