We have issues labeled as Good First Issue and Help Wanted which are good opportunities for new contributors.
Rust, a C compiler, and CMake are required to build uv.
On Ubuntu and other Debian-based distributions, you can install the C compiler and CMake with:
sudo apt install build-essential cmake
You can install CMake with Homebrew:
brew install cmake
See the Python section for instructions on installing the Python versions.
You can install CMake from the installers or with pipx install cmake
.
For running tests, we recommend nextest.
If tests fail due to a mismatch in the JSON Schema, run: cargo dev generate-json-schema
.
Testing uv requires multiple specific Python versions; they can be installed with:
cargo run toolchain install
The storage directory can be configured with UV_TOOLCHAIN_DIR
.
You can invoke your development version of uv with cargo run -- <args>
. For example:
cargo run -- venv
cargo run -- pip install requests
When testing debug builds on Windows, the stack can overflow resulting in a STATUS_STACK_OVERFLOW
error code.
This is due to a small stack size limit on Windows that we encounter when running unoptimized builds — the release
builds do not have this problem. We added a UV_STACK_SIZE
variable to
bypass this problem during testing. We recommend bumping the stack size from the default of 1MB to 2MB, for example:
$Env:UV_STACK_SIZE = '2000000'
Source distributions can run arbitrary code on build and can make unwanted modifications to your system ("Someone's Been Messing With My Subnormals!" on Blogspot, "nvidia-pyindex" on PyPI), which can even occur when just resolving requirements. To prevent this, there's a Docker container you can run commands in:
docker buildx build -t uv-builder -f builder.dockerfile --load .
# Build for musl to avoid glibc errors, might not be required with your OS version
cargo build --target x86_64-unknown-linux-musl --profile profiling
docker run --rm -it -v $(pwd):/app uv-builder /app/target/x86_64-unknown-linux-musl/profiling/uv-dev resolve-many --cache-dir /app/cache-docker /app/scripts/popular_packages/pypi_10k_most_dependents.txt
We recommend using this container if you don't trust the dependency tree of the package(s) you are trying to resolve or install.
Please refer to Ruff's Profiling Guide, it applies to uv, too.
We provide diverse sets of requirements for testing and benchmarking the resolver in scripts/requirements
and for the installer in scripts/requirements/compiled
.
You can use scripts/bench
to benchmark predefined workloads between uv versions and with other tools, e.g.
python -m scripts.bench \
--uv-path ./target/release/before \
--uv-path ./target/release/after \
./scripts/requirements/jupyter.in --benchmark resolve-cold --min-runs 20
You can use tracing-durations-export to visualize parallel requests and find any spots where uv is CPU-bound. Example usage, with uv
and uv-dev
respectively:
RUST_LOG=uv=info TRACING_DURATIONS_FILE=target/traces/jupyter.ndjson cargo run --features tracing-durations-export --profile profiling -- pip compile scripts/requirements/jupyter.in
RUST_LOG=uv=info TRACING_DURATIONS_FILE=target/traces/jupyter.ndjson cargo run --features tracing-durations-export --bin uv-dev --profile profiling -- resolve jupyter
You can enable trace
level logging using the RUST_LOG
environment variable, i.e.
RUST_LOG=trace uv
Releases can only be performed by Astral team members.
Changelog entries and version bumps are automated. First, run:
./scripts/release.sh
Then, editorialize the CHANGELOG.md
file to ensure entries are consistently styled.
Then, open a pull request e.g. Bump version to ...
.
Binary builds will automatically be tested for the release.
After merging the pull request, run the release workflow
with the version tag. Do not include a leading v
.
The release will automatically be created on GitHub after everything else publishes.