Regression testing of enterprise software is expensive as the whole system needs to be retested after code changes. Most of regression testing can be automated, but it still requires a lot of resources and time. Previous work at Ultimate Software designed and implemented a taxonomy manager to allow domain experts to categorize and organize tests by modeling the application domain as a set of hierarchical features, and any cross-dependencies among them. One of the long term goals of this type of modeling is to provide cost effective testing during regression as modifications to the software are validated over time. To enable this type of testing, this work presents the design and implementation of an approach for reestablishing traceability links between features and source code. With that, code metrics can be collected on the feature level and used for assessing risks and the current state of the application on the system level. It is shown that categorizing tests can greatly reduce the amount of time needed to select and run regression tests using an industrial case study. By applying knowledge gained from the traceability links between features and source code to risk-based testing it is also possible to minimize test suites by simplifying or removing tests of low risk features.
