Exploring Program Synthesis in Model Driven Engineering

Abstract

Today, most software is still developed through conventional methods. That is, the greater the demand for software, the more individuals are hired to carry out the development. This is feasible when the scope of the project is limited and there is no need for prolonged maintenance or enhancements. But, when it is necessary to continue developing a family of products, this makes the final product more expensive because of an increasing number of developers who build the software and maintain it over time, with the counterpart of possibly introducing new bugs in the software of these products due to the repetitive implementation of the features that define it, instead of reusing and modifying the existing ones. By automating these processes, human error is greatly reduced, and production costs are lowered. This project aims to help software developers, particularly, to those who do Model-Driven Engineering (MDE). More specifically, through Program Synthesis (PS) techniques, assist in the tasks that an MDE engineer carries out. We then present a survey covering the state-of-the-art at the intersection of MDE and PS. Finally, we propose an application of Machine Learning (ML) as a synthesis technique in combination with models. Thus, an ML-based assistant, capable of extracting the patterns behind the relationships between model elements, has been developed.