Fixing flakiness of TextureLayerChangeInvisibleMailboxTest

content/browser/renderer_host/compositor_impl_android.cc

Index: content/browser/renderer_host/compositor_impl_android.cc
diff --git a/content/browser/renderer_host/compositor_impl_android.cc b/content/browser/renderer_host/compositor_impl_android.cc
index c98051dcee1e175c034707bf08fd6c501bc77845..6b551f242ea4ca057c5d5b6ce6385e5dda880e1a 100644
--- a/content/browser/renderer_host/compositor_impl_android.cc
+++ b/content/browser/renderer_host/compositor_impl_android.cc
@@ -806,6 +806,8 @@ void CompositorImpl::DidSubmitCompositorFrame() {
 
 void CompositorImpl::DidReceiveCompositorFrameAck() {
   TRACE_EVENT0("compositor", "CompositorImpl::DidReceiveCompositorFrameAck");
+  if (!has_compositor_frame_sink_)

[danakj, 2017/04/04 22:14:47]

This would be racey if CompositorImpl can has_compositor_frame_sink_ again
before the ack arrives, which is why I suggested the weakptrfactory to cancel
the callback mid-flight. But maybe you can convince us that that wouldn't
happen?

[Saman Sami, 2017/04/04 22:23:47]

Interesting. No, it's not obvious to me why it won't be racey. I'll implement
this tomorrow.

+    return;
   DCHECK_GT(pending_frames_, 0U);
   pending_frames_--;
   client_->DidSwapFrame(pending_frames_);