epassaro / cache-conda-envs Goto Github PK

Thanks for the examples and blog post about this. I had trouble working with your example because you defined each matrix environment separately so you could define a label and prefix. However, I usually use lists for my matrix entries like:

      matrix:
        os: ["windows-latest", "ubuntu-latest", "macos-latest"]
        python-version: ["3.9", "3.10"]

which means I can't easily define prefix/label (as far as I know). I cam up with an alternative that doesn't require the user to know what platform they are on by updating the DATE job:

      - name: Set cache environment variables
        shell: bash -l {0}
        run: |
          echo "DATE=$(date +'%Y%m%d')" >> $GITHUB_ENV
          CONDA_PREFIX=$(python -c "import sys; print(sys.prefix)")
          echo "CONDA_PREFIX=$CONDA_PREFIX" >> $GITHUB_ENV

I also updated my cache entry to include the python version:

      - uses: actions/cache@v3
        with:
          path: ${{ env.CONDA_PREFIX }}
          key: ${{ matrix.os }}-${{matrix.python-version}}-conda-${{ hashFiles('environment.yml') }}-${{ env.DATE }}-${{ env.CACHE_NUMBER }}
        id: cache

Hello, thanks for this code and the great blog post explaining it! This has been very helpful for setting up decent GitHub actions with Conda environments.

I'm wondering if you might be able to explain here, or add to the documentation, the process for manually triggering a new build via the CACHE_NUMBER variable? That is, how is this variable incremented to invalidate the cache? Does this require changing the variable value in the workflow script (therefore requiring a new commit), or can it be done externally somehow when running the workflow? I think the latter might be possible using the repository secret variables, but I might be missing how to do it with the regular environment variables in the workflow as designed. Thanks!