nixos/fish: generate autocompletions from man pages

[hyperfekt]

Motivation for this change

#52464, but fixed.
Any colliding completions will override just as the system path does, but we prevent generating collision warnings for completions that are actually identical by not writing the source path into the completion file.
Also we have a small test now that should cover the effect of this PR.

Things done

• Tested using sandboxing (nix.useSandbox on NixOS, or option sandbox in nix.conf on non-NixOS)
• Built on platform(s)
  • NixOS
  • macOS
  • other Linux distributions
• Tested via one or more NixOS test(s) if existing and applicable for the change (look inside nixos/tests)
• Tested compilation of all pkgs that depend on this change using nix-shell -p nox --run "nox-review wip"
• Tested execution of all binary files (usually in ./result/bin/)
• Determined the impact on package closure size (by running nix path-info -S before and after)
• Assured whether relevant documentation is up to date
• Fits CONTRIBUTING.md.

---

[samueldr]

Here's the actual diff (excluding test), compared to the state before the revert:

(Made this by squashing the revert commit and a commit of these changes together.)

diff --git a/nixos/modules/programs/fish.nix b/nixos/modules/programs/fish.nix
index 03d6c26c8c8..1fe59407943 100644
--- a/nixos/modules/programs/fish.nix
+++ b/nixos/modules/programs/fish.nix
@@ -182,11 +182,38 @@ in
 
     environment.etc."fish/generated_completions".source =
       let
+        patchedGenerator = pkgs.stdenv.mkDerivation {
+          name = "fish_patched-completion-generator";
+          srcs = [
+            "${pkgs.fish}/share/fish/tools/create_manpage_completions.py"
+            "${pkgs.fish}/share/fish/tools/deroff.py"
+          ];
+          unpackCmd = "cp $curSrc $(basename $curSrc)";
+          sourceRoot = ".";
+          patches = [ (pkgs.writeText "fish_completion-generator_patch" ''
+            --- a/create_manpage_completions.py
+            +++ b/create_manpage_completions.py
+            @@ -776,8 +776,6 @@ def parse_manpage_at_path(manpage_path, output_directory):
+
+                         built_command_output.insert(0, "# " + CMDNAME)
+
+            -            # Output the magic word Autogenerated so we can tell if we can overwrite this
+            -            built_command_output.insert(1, "# Autogenerated from man page " + manpage_path)
+                         # built_command_output.insert(2, "# using " + parser.__class__.__name__) # XXX MISATTRIBUTES THE CULPABILE PARSER! Was really using Type2 but reporting TypeDeroffManParser
+
+                         for line in built_command_output:
+          '') ];
+          dontBuild = true;
+          installPhase = ''
+            mkdir -p $out
+            cp * $out/
+          '';
+        };
         generateCompletions = package: pkgs.runCommand
-          "${package.name}-fish-completions"
+          "${package.name}_fish-completions"
           (
             {
-              src = package;
+              inherit package;
               nativeBuildInputs = [ pkgs.python3 ];
               buildInputs = [ pkgs.fish ];
               preferLocalBuild = true;
@@ -196,13 +223,14 @@ in
           )
           ''
             mkdir -p $out
-            if [ -d $src/share/man ]; then
-              find $src/share/man -type f | xargs python ${pkgs.fish}/share/fish/tools/create_manpage_completions.py --directory $out >/dev/null
+            if [ -d $package/share/man ]; then
+              find $package/share/man -type f | xargs python ${patchedGenerator}/create_manpage_completions.py --directory $out >/dev/null
             fi
           '';
       in
         pkgs.buildEnv {
           name = "system-fish-completions";
+          ignoreCollisions = true;
           paths = map generateCompletions config.environment.systemPackages;
         };
 
diff --git a/nixos/tests/all-tests.nix b/nixos/tests/all-tests.nix
index 65227857a38..2ddb54bcc3d 100644
--- a/nixos/tests/all-tests.nix
+++ b/nixos/tests/all-tests.nix
@@ -74,6 +74,7 @@ in
   ferm = handleTest ./ferm.nix {};
   firefox = handleTest ./firefox.nix {};
   firewall = handleTest ./firewall.nix {};
+  fish = handleTest ./fish.nix {};
   flannel = handleTestOn ["x86_64-linux"] ./flannel.nix {};
   flatpak = handleTest ./flatpak.nix {};
   fsck = handleTest ./fsck.nix {};

[dotlambda]

Thanks a lot for this!

[dotlambda]

I can confirm that this PR produces no errors with my current system configuration. 🎉

[dotlambda]

You forgot to git add the patch.

[hyperfekt]

You forgot to git add the patch.

Argh. Fixed.

[gebner]

Also resolves the issue with the collisions for me.

[rnhmjoj]

I was wondering if the generation could be implemented in a way such that the completion are in a single derivation. Currently I'm seeings hunders of _fish-completion derivations being built: it's not a big deal but it's quite noisy. Also there is no option to disable the generation, right?

[hyperfekt]

This is fundamentally a tradeoff between noise and performance on incremental package additions to the environment - creating completions for everything in Python takes a significant amount of time which I did not want to burden every new package installation with.
There is indeed no option to disable the generation because the generated completions shouldn't have a negative effect given they have a low priority.
If others find as well that the noise when initially building a system is too high to tolerate it could be worth the UI complexity of adding a toggle. If not, you might just (which reminds me we don't really have a good mechanism for that) want to patch the module to disable the generation.

[rnhmjoj]

creating completions for everything in Python takes a significant amount of time which I did not want to burden every new package installation with.

I didn't think of that, you are right: a single derivation would have to be rebuilt every time and it's definitely worse.

If others find as well that the noise when initially building a system is too high to tolerate it could be worth the UI complexity of adding a toggle.

Probably not, as I said it's not a big deal.

Thank you for replying and sorry for the trouble.