Workarounds for lazy_html

[tjarratt]

Based on the work started in #22, I identified what was causing the build for fail for vix. Copying my comment from that PR below

Having a look at this, I think I see what happened. When we apply the elixir_make workaround for lazy_html, that applies to vix (because it depends upon elixir_make), and since our override sets the nativeBuildInputs for the derivation, the ones we had previously set for vix end up getting overriden.

I think I see a few ways to handle this, I'll take a pass at it.

I had thought to use overrrideAttrs so that we could append onto the nativeBuildInputs, but I tested whether we could build lazy_html and vix without setting pkgs.gnumake in the nativeBuildInputs and was pleasantly surprised to see this wasn't necessary for it to build. This might have side effects I can't see, or be undesirable for a number of reasons, let me know if there's any way of improving this and I can have a shot.

Closes #22 #30

[camelpunch]

Thanks! Can you take a look at the Linux build please?

[tjarratt]

For sure !

I was waiting around last night for the build to go through, but it took too long and I went to bed first. Will have a look at the linux build and get this in a good shape :-D

[camelpunch]

It is a silly slow build indeed.

[tjarratt]

Made some progress with the initial issue (it was missing ca certs) but now I'm running into issues with how elixir_make is setup in lazy_html. My next step is to compare how the precompilation is setup in vix and lazy_html to figure out what it would take to compile lazy_html on linux.

And then once that's done, I intend to try tearing out bits and pieces of the elixirMake workaround to ensure all the additions are necessary.

[ju1m]

Hi, if that may help you, for lazy_html on Linux I'm using:

{ lexbor, ... }:
finalMixPkgs: previousMixPkgs: {
  lazy_html = previousMixPkgs.lazy_html.overrideAttrs (previousAttrs: {
    # Explanation: somehow `mix compile --no-deps-check`
    # replaces Fine.include_dir() by "/build/fine-0.1.4/c_include"
    # a path which is not available when building lazy_html there.
    #
    # Explanation: lazy_html being built in a sandbox
    # it cannot download its precompiled binary,
    # it then attempt to compile from source by first git cloning lexbor,
    # but lexbor is already packaged in nixpkgs,
    # and to let the Makefile reuse it, it's enough to empty @lexbor_git_sha.
    postPatch = ''
      substituteInPlace mix.exs \
        --replace-fail "Fine.include_dir()" '"${finalMixPkgs.fine}/src/c_include"' \
        --replace-fail '@lexbor_git_sha "244b84956a6dc7eec293781d051354f351274c46"' '@lexbor_git_sha ""'
    '';

    # Explanation: workaround:
    # (File.Error) could not make directory (with -p) "/homeless-shelter/.cache/elixir_make":
    # no such file or directory
    preConfigure = previousAttrs.preConfigure or "" + ''
      export ELIXIR_MAKE_CACHE_DIR="$TMPDIR/.cache"
    '';

    # Explanation: nix provides lexbor.
    preBuild = previousAttrs.preBuild or "" + ''
      export LEXBOR_GIT_SHA=
      install -Dm644 \
        -t _build/c/third_party/lexbor/$LEXBOR_GIT_SHA/build \
        ${lexbor}/lib/liblexbor_static.a
    '';
    buildInputs = previousAttrs.buildInputs or [ ] ++ [
      lexbor
    ];
  });
}

[tjarratt]

@ju1m you're a livesaver, thanks for sharing that ! I'll give it a shot and see if it gets the build green here.

[tjarratt]

Builds are green ✅ 🎉

Big shout outs to @codyps for giving me a good starting point to contribute, and to @ju1m for giving me the nudge that got this over the line. I knew there was SOMETHING about how the 'mix.exs' file in lazy_html configured elixir_make that made it prefer downloading over building it directly, as vix does, but I couldn't quite figure it out.

And of course, a thank you to @camelpunch for support and guidance throughout. Lemme know if you have any feedback on how this landed, I wouldn't mind making a few more changes if you see any way this could be improved.