A framework for regression testing

Regression testing is a testing process used to revalidate a modified software system. Current research has focused mainly on unit (module) level regression testing. This dissertation develops a comprehensive strategy for software regression testing at unit, integration and system level, based on selecting a subset of the previous tests (selective retest). As part of the development of these strategies, I have also classified the common errors and faults in combining modules into a working system, and describe a test selection strategy for integration. The concept of a firewall is introduced to reduce regression integration effort. It is shown that re-integration of all modules is not needed for all modifications. This result implies that large saving can be realized through selective retest.;Global variables continue to be widely used despite their many undesirable effects on software maintenance. This dissertation also analyzes the complication in regression testing associated with the use of global variables. It is shown that global variables testing is not more difficult than parameters testing.;This dissertation also answers the question on the economical advantage of using a selective retest strategy and the retest-all strategy. A cost model has been developed and the conditions under which the selective strategy will be more economical than the retest-all strategy have been established. This result can be used to identify the most cost-effective regression strategy to be used for a given change situation.;The proposed regression testing framework has been evaluated based on four successive modifications of a program. The experimentation shows that this strategy is more cost-effective than the retest-all strategy which reruns all previous test cases, using only 34 percent of the tests used by the retest-all strategy.