[JacobStephens2]

Summary

 release builds
   x86_64-unknown-linux-musl
+  aarch64-unknown-linux-musl (ubuntu-24.04-arm)
   aarch64-apple-darwin
+ ARM static/version smoke check before publication
+ ARM full suite on main/nightly/manual and test:linux-arm PRs

Closes #1663

Test seams

• cargo-dist's release plan and generated shell installer.
• The reusable release smoke workflow's shell steps, run against archive fixtures with the host commands and binary at the external process seam.
• GitHub Actions' CI job selection and publication dependencies.

Evidence

• Before: the ARM smoke regression failed with KeyError: 'arm-linux'; the release plan omitted ARM Linux.
  After: all five ARM smoke tests pass, including rejection of dynamic linkage and the wrong version. The actual dist plan selects aarch64-unknown-linux-musl on native ubuntu-24.04-arm; the generated installer maps aarch64 GNU and musl hosts to the static ARM archive and retains its receipt writer.
• All 34 Python tooling tests pass. python3 scripts/generate-release.py --check passes with cargo-dist 0.33.0. Regeneration leaves release.yml unchanged because its matrix already comes from the plan.
• Full hosted CI passes: ARM Linux 18m40s, x86_64 Linux 22m48s, macOS gate and both browser suites. The test:linux-arm label demonstrates the PR opt-in. ARM finishes before the existing Linux job, so no sharding is needed on this evidence. The reviewed event condition also selects the full ARM suite on main pushes and nightly.
• Release dry run passes all three builds and ARM/Rocky/WSL smoke checks. The native ARM smoke log reports an ARM aarch64 statically linked executable and prints the plan's expected thirdshift 0.18.8 from the PR merge build. ARM builds in 2m44s; smoke takes 6s. Publication jobs are skipped on the PR.
• Local scripts/affected-tests main: site PASS, 60 tests; the initial cargo test attempt FAIL after 1,390.62s, including 4m04s compilation, on existing process_tree::interrupting_each_migrated_check_stops_its_owned_detached_descendant (owned check process … survived). cargo test --test process_tree interrupting_each_migrated_check_stops_its_owned_detached_descendant -- --exact --nocapture then passed in isolation (1.71s), and cargo test --test process_tree -- --test-threads=3 passed all 20 active tests (21.51s), following the repository's local-concurrency guidance. Full hosted CI for this head is green, so its successful full suites are reused instead of repeating another full run. No product or test-fixture Rust code changed.
• Live ARM installer/update verification waits for the first release containing this target. The generated installer platform table and receipt-writing path were checked, as the issue permits before that release exists. No release is made here.

Local setup uses Python 3.11 (PATH=/tmp/issue-1663-tools:$PATH) and the installed Playwright cache (PLAYWRIGHT_BROWSERS_PATH=/home/jstephens/.cache/ms-playwright).

Unaddressed findings

Standards

• S1, P3 possible Duplicated Code — .github/workflows/ci.yml:100–113: declined as a design judgement. Keep the two Linux jobs explicit for this change. The x86_64 job owns formatting, Python tooling and release-generator checks (ci.yml:76–81); ARM runs only the requested clippy and suite, behind its own job-level PR gate (ci.yml:98). Sharing these short sequences would add a wrapper and platform-specific orchestration. This is an implementation decision, not a deferred Spec or Day shift decision. The reviewer reported no wrong behavior or hard standard breach.

Spec

None; the reviewer raised no findings.

Security

None; the guidance-mode review found no introduced Security findings against pinned merge base 44c99f7977f78ce4969c15d4d137db4c1a6f700b.

Reviewed all 10 changed files and supporting release generator, publication dependencies, installer platform/receipt handling, updater and test helpers. Applied the attack-class, supply-chain/release and cloud/deployment guidance, ADRs 0003/0013/0015 and the repository's run-isolation threat-model discussion. Checked shell input handling, archive/artifact selection, cache/event trust and the smoke-to-publication gate. GitHub's download-artifact v8 inputs and cache scope rules support the reviewed workflow boundaries.

Local validation: python3 -m unittest discover -s scripts -p 'test_*.py' -v passed all 34 tests, including all 5 ARM smoke tests. Execution used an empty allowlisted environment, no network, read-only repository/toolchain, fresh home/cache/temp, and a disposable 128 MiB scratch filesystem; limits were 1 CPU, 512 MiB memory with no swap, 64 tasks, 16 MiB per file and 60 seconds wall time. No dependencies were fetched. Reused the successful full CI and release dry run for reviewed head bb8f68943430c7f29503fc1e98437ec1128c433d, as docs/testing.md directs. No source changes or fix commits were needed. This was a focused diff review in one session.

Security review outcome

No unaddressed introduced findings.

Review coverage

Independent thirdshift-code-review axes reviewed git diff main...HEAD, with main pinned to 44c99f7977f78ce4969c15d4d137db4c1a6f700b. Standards: 1 judgement call, 0 hard violations. Spec: 0 findings. Each reviewer read all 10 changed files, including the generated manual hunks; no changed file was left unread and no generated output was excluded.

Reports and final files-read lists: .thirdshift-review-8RltMQ/standards.md, .thirdshift-review-8RltMQ/spec.md; counts: .thirdshift-review-8RltMQ/summary.json ({"standards":1,"spec":0}).

Merge Danger

Door: two-way

Revert the workflow/configuration changes before the next release to withdraw the new target. No release is made by this PR.

Blast Radius: releases

ARM build and smoke failures block publication. Ordinary PRs skip the ARM full suite unless labelled.

Built with codex codex-cli 0.166.0 · gpt-6.1-sol · xhigh