Can Source Code Analysis Indicate Programming Skills? A Survey with Developers.

Background: Both open-source and proprietary software systems have become increasingly complex. Despite their growing complexity and increasing size, software systems must satisfy strict release requirements that impose quality, putting significant pressure on developers. Therefore, software projects' success depends on the identification and hiring of qualified developers. Several approaches aim to address this problem by automatically proposing models and tools to automatically identify programming skills through source code. However, we still lack empirical knowledge on the applicability of these models in practice. Aims: Our goal is to evaluate and compare two models proposed to support programming skill identification. Method: This paper presents a survey with 110 developers from GitHub. This survey was conducted to evaluate the applicability of two models for computing programming skills of developers based on the metrics Changed Files and Changed Lines of Code. Results: Based on the survey results, we conclude that both models often fail to identify the developer's programming skills. Concerning precision, the Changed Files model obtained 54% to identify programming languages, 53% for back-end & front-end profiles, and 45% for testing skills. About the Changed Lines of Code model, we obtained 36% of precision to identify programming languages, 45% for back-end & front-end profiles, and 30% for testing. Conclusion: Practitioners can use our survey to refine the practical evaluation of professional skills for several purposes, from hiring procedures to the evaluation of team.