[DeviceService] Add service tests for VibrationManager.

services/device/android/register_jni.cc

Index: services/device/android/register_jni.cc
diff --git a/services/device/android/register_jni.cc b/services/device/android/register_jni.cc
index 345a8699e79c2e501e797720985e6f9863839f3d..8c3c5a5e68d87f290fe6990f5c63b4eeb7d2e026 100644
--- a/services/device/android/register_jni.cc
+++ b/services/device/android/register_jni.cc
@@ -5,16 +5,33 @@
 #include "services/device/android/register_jni.h"
 
 #include "base/android/jni_android.h"
+#include "base/android/jni_registrar.h"
+#include "mojo/android/system/core_impl.h"
+#include "mojo/android/system/watcher_impl.h"
 #include "services/device/time_zone_monitor/android/time_zone_monitor_jni_registrar.h"
 
 namespace device {
 
+namespace {
+
+const base::android::RegistrationMethod kRegisteredMethods[] = {

[leonhsl(Using Gerrit), 2017/03/28 07:27:52]

We must register these JNIs to enable
InterfaceRegistrar::createInterfaceRegistryForContext to work well when running
service_unittests apk on Android.

Of course when Device Service is embedded by content_browser, content_browser
also registers these JNIs, but I think duplicated registration here has no bad
effect and Device Service should register all its necessary JNIs by itself.
WDYT? Thanks.

[blundell, 2017/03/28 12:40:47]

It seems like we should put this somewhere inside //mojo where it will just
naturally be depended on by any target that needs it, similar to e.g.
base_jni_registrar.* in //base. Is there some reason that won't work?

[leonhsl(Using Gerrit), 2017/03/29 07:37:16]

To make sure I understand the suggestion clearly:

When any one target X needs to use JNIs of //base, usually
  - target X depends on //base which packages base_jni_registrar.* and needed
C++ sources, and calls base::android::RegisterJni() on initiation.
  - corresponding android_library target X_java depends on //base:base_java
which packages needed java sources.

For Mojo case, it's similar that
  - X needs to depend on //mojo/android:libsystem_java which packages needed C++
sources, and calls both mojo::android::RegisterCoreImpl() and
mojo::android::RegisterWatcherImpl().
  - android_library X_java needs to depend on //mojo/android:system_java which
packages needed java sources.

So, the only difference with //base case is that //mojo/android:libsystem_java
is not exposing a mojo::android::RegisterJni() API consolidating calls to
mojo::android::RegisterCoreImpl() and mojo::android::RegisterWatcherImpl().

Do you mean that we'd better provide such a mojo::android::RegisterJni()?
Thanks!
If so I'll create another CL to do this task because it will also touch other
//mojo/android users unrelated with Device Service.

+    {"CoreImpl", mojo::android::RegisterCoreImpl},
+    {"WatcherImpl", mojo::android::RegisterWatcherImpl},
+};
+
+}  // namespace
+
 bool EnsureJniRegistered() {
   static bool g_jni_init_done = false;
 
   if (!g_jni_init_done) {
     JNIEnv* env = base::android::AttachCurrentThread();
 
+    if (!RegisterNativeMethods(env, kRegisteredMethods,
+                               arraysize(kRegisteredMethods))) {
+      return false;
+    }
+
     if (!android::RegisterTimeZoneMonitorJni(env))
       return false;