Note, tox does support this already, so the stubtest job is the only one (using dependencies) that does work right now.

I rebased and fixed the comments. pip doesn't yet support this, but pypa/pip#13065 is scheduled for 25.1 next weekend.

Added a few notes for the pip 25.1 requirement.

Coming over from #30877, this branch works for a modified uv workflow:

uv venv .venv

# Activate the environment (NOT optional)
source .venv/bin/activate  # Linux/macOS
# or: .venv\Scripts\activate  # Windows

# Install dev dependencies
uv pip install --group dev

# Install matplotlib in editable mode
uv pip install --no-build-isolation --group dev --editable .

# Verify installation
python -c "import matplotlib; print(matplotlib.__file__)"

However, solving dependencies using a strict solver that tries to make a lock file (this includes uv sync or using uv run for anything) is running into repeated issues and dependency conflicts with the "test-minver" group, which go away if that group is removed. I haven't tried pixi, but I expect that this is going to happen with any tool making a lock file that tries to resolve all the dependencies in the base pyproject.toml.

Could we split that group off into its own subproject? Eg something like:

matplotlib/
├── pyproject.toml
└── requirements/
    └── test-minver/
        └── pyproject.toml

requirements/minver-test/pyproject.toml:

[project]
name = "matplotlib-test-minver"
version = "0.0.0"
requires-python = ">=3.11,<3.12"
dependencies = [
    "contourpy==1.0.1",
    "cycler==0.10",
    "fonttools==4.22.0",
    "importlib-resources==3.2.0",
    "kiwisolver==1.3.2",
    "meson-python==0.13.1",
    "meson==1.1.0",
    "numpy==1.25.0",
    "packaging==20.0",
    "pillow==9.0.1",
    "pyparsing==3.0.0",
    "pytest==7.0.0",
    "python-dateutil==2.7",
    "ipython==7.29.0",
    "ipykernel==5.5.6",
    "matplotlib-inline<0.1.7",
]

Consensus on call is to put the minver back in a requirments.txt as we only use it one CI test to keep things simple and unbreak uv.

OK, moved the minimum version requirements to the ci directory.

@scottshambaugh let me know if there are any other problems with uv.

uv installation is super simple now and works without errors!

# Create the environment and install (includes dev group in the env)
uv sync

# Verify installation
uv run python -c "import matplotlib; print(matplotlib.__file__)"

# OR
source .venv/bin/activate
python -c "import matplotlib; print(matplotlib.__file__)"

And pixi works out of the box as well! @tacaswell FYI

# Add tool.pixi sections to pyproject.toml (can skip if preconfigured)
pixi init

# Create the environment and install (creates both default and dev environments)
pixi install

# Verify installation
pixi run -e dev python -c "import matplotlib; print(matplotlib.__file__)"

# OR
pixi shell -e dev
python -c "import matplotlib; print(matplotlib.__file__)"

These are the non-empty sections added to pyproject.toml with pixi init. We could add this by default to simplify setup, though platforms should be extended. I'm not as familiar with pixi so someone else should weigh in.

[tool.pixi.workspace]
channels = ["conda-forge"]
platforms = ["linux-64"]

[tool.pixi.pypi-dependencies]
matplotlib = { path = ".", editable = true }

[tool.pixi.environments]
default = { solve-group = "default" }
dev = { features = ["dev"], solve-group = "default" }

@QuLogic were there any outstanding discussions from the call on this? IMO it's good to merge

I don't recall anything, and I'm not seeing anything in the meeting notes.