A Semantic-Based Approach for Detecting and Decomposing God Classes

TestElement ProxyControlGUI 22 46 1.12 29 57 0.98 23 252 0.98 24 30 1.20 22 58 1.08 Figure 8: Measures of CBO, NOM, and IC For each candidate class, a complete weighted graph is built. Figure 9(a) shows the graph for the MenuFactory class. The graph involves 22 nodes representing the methods defined in the class and 231 edges connecting the nodes. The threshold for the class ranges from 0.95 to 2.09 and we experimented a threshold ranging from 1 to 2.1 incremented by 0.1. No edge is removed with threshold 1.0 causing no split and 217 edges are removed with threshold 2.1 resulting in three sub-graphs. In a manually verification, threshold 1.5 produces the best result, which decomposes into two subgraphs with 157 edges removed, each sub-graph representing a single responsibility. One sub-graph involves 18 methods and the other involves 4 methods. Figure 9(b) shows the resulting decomposition. Table 8 shows the threshold ranges used in experiments for each of the candidate God classes. 6.1 Results Analysis From the case study, we observe three types of God classes shown in Figure 10. Type A has two heterogeneous responsibilities, each having an independent set of fan-in and fanout interactions, which should be put in a separate class. This is an example of a obvious need for responsibility decomposition. The MenuFactory, SampleResult , and AbstractTestElement classes belong to Type A. The ManuFactory class has responsibilities of 1) creating menus and 2) controlling the drag and drop function. While the drag and drop function supports menus, its controlling responsibility is not directly related to menus. The SampleResult class involves responsibilities of 1) collecting and storing sample results and 2) measuring the time taken to collect sample results. The collecting and storing functions in the first responsibility is quite heterogeneous to the measuring function in the second responsibility. The AbstractTestElementclass has responsibilities of 1) configuring properties of tested elements and 2)configuring thread context. The target objects concerned in the two responsibilities are completely different types, and thus the responsibilities share no commonality. Table 8: Thresholds and ISS Statistics God Class Item Mean Std. Dev. Min Max Mean Std. Dev. Min Max Mean Std. Dev. Min Max Mean Std. Dev. Min Max Mean Std. Dev. Min Max Value MenuFactory 0.95 ~ 2.09