[teamleaderleo]

Merged (1355c9c): the first PR of the ci-linux-move lane, which moves CI jobs that don't need macOS off the Mac runners. This PR moves cmux-next-shim-cpp.yml; later workflows get their own PRs.

Summary

• CEF shim download state (C++) now runs on blacksmith-4vcpu-ubuntu-2404 (vars.LINUX_RUNNER when set; ubuntu-24.04 on forks). It used to take a light or side mini, or Blacksmith macOS as the fallback. The job keeps its name, so required checks and the stall and merge-audit tools still match it.
• The ten scripts/cmux-next/test-*-cpp.sh compile header-only rules that need no CEF and no Apple SDK. The job selected Xcode only to find clang++. The new scripts/cmux-next/lib/cxx.sh picks the compiler:
  • CXX when set;
  • else xcrun clang++ on a Mac, so local runs are unchanged;
  • else clang++.
• cmux_page_path_test.cpp used macOS's /tmp symlink to pass the root through a symlink. The test now makes its own symlink, so that case is tested on Linux as well.
• tests/test_cmux_next_shim_cpp_linux.py is a new linux-guard test, committed red first. It checks that:
  • every script takes the shared compiler and calls no xcrun itself;
  • the workflow runs every script;
  • the job stays on Linux under its name, with no Xcode step.
• tests/test_ci_owned_pool_sources.py: on feat-cmux-next, a stale entry in main's rescue list is now a warning instead of a failure.
  • Main's list can't change in the same commit that moves a feat-cmux-next workflow off the minis, so without this change no move could ever land green.
  • A stale entry is harmless: the rescue only watches a run that has no owned job.
  • It is still a failure on main, and an owned workflow that the rescue doesn't watch still fails on both branches.
  • A follow-up PR to main removes cmux-next-shim-cpp.yml from SIDE_WORKFLOW_PATHS once this one lands.
• 208993d (after the first CI run): the job's runs-on takes the fork-safe Linux expression the guard jobs use, the job's .github/runners.json macOS entry is gone, and a ci-guards.yml step runs tests/test_cmux_next_shim_cpp_linux.py on every PR.

Testing

• Red at 4ab761e (test only): python3 tests/test_cmux_next_shim_cpp_linux.py failed: 11 failures in 4 tests.
• Green at the head: the same command passes.
• python3 tests/test_ci_owned_pool_sources.py passes (7 tests, stale-entry warning printed), and so does python3 tests/test_ci_owned_pool_rescue.py.
• All ten scripts pass on cmux16 through both paths: xcrun clang++ (the default) and CXX=/usr/bin/clang++. The case counts match: 62, 18, 101, 20, 28, 42, 7, 18, 50 and 10. No /tmp/cmux-page-path.* directory is left behind.
• Proven on Linux: this PR's run at 208993d and the feat-cmux-next push run at 1355c9c both passed on Blacksmith Linux.

Before and after (job CEF shim download state (C++), from the Actions jobs API)

Run	Event	Runner	Queue	Run
39674915952	push 1355c9c (after)	blacksmith-4vcpu-ubuntu-2404	3 s	25 s
39674287779	PR 208993d (after)	blacksmith-4vcpu-ubuntu-2404	4 s	23 s
39604308183	push 001b266 (before)	cmux19 owned mini	10 s	23 s
39575400940	push 01797cf (before)	cmux24 owned mini	9 s	21 s
39531504619	push 1219ce8 (before)	cmux17 owned mini	4 s	80 s
39508705780	push 9bdb297 (before)	cmux21 owned mini	10,846 s	14 s
39566718746	PR (before)	blacksmith-6vcpu-macos-26	11,162 s	50 s
39399414046	PR (before)	cmux18 owned mini	1,941 s	129 s

The test time is the same (about 20 to 25 s). What changes is the wait: on the Mac pool the job waited up to 3 h behind the side lane, and on Linux it starts in seconds. The Mac time saved is small, about 1.8 owned-mini hours a week (ci-dash), plus the queue slot.

Observed / Not run

• Observed: everything listed under Testing.
• Not run locally: Linux. The Linux compile (Ubuntu clang with libstdc++) is proven by the CI runs above. actionlint and shellcheck aren't installed on cmux16, so the post-edit hooks and CI check those.

Inventory: Mac-runner jobs and what they need

I read these from origin/main and origin/feat-cmux-next and ran nothing. The hours are owned-mini job hours over 7 days, from ci-dash (jobs-group.json?since=7d, about 2,142 h in total).

The linux-swift lane (cosmic-gray) took every Swift/SwiftPM job. This lane keeps lint, Rust, web/bun, Python, C++/script tests, generators and diff checks.

Workflow / job	Needs	Mini h/7d	Verdict
cmux-next-shim-cpp / shim-downloads-cpp	nothing Mac-specific: header-only C++	1.8	MOVE (this PR)
signing-keychain-macos / bash32-syntax	/bin/bash 3.2 only (CMUX_PARSE_BASH can point elsewhere)	<1	MOVE: GNU bash 3.2.57 built on Linux; run it on both once
macos-matrix-fuzz / fuzz	nothing on the runner: curl and python to the fleet controller, holds a Mac for up to 240 min	—	MOVE, if Blacksmith Linux can reach CMUX_FLEET_CONTROLLER
cmux-next-source-archive / archive	xcrun llvm-dwarfdump; macOS fetch lists	—	PARTIAL: notices and archive on Linux (apt llvm); the macOS fetch lists stay on Mac
cmux-tui-artifacts / cmux-next-daemon-tests	macOS runtime of the shipped daemon (cargo only; no SDK or GUI)	292	PARTIAL, but it gates nightly promotion, so moving it means giving up macOS runtime coverage. Not mine to decide alone; ci-speed's #23660 also edits this file
ci.yml (main) / claude-wrapper	BSD stat -f, /bin/zsh spool path	1.6	PARTIAL: needs a portable stat
cmux-tui / cdp-browser-smoke (macos entry)	nothing Mac-specific; a Linux entry exists	—	PARTIAL: drop it only if Mac CDP coverage isn't wanted
cmux-tui-build-package / build (darwin)	mostly cross-built on Linux already; Mac for npm/PyPI packaging, the agent plugin and the stale-stub fallback	—	PARTIAL
cmux-tui / macos, macos-relay, lint/test (macos)	cfg(target_os="macos") code, launchd, PTY	152 + 2.4 + 3.5	KEEP (the Linux entries already exist)
cmux-next-terminal / rust-terminal	cmux-pty Apple cfgs; corpus digests recorded on macOS	—	KEEP
remote-daemon / remote-daemon-macos-tests	darwin-only Go files	2.1	KEEP
relay-publish-npm / smoke (darwin); coderouter-release / build (darwin)	darwin binaries, codesign	—	KEEP
relay-tls / system-keychain	System keychain, security	—	KEEP
cmux-next / generated-files, swift-canary, daemon-test; pr/base-autofix action-exports; batch regen; seed-swiftpm-manifests	CmuxNext swift build/swift test (AppKit, macOS 14 platform)	108 + 108 + 30 + 77	KEEP
Swift package jobs: agent-brands, auth-refresh, iroh-v2, relay-tls diagnostic, terminal-hang, test-feed-reply-providers, mobile-core-package, car-guards mac, cloud-machine-tests, cloud-command-deadlines	Apple frameworks (CMUXMobileCore imports Darwin and CoreImage unguarded)	—	the linux-swift lane's
App builds and tests: notarize, stapling, Sparkle, GhosttyKit and xcframeworks, compile admission, release/scheme compile, fast-tier, dogfood, E2E, media tour, frame pacing, iOS, nightly*, release, reload-build, perf-activation, test-macos-suite, compat	Xcode, codesign, a display or simulators	—	KEEP

The Mac side-lane queue before the first merge (ci-dash, 2026-10-21 about 20:24Z):

Pool	Running	Queued	Oldest wait
glaeda-side-std-xcode-26.6	28	86	3,667 s
glaeda-gui-std	9	44	3,674 s
blacksmith-6vcpu-macos-26	11	21	—

Changelog

none

Risk: low. CI-only change: one test job's runner, its test scripts, and a CI guard's warning.

Agent session: cmux16

🤖 Generated with Claude Code

[github-actions]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

Review 208993d: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr23712 -f ref=cmux/ci-linux-move -f platform=macos -f backend_mode=local

Not verified: macOS jobs did not run: compile and app tests were skipped on 208993d239; ci-status is green without them.

CI passes on 208993d239 (run 39674287730 attempt 1).
macOS jobs did not run: compile and app tests were skipped: Fast static checks skipped (the macos job needs both to succeed). Re-run CI once they pass, before merging.

Written by scripts/ci/classify_failures.py (ci-failure-attribution.yml); signatures are its SIGNATURES table. A machine verdict is the runner's fault, not this PR's; yours means the failing file is one this PR changes, also red on main that main's latest full suite fails the same way, seen on other PRs that it failed on another pull request's run lately.

red: cmux-next checks (god files, concurrency, crash safety, l10n); base red: the same failure on the base 2f1920a; does not block

[teamleaderleo]

Owned by ci-linux-move on cmux16 · session

[blacksmith-sh]

Found 10 test failures on Blacksmith runners:

Failures

Test	View Logs
/tmp/tmp944jm63j/summary-fail.json (live):/summary-fail.json (live):	View Logs
/tmp/tmpz8rgtf28/summary-missing.json (live):/summary-missing.json (live):	View Logs
[new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows/ [new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows	View Logs
[new_workspaces] the window draws no frames/[new_workspaces] the window draws no frames	View Logs
[tabs] debug.tab_switch timed 0 of 40 switches/ [tabs] debug.tab_switch timed 0 of 40 switches	View Logs
[tabs] the window draws no frames/[tabs] the window draws no frames	View Logs
compile admission must pass the owned Mac's GhosttyKit archive cache/ compile admission must pass the owned Mac's GhosttyKit archive cache	View Logs
convention violations found (ERROR lines above)./ convention violations found (ERROR lines above).	View Logs
test_a_file_a_test_reads_selects_that_test (main.PullRequestTiers.test_a_file_a_tes t_reads_selects_that_test)/ test_a_file_a_test_reads_selects_that_test (main.PullRequestTiers.test_a_file_a_tes t_reads_selects_that_test)	View Logs
test_a_shared_module_selects_its_dependents (main.PullRequestTiers.test_a_shared_mo dule_selects_its_dependents)/ test_a_shared_module_selects_its_dependents (main.PullRequestTiers.test_a_shared_mo dule_selects_its_dependents)	View Logs

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[teamleaderleo]

Bead cx-q6yds created for this PR by the cmux coordinator.

[teamleaderleo]

Independent review (claude subagent) of the owned-pool guard change, read together with #23719.

• The test file is byte-identical to test(owned-pool sources): a stale entry only warns for a workflow main does not have #23719's at f26b6ab.
• A stale entry warns only when main's tree lacks the workflow. A stale entry for a workflow main does have still fails on both branches, and an owned workflow the rescue doesn't watch still fails.
• Land order simulated from the main worktree: (1) test(owned-pool sources): a stale entry only warns for a workflow main does not have #23719 merges, nothing is stale; (2) this PR merges, warning only for cmux-next-shim-cpp.yml, no failure on either branch; (3) the main PR that drops the entry: clean. Without test(owned-pool sources): a stale entry only warns for a workflow main does not have #23719 first, main goes red once this PR lands.
• python3 tests/test_ci_owned_pool_sources.py: 7 tests pass, with branch detection and with CMUX_OWNED_POOL_BRANCH=main.

Review (claude subagent): SHIP at a0048a6

[teamleaderleo]

Independent review of 208993d (the guard-fix commit), by a claude subagent:

• The fork-safe runs-on matches the expression already used by ci-guards.yml, ci-fast-guards.yml and cmux-next-guards.yml; tests/test_ci_fork_runner_routing.py passes.
• The runners.json entry is gone, the file is valid JSON, tests/test_runner_routing_table.py passes and runner_routing_table.py check exits 0.
• The new ci-guards step matches its neighbours; cmux-next-guards.yml runs the ci group on every feat-cmux-next PR, so the registry no longer flags the shim test.
• CEF shim download state (C++) passed on Blacksmith Linux in 23 s on this head.

Review (claude subagent): SHIP at 208993d

[teamleaderleo]

base-red-watch: cmux-next guards / cmux-next CI guard group went red on feat-cmux-next in https://github.com/manaflow-ai/cmux/actions/runs/39674917051 (first red at 1355c9c, last green at 011b014). Suspect: #23712 (commit 1355c9c). If it is yours, fix forward on feat-cmux-next.

[github-actions]

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.3968001295301 (8cba1da27d)