Determinants of program repair maintenance requirements

Abstract

Considerable resources are devoted to the maintenance of programs including that required to correct errors not discovered until after the programs are delivered to the user. A number of factors are believed to affect the occurrence of these errors, e.g., the complexity of the programs, the intensity with which programs are used, and the programming style. Several hundred programs making up a manufacturing support system are analyzed to study the relationships between the number of delivered errors and measures of the programs' size and complexity (particularly as measured by software science metrics), frequency of use, and age. Not surprisingly, program size is found to be the best predictor of repair maintenance requirements. Repair maintenance is more highly correlated with the number of lines of source code in the program than it is to software science metrics, which is surprising in light of previously reported results. Actual error rate is found to be much higher than that which would be predicted from program characteristics.