Pin setuptools version to fix installation with pip

[piiq]

Description

This PR pins the setuptools version to the latest version that's used in the project's CI.
Resolves #1950 that was introduced due to pypa/setuptools#3693

Type of change

• New feature (backwards compatible)
• New feature (breaking API changes)
• Bug fix
• Other (Documentation, CI, ...)

Contribution checklist

• [ ] Add our license to new files.
• Check that your code adheres to our coding style.
• [ ] Write unittests.
• Run the unittests and check that they’re passing.
• Run the benchmarks to make sure your change doesn’t introduce a regression.

Test procedure:

• Install package from source with python setup.py develop && pip install .
• Run unit tests with cd cvxpy/tests && pytest
• Run benchmarks with ``

Test results:

• Unit tests: 852 passed, 289 skipped, 543 warnings in 41.25s
• Benchmark: 9 passed, 3 skipped, 1 warning in 6.43s

[CLAassistant]

All committers have signed the CLA.

[piiq]

@phschiele @SteveDiamond I would appreciate a review and let me know if there is anything I can do to have this patch released to pypi

[SteveDiamond]

@piiq can you clarify more what happened? I saw your link, but is this bug in setuptools going to be fixed soon, or is this a permanent problem due to deprecation?

[github-actions]

Benchmarks that have stayed the same:

   before           after         ratio
 [475c2d43]       [fdf67278]
      25.8±0s          26.3±0s     1.02  cvar_benchmark.CVaRBenchmark.time_compile_problem
      18.6±0s          18.8±0s     1.01  qp_1611_benchmark.QP1611Benchmark.time_compile_problem
      8.31±0s          8.37±0s     1.01  huber_regression.HuberRegression.time_compile_problem
      8.01±0s          8.05±0s     1.00  semidefinite_programming.SemidefiniteProgramming.time_compile_problem
      8.46±0s          8.49±0s     1.00  optimal_advertising.OptimalAdvertising.time_compile_problem
      3.74±0s          3.74±0s     1.00  simple_LP_benchmarks.SimpleFullyParametrizedLPBenchmark.time_compile_problem
      40.0±0s          39.6±0s     0.99  sdp_segfault_1132_benchmark.SDPSegfault1132Benchmark.time_compile_problem
      21.1±0s          20.6±0s     0.98  simple_LP_benchmarks.SimpleLPBenchmark.time_compile_problem
      9.88±0s          9.61±0s     0.97  svm_l1_regularization.SVMWithL1Regularization.time_compile_problem
      1.65±0s          1.58±0s     0.96  simple_LP_benchmarks.SimpleScalarParametrizedLPBenchmark.time_compile_problem

[piiq]

Thanks for the reply @SteveDiamond

If i understand correctly it's a permanent problem due to deprecation on the setuptools side that the numpy maintainers are not planning to adjust to. From this comment in a numpy issue thread I understood that numpy does not and will not support setuptools>60, but all versions up to the 65.5.0 were ok.

For cvxpy the problem is that it breaks the build process of SCS when installing with pip on a platform for which wheels with pre-built SCS binaries are not provided.

I've explicitly pinned the versions in pyproject.toml, setup.py and setup.cfg to be lower than the highest version that's installed onto the build system in continuous_integration/install_dependencies.sh. This solves the failing build issue because this pinned setuptools version is lower than 65.5.0.

For a test that the patch works as expected on an Apple Silicon machine I run pip install cvxpy or pip install git+https://github.com/cvxpy/cvxpy.git@master and see the SCS build fail and I run pip install git+https://github.com/piiq/cvxpy.git@bugfix/pin-setuptools and see that the build succeeds

[bodono]

By the way, I accepted this PR that makes a similar fix to SCS a few days ago.