I have problem with updating entites that have many-to many relationship. Below my User and category class:
public class User : IEntity
{
[Key]
public virtual long Id { get; set; }
private ICollection<Category> _availableCategories;
public virtual ICollection<Category> AvailableCategories
{
get { return _availableCategories ?? (_availableCategories = new List<Category>()); }
set { _availableCategories = value; }
}
}
public class Category : IEntity
{
[Key]
public long Id { get; set; }
/// <summary>
/// Full name or description of a category
/// </summary>
[StringLength(255)]
public string FullName { get; set; }
}
This is code snippet from my repository
public override void Edit(User user)
{
var dbUser = _context.Users.Include(x => x.AvailableCategories)
.Single(x => x.Id == user.Id);
var categories = _context.Categories;
dbUser.AvailableCategories.Clear();
foreach (var cat in user.AvailableCategories)
{
dbUser.AvailableCategories.Add(cat);
}
_context.Entry(dbUser).State = EntityState.Modified;
}
However the categories don't get updated. What EF does is insert empty rows into category table and sets relations to this new rows with user.
How can I update User so that I change only categories that already exist in the database?
User that I pass to Edit method has AvailableCategories with only Ids set (rest of properties are empty).
When you're doing something like posting back M2M relationships, you either must post the full object, as in every single property on those objects, or simply post a list of ids and then use those to query the associated objects back from the database. Otherwise, Entity Framework understands your purpose to be to update the properties on the objects as well, in this case with empty values.
Obviously the first option is quite unwieldy, so the second way is the preferred and standard way. Generally, for this, you'd want to use a view model so you could have a property like the following, that you would post into:
public List<long> SelectedCategories { get; set; }
But, if you insist on using the entity directly, you can get much the same result by simply doing:
var selectedCategories = user.AvailableCategories.Select(m => m.Id)
Once you have the ids:
var newAvailableCategories = _context.Categories.Where(m => selectedCategories.Contains(m.Id));
And then finally set that on your user:
dbUser.AvailableCategories = newAvailableCategories;
I notice you are also adding the user.AvailableCategories directly into dbUser.AvailableCategories. I've noticed when binding back complex objects from an MVC view that DB Entities are no longer attached to the DbContext. If you look at the entity, you can verify by checking dbContext.Entry(cat).State is "detached" (or something unexpected) I believe.
You must query those entities back out of the dbContext (possibly by using the returned cat.Id's). Or otherwise manually set the entities as "unchanged". And then add those "non-detached" items into dbUser.AvailableCategories. Please see Chris's answer as it shows with specific code how to get this done.
Also, I might use a linking entity. Possibly something like this:
public class UserCategory
{
public User User {get;set;}
public Category Category {get;set;}
}
And add it to DB context. Also, drop the linking lists in your current User and Category class. This way you can manipulate the UserCategory class (and DbSet) to manage your many-to-many relationship.
Related
I'm using Entity Framework Database First approach. Let's say I have a model class called Product and that class has a NumberOfViews property. In the Edit page I pass an instance of the product class to the controller.
The problem is I can't add #Html.EditorFor(model => model.NumberOfViews) in the Edit page, because it's supposed that NumberOfViews is updated with every visit to the product page, and NOT by the website Admin.
And I can't add it as #Html.HiddenFor(model => model.NumberOfViews), because if the Admin Inspected the element, he can edit it manually.
Also If I try to programmatically set the value on the server-side (e.g., Product.NumberOfViews = db.Products.Find(Product.Id).NumberOfViews;), I get the following error:
An object with the same key already exists in the ObjectStateManager. The ObjectStateManager cannot track multiple objects with the same key.
And if I don't add it to either the view or the controller, the value will be null, thus overriding any previous value.
So what should I do?
I have noticed a lot of people use the same model for their Entity Framework as they do for their MVC Controller. I generally discourage this practice. In my opinion, a database model is not the same as a view model.
Sometimes a view needs less information than what the database model is supplying. For example while modifying account password, view does not need first name, last name, or email address even though they may all reside in the same table.
Sometimes it needs information from more than one database table. For example if a user can store unlimited number of telephone numbers for their profile, then user information will be in user table and then contact information with be in contact table. However when modifying user profile, they may want to add/edit/delete one or more of their numbers, so the view needs all of the numbers along with first name, last name and email address.
This is what I would do in your case:
// This is your Entity Framework Model class
[Table("Product")]
public class Product
{
[DatabaseGenerated(DatabaseGeneratedOption.Identity)]
public int ProductId { get; set; }
public string Name { get; set; }
public int NumberOfPageViews { get; set; }
}
// This is the model you will use in your Edit action.
public class EditProductViewModel
{
public int ProductId { get; set; }
public string Name { get; set; }
}
public class ProductController : Controller
{
IProductService service;
//...
[HttpGet]
public ActionResult Edit(int productId)
{
var product = service.GetProduct(productId);
var model = new EditProductViewModel()
{
ProductId = product.ProductId,
Name = product.Name
};
return View(model);
}
[HttpPost]
public ActionResult Edit(EditProductViewModel model)
{
if (ModelState.IsValid)
{
var product = service.GetProduct(model.ProductId);
product.Name = model.Name;
service.Update(product);
}
// ...
}
}
I have a User table and an Avatar table. One User can have many avatars (or null). But I need to mark which avatar is the current, so I have an Avatar_Id in User table that is the current avatar. And a ForeignKey User_Id in Avatar to tell me which User is the owner.
Trying to do that is generating me a lot of errors and headaches when I try to populate some data in order to test the relationship.
public class User
{
[Key]
public int Id { get; set; }
public Avatar Avatar { get; set; }
public virtual ICollection<Avatar> Avatars { get; set; }
}
public class Avatar
{
[Key, ForeignKey("User")]
public int Id { get; set; }
public User User { get; set; }
}
Test part:
var user = new User();
var avatar = new Avatar()
{
User = user
};
// user.Avatar = avatar; // <- this gives [a circular] error; without this I have null.
db.Users.Add(user);
db.Avatars.Add(avatar);
db.SaveChanges();
This is resulting me with Avatar_Id = NULL within User table, and User_Id = NULL in Avatar table. I expected these fields filled (well, Avatar_Id can be null).
Its better to make boolean field 'IsDefault' in table with avatar and check while add/update avatars that no more default avatars for this user. Also you can add same property in avatar class.
#Fabricio I can't test this code before post, but I'm pretty convinced it will work.
public class User
{
[Key]
public int UserId { get; set; }
public int AvatarId { get; set; }
[ForeignKey("AvatarId")]
public Avatar Avatar { get; set; }
public ICollection<Avatar> Avatars { get; set; }
}
public class Avatar
{
[Key]
public int AvatarId { get; set; }
[ForeignKey("User")]
public int UserId { get; set; }
public User User { get; set; }
}
The problem is when you put two foreign keys merge like one. Now you have a foreign key in Avatar table and other in User table, each one represents one mode of relationship.
The foreign key "AvatarId" represents a special form of foreign key, a unique + foreign key, (a second form to build the one to one relationship). You can read more about this in here: http://weblogs.asp.net/manavi/archive/2011/05/01/associations-in-ef-4-1-code-first-part-5-one-to-one-foreign-key-associations.aspx
I've given this a bit of though because I modeled a similar case once and didn't mind to reevaluate the options.
Look closely at your premises:
One User can have many avatars (or null)
This short sentence implies that the one-to-one association User-Avatar must be optional both ways, because a User without Avatars can’t possibly refer to one its own avatars, and when a user has more than one avatar only one of them can refer to User as being the user's default. (They all refer to user as owner).
So you can only model it as a 0..1 – 0..1 association. So Avatar’s primary key can't be a foreign key to user. (It couldn't anyway, otherwise a user could only have one avatar).
Maybe this could have been done by Jonny Piazzi's model if this wouldn't throw the infamous "may cause cycles or multiple cascade paths" exception. Both user and Avatar refer to one another and you have to tell EF explicitly which of the FKs is not cascading. This can only be done by fluent mapping:
protected override void OnModelCreating(DbModelBuilder modelBuilder)
{
modelBuilder.Entity<User>()
.HasOptional(u => u.Avatar)
.WithOptionalDependent()
.Map(m => m.MapKey("AvatarId"))
.WillCascadeOnDelete(false);
...
}
This puts a nullable, non-cascading FK column AvatarId in User (that's why User is the dependent of Avatar).
Now your second issue, the chicken-eg problem when populating the model.
This can only be done when you call SaveChanges twice and wrap these calls in a transaction scope. For example:
using (var tran = new TransactionScope())
{
var user = new User();
var avatar = new Avatar();
user.Avatars = new HashSet<Avatar>();
user.Avatars.Add(avatar);
user.Avatars.Add(new Avatar());
user.Avatars.Add(new Avatar());
db.Users.Add(user);
db.SaveChanges();
user.Avatar = avatar; // set FK
db.SaveChanges();
tran.Complete();
}
Now EF can decide which key to generate first (User's) before referring to it by foreign keys. Subsequently you set the FK in User.
But... is this the best model?
Maybe, maybe not.
The issue is that your model does not enforce the business rule that a user can only have one of its own avatars as default avatar. User.AvatarId can refer to any avatar. So you have to write business logic to enforce the business rule.
With YD1m's solution (no User.AvatarId, but a column Avatar.IsDefault) this business rule is enforced implicitly. But now you have to write business logic to enforce that only one avatar is the default.
It's up to you to decide what you think is more feasible.
(for the record: way back, I took the latter option)
I have a user class in EF Code First that contains a lot of properties, and each user has a collection of "Contacts" which are other users as a subset of the total user population. The other collection "ContactOfOthers" is just the reverse showing who has this user as a contact as this is a many-to-many relationship.
public class User {
[Key]
public string UserName { get; set; }
// Lots of other properties not relevant to this question...
[NotMapped]
public bool IsMyContact { get; set; }
public virtual ICollection<User> Contacts { get; set; }
public virtual ICollection<User> ContactOfOthers { get; set; }
}
I introduced a not-mapped (not mapped to DB) property called IsMyContact. This is for cases when the user queries for a bunch of users and I need to show in the View which users are already in their contacts list. So this property should be true if the User is part of their "Contacts" collection. It shouldn't be saved to the DB since it can be different for the same user, depending on the user doing the query.
Is there a nice way to do this in a query from the context? It could of course be brute-forced by doing two queries then iterating through the main one, looking for matches to the user's Contacts collection, but I'm wondering if there's a more elegant way to do this from one query, projecting a run-time computed column onto this property?
I don't know a way how to populate the IsMyContact property in the User directly within the query. But an alternative approach could be to introduce a ViewModel which wraps the User and has in addition the IsMyContact flag:
public class UserViewModel
{
public bool IsMyContact { get; set; }
public User User { get; set; }
}
(The class User would not have the IsMyContact flag anymore.)
You could then project into this type when you run your query:
string me = "That's me"; // name of the user who is selecting
List<UserViewModel> list = context.Users
.Where(u => ...some filter...)
.Select(u => new UserViewModel
{
IsMyContact = u.ContactOfOthers.Any(c => c.UserName == me),
User = u
})
.ToList();
The benefits would be: You need only one round trip and you are not forced to load the whole collection of Contacts to determine the IsMyContactFlag (but you can if you want to).
The drawback: You need this additional ViewModel type.
It is possible to do this but it will be far from a "nice way" because you cannot return instances of your User type. You must write custom linq-to-entities query and you must solve two problems:
You cannot project to mapped types in linq-to-entities
You cannot access non mapped properties in linq-to-entities
So my high level untested idea about doing this is:
var query = from u in ctx.Users
where u.Id != id // don't include current user - you can add other condition
join c in ctx.Users
.Where(x => x.Id == id) // current user
.SelectMany(x => x.Contacts)
on u.Id equals c.Id into leftJoin
from y in leftJoin.DefaultIfEmpty()
select new
{
UserName = u.UserName,
IsMyContact = y != null
};
This should be a query which will load pairs of UserName and information if the user is contact or not. If you want User instance instead you must do something like this:
var users = query.AsEnumerable()
.Select(new User
{
// Project to list in linq-to-objects
});
I just wondered how people were approaching this situation. It's something that seems like a weak point in my usage of MVC with ORMs (NHibernate in this case)...
Say you have a fine-grained and complicated entity in your model. You will likely have an admin page to manage objects of this type. If the entity is complicated, it is unlikely that you will be modifying the whole entity in one form. You still need to pass the relevant properties to the view, and incorporate changes to those properties in the model when the view returns them.
What does anyone do in this situation?
Create a view model which is (or contains) a subset of the entities properties. Pass this to and from the view. In 'edit' action method in controller, get the object from repository, go though all the properies in the ViewModel and apply them to the Model object (model.a = viewmodel.a, modelb = viewmodel.b). This seems the obvious sensible route, but generates a lot of tedious plumbing code. Also this complicates validation a bit.
Something else?
I've looked briefly at automapper - but this doesn't seem to fit the bill exactly, maybe I'm wrong?
Thanks.
This sounds like the perfect scenario for automapper. You create a view model class which contains a subset of the fields or your real model, and you let AutoMapper take care extraccting values from the domain model object into your view model object. What issues are you having with this approach?
Consider this example:
Here is your domain model and your view model
public class Person
{
public string FirstName
{ get; set; }
public string LastName
{ get; set; }
public string HomeNumber
{ get; set; }
public string Address1
{ get; set; }
public string Address2
{ get; set; }
}
public class PersonViewModel
{
public string FirstName
{ get; set; }
public string LastName
{ get; set; }
public string HomeNumber
{ get; set; }
}
Here is your mapping, you have to create a mapping in both directions from dm->vm and vm->dm.
From what I've seen when using Automapper is that if you map from object A to B and B has a property which A doesn't have, it will be reset. So when I create the map I direct it to ignore those missing properties. I'm not a Automapper expert so I may be using it wrong.
Mapping
Mapper.CreateMap<Person, PersonViewModel>();
// Automapper will reset values in dest which don't exist in source, so be sure to ignore them!
Mapper.CreateMap<PersonViewModel, Person>()
.ForMember(dest => dest.HomeNumber, opt => opt.Ignore());
Finally usage:
Person p = new Person()
{
FirstName = "First",
LastName = "Last",
Address1 = "add 1",
Address2 = "add 2"
};
PersonViewModel pvm = Mapper.Map<Person, PersonViewModel>(p);
// Map to a new person
Person p2 = Mapper.Map<PersonViewModel, Person>(pvm);
// Map to the existing person just to update it
Person p3 = new Person()
{
HomeNumber = "numberHere"
};
// This will update p3
Mapper.Map<PersonViewModel, Person>(pvm, p3);
Because of the exclusion, this is obviously less than ideal, but much better than manually doing the whole thing.
Have your view model map one-to-one with your domain model.
Specify Model as argument for the routeValues as below. This means your view model will be initialized with the values from the domain model. Only the sub set of fields in the form will be overwritten in the resulting personViewData.
Update View:
#model ViewModel.PersonView
#using (Html.BeginForm("Update", "Profile", Model, FormMethod.Post))
{
...Put your sub set of the PersonView fields here
}
ProfileController:
public ActionResult Update(string userName)
{
Person person = _unitOfWork.Person.Get().Where(p => p.UserName == userName).FirstOrDefault();
PersonView personView = new PersonView();
Mapper.Map(person, personView);
return View(personView);
}
[HttpPost]
public ActionResult Update(PersonView personViewData)
{
Person person = _unitOfWork.Person.Get().Where(p => p.UserName == personViewData.UserName).FirstOrDefault();
Mapper.Map(personViewData, person);
_unitOfWork.Person.Update(person);
_unitOfWork.Save();
return Json(new { saved = true, status = "" });
}
Why don't you use TryUpdateModel with the form collection.
If your view is editing a person
public class Person
{
public string ID { get; set; }
public string FirstName { get; set; }
public string LastName { get; set; }
public string Email { get; set; }
public string Address { get; set; }
}
And your view is only editing first name and last name, you can do this:
public ActionResult Action(FormCollection form)
{
Person personToUpdate = Repository.GetPerson(form["ID"]);
TryUpdateModel<Person>(personToUpdate, form);
Repository.Update(personToUpdate)
return View();
}
That will only update Person with the items that a part of the form collection. If you don't want a field updated, don't submit it with the form.
What if you have full model but each page uses and updates only the required part? Then you update the business model using complete view data at the last page.
I use a similar approach to yours (in my case Entity Framework) with Entity -> ViewModel -> View but only on views with "complex" entities that have either 1:M or M:M relationships. In most cases I took the low road and went for Entity->View when I have a simple entity.
My ViewModel is defined as Entity+supporting properties: SelectList or MultiSelectList and either a string or List<string>. I'll also use a ViewModel for instances where I have properties I need for the view but may not necessarily need in the entity (database).
Http Get controller methods are straightforward ActionResults with return View(repository.FetchNewViewModel()) for Create or repository.FetchModelById(id) for Edit. In both instances I'm initializing my entities before passing them to the view.
Create([Bind(Exclude = "Entity.EntityId")] ViewModel model) and Edit(ViewModel model) are the Http Post controller methods of Create and Edit. My Edit view has a hidden input field for EntityId to pass it back and forth.
By the time the Http Post method has the viewmodel, I lose all Entity.Relation and ViewModel.(Multi)SelectList values. I have to rebuild the object if I want my view to display properly:
`
try
{
var tags = model.TagIds; // List<string> or <int> depending on your Id type
if (model.TagsList == null) // It will be
{
model.TagsList = _repository.FetchSelectedTags(tags); // Build a new SelectList
}
if (!ModelState.IsValid)
{
return View(model);
}
_repository.Add(model.Article, tags); // or Save() for Edit
}
catch
{
return View(model); // Generally means something screwed in the repository class
}
return RedirectToAction("Index");
`
There is maybe 30% of my entity base using a ViewModel so I definitely only use it as needed. If you have complex views and model data in most instances you can probably break it down to smaller views.
Right now i´m working on a large project using S#arp Architecture and im also using the approach:
Model -> ViewModel -> Model
I use the ViewModel for the Binding part and Validations, the other approach is to use the Model Directly (with tryUpdateModel / UpdateModel which we used during the prototype develop) but for complex scenarios we end up handling special situation like SelectLists/Checkbox/Projections/HMAC Validations in a little ViewModel anyway and using a lot of Request.Form["key"] =( , the other drawback is handling the errors situations where you want to repopulate the form with the user input, i found it a little more complicated using the Model directly (using a ViewModel we take a lot of advantage of ModelState attempted value, saving us a couple of trips to the DB, anyone who have faced this scenario will know what i mean).
This approach is a bit time consuming, just like you said, you end up matching properties, but in my opinion is the way to go for complex forms.
It worth mentioning that we just use ViewModels for the Create/Edit scenarios, for almost everything else we use directly the model.
I have not use autommapers so far, but definitely i ll give it a try.
Does it make sense create an object that contains only those properties that the user will input on the webpage, use that for binding in the controller, and then map to the full Entity Object? Or should you just use the entity object, and use Include and Exclude to make restrictions on what gets bound on input?
I have come to like the idea of using interfaces to segregate which properties should be included when the object is updated.
For example:
To create and update an person object:
interface ICreatePerson
{
string Name { get; set; }
string Sex { get; set; }
int Age { get; set; }
}
interface IUpdatePerson
{
string Name { get; set; }
}
class Person : ICreatePerson, IUpdatePerson
{
public int Id { get; }
public string Name { get; set; }
public string Sex { get; set; }
public int Age { get; set; }
}
Then, when binding model, just use the appropriate interface as the type and it will only update the name property.
Here is an example controller method:
public ActionResult Edit(int id, FormCollection collection)
{
// Get orig person from db
var person = this.personService.Get(id);
try
{
// Update person from web form
UpdateModel<IUpdatePerson>(person);
// Save person to db
this.personService.Update(person);
return RedirectToAction("Index");
}
catch
{
ModelState.AddModelErrors((person.GetRuleViolations());
return View(person);
}
}
See this article (and the comments) for a very good discussion of the options.
I recommend using a separate presentation model type in most cases. Aside from the issue of binding (which is important, but there are other ways around this issue), I think that there are other reasons why using presentation model types is a good idea:
Presentation Models allow "view-first" development. Create a view and a presentation model at the same time. Get your user representative to give you feedback on the view. Iterate until you're both happy. Finally, solve the problem of mapping this back to the "real" model.
Presentation Models remove dependencies that the "real" model might have, allowing easier unit testing of controllers.
Presentation Models will have the same "shape" as the view itself. So you don't have to write code in the view to deal with navigating into "detail objects" and the like.
Some models cannot be used in an action result. For example, an object graph which contains cycles cannot be serialized to JSON.