ResolutionImpossible error for Numpy

[sr-murthy]

• [Y ] I have searched the issue tracker and believe that this is not a duplicate.

Steps to reproduce

The PDM command, which is run in a Py 3.10.1 Docker container, is to install all project dependencies (all project source files, including pyproject.toml, are present in the container working directory, except the lock file which PDM creates from scratch):

pdm install --verbose --project . --dev --no-editable --no-self --no-isolation

The default deps (in pyproject.toml) include Numpy (unpinned).

Actual behavior

This used to work fine until yesterday, today I'm not sure what's wrong. I keep geting the following error:

#16 138.3 resolvelib.resolvers.ResolutionImpossible: [RequirementInformation(requirement=PythonRequirement(name='python', marker=None, extras=None, specifier=<PySpecSet >=3.10.1>, editable=False, prerelease=False), parent=None), RequirementInformation(requirement=PythonRequirement(name='python', marker=None, extras=None, specifier=<PySpecSet >=3.9,<3.13>, editable=False, prerelease=False), parent=<Candidate numpy@1.26.1>)]

I've tried restricting Numpy to earlier versions, and using even using later versions of Python for the Docker image. Same result.

Expected behavior

Successful installation of all project deps.

Environment Information

This is a Docker environment:

#16 0.865 python.use_venv is on, creating a virtualenv for this project...
#16 2.278 Virtualenv is created successfully at /<project name>/.venv
#16 2.328 PDM version:
#16 2.328   2.10.0
#16 2.328 Python Interpreter:
#16 2.328   /<project name>/.venv/bin/python (3.10)
#16 2.328 Project Root:
#16 2.328   /<project name>
#16 2.329 Local Packages:
#16 2.329   
#16 3.212 {
#16 3.212   "implementation_name": "cpython",
#16 3.212   "implementation_version": "3.10.13",
#16 3.212   "os_name": "posix",
#16 3.212   "platform_machine": "x86_64",
#16 3.212   "platform_release": "5.4.0-162-generic",
#16 3.212   "platform_system": "Linux",
#16 3.212   "platform_version": "#179-Ubuntu SMP Mon Aug 14 08:51:31 UTC 2023",
#16 3.212   "python_full_version": "3.10.13",
#16 3.212   "platform_python_implementation": "CPython",
#16 3.212   "python_version": "3.10",
#16 3.212   "sys_platform": "linux"
#16 3.212 }

I've tried to run the same job number of times - it fails with the same error.

[pawamoy]

It's because the latest Numpy version added <3.13 to its requires-python metadata, so your accepted range is no longer a subset of what your dependencies support. You have to add the same upper bound to your metadata.

• Numpy 1.25.2: https://github.com/numpy/numpy/blob/v1.25.2/setup.py#L523
• Numpy 1.26.0: https://github.com/numpy/numpy/blob/v1.26.0/pyproject.toml#L29 (though setup.py still has Python >= 3.9, not sure which one of pyproject.toml or setup.py is now used to build distributions)

[sr-murthy]

@pawamoy Many thanks. My requires-python was just >=3.9.6, but I've amended this in my TOML. The install works now.

A more general point: I guess this kind of thing can happen very easily, although I haven't come across a case where a dep specifies an upper bound on a Python beta version.

[sr-murthy]

I should have read the docs first 😊

https://pdm.fming.dev/latest/usage/project/#set-requires-python-value