🐛 conda-pupa 🦋

/ˈpjuː.pə/, plural /ˈpjuː.piː/

Convert PyPA package "caterpillers", whether they use setup.py or pyproject.toml, to beautiful .conda butterflies.

conda-pupa is handy for filling in a handful of PyPI exclusive dependencies in your conda environment. You can share the resulting packages with other conda users even if they are not using conda-pupa.

conda-pupa is the easiest way to build a Python project as a conda package without using conda-build or a separate recipe.

conda-pupa replaces pip install -e ., but conda treats the Python package the same as any other conda package, because it is a conda package, and it can use dependencies from conda. It is a Python-standards way to replace the simplistic conda-build develop; conda-pupa understands pyproject.toml as well as setup.py.

Vision

Install pypi projects into conda environments, converting them all to .conda on the fly before using conda to do the installation.

Recursively add missing pypi packages into the conda environment, in the same way that "pip wheel " populates a directory with .whl packages of it and all of its dependencies.

Control which packages are built from pypi and cached versus which come from pre-existing conda channels.

Replace conda-build develop with something compatible with all modern Python projects, more like pip install -e .

Use Python standards to build packages, handle metadata for both setup.py, pyproject.toml. Glue together libraries to process packages and their metadata.

Have 100% test coverage.

Installation

conda-pupa depends on conda. It provides a conda plugin for use without having to install into the active environment. To make it available to the main conda,

conda install -n base -c dholth conda-pupa

Alternatively, python -m conda_pupa.

Usage

conda-pupa works on environments that already contain python.

conda pupa -e .

Create and install a conda package linking a pyproject.toml or setup.py project at . into the Python environment, like pip install -e .

conda pupa --build .

Build a wheel for a Python project and convert to .conda.

conda pupa --prefix $CONDA_PREFIX twine==6.0.1

If a dependency twine==6.0.1, passed as a conda-format MatchSpec, is missing from $CONDA_PREFIX, then convert it and its dependencies from pypi wheels to conda packages. Collect the new .conda packages into a local channel.

Add --override-channels to convert all dependencies missing from --prefix from pypi and not just the ones missing from your default conda channels.

🐞 Note conda-pupa currently packages all wheels as noarch even if they should be arch-specific.

% conda pupa --help
Usage: conda pupa [OPTIONS] [PACKAGE_SPEC]...

Options:
  -c, --channel TEXT            Additional channel to search for packages.
  -O, --override-channels TEXT  Do not search default or .condarc channels.
                                Will search pypi.
  -e, --editable TEXT           Build named path as editable package; install
                                to link checkout to environment.
  -b, --build TEXT              Build named path as wheel converted to conda.
  -p, --prefix TEXT             Full path to environment location (i.e.
                                prefix).
  -n, --name TEXT               Name of environment.
  -h, --help                    Show this message and exit.

Bugs ====

conda-pupa does not have optimizations.
conda-pupa packages all packages as noarch even if they should be platform specific; this works okay for local single-platform use.
conda-pupa does not consider Python extras.
... and more!

Future Improvements

When available, convert pypi metadata files only to generate repodata.json during dependency discovery, like pip. Once a solution has been found, download and convert the necessary wheels.
Use conda's build hash to tag wheels that may convert to more than one conda package in case of Python markers.
Handle Python source distributions. Use conda to install the compiler too?
...