Also, one thing I need to actually test out, and don’t understand it from a static read. What does “we maintain one cached venv per interpreter+resolve” Is that in the global cache? As in, if I have 2 identical resolves across projects (which is common for me), does that mean they share a cached venv? Or are the cached venvs still isolated on top of that?

We can isolate each version of each venv (by adding a hash of the inputs), but that will cause a lot of cache bloat and is usually unnecessary

adding some entropy based on the full path to the repo root will handle that case?

Haven’t tested on the latest, but something has been manifesting in the weirdest way (which I think is related to the cache thing I flagged above). However, it wasn’t that multiple projects used the wrong cache, it was that even after re-generating lockfiles in the current project, it ran using the OLD deps 🤯

So, need to upgrade and re-try, but I saw that the lockfiles were updated in the diff, but not updated in the lockfile itself (e.g. the .metadata and the default.lock).