Add preliminary support for Python 3.12a7.

[icemac]

No description provided.

[icemac]

o no, we have breaking tests on Python 3.12a7.

[icemac]

And of course the test failures do not happen locally neither on MacOS Intel nor on M2.

I have no idea how to solve this.
@dataflake Do you have any suggestion?

[dataflake]

The issues are described at zopefoundation/meta#181. Both zc.buildout and setuptools are at fault but there is no reaction to the issues I have filed.

The only suggestion I have would be to not install zope.proxy via setup_requires to get at the C header files, but to just copy them straight into zope.security. This has the drawback of them going out of sync with zope.proxy but I think it would get rid of the build issue. Or maybe there's a saner way to get at those header files during the build.

[d-maurer]

Michael Howitz wrote at 2023-4-17 23:06 -0700:

And of course the test failures do not happen locally neither on Apple Intel nor on M2.

Is it possible that the GH test uses the prebuild `zope.security` wheel rather than build the extensions anew from a `zope.security` checkout? In the output I see `Install zope.security 3.12.0-alpha7` which apparently extracts the wheel content while see `Install zope.security` step is preceeded by an icon which suggests "skipped". Should this be really the case, it could explain differences between a local test run and the GH run.

[d-maurer]

Jens Vagelpohl wrote at 2023-4-17 23:50 -0700:

The issues are described at zopefoundation/meta#181. Both `zc.buildout` and `setuptools` are at fault but there is no reaction to the issues I have filed.

The `test (3.12.0-alpha.7, ubuntu-20.04)` is also failing which likely is not Mac related and likely not due to `setuptools/buildout` problems.

[dataflake]

The test (3.12.0-alpha.7, ubuntu-20.04) is also failing which likely is not Mac related and likely not due to setuptools/buildout problems.

This is already understood. The issue I am talking about cropped up at zopefoundation/AccessControl#139. I think the way dependency header files are provided before the build step in some packages is the problem. Packages that have a copy of their dependencies' headers don't show the issue.

[icemac]

@dataflake wrote:

The only suggestion I have would be to not install zope.proxy via setup_requires to get at the C header files, but to just copy them straight into zope.security

I implemented this in #98. I am planning to release that changes as a dev version as some tests break but it should unblock us in AccessControl.

[icemac]

Superseded by #98.