Sync aborts the app process (SIGABRT in libaw_sync.so) on Android 16 — syncBoth and syncPush alike

[Lorite]

Summary

With sync enabled, the app is killed by SIGABRT roughly half a second into every sync cycle. The abort happens inside libaw_sync.so, so performSyncAsync never returns from syncFn() and the SAF mirror added in #209 is never reached — the sync directory is populated internally but nothing is ever copied to the user-chosen SAF directory.

This affects both JNI entry points: syncBoth and syncPush abort identically, so it is not specific to the pull path.

Environment

• Device: OnePlus Nord CE 3 5G (CPH2493 / OP556FL1), arm64
• Android 16 (BP2A.250605.015)
• Built from master @ 53a2fb6 (includes feat(sync): add SyncSettingsActivity — enable/disable toggle + SAF directory picker #204 and feat(sync): bridge SAF-granted URI to aw-sync output (Phase 2) #209), debug variant
• libaw_sync.so / libaw_server.so taken unmodified from the CI Build workflow artifacts

Steps to reproduce

1. Install a build containing feat(sync): add SyncSettingsActivity — enable/disable toggle + SAF directory picker #204 + feat(sync): bridge SAF-granted URI to aw-sync output (Phase 2) #209.
2. Open Sync Settings, enable the toggle, choose an SAF directory in shared storage (e.g. /sdcard/ActivityWatchSync).
3. Wait ~1 minute for SyncScheduler's first sync.

Observed

08:04:29.565 I SyncInterface: Starting sync operation: Full Sync
08:04:30.277 F DEBUG   : signal 6 (SIGABRT), code -1 (SI_QUEUE), fault addr --------
08:04:30.277 F DEBUG   : pid: 25108, tid: 25696, name: pool-6-thread-1
08:04:30.277 F DEBUG   :   #00 pc 000000000008b70c  libc.so (abort+156)
08:04:30.277 F DEBUG   :   #01 pc 00000000012eb124  base.apk (offset 0x2f30000)
08:04:30.277 F DEBUG   :   #02 pc 00000000012e7d5c  base.apk (offset 0x2f30000)
08:04:30.277 F DEBUG   :   #03 pc 00000000012d8c80  base.apk (offset 0x2f30000)
08:04:30.277 F DEBUG   :   #04 pc 00000000012e81f8  base.apk (offset 0x2f30000)
08:04:30.277 F DEBUG   :   #05 pc 00000000012d7fa4  base.apk (offset 0x2f30000)
08:04:30.277 F DEBUG   :   #15 pc 00000000006d4c28  base.apk (Java_net_activitywatch_android_SyncInterface_syncBoth+260)
08:04:30.277 F DEBUG   :   #25 ... SyncInterface$syncBothAsync$2.invoke
08:04:30.277 F DEBUG   :   #33 ... SyncInterface.performSyncAsync$lambda$4

Patching SyncScheduler.performSync() to call syncPushAsync instead produces the same abort, twice in one capture:

08:14:10.059 I SyncInterface: Starting sync operation: Push
08:14:10.579 F libc    : Fatal signal 6 (SIGABRT) in tid 8024 (pool-6-thread-1)
             #15 ... (Java_net_activitywatch_android_SyncInterface_syncPush+260)
08:15:23.946 I SyncInterface: Starting sync operation: Push
08:15:24.024 F libc    : Fatal signal 6 (SIGABRT) in tid 9839 (pool-5-thread-1)
             #15 ... (Java_net_activitywatch_android_SyncInterface_syncPush+260)
08:15:24.528 I Zygote  : Process 8448 exited due to signal 6 (Aborted)

Note both entry points abort at the same +260 offset.

What this narrows it down to

The staging tree is created before the crash:

files/sync/<hostname>/<device-id>/test.db   45056 bytes

In aw-sync/src/sync.rs, sync_run fetches info from the local server, takes info.device_id, and only then calls setup_local_remote(path, device_id). Since the directory is named with the correct device id, both the get_info() call against the in-app aw-server-rust and the datastore creation must have succeeded. That places the abort downstream of both, in the bucket-sync step itself rather than in startup, the HTTP client, or directory setup.

SIGABRT with no fault address is consistent with a Rust panic, but the panic message never reaches logcat (Rust writes panics to stderr, which Android discards), and the shipped .so has no .debug_info, so addr2line only resolves to the nearest exported symbol. A debug build, or routing the Rust panic hook to android_log, would identify the frame immediately.

Suggested follow-ups

1. Install a panic hook that logs via android_log so panics are diagnosable from logcat at all — useful well beyond this bug.
2. Publish a symbol-bearing build (or upload symbols) so tombstones can be symbolicated.

Secondary observations

Two smaller things found while investigating, happy to open separate issues or PRs if useful:

SyncSettingsActivity is unreachable from the UI. It is registered only in res/menu/main.xml (the options menu), but MainActivity never calls setSupportActionBar(), the theme is AppTheme.NoActionBar, and the Toolbar in res/layout/app_bar_main.xml has been commented out since 4a02a0b6 (2018). So onCreateOptionsMenu inflates a menu with no host and no overflow button is rendered, while activity_main_drawer.xml has no sync entry. Adding a nav_sync_settings item to the drawer (mirroring nav_auth_settings) makes it reachable.

The SAF mirror in #209 copies nothing even when sync succeeds. copySyncFilesToSafDir() does a flat File(syncDir).listFiles()?.filter { it.isFile }, but aw-sync never writes a regular file at the root of the sync directory — setup_local_remote writes <sync_dir>/<device_id>/test.db, and on Android the observed tree is <sync_dir>/<hostname>/<device_id>/test.db. So the filter matches zero entries. The receiving side also requires the nesting: find_remotes in aw-sync/src/util.rs keeps only directories and looks for *.db one level inside them, so a flattened copy would be ignored even if it were made. A recursive, structure-preserving mirror fixes both halves.