feat(sync): automatic one-shot staging reset (#296)

[TimeToBuildBob]

Summary

Kotlin half of #296: an automatic one-shot reset of the phone's push staging db, so pre-ActivityWatch/aw-server-rust#713 duplicate copies stop being re-imported by desktops. Item 2 of ActivityWatch/aw-server-rust#717 (item 1, desktop dedupe, merged in ActivityWatch/aw-server-rust#718).

What it does

• StagingReset.runOnce (Android-free, unit-tested): skips if this hostname's flag is set; otherwise calls the native reset and sets the flag only on {"success": true}. Failures leave the flag unset, so the reset retries on the next sync. It never throws, including UnsatisfiedLinkError.
• SyncInterface: declares external fun resetStaging(hostname), and runs the reset inside the full-sync executor task, after the legacy-folder migration and before syncBoth. That path holds syncInFlight, so the reset cannot race a push writing the same staging db. (The push/pull-all wrappers have no callers and no guard, so they are left alone.)
• AWPreferences.isStagingResetDone/setStagingResetDone(hostname): the flag is keyed per hostname because staging lives under {sync_dir}/{hostname}/.

Dependency: the Rust side is not written yet

resetStaging must be added to aw-sync/src/android.rs (spec in #296: delete test.db/-wal/-shm for the own device dir only, refuse paths outside it). Until a build carries it, the native call throws UnsatisfiedLinkError. That error is caught and logged, the flag stays unset, and sync carries on, so this PR is safe to merge before the Rust PR. The reset then fires on the first sync after the submodule bump.

Verify before closing #296

• Confirm the phone's local android-test / android-unlock buckets are not duplicated. If they are, re-staging reproduces the copies.
• After the Rust side ships: a pull on the desktop peer should add zero events.

Relates to #296, ActivityWatch/aw-server-rust#717, ActivityWatch/aw-server-rust#713, #295.

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Adds automatic staging database reset to sync flow.

The Android changes appear safe to merge before the native reset ships.

What we checked:

• Missing native method stops sync: runOnce catches LinkageError and returns Failed. The caller logs that result, then continues to syncBoth.

Summary

Adds Android-side support for a one-time staging reset before full sync.

• Saves completion separately for each hostname, only after the native call reports success.
• Catches a missing native method so sync can continue and retry the reset later.
• Replaces the broken Mockito tests with direct tests of StagingReset.runOnce.
• The three earlier, unnumbered findings are addressed: preference methods now exist, Mockito imports are gone, and the replacement tests exercise the reset logic.
• The native reset still requires a separate Rust change. No new actionable issues were found.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Full sync takes shared guard] --> B[Migrate legacy folders]
    B --> C{Reset done for hostname?}
    C -- Yes --> G[Run native full sync]
    C -- No --> D[Call native reset]
    D --> E{Reset reports success?}
    E -- Yes --> F[Save completion flag]
    E -- No or missing method --> G
    F --> G
    G --> H[Report sync result and release guard]

Loading

_{Reviews (2) · Last reviewed commit: "fix(sync): make staging reset compile, t..."}

[TimeToBuildBob]

@greptileai review

[TimeToBuildBob]

Ready for maintainer review: all three Greptile P1s are fixed in e0bcfec. The re-review on that head scores 5/5, CI is green (Test + E2E), and every thread is resolved.

I can't merge this one myself (pull-only access on aw-android), so it's waiting on a maintainer.

[TimeToBuildBob]

CI-green and mergeable (Greptile 5/5) — waiting only on a maintainer click.

This PR is ready to merge, but the bot has pull-only access to this repo and can't self-merge — surfacing it here so it isn't lost. The monitoring loop will stop re-flagging it now that this note is posted.

[TimeToBuildBob]

🤖 AI code review

Safe to merge — 1 P1 disposed (accepted-tradeoff)

_{Updated after inline dispositions on finding threads — this is the current state; the verdict below is frozen at review time and is kept as the historical record of that pass.}

Finding disposition

Finding	Severity	State
mobile/src/main/java/net/activitywatch/android/StagingReset.kt:43	P1	accepted-tradeoff

Adds a one-shot staging-db reset for the full-sync path. StagingReset.runOnce is a JVM-free helper that calls a native resetStaging(hostname) and records per-hostname completion in SharedPreferences only on a success response. SyncInterface declares the external resetStaging and invokes resetStagingOnce inside the syncBoth executor task before syncBoth. AWPreferences gains isStagingResetDone/setStagingResetDone keyed by hostname. Adds unit tests for the runOnce flag logic.

Not safe to merge — 1 P1 open

Confidence 3/5

1 finding · ❌ 1 P1

❌ P1 high — mobile/src/main/java/net/activitywatch/android/StagingReset.kt:43

The resetStagingOnce function is called inside the syncBoth executor task, but it is not wrapped in a try-catch itself. StagingReset.runOnce catches Exception and LinkageError, but the reset lambda calls resetStaging(hostname), which is a JNI call. If the native library throws any other Error (e.g. an AssertionError from the native side, or a NoClassDefFoundError), it would escape runOnce and propagate up through the executor task. The executor task in performSyncAsync catches Exception, but not Error. An Error escaping would kill the executor thread before the callback clears syncInFlight, leaving syncInFlight true forever and rejecting all future syncs as 'already in flight'. The PR description explicitly mentions UnsatisfiedLinkError being caught, but other LinkageError subclasses are caught, yet other Errors are not. The comment in StagingReset says 'Never throws', but it only catches Exception and LinkageError, not all Throwable. A native crash or an OutOfMemoryError would still escape. This is a real defect: the sync executor can be permanently wedged by an Error from the native reset.

return try {
    if (isDone()) return Result.AlreadyDone
    // ...rest of current body...
} catch (e: Exception) { ... } catch (e: LinkageError) { ... }

_{How this was verified: Checked StagingReset.kt: catches Exception and LinkageError only. performSyncAsync in SyncInterface.kt catches Exception only (line 345). An Error from resetStaging would propagate out of the executor task, skipping the finally? Actually finally runs, but the callback is not posted, so syncInFlight remains true. The comment in StagingReset claims 'Never throws' but that is not true for Errors other than LinkageError.}

_{Consensus: 3/3 passes agreed}
Distinct keys: 1 (general)

Files changed (4) — the diff as I read it

• mobile/src/main/java/net/activitywatch/android/AWPreferences.kt — Adds per-hostname staging-reset-done flag getter/setter backed by SharedPreferences.
• mobile/src/main/java/net/activitywatch/android/StagingReset.kt — New object with runOnce() that invokes a reset lambda, parses JSON success, and marks done only on success; catches Exception and LinkageError.
• mobile/src/main/java/net/activitywatch/android/SyncInterface.kt — Declares external resetStaging, adds resetStagingOnce() and calls it before syncBoth in the full-sync executor task.
• mobile/src/test/java/net/activitywatch/android/StagingResetTest.kt — New unit tests covering runOnce success, already-done, native failure, malformed response, UnsatisfiedLinkError, and pref key.

_{Reviewed e0bcfec8d72e · openrouter/deepseek/deepseek-v4-flash-0731 · llm engine · 1195s · about this reviewer}

_{Maintainer commands}

_{@TimeToBuildBob review (own line) — fresh review · @TimeToBuildBob fix — a worker acts on the findings. Once per comment; 👀 = received.}