Hold video frame in Bitmap instead of keeping a ByteBuffer reference.

remoting/client/jni/chromoting_jni_runtime.cc

Index: remoting/client/jni/chromoting_jni_runtime.cc
diff --git a/remoting/client/jni/chromoting_jni_runtime.cc b/remoting/client/jni/chromoting_jni_runtime.cc
index 108bb71339d937272e1885c7dd39b03e04d73a1a..7aa5302c7bedc0aa21e2ecee6ba5b9b449d5191f 100644
--- a/remoting/client/jni/chromoting_jni_runtime.cc
+++ b/remoting/client/jni/chromoting_jni_runtime.cc
@@ -171,22 +171,23 @@ void ChromotingJniRuntime::CommitPairingCredentials(const std::string& host,
 
 void ChromotingJniRuntime::UpdateImageBuffer(int width,
                                              int height,
-                                             jobject buffer) {
+                                             uint8* data) {
   DCHECK(display_task_runner_->BelongsToCurrentThread());
 
   JNIEnv* env = base::android::AttachCurrentThread();
-  env->SetStaticIntField(
-      class_,
-      env->GetStaticFieldID(class_, "sWidth", "I"),
-      width);
-  env->SetStaticIntField(
-      class_,
-      env->GetStaticFieldID(class_, "sHeight", "I"),
-      height);
-  env->SetStaticObjectField(
+
+  base::android::ScopedJavaLocalRef<jobject> buffer(env,
+      env->NewDirectByteBuffer(data, width * height * kBytesPerPixel));

[Sergey Ulanov, 2013/09/25 23:32:08]

frame stride may not be equal width*kBytesPerPixel (though it's not the case
with BasicDesktopFrame). I think you also need to pass frame stride to
setVideoFrame.

[solb, 2013/09/26 14:21:36]

I can see the argument for this, but at the same time, the setVideoFrame()
method's choice of BitmapConfig assumes kBytesPerPixel will always be 4, and
supporting different values would require additional code in that method to
handle cases we don't need.  At the very most, we'd be using the passed value to
print an error/warning if it didn't match our expectations.

[Lambros, 2013/09/26 22:29:37]

It's possible to deal with strides that are different from the width, but it may
be less efficient. The Bitmap ctors that accept a stride argument also require
passing an int[] array holding the pixels. That means the new Bitmap will read
and process the pixels from int[], as well as during copyPixelsFromBuffer().

Also, Bitmap expects a stride measured in pixels but DesktopFrame exposes its
stride in bytes. So we'd have to divide it by 4 before passing it to Java (and
DCHECK that it's a multiple of 4).

I'm not sure if it's worth supporting different strides, so I've just added a
DCHECK that the stride is width * kBytesPerPixel.

+
+  env->CallStaticVoidMethod(
       class_,
-      env->GetStaticFieldID(class_, "sBuffer", "Ljava/nio/ByteBuffer;"),
-      buffer);
+      env->GetStaticMethodID(
+          class_,
+          "setVideoFrame",
+          "(IILjava/nio/ByteBuffer;)V"),
+      width,
+      height,
+      buffer.obj());
 }
 
 void ChromotingJniRuntime::UpdateCursorShape(