Right now, gfx code uses AndroidBridge directly to access the widget/Java layer code, but we need a solution that works with individual widgets.

To make code simpler on the C++ side, this patch adds a CreateEGLSurface call in GeckoLayerClient, so that the C++ side only has to deal with managing GeckoLayerClient.

mozilla::widget::ILayerClient will be the interface that C++ gfx code uses to interact with Java gfx code, through an intermediary in LayerClientBridge. Using a standalone interface lets us support multiple windows, whereas the existing approach using AndroidBridge calls does not support multiple windows.

Switch existing graphics code from using AndroidBridge calls to using ILayerClient calls, using an instance obtained from nsIWidget. Because this is an Android-specific interface, I didn't want to add a getter to nsIWidget itself. Also, this interface is used on the compositor thread, so the implementation has to be thread-safe, which means we cannot just have nsWindow implement the interface. Due to these reasons, this patch uses nsIWidget::GetNativeData as the means to get the interface from widget.

Remove the AndroiBridge calls made obsolete by the new ILayerClient interface.

Comment on attachment 8692031 [details] [diff] [review] Use ILayerClient instead of AndroidBridge (v1) Review of attachment 8692031 [details] [diff] [review]: ----------------------------------------------------------------- ::: gfx/layers/composite/AsyncCompositionManager.cpp @@ +1388,5 @@ > +{ > + Compositor* const compositor = mLayerManager->GetCompositor(); > + MOZ_ASSERT(compositor); > + > + nsIWidget* const widget = compositor->GetWidget(); Note that this gets run on the compositor thread so you should probably use a RefPtr for holding the widget at least.

(In reply to Kartikaya Gupta (email:kats@mozilla.com) from comment #5) > Comment on attachment 8692031 [details] [diff] [review] > Use ILayerClient instead of AndroidBridge (v1) > > Review of attachment 8692031 [details] [diff] [review]: > ----------------------------------------------------------------- > > ::: gfx/layers/composite/AsyncCompositionManager.cpp > @@ +1388,5 @@ > > +{ > > + Compositor* const compositor = mLayerManager->GetCompositor(); > > + MOZ_ASSERT(compositor); > > + > > + nsIWidget* const widget = compositor->GetWidget(); > > Note that this gets run on the compositor thread so you should probably use > a RefPtr for holding the widget at least. I tried that, but turns out nsWindow's ref-counting is non-threadsafe. We could switch it to be threadsafe, but I wasn't sure how much of a perf impact that would be. In Compositor, the widget is stored as a raw pointer presumably for the same reason.

Jim and I discussed, and we shouldn't need the ILayerClient once we kill all the Java stuff.

Gonna merge this bug with others.