Python version limited requirement in universal locking

[davidfritzsche]

With uv 0.2.18 the following fails

$ echo 'uv;python_version>="3.8"' | uv pip compile -p 3.7 --universal -
warning: uv is only compatible with Python 3.8+, found Python 3.7.17.
  × No solution found when resolving dependencies:

Which on one hand makes sense, as there really is no version of uv supporting Python 3.7. On the other hand, uv==X; python_version>="3.8" could be installed under Python 3.7 for any version X.

In non-universal locking, uv;python_version>="3.8" can be compiled for Python 3.7:

$ echo 'uv;python_version>="3.8"' | uv pip compile -p 3.7 -
warning: uv is only compatible with Python 3.8+, found Python 3.7.17.
Resolved 0 packages in 0.77ms
# This file was autogenerated by uv via the following command:
#    uv pip compile -p 3.7 -

[charliermarsh]

Oh thank you, I think this is the same as #4669 -- sorry, I missed that you created this.

[charliermarsh]

I'm going to merge into #4669. Good bug!

[charliermarsh]

Re-opening because this case is a little different than the NumPy one and they'll have separate PRs.

[houbie]

I'm facing a related issue: using python 3.9, uv pip install pdm correctly installs pdm version 2.17.3, but echo 'pdm==2.17.3'| uv pip compile --universal - fails:

(px-uvdir) % echo 'pdm==2.17.3'| uv pip compile --universal -
  × No solution found when resolving dependencies:
  ╰─▶ Because only the following versions of truststore{python_version >= '3.10'} are available:
          truststore{python_version >= '3.10'}==0.1.0
          truststore{python_version >= '3.10'}==0.2.0
          truststore{python_version >= '3.10'}==0.3.0
          truststore{python_version >= '3.10'}==0.4.0
          truststore{python_version >= '3.10'}==0.5.0
          truststore{python_version >= '3.10'}==0.6.0
          truststore{python_version >= '3.10'}==0.6.1
          truststore{python_version >= '3.10'}==0.7.0
          truststore{python_version >= '3.10'}==0.8.0
          truststore{python_version >= '3.10'}==0.9.0
          truststore{python_version >= '3.10'}==0.9.1
      and the requested Python version (>=3.9.17) does not satisfy Python>=3.10, we can conclude that any of:
          truststore{python_version >= '3.10'}<0.9.0
          truststore{python_version >= '3.10'}>0.9.0
       are incompatible.
      And because truststore{python_version >= '3.10'}==0.9.0 was yanked (reason: Doesn't work for CPython 3.13, upgrade to 0.9.1), we can conclude that all versions of truststore{python_version >= '3.10'} are incompatible.
      And because pdm==2.17.3 depends on truststore{python_version >= '3.10'} and you require pdm==2.17.3, we can conclude that the requirements are unsatisfiable.

pdm has a dependency on truststore Requires-Dist: truststore; python_version >= "3.10"
uv pip compile should retain this dependency with the python constraint instead of failing on it

[charliermarsh]

@houbie -- Think that's the same as #4668.

[charliermarsh]

On main there's now a workaround for this:

[project]
name = "foo"
version = "0.1.0"
description = "Add your description here"
readme = "README.md"
requires-python = ">=3.7"
dependencies = ["uv ; python_version >= '3.8'"]

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

[tool.uv]
environments = ["python_version < '3.8'", "python_version >= '3.8'"]

[DanCardin]

I'm encountering something similar, where the environments workaround doesn't appear to work. It works with only the "numpy >= 2.0; python_version >= '3.9'" dependency, but not with certain other dependencies in combination.

[project]
name = "foo"
version = "0.0.0"
description = ""
requires-python = ">=3.8,<4"
dependencies = [
  "pydantic",
  "numpy >= 2.0; python_version >= '3.9'",
]

[tool.uv]
environments = ["python_version < '3.9'", "python_version >= '3.9'"]

This ^ will attempt to install numpy (which then fails for me), which i think it shouldnt even be attempting.

I'm not sure what significance pydantic has here, it only depends on pydantic-core, typing-extensions and annotated-types. And explicitly stating those dependencies (locked to the locked versions) individually instead doesn't fail.

[charliermarsh]

Can you clarify what the issue is? It's attempting to install NumPy on which Python versions? (I can uv lock that successfully.)

[DanCardin]

oh right sorry! uv sync -p 3.8 fails trying to install numpy (when it shouldn't be), uv sync -p 3.9 succeeds and does.

admittedly the failing part is because it's trying to build the wheel and i dont have the requirements necessary, but the point is that it's installing numpy when it shouldn't anyway.

[mkniewallner]

On main there's now a workaround for this:

[project]
name = "foo"
version = "0.1.0"
description = "Add your description here"
readme = "README.md"
requires-python = ">=3.7"
dependencies = ["uv ; python_version >= '3.8'"]

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

[tool.uv]
environments = ["python_version < '3.8'", "python_version >= '3.8'"]

It looks like the workaround suggested here does not work anymore on 0.4.1, with the changes made in #6841:

$ uv lock
error: Environment markers `python_full_version >= '3.8'` don't overlap with Python requirement `>=3.7`

Since the workaround was suggested a few weeks ago, is there maybe a better way to work around the issue now?

[charliermarsh]

If that's not working, it looks like a bug. Let me take a look.

[charliermarsh]

Apologies @mkniewallner, really dumb bug: #6902

[DanCardin]

Fwiw, if anyone else runs into #4668 (comment) (because it was blocking by ability to adopt uv), a newer release of uv (0.4.20), given the referenced pyproject.toml, produces

  × No solution found when resolving dependencies for split (python_full_version >= '3.9'):
  ╰─▶ Because the requested Python version (>=3.8, <4) does not satisfy Python>=3.10 and the requested Python version (>=3.8, <4) does not satisfy
      Python>=3.9,<3.10, we can conclude that Python>=3.9 is incompatible.
      And because only the following versions of numpy{python_full_version >= '3.9'} are available:
          numpy{python_full_version >= '3.9'}<=2.0.2
          numpy{python_full_version >= '3.9'}>=2.1.0
      we can conclude that all of:
          numpy{python_full_version >= '3.9'}>=2.0.0,<2.1.0
          numpy{python_full_version >= '3.9'}>2.1.0,<2.1.1
          numpy{python_full_version >= '3.9'}>2.1.1,<2.1.2
          numpy{python_full_version >= '3.9'}>2.1.2
       are incompatible.
      And because the requested Python version (>=3.8, <4) does not satisfy Python>=3.10 and your project depends on numpy{python_full_version >=
      '3.9'}>=2.0, we can conclude that your project's requirements are unsatisfiable.

I took that to imply that I need to match the upper bound with the upper bound of my project: environments = ["python_version < '3.9'", "python_version >= '3.9' and python_version < '4'"], and that solved the issue, at least in the newer version of uv.

[charliermarsh]

This should "just work" without any workarounds in the next version (see: #8628).

[potiuk]

This should "just work" without any workarounds in the next version (see: #8628).

Cool . We are getting way closer now with uv being the default tool for Airflow development.