How to set up your development environment

With Pixi

The development workflow is built around Pixi. Install Pixi and Git, then clone this repository alongside conda/conda, whose test data is used by the test and benchmark tasks:

git clone https://github.com/conda/conda.git
git clone https://github.com/conda/conda-rattler-solver.git
cd conda-rattler-solver

The tasks expect conda’s test data at ../conda/tests/data. Pixi installs the locked dependencies and an editable copy of this repository when you run them. The development dependencies bootstrap conda with a compatible solver from conda-canary/label/dev, then replace it with this checkout for tests and benchmarks.

For example, to run the test suite with Python 3.13:

pixi run --locked -e test-py313 test

The available environments are defined in pyproject.toml. The ones you will use most often are:

Environment	Purpose
test-py311, test-py312, test-py313, test-py314	Run the tests and benchmarks on a given Python version
lint	Run the pre-commit hooks
docs	Build this documentation
build	Build the conda package from recipe/

See Typical development workflows for the tasks each one offers.

With devcontainer in VS Code

A devcontainer configuration is bundled in this repository for a disposable Linux environment. You’ll need Docker and VS Code with the Dev Containers extension:

1. Clone conda/conda and conda/conda-rattler-solver to your preferred locations, for example, ~/devel/conda and ~/devel/conda-rattler-solver. The location does not matter as long as both repositories have the same parent directory.

2. Open your conda-rattler-solver clone with VS Code.

3. Connect to the DevContainer image via the bottom-left menu (❱❰) and click on “Reopen in Container”. The image is based on Miniforge.

4. The image will be built and after a couple of minutes, you’ll be dropped into a Bash shell. Since the local repositories are mounted, you can make modifications to the source live, and they will be reflected in the Docker instance automatically. Run pytest or conda as needed, no need to reload Docker!

5. If the development environment breaks, click again on ❱❰ and, this time, choose “Rebuild container”. You might need to retry a couple of times.

With regular Docker

You can reuse the devcontainer scripts with regular Docker too.

1. Clone conda/conda and conda/conda-rattler-solver to your preferred locations, as above.

2. Start a new Docker instance with this command. Adjust the local mounts as necessary.

   docker run -it --rm \
     -v ~/devel/conda:/workspaces/conda \
     -v ~/devel/conda-rattler-solver:/workspaces/conda-rattler-solver \
     condaforge/miniforge3:latest \
     bash
   

3. Run the post_create and post_start scripts:

   bash /workspaces/conda-rattler-solver/.devcontainer/post_create.sh
   bash /workspaces/conda-rattler-solver/.devcontainer/post_start.sh
   

Manual setup

If you prefer to manage the environment yourself, be careful with overwriting your existing conda installations, especially when it comes to shell initialisation.

1. Get yourself familiar with the “Development environment” guide for conda itself.

2. Fork and clone the conda-rattler-solver repository to your preferred location:

   git clone "git@github.com:$YOUR_USERNAME/conda-rattler-solver" "$REPO_LOCATION"
   cd "$REPO_LOCATION"
   

3. Install the required dependencies for conda-rattler-solver:

   CONDA_CHANNEL_PRIORITY=flexible conda install \
     --file "$REPO_LOCATION"/dev/requirements.txt \
     --file "$REPO_LOCATION"/tests/requirements.txt
   

4. Install conda-rattler-solver in editable mode:

   cd "$REPO_LOCATION"
   python -m pip install --no-deps -e .
   

Debugging

Once you have followed the steps above you may need to investigate the state at a particular point. Insert a breakpoint() within the code and run a test or conda directly to hit the breakpoint.

Most of the solving logic is in Rust code, inside the rattler and resolvo libraries. The Python side of this plugin is responsible for translating conda’s view of the request (installed packages, history, pins, and channel configuration) into py-rattler objects, running the solve, and translating the result back. If you suspect a bug in the solver itself, we suggest investigating against the rattler repository.