Support for uv

[t-kalinowski]

uv is a fast and efficient package management tool. We should consider adding support for it in reticulate.

One of its key advantages is that it allows for installing a specific Python version on demand in the same motion as setting up the virtual environment. This capability eliminates the need for users to first ensure a suitable Python venv starter is pre-installed (e.g., by running install_python()). It also avoids the need for reticulate to search for and locate suitable virtualenv_starter()s. This feature could streamline the user experience.

The ability to install a specific Python version on demand means that uv could serve as a drop-in replacement for many uses of conda. Importantly, it sidesteps the potential risk of binary incompatibilities that conda usage introduces.

[vincentarelbundock]

I would also be very interested in this feature. uv has taken the Python world by storm, and has become incredibly popular in a short amount of time.

I believe that the main stumbling is that reticulate does not detect the location of libpython.

In this minimal example, I install uv, create a new project, and then call reticulate::py_discover_config() in R. There are two different calls:

1. Rscript without preamble uses the default Python install on my machine.
2. Rscript with a uv run preamble uses the virtual environment and python install set up by uv.

Notice that we get [NOT FOUND] for libpython in the second case.

# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# Initialize a new project
uv init mwe
cd mwe

# Add some python dependencies
uv add pandas

default reticulate python config

Rscript -e "reticulate::py_discover_config()"

python:         /home/username/.virtualenvs/global/bin/python
libpython:      /home/username/.pyenv/versions/3.12.4/lib/libpython3.12.dylib
pythonhome:     /home/username/.virtualenvs/global:/home/username/.virtualenvs/global
virtualenv:     /home/username/.virtualenvs/global/bin/activate_this.py
version:        3.12.4 (main, Jun  9 2024, 13:30:56) [Clang 15.0.0 (clang-1500.3.9.4)]
numpy:          /home/username/.virtualenvs/global/lib/python3.12/site-packages/numpy
numpy_version:  2.0.2

uv reticulate python config

uv run Rscript -e "reticulate::py_discover_config()"

python:         /home/username/mwe/.venv/bin/python
libpython:      [NOT FOUND]
pythonhome:     /home/username/mwe/.venv:/home/username/mwe/.venv
virtualenv:     /home/username/mwe/.venv/bin/activate_this.py
version:        3.12.6 (main, Sep  9 2024, 21:36:32) [Clang 18.1.8 ]
numpy:          /home/username/mwe/.venv/lib/python3.12/site-packages/numpy
numpy_version:  2.1.2

[t-kalinowski]

@vincentarelbundock Thanks for reporting this! Reticulate should now work with a venv created by uv after #1678.

I'll keep this issue open since we'd still like to add convenient wrappers in reticulate around uv, similar to the existing conda_* and virtualenv_* functions.

[vincentarelbundock]

Thanks a lot for the quick fix @t-kalinowski . Looks like the dev version is working for me!