You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Dependencies cache key has to be specific to OS version #432
A cache key looks like this setup-python-Linux-python-3.11.0-beta.3-pip-60d374041db12689c3d346e149aa43744f06aa781ab1c8cbbe70e0b09b98dd70.
It is specific to a Python version and OS type. It has to be specific to a version of an operating system too.
Justification:
PyPI provides source distributions of packages (archives with code) and built distributions (wheels). When a wheel suitable for an environment is not found on PyPI, a source distribution is usually downloaded and a wheel is built locally. The source may contain some code that needs to be compiled (C, Rust, etc.)
On Linux, a compiled wheel may depend on a specific version of a shared library. Different Ubuntu releases used on GitHub Actions may provide different versions of shared libraries. Therefore, a wheel compiled on Ubuntu 20.04 may be incompatible with Ubuntu 22.04. This makes the cache feature unusable when a workflow runs jobs on multiple Ubuntu versions or when an Ubuntu version is changed in a workflow.
https://github.com/urllib3/urllib3/runs/6817232938 is an example of a failed run.
The error is caused by a cffi wheel that was compiled and cached in Ubuntu 20.04 previously where it depended on libffi.so.7.
Ubuntu 22.04 (that provides libffi.so.8) restored that cache archive because it composed the same cache key, and cffi become unusable there (urllib3/urllib3#2626 (comment) provides a little bit more info.)
For now, I will close this issue, if there are any questions or other issues please feel free to continue the conversation here, reopen the issue, or create a new one.
Hi @illia-v, sorry for the delayed answer. At first I have implemented that part as well. But then as I and my team discussed, we realized we are saving virtualenv for poetry and pipenv. For pip we are saving global cache and that's why we had to do it.
Activity
vsafonkin commentedon Jun 13, 2022
Hi @illia-v, we will take a look at this, thank you.
dsame commentedon Jun 27, 2022
@e-korolevskii implementing this feature should solve this issue and this
panticmilos commentedon Jul 19, 2022
Hi @illia-v, we merged PR https://github.com/actions/setup-python/pull/467 that is solving this issue.
For now, I will close this issue, if there are any questions or other issues please feel free to continue the conversation here, reopen the issue, or create a new one.
Cheers
illia-v commentedon Jul 21, 2022
@panticmilos thank you!
illia-v commentedon Jul 21, 2022
@panticmilos what do you think about updating these values too?
setup-python/src/cache-distributions/poetry-cache.ts
Line 36 in 592a7a7
setup-python/src/cache-distributions/pipenv-cache.ts
Line 35 in 592a7a7
panticmilos commentedon Aug 10, 2022
Hi @illia-v, sorry for the delayed answer. At first I have implemented that part as well. But then as I and my team discussed, we realized we are saving virtualenv for poetry and pipenv. For pip we are saving global cache and that's why we had to do it.