python3Packages.llama-cpp-python: disable removeStubsFromRunpathHook to fix import when cudaSupport is enabled

[GaetanLepage]

Things done

Better alternative to #465751.
This fixes all reverse dependencies.

Ideally, we would need a cleaner toggle to disable removeStubsFromRunpathHook.

cc @ConnorBaker @SomeoneSerge

• Built on platform:
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• Tested, as applicable:
  • NixOS tests in nixos/tests.
  • Package tests at passthru.tests.
  • Tests in lib/tests or pkgs/test for functions and "core" functionality.
• Ran nixpkgs-review on this PR. See nixpkgs-review usage.
• Tested basic functionality of all binary files, usually in ./result/bin/.
• Nixpkgs Release Notes
  • Package update: when the change is major or breaking.
• NixOS Release Notes
  • Module addition: when adding a new NixOS module.
  • Module update: when the change is significant.
• Fits CONTRIBUTING.md, pkgs/README.md, maintainers/README.md and other READMEs.

---

Add a 👍 reaction to pull requests you find important.

[GaetanLepage]

nixpkgs-review result

Generated using nixpkgs-review.

Command: nixpkgs-review pr 468659 --extra-nixpkgs-config '{ allowUnfree = true; cudaSupport = true; }'
Commit: c8cb3ca5f3d163e32fd3de4c3dc7574a79cfddb6

---

x86_64-linux

✅ 25 packages built:

• nixpkgs-manual
• python312Packages.kserve
• python312Packages.kserve.dist
• python312Packages.llama-cpp-python
• python312Packages.llama-cpp-python.dist
• python312Packages.llm-gguf
• python312Packages.llm-gguf.dist
• python312Packages.outlines
• python312Packages.outlines.dist
• python312Packages.torchrl
• python312Packages.torchrl.dist
• vllm (python312Packages.vllm)
• vllm.dist (python312Packages.vllm.dist)
• python313Packages.kserve
• python313Packages.kserve.dist
• python313Packages.llama-cpp-python
• python313Packages.llama-cpp-python.dist
• python313Packages.llm-gguf
• python313Packages.llm-gguf.dist
• python313Packages.outlines
• python313Packages.outlines.dist
• python313Packages.torchrl
• python313Packages.torchrl.dist
• python313Packages.vllm
• python313Packages.vllm.dist

---

aarch64-linux

✅ 9 packages built:

• nixpkgs-manual
• python312Packages.llama-cpp-python
• python312Packages.llama-cpp-python.dist
• python312Packages.llm-gguf
• python312Packages.llm-gguf.dist
• python313Packages.llama-cpp-python
• python313Packages.llama-cpp-python.dist
• python313Packages.llm-gguf
• python313Packages.llm-gguf.dist

[ConnorBaker]

IMHO this is the wrong approach.

References to stubs should never be present in runtime binaries because they cannot and will not work. At best other runpath entries supply the full library before the stub is tried. At worst (think non-NixOS, libraries which don’t link against the driver but need the driver, or when we need to use cuda_compat) the stub file is loaded and tracking down the runtime failure becomes much more difficult.

The root of the problem is that importing the module immediately tries to dlopen a library — that should be lazy.

By disabling the hook, all that happens is that the module succeeds in immediately dlopening a library (which it shouldn’t do) corresponding to a stub library (which will never work and shouldn’t be in the closure).

[GaetanLepage]

I applied the same process as in #465751 to outlines in #471697.

I'll be closing this PR.