Incorrect wheel dependencies when optional packages are combined with platform markers

[anaoum]

• OS version and name: Ubuntu 20.04.2 LTS
• Poetry version: 1.2.0
• Link of a Gist with the contents of your pyproject.toml file: https://gist.github.com/anaoum/1dc485b52bc33bff4e729ee9a3959d50

Issue

The above pyproject.toml has two versions of torch as a dependency, one for platform==linux, and one for platform==darwin. Both are marked as optional=true, and appear in the "mle" and "all" extras groups. When I generate the wheel package, the following METADATA is produced:

Metadata-Version: 2.1
Name: platras
Version: 0.1.0
Summary: 
Author: anaoum
Requires-Python: >=3.8,<3.9
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.8
Provides-Extra: all
Provides-Extra: mle
Requires-Dist: loguru (>=0.6.0,<0.7.0)
Requires-Dist: torch @ https://download.pytorch.org/whl/cpu/torch-1.12.1-cp38-none-macosx_10_9_x86_64.whl ; sys_platform == "darwin"
Requires-Dist: torch @ https://download.pytorch.org/whl/cu113/torch-1.12.1%2Bcu113-cp38-cp38-linux_x86_64.whl ; (sys_platform == "linux") and (extra == "mle" or extra == "all")

Note that the darwin version of the torch dependency does not also condition on (extra == "mle" or extra == "all")

This can be easily reproduced using the linked pyproject.toml above and running poetry build.

This issue causes problems in downstream projects that attempt to install the package with extras, as poetry will see two incompatible versions of the same dependency and fail.

[dimbleby]

I think the fix should be as straightforward as removing this line if you - or anyone else - care to put together an MR.

Writing a testcase will be harder than making the fix

[anaoum]

Thanks for the direction @dimbleby. I have confirmed that this resolves the issue, and implemented the fix/unit test in python-poetry/poetry-core#462

[github-actions]

This issue has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.