Limiting Child Objects in Class Diagram

Question

I want to model a simple system using a Class Diagram:

• I have 3 possible classes: Company, Employee, Manager.
• The Company must have 20 Employees (aggregation?).
• The Company must have 1 Manager (aggregation?).
• A Manager is an Employee (generalization?).
• Each Employee can only be in 1 Company.

In other words, I want to limit this system to have 20 Employees, 1 of which must be a Manager. However ONLY 1 can be a Manager. This would make it so that there are 19 Employees and 1 Manager objects at all times.

I have the setup in my head of how I want this system to work, but I can't get the model quite right. This is what I got:

I feel like I am very close, but my issue is that although the Company 1 to 1 relationship to the Manager seems correct, the 1 to 19 to the Employee seems off. Since a Manager is an Employee, I have no way of limiting how many of those 19 Employees are Managers. I am trying to do this without splitting Employees up into Non-Manager and Manager classes.

Am I on the right track? Is there something I am missing? Or is it clear enough that 19 MUST be Employee objects and 1 MUST be Manager object?

Answer 1

The problem

Your model has a weakness: a Manager of one Company could be Employee of another, because nothing in your model says that the employment relationship with the Company is the same for both classes.

Why? If you apply the UML generalization semantics:

• Employee has an association with Company,
• Manager is a specialization of Employee, and therefore inherits all its properties, operations and associations, including the association with Company.
• Manager has in addition its own association with the Company. So it has two distinct associations: the inherited one and its own one.

Potential solutions

It would be helpful to label the ends of the association, e.g. employer/ employee and company/manager:

• The simplest solution could be to remove the association between Manager and Company, since it is inherited. But there is no easy way to tell that there must be a Manager among the employees. Moreover, there wouldn't be an easy way to find the Manager of a Company. So this solution does not appear appropriate.

• Another solution would be to add a constraint that specifies that for the Manager, manager.company is the same than employee.employer. Since a Manager manages one and only one company, no other manager could by deduction exist among the employees. But this sounds somewhat artificial.

• The best solution in my view is therefore to keep both associations but use UML smeantics to explain that company {subsets employer} and manager {subsets employee} Be aware that this solution requires 20 employees, since it makes clear that the manager is one amont the 20.

If you're interested to know more about subsetting, I advise you this artile about redefinition, specialization and subsetting of associations, which could also inspire you other variants.

Minor remarks

The aggregation are fine. However, aggregation is a modelling placebo since the UML specs clearly say page 110:

Precise semantics of shared aggregation varies by application area and modeler.

Therefore, I'd suggest to avoid them when possible. You could therefore as well use normal UML associations. Especially for the manager where there is only one.

Another remark made by qwerty_so in the comments is that the + should be removed in +has : The + is about public visibility. It’s not the association itself that is public or private, but the association end (e.g. employer and employee in my proposal).