Closed mmmarinho closed 9 months ago
This was a problem that started happening somewhere between https://pypi.org/project/dqrobotics/21.4.0a75/#files and https://pypi.org/project/dqrobotics/20.4.0.79/#files so it's an old issue that never surfaced (or nobody complained about).
It can be tracked to other repositories such as https://github.com/pypa/setuptools/issues/2520
Building a truly universal2
wheel is something that can be considered, but at this stage it's better to change the naming of the wheels being built by the Github-hosted runners to represent what is actually being built, i.e. x86_x64
, in particular because we already have a self-hosted machine building the arm64
wheels.
This is to track recent issues that can be traced to the newly built
universal2
wheels. I am not sure when those begun to be compiled by thex86_x64
GithubActions machines, but they are not in factuniversal2
and that is causing problems.To replicate this issue, in a M2 MacbookPro running
Ventura 13.6.3
, we simply install the latest version of the library. It will, as shown below, installmacosx_10_15_universal2.whl
. I suppose that the versioning issue of not having a10_15
wheel available inarm64
caused the installation to default to theuniversal2
wheel.That would be fine as long as it worked, but the output of the simplest of scripts, e.g.
is an error, such as,
The wheel is being built
x86_64
but self entitling asuniversal2
and messing up thepip
installation. The simplest solution is forcing a rename to the wheels that are output byx86_64
GithubAction runners. However, at this stage, I do not know why the wheels are being built asuniversal2
, so such a naive solution might cause more problems.