flutter_perf_radar 0.1.1

On-device performance and stability tracer for Flutter. Frame timing, startup, error capture, and main-thread stall detection in debug/profile; complete no-op in release.

flutter_perf_radar

On-device performance and stability tracer for Flutter. Measures frame timing, detects jank, captures unhandled errors, watches for main-thread stalls, and counts per-subtree widget rebuilds — in debug and profile builds. Complete no-op in release — no guards, no build flavours required.

---

Installation

dependencies:
  flutter_perf_radar: ^0.1.0

---

Quick start

1. Initialise in main()

import 'package:flutter_perf_radar/flutter_perf_radar.dart';

void main() async {
  WidgetsFlutterBinding.ensureInitialized();
  await PerfRadar.init(PerfRadarConfig.standard());
  runApp(const MyApp());
}

PerfRadarConfig.standard() enables the engine in debug and profile builds (kDebugMode || kProfileMode) with a 250 ms stall threshold.

2. Add the overlay badge (optional)

// In your root widget's build method:
return PerfRadar.overlay(child: const MyApp());

The draggable badge shows live frame count and jank count. Tap to open the full PerfRadarScreen dashboard.

3. Instrument custom operations

// Synchronous — span is recorded automatically.
final result = PerfRadar.trace('parse_json', () => jsonDecode(raw));

// Async.
final user = await PerfRadar.traceAsync('fetch_user', () => api.getUser(id));

// Manual start/stop for callback-bounded code.
final handle = PerfRadar.start('image_decode');
decoder.decode(bytes, onDone: () => handle.stop());

4. Open the inspector screen

Navigator.of(context).push(
  MaterialPageRoute(builder: (_) => const PerfRadarScreen()),
);

The screen shows frame stats, span statistics with histograms, stability counters (errors + stalls), and per-key rebuild counts.

---

Frame timing and jank

PerfRadar hooks into SchedulerBinding.addTimingsCallback. Every frame duration is recorded. Frames longer than jankThresholdMicros (default 16 667 µs ≈ 60 fps) are counted as jank.

final stats = PerfRadar.frameStats;
print('frames: ${stats.frameCount}  jank: ${stats.jankCount}');

Resetting the counters

Frame and jank counts, the recent-frame ring, and the build/raster/total latency histograms all accumulate since launch. Reset them to zero to measure a specific interval — a single screen transition, a scroll, one network round-trip — instead of since-launch totals:

PerfRadar.resetFrameStats(); // start of the window
// ... exercise the code path you want to measure ...
final stats = PerfRadar.frameStats; // stats for just this interval

PerfRadar.resetFrameStats() clears the counters, the recent-frame ring, and every latency histogram; jankThresholdMicros is left untouched and the engine keeps running. It delegates to FrameStats.reset() on the underlying accumulator.

Three ways to trigger a reset:

• Programmatically — PerfRadar.resetFrameStats() from your own code.
• From the dashboard — the Frames tab shows a reset button (FramesTab.onReset, wired by PerfRadarView to resetFrameStats() plus an immediate refresh) whenever it is displayed inside PerfRadarScreen.
• Over the VM service — the ext.perf_radar.resetFrames extension (registered alongside ext.perf_radar.snapshot) calls PerfRadar.resetFrameStats() and acknowledges with {"reset": true}, so DevTools or any VM service client can zero the counters remotely.

All reset paths are no-ops in release builds.

---

Stability: errors and stall watchdog

Unhandled errors are captured via FlutterError.onError and PlatformDispatcher.instance.onError:

final snapshot = PerfRadar.stabilitySnapshot;
print('errors: ${snapshot.errorCount}  stalls: ${snapshot.stallCount}');

for (final err in snapshot.recentErrors) {
  print('${err.timestamp}: ${err.summary}');
}

The stall watchdog fires a periodic heartbeat on the main isolate. When the heartbeat is delayed by more than stallThresholdMicros, a StallRecord is emitted and retained:

for (final stall in snapshot.recentStalls) {
  print('stall: ${stall.durationMicros} µs at ${stall.timestamp}');
}

---

TracedSubtree — rebuild counting

Wrap any widget subtree to count how many times it rebuilds:

TracedSubtree(
  label: 'home_feed',
  child: const HomeFeed(),
)

Each rebuild increments the span counter for key rebuild:home_feed. Read counts from the span snapshot:

final snap = PerfRadar.snapshot();
for (final entry in snap.stats.entries) {
  if (entry.key.name.startsWith('rebuild:')) {
    print('${entry.key.name}: ${entry.value.histogram.count} rebuilds');
  }
}

When PerfRadar is disabled, TracedSubtree is a transparent pass-through with zero overhead.

---

Configuration reference

Field	Type	Default	Description
enabled	bool	kDebugMode || kProfileMode	Master switch.
showOverlay	bool	false	Whether PerfRadar.overlay() renders the badge.
jankThresholdMicros	int	16667	Frames longer than this are counted as jank (~60 fps).
stallThresholdMicros	int	250000	Main-thread delays longer than this are counted as stalls.
maxStallsRetained	int	50	Rolling buffer depth for stall records.
maxErrorsRetained	int	100	Rolling buffer depth for error records.

await PerfRadar.init(PerfRadarConfig(
  enabled: kDebugMode || kProfileMode,
  showOverlay: true,
  jankThresholdMicros: 8333,  // 120 fps threshold
  stallThresholdMicros: 100000,
));

---

Debug/profile-only guarantee

The engine starts only when PerfRadarConfig.enabled is true and the build is not release (kPerfEnabled guard). In release builds every call (init, trace, traceAsync, start, overlay, frameStats, stabilitySnapshot) is a synchronous no-op returning a safe default. Nothing is conditionally compiled — no tree-shaking or build flavours required.

---

Features

• Frame timing and jank detection via SchedulerBinding timing callbacks.
• Duplicate call detection — pass an optional dedupKey to trace/traceAsync/start, surfaced as a "N dup" count in the trace detail.
• Stall correlation — tapping a stall opens a detail screen correlating its blocking window with the instrumented spans that overlapped it.
• Stall watchdog — periodic heartbeat detects main-thread freezes above a configurable threshold.
• Error capture — FlutterError.onError + PlatformDispatcher.onError with rolling retention.
• Span instrumentation — trace / traceAsync / start → SpanHandle backed by radar_trace histograms and Zone-based async nesting.
• TracedSubtree — zero-overhead rebuild counter for any widget subtree.
• PerfRadarScreen — self-contained dark-theme dashboard with frame, span, stability, and rebuild panels.
• Draggable overlay badge — live frame/jank indicator without leaving the running app.
• Zero-throw contract — the engine never throws into the host app.

---

Related packages

Package	Purpose
radar_trace	The underlying span/histogram engine used by this package.
radar	Umbrella: one import for both flutter_leak_radar + flutter_perf_radar.
flutter_leak_radar	On-device memory leak detector — heap growth, precise retention, overlay.

---

License

MIT — see LICENSE.