Functional over related classes bad smell detection and refactoring suggestions

International Journal of Software Engineering & Applications (IJSEA), Vol.5, No.2, March 2014
DOI : 10.5121/ijsea.2014.5203 29
FUNCTIONAL OVER-RELATED CLASSES BAD SMELL
DETECTION AND REFACTORING SUGGESTIONS
Jiang Dexun1
, Ma Peijun2
, Su Xiaohong3
, Wang Tiantian4
School Of Computer Science and Technology, Harbin Institute of Technology,Harbin,
China
ABSTRACT:
Bad phenomena about functional over-related classes and confused inheritances in programs will cause
difficulty in programs comprehension, extension and maintenance. In this paper it is defined as a new bad
smell Functional over-Related Classes. After the analysis, the characteristics of this new smell are
transformed to the large number of entities dependency relationships between classes. So after entities
dependency information collection and analysis, the bad smell is detected in programs, and corresponding
refactoring suggestions are provided based on detection results. The experiments results of open source
programs show that the proposed bad smell cannot be detected by current detection methods. The proposed
detection method in this paper behaves well on refactoring evaluation, and the refactoring suggestions
improve the quality of programs.
KEYWORDS:
Function over-Related; dependency relationship; bad smell detection; refactoring
1 INTRODUCTION
Good design quality of software will ease the maintenance and reusing, and reusability,
flexibility, understandability, functionality and extendibility will be improved. On the other hand,
if too many bad smells exist in programs, the quality of the software would be very low.
Bad smells [1] are signs of potential problems in codes. It causes difficulty for understanding and
modifying of programs. Bad smells are not mistakes or defects in codes, but may cause them.
Indeed bad smells are alerting of codes. So bad smells should be removed if found. Refactoring is
a programming technique for optimizing the structure or pattern of an existing body of code by
altering its internal nonfunctional attributes without changing its external behavior. On the
premise of preserving observed behaviors, refactoring can improve the quality through increasing
the understanding and decreasing the reuse cost of programs.
Inheritance is a fundamental feature of the Object Oriented (OO) paradigm. It is used to promote
extensibility and reuse in OO systems. Oppositely, in OO programs, the two classes are designed
without inheritance relationships, but they should have this inheritance. This situation may cause
bad results. The changing of one class will not cause the change of the other class automatically.

30
But in functional request, the two classes should change together for this modification. Thus it is
a bad phenomenon, because it decreases the understandability and maintenance of the program.
In this paper, a new bad smell is defined to describe the bad phenomenon above. For this new bad
smell, the characteristics, classification, impact and comparison to other similar and relevant bad
smells are discussed. Based on the features of this bad smell, the detection method is proposed,
and corresponding refactoring suggestion is provided.
The rest of the paper is organized as follows. Section 2 presents a short overview of related work.
In Section 3, the expression and characteristics are described, and the new bad smell is defined.
This bad smell is classified and detected, and the refactoring suggestion is provided in Section 4.
Section 5 shows the experimental results and analysis. The conclusion is provided in Section 7.
2 RELATED WORK
2.1 Classification of bad smells
Before bad smell, program analyzer use anti-pattern for the potential problems in the programs. In
[2-4], some of the bad smells in [1] are classified as anti-pattern. But Luo [5] discussed the
difference and correlation of anti-pattern and bad smells. Anti-pattern is the problem in design
level, and bad smells are in presentation level. Anti-pattern causes bad smells.
Fowler classified 22 kinds of bad smell, gave the description and improving process. Kerievsky
[6] introduced 6 new bad smells, and Abebe [7] introduced 7 more.
Wake [8] simply classified bad smells as “inner class” and “outside class”. Mantyla [9] classified
the 22 bad smells in [1]: the bloaters, the object-orientation abusers, the change preventers, the
dispensable, encapsulators and others. The significance of classification is that bad smells are
more understandable, the relations between bad smells are more obvious, and the researches of
bad smell can be further.
Raul [10] analyzed these 6 kinds of bad smell from the aspects of Granularity, Intra vs Inter-
relations, Inheritance and Access Modifiers, and dividing common bad smells with relevant
metrics.
2.2 Bad smell detection and refactoring
Fowler [1] introduced bad smells and gave detecting and refactoring methods. But these
processes should be achieved manually.
Hayden and Ewan [11] detected bad smells with dependency graphs. Their method can detect
Feature Envy, Data Dump and other bad smells. This method has shortage, and the collection of
dependency relationships is not completely right, such as duplication names of classes and fields.
This causes limited precision in bad smell detection.

31
Tourwe and Mens [12] introduced a bad smell detection method independent from programming
languages. They consider the key components of programs are expressions and principles, and the
principles should be redacted manually. The disadvantage of this method is the principle
transformation between program languages.
Atkinson and King [13] proposed syntactic based bad smell detection approach with low-cost.
The judgment of this approach is too simple, and the recall rate is low.
Simon [14] proposed a metrics based visual approach to assist the software engineer to detect bad
smells in programs. The cohesion of programs can be collected with metrics and reflected as
visual images. But bad smells detection needs manual judgment.
There are many open source tools for bad smell detection, such as CheckStyle1, PMD2, iPlasma
[15], JDeodorant [16] and so on. There tools are convenient, automatic, easy to obtain, and have
high detecting speed. But just small kinds of bad smells can be detected, such as Duplicated
Code, Data Class, Feature Envy, Large Class, Long Method and so on. Furthermore, the detection
processes are simple. For example, for Duplicated Code, only text duplication is considered, for
Large Class, fixed thresholds are used for detection. All of these causes lower precision of the
detection results.
3 FUNCTIONAL OVER-RELATED CLASSES
3.1 Bad phenomenon in codes
In object oriented language, the functions of programs are stored in different methods of classes.
Through the invoking and executing related methods, the functions of programs are achieved. If
the same functions are achieved in different places, the same methods should be used.
Furthermore, if the more similar the methods being used are, the more similar their functions are.
Therefore, the similarity of entities invoking leads to that of functions.
One typical sample of bad smells in open source program HSQLDB2.2.7 is displayed in Figure 1.
There are two classes in this sample, and they have no inheritance relationships, means they are
not parent-child or brothers relationships. But they all invoke some entities outside the classes,
and some of the entities are same, although the usages of the same entities are different. In class
SchemaObjectSet, method getSQL() invokes attribute dataImpact in class Routine, and method
addAllSQL() invokes attribute name and method schema(). In class ParserRoutine, method
compileRepeat() invokes attribute name in class HsqlName, and method
readLocalTableVariableDeclarationOrNull() invokes attribute name and method schema() in
class HsqlName. Other entities are both invoked by these two classes SchemaObjectSet and
ParserRoutine, but they are not listed here because of space cause.
1
checkstyle.sourceforge.net
2
pmd.sourceforge.net

32
Figure 1 Bad phenomenon example existing in open source programs

33
If the number of the same entities invoking relationships is high in the total invoking relationships
between two classes, theoretically the two classes should be designed in the same inheritance tree,
or bad smells occur.
3.2 Definition of new bad smell
Definition 1 (Entity): the entity is the attribute a or the method m in one class, which is signed
as E .
Definition 2 (Dependency relationship): if in a method 1m attribute a or method 2m is used, it is
called 1m invokes a or 2m . Maybe a or 2m is in the same class with 1m , or the opposite. When
entity 1E invokes entity 2E , dependency relationship occurs between them.
The dependency relation contains two forms: one is to instantiate other classes, one is to
parameterize other classes. The two forms are shown in Figure 2.
Figure 2 Sample codes of dependency relationship
If methods invoke attributes or methods in same classes, the form should be “this.attrbute” or
“this.method()”. If methods invoke attributes or methods in father classes, the form should be
“super.attrbute” or “super.method()”.
The dependency relationship is represented as 1 2( , )I E E , and 2E and 1E are the entities. Invoking is
directed, but dependency relationship is undirected, so 1 2( , )I E E = 2 1( , )I E E .
Dependency relationship is the undirected chain structure with two points.
Definition 3(Property set): The property set of an entity is the set of all the entities which have
dependency relationships with this entity. The entity itself is the element of its property set. The
property set of entity E is signed as ( )P E
.
Definition 4 (Dependency chain): Dependency chain is the chain structure composed of invoking
relationships. If there are two dependency relationships 1 2( , )I E E and 2 3( , )I E E , the length of chain
1 2 3E E E− −
is 2. In this chain 1 3E E≠
.
Definition 6 (Function related): two classes are function related, if dependency relationship
occurs between entities in respective classes.

34
Definition 6 (Function over-Related Classes): two classes are not in the same inheritance tree. If
there are dependency relationships between the entities in these respective classes, and the rate 
is high, the two classes are function over-related. And it is seen as a new bad smell, named
“Functional over-Related Classes”, FRC for short. And the equation of  is:
number of the entities with dependency relationships
number of all the entities in the class
 =
3.3 Features and comparison of FRC
3.3.1 What is functional related?
There are two food factories which separately produce chicken sausages and chicken tines.
Different foods are in accordance with the tastes of different people, but the food materials are
both live chicken. On the other hand, these two factories are “related” but not “same”. Because of
the same materials, the two factories can be considered to locate together, such as the chicken
farms or places with convenient transportation.
The function related classes judging are based on whether they have dependency relationships
between them, or they have dependency relationships to the entities in a third class. But they may
invoke the entities for different using, and it is not considered in this paper.
3.3.2 Classification of FRC bad smell
FRC bad smell can be classified into direct related and indirect related.
（1）Direct related bad smell
There is dependency relationship between AE and BE , and AE is in Class A, BE is in class B . If
this kind of invoking behaviors is frequent, and the share of the dependency relationships is high,
it is considered to be direct related FRC bad smell. In practice there are two kinds of situations,
one is that too many entities in Class A invoke entities in class B , as shown in Figure 3(a); the
other is that entities in Class A invoke too many entities in class B , as shown in Figure 3(b).
Figure 3 Direct related FRC bad smells

35
（2）Indirect related bad smell
There are dependency relationships between AE and CE , and also between BE and CE . CE is in
class C . So it is a dependency chain between class A and B, with the length of 2. If there are
other dependency chain between class A and B, and the share is high, it is considered to be
indirect related FRC bad smell, as shown in Figure 4.
Figure 4 Indirect related FRC bad smell
In this paper the dependency chain with only length 2 is considered in indirect related bad smell.
If the dependency chain is longer, the correlation between classes and entities is decreased.
3.3.3 Features and impacts of FRC bad smell
After analysis, the features are listed as follows:
(1)This bad smell occurs between two classes, and the dependency relationships are between
entities form different classes. The two classes are named original class.
(2)The entities in one original class have dependency relationships with entities outside the class,
or the entities in both of the original classes do.
(3)If there are dependency relationships between entity AE in class A and entity BE in class B ,
the intersection of entity AE ’s property set and entity BE ’s property set is not empty. For
example, entity AE in class A invokes entity CE in class C, and entity BE in class B also invokes
CE . Then
( ) ( )A B CP E P E E∩ = ≠ ∅ .
(4)The share of the element number in the intersection is high. The most extreme case is that all
the entities in one class are invoked by entities in another class. And it is obviously a wrong
design, for the class being invoked should be a superclass.
And the negative effects of FRC bad smell are:
(1) It is consumed that the FRC bad smell occurs at Class 1 and Class 2. When function request
changes, the program should be modified at Class 1 and Class 2. The dependency relationship
causes similar modifications at Class 1 and Class 2. And this will increase the workload of codes

36
modification. If Class 1 and Class 2 are in same inheritance hierarchy, the modification process
would be simplified.
(2) If there is dependency relationship between Class 1 and Class 2, but they are not in the same
inheritance hierarchy, this program will be difficult to understand, extend and reuse.
4 FRC BAD SMELL DETECTION AND REFACTORING
The detection of FRC bad smell is based on the analysis of this bad smell. The features of FRC
bad smell is collected as the standard of static analysis and bad smell detection. The total
algorithm of FRC detection is shown in Figure 5.
The inputs of the algorithm are: every class iC in the programs being detected, the entities ikE in
iC , and the two-dimensional array Inv recording all the dependency relationships of ikE . The
gain process of these input data is provided in [17], so the same will not be repeated here.
Figure 5 Total algorithm of FRC bad smell detection
For different kinds of FRC bad smells, the detection and refactoring processes are not the same.
The main approach of refactoring is adding new inheritance and changing existing inheritance.
But other simple refactoring operations are also necessary for the rationality and readability of
programs, such as changing names about the classes and entities.

37
4.1 Direct FRC
First the detection approach and refactoring suggestion about directed related FRC bad smell are
represented. In some cases, indirect bad smell may disappear along with removing direct bad
smell.
Just as shown in Figure 3(a), if some entities in class B is invoked by methods in class A, the rate
1 is computed for bad smell detection:
1 = number of entities in class B which are invoked by methods in class A
And the thresholds should be added.
There are dependency relationships between entities. In the class if all the entities with
dependency relationships are clustered, finally there are one or several clustering groups. For
example, 1e
, 2e
and 3e
are in one cluster group. If there is one dependency relationship between
1e
and 4e
, 4e
is pushed on this cluster group. In class A, the number of all the entities is AE
, and
the entities number of the largest cluster group is AY . Similar values in class B are BE and BY .
The attributes and methods in classes may be private or protected. The number of
private/protected attributes and methods in class A is AM
. Similar value in class B is BM
.
Then two thresholds should be computed for bad smell detection.
2
1 2
( )
( )
A B A
B A A A
E Y Y
E E Y E

⋅ −
=
⋅ − −
2
A B A B A B
A
Y E M E E M
E

⋅ + ⋅ − ⋅
=
And B AY Y>
.
1 is compared with 1 and 2 .
a) When 0< 1 < 1 , there is no bad smells;
b) When 1 < 1 < 2 and A A AY M E+ <
, FRC bad smell exists. And the refactoring operation is to
extract relevant entities to create a new class, as the superclass of both class A and B. The
situation is shown in Figure 6(a);
c) When 1 > 2 , FRC bad smell exists. And the refactoring operation is to set class B as the
superclass of A, as shown in Figure 6(b).

38
Figure 6 Refactoring suggestion of direct related FRC bad smell
4.2 Indirect FRC
The methods in class A and B both invoke the entities in class C, just shown as Figure 4. Then
this rate should be computed:
2 =entities number invoke by methods in both A and B
The entities number in class A, B, C are separately AE , BE and CE . The entities number of the
largest cluster group is AY
. The number of private/protected attributes and methods in class A is
AM . Similar values in class B are BY and BM . Then two thresholds are computed.
3
C
A
A
E
Y
E
 = ⋅
， 4
C
B
B
E
Y
E
 = ⋅
When 2 3 > , 2 4 > and A A B B A BY M Y M E E+ + + < + , indirect FRC exists. Then the
refactoring suggestion is to set class C as the superclass of class A and B, and it is shown in
Figure 7.
Figure 7 Refactoring suggestion of indirect related FRC bad smell

39
5 EXPERIMENTS AND ANALYSIS
5.1 FRC bad smell detection
According to the approach of FRC bad smell detection described in Section 4, a FRC bad smell
detection tool FRC Detector is built in .net environment. Besides FRC bad smell detection, this
tool can provide refactoring suggestions for existing FRC bad smell. The working process of FRC
Detector is represented in Figure 8.
Figure 8 Working process of FRC Detector
Open source programs HSQLDB, Tyrant, ArgoUML and JFreeChart are the object programs for
FRC bad smell detection. HSQLDB3is a Java database. The version is 2.2.8(2012-4-18), and the
codes are over 130,000 lines. Tyrant4 is a game, and the version is 0.96. ArgoUML5 is a open
source UML model tool, with the version 19912. JFreeChart6 is a free chart library for the Java
platform, with the version 1.0.13.
Meanwhile, bad smells detection tools are used towards same programs. Smelly7 is used for the
detection of Data Class, God Class, Long Parameter List; Code Bad Smell Detector8 is used for
the detection of Data Clumps，Switch Statements, Speculative Generality, Message Chains,
Middle Man; PMD9 is used for Duplicate Code; the method in [17] is used for Feature Envy; the
3
hsqldb.org
4
sourceforge.net/projects/tyrant
5
argouml.tigris.org/source/browse/argouml
6
sourceforge.net/p/jfreechart/code
7
smelly.sourceforge.net
8
cbsdetector.sourceforge.net
9
pmd.sourceforge.net

40
method in [18] is used for Parallel Inheritance Hierarchy; FRC Detector proposed in this paper is
used for FRC bad smell.
The experimental results are displayed in Table 1.
Table 1 Result of FRC bad smell detection
Detecting tools Bad smells Number of bad smells
HSQLDB Tyrant ArgoUML JFreeChart
Smelly Data Class 1 1 5 4
God Class 0 0 7 3
Long Parameter List 2 0 7 4
Code Bad Smell
Detector
Data Clumps
1 2 0 12
Switch Statements 6 4 32 15
Speculative Generality 0 0 1 5
Message Chains 3 4 9 8
Middle Man 0 0 5 1
PMD Duplicate Code 1 0 48 33
Method in [17] Feature Envy 3 2 0 16
Method in [18] Parallel Inheritance
Hierarchy 1 0 1 1
FRC Detector FRC 4 6 32 5
FRC bad smell cannot be detected by Smelly, Code Bad Smell Detector and PMD. Meanwhile,
the programs without other bad smells still have FRC bad smells. Therefore, FRC bad smell
cannot be detected by existing bad smell detection tools. And the improvements of other bad
smells cannot remove the decrease of FRC bad smell in programs design quality.
5.2 Refactoring impact analysis
5.2.1 Refactoring suggestions of FRC bad smell
FRC Detector can provide corresponding refactoring suggestions for existing FRC bad smells.
One suggestion is composed of one or more refactoring operations. Contents of the refactoring
suggestions are created through the approach in this paper by FRC Detector, and the executing of
the refactoring suggestions is achieved manually. The refactoring operations of partly existing
FRC bad smells are shown in TABLE 2.
Table 2 Refactoring operations of existing FRC bad smells
Program Class Type Refactoring operations
HSQLDB RangeIteratorBase，
RangeVariable
Direct RangeIteratorBase is changed as
the subclass of RangeVariable
ArgoUML StateBodyNotationUml,
TransitionNotationUml,
OperationNotationUml
Indirect Part of the entities are extracted a
new class as superclass

41
The refactoring operations of all the existing FRC bad smells are not listed entirely because of
space cause.
5.2.1 Program quality property computing
In this paper, the QMOOD (Quality Model for Object Oriented Design) model in paper [19] is
used for refactoring impact analysis. The metrics of DesignSize, Hierarchies, Abstraction,
Encapsulation, Coupling, Cohesion, Composition, Inheritance, Polymorphism, Messaging, and
Complexity are extracted from the programs before and after refactoring. Through the QMOOD
model, six program quality properties are computed.
According to paper [20], six properties about programs quality can be measured for total quality
analysis:
 (Reusability)：Object oriented design characteristics that allow a design to be reapplied to
a new program without significant effort
 (Flexibility)：Characteristics that allow the incorporation of changes in a design.
 (Understandability)：Properties of the design that enables it to be easily learned and
comprehended, which relates to the complexity of the design structure.
 (Functionality)：Responsibilities assigned to the classes of a design, which are made
available by the classes through their public interfaces.
 (Extendibility)：Characteristics refer to the presence and usage of properties in an existing
design that allow for the incorporation of new requirements in the design.
 (Effectiveness)：Characteristics refer to the design ability to achieve the desired
functionality and behavior using object oriented design concepts and technique.
The structure, relationships and functions of programs are reflected by these design quality
properties. The description of design quality metrics are displayed in TABLE 3.

42
Table 3 Description and computation of design quality metrics
Design quality
property
Description
DesignSize The count of the total number of classes in the program
Hierarchies The count of the number of class hierarchies in the program
Abstraction Number of classes along all paths from the root classes to all
classes in an inheritance structure.
Encapsulation The ratio of the number of private/protected entities to the total
number of entities declared in the class.
Coupling Count of the different number of classes that a class is directly
related to.
Cohesion The summation of the intersection of dependency relationships
of an entity with the maximum clustering set of all entities in the
class.
Composition The count of the number of data declarations whose types are
user defined classes.
Inheritance The ratio of the number of entities inherited by a class to the
total number of entities accessibly by member methods of the
class.
Polymorphism The count of the methods that exhibit polymorphic behavior.
Messaging The count of the number of public methods in a class
Complexity The count of all the methods defined in a class.
According to paper [19], the values of design quality properties can be computed by design
quality metrics. The formulas are listed in TABLE 4.
Table 4 Computing formula of design quality properties
Design quality property Formula
(Reusability)[] -
0.25*Coupling+0.25*Cohesion+0.5*Messaging+0.5*Desi
gnSize
(Flexibility)[] 0.25*Encapsulation-
0.25*Coupling+0.5*Composition+0.5*Polymorphism
(Understandability) -0.33*Abstraction+0.33*Encapsulation-
0.33*Coupling+0.33*Cohesion-0.33*Polymorphism-
0.33*Complexity-0.33*DesignSize
(Functionality) 0.12*Cohesion+0.22*Polymorphism+0.22Messaging+0.22
*DesignSize+0.22*Hierarchies
(Extendibility) 0.5*Abstraction-
0.5*Coupling+0.5*Inheritance+0.5*Polymorphism
(Effectiveness) 0.2*Abstraction+0.2*Encapsulation+0.2*Composition+0.2
*Inheritance+0.2*Polymorphsm
The standard value of each design quality metric in the formulas is 1. Then the values of each
design quality property are 1 expect the understandability property, and it is -1.

43
5.2.2 Quality properties results before and after refactoring
The real values of each design quality metric before and after refactoring are computed and
shown in Table 5. And “Before” means the program before refactoring, and “After” means that
after refactoring.
Table 5 Real values of each design quality metric before and after refactoring
Design quality
metrics
Before After Before After Before After Before After
DesignSize
110.000
111.00
0
116.00
0
118.000 1873.000 1881.000 583.000 587.000
Hierarchies 5.000 6.000 1.000 1.000 12.000 24.000 9.000 13.000
Abstraction 1.088 1.102 1.817 1.860 1.507 1.610 2.253 2.299
Encapsulation 0.610 0.723 0.480 0.566 0.710 0.728 0.630 0.753
Coupling 6.209 5.789 4.399 3.669 3.281 3.135 5.276 4.799
Cohesion 0.228 0.255 0.199 0.220 0.263 0.275 0.260 0.266
Composition 2.103 1.986 1.112 1.195 2.111 2.184 0.496 0.541
Inheritance 0.318 0.336 0.530 0.546 0.349 0.371 0.065 0.065
Polymorphism 2.076 2.196 2.891 2.748 3.047 2.765 2.892 2.762
Messaging 33.000 35.000 37.000 39.000 22.000 22.000 33.000 36.000
Complexity
172.000
204.00
0
379.00
0
434.000 6551.000 7175.000 876.000 1002.000
The measure units of each design quality metric are not unified. For example, Design Size
represents the number of classes in a program, so it is a value larger than 1 (actually it is always
much larger than 1). But some metrics like Encapsulation and Inheritance are the ratio so the
value is between 0 and 1. So the design quality metrics need standardizing for property
computing. After simplification, the design quality metric before refactoring is set 1, and the
design quality metric after refactoring is the ratio of original after value and before value. The
results after standardization are displayed in Table 6.

44
Table 6 Standard values of each design quality metric before and after refactoring
Design quality
metrics
(standardization)
Design Size 1.000 1.009 1.000 1.017 1.000 1.004 1.000 1.007
Hierarchies 1.000 1.200 1.000 1.000 1.000 2.000 1.000 1.444
Abstraction 1.000 1.013 1.000 1.024 1.000 1.068 1.000 1.020
Encapsulation 1.000 1.185 1.000 1.179 1.000 1.026 1.000 1.195
Coupling 1.000 0.932 1.000 0.834 1.000 0.955 1.000 0.910
Cohesion 1.000 1.116 1.000 1.103 1.000 1.045 1.000 1.023
Composition 1.000 0.944 1.000 1.074 1.000 1.034 1.000 1.091
Inheritance 1.000 1.057 1.000 1.032 1.000 1.063 1.000 1.012
Polymorphism 1.000 1.058 1.000 0.951 1.000 0.907 1.000 0.955
Messaging 1.000 1.061 1.000 1.054 1.000 1.000 1.000 1.091
Complexity 1.000 1.186 1.000 1.145 1.000 1.095 1.000 1.144
According to the formulas in Table 4, the values of design quality properties are computed and
listed in Table 7.
Table 7 Design quality properties of each program
Take HSQLDB 2.2.8 as the example, the design quality properties changing is shown in Figure 9.
Design quality
property
Reusability 1.000 1.081 1.000 1.103 1.000 1.025 1.000 1.077
Flexibility 1.000 1.064 1.000 1.099 1.000 1.008 1.000 1.094
Understandability -0.990 -0.956 -0.990 -0.887 -0.990 -0.976 -0.990 -0.930
Functionality 1.000 1.064 1.000 1.005 1.000 1.095 1.000 1.099
Extendibility 1.000 1.098 1.000 1.086 1.000 1.042 1.000 1.039
Extendibility 1.000 1.052 1.000 1.052 1.000 1.020 1.000 1.055
Amount 4.010 4.402 4.010 4.457 4.010 4.213 4.010 4.434

45
-1.5
-1
-0.5
0
0.5
1
1.5
Reusability Flexibility Understandability Functionality Extendibility Effectiveness
Designqualitypropertyvalue
Before
After
Figure 9 Design quality properties changing of HSQLDB 2.2.8
The refactoring suggestions of FRC bad smell are usually creating new classes then new
inheritance between new and original classes, or adding new inheritance between existing classes.
From Table 5, after refactoring the metrics of DesignSize, Hierarchies, Abstraction,
Encapsulation, Cohesion, Inheritance and Complexity increased; Coupling decreased; the metrics
of Composition, Polymorphism, Messaging are mainly changeless. From the formulas of Table 4,
the changing of all the metrics above increases the value of quality properties except
Understandability property. The increasing of Design Size, Abstraction and Complexity would
decrease Understandability. But in another hand the removing of FRC bad smell increases
Cohesion and decreases Coupling, so totally the value of Understandability is increased.
Thus, after FRC bad smell detection and corresponding refactoring operation, the values of each
design quality properties are increased, so the total quality of the programs would be improved.
In large scale codes of over 300,000 lines and 1873 classes (ArgoUML), FRC bad smell is
detected to exist in 32 places. Just executing these 32 refactoring operations (about 108 relevant
classes should be changed), the quality standard value of the program is changed from 4.010 to
4.213, and the increasing percentage is about 5%. It is seen that the cost of FRC refactoring is
lower, and benefit is higher.
6 CONCLUSION
In open source programs, the phenomena of function over related between classes without
inheritance relationships, express the essence of frequently invoking between entities of classes.
Actually, some public functional sources (data or operations) can be used by every member, and
this is achieved by inheritance in program design. The classes which should have inheritance
relationships but have no relationships, may cause the difficulty of understanding and
maintenance, and decrease the program quality.
In this paper, the bad code phenomena above are defined as a new bad smell called FRC. This
bad smell exists much in codes and decreases the quality of programs. It is necessary to study the
detecting method and refactoring suggestion for this bad smell.

46
In this paper, the detection approach of FRC bad smell is proposed. The basic process is to collect
the dependency relationships between classes, and compute the invoking rates and compare with
dynamic thresholds. Finally the bad smell is judged to exist or not. Furthermore, according to the
detection results, refactoring suggestions are provided.
The limitation of this paper is the thresholds in FRC bad smell detection. The preset thresholds
decrease the veracity of detection results. In this paper dynamic thresholds are used for bad smell
detection, and different collection and computing process of thresholds are executed according to
the different situation of programs. Therefore the subjectivity is lower, and the detection is more
accurate.
ACKNOWLEDGEMENT
This research is supported by the National Natural Science Foundation of China under Grant
No.61173021 and the Research Fund for the Doctoral Program of Higher Education of China
(Grant No. 20112302120052).
REFERENCE
[1] Fowler M, Beck K. Refactoring: improving the design of existing code. Addison-Wesley Professional.
1999.
[2] Abbes M, Khomh F, Gue W G, Antoniol, G. An empirical study of the impact of two antipatterns,
blob and spaghetti code, on program comprehension. in Software Maintenance and Reengineering
(CSMR), 2011 15th European Conference on. IEEE, march 2011, pp:181 -190
[3] Counsell S. The ‘deception’ of code smells: An empirical investigation. International Conference on
Information Technology Interfaces(ITI). 2010, pp:683-688
[4] Counsell S, Hierons RM, Hamza H, Black S, Durrand M. Exploring the eradication of code smells:
An empirical and theoretical perspective. In: Proc. of the Advances in Software Engineering, 2010.
[5] Luo Yixin. An Ontological Identification of Relationships between Anti-Patterns and Code Smells.
Aerospace Conference. 2010, pp:1-10
[6] J. Kerievsky. Refactoring to patterns. Addison-Wesley , 2004
[7] Abebe S L, Haiduc S, Tonella P, Marcus A. Lexicon bad smells in software. in Proc. Working Conf.
on Reverse Engineering. IEEE, 2009, pp: 95–99
[8] Wake W C. Refactoring Workbook. Addison-Wesley, 2003.
[9] Mantyla M. Bad smells in software: a taxonomy and an empirical study. Ph.D. dissertation, Helsinki
University of Technology, 2003.
[10] Raul M, Carlos L, Yania C. Extending a Taxonomy of Bad Code Smells with Metrics. WOOR’06,
Nantes, 4th July, 2006.
[11] Hayden M, Ewan T. Identifying refactoring opportunities by identifying dependency cycles. In
Proceedings of the 29th Austrasian Computer Science Conference - Volume 48, ACSC '06. 2006.
[12] Tom Tourwe and Tom Mens. Identifying refactoring opportunities using logic meta programming. In
Proceedings of the 7th European Conference on Software Maintenance and Reengineering(CSMR
'03), Benevento, Italy, March 26{28, pages 91{100. IEEE Computer Society, Los Alamitos,
California, March 2003.
[13] Atkinson, D.C., King, T., 2005. Lightweight detection of program refactorings. In: Proceedings of the
12th Asia–Pacific Software Engineering Conference. IEEE CS Press, Taipei, Taiwan, pp:663–670.
[14] Simon, F., Steinbr, F., Lewerentz, C., 2001. Metrics based refactoring. In: Proceedings of the 5th
European Conference on Software Maintenance and Reengineering. IEEE CS Press, Lisbon, Portugal,
pp:30–38.

47
[15] Cristina M, Radu M, Petru M, Daniel R. iPlasma: An integrated platform for quality assessment of
object-oriented design. In: Proceedings of the 21st IEEE International Conference on Software
Maintenance (ICSM 2005). IEEE CS Press, Budapest, Hungary, pp: 77–80.
[16] Nikolaos Tsantalis, Alexander Chatzigeorgiou. Identification of extract method refactoring
opportunities for the decomposition of methods. Journal of Systems and Software, 2011,
84(10):1757–1782.
[17] Jiang Dexun, Ma Peijun. Detecting Bad Smells with Weight Based Distance Metrics Theory. Proc. of
2nd International Conference on Instrumentation, Measurement, Computer, Communication and
Control (IMCCC). 2012:299-304
[18] Marticorena, C López, Y Crespo. Parallel Inheritance Hierarchy- Detection from a Static View of the
System. 6th International Workshop on Object Oriented Reengineering (WOOR), Glasgow, UK.
2006,6.
[19] Jagdish Bansiya, Carl G Davis. A Hierarchical model for object-oriented design quality assessment.
IEEE Transactions on Software Engineering.2002, 28(1):4-17.
[20] IEEE. Standard Glossary of Software Engineering Terminology 610. 12-1990, Vol. 1. Los Alamitos:
IEEE Press , 1999
AUTHOR
Miraldi Fifo was born on July 8th, 1990 in Korçë, Albania. He has graduated the Faculty of Information
Technology at the Polytechnic University of Tirana in October 2011. He holds a Bachelor diploma in
Computer Engineering from 2011 and he completed the first year of the 2-year master program studies in
Computer Engineering (also at the Polytechnic University of Tirana).He has always been interested in
Operating Systems, Computer Networks, Software Engineering etc. He is currently studying for a 2-year
master in Software Engineering at Mälardalen University in Sweden. He has been involved in many
projects during his studies in Sweden and one of them was successfully performed in a distributed
environment which included the Polytechnic University of Milano in Italy and the University of Zagreb in
Croatia. He is currently planning to start his Industrial Master Thesis in Elektronik Partner AB in
Eskilstuna, Sweden.

Functional over related classes bad smell detection and refactoring suggestions

More Related Content

Similar to Functional over related classes bad smell detection and refactoring suggestions (20)

Recently uploaded (20)

Functional over related classes bad smell detection and refactoring suggestions