在 NixOS 中使用 axel 获取源代码,而不是使用curl 获取 fetchurl

在 NixOS 中使用 axel 获取源代码,而不是使用curl 获取 fetchurl

我正在尝试使用 Cuda 启动 shell。我有以下 shell.nix 文件

{ pkgs ? import <nixpkgs> { } }:

pkgs.mkShell {
  name = "cuda-env-shell";
  buildInputs = with pkgs; [
    unzip
    # other packages
    cudatoolkit
    # other packages
    binutils
  ];
}

但这由于以下错误而失败。我以前曾遇到过像 davinci-resolve 这样的大文件的情况。对于 davinci-resolve,我的解决方案是使用具有 10 个连接的 axel 创建一个新的 flake,并且每次都运行良好。现在我正面临着 Cuda 的这个问题。

[1/0/3 built, 4 copied (183.8/183.8 MiB), 42.8 MiB DL] building cuda_11.8.0_520.61.05_linux.run:                                  Dload  Upload   Total   Spent    Left
error: builder for '/nix/store/idz14mzbsv4k5v5ims095jcjk2r50glr-cuda_11.8.0_520.61.05_linux.run.drv' failed with exit code 1;
       last 7 log lines:
       >
       > trying https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run
       >   % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
       >                                  Dload  Upload   Total   Spent    Left  Speed
       >  49 4135M   49 2044M    0     0   772k      0  1:31:20  0:45:08  0:46:12     0
       > curl: (56) Recv failure: Connection reset by peer

我想要 Cuda,但是我可以使用 axel 而不是curl 检索 src 吗?我尝试了类似以下的操作,但它似乎不是有效的语法?

    cudatoolkit.overrideAttrs
    (finalAttrs: previousAttrs: {
      pname = previousAttrs.pname + "-bar";
      src = runCommandLocal "${previousAttrs.pname}.run" rec { } ''
        axel \
          --num-connections=10 \
          --verbose \
          --insecure \
          --output $out \
          "https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run"
      '';
    })

错误:

        at /nix/store/0g53xyh39z3y90p4d8r341wbqyjy1zhl-source/pkgs/stdenv/generic/make-derivation.nix:395:7:

          394|       depsHostHost                = elemAt (elemAt dependencies 1) 0;
          395|       buildInputs                 = elemAt (elemAt dependencies 1) 1;
             |       ^
          396|       depsTargetTarget            = elemAt (elemAt dependencies 2) 0;

       error: Dependency is not of a valid type: element 13 of buildInputs for cuda-env-shell

答案1

第一步,需要在表达式两边添加括号(cudatoolkit.overrideAttrs (finalAttrs: previousAttrs: { .. }))以避免破坏列表。

第二步,需要添加一个outputHash使构建器能够接受下载的源代码。最初,您可以使用

src = runCommandLocal "${previousAttrs.pname}.run" {
  outputHashMode = "flat";
  outputHashAlgo = "sha256";
  outputHash = "";
}

触发显示源哈希的错误。在这种情况下,我获得了哈希值:sha256-kiPErzrr5Ke77Zq9mxY7A6GzS4VfvCtKDRtwasCaWhY=

最后,需要添加axel到执行命令的环境中。由于它仅用于下载源码,因此您可以将其添加nativeBuildInputsrunCommandLocal.此外,您可以axel在脚本中替换为${pkgs.axel}/bin/axel.

{pkgs ? import <nixpkgs> {}}:
pkgs.mkShell {
  name = "cuda-env-shell";
  buildInputs = with pkgs; [

    (cudatoolkit.overrideAttrs (finalAttrs: previousAttrs: {
      pname = previousAttrs.pname;
      src =
        runCommandLocal "${previousAttrs.pname}.run" {
          nativeBuildsInputs = [axel];
          # outputHashMode = "flat";
          # outputHashAlgo = "sha256";
          # outputHash = "";
          outputHash = "sha256-kiPErzrr5Ke77Zq9mxY7A6GzS4VfvCtKDRtwasCaWhY=";
        } ''
          # ${pkgs.axel}/bin/axel \
          axel \
            --num-connections=10 \
            --verbose \
            --insecure \
            --output "$out" \
            "https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run"
        '';
    }))
  ];
}

相关内容