Repository navigation
Harden status recovery and add staged toolbar updates - #7
Vivek-Raman wants to merge 1 commit into
Conversation
- Add cross-platform checks and tagged-release verification - Improve retention, pin recovery, integrations, and viewer health - Upgrade frontend tooling and document validation workflows
Vivek-Raman
left a comment
Vivek-Raman
left a comment
There was a problem hiding this comment.
Found two verified P2 issues on modified lines: repeated migration backups during mixed-version operation, and native always-on-top updates that can execute out of settings-commit order. Both findings were independently verified. No other actionable findings met the confidence threshold across correctness, architecture, security, performance, dependency declarations, repository conventions, and history.
| .get("points") | ||
| .and_then(|v| v.as_array()) | ||
| .is_some_and(|points| { | ||
| points.iter().any(|point| point.get("pinned").is_some()) |
There was a problem hiding this comment.
[P2] Avoid a new migration backup for every false-only legacy report
A still-running pre-PR v2.5 MCP serializes pinned: false on every ordinary report. The new reader treats the field's presence as a migration, so each subsequent legacy report reaches the UUID backup creation below. Two independent temporary-home reproductions produced one additional full-file backup per report, even with no pinned points. These files are never trimmed and can each approach the 1 MiB status limit, bypassing the bounded status/history storage.
Only preserve a migration backup when actual inline pins need preservation, or keep a bounded migration backup. False-only fields can still be stripped without creating another backup:
| points.iter().any(|point| point.get("pinned").is_some()) | |
| points.iter().any(|point| { | |
| point.get("pinned").and_then(serde_json::Value::as_bool) | |
| == Some(true) | |
| }) |
| owner.set_pinned(pinned)?; | ||
| app.get_webview_window("main") | ||
| .ok_or_else(|| "Main window unavailable".to_string())? | ||
| .set_always_on_top(pinned) |
There was a problem hiding this comment.
[P2] Order native pin effects with the settings commits
Moving these handlers onto parallel blocking workers allows A to persist true, B to persist false and enqueue the native false update, then A to enqueue true. SettingsOwner releases its mutex before the native call, and the settings checkbox permits overlapping requests. Tauri queues worker-originated native calls independently, so the saved setting and UI can finish unchecked while the viewer remains always on top. The previous synchronous handlers ran serially on the main thread.
Serialize the settings commit and native enqueue through one shared PinStateCoordinator used by both set_pinned and reset_settings, which has the same gap. Reading the latest snapshot before enqueueing does not close the race unless both steps share that ordering boundary.
Sideglance could freeze during filesystem and agent CLI operations, lose user data during interrupted updates, and show stale reports without explaining their state. This change makes status storage recoverable, moves blocking operations off the UI thread, and adds an explicit check → download → install flow to the main toolbar.
Changes
Validation
Limits and rollout
Restart existing agent sessions after upgrading so they load the new MCP executable. Preserve both status files during backup or recovery; automatic-deletion history is bounded rather than permanent storage.
Two low upstream npm build-tool advisories remain with no fix reported by npm; there are no high, moderate, or critical npm findings. RustSec tooling was unavailable locally. Windows/Linux runtime behavior and native window profiling remain for CI/manual verification. Background throttling remains unchanged pending native event-delivery validation.