fix: replace fixed 15s WebSocket reconnect with exponential backoff + jitter

[kodjima33]

Summary

Replaces the fixed 15-second WebSocket reconnection timer with exponential backoff and full jitter to prevent synchronized reconnect storms.

Changes

• Added _reconnectAttempt counter and _maxBackoffSeconds = 120
• Added _getReconnectDelay() — exponential backoff (1s→2s→4s→...→120s cap) with full jitter
• Replaced Timer.periodic(Duration(seconds: 15)) with single-shot Timer using computed delay
• Reset _reconnectAttempt = 0 on successful connection in onConnected()

Note: The diff includes some auto-formatting changes alongside the substantive logic changes.

Fixes #5527

[greptile-apps]

Greptile Summary

This PR replaces the fixed 15-second Timer.periodic WebSocket reconnection in CaptureProvider with a single-shot Timer using exponential backoff (1s→2s→4s→...→120s cap) and full jitter to prevent synchronized reconnect storms across clients.

• Adds _reconnectAttempt counter and _maxBackoffSeconds = 120 cap
• Adds _getReconnectDelay() implementing the AWS-style "full jitter" pattern
• Replaces Timer.periodic with one-shot Timer using computed delay
• Resets _reconnectAttempt to 0 on successful connection in onConnected()
• Removes the now-unnecessary _keepAliveLastExecutedAt rate-limiting field
• Bug: pow(2, _reconnectAttempt).toInt() is called before min() caps the value — this will crash with UnsupportedError when _reconnectAttempt >= 1024 (reachable after ~34 hours of persistent disconnection)
• The bulk of the diff is auto-formatting changes (indentation normalization) with no behavioral impact

Confidence Score: 3/5

• The PR improves reconnection behavior but has a latent crash bug in the backoff calculation that manifests after prolonged disconnection.
• The core idea is sound (exponential backoff with jitter), and the reset logic is correct. However, the operator ordering bug in _getReconnectDelay() will cause a runtime crash after ~34 hours of disconnection, which is a realistic scenario for a mobile/desktop app. The fix is a one-line change. The rest of the diff is safe auto-formatting.
• app/lib/providers/capture_provider.dart — the _getReconnectDelay() method needs the .toInt() moved after min() to prevent overflow crash.

Important Files Changed

Filename	Overview
app/lib/providers/capture_provider.dart	Replaces fixed 15s periodic WebSocket reconnect with exponential backoff + jitter. The backoff logic has a potential crash: pow(2, n).toInt() is called before the min() cap, which will throw UnsupportedError when _reconnectAttempt >= 1024 (~34h of disconnection). Most other changes are auto-formatting only.

Sequence Diagram

sequenceDiagram
    participant CP as CaptureProvider
    participant T as Timer (single-shot)
    participant WS as WebSocket

    Note over CP: WebSocket disconnects
    CP->>CP: onClosed() / onError()
    CP->>CP: _startKeepAliveServices()
    CP->>CP: _getReconnectDelay()<br/>(2^attempt capped at 120s, with jitter)
    CP->>CP: _reconnectAttempt++
    CP->>T: Schedule Timer(delay)
    T-->>CP: Timer fires
    CP->>WS: _initiateWebsocket()

    alt Socket connects successfully
        WS-->>CP: onConnected()
        CP->>CP: _reconnectAttempt = 0
    else Socket creation fails (null)
        CP->>CP: _startKeepAliveServices()<br/>(reschedule with increased backoff)
    else Socket connects then drops
        WS-->>CP: onClosed() / onError()
        CP->>CP: _startKeepAliveServices()<br/>(reschedule with increased backoff)
    end

Loading

_{Last reviewed commit: a67c19b}

[greptile-apps]

toInt() before min() will crash on high attempt counts

.toInt() is called on pow(2, _reconnectAttempt) before min clamps the value. Since pow returns a double, once _reconnectAttempt reaches 1024, pow(2, 1024) returns double.infinity, and double.infinity.toInt() throws an UnsupportedError. With the backoff cap at 120s and full jitter, attempt 1024 can be reached in ~34 hours of persistent disconnection.

Move .toInt() after the min call so the cap is applied first:

Suggested change

	final baseDelay = min(pow(2, _reconnectAttempt).toInt(), _maxBackoffSeconds);
	final baseDelay = min(pow(2, _reconnectAttempt), _maxBackoffSeconds).toInt();

[beastoin]

@kodjima33 The switch from periodic to one-shot timer means early returns (user not signed in, device not ready) silently abandon the retry chain with no follow-up timer scheduled, which can cause a permanent WebSocket disconnection during active recording until manual intervention. Could you revert this and add tests covering the early-return paths to ensure the retry always re-schedules on transient failures before re-merging?

Let us know if you need a hand with testing.

---

by AI for @beastoin