Object Oriented Software

Metric
 Requirements
• Understanding of class relationship
• Understanding of class diagram
– Association A B

• Aggregation A B
• Composition A B
– Generalization A B
• Inheritance
• Sub class – super class
• Parent – Child
 Weighted Methods per Class
(WMC)
• The effort in developing a class will in some sense will be determined by
the number of methods the class has and the complexity of the methods.
• Suppose a class C has methods M1, M2... Mn defined on it. Let the
complexity of the method M1 be c1, then
i n
WMC =
c
i 1
i
• If the complexity of each method is considered 1, WMC gives the total
number of methods in the class.
• A separate matrix will be required to calculate complexity, like,
cyclomatic complexity.
• The data based on evaluation of some existing programs, shows that in
most cases, the classes tend to have only a small number of methods,
implying that most classes are simple and provide some specific
abstraction and operations.
• WMC metric has a reasonable correlation with fault-proneness of a
class. As can be expected, the larger the WMC of a class the better the
chances that the class is fault-prone.
 Depth of Inheritance Tree (DIT)
• Inheritance is one of the unique features of the object- oriented paradigm.
• Inheritance is one of the main mechanisms for reuse
• The deeper a particular class is in a class hierarchy, the more methods it
has available for reuse, thereby providing a larger reuse potential
• Inheritance increases coupling, which makes changing a class harder
• A class deep in the hierarchy has a lot of methods it can inherit, which
makes it difficult to predict its behavior
• The DIT of a class C in an inheritance hierarchy is the depth from the root
class in the inheritance tree
• In case of multiple inheritance, the DIT metric is the maximum length from a
root to C
• Statistical data suggests that most classes in applications tend to be close
to the root, with the maximum DIT metric value being around 10
• Most the classes have a DIT of 0 (that is, they are the root).
• Designers might be giving upon reusability in favor of comprehensibility
• The experiments show that DIT is very significant in predicting defect-
proneness of a class: the higher the DIT the higher is the probability that
the class is defect-prone.
 Number of Children (NOC)
• The number of children (NOC) metric value of a class C is the
number of immediate subclasses of C
• This metric can be used to evaluate the degree of reuse, as a higher
NOC number reflects reuse of the definitions in the superclass by a
larger number of subclasses
• It also gives an idea of the direct influence of a class on other
elements of a design
• The larger the influence of a class, the more important the class is
correctly designed
• In the empirical observations, it was found that classes generally
had a small NOC metric value, with a vast majority of classes
having no children
• This suggests that in the systems analyzed, inheritance was not
used very heavily
• The data suggest that the larger the NOC, the lower the probability
of detecting defects in a class
• The higher NOC classes are less defect-prone. The reasons for
this are not very clear or definite.
Coupling between Classes (CBC)
• Coupling between classes of a system reduces modularity and
make class modification harder
• It is desirable to reduce the coupling between classes
• The less coupling of a class with other classes, the more
independent the class, and more easily modifiable
• Coupling between classes (CBC) is a metric that tries to quantify
coupling that exists between classes
• Two classes are considered coupled if methods of one class use
methods or instance variables defined in the other class
• There are indirect forms of coupling (through pointers, etc.) that are
hard to identify
• The experimental data indicates that most of the classes are self-
contained and have low CBC value.
• Some types of classes, for example the ones that deal with
managing interfaces, generally tend to have higher CBC values
• The data found that CBC is significant in predicting the fault-
proneness of classes, particularly those that deal with user
interfaces.
 Lack of Cohesion in Methods
(LCOM)
• Cohesion captures how closely bound the different methods of the class
are
• Two methods of a class C can be considered “cohesive” if the set of
instance variables of C that they access have some elements in common
• High cohesion is a highly desirable property for modularity
• Let Ii and Ij be the set of instance variables accessed by the methods M1
and M2, Q be the set of all cohesive pairs of methods, that is, all (Mi, Mj)
such that Ii and Ij have a non-null intersection. Let P be the set of all
noncohesive pairs of methods, that is, pairs such that the intersection of
sets of instance variables they access is null. Then LCOM is defined as
LCOM = |P|-|Q|, if |P| > |Q|, otherwise 0.
• If there are n methods in a class C, then there are n(n — 1) pairs, and
LCOM is the number of pairs that are noncohesive minus the number of
pairs that are cohesive
• The larger the number of cohesive methods, the more cohesive the class
will be, and the LCOM metric will be lower
• A high LCOM value may indicate that the methods are trying to do different
things and operate on different data entities
• If this is validated, the class can be partitioned into different classes
• The data in [BBM95] found little significance of this metric in predicting the
fault-proneness of a class
Example 1 (NOC, DIT, CBC)
A B F H

C D E G

CLASS NOC Influence on Design DIT Reuse Potential CBC (appx.)

A 0 Low 0 Low Low (2)

B 3 Highest 0 Low Highest (4)

C 0 Low 0 Low Lowest (1)

D 0 Low 1 Moderate Lowest (1)

E 0 Low 1 Moderate Low (2)

F 1 Moderate 1 Moderate Low (2)

G 0 Low 0 Low Lowest (1)

H 0 Low 2 Highest Lowest (1)

G C D E

CLASS NOC Influence on Design DIT Reuse Potential CBC (appx.)

A 0 Low 0 Low Low (2)

B 3 Highest 0 Low Highest (4)

C 0 Low 0 Low Low (2)

D 0 Low 1 Moderate Lowest (1)

E 0 Low 3 Highest Low (2)

F 1 Moderate 1 Moderate Low (2)

G 0 Low 0 Low Lowest (1)

H 1 Moderate 2 High Low (2)

a1 Pairs:
a2 (O1, O2), (O1, O3), (O1, O4), (O2, O3), (O2, O4),
a3 (O3, O4)
a4
P = 2 (Non-Cohesive pairs)
Q = 4 (Cohesive pairs)
O1(a1, a2)
O2(a1)
Q>P
O3 (a4)
O4 (a1, a4)
LCOM = 0