I'm looking for advice on this problem and whether service locator and class naming conventions are an ok solution (I tend to avoid these anti-patterns), and potential performance ramifications.
An app has a collection of objects implementing the same interface, distinguished by name. For example:
public interface IDog {
void Bark();
}
public class Pug: IDog {
public void Bark() {
// Pug bark implementation
}
}
public class Beagle: IDog {
public void Bark() {
// Beagle bark implementation
}
}
In the code, when you need an IDog, you only know a string name that is passed to you, for example "Pug" or "Beagle". In this case the string may contain special characters (example: <breed:pug />)
There are a few proposed solutions that have come about:
Using reflection, find the implementation needed where the string name == implementation name.
Add an addribute to each class, use reflection where string name == attribute property. Ex [DogBreed("Pug")]
Add a Breed property to the IDog interface. Inject a IList into a factory class, and have it retrieve the matching dog. Ex.
Private IList _dogs;
Public DogFactory(IList<IDog> dogs) {
_dogs = dogs;
}
Public IDog GetDog(string dogBreed) {
return _dogs.First(x => x.Breed == dogBreed);
}
1 and 2 use service locator. 1 uses an implied naming convention that you will only know by seeing the reflection code. 3 the concern is that all of the objects will be built in memory even though you only need a single implementation.
I personally have leaned towards #3 in the past. Object creation should be cheap. However, this is a legacy web app and objects down the chain may have heavy initialization cost. This application uses Unity for IoC.
Option 1.
This option sounds like the Partial Type Name Role Hint idiom. If you inject the list of candidates and find the appropriate Strategy among those candidates, it's just plain old Constructor Injection, and has nothing to do with Service Locator (which is a good thing).
Option 2.
This option sounds like the Metadata Role Hint idiom. Again, if you inject the list of candidates via the constructor, Service Locator is nowhere to be seen.
Option 3.
This options sounds like a variation of the Role Interface Role Hint idiom. Still supports use of good old Constructor Injection.
Personally, I tend to favour Partial Type Name Role Hint because this design doesn't impact the implementation of any business logic. All the selection logic becomes a pure infrastructure concern, and can be defined independently of the implementations and clients.
When it comes to the cost of composing the relevant object graphs, there are ways to address any issues in clean ways.
I have searched a little bit and I am confused.
First Approach uses a repository and a service for each entity with Autofac. Unit of work class does not have repositories. Therefore, you should create each repository instead of just creating one unit of work class in caller constructor.
OrderService(IUnitOfWork unitOfWork, IUserRepository userRepository,IOrderRepository orderRepository,IBalanceRepository balanceRepository)
Second Approach uses just a generic repository. It uses extension classes instead of using one repository for each entity. Unit of work class has generic repositories.Therefore, you can just create a unit of work class on caller class constructor.
OrderService(IUnitOfWork unitOfWork)
In this approach we use one generic class for repositories but we create a repository object for each entity. If this approach is fine how can I implement it with Autofac ?
Third Approach uses one generic repository and one object for generic repository with Autofac. It uses generic methods instead of generic class. But generic repository has unit of work class instead of opposite. Is this anti pattern ?
OrderService(IUnitOfWork unitOfWork,IGenericRepository repository)
Which approach should I use ?
The sense of the unit of work and the repository pattern is to describe exactly what is needed for example for a use case. So the unit of work which has repositories for every entity or can create a repository for any entity by a generic method is as much to avoid, as a repository that returns an IQueryable. The last flaw will move your Dal to your domain model or even UI (imagine where the filter logic is written and exactly when the filter is executed and where any exceptions will be thrown), the first creates a kind of monolytic application and makes it hard to write unit tests. A unit of work (interface) that only has the 3 Repositories your use case needs and the three repository (interfaces) that have only the methods needed, returning either a single object or a list of objects is more easy to mock and test and specifies exactly what the use case needs and communicates it to your fellow developers (or yourself in 2 years). The interfaces could be implemented (if you choose to) by one big unit of work class and perhaps a few standard repository classes, but thats a different decision that should be guided by the technology (EF code first in earlier versions was not able to have multiple contexts in one database) and the complexity of your application.
i use the 2nd,IUnitOfWork is only an interface,the behind i use ef.
public interface IUnitOfWork:IDisposable
{
IRepository<T> GetRepository<T>() where T:class;
int SaveChanges();
}
and create dbcontext class
public class DataContext:DbContext,IUnitOfWork
implement the GetRepository method,you should add a adapter class,from dbset to irepository
public IRepository<T> GetRepository<T>() where T:class
{
return new RepositoryEfAdapter<T>(Set<T>(), this);
}
this is a sample,you can register datacontext as iunitofwork
I'm trying to get dependencies set up correctly in my Workflow application. It seems the best way to do this is using the Service Locator pattern that is provided by Workflow's WorkflowExtensions.
My workflow uses two repositories: IAssetRepository and ISenderRepository. Both have implementations using Entity Framework: EFAssetRepository, and EFSenderRepository, but I'd like both to use the same DbContext.
I'm having trouble getting both to use the same DbContext. I'm used to using IoC for dependency injection, so I thought I'd have to inject the DbContext into the EF repositories via their constructor, but this seems like it would be mixing the service locator and IoC pattern, and I couldn't find an easy way to achieve it, so I don't think this is the way forward.
I guess I need to chain the service locator calls? So that the constructor of my EF repositories do something like this:
public class EFAssetRepository
{
private MyEntities entities;
public EFAssetRepository()
{
this.entities = ActivityContext.GetExtension<MyEntities>();
}
}
Obviously the above won't work because the reference to ActivityContext is made up.
How can I achieve some form of dependency chain using the service locator pattern provided for WF?
Thanks,
Nick
EDIT
I've posted a workaround for my issue below, but I'm still not happy with it. I want the code activity to be able to call metadata.Require<>(), because it should be ignorant of how extensions are loaded, it should just expect that they are. As it is, my metadata.Require<> call will stop the workflow because the extension appears to not be loaded.
It seems one way to do this is by implementing IWorkflowInstanceExtension on an extension class, to turn it into a sort of composite extension. Using this method, I can solve my problem thus:
public class UnitOfWorkExtension : IWorkflowInstanceExtension, IUnitOfWork
{
private MyEntities entities = new MyEntities();
IEnumerable<object> IWorkflowInstanceExtension.GetAdditionalExtensions()
{
return new object[] { new EFAssetRepository(this.entities), new EFSenderRepository(this.entities) };
}
void IWorkflowInstanceExtension.SetInstance(WorkflowInstanceProxy instance) { }
public void SaveChanges()
{
this.entities.SaveChanges();
}
}
The biggest downside to doing it this way is that you can't call metadata.RequireExtension<IAssetRepository>() or metadata.RequireExtension<ISenderRepository>() in the CacheMetadata method of a CodeActivity, which is common practice. Instead, you must call metadata.RequireExtension<IUnitOfWork>(), but it is still fine to do context.GetExtension<IAssetRepository>() in the Execute() method of the CodeActivity. I imagine this is because the CacheMetadata method is called before any workflow instances are created, and if no workflow instances are created, the extension factory won't have been called, and therefore the additional extensions won't have been loaded into the WorkflowInstanceExtensionManager, so essentially, it won't know about the additional extensions until a workflow instance is created.
I am about to implement an Entity Framework 6 design with a repository and unit of work.
There are so many articles around and I'm not sure what the best advice is: For example I realy like the pattern implemented here: for the reasons suggested in the article here
However, Tom Dykstra (Senior Programming Writer on Microsoft's Web Platform & Tools Content Team) suggests it should be done in another article: here
I subscribe to Pluralsight, and it is implemented in a slightly different way pretty much every time it is used in a course so choosing a design is difficult.
Some people seem to suggest that unit of work is already implemented by DbContext as in this post, so we shouldn't need to implement it at all.
I realise that this type of question has been asked before and this may be subjective but my question is direct:
I like the approach in the first (Code Fizzle) article and wanted to know if it is perhaps more maintainable and as easily testable as other approaches and safe to go ahead with?
Any other views are more than welcome.
#Chris Hardie is correct, EF implements UoW out of the box. However many people overlook the fact that EF also implements a generic repository pattern out of the box too:
var repos1 = _dbContext.Set<Widget1>();
var repos2 = _dbContext.Set<Widget2>();
var reposN = _dbContext.Set<WidgetN>();
...and this is a pretty good generic repository implementation that is built into the tool itself.
Why go through the trouble of creating a ton of other interfaces and properties, when DbContext gives you everything you need? If you want to abstract the DbContext behind application-level interfaces, and you want to apply command query segregation, you could do something as simple as this:
public interface IReadEntities
{
IQueryable<TEntity> Query<TEntity>();
}
public interface IWriteEntities : IReadEntities, IUnitOfWork
{
IQueryable<TEntity> Load<TEntity>();
void Create<TEntity>(TEntity entity);
void Update<TEntity>(TEntity entity);
void Delete<TEntity>(TEntity entity);
}
public interface IUnitOfWork
{
int SaveChanges();
}
You could use these 3 interfaces for all of your entity access, and not have to worry about injecting 3 or more different repositories into business code that works with 3 or more entity sets. Of course you would still use IoC to ensure that there is only 1 DbContext instance per web request, but all 3 of your interfaces are implemented by the same class, which makes it easier.
public class MyDbContext : DbContext, IWriteEntities
{
public IQueryable<TEntity> Query<TEntity>()
{
return Set<TEntity>().AsNoTracking(); // detach results from context
}
public IQueryable<TEntity> Load<TEntity>()
{
return Set<TEntity>();
}
public void Create<TEntity>(TEntity entity)
{
if (Entry(entity).State == EntityState.Detached)
Set<TEntity>().Add(entity);
}
...etc
}
You now only need to inject a single interface into your dependency, regardless of how many different entities it needs to work with:
// NOTE: In reality I would never inject IWriteEntities into an MVC Controller.
// Instead I would inject my CQRS business layer, which consumes IWriteEntities.
// See #MikeSW's answer for more info as to why you shouldn't consume a
// generic repository like this directly by your web application layer.
// See http://www.cuttingedge.it/blogs/steven/pivot/entry.php?id=91 and
// http://www.cuttingedge.it/blogs/steven/pivot/entry.php?id=92 for more info
// on what a CQRS business layer that consumes IWriteEntities / IReadEntities
// (and is consumed by an MVC Controller) might look like.
public class RecipeController : Controller
{
private readonly IWriteEntities _entities;
//Using Dependency Injection
public RecipeController(IWriteEntities entities)
{
_entities = entities;
}
[HttpPost]
public ActionResult Create(CreateEditRecipeViewModel model)
{
Mapper.CreateMap<CreateEditRecipeViewModel, Recipe>()
.ForMember(r => r.IngredientAmounts, opt => opt.Ignore());
Recipe recipe = Mapper.Map<CreateEditRecipeViewModel, Recipe>(model);
_entities.Create(recipe);
foreach(Tag t in model.Tags) {
_entities.Create(tag);
}
_entities.SaveChanges();
return RedirectToAction("CreateRecipeSuccess");
}
}
One of my favorite things about this design is that it minimizes the entity storage dependencies on the consumer. In this example the RecipeController is the consumer, but in a real application the consumer would be a command handler. (For a query handler, you would typically consume IReadEntities only because you just want to return data, not mutate any state.) But for this example, let's just use RecipeController as the consumer to examine the dependency implications:
Say you have a set of unit tests written for the above action. In each of these unit tests, you new up the Controller, passing a mock into the constructor. Then, say your customer decides they want to allow people to create a new Cookbook or add to an existing one when creating a new recipe.
With a repository-per-entity or repository-per-aggregate interface pattern, you would have to inject a new repository instance IRepository<Cookbook> into your controller constructor (or using #Chris Hardie's answer, write code to attach yet another repository to the UoW instance). This would immediately make all of your other unit tests break, and you would have to go back to modify the construction code in all of them, passing yet another mock instance, and widening your dependency array. However with the above, all of your other unit tests will still at least compile. All you have to do is write additional test(s) to cover the new cookbook functionality.
I'm (not) sorry to say that the codefizzle, Dyksta's article and the previous answers are wrong. For the simple fact that they use the EF entities as domain (business) objects, which is a big WTF.
Update: For a less technical explanation (in plain words) read Repository Pattern for Dummies
In a nutshell, ANY repository interface should not be coupled to ANY persistence (ORM) detail. The repo interface deals ONLY with objects that makes sense for the rest of the app (domain, maybe UI as in presentation). A LOT of people (with MS leading the pack, with intent I suspect) make the mistake of believing that they can reuse their EF entities or that can be business object on top of them.
While it can happen, it's quite rare. In practice, you'll have a lot of domain objects 'designed' after database rules i.e bad modelling. The repository purpose is to decouple the rest of the app (mainly the business layer) from its persistence form.
How do you decouple it when your repo deals with EF entities (persistence detail) or its methods return IQueryable, a leaking abstraction with wrong semantics for this purpose (IQueryable allows you to build a query, thus implying that you need to know persistence details thus negating the repository's purpose and functionality)?
A domin object should never know about persistence, EF, joins etc. It shouldn't know what db engine you're using or if you're using one. Same with the rest of the app, if you want it to be decoupled from the persistence details.
The repository interface know only about what the higher layer know. This means, that a generic domain repository interface looks like this
public interface IStore<TDomainObject> //where TDomainObject != Ef (ORM) entity
{
void Save(TDomainObject entity);
TDomainObject Get(Guid id);
void Delete(Guid id);
}
The implementation will reside in the DAL and will use EF to work with the db. However the implementation looks like this
public class UsersRepository:IStore<User>
{
public UsersRepository(DbContext db) {}
public void Save(User entity)
{
//map entity to one or more ORM entities
//use EF to save it
}
//.. other methods implementation ...
}
You don't really have a concrete generic repository. The only usage of a concrete generic repository is when ANY domain object is stored in serialized form in a key-value like table. It isn't the case with an ORM.
What about querying?
public interface IQueryUsers
{
PagedResult<UserData> GetAll(int skip, int take);
//or
PagedResult<UserData> Get(CriteriaObject criteria,int skip, int take);
}
The UserData is the read/view model fit for the query context usage.
You can use directly EF for querying in a query handler if you don't mind that your DAL knows about view models and in that case you won't be needing any query repo.
Conclusion
Your business object shouldn't know about EF entities.
The repository will use an ORM, but it never exposes the ORM to the rest of the app, so the repo interface will use only domain objects or view models (or any other app context object that isn't a persistence detail)
You do not tell the repo how to do its work i.e NEVER use IQueryable with a repo interface
If you just want to use the db in a easier/cool way and you're dealing with a simple CRUD app where you don't need (be sure about it) to maintain separation of concerns then skip the repository all together, use directly EF for everything data. The app will be tightly coupled to EF but at least you'll cut the middle man and it will be on purpose not by mistake.
Note that using the repository in the wrong way, will invalidate its use and your app will still be tightly coupled to the persistence (ORM).
In case you believe the ORM is there to magically store your domain objects, it's not. The ORM purpose is to simulate an OOP storage on top of relational tables. It has everything to do with persistence and nothing to do with domain, so don't use the ORM outside persistence.
DbContext is indeed built with the Unit of Work pattern. It allows all of its entities to share the same context as we work with them. This implementation is internal to the DbContext.
However, it should be noted that if you instantiate two DbContext objects, neither of them will see the other's entities that they are each tracking. They are insulated from one another, which can be problematic.
When I build an MVC application, I want to ensure that during the course of the request, all my data access code works off of a single DbContext. To achieve that, I apply the Unit of Work as a pattern external to DbContext.
Here is my Unit of Work object from a barbecue recipe app I'm building:
public class UnitOfWork : IUnitOfWork
{
private BarbecurianContext _context = new BarbecurianContext();
private IRepository<Recipe> _recipeRepository;
private IRepository<Category> _categoryRepository;
private IRepository<Tag> _tagRepository;
public IRepository<Recipe> RecipeRepository
{
get
{
if (_recipeRepository == null)
{
_recipeRepository = new RecipeRepository(_context);
}
return _recipeRepository;
}
}
public void Save()
{
_context.SaveChanges();
}
**SNIP**
I attach all my repositories, which are all injected with the same DbContext, to my Unit of Work object. So long as any repositories are requested from the Unit of Work object, we can be assured that all our data access code will be managed with the same DbContext - awesome sauce!
If I were to use this in an MVC app, I would ensure the Unit of Work is used throughout the request by instantiating it in the controller, and using it throughout its actions:
public class RecipeController : Controller
{
private IUnitOfWork _unitOfWork;
private IRepository<Recipe> _recipeService;
private IRepository<Category> _categoryService;
private IRepository<Tag> _tagService;
//Using Dependency Injection
public RecipeController(IUnitOfWork unitOfWork)
{
_unitOfWork = unitOfWork;
_categoryRepository = _unitOfWork.CategoryRepository;
_recipeRepository = _unitOfWork.RecipeRepository;
_tagRepository = _unitOfWork.TagRepository;
}
Now in our action, we can be assured that all our data access code will use the same DbContext:
[HttpPost]
public ActionResult Create(CreateEditRecipeViewModel model)
{
Mapper.CreateMap<CreateEditRecipeViewModel, Recipe>().ForMember(r => r.IngredientAmounts, opt => opt.Ignore());
Recipe recipe = Mapper.Map<CreateEditRecipeViewModel, Recipe>(model);
_recipeRepository.Create(recipe);
foreach(Tag t in model.Tags){
_tagRepository.Create(tag); //I'm using the same DbContext as the recipe repo!
}
_unitOfWork.Save();
Searching around the internet I found this http://www.thereformedprogrammer.net/is-the-repository-pattern-useful-with-entity-framework/ it's a 2 part article about the usefulness of the repository pattern by Jon Smith.
The second part focuses on a solution. Hope it helps!
Repository with unit of work pattern implementation is a bad one to answer your question.
The DbContext of the entity framework is implemented by Microsoft according to the unit of work pattern. That means the context.SaveChanges is transactionally saving your changes in one go.
The DbSet is also an implementation of the Repository pattern. Do not build repositories that you can just do:
void Add(Customer c)
{
_context.Customers.Add(c);
}
Create a one-liner method for what you can do inside the service anyway ???
There is no benefit and nobody is changing EF ORM to another ORM nowadays...
You do not need that freedom...
Chris Hardie is argumenting that there could be instantiated multiple context objects but already doing this you do it wrong...
Just use an IOC tool you like and setup the MyContext per Http Request and your are fine.
Take ninject for example:
kernel.Bind<ITeststepService>().To<TeststepService>().InRequestScope().WithConstructorArgument("context", c => new ITMSContext());
The service running the business logic gets the context injected.
Just keep it simple stupid :-)
You should consider "command/query objects" as an alternative, you can find a bunch of interesting articles around this area, but here is a good one:
https://rob.conery.io/2014/03/03/repositories-and-unitofwork-are-not-a-good-idea/
When you need a transaction over multiple DB objects, use one command object per command to avoid the complexity of the UOW pattern.
A query object per query is likely unnecessary for most projects. Instead you might choose to start with a 'FooQueries' object
...by which I mean you can start with a Repository pattern for READS but name it as "Queries" to be explicit that it does not and should not do any inserts/updates.
Later, you might find splitting out individual query objects worthwhile if you want to add things like authorization and logging, you could feed a query object into a pipeline.
I always use UoW with EF code first. I find it more performant and easier tot manage your contexts, to prevent memory leaking and such. You can find an example of my workaround on my github: http://www.github.com/stefchri in the RADAR project.
If you have any questions about it feel free to ask them.
I have a WPF application based on MVVM with Caliburn.Micro and Ninject. I have a root viewmodel called ShellViewModel. It has a couple of dependencies (injected via constructor) which are configured in Caliburn's Bootstrapper. So far so good.
Somewhere down the line, there is a MenuViewModel with a couple of buttons, that in turn open other viewmodels with their own dependencies. These viewmodels are not created during creation of the root object, but I still want to inject dependencies into them from my IoC container.
I've read this question on service locator vs dependency injection and I understand the points being made.
I'm under the impression however that my MenuViewModel needs to be able to access my IoC container in order the properly inject the viewmodels that are being made dynamically..which is something I'm trying to avoid. Is there another way?
Yes, I believe you can do something a bit better.
Consider that if there was no on-demand requirement then obviously you could make those viewmodels be dependencies of MenuViewModel and so on up the chain until you get to the root of the object graph (the ShellViewModel) and the container would wire everything up.
You can put a "firewall" in the object graph by substituting something that can construct the dependencies of MenuViewModel for the dependencies themselves. The container is the obvious choice for this job, and IMHO from a practical standpoint this is a good enough solution even if it's not as pure.
But you can also substitute a special-purpose factory instead of the container; this factory would take a dependency on the container and provide read-only properties for the real dependencies of MenuViewModel. Accessing the properties would result in having the container resolve the objects and returning them (accessor methods would also work instead of properties; what's more appropriate is another discussion entirely, so just use whatever you think is better).
It may look like that you haven't really changed the status quo, but the situation is not the same it would be if MenuViewModel took a direct dependency on the container. In that case you would have no idea what the real dependencies of MenuViewModel are by looking at its public interface, while now you would see that there's a dependency on something like
interface IMenuViewModelDependencyFactory
{
public RealDependencyA { get; }
public RealDependencyB { get; }
}
which is much more informative. And if you look at the public interface of the concrete MenuViewModelDependencyFactory things are also much better:
class MenuViewModelDependencyFactory : IMenuViewModelDependencyFactory
{
private Container container;
public MenuViewModelDependencyFactory(Container container) { ... }
public RealDependencyA { get { ... } }
public RealDependencyB { get { ... } }
}
There should be no confusion over what MenuViewModelDependencyFactory intends to do with the container here because it's so very highly specialized.