[lawrencecchen]

Summary

REOPEN-F7 (first window frame after a reopen, target p95 < 300 ms on the realistic launch bench). Profile 8 (xctrace, cmux19, job 72f71795e8b896e2f54c5ad4, 4 warm realistic reopens, both earlier fixes in) puts SidebarIconButton.init at 9-12 ms (profiled) of the window build before the first frame. renderIcon is about 4 ms of that, and the buttons' own layout another 2 ms.

Cause: each workspace row makes 5 icon buttons (close, audio, pin, tree fold, more) that stay hidden until the pointer is on the row. Each button made its template image (an NSImage with a bitmap representation per scale) at init.

Fix (1): renderIcon returns while the button is hidden or in a hidden row. It runs when the button shows (viewDidUnhide, which AppKit also sends when an ancestor unhides), joins a window, lays out, or is asked its intrinsic size. An icon, label or symbol change on a hidden button draws when it shows. A shown button has its image before it draws, so the look does not change (Sidebar-Look: no visual change). The shared glyph bitmaps from #32481 do not change.

Fix (2), from review at bc142a2 (P3): the row's Unpin button was not hidden at init, so it made its image when the row joined the window, before the row's first layout hid it. It now starts hidden, as the close, audio and tree fold buttons; it shows only through layoutPin / setHoverShown, which unhides a hidden view the same way. The More button is an OverflowButton, not changed here. So the saving is the 4 SidebarIconButtons per row (close, audio, pin, tree fold), not 5.

Gain

Gain: 108-112 fewer icon images per launch (4 hidden icon buttons per workspace row: close, audio, pin, tree fold; 24-25 rows plus 3 spare rows in the realistic launch, from debug.sidebar_perf row_views_made). Count tests: SidebarIconButtonLazyImageTests red on base (steps d90f890c, 9ffa7f7a), green on head (step 9d21cdb9). Time gain 1-4 ms main-thread CPU, under the bench's noise; no regression (A/B cmux21, 20 launches per arm). Accepted under the HOT PATH AUDIT rule (handoff/hot-path-audit/README.md item 3: a deterministic count when a time is too noisy).

Testing

• Regression test SidebarIconButtonLazyImageTests: a hidden button, and a button in a hidden row, have no image after layout and get one when shown; a shown button in a window, and a loose one asked its size, have an image; a new icon on a hidden button waits too.
• Red for the pin start state: 31b18bf (the new test aNewRowsHoverControlsStartHidden before the pin change): step 9ffa7f7a54d23a8fec1eb6b5 failed as expected (the pin button is shown at init).
• Green at 8a8ba37: step 9d21cdb9a7be734d83988d7a passed all 5 suites (SidebarIconButtonLazyImageTests, ThemedTooltipTests, SidebarRowTitleAndCloseTests, SidebarSearchColumnTests, SidebarTrailingColumnTests). Compile at 8a8ba37: job b2174381738b6f74625372b7 passed.
• Red: ref ff8da2e (base e674d0b + this test, no fix; compile job 35499498 passed): step d90f890cd6c4c148d967f04e (Blacksmith run 41597837375) failed as expected: 3 of 4 tests, hidden buttons have their 14 pt icon at the base.
• Green at d567840 (step 1844f51d, run 41597808453) found a bug in the fix: a loose button asked its size kept AppKit's 1 x 1 placeholder image (the check was image == nil). Fixed at 5263ab5 (renderedSize == 0). Merged origin/feat-cmux-next 8f963a6 (plain merge): head bc142a2. Green at bc142a2: step f4ebf1cdbb7008170ae1ab1b passed, all 3 suites (SidebarIconButtonLazyImageTests, ThemedTooltipTests, SidebarRowTitleAndCloseTests), exit 0.
• Compile at 5263ab5: job 7acd09168f78ddd4b2bdcbee passed (CmuxNext, app, cmux-tui). At bc142a2: job 361c46840c1e8bd9270193f4 passed (CmuxNext, app, cmux-tui).
• A/B on cmux21 (dialog on screen; launch timing needs no clicks), warm realistic launch, steps A B A B, 10 launches each, ms from main() to the first window frame. A = feat 8f963a6 (jobs 1f23152e, 678f05fc), B = 8a8ba37 (jobs 9f0199ab, 503a5eec):
  • A: n=20 p50 271.6 / p95 312.0 / max 329.2, 3 over 300. B: n=20 p50 271.5 / p95 299.3 / max 311.4, 1 over 300. Start tier Super in all 40.
  • Main-thread CPU (main() to frame) p50 by step in run order: A 234.5, B 232.2, A 231.2, B 225.8. Every later step was faster, whatever the arm; with that drift taken out, B saves about 1-4 ms of main-thread CPU. The p50 does not move, and the lower B tail follows the drift. No regression; no wall-time gain this bench can resolve.
  • Sidebar: 24-25 row views per launch for 24 seeded workspaces (+3 spare) in both arms.
• Launch profiles of feat and of REOPEN-F7: a hidden sidebar icon button makes its image when it first shows #33112+REOPEN-F7: a hidden sidebar icon button registers its tooltip when it first shows (stacked on #33112) #33113 on cmux21 (scratch specs reopen-f7-prof9A/B): running, to show the removed work directly.
• Before the feat merge (bc142a2), each Swift suite ended "failed (exit 1)" about 660 s after its tests passed: the test-bundle run-loop guard held xctest's main thread, fixed on feat by test(next): the run loop guard never holds the XCTest leg's main loop (signal-6 abort on some fleet Macs) #33023/base red: the test run-loop guard holds only the Swift Testing helper, never xctest (xctest signal 6 after every CmuxNext suite) #33013 and test(cmux-next): the XCTest-phase regression test also starts xctest without -XCTest (follow-up of #33013) #33276. The results above before bc142a2 are read from the test lines.

Review

Review subagent: SHIP at 8a8ba37 (comment 6601535817); the earlier SHIP at bc142a2 (comment 6600179225) found the pin P3, fixed at 8a8ba37.

Observed / Not run

• Observed: the fleet runs above.
• Not run: the sidebar budgets suite (split-spring.sh --suite=budgets) runs on cmuxs-Mac-mini, which has the LaunchServices fault (STATUS 7775), so its timings are not valid now. No gallery capture: no look change.

Changelog

none

[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 8a8ba37: 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=pr33112 -f ref=fix/reopen-f7-lazy-button-image -f platform=macos -f backend_mode=local

web and size checks: green

Shipped in:

• 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)
• cmux NEXT nightly 1.0.0-nightly.4166228559901 (03b4a57138)

[blacksmith-sh]

Found 1 test failure on Blacksmith runners:

Failure

Test	View Logs
test_every_listed_hook_is_one_glaeda_released_with_the_check (main.SigningHooksInSt epWithGlaedaTests.test_every_listed_hook_is_one_glaeda_released_with_the_check)/ test_every_listed_hook_is_one_glaeda_released_with_the_check (main.SigningHooksInSt epWithGlaedaTests.test_every_listed_hook_is_one_glaeda_released_with_the_check)	View Logs

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

[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.

[lawrencecchen]

Review (subagent opus): SHIP at bc142a2

• Every path that shows a button makes its image before it draws. viewDidUnhide fires each time isHiddenOrHasHiddenAncestor goes from true to false, so it covers the button's own unhide, an ancestor's unhide (setHoverShown, the pool's dequeue, spare rows) and both unhide orders. Joining a window while visible renders in viewDidMoveToWindow. An alpha-0 button that is not hidden renders when it joins the window. No change from the base there.
• The hidden guard returns before renderedSize is read or written. A density or tunable change, or an icon, label or symbol change made while the button is hidden, is applied by the size compare on unhide. Nothing is lost.
• No caller reads a SidebarIconButton's .image. Row, card and header layouts set frames and do not use intrinsic size. The emoji button in the stack view stays visible and gets its image from the intrinsicContentSize override. Snapshot and gallery paths draw in-window, where hidden views do not draw. VoiceOver uses setAccessibilityLabel, which init still sets.
• The red/green logic is sound. On the base, 3 of 4 tests fail (test 3 passes there, as it should) and all pass at the head.
• P3 (perf claim only): pinButton is not hidden at init, so it still makes its image when its row joins the window, before layout hides it. moreButton is an OverflowButton, which this PR does not change. Only close, audio and tree fold get the full saving. The pending A/B will show the real gain.
• P3: the intrinsicContentSize getter sets image once, which invalidates the intrinsic size from inside a size read. This is harmless, but it is a side effect in a getter.

[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.

[lawrencecchen]

Review (subagent opus): SHIP at 8a8ba37

• New since bc142a2: only pinButton.isHidden = true in WorkspaceRowView.init plus the test aNewRowsHoverControlsStartHidden. SidebarIconButton.swift and the rest of the PR are unchanged from the reviewed SHIP (merge-base 8f963a6, 3 files).
• Show paths still work: layoutPin -> setHoverShown treats a hidden pin as appears (alpha 1 when not fading, a fade-in from 0 when fading, the same as close/more), and viewDidUnhide makes the image. The frame is set after that in both close-spot branches. placeCloseSpot's offset only moves frames. prepareForReuse does not touch the pin. The tests that expect a visible pin (SidebarSearchColumnTests, SidebarGridAlignmentTests, SidebarTrailingColumn*) lay out first.
• No visual change: before, the row's first layout already hid an unshown pin, and its frame was .zero until then. Now it starts in that hidden state, so the person sees nothing different.
• P3 (no action): the new test checks a row that is not in a window, so it proves the start state, not "no image when the row joins a window". The red/green on the pin's start state (9ffa7f7a / 9d21cdb9) matches the change. I did not compile or run anything; I read the code and used the author's fleet results.