Update for CUDA 12.3.2

[adibbley]

xref: conda-forge/cuda-feedstock#15

[conda-forge-webservices]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[adibbley]

@conda-forge-admin, please rerender

[jakirkham]

Thanks Alex! 🙏

Will take a closer look at CI tomorrow

[jakirkham]

Thanks Alex! 🙏

Had a couple questions below

[jakirkham]

Would it work if we used the same approach as we did to resolve issue ( conda-forge/cuda-feedstock#10 )?

Suggested change

	patchelf --set-rpath '$ORIGIN/../lib' $PREFIX/$i/$j
	patchelf --set-rpath '$ORIGIN' $PREFIX/$i/$j

[adibbley]

No, these will turn back to warnings:

INFO (cuda-sanitizer-api,compute-sanitizer/libTreeLauncherTargetInjection.so): Needed DSO lib/libgcc_s.so.1 found in conda-forge/linux-64::libgcc-ng==13.2.0=h807b86a_3

[jakirkham]

Wonder if we could symlink libgcc_s.so into the same directory as libTreeLauncherTargetInjection.so

Maybe that would make the patchelf call unneeded

For more context, thinking about the patchelf call with '$ORIGIN/../lib', am wondering if that will cause issues with the targets structure used in other packages (though admittedly not as relevant here)

Edit: Also not using patchelf might allow us to keep the overlinking check

[adibbley]

I don't see how managing that symlink could be at all preferable to what I have here.

Why are patchelf and the overlink check a concern here when we've already done that for every other cuda library package?

The libraries here are not installed to targets, or even lib, so you are right that it is not relevant here.

[jakirkham]

Are we able to revert this after the patchelf change?

Suggested change

	error_overlinking: false
	error_overlinking: true

[adibbley]

No, we've needed this anytime we update rpath (example: https://github.com/conda-forge/libcurand-feedstock/pull/2/files#diff-478786365dd93f740eb520c2faa03d7a0623273663b75472272c8ef94297bbe2)

This will be the error:

WARNING (cuda-sanitizer-api,compute-sanitizer/libsanitizer-public.so): runpaths ['$ORIGIN/../lib'] found in /home/conda/feedstock_root/build_artifacts/cuda-sanitizer-api_1706216449799/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_pl/compute-sanitizer/libsanitizer-public.so

[jakirkham]

Was trying to get a better sense of what is going on here. So downloaded the redist archive. In looking at its contents, a couple things stood out:

1. The compute-sanitizer directory has libraries not under a lib directory (unlike the headers in include)
2. There are 2 layers of what appear to be the same SOs

Do we know why these are happening?

Have included the abbreviated results from running tree below for clarity:

cuda_sanitizer_api-linux-x86_64-12.3.101-archive
...
├── compute-sanitizer
│   ...
│   ├── libInterceptorInjectionTarget.so
│   ├── libsanitizer-collection.so
│   ├── libsanitizer-public.so
│   ├── libTreeLauncherPlaceholder.so
│   ├── libTreeLauncherTargetInjection.so
│   ├── libTreeLauncherTargetUpdatePreloadInjection.so
│   ├── TreeLauncherSubreaper
│   ├── TreeLauncherTargetLdPreloadHelper
│   └── x86
│       ├── libInterceptorInjectionTarget.so
│       ├── libTreeLauncherPlaceholder.so
│       ├── libTreeLauncherTargetInjection.so
│       ├── libTreeLauncherTargetUpdatePreloadInjection.so
│       ├── TreeLauncherSubreaper
│       └── TreeLauncherTargetLdPreloadHelper
...

[jakirkham]

Interestingly was able to get the package to build locally with only this change (adding the sysroot to requirements/build)

Given this, wonder if we can drop the patchelf bits

[adibbley]

Not sure how you tested it, this is not sufficient when running build-locally.py

[jakirkham]

Did the following locally

docker run --rm -it quay.io/condaforge/linux-anvil-aarch64

cat >>~/.condarc <<EOF
conda-build:
  error_overlinking: true
EOF

git clone https://github.com/adibbley/cuda-sanitizer-api-feedstock -b cuda-12.3.2
cd cuda-sanitizer-api-feedstock
git reset --hard 25c14ad89edd769dfe5e4955f7d44f5313b91926

git apply -
diff --git a/recipe/meta.yaml b/recipe/meta.yaml
index 71d935e..9a0d82b 100644
--- a/recipe/meta.yaml
+++ b/recipe/meta.yaml
@@ -30,6 +30,7 @@ requirements:
   build:
     - {{ compiler("c") }}
     - {{ compiler("cxx") }}
+    - sysroot_{{ target_platform }} 2.17    # [linux]
     - arm-variant * {{ arm_variant_type }}  # [aarch64]
   host:
     - cuda-version {{ cuda_version }}
<ctrl + d>

conda build -m .ci_support/linux_aarch64_.yaml recipe/

[adibbley]

Looks like you tested aarch64, this error only occurred on x86_64 AFAICT. First CI from before any changes: https://github.com/conda-forge/cuda-sanitizer-api-feedstock/pull/5/checks?check_run_id=20821915045

[jakirkham]

Oops thanks for catching that. Am on an ARM machine where this is a little hard to test

Let's stick with what we have already. If we come up with a better solution, we can always do another PR

[bdice]

This looks fine to me. I have no significant reservations about the changes here. Happy to merge once @jakirkham and @adibbley are ready.

[jakirkham]

Let's go ahead and merge. We can iterate on any improvements separately