Open mayeut opened 5 months ago
henryiii
commented
5 months ago While MACOSX_DEPLOYMENT_TARGET makes sense, the others might cause it to be harder to build wheels in the test step if wheels are missing from test dependencies. Not the best idea to build here, in general, but it may currently work because the correct variables are set for building.
mayeut
commented
5 months ago Maybe we should adapt them all then ?
If we're thinking that we might break some edge case that requires building from sources by removing any of those environment variable then IMHO we should update them all:
MACOSX_DEPLOYMENT_TARGET to the current running OS (this one is already done but was meant to be temporary, required by uv for now)_PYTHON_HOST_PLATFORM/ARCHFLAGS to the current running test architecture as it's more likely to fail an universal2 build than a specific one. In fact, that was the issue when building matplotlib from sources in https://github.com/pypa/cibuildwheel/pull/1856#issuecomment-2156380899, it might have failed later on of course.
joerick
commented
1 month ago I think removing these variables for the test step makes sense in general.
Probably the right approach would be to start again from os.environ rather than copying env here:
And equivalents for the other platforms.
MACOSX_DEPLOYMENT_TARGETto the current running OS (this one is already done but was meant to be temporary, required byuvfor now)
I don't follow this - normally on a user's machine MACOSX_DEPLOYMENT_TARGET isn't set and installing with uv works.
Description
macOS builds are using multiple environment variables specific to the build environment:
MACOSX_DEPLOYMENT_TARGET,_PYTHON_HOST_PLATFORM,ARCHFLAGS,SDKROOTThose are propagated to the test environments. They should probably be removed from those.
see https://github.com/pypa/cibuildwheel/pull/1856#issuecomment-2156493426 and the 2 following comments for more context.
Build log
No response
CI config
No response