[lawrencecchen]

Summary

The fleet films and benches Release builds with bench hooks (CMUX_BENCH_HOOKS, cmux-ci build cmux --release). Their launch never turns the tunable store on: DebugSettingsService.start() returns unless DevTools.isEnabled (a Debug compile or a nightly bundle), so TunableStore.shared stays off with no descriptors. debug.tunables set then changes nothing and state says "active": false (film round px-col-drag-2 on FEAT DOGFOOD BUILD 1d8e759e). One build can't film a Debug switch off and on, and the speed sweep can't scale the springs.

debug.tunables gets an activate action. On a store that launch left off, it registers TunableCatalog.all and calls activate(file: nil), so overrides apply for this launch, in memory only. On a store that's already on (DEV, NIGHTLY), it does nothing. It answers state (active: true). Nothing changes unless a client calls it, and it exists only where the method does (#if DEBUG || CMUX_BENCH_HOOKS): Release, RC and NIGHTLY don't have it.

A film kit calls debug.tunables {"action": "activate"} once after launch, checks active, then sets the stage switch or motion.spring.<token>. available still reports the Debug Settings window (DevTools), so it stays false in a bench build.

Why memory only: nothing is written and the next launch starts off again. A file wouldn't help (the next launch doesn't read it) and a stale one could change another owner's bench.

What a set after activate reaches is now in plans/cmux-next/debug-settings.md:

• a tunable read at each use (a switch read per frame, MotionSpring.base when a spring starts): its next read;
• a reader that tracks the key after activate: told of each change. One that tracked it before isn't (an off store returns before it registers the read): SidebarTunablesWatch keeps its launch snapshot of 8 row values until a relaunch, and an ObservationStream loop that started before stays put;
• browserResize.dragResize's synced choice is read once, when Chromium starts. cmux pages (New Tab and agent chats included) show in Chromium by default, and the launch's New Tab spare page starts it about 250 ms after the first frame, normally before a socket client has sent activate, so a bench build can't count on reaching synced this way (debug.cef state, trigger and ready_after_launch_s say whether, why and when Chromium started); layerHosting applies at a page's next show; the first page that takes the launch spare (a New Tab, a new workspace's first page, an existing chat) keeps the old icons.family;
• readers gated on DevTools.isEnabled or wired only by DebugSettingsService.start() keep their Release behavior: auth.signInWindow, home.chiefViewGallery, sidebar.sections.localPrototype, home.flightRecorder.

With no file, loading stays false, settledOverride gives the override and nothing is written.

DebugTunables.handle and .state take the store as a parameter (default .shared) so the test drives a store of its own; the socket method and every other caller still pass the shared one. FRAME OWNER REFACTOR stage S0 (refactor owner pixel-perfect).

Testing

New suite DebugTunablesActivateTests (its own file, a fresh TunableStore per test because other suites turn the shared store on):

• without activate, a set changes nothing and state says off (passes on the base too: benches that never call activate stay the same);
• activate turns the store on, and each later set reaches the next read of browserResize.pageBackdrop and of the move spring;
• a reader that starts tracking after activate is told of the next set;
• activate leaves a store that's already on as it is.

Command (fleet, both runs; the controller runs package-test-lane.sh only as class light, it refuses isolated):
cmux-ci run --class light --script scripts/ci/package-test-lane.sh --arg=suite --arg=Packages/macOS/CmuxNext --arg=DebugTunablesActivateTests --ref <sha> --key <lane>-<sha12> (submitted through the lead's proof-step.py, which adds the key)

• Red at 9316d72 (feat 1a7e2c1 merged into the test commit 4ee6acb, no fix): step 36c6e3dff7bf61a7f25ca21e (cmux10) failed as expected, 3 of 4 tests: activate answers "unknown action", active and the pageBackdrop / move spring values don't change after a set, and the tracking reader is told 0 times. withoutActivateASetChangesNothing passed, as it must on both sides.
• Green, 4 of 4 passed each time: at 3be0406 (the fix) step a122187239af62861d765bd1 (aws-m4pro-5); at 90b21bb step f1ac080a739723faf9edf6e6 (aws-m4pro-3); at 9145aba step 5a0eae3b8770b2e2768274e5 (Blacksmith overflow, hq run 41635215511). The head 8893b5a changes only plans/cmux-next/debug-settings.md over 9145aba, so it runs the same Swift. Since the red, the Swift changed only by dropping one assertion that passed on the red ref too (file is null there as well), so which tests fail without the fix doesn't change.
• Compile-checks (cmux-ci compile-check --app --package CmuxNext --recipe-version 2: CmuxNext with its tests, the app, cmux-tui): red ref 9316d72 passed (817e15f36602047c4f49687e), 3be0406 passed (8faf68223e4eefd3abda2a69), 90b21bb passed (997c8ae58d61ba6be177c60c), 9145aba passed (d74700306ab801c4e670d90b). The test commit alone (4ee6acb, on feat 6479948) did not build its tests: 38feb5580a6424b18c50a734 failed only on the base red RowBadgeSpinnerDensityTests.swift:32 that base red: RowBadgeSpinnerDensityTests' argument table is nonisolated (from #33120) #33377 fixed, hence the merge of feat.

Observed / Not run

Observed: the red step failing on exactly the 3 tests that need activate, with the messages above; the green steps passing 4 of 4 at 3be0406, 90b21bb and 9145aba; the compile-checks above. In PR CI, cmux-next swift test ran DebugTunablesActivateTests first and it passed; that job and the release-notary guard are red on base findings (linked in the PR comments).

Not run: a live bench build round (the refactor owner's film kit calls activate in its next round); Release/RC compiles beyond CI's (the change is inside #if DEBUG || CMUX_BENCH_HOOKS).

Changelog

none

Proof

No capture: debug and bench socket method only, no UI.

Checklist

• Behavior changes have added or updated tests
• No user-facing strings (developer socket method); no localization change
• Not a relay-allowlisted method: debug.tunables is registered only on the local control router
• Reviewed with a subagent before merge (FIX on the doc at 3be0406 and at 90b21bb, SHIP with doc nits at 9145aba, all fixed; no code defect found in any review; the receipt for the head is a PR comment)

[cursor]

Bugbot needs on-demand usage enabled

Bugbot uses usage-based billing for this team and requires on-demand usage to be enabled.

A team admin can enable on-demand usage in the Cursor dashboard.

[github-actions]

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

Review 8893b5a: 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=pr33553 -f ref=fix/frame-owner-s0-bench-tunables -f platform=macos -f backend_mode=local

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4165528235801 (b08cc6f4d3)
• cmux NEXT nightly 1.0.0-nightly.4165709173201 (7f6fe51cc3)
• cmux NEXT nightly 1.0.0-nightly.4165880669801 (de5076464a)
• cmux NEXT nightly 1.0.0-nightly.4166004947701 (abcde7ea04)

[blacksmith-sh]

Found 2 test failures on Blacksmith runners:

Failures

Test	View Logs
test_every_compile_job_names_its_warm_sources (main.ThisTree.test_every_compile_job _names_its_warm_sources)/ test_every_compile_job_names_its_warm_sources (main.ThisTree.test_every_compile_job _names_its_warm_sources)	View Logs
test_the_repository_highlights_validate (main.DigestTests.test_the_repository_highl ights_validate)/ test_the_repository_highlights_validate (main.DigestTests.test_the_repository_highl ights_validate)	View Logs

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

[lawrencecchen]

Base finding, not this PR: nightly-guards / workflow-guard-tests / release-notary fails on tests/test_ci_warm_sources.py (stable-next.yml:swift-suites is exempt but its steps wire ['deriveddata-reuse']). The same check fails the same way on the feat-cmux-next push run of 0af7ab7 (run 41621534657); its culprit is #33549 (864f4a3), and #33648 fixes it. This PR changes no workflow or test-infrastructure file (5 files: Settings/DebugTunables.swift, Settings/DebugSettings.swift, the TunableStore.swift doc comment, plans/cmux-next/debug-settings.md, the new DebugTunablesActivateTests.swift).

[lawrencecchen]

Review (claude-subagent): FIX at 3be0406

Independent review subagent (correctness first, then repository rules). The code was found correct: register before activate, active stored releasing after the state, file: nil leaves loading false and no writer, the where !store.isActive guard skips DEV and NIGHTLY, the default .shared keeps every other caller as before, and every new line is under #if DEBUG || CMUX_BENCH_HOOKS. Findings, all fixed in 90b21bb:

1. (medium, doc) debug-settings.md said dragResize's synced choice applies if set "before the launch's first Chromium page". Chromium is the default engine, and ChromiumWarmup starts it about 3 s after launch at the first idle moment, so activate and set must come before that.
2. (low-medium, doc) A reader that tracked a key before activate never hears of a change: SidebarTunablesWatch keeps its 8 launch row values until a relaunch (only sidebar.rowCloseSpot follows), and an ObservationStream loop started before stays put.
3. (low, doc) icons.family reaches agent pages made after the set; four DevTools-gated tunables were not named (home.chiefViewGallery, sidebar.sections.localPrototype, home.flightRecorder and its captures).
4. (low, test) The state["file"] null check proved nothing (DebugSettingsService.fileURL, null on the red ref too); removed. activateLeavesAStoreThatIsOnAsItIs passes with or without the guard: not a correctness risk.

[lawrencecchen]

Base findings, not this PR: cmux-next swift test (job 125859300187, head 90b21bb) fails in 9 suites, none of which reads the tunable store or calls debug.tunables: SidebarLookContractTests, OmnibarFocusLayoutShiftTests, TabSelectionKeepsTabWidthsTests, PaneChromeEvenInsetTests, BrowserChromeIconsTests, MarkdownLiveShareTests, TabSwitchWorkBudgetTests, SyncBarrierUnderLoadTests, WindowCleanupCoverageTests.

• The same job on an unrelated PR in the same hour (lane/preview-cli, run 41618184473) fails 7 of them: BrowserChromeIconsTests, PaneChromeEvenInsetTests, SidebarLookContractTests, SyncBarrierUnderLoadTests, TabSelectionKeepsTabWidthsTests, TabSwitchWorkBudgetTests, WindowCleanupCoverageTests.
• SyncBarrierUnderLoadTests is a known base red from Load no longer fails app requests: identify skips the daemon queue, the read barrier waits on liveness, the CLI waits for a live app (#32724) #33273 (the router never publishes a loaded topology); its owner has branch base-red/sync-barrier-test.
• WindowCleanupCoverageTests lists 7 other suites as offenders; DebugTunablesActivateTests is not one (it opens no window).
• DebugTunablesActivateTests passed in this job (changed suites first, 4 of 4) and in fleet step f1ac080a739723faf9edf6e6 at this head.

[lawrencecchen]

Review (claude-subagent): FIX at 90b21bb

Second independent review subagent (the repair of the first review's doc findings, then a fresh pass over the whole diff). Code: no defects found (register writes under the state mutex before the releasing store of active; the action and method exist only under #if DEBUG || CMUX_BENCH_HOOKS; each test uses its own store; the removed file assertion passed on the red ref, so the failing set is the same). Doc findings, all fixed in 9145aba:

1. (medium, doc) "Chromium starts about 3 s after launch" was wrong: every cmux page shows in Chromium by default (browser.defaultEngine, agent.chromium), and the launch's New Tab spare page starts Chromium about 250 ms after the first frame (NewTabSparePool -> PageWebView.startChromium -> CEFRuntime reads dragResizeAtLaunch), so a bench build cannot count on reaching synced with activate.
2. (nit, doc) The first New Tab after a set takes the spare page made at launch, which keeps the old icon family.
3. (nit, doc) An observation made before activate stays silent only when it reads tunables alone.

[lawrencecchen]

Review (claude-subagent): SHIP at 8893b5a

Independent review subagents, correctness first, then repository rules. Four passes:

• 3be0406 (the fix): FIX, doc only. Code correct: register writes under the state mutex before activate stores active (releasing); file: nil leaves loading false and no writer, so nothing is written; the where !store.isActive guard skips DEV and NIGHTLY; the default store: .shared keeps every other caller as before; every new line is under #if DEBUG || CMUX_BENCH_HOOKS, so Release, RC and NIGHTLY are unchanged.
• 90b21bb: FIX, doc only (Chromium starts through the launch's New Tab spare page about 250 ms after the first frame, not ~3 s; the launch spare keeps the old icon family; an observation made before activate stays silent only when it reads tunables alone). No code defect.
• 9145aba: SHIP with three doc nits (not every cmux page is Chromium; the other paths that take the launch spare; ready_after_launch_s says when Chromium started).
• 8893b5a (this head; only plans/cmux-next/debug-settings.md over 9145aba): SHIP. Optional, not applied: the doc could also name the two pages that stay on WebKit (the update island, the icon picker).

Tests: each of the 3 tests that need activate fails without the fix for the right reason (red step 36c6e3dff7bf61a7f25ca21e at 9316d72); each test uses its own store and none sleeps. Green 4/4 at 3be0406, 90b21bb and 9145aba (the head's Swift). Not checked by the reviewers: no compile or test run of their own (fleet results above), no bench build run.