ui: Move software fallback decisions to GpuProcessTransportFactory.

ui/compositor/compositor.cc

Index: ui/compositor/compositor.cc
diff --git a/ui/compositor/compositor.cc b/ui/compositor/compositor.cc
index 4793cbb136113dd161f7d323548426d5c845d84f..287eaf19751ffc5e0d4f4fe09826482f812fef7c 100644
--- a/ui/compositor/compositor.cc
+++ b/ui/compositor/compositor.cc
@@ -76,7 +76,6 @@ Compositor::Compositor(gfx::AcceleratedWidget widget,
       device_scale_factor_(0.0f),
       last_started_frame_(0),
       last_ended_frame_(0),
-      num_failed_recreate_attempts_(0),
       disable_schedule_composite_(false),
       compositor_lock_(NULL),
       layer_animator_collection_(this),
@@ -302,27 +301,16 @@ void Compositor::Layout() {
 }
 
 void Compositor::RequestNewOutputSurface() {
-  bool fallback =
-      num_failed_recreate_attempts_ >= OUTPUT_SURFACE_RETRIES_BEFORE_FALLBACK;
-  context_factory_->CreateOutputSurface(weak_ptr_factory_.GetWeakPtr(),
-                                        fallback);
+  context_factory_->CreateOutputSurface(weak_ptr_factory_.GetWeakPtr());
 }
 
 void Compositor::DidInitializeOutputSurface() {
-  num_failed_recreate_attempts_ = 0;
 }
 
 void Compositor::DidFailToInitializeOutputSurface() {
-  num_failed_recreate_attempts_++;
-
-  // Tolerate a certain number of recreation failures to work around races
-  // in the output-surface-lost machinery.
-  if (num_failed_recreate_attempts_ >= MAX_OUTPUT_SURFACE_RETRIES)
-    LOG(FATAL) << "Failed to create a fallback OutputSurface.";
-
-  base::MessageLoop::current()->PostTask(
-      FROM_HERE, base::Bind(&Compositor::RequestNewOutputSurface,
-                            weak_ptr_factory_.GetWeakPtr()));
+  // The OutputSurface should already be bound/initialized before being given to
+  // the Compositor.
+  NOTREACHED();

[no sievers, 2015/02/25 01:51:00]

Although there is also |worker_context_provider_| being initialized in
OutputSurface::BindToClient() which could still fail.

Maybe we shouldn't make the assumption here about what else LTH(Impl)
initializes (or what GLRenderer does) that could fail and trigger this callback?

It would be a problem if it happens to create an infinite retry loop without
generating a crash report, I guess.

Should this maybe just also have a simple counter for successive attempts? (for
example reset in DidInitializeOutputSurface)

[enne (OOO), 2015/02/25 19:37:33]

I think it'd be fine to document that this can only be called if the output
surface fails to bind.  That's definitely the case right now.

[no sievers, 2015/02/25 20:52:12]

Sounds good then. I guess it is in the name :)

 }
 
 void Compositor::DidCommit() {