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

https://discourse.nixos.org/t/2024-05-21-documentation-team-meeting-notes-129/45796/1

@fricklerhandwerk @alejandrosame What do you two think of the following?

Can be checked with nix eval -f <file> --raw.

I simplified the sorting by doing it when I get the list of interpreters.
Getting the "cute" name and aliases are turned into functions that only need knowledge of the interpreter name.

let

  pkgs = import ./. {
    config = { };
    overlays = [ ];
  };
  lib = pkgs.lib;

  inherit (lib.attrsets) attrNames filterAttrs;
  inherit (lib.lists) naturalSort;
  inherit (lib.strings) concatStringsSep;

  isPythonInterpreter = name: (lib.strings.match "(pypy|python)([[:digit:]]*)" name) != null;
  interpreters = naturalSort (
    builtins.filter isPythonInterpreter (builtins.attrNames pkgs.pythonInterpreters)
  );

  getCuteName =
    interpreter:
    let
      cuteName = {
        cpython = "CPython";
        pypy = "PyPy";
      };
      pythonInterpreter = pkgs.${interpreter};
    in
    "${cuteName.${pythonInterpreter.implementation}} ${pythonInterpreter.pythonVersion}";

  getAliases =
    interpreter:
    builtins.attrNames (
      filterAttrs (
        name: value:
        if isPythonInterpreter name then
          value.outPath == pkgs.${interpreter}.outPath && name != interpreter
        else
          false
      ) pkgs
    );

  result = builtins.map (interpreter: {
    inherit interpreter;
    cuteName = getCuteName interpreter;
    aliases = getAliases interpreter;
  }) interpreters;

  toMarkdown =
    data:
    concatStringsSep "" (
      map (interpreter: ''
        | ${interpreter.interpreter} | ${
          concatStringsSep ", " interpreter.aliases or [ ]
        } | ${interpreter.cuteName} |
      '') data
    );
in
''
  | Package | Aliases | Interpeter |
  |---------|---------|------------|
  ${toMarkdown result}
''

I get this markdown output

| Package | Aliases | Interpeter |
|---------|---------|------------|
| pypy27 | pypy, pypy2 | PyPy 2.7 |
| pypy39 | pypy3 | PyPy 3.9 |
| pypy310 |  | PyPy 3.10 |
| python27 | python, python2 | CPython 2.7 |
| python39 |  | CPython 3.9 |
| python310 |  | CPython 3.10 |
| python311 | python3 | CPython 3.11 |
| python312 |  | CPython 3.12 |
| python313 |  | CPython 3.13 |

Rendered

@djacu yeah nice!

@djacu Yeah, and checking for outPath also simplifies the code and makes the alias collection more generic

@djacu wait, how did you get past the Python 2.7 EOL error? We have to evaluate derivations if we want to access outPath, and this is where the allowInsecure assertion will hit.

@djacu wait, how did you get past the Python 2.7 EOL error? We have to evaluate derivations if we want to access outPath, and this is where the allowInsecure assertion will hit.

Meh, so still there's the point that going through outPath is "unreliable": not guaranteed to work at the level we care here, identifying aliases from an evaluated tree.

Is parsing the nixpkgs AST the only truly general solution to discover aliases? (To be clear, it would be out of scope of this PR, but in the end this is the underlying issue that surfaces when traversing pkgs to get aliases).

@djacu wait, how did you get past the Python 2.7 EOL error? We have to evaluate derivations if we want to access outPath, and this is where the allowInsecure assertion will hit.

export NIXPKGS_ALLOW_INSECURE=1 :P

outPath was the one thing I was suspicious of and forgot to test the script without export NIXPKGS_ALLOW_INSECURE=1. If you run it without the value.outPath == pkgs.${interpreter}.outPath expression, you don't need the insecure flag, but it obviously gets the wrong answer. I was mostly trying to figure out if I could simplify the logic. If we can resolve the insecure/outPath issue, I think we have something really nice.

Wow, this got a lot shorter