processFrame method
Process one presented frame from FrameTiming.
Called once per frame on every tier (no VM needed), at display rate, for every enabled detector except the frame-timing producer itself. A batch of N timings delivered in one engine callback yields N calls in presentation order. Keep the work constant-time and allocation-free; accumulate here and emit from finalizeScan. A detector that throws is reported once and skipped until the next structural scan. Default is a no-op.
Implementation
@override
void processFrame(FrameStats frame) {
if (!_isEnabled) return;
if (_insideStartupWindow()) return;
_windowFrameCount++;
final atUs = frame.vsyncStartUs ?? frame.rasterFinishUs;
final newest = _newestFrameAtUs;
if (atUs != null && (newest == null || atUs > newest)) {
_newestFrameAtUs = atUs;
}
final uiUs = frame.uiDuration.inMicroseconds;
final rasterUs = frame.rasterDuration.inMicroseconds;
if (uiUs <= 0 || rasterUs <= 0) return;
// The frame missed its deadline on the raster thread, and raster, not
// the UI thread, was the larger cost.
if (rasterUs <= frame.frameBudgetUs) return;
if (rasterUs <= maxFrameRasterFloorUs) return;
if (rasterUs <= rasterMultiplierThreshold * uiUs) return;
_dominantCount++;
if (rasterUs > _worstFrameRasterUs) _worstFrameRasterUs = rasterUs;
final i = _ringWrite;
_ringRasterUs[i] = rasterUs;
_ringUiUs[i] = uiUs;
_ringHasAt[i] = atUs != null;
_ringAtUs[i] = atUs ?? 0;
final identityUs = frame.rasterFinishUs ?? frame.vsyncStartUs;
_ringHasIdentity[i] = identityUs != null;
_ringIdentityUs[i] = identityUs ?? 0;
_ringFresh[i] = true;
_ringWrite = (i + 1) % _ringCapacity;
if (_ringCount < _ringCapacity) _ringCount++;
}