From April 12 to 18, 2026, Rio de Janeiro welcomed the international software-engineering community for the 48th IEEE/ACM International Conference on Software Engineering (ICSE 2026).

Artificial or Just Artful? Do LLMs Bend the Rules in Programming?

Authors: Oussama Ben Sghaier, Kevin Delcourt, and Houari Sahraoui

Venue and presentation: Journal Ahead Workshop (JAWs 2026), AI4SE session, Monday, April 13, 10:05–10:10, Bora Bora I.

Read the paper: arXiv preprint · arXiv DOI: 10.48550/arXiv.2512.21028

Code-generating large language models are often evaluated by whether their output passes tests. But what happens when those same tests are visible to the model—even when the prompt tells it not to exploit them? The authors study five models under five prompting conditions that vary test visibility and the restrictions placed on using test information. The results show that access to tests can substantially change model behaviour and, for some models, nearly double correctness. Explicit instructions not to rely on the tests only partly reduce the effect. The study highlights a practical limitation of prompt-only alignment: an agent may continue to exploit useful contextual signals even when instructed not to do so.

Modeling Sampling Workflows for Code Repositories

Authors: Romain Lefeuvre, Maïwenn Le Goasteller, Jessie Galasso-Carbonnel, Benoît Combemale, Quentin Perez, and Houari Sahraoui

Venue and presentation: 23rd International Conference on Mining Software Repositories (MSR 2026), Technical Papers, “Session 2-A: Ecosystems & Methods,” Tuesday, April 14, 14:10–14:20, Oceania V.

Read and reproduce: DOI: 10.1145/3793302.3793369 · HAL preprint · arXiv preprint · source code and replication material

Empirical software-engineering studies frequently construct datasets through several filtering, grouping, random-selection, and set-operation steps. When those decisions are described only informally, it becomes difficult to reproduce a sample or reason about how well it represents the broader population of repositories. This paper introduces an executable domain-specific language, implemented as a Python fluent API, for expressing sampling workflows as explicit and composable operations. The formalized workflow can also produce statistical indicators that help researchers assess representativeness and discuss the generalizability of their findings. A case study shows that the language can model sampling strategies reported in recent MSR research.

Maven-Lockfile: High Integrity Rebuild of Past Java Releases

Authors: Larissa Schmid, Elias Lundell, Yogya Gamage, Benoit Baudry, and Martin Monperrus

Venue and presentation: ICSE 2026 Demonstrations, “Dependability and Security 5,” Thursday, April 16, 11:00–11:15, Oceania X.

Read and try the tool: arXiv preprint · arXiv DOI: 10.48550/arXiv.2510.00730 · source code

Modern projects depend on large networks of third-party libraries, making it difficult to reproduce a past build or verify that its dependencies have not been modified. Although lockfiles address this problem in several package ecosystems, Maven does not provide native lockfile support for Java projects. Maven-Lockfile records direct and transitive dependencies together with checksums. It supports rebuilding historical project versions and detects tampered artifacts, bringing dependency pinning, reproducibility, and integrity checking to Maven-based builds with limited configuration.

Creativity Beyond Functionality: An Observational Study of Easter Eggs in Video Games

Authors: Cristiano Politowski, Claudio Di Sipio, Benoit Baudry, Michalis Famelis, and Houari Sahraoui

Venue and presentation: 10th International Workshop on Games and Software Engineering (GAS 2026), “Paper Session 2—AI, Gameplay Systems, and Serious Games,” Saturday, April 18, 14:15–14:30, Capri III.

Read and explore: DOI: 10.1145/3786171.3788382 · ACM Digital Library · project page

Easter eggs are deliberately hidden pieces of content that often serve no functional purpose, yet they can require substantial creative and technical work. The paper turns this underexplored aspect of game development into an object of systematic software-engineering study. Using a collection of 4,277 Easter eggs from 1,489 games, the authors develop and manually validate a taxonomy of 26 categories. They then examine relationships between Easter eggs and characteristics such as genre, platform, game mode, and rating. The result is both a large-scale empirical account of hidden game content and a foundation for future research on creativity, maintenance, and non-functional features in games.