Description:

When Poetry 1.2 dependency groups are defined in pyproject.toml and actions/setup-python@v4 caching is enabled with cache: "poetry", the caching does not work very well.

Let's say you have a few dependency groups which you install in your CI with different kind of commands:

Defined like this (semi-real world example) in pyproject.toml:

[tool.poetry.dependencies]
python = ">=3.10, <3.11"
fastapi = "^0.79.0"
uvicorn = {extras = ["standard"], version = "^0.18.2"}
python-dateutil = "^2.8.2"
pytz = ">2018.5"
redis = "^4.0.2"

[tool.poetry.group.docs.dependencies]
mkdocs = "^1.3.1"
mkdocs-awesome-pages-plugin = "^2.7.0"
mkdocs-material = "^8.2.11"

[tool.poetry.group.linting.dependencies]
black = "^22.1.0"
flake8 = "^4.0.1"
flake8-bandit = "^3.0.0"
flake8-bugbear = "^22.1.11"
isort = "^5.10.1"
pre-commit = "^2.15.0"

[tool.poetry.group.testing.dependencies]
coverage = "^6.2"
factory-boy = "2.12.0"
freezegun = "~=1.0"
pytest = "^7.1.1"
pytest-cov = "^3.0.0"
pytest-dotenv = "^0.5.2"
pytest-freezegun = "^0.4.2"
pytest-mock = "^3.7.0"
pytest-randomly = "^3.11.0"
responses = "0.20.0"

[tool.poetry.group.typing.dependencies]
mypy = "0.*"
types-freezegun = "^1.1.3"
types-protobuf = "^3.19.21"
types-python-dateutil = "^2.8.2"
types-pytz = "^2022.2.1"
types-redis = "^4.3.19"
types-requests = "^2.26.0"

Let's say you then use the following for all these CI jobs:

    - uses: actions/setup-python@v4
      with:
        python-version: ${{ inputs.python-version }}
        cache: 'poetry'

    - run: <install-command>

Whichever cache is written first will be used for all these CI jobs. Let's say the docs dependency group finishes first and will be written as the cache.

The end result seems to be here that all the other dependency groups will only get the docs deps cached. And in this example they are not needed for any of the other dependency groups. So for example, the typing CI job will always have to install all its dependencies even if a cache exists.

Now, I'm wondering if I can somehow utilize the cache-dependency-path option to e.g. set this for each CI job so to produce one cache per dependency group. Or is this use case simply not supported today?

It's possible that this won't be supported and if so, I think the docs should include a recommendation on e.g. not using the built-in caching and instead rely on something like this:

env:
    python-version: ???
    dependency-group: ???

...

    - uses: actions/cache@v3
      with:
        path: |
          ~/.cache/pip
          ~/.cache/pypoetry/virtualenvs
          ~/.cache/pypoetry/cache
          ~/.cache/pypoetry/artifacts
        key: ${{ runner.os }}-${{ runner.arch }}-python${{ env.python-version }}-${{ hashFiles('poetry.lock') }}-depgroup${{ env.dependency-group }}

Action version:
Specify the action version

Platform:

• Ubuntu
• macOS
• Windows

Runner type:

• Hosted
• Self-hosted

Tools version:

Repro steps:

1. Add a dependency group no#1 with just one dependency, which will be the one to generate the cache.
2. Add another dependency group no#2 which will use a lot of dependencies.
3. Set up one GHA CI job for each group using actions/setup-python@v4 and specify cache: 'poetry'.
4. See how the CI job using the dependency group no#2 will never get its dependencies cached and always keep installing them.

Expected behavior:

I'd expect to be able to append e.g. the dependency group name to the cache key.

Actual behavior:

I believe I can only add filepaths to the cache option of cache-dependency-path to control the cache key.