ci: keep the runner account out of the SwiftPM manifest cache key

[teamleaderleo]

Compile admission jobs on the glaeda Mac minis restore the SwiftPM manifest cache from #14257 but still evaluate every manifest. Their resolves take 52 to 56 s, against 13 to 21 s on Blacksmith macOS 26 in the same window (PR runs created 19:07 to 20:40Z).

For example, admission job 107819765887 on cmux11s-mac-mini-glaeda logged "Installed 162 SwiftPM manifest cache entries" (same Xcode 26.6 prefix), then resolved cold.

The cause is the fixed resolve environment in scripts/ci/swiftpm-manifest-cache.sh run. It kept HOME, USER and LOGNAME, and SwiftPM keys every manifest on the whole environment. Those three name the runner account: runner on Blacksmith, cmux on the minis. This drops them.

Without HOME, SwiftPM finds the same ~/Library/Caches/org.swift.swiftpm through the user database. I checked locally with xcodebuild:

• A resolve with no HOME, USER or LOGNAME succeeds.
• It writes to the usual manifest.db.
• A second run hits.
• Adding HOME back creates a separate entry.
• A resolve that fetches a remote package over git also succeeds without HOME.

install, stage and a new clear (used by both seeders) find the cache directory through the user database too, so they agree with xcodebuild even on a runner that points HOME elsewhere.

The script's hash is part of the cache key, so current seeds stop matching exactly:

• seed-derived-data.yml reseeds the macOS 26 pool on the next main push that changes the build.
• seed-swiftpm-manifests.yml reseeds macOS 15 on its schedule.

Until then, readers restore the newest seed for their Xcode, which is now keyed differently, so resolves run as they did before #14257.

Tests: tests/test_ci_swiftpm_manifest_cache.sh now runs run under two different accounts and requires identical environments with no HOME, USER or LOGNAME. These also pass: test_ci_test_compilation_cache_seed.sh, and test_ci_canonical_build_root.py with a realpath TMPDIR.

Signed: Manifold g1 ✨ (run_swiftpm-manifest-cache-20260924)

🤖 Generated with Claude Code

Summary by CodeRabbit

• Chores
  • CI workflows now use a consistent process to clear and restore the Swift package manifest cache. The cache location can be set explicitly and no longer depends on the HOME, USER, or LOGNAME environment variables.
• Tests
  • Expanded checks confirm cache operations use consistent environments and support a configured cache location.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: Repository: manaflow-ai/cmux/.coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: c8caa972-e70f-4c8e-a2b0-0c70f2837d14

📥 Commits

Reviewing files that changed from the base of the PR and between 150ba9f and ab4c0a1.

📒 Files selected for processing (4)

• .github/workflows/seed-derived-data.yml
• .github/workflows/seed-swiftpm-manifests.yml
• scripts/ci/swiftpm-manifest-cache.sh
• tests/test_ci_swiftpm_manifest_cache.sh

 ____________________________
< Try...catch me if you can. >
 ----------------------------
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[cursor]

Bugbot is paused — on-demand spend limit reached

Bugbot uses usage-based billing for this team and has hit its on-demand spend limit.

A team admin can raise the spend limit in the Cursor dashboard, or wait for the next billing cycle to continue.

[teamleaderleo]

Measured after the reseed. Push seed 36059263610 saved under the new key at about 21:10Z. The table covers PR runs created 21:55Z to 22:55Z.

lane	runner	resolve
admission	Blacksmith macOS 26	14 to 19 s, median 15 s (n=11)
admission	glaeda minis	13 to 16 s (n=4); before this PR: 52 to 56 s
cli-pipe-regressions	Blacksmith macOS 26	13 to 18 s, median 14 s (n=6)
cli-pipe-regressions	glaeda mini	41 s (n=1)

Before #14257, the medians were 63 s for admission and 48 s for cli-pipe.

The minis now hit the Blacksmith seed for admission, which resolves at the fixed canonical root. cli-pipe-regressions resolves in the runner workspace. That path differs between a mini (/Users/cmux/actions-runner-glaeda/_work/...) and Blacksmith (/Users/runner/_work/...), and each manifest's cache entry includes its absolute path, so it still misses there. No PR jobs landed on the macOS 15 pool in this window. Its seed (#14272) saved at 21:03Z.

Signed: Manifold g1 ✨