Marvin Muñoz Barón

Sebastian Baltes, Florian Angermeir, Chetan Arora et al.

Large Language Models (LLMs) are widely used in software engineering (SE) research and practice, yet their non-determinism, opaque training data, and rapidly evolving models threaten the reproducibility and replicability of empirical studies. We address this challenge through a collaborative effort of 22 researchers, presenting a taxonomy of seven study types that organizes how LLMs are used in SE research, together with eight guidelines for designing and reporting such studies. Each guideline distinguishes requirements (must) from recommended practices (should) and is contextualized by the study types it applies to. Our guidelines recommend that researchers: (1) declare LLM usage and role; (2) report model versions, configurations, and customizations; (3) document the tool architecture beyond the model; (4) disclose prompts, their development, and interaction logs; (5) validate LLM outputs with humans; (6) include an open LLM as a baseline; (7) use suitable baselines, benchmarks, and metrics; and (8) articulate limitations and mitigations. We complement the guidelines with an applicability matrix mapping guidelines to study types and a reporting checklist for authors and reviewers. We maintain the study types and guidelines online as a living resource for the community to use and shape (llm-guidelines$.$org).

Marvin Muñoz Barón, Marvin Wyrich, Stefan Wagner

Background: Developers spend a lot of their time on understanding source code. Static code analysis tools can draw attention to code that is difficult for developers to understand. However, most of the findings are based on non-validated metrics, which can lead to confusion and code, that is hard to understand, not being identified. Aims: In this work, we validate a metric called Cognitive Complexity which was explicitly designed to measure code understandability and which is already widely used due to its integration in well-known static code analysis tools. Method: We conducted a systematic literature search to obtain data sets from studies which measured code understandability. This way we obtained about 24,000 understandability evaluations of 427 code snippets. We calculated the correlations of these measurements with the corresponding metric values and statistically summarized the correlation coefficients through a meta-analysis. Results: Cognitive Complexity positively correlates with comprehension time and subjective ratings of understandability. The metric showed mixed results for the correlation with the correctness of comprehension tasks and with physiological measures. Conclusions: It is the first validated and solely code-based metric which is able to reflect at least some aspects of code understandability. Moreover, due to its methodology, this work shows that code understanding is currently measured in many different ways, which we also do not know how they are related. This makes it difficult to compare the results of individual studies as well as to develop a metric that measures code understanding in all its facets.