android webview: allow cc to fail hardware draw

cc/scheduler/scheduler_state_machine.cc

Index: cc/scheduler/scheduler_state_machine.cc
diff --git a/cc/scheduler/scheduler_state_machine.cc b/cc/scheduler/scheduler_state_machine.cc
index 661dc7aace4712f57995bb1be623b3bc953af1d8..4f2db8f50f73489852f8888a5bf0107292d4ee43 100644
--- a/cc/scheduler/scheduler_state_machine.cc
+++ b/cc/scheduler/scheduler_state_machine.cc
@@ -256,8 +256,6 @@ bool SchedulerStateMachine::PendingDrawsShouldBeAborted() const {
   // software draws could be scheduled by the Android OS at any time and draws
   // should not be aborted in this case.
   bool is_output_surface_lost = (output_surface_state_ == OUTPUT_SURFACE_NONE);
-  if (settings_.using_synchronous_renderer_compositor)
-    return is_output_surface_lost || !can_draw_;
 
   // These are all the cases where we normally cannot or do not want to draw
   // but, if needs_redraw_ is true and we do not draw to make forward progress,
@@ -265,7 +263,12 @@ bool SchedulerStateMachine::PendingDrawsShouldBeAborted() const {
   // This should be a superset of PendingActivationsShouldBeForced() since
   // activation of the pending tree is blocked by drawing of the active tree and
   // the main thread might be blocked on activation of the most recent commit.
-  return is_output_surface_lost || !can_draw_ || !visible_;
+  // Note deliverately not checking |visible_| here because |can_draw_| already
+  // incorporates |visible_| state.
+  // TODO(boliu): This contradicts comment above that

[boliu, 2015/10/09 18:09:51]

Question for Brian:

So this patch shoves resourceless software (which requires always drawing) and
visible into CanDraw. Seems like I can't really get away with scheduler being
oblivious to whether the current draw is resourceless software draw or not. I
had a DCHECK_IMPLIES(!visible_, !can_draw_); before, but then of course that's
obviously doesn't hold with resourceless software mode.

Are you happy with adding this to PendingActivationsShouldBeForced:
if (!PendingDrawsShouldBeAborted())
  return false;

The alternative is to plumb resourceless software draw through to the scheduler,
which means this patch needs a rewrite.

Any thoughts?

[boliu, 2015/10/09 18:39:53]

This goes a bit deeper, since resourceless software can flip between true and
false very often. Is it possible to force activate (!visible &&
!resourcless_software), and then resourceless flips to true, and the pending
draw is not aborted?

(Actually articulating that, not sure what's wrong with that picture..)

+  // PendingDrawsShouldBeAborted is a superset of
+  // PendingActivationsShouldBeForced.
+  return is_output_surface_lost || !can_draw_;
 }
 
 bool SchedulerStateMachine::PendingActivationsShouldBeForced() const {