Unit Test for WebView animating in and out of screen

android_webview/browser/browser_view_renderer_unittest.cc

Index: android_webview/browser/browser_view_renderer_unittest.cc
diff --git a/android_webview/browser/browser_view_renderer_unittest.cc b/android_webview/browser/browser_view_renderer_unittest.cc
index b4a9d42227241cef82d61d71a47608eb57d52001..a65aedb910a076312cfbe62b12012106d6385f99 100644
--- a/android_webview/browser/browser_view_renderer_unittest.cc
+++ b/android_webview/browser/browser_view_renderer_unittest.cc
@@ -59,4 +59,80 @@ class ClearViewTest : public RenderingTest {
 
 RENDERING_TEST_F(ClearViewTest);
 
+// - The first ondraw is onscreen. It reads the default draw constraints. The
+//   first DrawGL will post the new constraints to UI thread.
+// - The second onDraw is offscreen with an empty global visible rect. It reads
+//   the new draw constraints. The second DrawGL will be skipped because the
+//   WebView is offscreen.
+// - The third onDraw is the result of force invalidate from the second onDraw.
+//   It reads the same draw constraints as the second onDraw does.
+class TestAnimateInAndOutOfScreen : public RenderingTest {
+ public:
+  TestAnimateInAndOutOfScreen() : on_draw_count_(0u) {
+    new_transform_.Scale(2.0, 2.0);
+  }
+
+  void StartTest() override {
+    browser_view_renderer_->SetContinuousInvalidate(true);
+  }
+
+  void WillOnDraw() override {
+    on_draw_count_++;
+    if (on_draw_count_ == 2u) {
+      browser_view_renderer_->PrepareToDraw(gfx::Vector2d(), gfx::Rect());
+      browser_view_renderer_->SetContinuousInvalidate(false);

[boliu, 2015/03/13 23:51:55]

Hmm, doesn't posting of the next OnDrawHardware still race with posting the
updated DrawConstraints from RT?

Feels like we need a hook for when DrawConstraints is updated in BVR? Not
trivial to implement that one..

We can subclass BVR and make that method virtual, then the TestBVR can add in
the hook..

[hush (inactive), 2015/03/16 18:19:21]

It does not race I think. Both are post tasks to UI.
The previous DrawGL will post a task to set parent draw constraints on UI. Then
the condition variable signals, and the next FakeWindow::onDrawHardware is
posted by PostInvalidate.
So we're sure by the time onDrawHardware is called, the parent draw constraints
are already updated.

We're lucky that PostInvalidate is not immediate...
There is one more thing to address to make onDraw and DrawGL absolutely
interlaced and not racy: PostInvalidate could go ahead of DrawGL. I will take
care of that in the next path.

[boliu, 2015/03/16 20:25:08]

PostInvalidate happen at end of BVR::OnDrawHardware, which is definitely not the
end of FakeWindow::OnDrawHardware. From staring at the code, it's happening in
exactly the wrong order we want, that PostTask(&OnDrawHardware) happens before
PostTask(&UpdateParentDrawConstraints), no?

But in general, there are no order defined between these two events. So it's not
safe to assume they are ordered, because it might break in the future.

[hush (inactive), 2015/03/17 18:26:45]

Okay. I added a callback in browser view renderer client informing the test when
the parent draw constraints get updated.
And removed the bits about "wait for mode draw to finish"

+    }
+  }
+
+  void DidOnDraw(bool success) override {
+    ParentCompositorDrawConstraints initial_constraints;
+    ParentCompositorDrawConstraints new_constraints(
+        false, new_transform_, gfx::Rect(window_->surface_size()));
+    EXPECT_TRUE(success);
+    if (on_draw_count_ == 1u) {
+      EXPECT_TRUE(initial_constraints.Equals(
+          browser_view_renderer_->parent_draw_constraints_for_testing()));
+    }
+
+    if (on_draw_count_ == 2u) {
+      EXPECT_TRUE(new_constraints.Equals(
+          browser_view_renderer_->parent_draw_constraints_for_testing()));
+    }
+
+    if (on_draw_count_ == 3u) {
+      EXPECT_TRUE(new_constraints.Equals(
+          browser_view_renderer_->parent_draw_constraints_for_testing()));

[boliu, 2015/03/13 23:51:55]

DrawGL for on_draw_count_ == 3 uses the identity matrix again. Should wait for
that to come back to BVR before ending the test I think?

[hush (inactive), 2015/03/16 18:19:21]

Well.. the 4th onDraw will early out in "Previous frame unconsumed" because the
3rd DrawGL is skipped. So the 4th onDraw will not have the identity, until your
CL to remove the blocking pipeline is checked in.

For the time being,  I just add a logic to update the parent draw constraints
when the previous frame is unconsumed so that test will pass. And the test will
continue passing after your CL lands.

[boliu, 2015/03/16 20:25:08]

Hmm. Let's not change production code just to make a test pass.

Add a comment here instead, and I'll take care of it in my patch.

[hush (inactive), 2015/03/17 18:26:45]

Done.

+      EndTest();
+    }
+  }
+
+  void SetParentDrawConstraints(AwDrawGLInfo& draw_info) override {
+    draw_info.width = window_->surface_size().width();
+    draw_info.height = window_->surface_size().height();
+    draw_info.is_layer = false;
+
+    gfx::Transform transform;
+    if (on_draw_count_ == 1u)
+      transform = new_transform_;
+
+    transform.matrix().asColMajorf(draw_info.transform);
+  }
+
+  bool WillDrawOnRT(SharedRendererState* functor) override {
+    // The 2nd onDraw gets skipped because the WebView is offscreen.
+    if (on_draw_count_ == 2u)
+      return false;
+
+    return true;
+  }
+
+  bool WillWaitForModeDrawToFinish() override { return true; }
+
+ private:
+  size_t on_draw_count_;
+  gfx::Transform new_transform_;
+};
+
+RENDERING_TEST_F(TestAnimateInAndOutOfScreen);
+
 }  // namespace android_webview