perf(web): stop rendering hidden terminals

[t3dotgg]

Hidden terminal drawers and panels still build cell snapshots and paint frames for output no one can see.

After construction, pause frame scheduling, snapshots, canvas paint, and cursor timers while hidden. Keep the one-time startup background fill, output parsing, and VT replies. Refit and repaint the current state on reveal, including visibility changes during WASM startup.

Includes Wout Stiens' zero-size canvas guard from PR #9027, with author and co-author credit preserved. Its server and native changes remain separate.

Verified with 73 focused tests, the web typecheck, and targeted lint. Headless tests use the real Ghostty core and renderer to check hidden work, query replies, cursor and selection state, same-size recovery, and delayed WASM initialization. No browser, device, or dev server run.

Refs #9661.

Created with GPT-6 Astra (preview) in Codex.

---

Note

Medium Risk
Touches the Ghostty render loop and terminal focus/resize behavior in the right panel; regressions could show as stale UI or focus quirks when reopening, though behavior is covered by new tests.

Overview
Hidden thread terminals stay mounted for fast switching, but no longer waste work painting off-screen canvases.

GhosttyTerminalSurface gains an optional visible flag and setVisible(), which cancels animation frames, snapshots, and cursor blink timers while hidden while still parsing output and replying to VT queries. On show (or when the mount regains non-zero size), it refits and does a full repaint of the current buffer, including selection state.

The chat UI threads rightPanelOpen into PersistentThreadTerminalPanel → ThreadTerminalDrawer → TerminalViewport, syncing visibility on layout and skipping focus, fit, and resize-driven layout when the panel is closed. Creation can use a getter for visibility so WASM startup respects toggles that happen mid-load.

New headless surface tests cover hidden paint suppression, live query replies, selection preservation/clear-on-reveal, zero-size mounts, and delayed WASM init.

^{Reviewed by Cursor Bugbot for commit 16b81b4. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Skip rendering and fitting for hidden terminal surfaces in GhosttyTerminalSurface

• Adds visibility tracking to GhosttyTerminalSurface so hidden or zero-size surfaces stop snapshotting, painting, fitting, cursor blinking, and focus handling while the terminal core continues to process VT input and replies.
• Wires the right panel's open state from ChatView through PersistentThreadTerminalPanel and ThreadTerminalDrawer down to each TerminalViewport, which syncs visibility to the surface via a ref and layout effect.
• Adds a setVisible setter that cancels queued render frames and cursor timers on hide and refits with a full repaint on reveal.
• Expands surface.test.ts with a real-WASM harness covering hidden paint suppression, selection retention/clear, zero-size mounts, and visibility changes during async core initialization.
• Behavioral Change: hidden surfaces no longer render but still answer terminal queries; zero-size mounts defer rendering until a nonzero resize, even if dimensions return to a prior size; visibility changes during WASM init are applied to the created surface before first paint.

^{Macroscope summarized 16b81b4.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.6 KiB	13.6 KiB	+13 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	−6 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.6 KiB	+19 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.0 KiB	57.1 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	9	10	+1 (+11.1%)	21	✅
Claude	Total thread wire	13.6 KiB	13.5 KiB	−28 B (−0.2%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.5 KiB	6.5 KiB	−31 B (−0.5%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.9 KiB	57.8 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	10	9	−1 (−10.0%)	21	✅

Baseline: 9510390 · PR result: 16b81b4 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This performance change alters the lifecycle of mounted terminals by suppressing snapshots, canvas painting, focus, and cursor timers while the existing panel is hidden, while retaining terminal parsing and replies. The behavior is covered by focused WASM-backed tests, but the production render gate changes when substantial existing work occurs and merits human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Note

🤖 GPT-6 Astra (preview) responding on behalf of Theo

I reviewed both the drawer and right-panel paths, including split terminals and visibility changes during WASM loading. Hidden terminals still parse output and answer VT queries. The new gates stop post-construction snapshots, canvas paint, queued render frames, and cursor timers. Reveal renders the current state once. The existing startup background fill remains.

I independently reran all 73 focused surface/core/renderer/drawer tests. Web typecheck, CI, Cursor, and Macroscope correctness/conventions/UI checks pass on 16b81b4. Automatic eligibility is neutral because rendering lifetimes change. No browser, device, or GPU measurement is claimed.

The source guard's contributor credit remains intact.