search

astral-sh / uv

An extremely fast Python package and project manager, written in Rust.
https://docs.astral.sh/uv
Apache License 2.0
26.34k stars 766 forks source link

Add a `--universal-strategy`-like setting to the resolver #7190

Open charliermarsh opened 2 months ago

charliermarsh commented 2 months ago

In the universal resolver, we prioritize solving for as few versions as possible. Sometimes, users instead want to get the newest possible version for every fork (e.g., they'd prefer getting the newest versions of foo for Python 3.10+, even if it means they have two versions of foo in the lockfile, one for Python 3.9 and earlier, and one for Python 3.10 and later).

(--universal-strategy is obviously a bad name.)

One implementation of this would be to ignore preferences from other forks.

henryiii commented 2 weeks ago

I'd even think this might be a nice default. Currently, the following simple example:

$ mkdir example
$ cd example
$ cat > pyproject.toml << EOL
[project]
name = "example"
version = "0.1"
requires-python = ">=3.8"
dependencies = ["numpy"]
EOL
$ uv run python
Resolved 2 packages in 7ms
error: Failed to prepare distributions
  Caused by: Failed to download and build `numpy==1.24.4`
  Caused by: Build backend failed to determine requirements with `build_wheel()` (exit status: 1)

[stderr]
Traceback (most recent call last):
  File "<string>", line 8, in <module>
  File "/Users/henryschreiner/.cache/uv/builds-v0/.tmpjYjj2d/lib/python3.12/site-packages/setuptools/__init__.py", line 10, in <module>
    import distutils.core
ModuleNotFoundError: No module named 'distutils'
  Caused by: distutils was removed from the standard library in Python 3.12. Consider adding a constraint (like `numpy >1.24.4`) to avoid building a version of numpy that depends on distutils.

will break on Python 3.12 and newer. Any package that supports a narrower range than you do (and packages should be allowed to stop supporting old Python versions) will cause you do never get updates, and updates are often required for new Python versions.

This is one of the biggest drawbacks of a universal solver, as most users starting out don't realize that "dependencies = ["numpy"]` isn't going to give them the latest version on the version of Python they are running. I think this would really reduce the friction when starting out.

See https://github.com/astral-sh/uv/issues/8492 also.