CMake version 3.26.4 is unavailable in pip

[karanshah]

Hello! CMake has released 3.26.4 version, but is not available via pip. Can this version be added to pip?

[jcfr]

For reference:

• Update to CMake 3.26.4 #365

[jcfr]

@henryiii @mayeut Since the CI run ¹ associated with #365 was ✔️

Any objection if tag v3.36.4 ?

Footnotes

1. https://github.com/scikit-build/cmake-python-distributions/actions/runs/5019944234 ↩

[henryiii]

I think there's an issue with the Windows ARM wheels. I think the CMake inside the wheel is actually the wrong version, and just happens to work since Windows ARM can run x86 code via emulation. I'd like to fix that if possible first.

2023-05-19T02:02:07.9979465Z   -- Downloading...
2023-05-19T02:02:07.9980442Z      dst='D:/a/cmake-python-distributions/cmake-python-distributions/_skbuild/win-amd64-3.9/cmake-build/cmake-3.26.4-windows-i386.zip'
2023-05-19T02:02:07.9997851Z      timeout='none'
2023-05-19T02:02:08.0001737Z      inactivity timeout='none'
2023-05-19T02:02:08.0003023Z   -- Using src='https://github.com/Kitware/CMake/releases/download/v3.26.4/cmake-3.26.4-windows-i386.zip'

[jcfr]

I gonna have a look this afternoon

[henryiii]

Okay, great, it's probably just a case of fixing the download script to grab the right one.

I've also been working on Windows ARM at wjakob/nanobind_example#27 and have been working on the horror that is scikit-build/scikit-build-core#355 to support it, since FindPython doesn't seem to support setting Python_SOABI as input, which is necessary for cross-compilation (and could save a subprocess call even if not cross compiling). Setuptools supports overriding via the environment variable SETUPTOOLS_EXT_SUFFIX.

But all of this won't affect cmake/ninja, since they are not making Python extensions.

I might try to propose something in the future.

[jcfr]

just a case of fixing the download script to grab the right one.

Ditto. Once the linked pull request completes, we should be able to tag v3.26.4

[jcfr]

3.26.4 has been published¹, wheels will be uploaded shortly.

Footnotes

1. https://github.com/scikit-build/cmake-python-distributions/releases/tag/3.26.4 ↩

[jcfr]

I thought one of the job was "stuck" as it has been running for ~3hrs ... but looking at the log of 3.26.3, the timing is consistent.

3.26.3 ✔️	3.6.24 ⏳

[jcfr]

And here is a break down of the timing (from 3.26.3 log¹):

The overhead of building through the emulator ...

Footnotes

1. https://github.com/scikit-build/cmake-python-distributions/actions/runs/4658536917/jobs/8244345614#step:4:93 ↩