• An ML engineer wants to use a python package that can't run directly on our developer's machines (macos) because it is linux-only (due to cuda, etc.) • It seems like environments are a perfect solution to this: just create a "linux-docker" environment and try to run there. Except: ◦ Lockfile generation fails, because the package is incompatible with macos. Okay, fair enough. ◦ Idea: I can create a separate resolve for packages that will run in that environment, generate lockfiles there, and voila! Good to go. Except... ▪︎ Oh, right. Lockfile generation is not environment-aware. Hm.