Copilot commented on code in PR #50279:
URL: https://github.com/apache/arrow/pull/50279#discussion_r3498184093


##########
dev/tasks/linux-packages/apache-arrow/debian/control.in:
##########
@@ -30,7 +30,11 @@ Build-Depends:
   meson,
   ninja-build,
   nlohmann-json-dev | nlohmann-json3-dev,
-  nvidia-cuda-toolkit [!arm64],
+# nvidia-cuda-toolkit may be unavailable (e.g. removed from Debian forky).
+# The pkg.apache-arrow.nocuda build profile (set via DEB_BUILD_PROFILES, e.g.
+# in the debian-forky Dockerfile) skips this build dependency and the CUDA
+# packages below.
+  nvidia-cuda-toolkit [!arm64] <!pkg.apache-arrow.nocuda>,

Review Comment:
   The new pkg.apache-arrow.nocuda profile disables the nvidia-cuda-toolkit 
Build-Depends and skips building the CUDA binary packages, but debian/rules 
still enables ARROW_CUDA purely based on whether nvidia-cuda-toolkit is 
installed. If the toolkit is installed (e.g., once it returns to forky), the 
build can still install CUDA artifacts while the CUDA packages are disabled, 
which is likely to leave unshipped files and break the debhelper build 
(dh_missing / fail-missing). Consider making debian/rules force ARROW_CUDA=OFF 
when pkg.apache-arrow.nocuda is active (e.g., by checking DEB_BUILD_PROFILES).



##########
dev/tasks/linux-packages/apache-arrow/apt/debian-forky/Dockerfile:
##########
@@ -88,3 +88,8 @@ RUN \
     apt install -y -V ${quiet} nvidia-cuda-toolkit; \
   fi && \
   apt clean
+
+# nvidia-cuda-toolkit was removed from Debian forky amd64, so disable the
+# CUDA packages via the pkg.apache-arrow.nocuda build profile.See GH-50090.
+# Once nvidia-cuda-toolkit is re-added to Debian forky, this can be removed.
+ENV DEB_BUILD_PROFILES=pkg.apache-arrow.nocuda

Review Comment:
   With DEB_BUILD_PROFILES=pkg.apache-arrow.nocuda set, this image can still 
end up installing nvidia-cuda-toolkit once it reappears in Debian forky (the 
apt list check will start succeeding). debian/rules then sees the toolkit 
installed and turns ARROW_CUDA=ON, even though the CUDA binary packages are 
disabled via Build-Profiles, which is likely to leave unshipped CUDA install 
artifacts and break the package build. Either stop installing 
nvidia-cuda-toolkit while the nocuda profile is active, or ensure ARROW_CUDA is 
forced OFF when the profile is set. (Also fixes the missing space in 
"profile.See".)



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to