Classification for Object-Oriented Cohesion Metrics

Transcription

1 Volume 2, 5, May 2013 Classification for Object-Oriented Cohesion Metrics TejdedaAlhussenAlhadi, Dr. AbdualhafidShuwehdi,Sagaya Aurelia Xavier Patrick Kishore, Dr. Omer Jomah Abstract This paper presents a number classification criteria that may serve as an evaluation scheme for objectoriented cohesion metrics. Based on these criteria, we present a critical survey the state--the-art object-oriented cohesion metrics. This classification contributes to an increased understanding the state--the-art as it is a mechanism for:comparing measures and their potential use, integrating existing measures which examine the same concepts in different ways and facilitating more rigorous decision making regarding the definition new measures and the selection existing measures for a specific goal measurement. Index Terms Class, Cohesion metric, Criteria I. INTRODUCTION In object-oriented stware, classes replace modules with methods and attributes as their elements. In this context, cohesion is the degree to which the methods and attributes a class belong together. Again, recent researches have led to a large number new cohesion measures for object-oriented systems. However, because cohesion is a complex stware attribute in object-oriented systems (e.g., there are several different mechanisms which are considered to contribute to the cohesion a class), and there has been no attempt to provide a structured synthesis, because there is no standard terminology and formalism for expressing measures, many measures are not fully operationally defined, i.e., there is some ambiguity in their definitions. As a result, it is difficult to understand how different cohesion measures relate to one another. Moreover, it is also unclear what the potential uses many existing measures are and how these different measures might he used in a complementary manner. In this paper, we begin with discussion the classification criteria for object-oriented cohesion metrics and present a TejdedaAlhussenAlhadi, Department Computer Science, Faculty Education, Azzaytuna University, Baniwalid, Libya Dr. AbdualhafidShuwehdi, Department Computer Science,Libyan Academy,Tripoli, Libya. Sagaya Aurelia, Department Computer Science, Faculty Education, Azzaytuna University, Baniwalid, Libya , Xavier Patrick Kishore, Department Computer Science, Faculty Education, Azzaytuna University, Baniwalid, Libya , Dr. Omer Jomah, Department Computer Science,Faculty Education, Azzaytuna University, Baniwalid, Libya, critical analysis the existing object-oriented cohesion metric in the light the classification criteria. II.CLASSIFICATIONS OF COHESION MEASUREMENTS In this section, we identify some classification criteria that can be used to classify, assess, compare, and evaluate the existing cohesionmetrics. The classification criteria are two types: Factors and characteristics. Factors:these criteria identify what the metric considers in its calculation cohesion value. s:these criteria, as the name implies, capture the characteristics the cohesion metric i.e. the features the metric. As shown in Table 1, criterion 3, 7, 9, 10 and 11 can be found in [1]. The table presents a summary these criteria where new (in the source column) signifies new criteria that we identified in the course this paper. Table 1: Classification Criteria Classification Criteria Underlying Granularity Availability Sensitivity rmalization Validation Description Source Briand [1] Briand [1] Remark Braind et. al. called this usable or partially usable. 502

2 Volume 2, 5, May Interpretation Connection type Special s Inheritance Factor Factor Factor Modified Briand [1] Briand [1] Modified version Braind s cohesion criteria [1]. Braind et. al. called this known problem A. Underlying To better understand a concept, it isimportant to understand the underlying principle upon which the work is built; this may be obtained by knowing where the original idea is obtained. Underlying gives the reference the work where the original idea, upon which the is built, is obtained. B. Granularity Granularity refers to the granularity the metric, as in which component the system does the metric measure; method, class or package. C. Availability Availability determines the stwareengineering development process in which the metric can fully be used. Some metrics can only be used when coding is completed such metrics are implementation. While some metrics can be used at the end the design stage such metrics are design the stware development process. D. Sensitivity Sensitivity describes how sensitive the cohesion metric is to changes. How does a change in the module (or class) affect the measurement? Does the change have negative or positive impact on the result the metric? E. rmalization rmalization deter mines if the result the metric is normalized i.e. values returned by the metric is between 0 and 1; classes with zero cohesion value have the least cohesion while classes with cohesion value 1 have perfect cohesion. Or the reverse is the case for inverse metrics like the LCOM metrics. F. Validation Validation specifies whether the metric is validated or not; if it is validated how is it validated-theoretically or empirically. If it is not validated how complex is the validation process, as in what are the things required for it to be validated and whether or not the researchers have given a way that their metric can be validated. G. Interpretation Interpretation determines the difficulty surrounding the interpretation the results obtained from the metric. It also describes whether the researchers have given suggestions on how to interpret the values the metric or not. H. Connection Type Connection type specifies factors the cohesion metric considers in calculating the cohesion the module i.e. the process the researchers used in capturing the interactions among the different components a class. I. Special s s like constructor, access methods etc have an impact on the cohesion a class. The special method attribute captures whether the impact such methods is considered in the definition the cohesion measure. J. Inheritance Inheritance describes whether the considers the impact inheritance in proposing its metric. Inherited methods and attributes have an impact on the cohesion a class. Table 2 presents all the possible types connections used by the existing cohesion metrics. This table is a modified version the work presented by Briand et al. in [1]; here we identified three more connection types: type 3, 8 and 9. Table 2: Connection Type Element 1 m class c m class c m class c Eleme nt 2 Attribu te a Description Name Measures m references a m invokes m directly m relates to m indirectly via other methods that directly invoke each other. MAR DMMR IMMR LCOM5, COMA, LCOM4, Co, CCM, ECCM, OCC, PCC, COMA COMA 503

3 Datadeclara tion in Param eter 1 ISSN: m m, m m m, m m Datadeclaratio nin Datadeclaratio n in Parameter 2 m m relates to m indirectly via other methods that directly invoke each other. m an directly reference an attribute a in common m an indirectly reference an attribute a in common Data-data interaction Datamethod interaction The existence Parameter 1 and Parameter 2 in the same method or otherwise m writes to an attribute a and m reads a IMMR DAS IAS DDI DMI PPI MIBAT COMA LCOM1, LCOM2, LCOM3, LCOM4, Co, TCC, CCM, ECCM, OCC, PCC, LCC LCC RCI RCI CAMC PCC Volume 2, 5, May 2013 Where; MAR means - Referencing DMMR means Direct - Referencing IMMR means Indirect - Referencing DAS means Direct Sharing IAS means Indirect Sharing DDI means Data-Data Interaction DMI means Data- Interaction PPI means Parameter-Parameter Interaction (or Intersection). MIBAT: s Interactions Based on Access Types III. Critical analysis object oriented cohesion metrics In this section we critically analyze the different es; all the object-oriented es are scrutinized based on the classification criteria. A. Evaluating the Degree and Class Cohesion Eder et al. [2] Eder et al. define degrees for measuring method cohesion. In this section we discuss the proposeetrics in the light the attributes discussed in Table 1. Table 3: Eder's et al. 1 Underlying An extension the concepts cohesion developed initially for procedure-oriented systems by Yourdon and Constantine [3] 2 Granularity Measure cohesion at method and 3 Availability Implementation 4 Sensitivity t sensitive 5 rmalization t normalized 6 Validation t validated 7 Interpretation t explanation is given for the metric interpretation 8 Cohesion Criteria Provides ordinal scales for capturing method cohesion. 9 Special s consideration was given 10 Inheritance Inheriteethods and attributes are considered in the case inheritance cohesion. B.Evaluating the LCOM1 and LCOM2 Metrics [4][5] In their, proposed the lack cohesion in methods (LCOM) and later redefine this metric; we call the two metrics LCOM1 and LCOM2. 504

4 Volume 2, 5, May 2013 Table 4: LCOM1 1 Underlying Based on the notion degree similarity methods initially proposed by Bunge [6] 2 Granularity Measures cohesion at class 3 Availability Partially 4 Sensitivity Reacts positively to changes though not in all situations. That is if the cohesion a class is altered, the metric reflects this alteration in its computation sometimes. 5 rmalization t normalized 7 Interpretation explanation on how to interpret the result the metric 8 Cohesion Criteria DAS (Direct Sharing) see Table 2 9 Special s consideration was given 10 Inheritance t considered Table 5 : LCOM2 1 Underlying Based on the notion degree similarity methods initially proposed by Bunge [6] 2 Granularity Measures cohesion at class 3 Availability Partially. 4 Sensitivity Behaves positively to changes in the cohesion a class though not in all situations 5 rmalization t normalized and empirically 7 Interpretation The researchers here asserted that high value LCOM2 is not desirable because high value the metric indicates disparateness in the functionality provided by the class. 8 Cohesion Criteria DAS (Direct Sharing) see Table 2 9 Special s consideration was given for special methods 10 Inheritance t considered C.Evaluating The LCOM3, LCOM4 and Co Metrics [8][9] Hitz and Montazeri proposed three metrics based on the LCOM metric; their proposeetrics are: LCOM3, LCOM4 and Co. In Table 6, Table 7, and Table 8 these metrics are carefully analyzed based on the classification criteria. Table 6: LCOM3 1 Underlying Based on the work 2 Granularity Captures cohesion at the class 3 Availability Partially 4 Sensitivity t sensitive to changes, it does not always behave the way it should when the interactions among the components in a class are tempered with. 5 rmalization t normalized 6 Validation t validated 7 Interpretation Difficult to interpret the result 8 Cohesion Criteria DAS (Direct Sharing) see Table 2 9 Special s consideration was given 10 Inheritance t considered Table 7: LCOM4 1 Underlying Based on the work 2 Granularity Captures cohesion at the class 3 Availability Partially 505

5 Volume 2, 5, May Sensitivity t very sensitive to changes, it does not always behave the way it should when the interactions among the components in a class are tempered with. 5 rmalization t normalized 6 Validation t validated 7 Interpretation Difficult to interpret the result 8 Cohesion Criteria Uses DAS and DMMR Table 2 9 Special s Considers access method by including an edge between methods in the undirected graph whenever one the methods invokes the other. 10 Inheritance t considered To further discriminate classes that have the value LCOM = 1, Hitz and Montazeri proposed a thiretric called connectivity (Co). This metric is analyzed in Table 8. Table 8: The Connectivity Metric 1 Underlying Based on the work 2 Granularity Captures cohesion at the class 3 Availability Partially 4 Sensitivity t very sensitive to changes, it does not always behave the way it should when the interactions among the components in a class are tempered with. 5 rmalization normalized 6 Validation t validated 7 Interpretation interpretation is provided for this metric 8 Cohesion Criteria Uses DAS and DMMR Table 2 9 Special s Considers access method - this metric is an extension LCOM4. The metric is proposed in order to further discriminate classes with one connected component in the undirected graph by considering the number edges in the graph. 10 Inheritance t considered D.Evaluating the TCC and LCC Metrics [10] Bieman and Kang proposed two cohesion metrics: TCC (Tight Class Cohesion) and LCC (Loose Class Cohesion). TCC and LCC are critically discussed based on the classification criteria in Table 9 and Table 10, respectively. Table 9: Tight Class Cohesion 1 Underlying Based on the work 3 Availability Partially 4 Sensitivity t very sensitive 5 rmalization normalized 6 Validation t validated 7 Interpretation Difficult to interpret the result 8 Cohesion Criteria Uses IAS Table 2 9 Special s Recommended that constructors be excluded 10 Inheritance Suggest three alternatives for handling inherited attributes/methods. The seconetric; Loose Class Cohesion (LCC) considers - in its definition both direct and indirect connections that may exist among the components a class. Table 10: Loose Class Cohesion 1 Underlying Based on the work 3 Availability Partially 4 Sensitivity t very sensitive. 5 rmalization normalized 6 Validation t validated 7 Interpretation Difficult to interpret 8 Cohesion Criteria Uses IAS Table 2 9 Special s Recommended that 506

6 Volume 2, 5, May 2013 constructors be excluded in the analysis 10 Inheritance Suggest three alternatives for handling inherited attributes/methods. E.Evaluating the LCOM5 Metric [11] Henderson-Sellers et al. proposed a new metric by redefining the Lack Cohesion in s (LCOM) metric; we call their metric LCOM5. Below, the metric is discussed based on the classification criteria. Table 11: LCOM5 1 Underlying Based on the work 3 Availability Partially 4 Sensitivity Sensitive to changes 5 rmalization normalized 6 Validation t validated 7 Interpretation Difficult to interpret 8 Cohesion Criteria Uses MAR Table 2 9 Special s consideration was given 10 Inheritance t considered F.Evaluating the RCI Metric [12] Briand et al. proposed cohesion metric, RCI (Ratio Cohesive Interaction), based on the visualization a class as a collection data declarations anethods. This metric is critically discussed in Table 12. Table 12: The RCI Metric 1 Underlying Adapted from the early work Briand et al. for measuring cohesionin object-based systems. 3 Availability Fully design 4 Sensitivity t very sensitive to changes 5 rmalization rmalized and empirically 7 Interpretation interpretation was provided. 8 Cohesion Criteria DDI and DMI Table 2 9 Special s t considered 10 Inheritance In addition to the three options provide by Bieman and Kang, Briandet al. added a fourth alternative for handling inheritance: excluding inherited attributes but include inherited methods. According to themthis makes little sense because inheriteethods can only access inherited attributes G.Evaluating the CAMC Metric [13] Bansiya et al. [13], proposed a metric whereby the cohesion a class is determined by the types objects that methods take as input parameters. The metric CAMC (Cohesion Among s in Class) measures the extent intersections individual method parameter type lists with the parameter type list all methods in the class. This metric is discussed in Table 13 in light the classification criteria. Table 13: The CAMC Metric 1 Underlying Based on the premise that the parameters a method reasonably define the types interaction that methods may implement. 3 Availability Fully design 4 Sensitivity t very sensitive 5 rmalization rmalized 6 Validation Validated empirically 7 Interpretation explanation on how to interpret the result this metric is provided 8 Cohesion Criteria PPI Table 2 9 Special s t considered 10 Inheritance t considered H.Evaluating the CCM and ECCM Metrics [14] Jarallah et al. [14] conducted a research on some the objectoriented design features that can affect the cohesion a class and attempted to relate how cohesion can be used to assess these design features. Two metrics were proposed: (i) CCM 507

7 Volume 2, 5, May 2013 (Class Cohesion Metric) and (ii) ECCM (Enhanced Class Cohesion Metric). Table 14: The CCM Metric 1 Underlying Based four design principles proposed in [14] 3 Availability Partially 4 Sensitivity t very sensitive to changes 5 rmalization rmalized 7 Interpretation explanation on how to interpret the result this metric is provided 8 Cohesion Criteria DMMR and DAS Table 2 9 Special s consideration was given 10 Inheritance t considered Table 15: The ECCM Metric 1 Underlying Based four design principles proposed in [14] 3 Availability Implementation 4 Sensitivity t very sensitive to changes 5 rmalization rmalized 7 Interpretation explanation on how to interpret the result this metric is provided 8 Cohesion Criteria DMMR and DAS Table 2 9 Special s consideration was given 10 Inheritance Suggest (i) avoiding unused inheriteethods in a subclass (ii) avoiding re-implementation inheriteethods I.Evaluating the OCC and PCC [15] Aman et al. proposed two metrics: (i) OCC (Optimistic Class Cohesion) and (ii) PCC(Pessimistic Class Cohesion). In Table 16 and Table 17, we critically analyzed these metricsbased on the classification criteria. Table 16: The OCC Metric 1 Underlying Based on the work 3 Availability Partially 4 Sensitivity t very sensitive 5 rmalization rmalized 7 Interpretation interpretation was given 8 Cohesion Criteria DMMR, DAS and MRBAT Table 2 9 Special s Access methods are treated as normal methods 10 Inheritance t considered Table 17: The PCC Metric 1 Underlying Based on the work 3 Availability Partially 4 Sensitivity t very sensitive 5 rmalization rmalized 7 Interpretation interpretation was given 8 Cohesion Criteria DMMR, DAS and MRBAT Table 2 9 Special s Access methods are treated as normal methods 10 Inheritance t considered IV.OBSERVATIONS Based on the critical analysis state--the-art existing cohesion metrics, the following are our findings: 1. Most the es are based on the work Chidamber and Kemerer notion degree similarity methods initially proposed by Bunge. 508

8 Volume 2, 5, May Most the es are based on attribute usage, method invocations or both. 3. Most the metrics studied capture cohesion at the class. 4. ne the metrics studied captures cohesion at the package. 5. Most the cohesion metrics studied does not accurately capture the cohesion a class without violating at least one example. 6. Most the metrics are not validated and few researchers provide explanation on how to interpret the result their metrics. [10] Bieman J. M., Kang B., Cohesion and Reuse in an Object-Oriented System, in Proc. ACM Symp. Stware Reusability (SSR 94), ,, [11] Henderson-Sellers B., Constantine L. L., Graham I. M., Coupling and cohesion (towards a valietrics suite for object-oriented analysis and design), Object Oriented Systems 3, (1996) [12] Briand L., Morasca, Basili V., Defining and Validating High- Level Design Metrics, Computer Science Technical Report CS-TR 3301, University Maryland at College Park, [13] Bansiya J., Etzkorn L., Davis C., and Li W., A Class Cohesion Metric for Object Oriented Designs, Jornal Object-Oriented Programming (January), pages 47-52, [14] Jarallah A., Wasiq M., and Ahmed M. A., Principle and Metrics for Cohesion-Based Object-Oriented Component Assessment. Confidential Draft Copy, [15] Aman H., Yamasaki K., Yamada H. da M., A Proposal Class Cohesion Metrics Using Sizes Cohesive Parts, Knowledge-based Stware Engineering, T. welzer et al.(eds.), pp , IOS Press, Sept Some the metrics are not normalized. In an attempt to address some the short comings some the existing cohesion metrics V.CONCLUSION We have provided criteria for the comparison, evaluation, and definition cohesion measures in object oriented systems. These classification criteria for object-oriented cohesion metrics is intended to be exhaustive and integrates new ideas with existing measurement frameworks in the literature. Thus, detailed guidance is provided so that cohesion measures may be defined in a consistent and operational way and existing measures may he selected based on explicit criteria. REFERENCES [1] Briand L., Daly J., Wust J., A Unified Framework for Cohesion Measurement in Object Oriented Systems, Technical Report ISERN Kaiserlautern: Germany: Fraunher Institute for Experimental Stware Engineering, [2] Eder J., Kappel G., Schrefl M., Coupling and Cohesion in Object- Oriented Sysems, Technical Report University Klagenfurt, Institute Computer Science, [3] Yourdon E. and Constantine L., Structured Design, Prentice Hall, [4] Chidamber R., Kemerer C. F., Towards a Metrics Suite for Object Oriented Design, in A. Paepcke, (ed.) Proc. Conference on Object- Oriented Programming: Systems, Languages and Applications (OOPSLA 91), SIGPLAN tices 26(11), , [5] Chidamber, Kemerer C., A Metrics Suite for Object Oriented Design, IEEE Transactions on Stware Engineering, vol 20 no6, June [6] Bunge M., Treatise on Basic Philosophy: Ontology I: The Furniture the World, Boston, Riedel, [7] Basili V. R., Briand L. C., and Melo W. L., A Validation object- Oriented Design Metrics as Quality Indicators, IEEE Transactions on Stware Engineering, 22(10), (1996). [8] Hitz M., Montazeri B., Measuring Coupling and Cohesion in Object-Oriented systems, in Proc. Int. Symposium on Applied Corporate Computing, Monterrey, Mexico, October [9] Hitz M., Montazeri B., s Metrics Suite: A Measurement Theory Perspective, IEEE Transaction on Stware Engineering, vol 22 no 4, April