Bump to ghc 9.6.6 from Nix latest

[morucci]

First tentative to move our Haskell stack to a higher and stable version of the stack. We rely on podenv/hspkgs to provide a nix derivation for a Haskell package set.

Here we attempt a bump to hspkgs for ghc 9.6.4.

[TristanCacqueray]

Hey @morucci , we shouldn't need hspkgs anymore, ghc-9.6 is now well established in nixpkgs. Let's remove that dependency and import the few overrides that are still needed. I think it should mostly work, though we need to fix the paths for the tools like weeder/hlint/fourmolu, which were provided as hspkgs.$tool where nixpkkgs provide them as pkgs.haskellPackages.$tool.

[TristanCacqueray]

I remember doing that, but then a go package failed to build because the infrastructure change a bit. So let's replace the hspkgs import with github:NixOS/nixpkgs/d3780c92e64472e8f9aa54f7bbb0dd4483b98303 (which is the latest from https://github.com/NixOS/nixpkgs/tree/haskell-updates) to avoid updating the non-haskell things. Later we can fix the rest of the build to only pull a single nixpkgs.

In particular, we need to change:

monocle/nix/default.nix

Line 66 in 2246940

hsPkgs = hspkgs.hspkgs.extend haskellExtend;

Into

hsPkgs = latestNixpkgs.haskellPackages.extend haskellExtend;

And then pull the overrides (e.g. gerrit) in

monocle/nix/default.nix

Lines 21 to 22 in 2246940

	# Add monocle and patch broken dependency to the haskell package set
	haskellExtend = hpFinal: hpPrev: {

[morucci]

Hey @TristanCacqueray thanks for the guidance. I'll try to address that path in followup commits inside that PR. I understand that we'll have a bit of delay using nix to benefit latest Haskell tool chains. So yes if 9.6 and its dependencies are now well integrated in nixpkgs then let's try to use it directly.

[TristanCacqueray]

It's not really because of nix, it's actually stackage which govern the set of "stable" versions. I think the default (9.6) should be fine, but you can try a more recent version by replacing pkgs.haskellPackages with pkgs.haskell.ghc910 for example.

Checkout how it was done to pull ghc9.6 when the default was 8.10: https://github.com/podenv/hspkgs/blob/7a46854f28ab9b99c51353c81d5967f1f6fd9a9b/flake.nix#L148

[TristanCacqueray]

It's odd that the formatting changed, are you using?

(use-package nix-mode
  :config
  ;; auto format with nixfmt by default
  (reformatter-define nixfmt-format
    :program "nixfmt")
  (add-hook 'nix-mode-hook 'nixfmt-format-on-save-mode))

With nix profile install nixpkgs#nixfmt

[morucci]

Looks like the effectful requirement in cabal is not compatible anymore because of effectful being 2.3.1 in the Nix set.

[TristanCacqueray]

Should be fine to use according to https://github.com/haskell-effectful/effectful/blob/master/effectful-core/CHANGELOG.md#effectful-core-2310-2024-06-07 and https://github.com/haskell-effectful/effectful/blob/master/effectful/CHANGELOG.md#effectful-2310-2024-06-07 . Let's bump the bound in monocle.cabal

[morucci]

(use-package nix-mode
  :config
  ;; auto format with nixfmt by default
  (reformatter-define nixfmt-format
    :program "nixfmt")
  (add-hook 'nix-mode-hook 'nixfmt-format-on-save-mode))

No, I was using the default nix-mode-format function. Now using your config, I'm getting the autoformat with nixfmt at save, and looks like it reverted to the initial format.

[morucci]

Should be fine to use according to https://github.com/haskell-effectful/effectful/blob/master/effectful-core/CHANGELOG.md#effectful-core-2310-2024-06-07 and https://github.com/haskell-effectful/effectful/blob/master/effectful/CHANGELOG.md#effectful-2310-2024-06-07 . Let's bump the bound in monocle.cabal

This seems to cause issue with withEffToIO. Let's see the CI result.

[TristanCacqueray]

Almost there, try to add SeqUnlift to the first param of withEffToIO

[TristanCacqueray]

That reformat seems wrong, you'll have to manually discard all these chunks.

[morucci]

Just did a drop of both formating commits. Then re-run the nixfmt (this is the last commit here). That looks fine I guess.

[TristanCacqueray]

One last error!

src/CLI.hs:32:41: error:
    Module ‘Options.Applicative.Help.Pretty’ does not export ‘string’
   |
32 | import Options.Applicative.Help.Pretty (string)
   |                                         ^^^^^^

[TristanCacqueray]

It looks like the hash function changed, you'll have to update the expected value in the test.

[TristanCacqueray]

For fourmolu, here is some elisp for you:

(use-package haskell-mode
  :config
  ;; ensure run-haskell uses the simplest ghci subprocess
  (setq-default haskell-process-type 'auto)
  (add-hook 'haskell-mode-hook 'haskell-auto-insert-module-template)
  ;; auto format with fourmolu by default
  (reformatter-define fourmolu-format
    :program "fourmolu"
    :args `("--stdin-input-file" ,buffer-file-name))
  (add-hook 'haskell-mode-hook 'fourmolu-format-on-save-mode))

Thanks calling fourmolu-format-buffer I see the buffer to be formatted. However C-x C-s does not trigger fourmolu. Does that work for you ?

[TristanCacqueray]

well done! Glad to see hspkgs is out of the equation 🥳