Meet pup, your new best friend

[liquidcarbon]

IMO, Pixi and uv belong together.

State of Pixi and UV as of now outdated

	pixi	uv	pup
installs python	✔	❌	✔
pypi support	almost ✔	✔	✔
conda support	✔	❌	✔
freeze	hacky	✔	✔
build	in progress	?	TBD

Discussion in Pixi: prefix-dev/pixi#1097

Good luck doing this without Pixi:

[liquidcarbon]

a blooper :)

[liquidcarbon]

TLDR

https://github.com/liquidcarbon/puppy is a transparent wrapper around pixi and uv, with simple APIs and recipes for using them to help write reproducible, future-proof scripts and notebooks.

From 0 to rich toolset in one command:

Start in an empty folder.

curl -fsSL "https://pup-py-fetch.hf.space?python=3.12&pixi=jupyter&env1=duckdb,pandas" | bash

installs python and dependencies, in complete isolation from any existing python on your system. Mix and match URL query params to specify python version, tools, and venvs to create.

The above also installs puppy's CLI (pup --help):

CLI - "uv-lite"

• pup add myenv pkg1 pkg2 (install packages to "myenv" folder using uv)
• pup list view what's installed across all projects
  - pup clone and pup sync clone and build external repos (must have buildable pyproject.toml files)

Pup as a Module - no more notebook kernels

The original motivation for writing puppy was to simplify handling kernels, but you might just not need them at all. Activate/create/modify "kernels" interactively with:

import pup
pup.fetch("myenv")  # "activate" - packages in "myenv" are now importable
pup.fetch("myenv", "pkg1", "pkg2")  # "install and activate" - equivalent to `pup add myenv pkg1 pkg2`  

Of course you're welcome to use !uv pip install, but after 10 times it's liable to get messy.

Target Audience

Loosely defining 2 personas:

1. Getting Started with Python (or herding folks who are):

   1. puppy is the easiest way to go from 0 to modern python - one-command installer that lets you specify python version, venvs to build, repos to clone - getting everyone from 0 to 1 in an easy and standardized way
   2. if you're confused about virtual environments and notebook kernels and install full jupyter into every project

2. Competent - check out Multi-Puppy-Verse and Where Pixi Shines sections:

   1. you have 10 work and hobby projects going at the same time and need a better way to organize them for packaging, deployment, or even to find stuff 6 months later
   2. you need support for conda and non-python stuff - you have many fast-moving external and internal dependencies - check out pup clone and pup sync workflows and dockerized examples

Filesystem is your friend

Puppy recommends a sensible folder structure where each outer folder houses one and only one python executable - in isolation from each other and any other python on your system. Pup is tied to a python executable that is installed by Pixi, along with project-level tools like Jupyter, conda packages, and non-python tools (NodeJS, make, etc.) Puppy commands work the same from anywhere within this folder.

The inner folders are git-ready projects, defined by pyproject.toml, with project-specific packages handled by uv.

# ├── puphome/  # python 3.12 lives here
# │   ├── public-project/
# │   │   ├── .git  # this folder may be a git repo (see pup clone)
# │   │   ├── .venv
# │   │   └── pyproject.toml
# │   ├── env2/
# │   │   ├── .venv/  # this one is in pre-git development
# │   │   └── pyproject.toml
# │   ├── pixi.toml
# │   └── pup.py
# ├── pup311torch/  # python 3.11 here
# │   ├── env3/
# │   ├── env4/
# │   ├── pixi.toml
# │   └── pup.py
# └── pup313beta/  # 3.13 here
#     ├── env5/
#     ├── pixi.toml
#     └── pup.py

Puppy embraces "explicit is better than implicit" from the Zen of python; it logs what it's doing, with absolute paths, so that you always know where you are and how you got there.

PS I've benefited a great deal from the many people's OSS work - now trying to pay it forward. The ideas laid out in puppy's README and implementation have come together after many years of working in different orgs, where average "how do you rate yourself in python" ranged from zero (Excel 4ever) to highly sophisticated. The matter of "how do we build stuff" is kind of never settled, and this is my take.

Thanks for checking this out! Suggestions and feedback are welcome!

[epicslapper]

Never heard of pixi , but explorer showed me what the pixi command did ,
and i fully agree ,! " sensible folder structure where 1. each outer folder houses one and only one python executable - 2. in isolation from each other and any other python on your system. Pup is tied to a python executable" , and 3. puppy works the same from within this folder

[epicslapper]

Hi , this is fantastic .....

too bad the linux nerds don't understand it .... ;-)
( btw , i can just type : activate and deactivate in VSC , nice )