Automatically Identifying Special and Common Unit Tests Based on Inferred Statistical Algebraic Abstractions

Common and special test inputs can be created to exercise some common and special behavior of the class under test, respectively. Although manually created tests are valuable, programmers often overlook some special or even common test inputs. We have developed a new approach for automatically identifying special and common unit tests for a class without requiring any specification. Given a class, we automatically generates test inputs and identifies common and special tests among the generated tests. Programmers can inspect these identified tests and use them to augment existing tests. Our approach is based on statistical algebraic abstractions, program properties (in the form of algebraic specifications) dynamically inferred from test executions. We use statistical algebraic abstractions to characterize program behavior and identify special and common tests. Our initial experience has shown that many interesting test inputs could be identified among a large number of generated tests.