Friday 5th December at 9a.m. Paris time, Gabriel Darbord, Ph.D. Candidate has defended his thesis named “Automatic test generation to help modernize our applications”. His thesis defense took place in Lille, France. Take a look at the summary below.
This thesis is fully in line with the partnership between Berger-Levrault and Inria, which aims to accelerate the development of digital solutions based on the use of artificial intelligence. As part of the joint EVREF project team, Gabriel devoted his software engineering thesis to modernizing the group’s legacy software infrastructure.
Summary
Legacy software systems often lack automated unit testing, making it difficult to detect regressions during maintenance or migration. Our industry partner, Berger-Levrault, faces this challenge as its applications are currently being migrated from monolithic Java systems to a Software-as-a-Service architecture with an Angular interface. To address this issue, this thesis studies the automatic generation of unit tests from program execution traces in order to capture the existing behavior of the system.
We use a technique called “test carving” to extract tests from execution traces. A trace is a record of the behavior observed during execution that includes input and output values. Each trace can serve as the basis for a unit test by replaying the recorded behavior with the same inputs and verifying that the result matches the original output. To facilitate adoption by developers, we generate tests that reconstruct the recorded values as code, improving their readability and maintainability.
These abstractions allow the approach to be specialized for different programming languages and testing frameworks. We implemented our approach in a concrete tool and evaluated it on systems written in Java and Pharo, covering both open-source and industrial software. The results show that it can generate readable tests that capture realistic system behavior.
Overall, this work provides a practical solution for deriving regression tests for legacy systems, while demonstrating how metamodel-based abstractions can be used to create tests in an extensible manner in various technical contexts.
Thesis advisors:
- Etien Anne – Director, Inria Lille Center
- Anquetil Nicolas – Co-Director, Inria Lille Center
Guests:
- Verhaeghe Benoit – Scientific project manager, Berger-Levrault
Some of his Scientific Publications:
- (2025) Pharo Sound Design: Adding Auditory Feedback to a Live Programming Environment: https://cel.hal.science/JIM/hal-05102350v2
- (2025) A Multi-Language Tool for Generating Unit Tests from Execution Traces: https://hal.science/hal-04841805v2
- (2024) Modest-Pharo: Unit Test Generation for Pharo Based on Traces and Metamodels: https://hal.science/hal-04622256v2
- (2023) Migrating the Communication Protocol of Client-Server Applications: https://inria.hal.science/hal-04050310
- (2024) Towards Efficient Belt Conveyor Maintenance: Leveraging Federated Learning: https://ieeexplore.ieee.org/document/10839891
