nixos/i18n: don't build all supportedLocales by default

[SuperSandro2000]

Description of changes

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/)
• 22.11 Release Notes (or backporting 22.05 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.

[gebner]

This broke any custom extraLocaleSettings (like LC_PAPER or LC_TIME etc.).

[gravndal]

Shouldn't this have been announced here: https://discourse.nixos.org/t/breaking-changes-announcement-for-unstable/17574?

Either way, wouldn't it make more sense to default to something like:

    supportedLocales = with config.i18n; lib.unique (
      (lib.mapAttrsToList (n: v: v + "/UTF-8")
        (lib.filterAttrs (n: v: n != "LANGUAGE") extraLocaleSettings)
      ) ++ [ (defaultLocale + "/UTF-8") ]
    );

[SuperSandro2000]

Shouldn't this have been announced here: discourse.nixos.org/t/breaking-changes-announcement-for-unstable/17574?

I didn't even know that thread existed.

Either way, wouldn't it make more sense to default to something like:

Please do a PR with that.

This broke any custom extraLocaleSettings (like LC_PAPER or LC_TIME etc.).

Presumable only when they are set to something other than your display language?

[gravndal]

#179194

[gebner]

This broke any custom extraLocaleSettings (like LC_PAPER or LC_TIME etc.).

Presumable only when they are set to something other than your display language?

That's their only use, yes. They default to the display language, so you don't need the override if it's the same.

[Artturin]

#179607

[Artturin]

#179486

[sternenseemann]

I think unconditionally generating C.UTF-8/UTF-8 is probably also a good idea.

[Profpatsch]

Either way, wouldn't it make more sense to default to something like:

Please do a PR with that.

No, that’s not how that works.

You can’t break other people’s systems and then tell them to fix it themselves.

[Profpatsch]

So we should revert this and reevaluate.

[Profpatsch]

Like, for example, not generating C.UTF-8 by default shows that there is a fundamental gap in understanding here.

[SuperSandro2000]

No, that’s not how that works.

If someone is presenting me a working code solution to a problem I think it is very reasonable to ask them to create a PR with that. The other option I would have would be to commit it as my own and then ask the same person to test their own code which is a bit weird.

You can’t break other people’s systems and then tell them to fix it themselves.

That is wrong. Every change has the potential to break something somewhere in some combination of the infinity of potential system combinations. We can only consider so much when doing a change. That's why I wrote a changelog entry because I knew before that if people relied silently on other locales their setup would break. There is just no other way to not break this. I and two reviewers didn't think about extraLocaleSettings or eo and only almost two weeks later we noticed that some things break.

Also my time is not infinite and between real life and other work and projects I might only find time to do this in a few days in which the fix could have been already be commited, merged and be on unstable.

So we should revert this and reevaluate.

Instead of hastily reverting changes and later reverting the revert, we should first consider if a small addition/change of code fixes those use cases or people need to adopt their config and explicitly mention a locale.

Like, for example, not generating C.UTF-8 by default shows that there is a fundamental gap in understanding here.

First of all, I know how locales work and the special meaning behind c.utf-8.

Then this must have been broken for a good amount of time in the minimal profile which is used in containers similar to environment.noXlibs.

---

Now back to fixing this instead of doing discussions.

First of all a big thanks to @gravndal for the two PRs fixing the situation when more complex settings are used. I'll really appreciate it.

Also regarding to #179486 : This must have been always broken but since we generated all locales by default this was never noticed. So to say we unconvered more bugs with this change that where always there.

Then why I didn't notice this while testing: I had C.UTF-8 set in i18n.supportedLocales so I couldn't. Happens when your configuration is multiple thousand lines long. Going to fix that in a PR right now.

[Profpatsch]

That is wrong. Every change has the potential to break something somewhere in some combination of the infinity of potential system combinations.

Given that this PR has neither any description, nor a commit description, I would not assume much thought has been put into possible breakage.

[dasJ]

It's called "unstable" for a reason…

[SuperSandro2000]

I wanted to write that, too. 😂

Yes, breakages can happen even if we all try our best to avoid them.

---

Given that this PR has neither any description, nor a commit description

yeah, I should do them more often but usually I just skip them because I don't read them that often when reviewing either.

---

C.UTF-8 fixup is in #180513

[Artturin]

#180935

[SuperSandro2000]

#180935

Could be #180368

[JonathanReeve]

If this is what's causing #179607, please revert this. It's causing my system to fail to build in my language, and everyone else who also uses Nixos in that language.

[SuperSandro2000]

I don't think that is good way to move this forward. I'd rather add an exception for that language instead of reverting improvements which where much needed and everyone benefits of.

[Artturin]

Another issue #183960

[gravndal]

@Artturin I don't think that's actually caused by this.

From some limited testing, it seems that if your current configured locales don't match the new supported locales reliably triggers those warnings.

A situation likes this can naturally occur if you're removing a locale from your extraLocaleSettings, e.g. I can trigger it by removing my LC_TIME=en_DK.UTF-8 (the only location where I reference this specific locale), and then do a rebuild switch.

This happens (I think) because my current shell sessions still exports the, now old, LC_TIME.

Edit: though, actually, that's just the warnings, not the error...
Edit2+: yeah, that's probably just a misconfigured locale, en_AT.UTF-8 isn't included with glibc.

Also:

 - channels(tmn): `"nixos-22.05-22.05.1806.e732e1fdbf7"`
 - channels(root): `"nixos-21.11.337967.573603b7fdb"`

This wasn't ever backported was it?