Add viz-host HitTestQuery.

components/viz/host/hit_test/hit_test_query.cc

Index: components/viz/host/hit_test/hit_test_query.cc
diff --git a/components/viz/host/hit_test/hit_test_query.cc b/components/viz/host/hit_test/hit_test_query.cc
new file mode 100644
index 0000000000000000000000000000000000000000..22471f95975d593a37b15e02836c0c6f2f489c47
--- /dev/null
+++ b/components/viz/host/hit_test/hit_test_query.cc
@@ -0,0 +1,53 @@
+// Copyright 2017 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "components/viz/host/hit_test/hit_test_query.h"
+
+namespace viz {
+namespace hit_test {
+
+HitTestQuery::HitTestQuery() {}
+
+HitTestQuery::~HitTestQuery() {}
+
+Target HitTestQuery::FindTargetForLocation(const gfx::Point& location_in_root) {
+  // TODO(riajiang): Assumptions I'm making about the list of DisplayHitTestData
+  // 1. The list is in descending z-order.

[rjkroege, 2017/06/15 19:58:35]

so: larger indicies mean closer to the eyepoint?

[riajiang, 2017/06/16 02:56:54]

Yes, I'm assuming that for now for the correct transform and child count. Once
that's updated for ascending z-order I'll update the code.

+  // 2. Children entries have their own FrameSinkId, but those that are just

[rjkroege, 2017/06/15 19:58:35]

entries here mean "region"?

[riajiang, 2017/06/16 02:56:54]

Yes. Updated.

+  // bounding boxes have the FrameSinkId of their parent.
+  // 3. Children count includes children of children.
+  // 4. After applying transform to the incoming point, point is in the same

[rjkroege, 2017/06/15 19:58:35]

you need ascii art.  Or an associated .md file with pictures. These need more
specificity.

[riajiang, 2017/06/16 02:56:54]

Added ascii art example for now. Maybe will move all these assumptions to a doc
later?

+  // coordinate system as the bounds it is comparing against.
+  // 5. Transform contains transform for the position.
+  // 6. HIT_TEST_MINE means the region can receive events, i.e., not just a
+  // bounding box.
+  DCHECK(!display_hit_test_data_list_.empty());
+  Target confirmed_target_so_far;

[rjkroege, 2017/06/15 19:58:35]

You should use Target* instead of copying whole Target objects.

Target can be a class. 

[riajiang, 2017/06/16 02:56:54]

Do you mean confirmed_target_so_far is copying the value of
tentative_target_so_far? We do need to copy the value in this case? If they are
pointers then they would point to the same object but we don't want to update
confirmed_target_so_far every time?

+  Target tentative_target_so_far;
+  tentative_target_so_far.location_in_target = location_in_root;

[rjkroege, 2017/06/15 19:58:35]

I'd put a blank line here.

[riajiang, 2017/06/16 02:56:54]

Done.

+  for (uint32_t i = 0; i < display_hit_test_data_list_.size(); ++i) {

[rjkroege, 2017/06/15 19:58:35]

use an iterator

[riajiang, 2017/06/16 02:56:54]

Done.

+    DisplayHitTestData data = display_hit_test_data_list_[i];
+    gfx::Point location(tentative_target_so_far.location_in_target);

[gklassen, 2017/06/15 19:51:20]

What happens if we transform a point for a child region that doesn't match?  Eg.
If A embeds B which embeds C and we test a point that overlaps all three but
does not match A or B - then the target is A and we need to return the point as
transformed to the coordinate space of A. 

We may need to use a recursive call or stack to handle this?

[riajiang, 2017/06/16 02:56:54]

Do you maybe mean that the point P does not match *C* or B and the target should
be A? This location conversion is for comparing the point with bounds (so they
are in the same coordinate system), but we only update the location if it's
actually inside the bounds (line 36)? I think I'll need to update this once we
change it to deal with front-to-back z-order.

+    if (!data.transform.IsIdentity())
+      data.transform.TransformPoint(&location);
+    if (data.bounds.Contains(location)) {
+      tentative_target_so_far.id = data.id;
+      tentative_target_so_far.location_in_target = location;

[rjkroege, 2017/06/15 19:58:36]

I think it's safe to begin by thinking about this in a purely 2d context but
this needs additional thought in a 3d context.

+      tentative_target_so_far.flags = data.flags;
+      // TODO(riajiang): Check properly once HitTestRegionFlags is well defined
+      // after https://codereview.chromium.org/2908783002/.
+      // HIT_TEST_MINE is set to be true, which means it can receive events.

[rjkroege, 2017/06/15 19:58:35]

if you encounter a region with hit_test_mine set, you should exit?

[riajiang, 2017/06/16 02:56:54]

We should also check child_count? If a region has hit_test_mine but still has
children left then we want to keep walking?

+      if (data.flags)
+        confirmed_target_so_far = tentative_target_so_far;
+      if (!data.child_count)
+        break;
+    } else {
+      i += data.child_count;
+    }
+  }
+  return confirmed_target_so_far;
+}
+
+}  // namespace hit_test
+}  // namespace viz