I'm developing an application using domain driven design. One of the patterns I've been using is Repository pattern. For the sake of simplicity, let's say I have following classes and interfaces.
Car - domain class representing car domain concept.
public class Car {
public int Id {get;private set;}
public string SomeUniqueCode {get;private set;}
}
ICarRepository - interface for adding, deleting or saving changes to Car objects.
public interface ICarRepository{
Car AddCar(Car c);
void DeleteCar(Car c);
}
My problem is, how to check uniqueness of SomeUniqueCode property among all Car objects in the database? That property is changed by user (not auto-generated) at any time during the object life-cycle. Of course, one solution would be to put the unique key in the database, but that is not the principle of DDD. I've seen Specification pattern used to validate single objects. How would that pattern be applied to a set of Car objects?
Is it legitimate that Specification class (let's call it CheckUniqueCarSpecification) accesses ICarRepository?
A repository mimics an in-memory collection. What I have used before is a Contains method as opposed to a Find method, I guess you could have either. A query layer could also be used for this. Just as you have a CarRepository you could have a CarQuery. Trying to check for uniqueness in the domain is somewhat pesky. I would do a check for the sake of convenience but still rely on the DB to raise the exception since you should also handle that case. Using the specification pattern for this may be more effort than it is worth.
Since repository is a 'collection' I wouldn't have Commit and Rollback on there.
Use DomainService ICarCodesLibrary.
public class Car {
ctor(string someUniqueCode, ICarCodesLibrary codes)
{
// the check
codes.IsValidCode(someUniqueCode)
}
public int Id {get;private set;}
public string SomeUniqueCode {get;private set;}
}
Implement the interface in the place where u create the Car object and inject it. Also get rid of the properties and use fields. The ID is OK to be a prop.
Related
Regarding below excerpt, concerning cqrs and ddd, from Patterns, Principles, and Practices of Domain-Driven Design by Nick Tune, Scott Millett
Does it mean that domain model on command side can omit most of business attributes ?
How would it look like for eg Customer Entity?
Could Customer entity omit FirstName, Surname etc?
If so, where would these business attributes be? Only in read model in CustomerEntity?
Or maybe apart from CustomerEntity containing all business attributes there would also be CustomerAggregate wrapping CustomerEntity with 1:1 relation, and command object would operate on CustomerAggregate? (seems strange to me).
What does it mean "Customer entity desn't make sense"?
The text you pointed means that you do not have to model a reusable Entity for your whole system or even for your whole bounded context (Do not model reusable real life things). Doing this is a bad design.
You have to model an Aggregate that performs an action. You feed the Aggregate with only, and just only, the data needed to perform that action and the aggregate response, the changes the domain suffered, is what you have to persist.
Why Entities and V.O.'s then?
To model consistency, encapsulation and decoupling is the basic part but these are implementation details. For DDD what matters is that are different roles (or concepts).
When feeding the aggregate (constructor, function call parameters, etc) the aggregate has to know if it is working with entities and/or with V.O. to build its response.
If the domain action means a change in an attribute of a entity (something with unique identification in your whole system) the response of the aggregate (once all rules and invariants has been checked) should include the new attribute value and the identification of that entity that allows persist the changes.
So, by default, every aggregate has its own entity with the unique identification and the attributes needed for the aggregate action.
One aggregate could have a Customer entity with ID and its Name.
Another aggregate could have a Customer entity with ID and its Karma points.
So every aggregate has its own inner Customer entity to work with. When you feed an aggregate you pass Customer data (i.e. ID and name or ID and Karma points) and the aggregate treats that info as a entity (It is a matter of implementation details if there is a struct, class, etc internally to the aggregate to represent the entity).
One important thing: If you just need to deal with entities ID's then treat it as a V.O. (CustomerIdentityVO) because the ID is immutable and, probably, in that action you just need to write this CustomerIdentityVO in some field in persistence, not change any Customer attribute.
This is the standard vision. Once you start to identify common structures relevant to several aggregates or one aggregate that can perform several actions with the same data fed you start to refactoring, reusing, etc. It just a matter of good OOP design and SOLID principles.
Please, note that I am trying to be higly above of implementation details. I know that you almost always will have unwanted artifacts that depends of programing paradigm type, chosen programing language, etc. but this approach helps a lot avoiding the worse artifact you could have.
Recommended readings:
http://blog.sapiensworks.com/post/2016/07/29/DDD-Entities-Value-Objects-Explained
http://blog.sapiensworks.com/post/2016/07/14/DDD-Aggregate-Decoded-1
http://blog.sapiensworks.com/post/2016/07/14/DDD-Aggregate-Decoded-2
https://blog.sapiensworks.com/post/2016/07/14/DDD-Aggregate-Decoded-3
and
https://blog.sapiensworks.com/post/2016/08/19/DDD-Application-Services-Explained
for a complete puzzle vision.
If you are using Event Sourcing then it's true that you can model aggregates without adding attributes that they don't need for implementing the business logic.
Here's an example:
class Customer {
public Guid ID { get; private set; }
public Customer(Guid id, firstName, lastName, ...) {
ID = id;
this.AddEvent(new CustomerCreatedEvent(id, firstName, ....);
}
public void ChangeName(firstName, lastName) {
this.AddEvent(new CustomerRenamedEvent(this.ID, firstName, lastName),
}
}
Custom only has ID attribute because it needs it to add it to every event that it generates. FirstName and LastName are omitted as they are not needed even when ChangeName method is called. It only records an event that this happened. If your logic requires the FirstName then you can add it. You can omit any properties that you don't need.
Your Repository in this case will save only the events and won't care about the values of the attributes of the Customer.
On the Read side you will probably need these properties as you will display them to your users.
If your aggregates are not event sourced, then you probably will need more attributes on your aggregate to implement it's logic and they will be saved to the database.
Here's an example:
class Customer {
public Guid ID { get; private set; }
public string FirstName { get; private set; }
public string LastName { get; private set; }
public void ChangeName(firstName, lastName) {
FirstName = firstName;
LastName = lastName;
}
}
In this case your Repository will need these properties as it will generate a query to update the database with the new values.
Not sure what "Customer entity doesn't make sense" means.
I am currently trying out DDD and reading Evans book. I have arrived at a model that has an aggregate whose root is Student. Now I need to have (or be able to distinguish) a RegisteredStudent and an EnrolledStudent (inherits RegisteredStudent). I don't know how to handle inheritance in DDD.
Should the 2 inherited classes be inside the aggregate? If so, are they also considered aggregate roots since their identity is the same as the root (there are only added properties to them)? If not, how do I give access to them from other entities?
Or should I not be using inheritance? Why?
And also, what if you have an entity in an aggregate that isn't a root, but you need it to inherit an entity outside? How should you go about it?
What you need to ask yourself here is whether a RegisteredStudent and an EnrolledStudent are different concepts. Are they not both students, but just in a different state?
In general, you should favor composition over inheritance.
Here's an example of what I would do. (Note that it's just my example, I don't know the domain, so it's not a definitive solution).
You could have a Student class, which is your aggregate root and then a few different state classes: Registered and Enrolled. That way you don't need to expose these state classes on the student but you could just expose methods on the Student. A small example in C#:
class Student
{
State _currentState;
void Enroll()
{
if(!_currentState is Registered)
throw new InvalidOperationException("Cannot enroll student who is not registered");
this._currentState = new Enrolled();
}
void Register(string name)
{
this._currentState = new Registered(name);
}
}
class StudentState{}
class Enrolled : StudentState
{}
class Registered : StudentState
{
public Registered(string name)
{
Name = name;
}
public string Name {get; private set;}
}
This is a simple application of the State design pattern, you could externalize more parts of it and build a complete state-machine, but I'll leave that up to you. (Also it's typed directly in to the SO-editor, so there could be syntax errors)
EDIT after comments
Whether you need to expose a State-property or not depends on the context. In general I would recommend not to do that, because you're exposing the internals of the Student. It would be better to expose a method called CanEnroll for example. That way you can change the internal implementation of your state pattern without affecting any clients.
As for question 3, it's hard to say without a use case. However, here are some guidelines:
Favor composition over inheritance (again, I know).
You can have a reference from inside an aggregate to the outer world, you shouldn't have a reference the other way around though.
I'm designing a new large scale application which needs to be as flexible as possible.
I chose designing mainly with DDD..
My question is about transferring DTO object's back to DO objects in my service layer.
i.e:
This is my domain object mapped to the DB (using ORM)
public class Cat
{
int ID {get; set;}
string Name {get; set;}
string BloodType {get; set;}
string Color {get; set;}
void Run(){...}
void Purr() {...}
}
Methods and some properties are only needed for server actions.
That's why I designed another, data transfer object for this cat type:
public class CatDTO
{
int ID {get; set;}
string Name {get; set;}
}
In the middle, I'll set up an object mapper to translate my DO's to DTO's (and vice versa).
When a client would like to update a cat's name he will call a service like this
public void UpdateCat(CatDTO cat)
{
// What will happen here?
Cat serverCat = Mapper.GetCat(CatDTO);
CatDao.SaveOrUpdate(serverCat);
}
When the mapper is translating the DTO object back to DO it will have to hit the DB in order to fill the missing properties of the Cat object (blood type, etc')
Needles to say this action is absurd but without filling the empty properties the rest of the server side cannot work with the Cat object because it relies on those missing properties (even if i just try to update the data in the DB, My ORM will update the bloodtype field as an empty string!)
I searched for this problem and couldn't find any explenation on the web (or at least someone who is bothered with the issue as I do)
Am I designing it the wrong way? Maybe I missed something in my DDD?
Thanks, Pavel.
The usual workflow for this use case is: retrieve mapped domain object by ID, apply updates specified by the DTO, commit unit of work. What you refer to as the DAO is normally called a repository in DDD. The code should look more like:
public void UpdateCat(CatDTO catDto)
{
Cat cat = this.catRepository.Get(cat.ID);
cat.Name = catDto.Name;
this.catRepository.Commit();
}
The Commit step can come in a variety of ways. It can either be an explicit save, or the unit of work can be committed outside of the UpdateCat method. This workflow applies to all related scenarios as well. Generally, domain behavior involves retrieving the appropriate entity, invoking some behavior on that entity and then committing the resulting changes to the database.
Also, DTOs shouldn't directly map back into existing entities. Instead, it is better to think of them as representing changes to be applied to existing entities and the code should reflect this. This is in part because an existing entity is "owned" by the repository and the repository is responsible for reconstitution, not a DTO mapper.
This is a long question so i am gonna go straight to the point. This is pseudo code for better illustration of the problem
DB Structure
User (UserID, Name, LastName)
Address(AddressID, UserID, Street, City, State, ZipCode) =>Many to One User relationship
Phone (PhoneID, UserID, Number, IsPrimary) =>Many to One User relationship
Domain Classes
class User:IEntity
{
public string Name {get;set;}
public string LastName {get;set;}
public ContactInfo{get;set;}
}
class Phone: IValueObject or IEntity? will see later.
{
public int id; // persistence ID, not domain ID
public string Number {get;set;}
}
class Address: IValueObject or IEntity? will see later.
{
public string Line1 {get;set;}
public string City {get;set;}
public string State {get;set;}
public string ZipCode {get;set;}
}
class ContactInfo: IValueObject or IEntity? will see later.
{
List<Address> Addresses {get;set;}
List<Phone> PhoneNumbers {get;set;}
}
So, so far we have a very basic representation of this domain and its models.
My question is the following. Let's say that i want to Update one of the addreses or fix the area code for one of the numbers because of misspelling wnen it was initially typed in.
If i follow Evan's bible about DDD, Value Objects should be immutable. Meaning, no changes to its properties or fields after it was created.
If that's the case, then i guess, none of my classes are a ValueObject, since i can't just recreate the whole ContactInfo class just because one portion of the string in the phone number is wrong. So, i guess that makes all my classes Entities?
Keep in mind that i have a "persistence id" for each of this classes since they are stored in a database.
Let's say that i decide to make Phone a value object, since it's easy to recreate in the constructor
public Phone(string newNumber)
so, it would be something like adding a method to User (agg root) AND contactinfo? (Demeter Law)
like...
User....
public void UpdatePrimaryPhoneNumber(string number)
{
this.ContactInfo.UpdatePrimaryPhoneNumber(number);
}
ContactInfo....
public void UpdatePrimaryPhoneNumber(string number)
{
var oldPhone = Phones.Where(p=>p.IsPrimary).Single();
var newPhone = new Phone(number, oldPhone.persistenceid???-> this is not part of the domain)
oldPhone = newPhone;
}
but i still have to deal with persistence id... grrrrr. what a headache.
Sometimes i feel when i read those blogs that most "ddd experts" that value objects are overused or i would say misused.
What would be the best solution to this scenario?
Thank you
If i follow Evan's bible about DDD, Value Objects should be immutable.
Meaning, no changes to its properties or fields after it was created.
If that's the case, then i guess, none of my classes are a
ValueObject, since i can't just recreate the whole ContactInfo class
just because one portion of the string in the phone number is wrong.
So, i guess that makes all my classes Entities?
While the VO itself may be immutable, a VO doesn't exist on its own - it is always part of an aggregate. Therefore, a VO can be immutable, but the object which references that VO doesn't have to be. What helped me understand VOs is to compare them to something like a primitive Int32 value. The value of each individual integer is immutable - a 5 is always a 5. But anywhere you have an Int32 you can set another value there.
For you domain, what that means is that you can have an immutable address VO, but a given use entity can reference any instance of an address VO. This is what will allow corrections and any other changes to be made. You don't change the individual fields on the address VO - you replace it with a whole new VO instance.
Next, "Persistence ids" shouldn't be expressed in anywhere in domain code. They exist solely to satisfy the needs of the relational databases and NoSQL databases don't require them at all.
The primary phone scenario should look more like this:
public void UpdatePrimaryPhoneNumber(string number)
{
var existingPrimaryNumber = this.Phones.FirstOrDefault(x => x.IsPrimary == true);
if (existingPrimaryNumber != null)
this.Phones.Remove(existingPrimaryNumber);
this.Phones.Add(new Phone(phoneNumber: number, isPrimary = true));
}
This method encapsulates the idea of updating an existing primary phone number. The fact that phone number VOs are immutable means that you have to remove an existing value and replace it with a new one. What usually happens on the database end, especially with ORMs like NHibernate, is it will issue a SQL delete and a subsequent insert to effectively replace all phone numbers. This is OK since the ID of the VOs doesn't matter.
An Entity has a rather unique and individual life-cycle. It has meaning when it stands alone.
The classic example of Order/OrderItem may help with this.
If an OrderItem becomes an Entity it would have a life-cycle of its own. However, this doesn't make too much sense since it is part of an Order. This always seems obvious when looking at an order but less so when looking at your own classes because there can be some references between classes. For instance, an OrderItem represents some Product that we are selling. A Product has a life-cycle of its own. We can have an independent list of Products. How we model the link between an OrderItem and the Product is probably another discussion but I would denormalize the Product data I require into the OrderItem and store the original Product.Id also.
So is the Address class an Entity or a Value Object? This is always an interesting one in that we have that favourite of answers: it depends.
It will be context-specific. But ask yourself whether you have (or need) an independent list of Addresss and then only have a need for the link to that Address in your User. If this is the case then it is an Entity. If, however, your Address makes sense only when it is part of your User then it is a Value Object.
The fact that a Value Object is immutable does not mean you need to replace more than just the specific Value Object. I don't know if I would have a ContactInfo class in your current design since it only wraps the two collections (Address/PhoneNumber) but I would keep it if there is more to it (probably is). So simply replace the relevant PhoneNumber. If you have something like primary/secondary then it is as simple as:
AR.ReplacePrimaryPhoneNumber(new PhoneNumber('...'))
If it is a list of arbitrary numbers then a Remove/Add would be appropriate.
Now for the persistence Id. You do not need one. When you have a primary/secondary scenario you know what your use case is and you can execute the relevant queries in your DB (to update the primary PhoneNumber, for instance). If you have an arbitrary list you may go for add all new numbers in my list and delete those numbers from the DB not in my list; else just delete all the numbers and add everything you have. If this seems like a lot of heavy movement: it is. Event sourcing would move a lot of this to in-memory processing and it is something I will be pushing for seriously going forward.
I hope this all makes sense. Getting away from focusing on the data side of things is rather difficult but necessary. Focus on the domain as though you have no database. When you find friction then do your utmost to not pull database thinking into your domain but try to think about ways you could keep your domain clean and still use your DB of choice.
I would create a class PhoneNumber which contains the String number of the current Phone class and use that as a Value object within your Phone class:
class Phone implements IEntity
{
public int id; // persistence ID, not domain ID
public PhoneNumber number {get;set;}
}
class PhoneNumber implements IValueObject
{
public String number {get;set;};
}
Later when your code evolves you will need (for example) phone number validation and you can put it in the PhoneNumber class. This class can then be reused over the whole application at different places.
The Address is in my opinion a Value object which you can treat like a whole. Although you could model Street, City, etc... which are normally entities, but this is probably over-modelling. No part of the address can change, the whole object is always replaced when changing after initial creation.
The User class is within this example with these boundaries an Aggregate root (and thus also an Entity).
The ContactInfo class is not a ValueObject (not immutable) and not an Entity (no real identity) but an Aggregate. It contains multiple classes which should be seen as a whole.
More info on http://martinfowler.com/bliki/DDD_Aggregate.html
Usually whenever a persistence id is there you should be thinking of an Entity.
If however you would want to add the persistence id's, I would start splitting like the Phone and PhoneNumber class. For example Address (Entity containing id) and AddressValue containing all the other fields (and logic about address values).
This should also solve the headache about managing the persistence identities, since you replace the whole value object and the persistence identity stays the same in case of the updatePrimaryPhoneNumber.
The shorter version of the Question: "Is it ok to have a superclass, with 2 subclasses, one is an entity the other is a Value Object?"
To longer version:
T have a Team superclass. The Team has the Master, Helpers and a Code.
Then i have the DefaultTeam, subclass of Team, which is an entity with an unique **Code**** has its domain identity.
Then i have the **ExecutionTeam, its a subclass of Team and has an extra attribute OriginalTeam:
public abstract class Team{
public string Code{ get; protected set; }
public Worker Master{ get; protected set; }
public IList<Worker > Helpers { get; protected set; }
...
}
public class DefaultTeam: Team
{
}
public class ExecutionTeam : Team
{
public virtual string Code { get { return OriginalTeam.Code; } }
public virtual DefaultTeam OriginalTeam { get; private set; }
...
}
The ExecutionTeam, is the team that executes a Task.
When a Task needs to be executed, we choose a DefaultTeam to execute it.
But we can change the Helpers from the DefaultTeam (the master never changes).
That team that executes the task, is a variation of the DefaultTeam (OriginalTeam), but with the Helpers that were chosen just for that Task.
The ExecutionTeam will have the same code has the OriginalTeam. So the ExecutionTeam has no unique identity.
If there are 10 executions of tasks by the same DefaultTeam, there will be 10 ExecutionTeams with the same code (with the same OriginalTeam). So ExecutionTeam is cannot be an Entity.
But having an Entity and a Value Object sharing the same superclass (both being Teams), is a bit strange. Maybe this domain model has something wrong.
Need opinions.
Thanks
What is it that makes the DefaultTeam a Value Object rather than an Entity? Isn't a DefaultTeam also an entity?
That being said, here are some comments:
Why do you need a special class for DefaultTeam? Can't a DefaultTeam simply be an ExecutionTeam, with certain specified values?
A DefaultTeam should probably be an instance of a Team that is associated with an application domain. For example, you might have a particular team that is generally used to solve problems with Project XYZ.
Instead of listing "DefaultTeam" as a property of the ExecutionTeam, you should probably have a "PreviousTeam" as a property of both the Team and ExecutionTeam classes.
This will be more generalizable, in case the team gets changed yet again.
Since Task is an important part of the domain and is assigned to a Team, it should probably be a property of Team.
"Helpers" doesn't seem an appropriate name for the team members. Why not just name them "Members" or "TeamMembers"?
"Master" is probably un-PC unless you are working in Dilbert land or dealing with a database :) You might want to change this to "Supervisor" or "Manager".
"Code" is probably a bad name in the context of your application, as it may easily be confused with programming code. You might want to use "Id" or "TeamId" instead.
Sounds like ExecutionTeam might be better modeled as an interface ICanExecuteTasks. Would that work for you? It would eliminate the issue you are struggling with..
As to your short question, if the ExecutionTeam was indeed a derived class of Team, (inheriting from team and representing an "IsA" relatoonship, then the answer is No, they cannot be of different types because of course, every ExecutionTeam isA Team, thgere is only one thing, which is both a Team and an ExecutionTeam at the same time... It cannot be both an entity Type and a value type at the same time.
But the way you have designed the classes, as you have structured things, ExcecutionTeam is not a derived class, it is a property of the DefaultTeam. This implies that they have a "HasA" relationship. THis implies that they are different, co-existing objects, one of which can be an entity and one of which can be a value type. But my gut tells me this is not an accurate mirror of your real domain model...