ci(nix): a stale bun.nix can no longer reach main unnoticed
ci / rust-arm64 (pull_request) Successful in 1m24s
apple / swift (pull_request) Successful in 1m32s
apple / screenshots (pull_request) Skipped
ci / web (pull_request) Failing after 2m19s
ci / docs-site (pull_request) Successful in 1m17s
android / android (pull_request) Successful in 3m10s
ci / bun-nix (pull_request) Successful in 1m6s
nix / flake (pull_request) Failing after 1m6s
ci / rust (pull_request) Failing after 6m2s

Moving the bun packages to bun2nix (1db8f763) removed the aggregate deps hash
that used to go stale, but not the second, quieter way a generated file rots.
bun2nix is a pure function of bun.lock, and `bun.nix` is committed alongside it
— but it regenerates ONLY from a local `bun install` that runs lifecycle
scripts. It does not regenerate under `bun install --ignore-scripts`, which is
what every bun install in CI uses, and it cannot regenerate on a merge or
rebase, where git carries someone else's lockfile change past a `bun.nix`
generated before it and reports no conflict.

That is not theoretical. web/bun.nix went onto main holding brace-expansion@5.0.7
(plus two nested entries the `^5.0.8` override had already collapsed) while
web/bun.lock said 5.0.8: the override landed in ec9aa415, the bun2nix branch had
generated bun.nix off the pre-override lock, and the merge kept both. The Nix
build fetches node_modules strictly from bun.nix, so the derivation's offline
`bun install` was asked for a tarball its store cache did not contain and
`nix build .#punktfunk-web` failed. It stayed broken for 553 commits
(2026-07-27 -> 2026-08-05) and was closed by accident, when an unrelated
advisory bump in b79d90b4 happened to rerun a real `bun install`.

Nothing caught it because nothing could: the string "nix" appeared in exactly
one workflow file, and only in a comment about bun2nix breaking a Windows step.

  * scripts/ci/check-bun-nix.sh regenerates each bun.nix from its committed
    bun.lock and diffs, with `--fix` to rewrite them. It reads which packages to
    check out of packages.nix's own `bunNix = src + ...` lines rather than a
    hardcoded list, so a third bun package is covered the day it is added, and
    an empty list is a hard error — a gate that checks nothing passes exactly
    like a clean tree. It also enforces the bun2nix version pin that flake.nix
    and the README only stated: bun.nix has no schema stability across releases,
    so the flake input ref and both package.json devDependencies must agree, and
    generation always uses the pinned version rather than a floating `bunx`.

  * ci.yml gains a `bun-nix` job running it. Deliberately unfiltered and not in
    nix.yml: it needs no Nix, takes under a minute, and the whole point is that
    this drift arrives through commits that look unrelated to Nix.

  * nix.yml is new and is the first CI that evaluates the flake at all —
    `nix flake check --no-build` over every output, then a real build of
    punktfunk-web + punktfunk-scripting, which are the two derivations whose
    inputs churn and the only ones that do not compile Rust. That leg is the
    end-to-end proof the drift gate cannot give: it catches a tarball the
    registry no longer serves, or the codegen going quietly message-less. It is
    path-filtered, has a workflow_dispatch opt-in for the hour-long Rust
    packages, and keeps `pull_request` — flatpak.yml shipped push-only and let
    manifest breakage reach main invisibly for weeks.

The script is POSIX sh (shellcheck-clean, exercised under dash, which is the
CI container's /bin/sh) and avoids process substitution, the parse-time failure
that silently disabled ci.yml's shader SPIR-V gate for 35 commits.
This commit is contained in:
2026-08-06 15:28:14 +02:00
parent 00d4026054
commit fb707b4956
4 changed files with 364 additions and 8 deletions
+28
View File
@@ -275,3 +275,31 @@ jobs:
run: bun run build
- name: Typecheck
run: bun run lint
# web/bun.nix and sdk/bun.nix are GENERATED from their bun.lock (bun2nix) and committed; the Nix
# build fetches node_modules from nothing else. They regenerate only on a local `bun install` that
# runs lifecycle scripts — never under CI's `--ignore-scripts`, and never on a merge or rebase,
# which happily carries a lockfile change past a bun.nix generated before it. That is not
# theoretical: web/bun.nix sat stale on main for 553 commits (2026-07-27 → 2026-08-05) with
# `nix build .#punktfunk-web` broken, and was repaired only by accident when an advisory bump
# happened to rerun a real `bun install`.
#
# Deliberately UNFILTERED and in ci.yml rather than nix.yml: it needs no Nix, takes well under a
# minute, and the whole point is that the drift arrives through commits that look unrelated to
# Nix. The Nix-toolchain gates (flake eval + building the bun packages) live in nix.yml.
bun-nix:
runs-on: ubuntu-24.04
container:
image: oven/bun:1
timeout-minutes: 15
steps:
# oven/bun ships neither git nor a real node, and the slim base has no CA bundle —
# actions/checkout needs all three (see the web job).
- name: Install git + node + CA certs
run: apt-get update && apt-get install -y --no-install-recommends ca-certificates git nodejs
- uses: actions/checkout@v4
# Regenerates each bun.nix from its committed bun.lock and diffs, and checks that the
# bun2nix version pin agrees across flake.nix and both package.json files (bun.nix has no
# schema stability across bun2nix releases). Fix with: scripts/ci/check-bun-nix.sh --fix
- name: bun.nix drift gate
run: sh scripts/ci/check-bun-nix.sh
+132
View File
@@ -0,0 +1,132 @@
# Nix packaging gate. Until this existed, NOTHING in CI ever evaluated flake.nix: the word "nix"
# appeared in exactly one workflow file, and only in a comment about bun2nix breaking a Windows
# step. Every Nix regression therefore reached main invisibly and was found by hand on a Nix box —
# `nix build .#punktfunk-web` was broken for 553 commits before anyone noticed (see the bun-nix job
# in ci.yml for that story).
#
# Two tiers, because a full `nix flake check` builds the whole Rust workspace with crane and would
# run for an hour on every push:
#
# * eval — `nix flake check --no-build`: instantiates every package, app, check, devShell and
# the NixOS module without building them. Catches the failures that actually happen to
# this flake — a renamed file, a callPackage argument that no longer exists, a syntax
# error, a package attribute dropped from packages.nix.
# * bun — actually BUILDS punktfunk-web + punktfunk-scripting. These are the two derivations
# whose inputs churn constantly (every dependency bump moves a lockfile) and they cost
# minutes, not hours, because neither compiles Rust. This is the end-to-end proof that
# the generated bun.nix really does materialise a working node_modules offline — it
# covers what the ci.yml drift gate cannot, e.g. a tarball the registry no longer
# serves, or the codegen going quietly message-less (see packages.nix's inlang note).
#
# The Rust packages (punktfunk-host, punktfunk-client) and punktfunk-gamescope are NOT built here.
# They are the expensive ones and their inputs are already gated by the `rust` job in ci.yml; build
# them by hand on a Nix box, or with the `build-rust` dispatch input below.
#
# ⚠ pull_request is deliberately present. flatpak.yml shipped with push-only triggers and manifest
# breakage reached main invisibly for weeks — do not "simplify" this workflow by dropping it.
# ⚠ The two path lists are duplicated on purpose: a YAML anchor would be tidier, but Gitea's
# workflow parser is not a place to bet on anchor support. Keep them in step by hand.
name: nix
concurrency:
group: ${{ github.workflow }}-${{ github.ref }}
cancel-in-progress: true
on:
push:
branches: [main]
paths:
- "flake.nix"
- "flake.lock"
- "packaging/nix/**"
- "**/bun.lock"
- "**/bun.nix"
- "**/package.json"
- "Cargo.lock"
- "Cargo.toml"
- "rust-toolchain.toml"
- ".gitea/workflows/nix.yml"
- "scripts/ci/check-bun-nix.sh"
pull_request:
paths:
- "flake.nix"
- "flake.lock"
- "packaging/nix/**"
- "**/bun.lock"
- "**/bun.nix"
- "**/package.json"
- "Cargo.lock"
- "Cargo.toml"
- "rust-toolchain.toml"
- ".gitea/workflows/nix.yml"
- "scripts/ci/check-bun-nix.sh"
workflow_dispatch:
inputs:
build-rust:
description: "Also build punktfunk-host + punktfunk-client (slow: full Rust workspace)"
type: boolean
default: false
jobs:
flake:
runs-on: ubuntu-24.04
container:
# Official Nix image (Docker Hub, like this fleet's other WAN images: oven/bun:1,
# fedora:43, node:22-bookworm). It ships nix and little else.
image: nixos/nix:latest
timeout-minutes: 90
env:
# The image defaults to stable Nix with the experimental features off; the flake needs both.
# Set at job level so every step — including the `nix profile install` below — sees it.
NIX_CONFIG: "experimental-features = nix-command flakes"
steps:
# actions/checkout is a JS action and needs node; a plain `run:` step executes through the
# container shell, so this must come BEFORE the checkout (same ordering, and the same
# reason, as flatpak.yml's fedora job).
- name: node + git for the JS actions
run: nix profile install nixpkgs#nodejs nixpkgs#git
- uses: actions/checkout@v4
# Nix reads the flake through libgit2 and refuses a checkout owned by another uid
# ("detected dubious ownership"), which is the normal case for a container job.
- name: Trust the checkout
run: git config --global --add safe.directory "$PWD"
# First-run diagnostics — cheap, and the difference between "the gate found a real problem"
# and "the runner had no disk" is otherwise a guess.
- name: Environment
run: |
nix --version
df -h /nix /tmp || true
# Evaluates + instantiates every flake output without building any of it.
- name: nix flake check (eval only)
run: nix flake check --no-build --show-trace
# The bun packages, built for real. This is the leg that would have caught the stale
# web/bun.nix end to end: the derivation's offline `bun install` runs against a store cache
# built strictly from bun.nix, so a lockfile that cache does not cover fails here.
- name: Build the bun packages
run: nix build --print-build-logs .#punktfunk-web .#punktfunk-scripting
# Both launchers exec pkgs.bun from the store; confirm they were produced and are real entry
# points rather than dangling wrappers.
- name: Smoke the built launchers
run: |
set -eu
web=$(nix path-info .#punktfunk-web)
scripting=$(nix path-info .#punktfunk-scripting)
test -x "$web/bin/punktfunk-web-server" || { echo "no punktfunk-web-server in $web" >&2; exit 1; }
test -x "$scripting/bin/punktfunk-scripting" || { echo "no punktfunk-scripting in $scripting" >&2; exit 1; }
# The console must be the bun bundle, not a node one — the same assertion packages.nix
# makes at build time, re-checked on the installed output.
grep -q 'Bun\.serve' "$web/share/punktfunk-web/.output/server/index.mjs" \
|| { echo "installed console is not a bun bundle" >&2; exit 1; }
echo "bun packages OK: $web $scripting"
# Opt-in only: the full Rust workspace through crane, which is the hour-long leg.
# `github.event.inputs.*` (string) rather than `inputs.*` — the portable spelling.
- name: Build the Rust packages (dispatch opt-in)
if: ${{ github.event.inputs.build-rust == 'true' }}
run: nix build --print-build-logs .#punktfunk-host .#punktfunk-client
+23 -8
View File
@@ -234,18 +234,33 @@ The shell exports `PF_FFVK_VULKAN_INCLUDE` (Vulkan headers for pf-ffvk bindgen)
already in the lockfile, via a generated-and-committed `bun.nix` (`web/bun.nix`, `sdk/bun.nix`).
There is **no aggregate deps hash to bump** — the previous design put `bun install` in a
fixed-output derivation whose single `outputHash` silently went stale on every lockfile change and
broke the build. `bun.nix` regenerates itself: `bun2nix` is a devDependency of both packages and
runs on every `bun install` (web's `postinstall`; the SDK's `prepare`, since sdk/ is the
*published* `@punktfunk/host` package and a `postinstall` would then fire on consumers' installs).
Regenerate by hand with `cd web && bunx bun2nix -o bun.nix` if a lockfile is ever edited directly.
broke the build. `bun2nix` is a devDependency of both packages and regenerates `bun.nix` on every
`bun install` (web's `postinstall`; the SDK's `prepare`, since sdk/ is the *published*
`@punktfunk/host` package and a `postinstall` would then fire on consumers' installs).
The `@unom` scope needs no special handling: `web/bun.lock` records those tarballs' full
`https://git.unom.io/api/packages/unom/npm/…` URLs and the registry is read-public (the same
anonymous pull CI's rpm/deb builds do).
> ⚠ **`bun.nix` has no schema stability across bun2nix versions.** The flake input is pinned
> (`github:nix-community/bun2nix?ref=2.1.2`) and the npm devDependency is pinned to the *same*
> exact version in `web/package.json` + `sdk/package.json`. Move both together, then rerun
> `bun install` in `web/` and `sdk/` to regenerate.
> **That devDependency hook is a convenience, NOT the guarantee — `bun.nix` still drifts.**
> It fires only on a local `bun install` that runs lifecycle scripts. It does *not* fire under
> `bun install --ignore-scripts`, which is what every bun install in CI uses; and it cannot fire
> on a **merge or rebase**, where git carries someone else's `bun.lock` change past a `bun.nix`
> generated before it and reports no conflict. That is how `web/bun.nix` shipped on main holding
> `brace-expansion@5.0.7` while `web/bun.lock` said `5.0.8` — for **553 commits** (2026-07-27 →
> 2026-08-05), with `nix build .#punktfunk-web` broken the whole time, until an unrelated
> advisory bump happened to rerun a real `bun install` and closed it by accident.
>
> The enforcement point is **`scripts/ci/check-bun-nix.sh`** (the `bun-nix` job in `ci.yml`,
> unfiltered so it sees the innocuous-looking commits drift arrives through). It regenerates each
> `bun.nix` from its committed `bun.lock` and diffs. Fix any report with:
>
> scripts/ci/check-bun-nix.sh --fix
>
> Never regenerate with a bare `bunx bun2nix`: **`bun.nix` has no schema stability across bun2nix
> versions**, and an unpinned `bunx` uses whatever is newest. The flake input
> (`github:nix-community/bun2nix?ref=2.1.2`) and the npm devDependency in `web/package.json` +
> `sdk/package.json` must name the *same exact version* — the script checks that too, and always
> generates with the pinned one. Move all three together, then rerun it with `--fix`.
Everything past the deps fetch is offline (the console's codegen + vite build; the runner's
`bun build --target=bun` bundle). Both launchers exec `pkgs.bun` from the store — unlike the
+181
View File
@@ -0,0 +1,181 @@
#!/bin/sh
# Drift gate for the generated bun2nix lockfile expressions (web/bun.nix, sdk/bun.nix).
#
# `bun.nix` is a DERIVED file: bun2nix is a pure function of `bun.lock` (it reads the lockfile text
# and emits one `fetchurl` per package, keyed by the lockfile's own integrity hashes — see
# packaging/nix/README.md). Nothing but the lockfile goes in, so any disagreement between the two
# committed files is drift, and it is always mechanically fixable.
#
# Why this exists: moving the bun packages to bun2nix (1db8f763) removed the *aggregate deps hash*
# that used to go stale, but not the second, quieter way a derived file rots. `bun.nix` regenerates
# only from a local `bun install` that runs lifecycle scripts (web's `postinstall`, the SDK's
# `prepare`). It does NOT regenerate on:
#
# * `bun install --ignore-scripts` — which is what EVERY bun install in CI uses (ci.yml,
# web-screenshots.yml, windows-host.yml, sdk-publish.yml), because web's `postinstall` shells
# out to a `bun` on PATH that CI's portable bun isn't;
# * a merge or rebase — git merges `bun.lock` and `bun.nix` as two unrelated files, so a branch
# that generated `bun.nix` before picking up someone else's lockfile change silently commits
# the pair out of step;
# * a lockfile edited or re-resolved by hand.
#
# That second case is not hypothetical: it is how `web/bun.nix` shipped on main carrying
# brace-expansion@5.0.7 (plus two nested entries the override had already collapsed) while
# `web/bun.lock` said 5.0.8 — the `^5.0.8` override from ec9aa415 landed in the lockfile, the
# bun2nix branch had generated `bun.nix` off the pre-override lock, and the merge kept both. The
# Nix build fetches node_modules strictly from `bun.nix`, so the offline `bun install` inside the
# derivation is then asked for a tarball the store cache does not contain and `punktfunk-web` fails
# to build — with a "package not found" that names npm, not the lockfile that actually drifted.
#
# The gate also enforces the version pin the flake and README only *state*: `bun.nix` has no schema
# stability across bun2nix releases, so the flake input ref and BOTH npm devDependencies must name
# the same exact version. Nothing checked that before; a half-moved pin regenerates the file with a
# generator the flake does not use.
#
# The list of packages to check is read out of packaging/nix/packages.nix (its `bunNix = src + …`
# lines) rather than hardcoded here, so a third bun package is covered the day it is added — and an
# empty list is a hard error, because a gate that checks nothing passes exactly like a clean tree.
#
# Usage:
# scripts/ci/check-bun-nix.sh # verify; non-zero on drift (CI)
# scripts/ci/check-bun-nix.sh --fix # regenerate the committed files in place
set -eu
FIX=0
if [ $# -gt 0 ]; then
case "$1" in
--fix) FIX=1 ;;
*) echo "usage: $0 [--fix]" >&2; exit 2 ;;
esac
fi
ROOT=$(CDPATH='' cd -- "$(dirname -- "$0")/../.." && pwd)
PACKAGES_NIX="$ROOT/packaging/nix/packages.nix"
FLAKE="$ROOT/flake.nix"
command -v bun >/dev/null 2>&1 || {
echo "check-bun-nix: bun is not on PATH (needed to run bun2nix and to read package.json)" >&2
exit 1
}
[ -f "$PACKAGES_NIX" ] || { echo "check-bun-nix: no $PACKAGES_NIX" >&2; exit 1; }
[ -f "$FLAKE" ] || { echo "check-bun-nix: no $FLAKE" >&2; exit 1; }
TMP=$(mktemp -d)
trap 'rm -rf "$TMP"' EXIT
# --- the pinned bun2nix version -------------------------------------------------------------------
# flake.nix: url = "github:nix-community/bun2nix?ref=2.1.2";
PINNED=$(sed -n 's/.*github:nix-community\/bun2nix?ref=\([^"]*\)".*/\1/p' "$FLAKE" | head -1)
[ -n "$PINNED" ] || {
echo "check-bun-nix: could not read the bun2nix input ref out of $FLAKE." >&2
echo "Expected a line like: url = \"github:nix-community/bun2nix?ref=<version>\";" >&2
exit 1
}
# --- which packages carry a generated bun.nix -----------------------------------------------------
# packages.nix: bunDeps = bun2nix.fetchBunDeps { bunNix = src + "/web/bun.nix"; };
sed -n 's/.*bunNix *= *src *+ *"\/\(.*\)\/bun\.nix".*/\1/p' "$PACKAGES_NIX" | sort -u > "$TMP/roots"
if [ ! -s "$TMP/roots" ]; then
echo "check-bun-nix: found no \`bunNix = src + \"/<dir>/bun.nix\"\` in $PACKAGES_NIX." >&2
echo "Either the bun packages were removed (delete this gate) or the expression changed shape" >&2
echo "and the gate silently stopped checking anything. Not passing vacuously." >&2
exit 1
fi
fail=0
checked=0
# --- version pin agreement ------------------------------------------------------------------------
# `bun.nix` has no schema stability across bun2nix versions, so the generator the flake builds with
# and the generator `bun install` runs must be the SAME exact version (packaging/nix/README.md).
while read -r dir; do
pkgjson="$ROOT/$dir/package.json"
[ -f "$pkgjson" ] || { echo "check-bun-nix: no $pkgjson" >&2; fail=1; continue; }
dev=$(bun -e "const d=require(process.argv[1]).devDependencies||{};console.log(d.bun2nix??'')" \
"$pkgjson")
if [ "$dev" != "$PINNED" ]; then
echo "check-bun-nix: bun2nix version pin disagrees." >&2
echo " flake.nix input ref : $PINNED" >&2
echo " $dir/package.json devDependency : ${dev:-<absent>}" >&2
echo "These must be the same exact version — bun.nix has no schema stability across" >&2
echo "bun2nix releases. Move both together, then rerun this script with --fix." >&2
fail=1
fi
done < "$TMP/roots"
# --- the generator ---------------------------------------------------------------------------------
# Prefer an already-installed bun2nix at the pinned version (fast, offline — the dev case); otherwise
# fetch exactly the pinned one, once, into $TMP. Never a floating `bunx bun2nix`: that would generate
# with whatever is newest, and `bun.nix` has no schema stability across releases.
BUN2NIX=""
while read -r dir; do
cand="$ROOT/$dir/node_modules/bun2nix/index.ts"
[ -f "$cand" ] || continue
have=$(bun -e "console.log(require(process.argv[1]).version??'')" \
"$ROOT/$dir/node_modules/bun2nix/package.json" 2>/dev/null || echo '')
if [ "$have" = "$PINNED" ]; then BUN2NIX="$cand"; break; fi
done < "$TMP/roots"
if [ -z "$BUN2NIX" ]; then
# Installed in its own scratch dir, so this never touches the repo's lockfiles or .npmrc.
mkdir -p "$TMP/gen"
if ! ( cd "$TMP/gen" && bun add --exact "bun2nix@$PINNED" ) > "$TMP/geninstall.log" 2>&1; then
echo "check-bun-nix: could not install bun2nix@$PINNED" >&2
cat "$TMP/geninstall.log" >&2
exit 1
fi
BUN2NIX="$TMP/gen/node_modules/bun2nix/index.ts"
[ -f "$BUN2NIX" ] || { echo "check-bun-nix: bun2nix@$PINNED installed but $BUN2NIX is absent" >&2; exit 1; }
fi
run_bun2nix() { # <lockfile> <outfile>
bun "$BUN2NIX" --lock-file "$1" --output-file "$2"
}
# --- regenerate + compare ---------------------------------------------------------------------------
while read -r dir; do
lock="$ROOT/$dir/bun.lock"
nix="$ROOT/$dir/bun.nix"
[ -f "$lock" ] || { echo "check-bun-nix: no $lock (packages.nix expects $dir/bun.nix)" >&2; fail=1; continue; }
out="$TMP/$(echo "$dir" | tr '/' '_').bun.nix"
run_bun2nix "$lock" "$out" >/dev/null
if [ "$FIX" -eq 1 ]; then
if [ ! -f "$nix" ] || ! cmp -s "$nix" "$out"; then
cp "$out" "$nix"
echo "check-bun-nix: regenerated $dir/bun.nix from $dir/bun.lock"
else
echo "check-bun-nix: $dir/bun.nix already in sync"
fi
checked=$((checked + 1))
continue
fi
if [ ! -f "$nix" ]; then
echo "check-bun-nix: $dir/bun.nix is MISSING — packages.nix fetches node_modules from it." >&2
fail=1
continue
fi
# Plain files, not `diff <(…) <(…)`: Gitea's runner executes a step's `run:` under `sh`, and
# dash has no process substitution — it would reject the script at parse time and the gate
# would never compare anything (exactly how the shader SPIR-V gate in ci.yml was lost).
if cmp -s "$nix" "$out"; then
echo "check-bun-nix: $dir/bun.nix matches $dir/bun.lock"
else
echo "check-bun-nix: $dir/bun.nix is STALE — it does not match $dir/bun.lock." >&2
echo "The Nix build fetches node_modules only from bun.nix, so punktfunk's bun packages" >&2
echo "would build against the wrong dependency set (or fail to fetch it at all)." >&2
echo "Regenerate and commit it: scripts/ci/check-bun-nix.sh --fix" >&2
echo "--- diff (committed -> regenerated from bun.lock) ---" >&2
diff -u "$nix" "$out" >&2 || true
fail=1
fi
checked=$((checked + 1))
done < "$TMP/roots"
[ "$checked" -gt 0 ] || { echo "check-bun-nix: checked nothing — refusing to report success" >&2; exit 1; }
if [ "$fail" -eq 0 ] && [ "$FIX" -eq 0 ]; then
echo "check-bun-nix: $checked bun package(s) in sync, bun2nix pinned at $PINNED everywhere"
fi
exit "$fail"