An Empirical Assessment of Error Masking Semantic Metric

Abstract

Semantic metrics are quantitative measures of software quality attributes based on the program functionality not only to the syntax. Different semantic metrics are proposed in literature and most of them are successfully used to assess internal quality attributes like complexity and cohesion. Among these metrics, a recent semantic suite for software testing is proposed to monitor software reliability. The purpose of this suite is to quantify an aspect of software testing and reliability that is fault tolerance by assessing the program redundancy. One of these metrics namely error masking is proposed to reflect the program non-injectivity and measures in bits the amount of erroneous information that can be masked by this program. However, to the best of our knowledge, this metric is only theoretically presented and manually computed. Also, its empirical validation as quantitative measure of erroneous information that a program may mask, still required. Hence, we aim in this paper to empirically assess this metric. So, we ought to propose an automated support tool to automatically generate it and to identify its statistical relationship with two other semantic metrics which are initial and final state redundancy. The experimental study we perform consists of a set of java programs from which we generate the value of these metrics. This study is benefit since the empirical assessment of this metric will help developers to identify the amount of erroneous information that can be masked by their programs.