The effect of interface complexity on program error density

The paper explains how to evaluate software maintainability by considering the effects of interfaced complexities between modified and unmodified parts. When software is maintained, designers develop software, considering not only functions of modified parts but also those of unmodified parts by reading specifications or source codes. So, not only does the volume and complexity of the unmodified and modified parts, but also the interfaced complexities between, them greatly affect the occurrence of software errors. The modified part consists of several subsystems, each of which is composed of several functionally related routines. The unmodified part also consists of several routines. A routine corresponds to a function of C-coding. We experimentally show by regression and discriminant analyses that the quality of each routine tends to decrease as the reuse-ratio increases. The optimal threshold of reuse-ratio is selected by applying AIC (Akaihe Information Criterion) procedures to discriminant analysis for software quality classification. We can reasonably separate routines into two parts, one modified in which most software errors occur and the other unmodified in which few errors occur. The interfaced complexities measured by the extended cyclomatic number between each sub system of the modified part and the unmodified part greatly affect the number of errors and error density. By applying regression analysis to medium size software, we have shown that 40% of variance of error density are represented by interfaced complexities among each subsystem of the modified part and the unmodified part