Pattern Recognition Methods for Determining Software Quality.

摘要

The On-Line Pattern Analysis and Recognition System (OLPARS) was demonstrated as a tool for evaluating program characteristics which contribute to program readability, freedom from errors, and development time. Structural features were extracted automatically from a data base of 155 PL/I programs and used as inputs to OLPARS. As expected, program length had a dominant effect on the time required to understand programs; additional features affecting understandability included GOTO and RETURN statements. Significant contributions of this study were methods for estimating program error rates from archival data, and reliable techniques for estimating understandability of programs. An interesting by-product was the ability to identify individual programmer styles. The principal outcome of the study was the demonstration that OLPARS could provide a facility for evaluating factors contributing to software quality. (Author)