Contributing to cuDF

July 6, 2026 ยท View on GitHub

Contributions to cuDF fall into the following categories:

  1. To report a bug, request a new feature, or report a problem with documentation, please file an issue describing the problem or new feature in detail. The RAPIDS team evaluates and triages issues, and schedules them for a release. If you believe the issue needs priority attention, please comment on the issue to notify the team.
  2. To propose and implement a new feature, please file a new feature request issue. Describe the intended feature and discuss the design and implementation with the team and community. Once the team agrees that the plan looks good, go ahead and implement it, using the code contributions guide below.
  3. To implement a feature or bug fix for an existing issue, please follow the code contributions guide below. If you need more context on a particular issue, please ask in a comment.

As contributors and maintainers to this project, you are expected to abide by cuDF's code of conduct. More information can be found at: Contributor Code of Conduct.

Documentation contributions

To make a contribution only to cuDF documentation:

  1. Create the cudf_dev conda environment following the steps in Building with a conda environment.
  2. Activate the environment and install a recent version of cudf
conda activate cudf_dev
conda install cudf -c rapidsai-nightly -c conda-forge
  1. Build and view the docs locally following the instructions in the Building documentation docs
  2. Follow steps 7-10 in the section Your first issue

Code contributions

Your first issue

  1. Follow the guide at the bottom of this page for Setting up your build environment.
  2. Find an issue to work on. The best way is to look for the good first issue or help wanted labels.
  3. Comment on the issue stating that you are going to work on it.
  4. Create a fork of the cudf repository and check out a branch with a name that describes your planned work. For example, fix-documentation.
  5. Write code to address the issue or implement the feature.
  6. Add unit tests and unit benchmarks.
  7. Create your pull request. To run continuous integration (CI) tests without requesting review, open a draft pull request.
  8. Verify that CI passes all status checks. Fix if needed.
  9. Wait for other developers to review your code and update code as needed. Changes to libcudf C++ files require at least 2 approvals from the cudf-cpp-codeowners before merging. Changes limited to libcudf_streaming C++ files require at least 1 approval from the rapidsmpf-cpp-codeowners before merging.
  10. Once reviewed and approved, a RAPIDS developer will merge your pull request.

If you are unsure about anything, don't hesitate to comment on issues and ask for clarification!

Seasoned developers

Once you have gotten your feet wet and are more comfortable with the code, you can look at the prioritized issues for our next release in our project boards.

Note: Always look at the release board that is currently under development for issues to work on. This is where RAPIDS developers also focus their efforts.

Look at the unassigned issues, and find an issue to which you are comfortable contributing. Start with Step 3 above, commenting on the issue to let others know you are working on it. If you have any questions related to the implementation of the issue, ask them in the issue instead of the PR.

Setting up your build environment

The following instructions are for developers and contributors to cuDF development. These instructions are tested on Ubuntu Linux LTS releases. Use these instructions to build cuDF from source and contribute to its development. Other operating systems may be compatible, but are not currently tested.

Building cudf with the provided conda environment is recommended for users who wish to enable all library features. The following instructions are for building with a conda environment. Dependencies for a minimal build of libcudf without using conda are also listed below.

General requirements

Compilers:

  • gcc version 13.3+
  • nvcc version 12.9+
  • cmake version 3.29.6+

CUDA/GPU Runtime:

You can obtain CUDA from https://developer.nvidia.com/cuda-downloads.

Create the build environment

  • Clone the repository:
CUDF_HOME=$(pwd)/cudf
git clone https://github.com/rapidsai/cudf.git $CUDF_HOME
cd $CUDF_HOME

Building with a conda environment

Note: Using a conda environment is the easiest way to satisfy the library's dependencies. Instructions for a minimal build environment without conda are included below.

  • Create the conda development environment:
# create the conda environment (assuming in base `cudf` directory)
conda env create --name cudf_dev --file conda/environments/all_cuda-133_arch-$(uname -m).yaml
# activate the environment
conda activate cudf_dev
  • Note: the conda environment files are updated frequently, so the development environment may also need to be updated if dependency versions or pinnings are changed.

Building without a conda environment

  • libcudf has the following minimal dependencies (in addition to those listed in the General requirements). The packages listed below use Ubuntu package names:

    • build-essential
    • libssl-dev
    • libz-dev
    • libpython3-dev (required if building cudf)

Build cuDF from source

  • A build.sh script is provided in $CUDF_HOME. Running the script without explicit targets builds and installs the default targets: libcudf, pylibcudf, cudf, libcudf_streaming, cudf_streaming, cudf_polars, and dask_cudf. By default, C++ libraries are installed to the $CONDA_PREFIX directory. To install them into a different location, set $INSTALL_PREFIX. Python packages are installed into the active Python environment. Finally, note that the script depends on the nvcc executable being on your path, or defined in $CUDACXX.
cd $CUDF_HOME

# Choose one of the following commands, depending on whether
# you want to build and install the libcudf C++ library only,
# or include Python libraries:

./build.sh  # libcudf, pylibcudf, cudf, libcudf_streaming, cudf_streaming, cudf_polars, and dask_cudf
./build.sh libcudf  # libcudf only
./build.sh libcudf cudf  # libcudf and cudf only
  • Other libraries like cudf-kafka and custreamz can be installed with this script. For the complete list of libraries as well as details about the script usage, run the help command:
./build.sh --help

Building for development

To build Python packages for development purposes, add the --pydevelop flag. To build C++ tests, you can also request that build.sh build the tests target. To build all libraries and tests, with Python packages in development mode, simply run

./build.sh --pydevelop tests libcudf libcudf_kafka libcudf_streaming pylibcudf cudf cudf_streaming cudf_polars dask_cudf cudf_kafka custreamz
  • Note: if Cython files (*.pyx or *.pxd) have changed, the Python build must be rerun.

To run the C++ tests, run

ctest --test-dir ${CUDF_HOME}/cpp/build  # libcudf
ctest --test-dir ${CUDF_HOME}/cpp/libcudf_kafka/build  # libcudf_kafka
ctest --test-dir ${CUDF_HOME}/cpp/libcudf_streaming/build  # libcudf_streaming

To run python tests, run

  • To run cudf tests:
cd $CUDF_HOME/python
pytest -v ${CUDF_HOME}/python/cudf/cudf/tests
pytest -v ${CUDF_HOME}/python/dask_cudf/dask_cudf/ # There are tests in both tests/ and io/tests/
pytest -v ${CUDF_HOME}/python/cudf_streaming/cudf_streaming/tests
pytest -v ${CUDF_HOME}/python/custreamz/custreamz/tests

cudf (Java):

  • First, build the libcudf C++ library following the steps above

  • Then, refer to the Java README

Done! You are ready to develop for the cuDF project. Please review the project's code formatting guidelines.

Debugging cuDF

Building in debug mode from source

Follow the instructions to build from source and add -g to the ./build.sh command.

For example:

./build.sh libcudf -g

This builds libcudf in debug mode which enables some assert safety checks and includes symbols in the library for debugging.

All other steps for installing libcudf into your environment are the same.

Debugging with cuda-gdb and cuda-memcheck

When you have a debug build of libcudf installed, debugging with the cuda-gdb and cuda-memcheck is easy.

If you are debugging a Python script, run the following:

cuda-gdb -ex r --args python <program_name>.py <program_arguments>

compute-sanitizer --tool memcheck python <program_name>.py <program_arguments>

Device debug symbols

The device debug symbols are not automatically added with the cmake Debug build type because it causes a runtime delay of several minutes when loading the libcudf.so library.

Therefore, it is recommended to add device debug symbols only to specific files by setting the -G compile option locally in your cpp/CMakeLists.txt for that file. Here is an example of adding the -G option to the compile command for src/copying/copy.cu source file:

set_source_files_properties(src/copying/copy.cu PROPERTIES COMPILE_OPTIONS "-G")

This will add the device debug symbols for this object file in libcudf.so. You can then use cuda-dbg to debug into the kernels in that source file.

Code Formatting

Using pre-commit hooks

cuDF uses pre-commit to execute all code linters and formatters. These tools ensure a consistent code format throughout the project. Using pre-commit ensures that linter versions and options are aligned for all developers. Additionally, there is a CI check in place to enforce that committed code follows our standards.

To use pre-commit, install via conda or pip:

conda install -c conda-forge pre-commit

pip install pre-commit

Then run pre-commit hooks before committing code:

pre-commit run

By default, pre-commit runs on staged files (only changes and additions that will be committed). To run pre-commit checks on all files, execute:

pre-commit run --all-files

Optionally, you may set up the pre-commit hooks to run automatically when you make a git commit. This can be done by running:

pre-commit install

Now code linters and formatters will be run each time you commit changes.

You can skip these checks with git commit --no-verify or with the short version git commit -n.

Summary of pre-commit hooks

The following section describes some of the core pre-commit hooks used by the repository. See .pre-commit-config.yaml for a full list.

C++/CUDA is formatted with clang-format.

doxygen is used as documentation generator and also as a documentation linter. In order to run doxygen as a linter on C++/CUDA code, run

./ci/checks/doxygen.sh

Python code runs several linters including Ruff with its various rules for Black-like formatting or Isort.

cuDF also uses codespell to find spelling mistakes, and this check is run as a pre-commit hook. To apply the suggested spelling fixes, you can run codespell -i 3 -w . from the repository root directory. This will bring up an interactive prompt to select which spelling fixes to apply.

Developer Guidelines

The C++ Developer Guide includes details on contributing to libcudf C++ code.

The Python Developer Guide includes details on contributing to cuDF Python code.

Attribution

Portions adopted from https://github.com/pytorch/pytorch/blob/master/CONTRIBUTING.md Portions adopted from https://github.com/dask/dask/blob/master/docs/source/develop.rst