ci: bump nixpkgs snapshot to f4f69867 - #1715

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7f787e9155

The change itself is clean. Regeneration is genuine — running the generator at head leaves the tree untouched, so the three flakes are byte-identical to generator output and nothing was hand-edited. The diff is the hash and only the hash: four one-line hunks, no reformatting, no Node version string, no input or output change. No stale copies — grepping the tree for the old hash returns zero occurrences.

I verified the version-invariance claim rather than taking it: nix isn't available here, so I read pkgs/development/web/nodejs/v{22,24,26}.nix at both pins — 22.23.2 / 24.19.0 / 26.7.0 at each. Also worth noting CI enforces it independently: the nix-flakes job asserts the exact version strings, so a wrong claim would fail rather than slip through.

f4f69867 is real, dated 2026-08-25, and is exactly the current nixos-26.05 channel revision — channels.nixos.org/nixos-26.05/git-revision returns that SHA. The old pin is a strict ancestor on the same branch, 75 commits and about two days back. No channel crossing.

Two convention items, both in the PR metadata:

The title has no <topic>: prefix. CONTRIBUTING.md:170-184 wants <topic>: <short description>, and the squash lands as <PR title> (#NNN) — the same section says not to fix it in the merge box, so it has to change here. Your own branch commit already has the right form (ci: bump nixpkgs snapshot). ci: bump nixpkgs snapshot to f4f69867 fits well inside the 72-char limit.

Trailer order is inverted. The body runs prose → bare session link → Changelog: none, but Changelog: has to be the last section before the trailer block, and session links are part of that block. Swap the last two. Content is right — a build-tooling pin bump changes no behaviour, and correctly no changelog/unreleased/1715.md was added. No duplicate trailers this time.

Gates: npm test 3393/0, tsc --noEmit exit 0, .github/ untouched (correctly — the workflow embeds the Node versions, not the nixpkgs hash, and pins no nixpkgs of its own).

Approving. Both findings are fixed — the title is now ci: bump nixpkgs snapshot to f4f69867 and conforms, and Changelog: none sits last before the session link. No duplicate trailers.

The delta is merge-only: the PR's own four files are byte-identical at both heads, and a110f069e..fc8e99359 is still exactly +4/−4. Since main brought in #1710, I re-ran the generator at the merged head — tree stays clean, so the merge didn't invalidate generated output. Four hunks are still only the hash; old hash still has zero occurrences anywhere. npm test 3396/0 (+3 = #1710's three new cbase32 proofs), tsc --noEmit exit 0.