It looks like something is broken with PR #214, which should have automatically downloaded the CUDA .run file, then extracted it to find the actual NVIDIA driver .run file to install, but instead it results in #224.
I'm not exactly sure what the issue is, but if we self-host the NVIDIA driver .run file, the old existing process should work just as well!
This should hopefully fix #224, and actually fix #213!
Testing methodology
I've tested this locally on an x86_64 machine by editing build.sh to only have the 545.23.08 driver version:
diff --git a/build.sh b/build.sh
index 65180ab..8411bb6 100755
--- a/build.sh
+++ b/build.sh
@@ -14,7 +14,7 @@ set -x
SDK_BRANCH=1.4
SDK_RUNTIME_VERSION=1.6
-for VER in $DRIVER_VERSIONS; do
+for VER in '545.23.08'; do
F="data/nvidia-$VER-$ARCH.data"
if [ ! -f $F ]; then
echo WARNING, no data file for $VER $ARCH
Then I ran ./build.sh x86_64 repo "" '--install --user' 'Fix CUDA 545.23.08 drivers (6157761)' and it worked.
It looks like something is broken with PR #214, which should have automatically downloaded the CUDA
.run
file, then extracted it to find the actual NVIDIA driver.run
file to install, but instead it results in #224.I'm not exactly sure what the issue is, but if we self-host the NVIDIA driver
.run
file, the old existing process should work just as well!These files are hosted at https://github.com/flathub/org.freedesktop.Platform.GL.nvidia/releases/tag/cuda, which seems okay under both NVIDIA's and GitHub's rules (see https://github.com/flathub/org.freedesktop.Platform.GL.nvidia/issues/224#issuecomment-1893017543).
This will only be required until we switch to using a normal package, and not extra-data (see https://github.com/flathub/org.freedesktop.Platform.GL.nvidia/pull/167).
This should hopefully fix #224, and actually fix #213!
Testing methodology
I've tested this locally on an
x86_64
machine by editingbuild.sh
to only have the545.23.08
driver version:Then I ran
./build.sh x86_64 repo "" '--install --user' 'Fix CUDA 545.23.08 drivers (6157761)'
and it worked.