Fix broken CMake-generated pkg-config files

[alexshpilkin]

Description of changes

Fixes #167528 and several other cases of similarly broken pkg-config files generated with CMake. The case of zxing-cpp was discussed in review comments of #172046.

I tried to patch things in such a way that the build won’t break even once an upstream fix is in place.

Upstream bugs: jiixyj/libebur128#121, libjxl/libjxl#1400, kcat/openal-soft#696, sctplab/usrsctp#662, Mindwerks/wildmidi#236, zxing-cpp/zxing-cpp#335, KhronosGroup/SPIRV-Tools#3905, openobex#66, dirkvdb/ffmpegthumbnailer#215, Matroska-Org/libmatroska#62, extra-cmake-modules!268, Matroska-Org/libebml#97.

I only patched things that I found by grepping the Nix store on my machine, so it’s probable there are many more cases of this problem in CMake build systems. Someone who has access to a full build of Nixpkgs should probably make a systematic review (rg -g '*.pc' '}//nix' /nix/store will probably find most cases).

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandbox = true set in nix.conf? (See Nix manual)
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 22.05 Release Notes (or backporting 21.11 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
  • (Release notes changes) Ran nixos/doc/manual/md-to-db.sh to update generated release notes
• Fits CONTRIBUTING.md.

[alexshpilkin]

I don’t even have the disk space to run nixpkgs-review on this, so if someone would do that I’d be grateful :-/

[gador]

Running rg -g '*.pc' '}//nix' /nix/store additionally outputs:

• openobex
• libeml
• libffmpegthumbnailer
• libmatroska
• libKActivities
• KF5CalendarCore
• libeml

[alexshpilkin]

@gador Thanks!

(Also, damn it, I’ve already read enough CMake today to last me a year... I’ll get to those as well, I promise, I really will.)

Right, my system also had kwayland, I forgot about that. The KDE people use some sort of .pc generator written in CMake that seems to produce uniformly broken pkg-config files, except I’m not sure if they are actually shipping them. Someone more versed in CMake should probably have a look at it.

[SuperSandro2000]

Either we find the package that is causing so many rebuilds or we send everything trough staging.

[alexshpilkin]

@SuperSandro2000 My money is on either libjxl or spirv-tools being deep in the dependency tree of some desktop environment or widget toolkit, but I’m honestly clueless here. Is there even a way to view the list ofborg computed?

[nh2]

Is there even a way to view the list ofborg computed?

You can click on

ofborg-eval — ^.^!

in the Github Actions list:

That brings up the changed paths: https://gist.github.com/GrahamcOfBorg/67f0505e1d8f9c25d56caa663fbb64ad

[nh2]

dependency tree of some desktop environment or widget toolkit

Yes, gtk4 is in the list because of gstreamer:

% nix why-depends nixpkgs.gtk4 nixpkgs.gst_all_1.gst-plugins-bad
/nix/store/699xjlyv5n4k4qcwhgxiv0vhv4m5y2kx-gtk4-4.4.1
╚═══lib/gtk-4.0/4.0.0/media/libmedia-gstreamer.so: …gins-base-1.18.4/lib:/nix/store/m3jdcmxgn44v5xz9qikj3hjkcda58awf-gst-plugins-bad-1.18.2/lib:/nix…
    => /nix/store/m3jdcmxgn44v5xz9qikj3hjkcda58awf-gst-plugins-bad-1.18.2

[alexshpilkin]

@nh2 Right, the smiley! ^^ I knew I saw that list at one point, but could not guess which line of the report to poke at and started doubting myself.

@SuperSandro2000 So, I was right that this is a GUI cascade but wrong about the source.

The majority of the rebuilds are because of both gtk4 and phonon-backend-gstreamer depending on gst-plugins-bad (which this touches in an incidental drive-by manner) which in turn depends on wildmidi (which this touches in an essential manner). Some of the Qt stuff is triggered by libkdegames depending on openal-soft (which this also touches, indepenently). Other parts of the Qt stuff seem to be spurious as they’re marked broken.

Overall, I think punting the wildmidi and the gst-plugins-bad changes to staging should redirect most of the avalanche—or we can just give up and decide this is going to be a large rebuild. Again, I’m clueless here, so please tell me if you want me to cherry-pick some things to a separate branch or something like that.

[nixos-discourse]

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/extend-include-search-path-for-a-specific-package/19438/2

[jtojnar]

Might be better to fetch it from a stable URL: https://github.com/AnotherFoxGuy/CMakeCM/raw/afe41f4536ae21f6f11f83e8c0b8b8c450ef1332/modules/JoinPaths.cmake

[Artturin]

done in #181875

[Artturin]

fix'd up and resolved conflicts #181875

[Artturin]

Merged #181875
Thanks @alexshpilkin !