d995750065
Keep connected MCP status dots green without a permanent pulse. Pending connect/disconnect operations retain their warning pulse and spinner; the shared reduced-motion and metadata refresh policies are unchanged. Add real StatusTab/SUID browser regressions at 150% scale, a bounded historical/current Chromium profiler, and measurement-limit documentation. Five browser scenarios, UI typecheck/build and repeatable profiler assertions passed. Fix the gatekeeper P2 by creating the temporary evidence parent on clean machines; the second independent review reported zero findings. The change addresses the idle-animation source in #804, not its unconfirmed generation GPU symptom or #805. Remote CI was pending at the explicitly requested admin merge.
2.9 KiB
UI motion profiling (#804)
Connected MCP servers describe a stable state, not active work. Their status dots stay green without pulsing. Only a pending connect/disconnect retains the warning pulse and spinner; success and failure both stop that feedback. This is shared UI behavior for web, Electron and Tauri.
The existing global prefers-reduced-motion: reduce rule limits animations to
one near-instant iteration and shortens transitions. The MCP fix does not add a
second motion policy or alter refresh/cache semantics.
Regression and bounded profiling
node --import tsx --test --test-concurrency=1 packages/ui/tests/browser/mcp-motion.test.ts packages/ui/tests/browser/status-panel.test.ts
node scripts/profile-ui-motion.mjs eb1bf48e
The optional Git revision substitutes only the historical MCP component. The
remaining source and installed frontend dependencies are identical. The script
writes Chromium traces, screenshots and results.json under a temporary
opencode/issue804-motion-* directory. No native daemon or user data is used.
Fixtures use the real StatusTab, SUID switches, full stylesheet and SessionView with the native event dispatcher. MCP calls and session HTTP are simulated. The session fixture seeds 620 short history messages, then samples idle, token-free single/grouped reasoning, 24 text deltas with a 50 ms delay between deliveries, and completion. Both motion settings use a 150% device scale.
Observed in the September 30 isolated Chromium run:
| State | Infinite animations | Layout activity in ~1.2 s sample |
|---|---|---|
| Eight connected MCPs, before fix | 8 | 0 layouts (opacity animation) |
| Eight connected MCPs, after fix | 0 | 0 layouts |
| Collapsed MCP section | 0 | 0 layouts |
| Single reasoning step, no deltas | 0 | 0 layouts |
| Grouped reasoning, no deltas | 1 spinner | 0 layouts |
| Grouped reasoning, reduced motion | 0 | 0 layouts |
| Session complete | 0 | 0 layouts |
Text streaming still generates rendering work with reduced motion: 52 layouts were recorded for 24 deliveries in both settings (~1.7 s including delivery overhead). The mounted transcript remained at 16 rows, rather than 620. These observations do not establish a full-transcript redraw on every animation frame.
Limits / follow-up
This is a headless Chromium rendering sample, not a WebView2/NVIDIA GPU measurement. Compositor-only animation can consume GPU without producing main thread layout/paint counters. No GPU percentage reduction is claimed.
The session fixture omits application/sidebar chrome and does not supply the full-history timeline RPC. Long reasoning, large tool results, remote HTTP bursts and full-app freezes still require a representative native recording. The generation symptom of #804 and the remote Electron freeze in #805 remain separate, unconfirmed problems; this change must not close either as fully fixed.