I don't think fq7IA5osMKsLx1jTA1iHZ2k972v0myJIWiwAvy4TbLN had anything to do with bad copy-pasting btw. That's the same hash I saw in the git blames, so either we broke it much earlier, or nvidia really has swapped the blob. Otherwise, we really should avoid editing these files manually. Instead, we want a deterministic and idempotent update script that we can run pre-commit style, instead of reviews. If two users run the updater and get the exact same output, the hashes pass the review. We should also add this to the CI probably

PR=273604; \
SYSTEM="aarch64-linux"; \
CUDA_SUPPORT="true"; \
CUDA_CAPABILITIES='[ "7.5" ]'; \
nixpkgs-review pr "$PR" \
  --system "$SYSTEM" \
  --no-shell \
  --checkout commit \
  --allow aliases \
  --build-args "--max-jobs 1" \
  --extra-nixpkgs-config "{
    allowUnfree = true;
    allowBroken = false;
    cudaSupport = ${CUDA_SUPPORT:-false};
    cudaCapabilities = ${CUDA_CAPABILITIES:-[]};
  }"

aarch64-darwin

Result of nixpkgs-review pr 273604 --extra-nixpkgs-config '{ allowUnfree = true; allowBroken = false; cudaSupport = false; cudaCapabilities = []; }' run on aarch64-darwin 1

x86_64-darwin

Result of nixpkgs-review pr 273604 --extra-nixpkgs-config '{ allowUnfree = true; allowBroken = false; cudaSupport = false; cudaCapabilities = []; }' run on x86_64-darwin 1

aarch64-linux

Jetson

Result of nixpkgs-review pr 273604 --extra-nixpkgs-config '{ allowUnfree = true; allowBroken = false; cudaSupport = true; cudaCapabilities = [ "7.2" ]; }' run on aarch64-linux 1

Non-Jetson

Result of nixpkgs-review pr 273604 --extra-nixpkgs-config '{ allowUnfree = true; allowBroken = false; cudaSupport = true; cudaCapabilities = [ "7.5" ]; }' run on aarch64-linux 1

Non-CUDA

Result of nixpkgs-review pr 273604 --extra-nixpkgs-config '{ allowUnfree = true; allowBroken = false; cudaSupport = false; cudaCapabilities = []; }' run on aarch64-linux 1

x86_64-linux

Non-Jetson

Result of nixpkgs-review pr 273604 --extra-nixpkgs-config '{ allowUnfree = true; allowBroken = false; cudaSupport = true; cudaCapabilities = [ "7.5" ]; }' run on x86_64-linux 1

Non-CUDA

Result of nixpkgs-review pr 273604 --extra-nixpkgs-config '{ allowUnfree = true; allowBroken = false; cudaSupport = false; cudaCapabilities = []; }' run on x86_64-linux 1

I don't think fq7IA5osMKsLx1jTA1iHZ2k972v0myJIWiwAvy4TbLN had anything to do with bad copy-pasting btw. That's the same hash I saw in the git blames, so either we broke it much earlier, or nvidia really has swapped the blob. Otherwise, we really should avoid editing these files manually. Instead, we want a deterministic and idempotent update script that we can run pre-commit style, instead of reviews. If two users run the updater and get the exact same output, the hashes pass the review. We should also add this to the CI probably

I agree, but I don't have the bandwidth at the moment to work on that.

I imagine such a script could be implemented as a DFS traversal of the pages at the CUDNN resist URL. If there's a new version which doesn't exist, add it. If there's a newer version of the same major/minor release, replace it with the newer version. Otherwise, skip it.

@SomeoneSerge can you confirm that jaxlib and cupy are broken on master?

@ConnorBaker yes https://hercules-ci.com/github/SomeoneSerge/nixpkgs-cuda-ci/jobs/7448

I agree, but I don't have the bandwidth at the moment to work on that.

Yea this is definitely not the top priority