UVLHub: An Open Science Ecosystem for the Universal Variability Language

Software product line engineering produces families of related software systems from a common set of assets that are systematically reused. This approach exploits the commonalities and manages the variabilities that distinguish one product from another. Feature models are the de facto notation in the community for representing those commonalities and variabilities. They have been a research topic for over three decades. Mature contributions exist in their automated analysis, which derives semantic properties such as satisfiability and dead features, as well as in their configuration and in tool support. Nevertheless, every major feature-modelling tool has historically used its own textual format. This has hindered interoperability and made it difficult to reuse models across studies. The Universal Variability Language (UVL) was proposed as a community effort to unify these serialisations. Its adoption by several research groups and tools has established the conditions for shared infrastructures. Yet the infrastructures around UVL models remain fragmented. Each activity is typically supported by a separate tool. Researchers must assemble ad-hoc chains of format conversions and manual steps that compromise reproducibility and reuse. This dissertation presents UVLHub, an Open Science ecosystem for the complete life cycle of a UVL dataset. The ecosystem is organised around three main components: UVLRepo for sharing, UVLGen for dataset generation, and UVLMng for management. UVLRepo deposits models in a domain-specific repository and validates them on upload. It maintains structured metadata, assigns persistent identifiers, and supports search over model properties. UVLGen produces new UVL models under a shared post-generation validation step. It offers parameterised random generation and natural-language generation through Large Language Models (LLMs). UVLMng coordinates three management capabilities, parsing, analysis, and other capabilities (visualisation and browser-based editing), as a single workflow. The workflow is triggered automatically for every model stored in the repository. UVLHub is, to our knowledge, the first integrated Open Science ecosystem for variability models. Its contribution lies in unifying sharing, generation, and management around a single domain-specific repository, an integration that no prior platform has provided. Every generated model can be deposited immediately. Every stored model becomes automatically analysable, visualisable, citable, and editable without leaving the environment. Every dataset is published with a persistent identifier that enables citation and long-term reuse. UVLHub is designed as an Open Science infrastructure aligned with the FAIR principles (Findable, Accessible, Interoperable, Reusable), which operationalise the three Open Science pillars within scope of this work (Open Data, Open Source, Open Educational Resources). It is anchored in the UVL standard and in open specifications for analysis and metadata rather than retrofitted onto pre-existing tools. At the time of writing, the ecosystem already hosts a live corpus of 1,614 UVL models and is in active international use. The repository component has been evaluated along usability, adoption, and FAIR-compliance axes, with positive results in all three; the generation component has been demonstrated through worked random and LLM-based examples under a shared post-generation validation step that admits only grammar-conforming UVL models; and the management workflow has been exercised through a large-scale solver benchmark over the full corpus. Together, these evaluations indicate that integrating sharing, generation, and management around a common variability language is feasible in practice and yields reproducibility and reuse properties that ad hoc toolchains have so far failed to provide.