Why am I ditching Poetry?

TL;DR; switch to Hatch+uv, pip is the best.

I like writing code. I don't like to relearn package managers, neither do I want to make a hobby out of learning a new one every year.

This is a short rant on why I am starting all my new Python projects with Hatch 🐣 instead of Poetry.

why poetry#

I started using Poetry when it was around v0.4.0. It was the first package manager which I started using after pip and it had some awesome benefits:

• virtual environment handling
• lock file

(The CLI was never intuitive to me.)

reasons to switch#

environment isolation#

The venv switching was never seamless. Countless times I ended up debugging a virtual environment. Time after time I need to manually destroy Poetry's venvs and completely rebuild the environment.

speed#

I work daily with a large project (600+ dependencies), let's call the project Apache Airflow.

Building the base Airflow image, with constraints file respected it takes somewhere around 600 sec to build it. That is a lot of time.

🎄 airflow ❯ poetry lock
Updating dependencies
Resolving dependencies... (71.3s)

Writing lock file

ease of use#

I spent my fair share of time explaining Poetry. Maybe it was a mistake moving away from low-tech pip and virtualenv.

the alternative#

speed#

I start all my new projects with Hatch and uv build backend. The comparison of the two tools/workflows are quite unfair, I am ditching the lock file as well.

However, I can build Apache Airflow in 5 sec range. In the below case, it's a 30x speed up.

🎄 airflow ❯ uv pip install --constraint constraints.txt -r requirements.txt
Resolved 203 packages in 1.96s

environment isolation#

I can isolate my docs, tests and other environments instead of relying on dependency groups.

ease of use#

To me, the simplicity of Hatch and the speed of uv are finally making it worthwhile to switch. After 6 years of Poetry, I have no regrets, switching it.

I don't like going back to projects, where poetry is used, but there is no value in switching if your dependencies resolve fast enough and CD pipelines perform reasonably.

the future#

pip is king. Luckily uv support most of the commands, so it makes it extremely easy to switch to uv pip If you are new to Python packaging, library development - get familiar with pip commands first and then the tools on top of it. The packaging user guide is really well written.

I will consider switching from hatch if there will be a cargo like tool for python.