Animated scroll shouldn't consume unhandled scrolls for OOPIFs.

content/browser/site_per_process_browsertest.cc

Index: content/browser/site_per_process_browsertest.cc
diff --git a/content/browser/site_per_process_browsertest.cc b/content/browser/site_per_process_browsertest.cc
index cb7a479d278c1781ffa6b7d373b786eb0e27a4cf..13533cabdfa12391d9dbf65438a69a5821b36dfc 100644
--- a/content/browser/site_per_process_browsertest.cc
+++ b/content/browser/site_per_process_browsertest.cc
@@ -1091,8 +1091,15 @@ IN_PROC_BROWSER_TEST_F(SitePerProcessBrowserTest,
 // Test that scrolling a nested out-of-process iframe bubbles unused scroll
 // delta to a parent frame.
 // Flaky: https://crbug.com/627238

[bokan, 2017/04/11 22:18:29]

If this used to be flaky, it means the functionality worked, right? So does that
mean that this test isn't testing bubbling from animated wheel scrolls? If so,
we need a test for that explicitly. If not, I'm confused.

[wjmaclean, 2017/04/12 15:54:47]

Yes and no. :-)
No.
It's a confusing story, but here goes:

First, the existing test (without this CL) works for --disable-smooth-scrolling.
Now, I've discovered that sometimes when the test runs, the scrolls sent to the
child *aren't* animated, and they bubble as expected. But on other runs, the
child scrolls are animated, and then the test fails without the code added by
this CL at the end of ScrollAnimated(), which sets the status to SCROLL_IGNORED
for unconsumed scrolls in children.

*** I do not know why the scrolls sent to the child have unpredictable behaviour
vis-á-vis animating or not.

A third failure mode exists with this CL enabled, namely that the animation
bubbled from the child hits a current animation in the parent. Normally this
should just combine and work, but it seems the new scroll delta gets ignored,
and the test timeouts waiting for the delta to be applied. I've filed a separate
bug for that case. The delay I've added to the test is meant to minimise that
case, but though it works well on my development machine, it still fails far too
often on the bots, so the test may have to be delayed until the
combining-animations bug is resolved.

+#if defined(OS_ANDROID)
+// Test failing on some Android builders, due to inaccurate coordinates on
+// some devices. See: https://crbug.com/700007.
+#define MAYBE_ScrollBubblingFromOOPIFTest DISABLED_ScrollBubblingFromOOPIFTest
+#else
+#define MAYBE_ScrollBubblingFromOOPIFTest ScrollBubblingFromOOPIFTest
+#endif
 IN_PROC_BROWSER_TEST_F(SitePerProcessBrowserTest,
-                       DISABLED_ScrollBubblingFromOOPIFTest) {
+                       MAYBE_ScrollBubblingFromOOPIFTest) {
   GURL main_url(embedded_test_server()->GetURL(
       "a.com", "/cross_site_iframe_factory.html?a(b)"));
   EXPECT_TRUE(NavigateToURL(shell(), main_url));
@@ -1192,6 +1199,17 @@ IN_PROC_BROWSER_TEST_F(SitePerProcessBrowserTest,
 
   filter->Reset();
 
+  // TODO(wjmaclean): This is ugly, but it seems that it stops the scroll
+  // we're about to send from occassionally colliding with the previous
+  // scroll animation.
+  {
+    base::RunLoop run_loop;
+    base::ThreadTaskRunnerHandle::Get()->PostDelayedTask(
+        FROM_HERE, run_loop.QuitClosure(), TestTimeouts::action_max_timeout());
+    run_loop.Run();
+    update_rect = filter->last_rect();
+  }
+
   // Scroll the parent down again in order to test scroll bubbling from
   // gestures.
   scroll_event.delta_y = -5.0f;