Exploring regression testing and software product line testing - research and state of practice

Abstract: In large software organizations with a product line development approach a selective testing of product variants is necessary in order to keep pace with the decreased development time for new products, enabled by the systematic reuse. The close relationship between products in product line indicates an option to reduce the testing effort due to redundancy. In many cases test selection is performed manually, based on test leaders’ expertise. This makes the cost and quality of the testing highly dependent on the skills and experience of the test leaders. There is a need in industry for systematic approaches to test selection. The goal of our research is to improve the control of the testing and reduce the amount of redundant testing in the product line context by applying regression test selection strategies. In this thesis, the state of art of regression testing and software product line testing are explored. Two extensive systematic reviews are conducted as well as an industrial survey of regression testing state of practice and an industrial evaluation of a pragmatic regression test selection strategy. Regression testing is not an isolated one-off activity, but rather an activity of varying scope and preconditions, strongly dependent on the context in which it is applied. Several techniques for regression test selection are proposed and evaluated empirically but in many cases the context is too specific for a technique to be easily applied directly by software developers. In order to improve the possibility for generalizing empirical results on regression test selection, guidelines for reporting the testing context are discussed in this thesis. Software product line testing is a relatively new research area. The understanding about challenges is well established but when looking for solutions to these challenges, we mostly find proposals, and empirical evaluations are sparse. Regression test selection strategies proposed in literature are not easily applicable in the product line context. Instead, control may be increased by increased visibility of the effects of testing and proper measurements of software quality. Focus of our future work will be on how to guide the planning and assessment of regression testing activities in large, complex reuse based systems, by visualizing the quality achieved in different parts of the system and evaluating the effects of different selection strategies when applied in various regression testing situations.