Support cache_save and cache_restore as separate options

[pulkit-apriori]

The current version supports a single cache flag which controls both saving and restoring.

In my workflow, I generally save caches only from the main branch to avoid trashing (non-configurable 10GB limit) but it is useful to have the ability to restore those caches to non-default branches.

Github's cache actions has two separate actions allowing this behavior - https://github.com/actions/cache/blob/main/restore/README.md and https://github.com/actions/cache/blob/main/save/README.md

[davidbuzinski]

Hi @pulkit-apriori ,

The point you make about saving and restoring makes sense, but I'd like to understand the use-case a little better in the case for a MATLAB installation. In this action, we only save if the installation completes successfully and no matching cache key exists. This means we're very rarely writing the cache entry. It typically only happens on first install or cases where the release or products inputs change.

We do it this way because we consider MATLAB closer to a tool that would be stored in a runner's "tool cache" (ex: Node, Python, etc.) than something like npm or pip dependencies, which may be updated far more frequently.

The ability to control when reads and writes occur is certainly possible as you noted, but comes with additional configuration options/ overhead so I want to understand the use-case a little better before moving forward with this request.

Thanks!
David B

[pulkit-apriori]

Hi @davidbuzinski , the problem I'm trying to avoid is thrashing due to the limited cache storage space. If I'm trying out changes to the CI set up / Matlab version in new branch, I'd prefer to not delete caches set up by the main branch by saving it. At the same time, loading from cache is desirable on all branches.