Many-Objective Test Database Generation for SQL

Generating test database for SQL queries is an important but challenging task in software engineering. Existing approaches have modeled the task as a single-objective optimization problem. However, due to the improper handling of the relationship between different targets, the existing approaches face strong limitations, which we summarize as the inter-objective barrier and the test database bloating barrier. In this study, we propose a two-stage approach MoeSQL, which features the combination of many-objective evolutionary algorithm and decomposition based test database reduction. The effectiveness of MoeSQL lie in the ability to handle multiple targets simultaneously, and a local search to avoid the test database from bloating. Experiments over 1888 SQL queries demonstrate that, MoeSQL is able to achieve high coverage comparable to the state-of-the-art algorithm EvoSQL, and obtain more compact solutions, only 59.47% of those obtained by EvoSQL, measured by the overall number of data rows.