cudaPackages: multiplex builder always provides attributes by ConnorBaker · Pull Request #406207 · NixOS/nixpkgs

ConnorBaker · 2025-05-11T16:02:20Z

#405707 broke evaluation of at least one attribute on Darwin (python3Packages.jax-cuda12-pjrt) because the multiplex builder would only add attributes for packages on supported platforms and access to cudaPackages was not guarded by platform checks.

As a quick fix, this PR changes the behavior of the multiplex builder so that it will add attributes for packages supporting the current CUDA version, even if they are unsupported on the current platform.

Tested with

nix-build ci -A eval.full   --max-jobs 24   --cores 16   --arg chunkSize 10000   --arg evalSystems '["x86_64-linux" "aarch64-linux" "x86_64-darwin" "aarch64-darwin"]'

EDIT:

The above didn't error on master, so I don't know that it's a good indication that this fixes the eval error.

However, I did verify manually:

[connorbaker@nixos-desktop:~/nixpkgs]$ git switch master
Switched to branch 'master'
Your branch is up to date with 'origin/master'.

[connorbaker@nixos-desktop:~/nixpkgs]$ nix eval --system aarch64-darwin .#python312Packages.jax-cuda12-pjrt
«error: attribute 'cudnn' missing»

[connorbaker@nixos-desktop:~/nixpkgs]$ git switch fix/cuda-darwin-eval 
Switched to branch 'fix/cuda-darwin-eval'

[connorbaker@nixos-desktop:~/nixpkgs]$ nix eval --system aarch64-darwin .#python312Packages.jax-cuda12-pjrt
«error: Package ‘python3.12-jax-cuda12-pjrt-0.6.0’ in /nix/store/1jnd81lxfm8rqk1zy25syqhpnpy2bklm-source/pkgs/development/python-modules/jax-cuda12-pjrt/default.nix:96 is not available on the requested hostPlatform:
  hostPlatform.config = "arm64-apple-darwin"
  package.meta.platforms = [
    "aarch64-linux"
    "armv5tel-linux"
    "armv6l-linux"
    "armv7a-linux"
    "armv7l-linux"
    "i686-linux"
    "loongarch64-linux"
    "m68k-linux"
    "microblaze-linux"
    "microblazeel-linux"
    "mips-linux"
    "mips64-linux"
    "mips64el-linux"
    "mipsel-linux"
    "powerpc64-linux"
    "powerpc64le-linux"
    "riscv32-linux"
    "riscv64-linux"
    "s390-linux"
    "s390x-linux"
    "x86_64-linux"
  ]
  package.meta.badPlatforms = [ ]
, refusing to evaluate.

a) To temporarily allow packages that are unsupported for this system, you can use an environment variable
   for a single invocation of the nix tools.

     $ export NIXPKGS_ALLOW_UNSUPPORTED_SYSTEM=1
     
   Note: When using `nix shell`, `nix build`, `nix develop`, etc with a flake,
         then pass `--impure` in order to allow use of environment variables.
    
b) For `nixos-rebuild` you can set
  { nixpkgs.config.allowUnsupportedSystem = true; }
in configuration.nix to override this.

c) For `nix-env`, `nix-build`, `nix-shell` or any other Nix command you can add
  { allowUnsupportedSystem = true; }
to ~/.config/nixpkgs/config.nix.
»

Things done

Add a 👍 reaction to pull requests you find important.

…n no attributes Signed-off-by: Connor Baker <ConnorBaker01@gmail.com>

SomeoneSerge · 2025-05-11T16:39:51Z

pkgs/development/cuda-modules/generic-builders/multiplex.nix

          ${majorMinorVersion} =
-            # Only keep the existing package if it is newer than the one we are considering.
-            if existingPackage != null && lib.versionOlder package.version existingPackage.version then
+            # Only keep the existing package if it is newer than the one we are considering or it is supported on the
+            # current platform and the one we are considering is not.
+            if
+              existingPackage != null
+              && (
+                packageOlder package existingPackage
+                || (!packageSupportedPlatform package && packageSupportedPlatform existingPackage)
+              )


https://github.com/NixOS/nixpkgs/pull/368366/files#diff-3f6a5eea0b39457aafc25a8802186e0685957214bce4cc2455d6f66e0d2822cbR83-R84

Unfortunately, though, we won't be able to see trofi's commit message in git-blame 9fd753e

Added a comment to reference it when calling lib.sort.

I'd still recover the named function

SomeoneSerge · 2025-05-11T16:41:14Z

pkgs/development/cuda-modules/generic-builders/multiplex.nix

+  # allPackages :: List (Package + { redistArch: String })
+  allPackages =
+    let
+      redistArchs = lib.attrNames evaluatedModules.config.${pname}.releases;
+    in
+    lib.concatMap (
+      redistArch:
+      let
+        packages = evaluatedModules.config.${pname}.releases.${redistArch};
+      in
+      lib.concatMap (
+        package: lib.optionals (satisfiesCudaVersion package) [ ({ inherit redistArch; } // package) ]
+      ) packages
+    ) redistArchs;
+
+  packageOlder = p1: p2: lib.versionOlder p1.version p2.version;
+  packageSupportedPlatform = p: p.redistArch == redistArch;


Or you keep the previous version but rename it to allPackages', and add allPackages = allPackages' ++ [(head ...)] where ... is flattened releases. The previous filter was more readable than 2x concatMap

Re-introduced allReleases and introduced allReleases' to make it more clear what's happening.

By "previous" I meant the one you introduced in the previous PR, where you only keep the matching platform's packages. That part looked fairly simple and comprehensible, we could keep it and append just one potentially-unsupported package at the end

Your call though, I can also look in more detail at the current version?

I'd say look more closely at the current version since it makes sure that the incompatible package used at least matches the version.

match the version

Good point. FYI my plan for the fallback is to runCommand pname { ... } "false", but it might be easier to do after the evalModules schema update

…her than no attributes fixup! cudaPackages: simplify multiplex builder's allPackages Signed-off-by: Connor Baker <ConnorBaker01@gmail.com>

ConnorBaker · 2025-05-12T13:37:59Z

Merging to unbreak Darwin eval.

SomeoneSerge · 2025-05-12T13:49:20Z

pkgs/development/cuda-modules/generic-builders/multiplex.nix

rebase --autosquash

github-actions bot added the 6.topic: cuda Parallel computing platform and API label May 11, 2025

ConnorBaker changed the title ~~cudaPackages: multiplex builder provides broken attributes rather tha…~~ cudaPackages: multiplex builder always provides attributes May 11, 2025

github-project-automation bot added this to CUDA Team May 11, 2025

github-project-automation bot moved this to New in CUDA Team May 11, 2025

ConnorBaker moved this from New to 👀 Awaits reviews in CUDA Team May 11, 2025

ConnorBaker marked this pull request as ready for review May 11, 2025 16:03

ConnorBaker mentioned this pull request May 11, 2025

cuda-modules: fix and clean up multiplex builder package selection logic #405707

Merged

13 tasks

ConnorBaker self-assigned this May 11, 2025

ConnorBaker force-pushed the fix/cuda-darwin-eval branch from d8270fe to 4bc9a05 Compare May 11, 2025 16:07

cudaPackages: multiplex builder provides broken attributes rather tha…

143898c

…n no attributes Signed-off-by: Connor Baker <ConnorBaker01@gmail.com>

ConnorBaker force-pushed the fix/cuda-darwin-eval branch from 4bc9a05 to 143898c Compare May 11, 2025 16:07

github-actions bot added 10.rebuild-darwin: 0 This PR does not cause any packages to rebuild on Darwin. 10.rebuild-linux: 0 This PR does not cause any packages to rebuild on Linux. labels May 11, 2025

ConnorBaker requested review from SomeoneSerge and WeetHet May 11, 2025 16:16

SomeoneSerge reviewed May 11, 2025

View reviewed changes

fixup! cudaPackages: multiplex builder provides broken attributes rat…

19d4f53

…her than no attributes fixup! cudaPackages: simplify multiplex builder's allPackages Signed-off-by: Connor Baker <ConnorBaker01@gmail.com>

ConnorBaker requested a review from SomeoneSerge May 11, 2025 17:28

ConnorBaker merged commit 1a120d4 into NixOS:master May 12, 2025
22 checks passed

github-project-automation bot moved this from 👀 Awaits reviews to ✅ Done in CUDA Team May 12, 2025

SomeoneSerge reviewed May 12, 2025

View reviewed changes

pkgs/development/cuda-modules/generic-builders/multiplex.nix

Copy link

Contributor

SomeoneSerge May 12, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

rebase --autosquash

This was referenced May 13, 2025

tests.cuda.db: init #406740

Closed

cudaPackages_11_0.autoAddCudaCompatRunpath.libcudaPath: fix the eval #406668

Merged

Uh oh!

Conversation

ConnorBaker commented May 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Things done

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ConnorBaker commented May 12, 2025

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

ConnorBaker commented May 11, 2025 •

edited

Loading