Eval report LGTM: https://github.com/NixOS/nixpkgs/actions/runs/21031311391?pr=480141#summary-60468411383

The last update (#464869) showed slightly faster numbers, implying Nixpkgs itself has grown slower in the last two months, although GHA's performance at the time is also a variable when comparing against the last bump.

The real purpose of the report is to compare how different versions of nix handle the current evaluation. In this regard, we can see nixVersions.git and nixVersions.nix_2_32 in the lead for linux, although nixVersions.nix_2_31 is slightly faster on darwin. Recent lix versions have comparable performance, except on x86_64-linux.

When CI tries to build the docs tests it failes with a weird fetcher error:

https://github.com/NixOS/nixpkgs/actions/runs/21031985076/job/60470081430?pr=480141#step:10:47

error: Cannot build '/nix/store/f4h6638q6m0ardx9y875jga2ab67qbaq-source-salted-bdkkpviaq89r.drv'.
       > aiohttp.client_exceptions.ConnectionTimeoutError: Connection timeout to host https://www.gnu.org/software/coreutils/manual/html_node/cat-invocation.html

nix-build-uncached: nix build failed: exit status 1

The URL it's fetching is accessible: https://www.gnu.org/software/coreutils/manual/html_node/cat-invocation.html

Testing locally I can reproduce the issue both on master and this PR.

To be extra conservative, I used nix-build-uncahced from the previous pinned nixpkgs:

$ git worktree add ../pinned --detach ee09932cedcef15aaf476f9343d1dea2cb77e261
$ fish_add_path -a (nix-build ../pinned -A nix-build-uncached --no-out-link)/bin

Then I tried to build ci's manual-nixpkgs-tests attr on master and this PR:

$ nix-build-uncached ci -A manual-nixpkgs-tests

I also tried building the test against the old pinned nixpkgs:

$ nix-build --dry-run ci --arg nixpkgs ../pinned -A manual-nixpkgs-tests

The workflow job runs when matrix.builds contains manual-nixpkgs-tests:

Which is true for the aarch64-linux run:

So maybe me testing locally on an x86_64-linux system is throwing me off.

Regardless, the fact that this job fails in my local testing, fails in this PR, but is successful for other PRs currently open is very confusing.

(I've temporarily pushed #482470 here for testing purposes, will drop shortly)

EDIT:

Interestingly the "test" pull_request workflow didn't run (maybe because of the merge conflicts?), so I wasn't able to validate this PR against #482470.

However, the pull_request_target did run and this time it had enough base objects to fetch the ci-pin-bump commit. Therefore, CI passes(!)

This doesn't mean #482470 is irrelevant; it is still likely that future ci-bumps will run into the same issue with being unable to fetch the ci-pin-bump commit in a shallow clone. It may also be an issue again here, once rebased.

I realize the rest of @NixOS/nixpkgs-ci wasn't initially pinged for review, because the treewide change touched too many files leading to too many reviewer requests.

Regardless, I'm going to merge this now, to avoid @dyegoaurelio needing to rebase for merge conflicts again.

I expect the backport will need to be done manually, as the treewide reformat will likely be different on 25.11.

The base branch has a hidden formatting conflict in pkgs/servers/x11/xorg/default.nix:

diff --git a/pkgs/servers/x11/xorg/default.nix b/pkgs/servers/x11/xorg/default.nix
index 0c1225237..803389ee8 100644
--- a/pkgs/servers/x11/xorg/default.nix
+++ b/pkgs/servers/x11/xorg/default.nix
@@ -219,7 +219,7 @@
   xwd,
   xwininfo,
   xwud,
-# keep-sorted end
+  # keep-sorted end
 }:
 _:
 {

keep-sorted has likely been added since this PR was last rebased, and its comments are now formatted differently with this PR.