Reflections on cross-cutting software tools born out of project work

Abstract

RSEs regularly collaborate on time-bounded projects, with code ownership remaining with researchers. Sometimes, however, such collaborations lead to the development of cross-cutting digital research infrastructure with potential for impact beyond a single project or domain.How can RSEs identify and nurture these pieces of software, and how can they use their unique position in research to drive these interconnections?ICCS RSEs work with climate scientists to improve the functionality and sustainability of climate modelling code. Much of this is written in Fortran, which remains relevant due to continued updates to the language standards and compilers, and considerable societal inertia, but lacks native support for machine learning (ML). With hybrid modelling being a popular research topic, a common request from collaborators is to facilitate inter-operation of their codebases with the popular ML package PyTorch. As a result we developed the FTorch interface. Given the cross-cutting nature of ML inter-operation, FTorch was spun out of the original project as a fully-fledged package in its own right and has since been deployed in several other projects and domains.In this talk, we reflect on our experiences developing FTorch. We discuss the decision to spin FTorch out beyond the original collaboration, and the subsequent journey to become an independent package developed and maintained by our team. Early on we faced the challenge of developing FTorch in ‘non-project time’ and managing it in-house (something not generally within ICCS’ remit for project work). We subsequently secured FTorch-specific funding to support development, promotion, and collaboration. Finally, we discuss how the unique role of the RSE is well suited to spotting opportunities for interconnected research.We hope that the talk will inspire attendees to focus on cross-cutting software that they have developed in their team and provide useful tips for how to grow such software into fully-fledged projects.Acknowledgements We acknowledge the valuable contributions of all current and past FTorch developers and contributors.ICCS is supported by Schmidt Sciences, LLC.This project was supported with funding from the Cambridge Centre for Data-Driven Discovery and Accelerate Programme for Scientific Discovery, made possible by a donation from Schmidt Sciences.A recording of this session is available on YouTube: https://youtu.be/fLVt3SAfXw0