Help detecting dependencies in pyproject.toml that are only specified as conda-dependencies

[Hofer-Julian]

Problem description

Problem description

Let's assume the following script:

# __init__.py
import numpy as np
x = np.array([1,2,3])

and the following pyproject.toml

[project]
dependencies = []
name = "pypi-test"
requires-python = ">= 3.11"
version = "0.1.0"

[build-system]
build-backend = "hatchling.build"
requires = ["hatchling"]

[tool.pixi.project]
channels = ["conda-forge"]
platforms = ["linux-64"]

[tool.pixi.pypi-dependencies]
pypi_test = { path = ".", editable = true }

[tool.pixi.dependencies]
numpy = ">=2.2.2,<3"

This works fine, when used with pixi, but when you publish it on PyPI, people will not be able to use it since numpy isn't added as dependency.

Suggested solutions

We already know which conda packages correspond to PyPI packages.
Therefore, we also know the list of packages that are missing in project.dependencies

We could:

• warn, but only if pyproject.toml is both the package and the manifest
• create pixi lint and add that as a lint rule
• let that handle by an external lint tool

References

Discord discussion

[ruben-arts]

I think if we implement this #532 it could be solved aswell.

[maurosilber]

I think if we implement this #532 it could be solved aswell.

I think that would require the inverse PyPI to conda mapping, while this works with the conda to PyPI mapping.

[theavey]

Also related to #2411 I think.

Would really like to not have to list (conda) dependencies in a bunch of places (recipe.yaml then twice in pyproject.toml), but would also like to stay PEP 621 compliant for ease of reading by other developers and whatever build backend I choose.