fonts broken for some applications

[yshui]

Describe the bug

root cause is #324516 removing the <include /> directive from fonts.conf. #338326 (and #338815) attempted at fixing this, but they are incomplete. as they only hot patched the fonts.conf installed to /etc, but not the one in the output of fontconfig, which is used by some packages.

for example:

nixpkgs/pkgs/applications/networking/browsers/vivaldi/default.nix

Lines 114 to 115 in b79ce4c

	--set-default FONTCONFIG_FILE "${fontconfig.out}/etc/fonts/fonts.conf" \
	--set-default FONTCONFIG_PATH "${fontconfig.out}/etc/fonts" \

(just search for FONTCONFIG_FILE, there are many more.)

feels wrong to me to hardcode a fonconfig path for the application instead of using /etc/fonts.

related: #338908

Notify maintainers

@teto @K900

---

Add a 👍 reaction to issues you find important.

[yshui]

a simple workaround for this is setting FONTCONFIG_FILE and FONTCONFIG_PATH in environment.sessionVariables.

(Even with this workaround, some applications still doesn't fully work, like librewolf. I don't know what's causing it.)

[yshui]

I reverted #324516 locally and now i have 700 derivations to build :(

[nixos-discourse]

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

https://discourse.nixos.org/t/tests-are-failing-due-to-missing-fontconfig-config/51487/4

[teto]

sorry about that. I usually run a nixpkgs-review on my PRs, especially of this magnitude but It's been some time so I dont remember the details.
I confess I missed that pkgs/development/libraries/fontconfig/default.nix was calling ./make-fonts-conf.xsl directly instead of makeFontsConf. If it stays like this, I fear a similar issue rises again: it should use the function.
Thanks for providing fixes to the issue.

[yshui]

don't worry about it, everyone makes mistakes/misses stuff from time to time, it's inevitable.

i am more disappointed by the fact nixpkgs doesn't have some process in place to catch mistakes like this. and mistakes like this isn't even that subtle. (well, maybe things like this is actually much harder to catch than it seems on the surface. i am not involved in nixpkgs development, so please correct me.)

personally, since i started using nixos-unstable, on average i have had easy-to-catch bugs like this around once per month. which feels way more frequent than it should be, and does erode my trust in nixpkgs.

[tukanoidd]

i am more disappointed by the fact nixpkgs doesn't have some process in place to catch mistakes like this. and mistakes like this isn't even that subtle. (well, maybe things like this is actually much harder to catch than it seems on the surface. i am not involved in nixpkgs development, so please correct me.)

personally, since i started using nixos-unstable, on average i have had easy-to-catch bugs like this around once per month. which feels way more frequent than it should be, and does erode my trust in nixpkgs.

Not involved either, just pure speculation, but i think it's just not feasible to have some kind of automated testing for these things because there are more than 100k packages, 10k module options + home-manager, which also interacts with nixpkgs for some things. And they all can be mix-and-matched in variaety of different configurations, which nobody can predict. The language itself is lazy-evaluated and doesn't have any static type safety that you could encode your logic with (lib.types are runtime-based and dont enforce any compile-time guarantees either, just fancy test "identity" functions from what I can gather) and get errors as soon as the code is not used properly (LSPs try to provide as much static typing as possible based on pre-evaluated nixpkgs, but that usually is not enough). Basically, everything is mostly runtime and configuration (in terms of personal preferences) - dependant, hard/impossible to automate (unless you can think of something, then you're only welcome to contribute and make everyone's lives easier!)

[eclairevoyant]

Er no, we have package tests and NixOS tests, it's documented in the manual.
https://nixos.org/manual/nixpkgs/stable/#var-meta-tests-packages

And half the time it's people not even checking if their PRs eval, so that's not even at the level of writing package tests. We just need more restraint and less self-merges and other rushed merges.

[tukanoidd]

Er no, we have package tests and NixOS tests, it's documented in the manual.
https://nixos.org/manual/nixpkgs/stable/#var-meta-tests-packages

And half the time it's people not even checking if their PRs eval, so that's not even at the level of writing package tests. We just need more restraint and less self-merges and other rushed merges.

My bad, should've looked.

[luiswirth]

This just broke my system...
Is there a workaround?
I tried adding this to my configuration.nix:

  environment.sessionVariables = {
    FONTCONFIG_FILE = "${pkgs.fontconfig.out}/etc/fonts/fonts.conf";
    FONTCONFIG_PATH = "${pkgs.fontconfig.out}/etc/fonts";
  };

but it doesn't help :(

[K900]

The fix should be making its way to channels very soon.

[MrSom3body]

This just broke my system... Is there a workaround? I tried adding this to my configuration.nix:

  environment.sessionVariables = {
    FONTCONFIG_FILE = "${pkgs.fontconfig.out}/etc/fonts/fonts.conf";
    FONTCONFIG_PATH = "${pkgs.fontconfig.out}/etc/fonts";
  };

but it doesn't help :(

You need to remove the .out part so it looks like this:

environment.sessionVariables = {                               
  FONTCONFIG_PATH = "${pkgs.fontconfig}/etc/fonts";            
  FONTCONFIG_FILE = "${pkgs.fontconfig}/etc/fonts/fonts.conf"; 
};                                                             

[K900]

No, you don't, and it's not actually going to help.

[couchpotato007]

This just broke my system... Is there a workaround? I tried adding this to my configuration.nix:

  environment.sessionVariables = {
    FONTCONFIG_FILE = "${pkgs.fontconfig.out}/etc/fonts/fonts.conf";
    FONTCONFIG_PATH = "${pkgs.fontconfig.out}/etc/fonts";
  };

but it doesn't help :(

if you use flakes you can simply use an old flake.lock file until the fix comes. That works for me.

[nakoo]

The fix should be making its way to channels very soon.

Is there a PR tracker related to this?

[eclairevoyant]

You'll have to wait for the next staging-next merge - #338840

[K900]

The fix for most applications should be in nixos-unstable now.

[luiswirth]

This did fix many of my applications. But wezterm is for instance still broken... So this issue is not fully fixed.

[K900]

What is broken about wezterm?

[eclairevoyant]

@luiswirth that's an unrelated issue, follow #336069 instead

[acid-bong]

Is this issue the reason why mono font is (was?) no longer an alias for monospace? I just updated today (using nixos-unstable) and Rofi (which uses mono instead of monospace) now falls back to sans-serif