you can solve sudokus in python packaging. not python code, python packaging. [project].name = "sudoku".version = "1.0.0".dependencies = [. "sudoku_3_1 == 2",. "sudoku_5_7 == 6",. "sudoku_0_7 == 5". ].

RT @astral_sh: Announcing the Astral OSS Fund. We're giving > $3,000 per Astral team member per year to open source projects, maintainers,….

RT @charliermarsh: But our goal is to evolve uv into a "Cargo for Python": a comprehensive project and package manager that's fast, reliabl….

RT @astral_sh: Announcing uv: an extremely fast Python package installer and resolver, written in Rust. uv is designed as a drop-in replac….

i'll also be at #PyConUS if you're interested in chatting about maturin, pyo3, evolving packaging standards, python tooling in rust or just python packaging in general.

i've reimplemented PEP 440 (the python version standard) and wrote about the depths of PEP 440 and how to evolve it for the future

RT @althonos: Hi #bioinformatics! With the developers at #PyO3 (#Python to #Rust bindings), we are working on a paper manuscript to make it….

@indygreg incidentally, i've also written a wheel installer that's faster than pip's singlethreaded (e.g. 5.8s vs. 7.6s for plotly on my laptop) and can additionally be multithreaded, if somebody is interested in reusing that

@indygreg there's still a lot missing and many things are broken, e.g. non-linux is badly tested, only pypi is supported, startup is slow, only some requirements.txt syntax is supported, some mysterious failures in jupyter/colab, windows scripts are wonky, etc.

@indygreg as another proof of concept, it supports the flat src layout you probably know from other ecosystems

@indygreg loads libpython, injects a custom meta path finder that knows the dependency resolution and then starts your program as you would from the cli. this guarantees that only exactly the dependencies you requested in the versions resolved are importable.

how does it work? internally, it looks for pyproject.toml or requirements.txt, if required calls a vendored poetry to resolve dependencies, installs each packages in a separate directory, downloads missing python binaries (thanks to @indygreg's great python-build-standalone),.

packages are installed once globally, so no more reinstalling torch for each project, no more venv/conda management, and it also means if you have all dependencies cached you can do a fresh clone of a project and your code will start immediately.

There's either a single standalone binary to download ( or a pip installable package (`pipx install monotrail`), depending on how you want to run it.

i've also made an interactive mode for jupyter notebooks (`monotrail.interactive(numpy="^1.21", pandas="^1")`), a pipx clone (`monotrail ppipx --extras jupyter black .`) and a tox clone (`monotrail run -p 3.8 -p 3.9 -p 3.10 command pytest`)

you can specify your dependencies through pyproject.toml/poetry.lock or requirements.txt, monotrail will resolve them, install missing packages in a global cache and hook the correct versions up your project, including installing python itself.

i've made a proof of concept for using python packages without virtualenv, or really any explicit installing at all: no virtualenv/pip/pyenv/conda, just ´monotrail run python` and it will set everything up.

Thanks to Michael Heinzinger, @thesteinegger, Burkhard Rost and all of @rostlab for making this project possible.