Python version constraints could lead to breaking dependency resolution

[m-martin-j]

Versions

river version: ^0.20.0 Python version: > 3.13 Operating system: any

Describe the bug

Is there any specific reason, why the maximally allowed Python version is limited to 3.13 from 0.20.0 releases of river? https://github.com/online-ml/river/blob/7b2f1b867db7b107f9f5498ec550bfe9cb228220/pyproject.toml#L19 I guess that this breaks dependency resolution for a fair portion of projects out there specifying their Python version as e.g. ^3.11, which is quite common. At least, this contradicts what is noted in the readme. (There, we should also correct the minimally required version from 3.8 to 3.9 to correspond what is required by the build system.) https://github.com/online-ml/river/blob/7b2f1b867db7b107f9f5498ec550bfe9cb228220/README.md?plain=1#L106

Steps/code to reproduce

The problem primarily concerns dependency resolution.

[m-martin-j]

A follow-up: After researching more, I understand that this point is heavily opinionated.

The capping is also done in scipy (https://github.com/scipy/scipy/issues/17957). Numpy seemed to have resolved by removing the upper bound (https://github.com/numpy/numpy/pull/24962) after some discussions (https://github.com/numpy/numpy/issues/24810).

[MaxHalford]

Hey @m-martin-j. Indeed, I can't remove this we depend on SciPy. Thanks for providing context. Maybe we can leave this issue open until SciPy changes something on their side.