content: Fix Context creation logic in ContextProviderFactoryImpl.

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 bc9028a6bbc7b25838370f3dbdc5f75757075abd..a025634c44963e3efe91b07a7931b5c238e8706f 100644
--- a/content/browser/renderer_host/compositor_impl_android.cc
+++ b/content/browser/renderer_host/compositor_impl_android.cc
@@ -648,26 +648,44 @@ void CompositorImpl::CreateVulkanOutputSurface() {
 #endif
 
 void CompositorImpl::CreateCompositorOutputSurface(
-    const scoped_refptr<cc::ContextProvider>& context_provider) {
-  // This callback should run only if we have a pending output surface request,
-  // since that is when we should have queued a context request.
-  // In case the surface was invalidated after the context request was queued,
-  // the request should have been dropped by the ContextProviderFactory.
+    const scoped_refptr<cc::ContextProvider>& context_provider,
+    ui::ContextProviderFactory::ContextCreationResult result) {
   DCHECK(output_surface_request_pending_);
-  DCHECK(host_->visible());
-  DCHECK(window_);
-  DCHECK_NE(surface_handle_, gpu::kNullSurfaceHandle);
-  DCHECK(context_provider);
-
-  scoped_refptr<ContextProviderCommandBuffer> context_provider_command_buffer =
-      static_cast<ContextProviderCommandBuffer*>(context_provider.get());
-  std::unique_ptr<cc::OutputSurface> display_output_surface(
-      new OutputSurfaceWithoutParent(
-          context_provider_command_buffer,
-          base::Bind(&CompositorImpl::PopulateGpuCapabilities,
-                     base::Unretained(this))));
-  InitializeDisplay(std::move(display_output_surface), nullptr,
-                    std::move(context_provider));
+
+  switch (result) {
+    // Don't retry if we are shutting down or if the Gpu Surface handle was
+    // lost. The Gpu Surface handle loss should happen only if we are invisible
+    // or this was from a previous request and the current surface has changed,
+    // in which case we would have made another request with the factory.
+    case ui::ContextProviderFactory::ContextCreationResult::
+        FAILURE_BROWSER_SHUTDOWN:
+    case ui::ContextProviderFactory::ContextCreationResult::
+        FAILURE_GPU_SURFACE_HANDLE_LOST:
+      break;
+    case ui::ContextProviderFactory::ContextCreationResult::
+        FAILURE_GPU_PROCESS_INITIALIZATION_FAILED:
+      // Retry only if we are visible.

[Khushal, 2016/08/17 21:24:56]

Should there be a check for consecutive failures? The way we do for consecutive
failures to initialize the output surface.

[no sievers, 2016/08/17 21:34:39]

Yes, I think a LOG(FATAL) here would be good if this happens twice in a row or
so.

[Khushal, 2016/08/17 21:55:54]

Done.

+      if (host_->visible())
+        HandlePendingOutputSurfaceRequest();
+      break;
+    case ui::ContextProviderFactory::ContextCreationResult::SUCCESS:
+      DCHECK(host_->visible());
+      DCHECK(window_);
+      DCHECK_NE(surface_handle_, gpu::kNullSurfaceHandle);
+      DCHECK(context_provider);
+      scoped_refptr<ContextProviderCommandBuffer>
+          context_provider_command_buffer =
+              static_cast<ContextProviderCommandBuffer*>(
+                  context_provider.get());
+      std::unique_ptr<cc::OutputSurface> display_output_surface(
+          new OutputSurfaceWithoutParent(
+              context_provider_command_buffer,
+              base::Bind(&CompositorImpl::PopulateGpuCapabilities,
+                         base::Unretained(this))));
+      InitializeDisplay(std::move(display_output_surface), nullptr,
+                        std::move(context_provider));
+      break;
+  }
 }
 
 void CompositorImpl::InitializeDisplay(