GOF class diagram for Builder pattern appears to contradict its corresponding sequence diagram

Question

In the Design Patterns: Elements of Reusable Object Oriented Software, the Gang of Four present the following canonical form for the Builder pattern:

In Appendix B the following is mentioned regarding notation:

Slanted type indicates that the class or operation is abstract

An object reference representing a part-of or aggregation relationship is indicated by an arrowheaded line with a diamond at the base

An arrowheaded line without the diamond denotes acquaintance (e.g., a LineShape keeps a reference to a Color object, which other shapes may share)

Martin Fowler's UML Distilled explains that composition is defined by two aspects:

• A composite should basically own its components. That is, deleting a Polygon object should ensure all of its Points get deleted as well.
• A composite should not share the objects it owns with other objects

Fowler uses the term "association" instead of "acquaintance". Regardless of which term is used, the idea of an associated object being shared by several other objects remains. Fowler goes on to state:

Aggregation is strictly meaningless; as a result, I recommend that you ignore it in your own diagrams

If we maintain the terminology "part-of" and "acquaintance" from the Gang of Four, this is what we are left with:

• Part-of relationship:
  • Denoted by an arrowheaded line with a diamond at the base
  • Point is a part-of Polygon in the snippet below:

class Polygon
{
    private Point m_point;
    public Polygon()
    {
        m_point = new Point();
    }
}

• Acquaintance
  • Denoted by an arrowheaded line without a diamond at the base
  • In the following snippet, LineShape is acquainted with Color

class LineShape
{
    private Color m_color;
    public LineShape(Color color)
    {
        m_color = color;
    }
}

Now, the GOF's diagram for the Builder pattern shows a part-of relationship between the Director and the Builder. However, as shown by the sequence diagram below, they also have an independent Client create separate instances of a Builder and Director:

Because the Client must send requests to both a Director instance as well as a ConcreteBuilder instance, it would have to instantiate both separately as shown below:

ConcreteBuilder aConcreteBuilder = new ConcreteBuilder();
Director aDirector = new Director(aConcreteBuilder);
aDirector.Construct();
Product product = aConcreteBuilder.GetResult();

This would imply that aConcreteBuilder can outlive aDirector and thus that they have a part-of instead of acquaintance relationship.

Is this a mistake? Or am I missing something?

Answer 1

The GOF is pre-UML. It uses a variant of OMT, a notation invented by J.Rumbaugh, before it was merged into UML:

• The current UML terminology is "association". OMT made a subtle difference between bidirectional "associations" (see footnote, GOF Annex B) and unidirectional "acquaintances" (implemented with reference). In modern UML, both would be associations. The latter could be expressed with navigable associations, although UML is implementation neutral.
• Moreover, although OMT and earlier version of UML both used the white diamond, to represent part-whole relationships, J.Rumbaugh himself called this shared aggregation thing a "modeling placebo". Since UML 2.0, UML has kept the white diamond, but no longer defines any semantic for it. So, white diamond or no diamond, end up having the same semantic in UML, and the simpler notation should be preferred.

The white diamond, the so called shared aggregation, does not imply ownership of the elements at the other end (not in OMT and not in UML): elements can perfectly outlive their aggregate. The lifecycle limitation is a characteristic of composition (more precisely, composite aggregation, which implies an exclusive ownership), which is represented with a black diamond.

---

Edit: additional information on OMT

To avoid any loss in translation, I acquired the legacy original OMT book. Rumbaugh uses the terms "association" and "aggregation" (no "acquaintance") and confirms my previous conclusions:

Aggregation is a special form of association, not an independent concept. (...) If the two objects are tightly bound by a part-whole relationship, it's an aggregation. If the two objects are usually considered as independent, even though they may be often linked, it is an association. (...) The decision to use aggregation is a matter of judgement and often arbitrary. (...)