Autofac generic decorator is duplicating components - c#

I have found what I think may be a bug in Autofac, but I wanted to see if anybody had a possible solution or workaround so I can make this work.
Basically I have set up a generic decorator, which works fine. The problem is that as soon as I call BeginLifetimeScope() with a configuration delegate, it erroneously resolves multiple components of the same type. If I don't use a configuration delegate with BeginLifetimeScope(), then it works correctly. Unfortunately, I need to add additional dependencies to my child scope, so not using a configuration delegate is not an option.
Here is an example that demostrates the problem:
var builder = new ContainerBuilder();
builder.RegisterType<Dependency>()
.Named<IDependency<object>>("service");
builder.RegisterGenericDecorator(
typeof(Decorator<>), typeof(IDependency<>), "service", "decorated");
var container = builder.Build();
// Returns 1
var scope1 = container.BeginLifetimeScope();
Console.WriteLine(
scope1.ResolveNamed<IEnumerable<IDependency<object>>>("decorated").Count());
// Returns 2 - notice the configAction doesn't even have to do anything
var scope2 = container.BeginLifetimeScope(r => { });
Console.WriteLine(
scope2.ResolveNamed<IEnumerable<IDependency<object>>>("decorated").Count());
And here are my fake types:
interface IDependency<T> { }
class Dependency : IDependency<object> { }
class Decorator<T> : IDependency<T> {}
Any help would be greatly appreciated!

It does just seem like a bug. As a workaround, I ended up doing the following:
var param = new TypedParameter(typeof(IDecoratorDependency), new DecoratorDependency());
var decorated = scope.ResolveNamed<IEnumerable<IDependency<object>>>("decorated", param);
That was good enough for my use case. However, this method is inflexible because it only allows me to supply parameters to the root object, in this case Decorator<T>, but not any of its dependencies.

Related

Unit Test EntityFrameworkCore IEntityTypeConfiguration<T>

How do I unit test EntityFrameworkCore.IEntityTypeConfiguration?
I want to make sure that my configuration is unit tested for several reasons:
avoid the burden of creating integration tests which depend more on the data setup
using a feature flag framework, I want to make sure that I map to different views and tables depending on the status of the flag
I was several custom converters which I must make sure are used
simply I want clear self contained tests to fail in case someone changes something that shouldn't, at that point they must intentionally remove a test
I was unable to find an answer to this question, so I had to dig it out myself.
var sut = new MyEntityTypeConfiguration();
var entityType = new EntityType("MyEntityName", typeof(MyEntityName), new Model(), false, ConfigurationSource.Convention);
var builder = new EntityTypeBuilder<MyEntityName>(entityType);
sut.Configure(builder);
// Metadata will contain all the set configurations
var meta = builder.Metadata;
var properties = builder.Metadata.GetDeclaredProperties();
var viewName = builder.Metadata.GetViewName();
[..]
I would recommend to use a method like this in your testing libraries to simplify the setup
private static EntityTypeBuilder<T> GetEntityBuilder<T>() where T : class
{
#pragma warning disable EF1001 // Internal EF Core API usage.
var entityType = new EntityType(typeof(T).Name, typeof(T), new Model(), false, ConfigurationSource.Convention);
var builder = new EntityTypeBuilder<T>(entityType);
return builder;
#pragma warning restore EF1001 // Internal EF Core API usage.
}
which you would use in this way:
var sut = new MyEntityTypeConfiguration();
var builder = GetEntityBuilder<MyEntity>();
sut.Configure(builder);
var meta = builder.Metadata;
var properties = builder.Metadata.GetDeclaredProperties();
var viewName = builder.Metadata.GetViewName();
Once the configure method is called, the builder Metadata property will contain all the details necessary for the correct EF configuration which are accessible and on which you can perform any desired assertion.

How to mock IConfiguration.GetValue

I tried in vain to mock a top-level (not part of any section) configuration value (.NET Core's IConfiguration). For example, neither of these will work (using NSubstitute, but it would be the same with Moq or any mock package I believe):
var config = Substitute.For<IConfiguration>();
config.GetValue<string>(Arg.Any<string>()).Returns("TopLevelValue");
config.GetValue<string>("TopLevelKey").Should().Be("TopLevelValue"); // nope
// non generic overload
config.GetValue(typeof(string), Arg.Any<string>()).Returns("TopLevelValue");
config.GetValue(typeof(string), "TopLevelKey").Should().Be("TopLevelValue"); // nope
In my case, I also need to call GetSection from this same config instance.
You can use an actual Configuration instance with in-memory data.
//Arrange
var inMemorySettings = new Dictionary<string, string> {
{"TopLevelKey", "TopLevelValue"},
{"SectionName:SomeKey", "SectionValue"},
//...populate as needed for the test
};
IConfiguration configuration = new ConfigurationBuilder()
.AddInMemoryCollection(inMemorySettings)
.Build();
//...
Now it is a matter of using the configuration as desired to exercise the test
//...
string value = configuration.GetValue<string>("TopLevelKey");
string sectionValue = configuration.GetSection("SectionName").GetValue<string>("SomeKey");
//...
Reference: Memory Configuration Provider
I do not have idea about NSubstitute, but this is how we can do in Moq.
Aproach is same in either cases.
GetValue<T>() internally makes use of GetSection().
You can Mock GetSection and return your Own IConfigurationSection.
This includes two steps.
1). Create a mock for IConfigurationSection (mockSection) & Setup .Value Property to return your desired config value.
2). Mock .GetSection on Mock< IConfiguration >, and return the above mockSection.Object
Mock<IConfigurationSection> mockSection = new Mock<IConfigurationSection>();
mockSection.Setup(x=>x.Value).Returns("ConfigValue");
Mock<IConfiguration> mockConfig = new Mock<IConfiguration>();
mockConfig.Setup(x=>x.GetSection(It.Is<string>(k=>k=="ConfigKey"))).Returns(mockSection.Object);
Mock IConfiguration
Mock<IConfiguration> config = new Mock<IConfiguration>();
SetupGet
config.SetupGet(x => x[It.Is<string>(s => s == "DeviceTelemetryContainer")]).Returns("testConatiner");
config.SetupGet(x => x[It.Is<string>(s => s == "ValidPowerStatus")]).Returns("On");
IConfiguration.GetSection<T> must be mocked indirectly. I don't fully understand why because NSubstitute, if I understand correctly, creates its own implementation of an interface you're mocking on the fly (in memory assembly). But this seems to be the only way it can be done. Including a top-level section along with a regular section.
var config = Substitute.For<IConfiguration>();
var configSection = Substitute.For<IConfigurationSection>();
var configSubSection = Substitute.For<IConfigurationSection>();
configSubSection.Key.Returns("SubsectionKey");
configSubSection.Value.Returns("SubsectionValue");
configSection.GetSection(Arg.Is("SubsectionKey")).Returns(configSubSection);
config.GetSection(Arg.Is("TopLevelSectionName")).Returns(configSection);
var topLevelSection = Substitute.For<IConfigurationSection>();
topLevelSection.Value.Returns("TopLevelValue");
topLevelSection.Key.Returns("TopLevelKey");
config.GetSection(Arg.Is<string>(key => key != "TopLevelSectionName")).Returns(topLevelSection);
// GetValue mocked indirectly.
config.GetValue<string>("TopLevelKey").Should().Be("TopLevelValue");
config.GetSection("TopLevelSectionName").GetSection("SubsectionKey").Value.Should().Be("SubsectionValue");
I could imagine in some rare scenarios it is needed but, in my humble opinion, most of the time, mocking IConfiguration highlight a code design flaw.
You should rely as much as possible to the option pattern to provide a strongly typed access to a part of your configuration. Also it will ease testing and make your code fail during startup if your application is misconfigured (instead than at runtime when the code reading IConfiguration is executed).
If you really(really) need to mock it then I would advice to not mock it but fake it with an in-memory configuration as explained in #Nkosi's answer
While Nkosi's answer works great for simple structures, sometimes you want to be able to have more complex objects (like arrays) without repeating the whole section path and to be able to use the expected types themselves. If you don't really care too much about performance (and should you in your unit tests?) then this extension method might be helpful.
public static void AddObject(this IConfigurationBuilder cb, object model) {
cb.AddJsonStream(new MemoryStream(Encoding.UTF8.GetString(JsonConvert.SerializeObject(model))));
}
And then use it like this
IConfiguration configuration = new ConfigurationBuilder()
.AddObject(new {
SectionName = myObject
})
.Build();
Use SetupGet method to mock the Configuration value and return any string.
var configuration = new Mock<IConfiguration>();
configuration.SetupGet(x => x[It.IsAny<string>()]).Returns("the string you want to return");
We need to mock IConfiguration.GetSection wichs is executed within GetValue extension method.
You inject the IConfiguration:
private readonly Mock<IConfiguration> _configuration;
and the in the //Arrange Section:
_configuration.Setup(c => c.GetSection(It.IsAny())).Returns(new Mock().Object);
It worked like a charm for me.
I've found this solution to work reliably for me for my XUnit C# tests & corresponding C# code:
In Controller
string authConnection = this._config["Keys:AuthApi"] + "/somePathHere/Tokens/jwt";
In XUnit Test
Mock<IConfiguration> mockConfig = new Mock<IConfiguration>();
mockConfig.SetupGet(x => x[It.Is<string>(s => s == "Keys:AuthApi")]).Returns("some path here");

Custom JsonConverter added to JsonSerializerOptions.Converters not used by default

I have implemented a custom json converter for one of my classes since I want to use the constructor for deserialization.
I have added the custom converter to my Startup.cs class:
services.AddControllers()
.AddJsonOptions(opt =>
{
opt.JsonSerializerOptions.WriteIndented = false;
opt.JsonSerializerOptions.Converters.Add(new MyJsonConverter());
});
My custom json converter looks like this:
public class MyJsonConverter : JsonConverter<MyClass>
{
...
}
The problem is that when running
var result = await JsonSerializer.DeserializeAsync<RoleEntity[]>(json);
my custom json converter isn't triggered.
However, if instead specifying it explicitly like the following
var serializerOptions = new JsonSerializerOptions
{
Converters = { new MyJsonConverter() }
};
var result = await JsonSerializer.DeserializeAsync<RoleEntity[]>(json, serializerOptions);
it works.
I just can't see what the problem is. Am I missing something trivial here?
This is the expected behaviour. When you use JsonSerializer.DeserializeAsync directly without specifying any options it, well, won't use any custom options. The options you add in are for MVC (see this answer). You need to either continue using it as you wrote or register options as singleton which you could then resolve/inject and use within your JsonSerializer.DeserializeAsync call.
To add a new singleton you can use IServiceCollection.AddSingleton.
There are many ways to get the result you want (singleton options is just one of them) but the important part of this answer is that the options you define with .AddJsonOptions will not be used for direct calls to JsonSerializer-methods.

How to resolve list of dependencies in Autofac?

I want register type, than resolve type, and then register instance using resolved values. Something like this:
//Register type:
builder.RegisterType<ValidateImportMandatoryColumns>().Named<IValidateImport>("MandatoryColumn").As<IValidateImport>();
builder.RegisterType<ValidateImportNonMandatoryColumns>().Named<IValidateImport>("NonMandatoryColumns").As<IValidateImport>();
//Resolve
var t1 = Container.ResolveNamed<IValidateImport>("MandatoryColumn");
var t2 = Container.ResolveNamed<IValidateImport>("NonMandatoryColumns");
//Create list with resolved values:
List<IValidateImport> allValidators = new List<IValidateImport>(){t1,t2};
//Register Instance:
builder.RegisterInstance(allValidators).As<List<IValidateImport>>();
This is not working. I can't resolve and than register again. Do you know how to do this with Autofac? Maybe is approach wrong, so please tell me if you have better idea. Goal is to inject list of validators with different types that use same interface.
Autofac has built in support for collection. If you want to resolve all IValidateImport, you can resolve IEnumerable<IValidateImport>
var allValidators = container.Resolve<IEnumerable<IValidateImport>>();
See Implicit Relationship Types for more information for more information.
By the way, if you want to update a container, which is not required here, you can use the following piece of code.
var builder = new ContainerBuilder();
// do some registration
var container = builder.Build();
var updater = new ContainerBuilder();
// do other registraitons
// update the container
updater.Update(container);

When do I pass the RegistrationBuilder to Catalog?

With the new fluent MEF programming model, if I have multiple catalogs:
To which catalog do I pass my RegistrationBuilder to???
Do I need to pass RegistrationBuilder to SatisfyImportsOnce call?
Which of SatisfyImportsOnce or ComposeParts do I use? (has anything changed with this in fluent mef?)
E.g. Here is an example to llustrate my confusion (see comments on the r.h.s):
// Get pre-wired registration builder
RegistrationBuilder rb = new MefCompositionRoot().CommonRegistrationBuilder();
// Register this WCF service class
rb.ForType<LogService>().Export<LogService>();
var assembly = typeof (LogService).Assembly;
var assemblyCatalog = new AssemblyCatalog(assembly, rb); // <-- HERE?
var dirCatalog = new DirectoryCatalog("bin", rb); // <-- and HERE?
// Combine catalogs
var catalog = new AggregateCatalog();
catalog.Catalogs.Add(dirCatalog);
catalog.Catalogs.Add(assemblyCatalog);
var container = new CompositionContainer(catalog);
container.SatisfyImportsOnce(this, rb);// THIS?
container.ComposeParsts(this); ///or THIS?
You can put the RegistrationBuilder to any of the catalogs you want to use MEF Conventions with. If you want to use only the Conventions Model, then use it in all catalogs. If the DirectoryCatalog in your sample code, will load assemblies containing only Attributed exports/imports then you do not need the RegistrationBuilder. Note that both models (Attributed and Contentions) can coexist. So, you could add the RegistrationBuilder to all catalogs unless there might be a type that satisfies a rule (configured using the RegistrationBuilder) that you do not want to use in your CompositionContainer.
This overload of SatisfyImportsOnce is a riddle. From the documentation (and from a quick look at MEF's source) it looks like you can use a specific RegistrationBuilder ad-hoc. In reality I have only managed to use it with the code sample that follows.
SatisfyImportsOnce disables recomposition. Check out this excellent answer on this subject.
Sample using SatisfyImportsOnce(Object, ReflectionContext)
private static void TestLateRegistration_SameBuilder_Ok()
{
var rb = new RegistrationBuilder();
var assemblyCatalog = new AssemblyCatalog(typeof(LogService).Assembly, rb);
using (var container = new CompositionContainer(assemblyCatalog))
{
rb.ForType<LogService>().Export();
var server = new TypeImportingLogService();
//Use the same RegistrationBuilder.
container.SatisfyImportsOnce(server, rb);
}
}

Categories

Resources