Diagnosis-Guided Regression Test Refinement

Abstract

The original goal of regression test refinement is to accommodate program changes to insure that new features are property implemented while existing features are not impacted. We found that regression test refinement was also essential for diagnosis of failed tests. This paper proposed and tried out an incremental diagnosis approach to refine regression tests of monotone systems where we use test output patterns to select the most effective test cases from the regression suite for diagnosis. That is, after detection of failed tests, more test data are generated into the regression suite to pinpoint the exact causes of failures. This refinement of the regression test suite is a training process from failed tests, so that the new regression suite is more likely to detect failures and diagnose the causes.