Feedback & Questions: Roche Case Study

[jmanitz]

This is a space for question, feedback, and comments by the community on the Roche case study

[dgkf-roche]

Adding a question that I grabbed from the session chat:

My question for the Roche approach would be, how does the validation manage the fact that users may use different combinations of packages that were not part of the test matrix. Who is responsible for making sure behavior is still correct given the resulting cohort of packages may have incompatible dependencies? Is anything done to check or enforce how dependencies are managed as validated packages are installed from “latest”. Thanks

This is a great question, and hits at the heart of some of the challenges with CRAN-style rolling releases. In all cases, we encourage analysts/developers to keep all packages up-to-date. We can put some guardrails on this by setting a default repos option to point to a snapshot, meaning that users, by default, will get a coherent, validated set of packages. If one chooses to track "latest", then some responsibility would be on their shoulders to stay up-to-date, and swap back over to a later snapshot once their analysis is stable.

Keeping this a bit dynamic allows for developers to be more free to develop rapidly early on (for example, if they're developing a companion package as part of an analysis), and focus on reproducibility when it matters, with some safe defaults for analysts that prefer not to worry about the details.