In almost all EF methods that use inheritance, I see that the parent class is marked as abstract. Is there any reason why this is done? Or is it just so the abstract class cannot be instantiated?
No, that is not mandatory, your base class could not be abstract, but as you already said it's the most common. It's like a inheritance by generalization, which is extracting shared characteristics from two or more classes, and combining them into a generalized superclass, that's why is common see the base class as abstract.
In EF there are three different approaches to representing an inheritance hierarchy:
Table per Hierarchy (TPH): Enable polymorphism by denormalizing the
SQL schema, and utilize a type discriminator column that holds type
information.
Table per Type (TPT): Represent "is a" (inheritance) relationships as
"has a" (foreign key) relationships.
Table per Concrete class (TPC): Discard polymorphism and inheritance
relationships completely from the SQL schema.
In the last approach if the base class was concrete, then an additional table would be needed to hold instances of that class.
In summary, if the base class is abstract or not depends more on your side if that make sense or not in the model you are trying to represent.
I'd assume because the methods you override are also abstract? Can't have an abstract method on a non-abstract class.
So, for example, it never makes sense to instantiate a plain old DbContext -- you always need collection properties and such. The base DbContext provides functionality which all DbContext derivatives would need (connecting to a database, etc), but isn't useful as a stand-alone object.
Related
I have a hard time understanding the concept of association class as explained in UML 2.5 specification. The thing that confuses me the most is the following sentence taken from page 199:
NOTE. Even when all ends of the AssociationClass have isUnique=true, it is possible to have several instances associating the same set of instances of the end Classes.
As noted here: https://issues.omg.org/issues/UMLR-757 this single sentence seems to undermine the usefulness of the concept. At the same time, it makes pretty much every text explaining the concept written before 2.5 version of the UML spec obsolete – see for example the discussion here: UML association class - clarifying
But how can this actually work conceptually? Assuming all ends of the association class have isUnique=true, how can one have more than one instance of association class associating the same set of instances of the end classes, when isUnique property of member ends clearly forbids existence of such links?
I’ve always thought that an association class is just a regular association with extra attributes and/or operations. That interpretation seems to invalid now. So what exactly is association class? How can it have uniqueness independent from that of association member ends? There seem to be some unspoken multiplicity lurking somewhere (uniqueness is irrelevant property without maximum multiplicity higher than 1), but I can't figure out where.
In very short
Indeed, this is not super-clear and would deserve a better explanation. To make it short: it’s just a consequence of the lack of integration in the dual semantics defined for the association class.
Detailed explanation
What is uniqueness for an association
According to UML 2.5.1 section 11.5.3.1, page 197:
When one or more ends of the Association have isUnique=false, it is possible to have several links associating the same set of instances.
We can deduct using logical contraposition that:
If it is not possible to have several links associating the same set of instances, all ends of the association have isUnique=true.
So we'd expect this to apply to an assocation class as well, since the association class is also an association.
Association classes are two distinct things at the same time
According to UML 2.5.1 section 11.5.3.2:
An AssociationClass is both an Association and a Class, and preserves the static and dynamic semantics of both.
So, an association class is not just "an association with extra attributes". If it were that simple, the association class could perfectly be a generalization of an association: the specialized association would just inherit the extra attributes. But this is explicitly prohibited page 199:
An AssociationClass cannot be a generalization of an Association or a
Class.
because any specialization into a class would loose the association semantics, and any specialization into an association would loose the class semantics.
And this duality, is the cause of our issue.
Impact of this duality on instances
According to UML section 11.5.3.2, page 199 (formatting from me):
An instance of an AssociationClass has the characteristics both of a link representing an instantiation of the AssociationClass as a kind of Association, AND of an object representing an instantiation of the AssociationClass as a kind of Class.
If isUnique=true for all association ends, the instances of the association are guaranteed to be unique. Remind however that the association is only about tuples made out of the association ends:
An Association declares that there can be links between instances whose types conform to or implement the associated types. A link is a tuple with one value for each memberEnd of the Association, where each value is an instance whose type conforms to or implements the type at the end.
However, nothing in the specs requires that the class instantiation (object) representing the association instantiation needs to be unique.
Imagine for example, that we have an association class between class A and class B, and a and b are instances of these classes. Imagine that the association ends have isUnique=true. This means that there can be only one tuple (a,b) since the association is guaranteed to be unique.
Let P be a property of the association class, and let (a,b,p1) and (a,b,p2) be two instances of the class in the association-class. The class knows no association-ends: from the point of view of the class, there is no unicity requirement. And from the point of view of the association, we have only one tuple (a,b) , so it's ok as well.
The NOTE just explains that this (unfortunate and ambiguous) situation is possible.
Are there contradictions or inconsistencies?
Formally, there is no contradiction. This is the result of the way the association class is defined: a class and an association at the same time, without further defining the relationship between the respective instances.
But this creates some issues in regard of the semantics of associations having non-unique ends:
When one or more ends of the Association have isUnique=false, it is possible to have several links associating the same set of instances. In such a case, links carry an additional identifier apart from their end values.
More precisely, this makes the the association-class with unique ends useless, since the same result can be achieved with non-unique ends:
for a simple association with non-unique ends, you can have duplicates, i.e. several links associating the same instances of the association ends that are distinguished with an additional identifier.
for association classes with unique ends, according to the note you can have duplicates, i.e. several objects (class instances) corresponding to a link made of unique association ends (association instances).
for an association-class with non-unique ends, you could have duplicates, i.e. several object instances correspond to the same set of member ends. It makes no difference if you interpret this as several links associating the same instances of the association ends, each associated with a single object, or, if you interpret this as one link associating a unique set of instances of the association end, that would be each be associated to mutliple object instances.
IMHO, this is unfortunate:
it does not match our mental model in which an association class with all association ends having isUnique=true should have a unique object instance of the class for a unique combination association ends. This clearly goes against the principle of least astonishment: I started with deny and it took me a while to accept this, since it was so terribly different from the traditional ways to implement association classes.
two different models, one with unique association ends, and one without could in fact express the same situation.
A simple solution to this issue would be to require a unique class instance (object corresponding to the association class) to correspond to a link that uniquely associates association ends. In that way, unique association ends would imply unique association object, without requiring other changes to the UML specs.
I'm drawing a class UML for a JAVA software design. I don't really understand what is a stereotype. The formal definition:
The stereotype is a profile class which defines how an existing metaclass may be extended as part of a profile. It enables the use of a platform or domain specific terminology or notation in place of, or in addition to, the ones used for the extended metaclass
Is that mean the stereotype should be the name of the parent class? But why there are some superclasses also has a stereotype?
So for example, I have a superclass Animal and a class dog which extends the Animal class, what could be the stereotype of both classes? Also, why there are some classes without stereotype.
What is a stereotype and what are its purpose?
A stereotype extends UML by allowing to categorize some elements in a more detailed way than foreseen natively by UML. It helps to make the model more expressive.
A typical example are the Entity, Control, and Boundary stereotypes. They are used in a use-case driven design: in a class diagram, you can immediately find-out which classes represents the use-case logic (marked as «control» stereotype), which classes represents user-interface with actors («boundary» stereotype), and which classes represents business/domain objects (marked as «entity»).
Another use of stereotypes is to indicate the way a class is supposed to be implemented, for example to mark as «table» classes that correspond to database tables that are related to in-memory objects, or to indicate among the objects which ones are «persistent». An even more common use is to show some concepts that are language specific and have no equivalent in UML. I think in particular to C# «struct» which have a value semantic wheres C# classes have a reference semantic, or C# «property» (which is a stereotype at attribute level).
How are stereotypes defined?
Sterotypes are defined in a profile. You can see a stereotype as a class in a profile diagram.
But it's not an ordinary class that you could relate to other classes in your model. So there is no inheritance at play here. You would never indicate a superclass as stereotype.
A stereotype is a class related to the "metamodel", which means the UML standard elements defined in the UML standard. So it's a class of classes (or a class of associations, or of any other UML element).
To elaborate on your example, with an abstract superclass Animal and concrete sub-classes Cat and Dog:
you could affect them all an «entity» profile, because they all belong to the domain model.
if you would want to store the objects in an RDBMS using the concrete table inheritance, you may use a stereotype «persistent» for Cat and Dog (since concrete table inheritance pattern does not implement persistence for the abstract classes)
When designing software (think UML diagrams for example) and real world objects.
How does one identify a suitable case for an Abstract class?
For example if we had an [Employee] and [Fireman] and [paidFireman] and [unpaidFireman]...I am having trouble seeing whether a Fireman or Employee should be abstract and why?
Abstract classes are one of those more esoteric constructs in UML. Since classes are already an abstraction of real world things, an abstract class is even one level higher. Abstract classes can not be instantiated (since it is assumed they miss something for a real life). Whether you say that Fireman is abstract while the paid/unpaid are not, is a pure point of view and must be argued in the specific domain.
As a rule of thumb: leave abstract classes out of the door until you come to a point where you feel the urgent need for it. Introducing abstractness limits your model (and can help to avoid some malformed results of it). But without those limits the model is still valid as long as the architect sticks to common sense rules.
It mainly depends on your functional requirements.
If it makes sense in your application just to have simple employees (without designating them as firemen, policemen, or craftsmen), then the class may not be abstract, as the application will have to make instances just of the Employee class.
If that doesn't make sense, i.e. the occupation of each of your employees needs to be known at creation time, abstract classes come into consideration. But still they aren't necessary in every case. The easiest way to make sure the occupation is known is to model it as a mandatory attribute. Introducing a subclass only makes sense if there is specialized behavior for each of those subclasses. If, e.g., the salary of the firemen is calculated as 50$ * count of the fires he exstinguished, but the salary of the policemen is 1000$ + 50 * rank, then you model an abstract operation getSalary() in the Employee class, which will be concretely specified and implemented in each of the subclasses.
As the concept of interface also got mentioned in one of the answers, an interface describes the obligation to implement certain operations in all classes realizing that interface. That's much the same as an abstract operation in an abstract class. But the abstract class can contain much more than an interface: attributes and non-abstract operations.
So the rule of thumb is: For concepts of your domain for which interface and behavior can be fully described, use non-abstract classes. For concepts for which only interfaces and no behavior can be described, use interfaces. For concepts for which interfaces and part of the behavior can be described, use abstract classes.
There are many uses for an abstract class. An abstract class is one that cannot have any direct instances.
In software design, it is one way to describe an interface. Some of the declared operations can be implemented in the superclass. Any remaining implementations must be specified in sub-classes. Regardless of where the implementations exist, an abstract class means there can be no direct instances, only instances of some non-abstract subclass.
In a domain analysis, an abstract class is a way of modeling an abstraction. For example, think of the abstraction Role. It is useful to say that a Person plays a number of Roles. However, there is no instance of a Role that makes sense, without it also being a more specific kind of Role, such as Employee, Fireman, or Teacher. For this situation, you not only want Role to be abstract, you also want a covering axiom. For more on that, please read https://stackoverflow.com/a/35950236/2596664.
I use the Entity Framework with Db-First approach. I used to have a table called Ranking that I mapped to an abstract base class with a few inherited concrete classes. Now I recently removed all the inheritance and choose to just use one concrete class called Ranking.
But since I changed it back I get the following runtime Exception:
Type 'DbModel.Ranking' in conceptual side cannot be mapped to type 'My.Application.Models.Ranking' on the object side. Both the types must be abstract or both must be concrete types.
In my code generation I have set Abstract to false, but I don't know how to change this on the conceptual side.
I even tried deleting the Ranking table from my table designer, and then update it again from the database. This didn't help either.
Anyone an idea?
Found it. There was a partial class defined where I forgot to remove the abstract modifier.
So the generated partial class didn't had the abstract modifier anymore, but the other partial still had.
Pretty stupid after all, but hopefully this prevents some other people wasting half an hour.
I have a TPH heirachy along the lines of:
A->B->C->D
A->B->C->E
A->F->G->H
A->F->G->I
I have A as Abstract, and all the other classes are concrete with a single discriminator column.
This works fine, but I want C and G to be abstract also. If I do that, and remove their discriminators from the mapping, I get error 3034 'Two entities with different keys are mapped to the same row'. I cannot see how this statement can be correct, so I assume it's a bug in some way.
Is it possible to do the above?
Lee
Seeing as they are abstract - so there are no entities in the database - have you tried just lying and pretending they do have a discriminator.
This should satisfy the mapping, but since the classes are abstract you still won't be able to create or materialize.
Alex