CI: make cuda tests even run to begin with

Arch profile.d scripts were converted to use an appending function that
disappears when /etc/profile exits, and overall are simply not suitable
-- any more -- for sourcing individually.

(I will freely admit I'm not really sure what the overall goal of
refraining from sourcing /etc/profile itself is. Arguably it's kind of
misuse of the profile...)

This silently broke the cuda tests, which never ran because the cuda
compiler was not detected as available.

While we are at it, I guess we can convert gentoo to use the same trick
of appending it in install.sh
This commit is contained in:
Eli Schwartz 2024-05-30 17:00:30 -04:00
parent f4577911b4
commit d9e3a3f09f
No known key found for this signature in database
GPG Key ID: CEB167EFB5722BD6
3 changed files with 3 additions and 6 deletions

View File

@ -80,11 +80,6 @@ class Builder(BuilderBase):
fi
'''
if self.data_dir.name == 'gentoo':
out_data += '''
source /etc/profile
'''
out_file.write_text(out_data, encoding='utf-8')
# make it executable

View File

@ -18,4 +18,4 @@ install_minimal_python_packages
# Manually remove cache to avoid GitHub space restrictions
rm -rf /var/cache/pacman
echo "source /etc/profile.d/cuda.sh" >> /ci/env_vars.sh
echo "source /etc/profile" >> /ci/env_vars.sh

View File

@ -156,3 +156,5 @@ rm /usr/lib/python/EXTERNALLY-MANAGED
python3 -m ensurepip
install_python_packages
python3 -m pip install "${base_python_pkgs[@]}"
echo "source /etc/profile" >> /ci/env_vars.sh