flake: update all inputs

[0xda157]

includes numtide/treefmt-nix#430 which fixes the ci failing

---

• I certify that I have the right to submit this contribution under the MIT license
• Commit messages adhere to Stylix commit conventions
• Theming changes adhere to the Stylix style guide
• Changes have been tested locally
• Changes have been tested in testbeds
• Each commit in this PR is suitable for backport to the current stable branch

[trueNAHO]

LGTM.

I think we need to re-run CI several times for it to pass because it times out due to uploading all the new artifacts to the cache.

This is good to merge for me once CI passes.

[trueNAHO]

I think we need to re-run CI several times for it to pass because it times out due to uploading all the new artifacts to the cache.

As of commit 647bb8d ("vscode: add more themed items (#1957)"), our aarch64-linux runner is nearly out of space:

2025-11-04T14:13:07.1872772Z Filesystem      Size  Used Avail Use% Mounted on
2025-11-04T14:13:07.1873087Z /dev/root        72G   71G  868M  99% /

This PR causes CI to fail because we are finally running out of space:

System.IO.IOException: No space left on device : '/home/runner/actions-runner/cached/_diag/Worker_20251108-151317-utc.log'

-- https://github.com/nix-community/stylix/actions/runs/19182118802/attempts/2

Here is the relevant code:

stylix/.github/workflows/check.yml

Lines 37 to 46 in 647bb8d

	# TODO: Lock this Action to a release tag once commit [1] ("fix: relocate
	# TMPDIR to /mnt to improve Nix installer compatibility") is part of a
	# release, resolving [2] ("does not work with cachix install nix action").
	#
	# [1]: https://github.com/wimpysworld/nothing-but-nix/pull/25
	# [2]: https://github.com/wimpysworld/nothing-but-nix/issues/24
	- uses: wimpysworld/nothing-but-nix@10c936d9e46521bf923f75458e0cbd4fa309300d # yamllint disable-line rule:line-length
	if: runner.os == 'Linux'
	with:
	hatchet-protocol: holster

[trueNAHO]

The problem is not the protocol not being aggressive enough. The problem is that it does nothing on aarch:

stylix/.github/workflows/check.yml

Line 44 in 647bb8d

if: runner.os == 'Linux'

IIRC, we excluded aarch because it was causing non-trivial problems. Seems like we have to resolve those now.

[0xda157]

As far as I can tell runner.os doesn't specify the runner architecture, only the os. Are there docs that point to the contrary?

[trueNAHO]

As far as I can tell runner.os doesn't specify the runner architecture, only the os. Are there docs that point to the contrary?

Commit e334b30 ("ci: check: add wimpysworld/nothing-but-nix Action for larger Nix store") simply replicates nothing-but-nix's internal guarding implementation, as mentioned in wimpysworld/nothing-but-nix#33 (comment).

@Defelo and @MattSturgeon, is Defelo/nixpkgs-review-gha#26 currently the best way for getting nothing-but-nix to work on darwin? What would an upstream PR look like? Considering we already pin a specific version, I would not mind pinning a patched mini-fork.

[0xda157]

Defelo and MattSturgeon, is Defelo/nixpkgs-review-gha#26 currently the best way for getting nothing-but-nix to work on darwin? What would an upstream PR look like? Considering we already pin a specific version, I would not mind pinning a patched mini-fork.

Darwin CI seems to be doing fine currently, I don't see why we need to reclaim more space there. The issue with CI is the runner seems to disconnect, which is likely caused by being starved of disk space, ram or cpu resources.

[danth]

This PR causes CI to fail because we are finally running out of space

What if we set min-free and max-free to trigger some garbage collections during the run?

[danth]

Also, NixOS/nix@ecdda57 is now released, so we could potentially switch to using a plain nix flake check which would be able to skip already built packages, at least after the first run.

[trueNAHO]

This PR causes CI to fail because we are finally running out of space

Not resolving this issue before the release of the 25.11 stable branch, would complicate our release-25.11 branch-off process.

This PR causes CI to fail because we are finally running out of space

What if we set min-free and max-free to trigger some garbage collections during the run?

Could be worth a try if a proper solution is not possible.

Also, NixOS/nix@ecdda57 is now released, so we could potentially switch to using a plain nix flake check which would be able to skip already built packages, at least after the first run.

Yes. Although, it would be better to clean this up in a follow-up PR. For reference, commit 29dc3dd ("gnome: update to GNOME 49 (#1966)") uses version 2.32.1.

[MattSturgeon]

includes numtide/treefmt-nix#430 which fixes the ci failing

You may also want to consider programs.biome.validate.enable = false, to avoid potential hash-mismatch issues reoccurring.

Validating your biome settings against a schema is unnecessary, if you test that the formatter actually works when updating its settings.

[0xda157]

we could switch to using namespace runners or buildbot, both provided by nix-community. nixvim and home-manager both have switched to buildbot.

[danth]

Switching to buildbot seems like a good idea and would save us time setting up our own CI. I assume things are also cached on the build server between runs making everything a lot faster.

[trueNAHO]

we could switch to using namespace runners or buildbot, both provided by nix-community. nixvim and home-manager both have switched to buildbot.

Considering Stylix consumes considerably more CI time than most other Nix projects 1 2 3 (maybe second to Nixpkgs), would we overload CI usage? IIRC, Stylix already runs on custom nix-community runners.

[danth]

Considering Stylix consumes considerably more CI time than most other Nix projects 1 2 3 (maybe second to Nixpkgs), would we overload CI usage?

This is definitely something to consider. Assuming packages can be cached locally on the build machine, our main consumption would be evaluation time. We could reduce this later by simplifying the testbed machines, for example using a more lightweight desktop environment, disabling documentation packages, and so on.

IIRC, Stylix already runs on custom nix-community runners.

AFAIK we're using standard GitHub hosted runners:

stylix/.github/workflows/check.yml

Lines 23 to 30 in b68e822

	- name: aarch64-linux
	runs-on: ubuntu-24.04-arm
	- name: aarch64-darwin
	runs-on: macos-15
	- name: x86_64-linux
	runs-on: ubuntu-24.04
	- name: x86_64-darwin
	runs-on: macos-15-intel

[trueNAHO]

CI is passing again in commit 2877806 after several manual retries. Maybe CI consistently passes now that the cache has been properly populated.

The CI of commit 727785e passed on the first retry. If the CI of commits 3e2c930 and 961948b both also pass on the first retry, then we can assume that CI is reliably working again.

[0xda157]

CI is passing again in commit 2877806 after several manual retries. Maybe CI consistently passes now that the cache has been properly populated.

This seems likely, but our ci should still be functional when dealing with a large flake update. I'd still like to pursue using buildbot.