An Application of Model Seeding to Search-Based Unit Test Generation for Gson

Abstract

Model seeding is a strategy for injecting additional information in a search-based test generation process in the form of models, representing usages of the classes of the software under test. These models are used during the search-process to generate logical sequences of calls whenever an instance of a specific class is required. Model seeding was originally proposed for search-based crash reproduction. We adapted it to unit test generation using EvoSuite and applied it to Gson, a Java library to convert Java objects from and to JSON. Although our study shows mixed results, it identifies potential future research directions.