Fix BGTask SIGSEGV from re-running php_embed_init under a live runtime

[shanerbaner82]

Problem

A WorkManager scheduler job (PHPSchedulerWorker) firing while the app process is alive SIGSEGVs in php_tsrm_startup_ex. Tombstone:

PHPSchedulerWorker.doWork → LaravelEnvironment.initializeForBackground → runBaseArtisanCommands
  → native_run_artisan_command → php_embed_init → php_tsrm_startup_ex → ts_resource_ex
  → zend_ini_refresh_caches → OnUpdateBool → null-pointer deref

The worker's initializeForBackground() runs runBaseArtisanCommands() (optimize:clear, storage:unlink, storage:link, migrate --force) whenever extractLaravelBundle() reports a change. Those route through native_run_artisan_command, which called php_embed_init() unconditionally. When the persistent runtime and/or the database queue worker already hold a process-wide TSRM context, the second php_embed_init() re-runs php_tsrm_startup_ex → zend_ini_refresh_caches → OnUpdateBool against live globals and null-derefs. WorkManager then retry-loops it (~2 min apart). It only stops if both the persistent runtime and the queue worker are disabled.

In DEBUG builds (NATIVEPHP_APP_VERSION="DEBUG") every extraction reports a change, so this fires on every background tick.

Fix

1. native_run_artisan_command is now TSRM-aware (php_bridge.c), mirroring ephemeral_embed_init:

   • wait_for_persistent_boot_settled() to avoid racing a mid-boot persistent thread;
   • hot path (a runtime is live): ts_resource(0) + php_embed_module.startup() + php_request_startup(), teardown via php_request_shutdown(NULL) + ts_free_thread() — leaving the global TSRM and php_initialized intact;
   • cold path (no live runtime — install-time setup or a WorkManager cold start): unchanged full php_embed_init() / safe_php_embed_shutdown().

   This is the root-cause fix and makes the classic artisan path safe for every caller.

2. Warm-path guard — new nativeIsPersistentRuntimeLive() JNI accessor (reads the process-wide persistent_initialized) lets initializeForBackground() skip the install-time artisan commands when a persistent runtime is already live — those were already run by MainActivity at app boot. The C fix remains the backstop if the guard ever races a mid-boot runtime.

Files

• resources/androidstudio/app/src/main/cpp/php_bridge.c — hot/cold TSRM split + JNI accessor + registration
• resources/androidstudio/app/src/main/java/com/nativephp/mobile/bridge/PHPBridge.kt — nativeIsPersistentRuntimeLive binding
• resources/androidstudio/app/src/main/java/com/nativephp/mobile/bridge/LaravelEnvironment.kt — guard in initializeForBackground()

Testing

Verified on an Android emulator with a DEBUG build, runtime_mode: persistent, and QUEUE_CONNECTION=database: firing a scheduled background task while the app is open no longer SIGSEGVs; logcat shows ⏭️ Skipping base artisan commands — persistent runtime already live. Reverting the fix reproduces Fatal signal 11 (SIGSEGV) in php_tsrm_startup.

iOS is unaffected — its scheduler already routes background work through ephemeral_php_boot / ephemeral_embed_init, which attaches to the existing TSRM.

🤖 Generated with Claude Code