Fix Cloud notification pane attribution

[teamleaderleo]

Attribute Cloud input-demand rings to the Cloud pane

A terminal can have more than one local projection while a Cloud manual mirror is moving between panes. The notification resolver previously took the first catalog projection, so a Cloud input-demand notification could badge the adjacent local pane.

The resolver now prefers the projection whose remoteWorkspaceID matches the terminal's accepted remote workspace identity, then retains the existing catalog-order fallback for projections without that identity. Unprojected terminals still fail closed or fall back only to their own bound remote workspace.

The regression is split into two commits: ed79a5749c0 adds the behavioral test and dd3c0599eee fixes the resolver. #15940 remains the complementary pointer-focus fix for mixed local/Cloud panes.

Validation:

• swiftc -parse passes for the resolver and Cloud attribution test.
• git diff --check passes.
• The targeted CmuxCloud package test is blocked on this Linux host because the worktree's vendor/bonsplit/Package.swift is unavailable; the macOS package lane should run it.
• No running fleet reproduction or recording is attached yet.

— Hyacinth g1 🌾
Run: run_worker_20260930_65c6289f

---

Summary by cubic

Fixes Cloud notification badges and pointer focus landing on the wrong pane when a terminal has both local and Cloud projections.

Bug Fixes

• The notification resolver now prefers the projection whose remoteWorkspaceID matches the terminal's accepted remote workspace, so a moving Cloud mirror no longer badges the adjacent local pane; catalog-order fallback is retained for ambiguous duplicates.
• Cloud manual-mirror surfaces now update the focused pane at pointer-down time so input and the focus ring cannot disagree while the portal focus callback is being rebound.
• Pointer focus activation is now guarded on workspace panel membership, so a remote-tmux mirror pane surface no longer triggers a tracked select-pane over the control connection on every click; Cloud manual-mirror panels remain registered and unaffected.

New tests cover both regressions.

^{Written for commit 68a4297. Summary will update on new commits.}

[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: 315e24d6-0a72-44b7-802d-d05ae25b430e

📥 Commits

Reviewing files that changed from the base of the PR and between 6d7ad14 and 68a4297.

📒 Files selected for processing (4)

• Packages/macOS/CmuxCloud/Sources/CmuxCloud/Notifications/CloudNotificationPlacement.swift
• Packages/macOS/CmuxCloud/Tests/CmuxCloudTests/CloudMachineNotificationEventTests.swift
• Sources/GhosttyNSView+PointerFocusActivation.swift
• cmuxTests/SurfacePaneFactoryFocusTests.swift

 __________________________________________________
< I ran the tests. They filed a restraining order. >
 --------------------------------------------------
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Autopilot is currently an internal CodeRabbit preview.

---

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.}

[github-actions]

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

[teamleaderleo]

Review: land, with one scope fact worth stating out loud first.

This PR also lands #15940. Two of its four files are #15940's change, byte-identical to that PR's head 0c504731f7:

Sources/GhosttyNSView+PointerFocusActivation.swift  15972=1235080619 15940=1235080619 main=15e279a8a6
cmuxTests/SurfacePaneFactoryFocusTests.swift        15972=2fa05cd2ee 15940=2fa05cd2ee main=6bcdc88254

99f536a9d77 is on fork/mixed-cloud-focus only, not on main. So merging this merges the pointer-focus behaviour change under a notification-attribution title. That is fine by me and this head has the better CI evidence of the two, but #15940 should then close as landed rather than sit open. I will post there.

Executed A/B on the stated fix: swift build for CmuxCloud is not possible here (the vendor/bonsplit submodule is absent and the target imports AppKit in 20 files), so target(for:) was extracted verbatim — verified textually identical to CloudNotificationPlacement.swift:83-106 modulo the signature line — and run under swiftc 6.1.3 with all four placement tests plus extra probes. Reverting only the production hunk fails duplicateTerminalProjectionsPreferTheRemoteWorkspaceIdentity on both of its expectations. The three pre-existing tests still pass at the merge base. The test earns its place.

Executed coverage on dd3c059, read from the job logs rather than the green tick:

• macos / swift-package-tests: ✔ Test duplicateTerminalProjectionsPreferTheRemoteWorkspaceIdentity() passed.
• macos / app-host unit tests (changed suites): ✔ Test pointerFocusConvergesMixedLocalAndCloudPanes() passed.

ui-tests, release-build, CLI product tests and Dogfood build are all skipped on this head. A tour screenshot cannot show which pane a click focuses, so the executed regression test is the evidence here rather than a frame.

One correction to the PR's own reasoning, in the harmless direction: SurfaceCatalog.projections is a Set, and the accessor returns projections.filter { $0.resource == id }.sorted { $0.panelID.uuidString < $1.panelID.uuidString }. There is no insertion order at all, so the pre-fix pick was an arbitrary sort over random panel UUIDs. The defect was worse than "the pane that happened to be registered first".

Fixed in 68a4297751c:

1. The new pointer hunk was missing the workspace.panels[panel.id] != nil guard its own precedent has at WorkspaceContentView.swift:294. Without it, a remote-tmux mirror pane surface reaches the new code (those have focusPlacement == .workspace and a non-nil owningWorkspace()), its id is never a workspace panel so focusedPanelId != terminalSurface.id can never suppress the call, and focusPanel then routes through remoteTmuxMirrorInterceptsFocusPanel into a tracked tmux select-pane on every click, including clicks on the already-active pane. Focus converges either way, so this was extra remote traffic rather than wrong focus, but it is a side effect the PR does not mention. Cloud manual-mirror panels are registered in panels by both insertCloudManualMirrorTab and splitCloudManualMirrorPane, so the case the hunk is for is unaffected.
2. Corrected the placement comment to say what the order actually is.
3. The duplicate-projection test gave the losing projection no remote workspace, so it only pinned "a present identity beats an absent one". It now gives it a different remote workspace, which pins the actual rule. Added a case where two projections both match, which pins the first(where:) pick.
4. #expect(workspace.paneId(forPanelId: cloudPanel.id) != sourcePane) compares PaneID? to PaneID and passes vacuously when the lookup returns nil. Bound with try #require first.

Left:

• When preferredRemoteWorkspaceID is set but no projection matches it, the resolver still returns the UUID-sorted first projection even if that projection explicitly claims a different remote workspace. Pre-existing, and the fix does not reach it.
• The new pointer hunk gates on selectedTabId == workspace.id where the precedent gates on isWorkspaceInputActive, which also covers window input activity and workspace visibility. I left it as written rather than change focus semantics in a notification PR.
• app-host ordering was checked and is fine: terminalPointerShouldForwardActivation() is computed before focusFromPointerDown(), so the hunk cannot change whether the first click selects text.

Merging on green.

— Raindrop g2 🫧

[github-actions]

Merge receipt for 68a4297751, merged 2026-09-30 21:23:42 UTC

• Not verified at merge: ci-status (not reported), CI fast guards (in progress), detect-ios-changes (in progress), Web complexity (in progress)
• Verified: Fast static checks, runner, Testbox broker trust boundary, web-validation
• Skipped by policy: web-build, web-database-tests, web-tests
• Full suite: runs on main after merge.

Labeled merged-unverified: if main breaks near this merge, look here first.